enhance: standardize error handling on merr + Sys/Input classification (#50221)

issue: #47420

## What this PR does

Project-wide migration of raw `fmt.Errorf` / `errors.New` in function
bodies onto
the `merr` framework, plus the Sys-vs-Input error classification and the
machinery it drives (retriability, fine-grained metrics, segcore
unification),
plus the convention docs and a linter that keeps it from regressing.

Scope: storage, proxy, coordinators (root/data/query), query node, data
node,
`pkg/util` & `internal/util`, expression parser, message queue,
streaming, and
misc packages. Bare raw-error usages went from ~3000 to a ~340 allowlist
(package-level sentinels / build-tag / test sites).

---

## How to review this PR

It is large but the vast majority is mechanical. Changes fall into three
tiers;
spend review budget on Part 2 and Part 3.

### Part 1 — Mechanical standardization (low risk, verify by rule)

Each converted call follows one of a small fixed set of rules. To
review, check
that each site obeys the matching rule rather than reading every line:

| Pattern | Rule |
|---|---|
| `fmt.Errorf("...")` originating a new error | →
`merr.WrapErrXxxMsg("...")` with a code matching the failure's meaning |
| Adding context to an existing typed error | → `merr.Wrap(err, "...")`
/ `merr.Wrapf(...)` — **preserves** the inner code (never `WrapErr*Err`,
which overwrites it) |
| Errors inside the streaming subsystem | → `status.New*` factories
(StreamingError), **not** merr — this is the component-internal dialect
(see `docs/dev/error_handling_guide.md`) |
| Low-level / control-flow signal caught by `errors.Is` | → kept as a
package-level `errors.New` sentinel (lowercase, same-package) |

Conventions are documented in `docs/dev/error_handling_guide.md`
(how-to) and
`docs/dev/error_sentinel_convention.md` (rules + audit). A
`gocritic`/`ruleguard`
rule (`rawmerrerror`, in `rules.go`) enforces "no raw `return
errors.New/fmt.Errorf`"
under `make verifiers`.

### Part 2 — Behavior changes (review these closely)

These are the sites where the wire contract or runtime behavior changes,
not just
the source text. Listed by category; representative locations given,
full set in
the diff.

**A. gRPC wire-code shifts: `UnexpectedError(1)/Code 65535` → typed
code.**
Where a handler previously returned a raw error (collapsed to
`Code=65535` on the
wire), it now returns a typed merr, so the client sees a real code. The
most
common shift is to `IllegalArgument(5)/Code 1100` (ParameterInvalid).
Touch
points include datanode task handlers (CreateTask/Query/Drop), proxy
Upsert,
querynode GetMetrics, datacoord CreateIndex, httpserver query-response
builder,
and typeutil schema validation. One code refinement: an index-param
validation
moved `1100` → `1101` (ParameterMissing). **Client/SDK assertions and
any code
that switched on `Code=65535` for these paths must be re-checked** (the
go_client
e2e assertions were already aligned in this PR).

**B. Prometheus `status` label contract change (externally visible).**
The proxy metric's coarse `fail` / `rejected` values are split into
`fail_input` / `fail_system` and `rejected_user` / `rejected_system` (in
`requestutil.ParseMetricLabel`; auth/privilege rejections count as
`rejected_user`), so dashboards can attribute a failure to caller vs
operator.
**Dashboards/alerts querying `status="fail"` must migrate to
`status=~"fail_.*"`, and `status="rejected"` to
`status=~"rejected_.*"`.** The
in-repo Grafana dashboard is already migrated; external dashboards built
on the
old values silently go empty after upgrade. This is the one change that
requires an ops-side migration.

**C. Retriability semantics.**
- C1: `merr.Status(err)` now forces `Retriable=false` when the error is
an
`InputError` — a malformed request can never succeed on blind retry, so
clients
never get the self-contradictory "your input is wrong but you may
retry".
- C2: `retry.Do` short-circuits an `InputError` (non-retriable) — **but
only when
  the caller did not pass a `RetryErr` predicate**. The check is an
`if c.isRetryErr != nil { ... } else if InputError { ... }` *mutually
exclusive*
branch (`pkg/util/retry/retry.go`): an explicit `RetryErr` takes
precedence and
bypasses the InputError abort. `retry.Handle` deliberately does **not**
apply
the InputError abort (its callers signal abort via `shouldRetry=false`).
Four
flusher startup callsites that must retry through transient "not ready"
errors
  were given explicit `RetryErr` escape hatches.

**D. segcore (C++→Go) error classification.**
A single shared Go-side table (`pkg/util/merr/segcore.go`) maps each
segcore code
to a merr sentinel + InputError/signal category, replacing scattered
hand-written
`if errorCode == ...` switches in the cgo wrappers. **Wire `Code` values
change
for every segcore pass-through error, not just the remapped ones.**
Named
sentinels remap (C++ `2003` → merr `2001`, `2033` → `2002`,
Folly/Knowhere codes
likewise); **all remaining pass-through codes (`2004`–`2043`, previously
surfaced to clients as raw C++ enum values) now serialize as `2000`**
(`ErrSegcore`), with the original C++ code preserved in the `Reason`
text
(`segcoreCode=...`); unknown/future codes collapse to `2000` as well
(pinned by
the `wire_code_projection` test). Transient segcore classes (object
storage /
file IO / OOM / mmap / FieldNotLoaded — 11 codes) now report
`Retriable=true`.
**Any client switching on raw segcore codes in the `2004`–`2043` range
must be
re-checked**; the in-Reason code remains available for diagnostics.
Signal
codes (PretendFinished / FollyCancel) are recognized centrally.
`errors.Is`-based
control flow on these (e.g. scheduler skip/retry) is preserved.

**E. InputError classification (25 sentinels + dynamic marks).**
25 sentinels in `errors.go` carry `WithErrorType(InputError)` (the
Collection /
ResourceGroup / Database families, `ErrIndexDuplicate`,
`ErrParameterInvalid`,
`ErrPrivilegeNotAuthenticated`, `ErrImportFailed`, `ErrQueryPlan`, ...),
plus dynamic
marks for the 8 segcore input codes (ExprInvalid, DimNotMatch,
MetricTypeInvalid, FieldIDInvalid, ...) and
`WrapErrAsInputError`. The widest blast radius is `ErrParameterInvalid`
(1100):
~2335 `WrapErrParameterInvalid*` callsites now classify as input /
non-retriable. Because of C1/C2 this changes retriability for
any path that returns these. **The audit to confirm no transient path
was
mis-marked is the single most important review item** (see Part 3). One
reverse
correction: storage field-stats parsing moved from `ErrParameterInvalid`
(input)
to `ErrDataIntegrity` — a corrupted stored stat is data corruption, not
user
input.

### Part 3 — Known risks & traps (called out proactively)

1. **`merr.Wrap` vs `WrapErr*Err` (code-masking).** `WrapErr*Err` builds
a
`wrappedMilvusError{sentinel: ErrServiceInternal}` whose `code()`
returns the
*outer* sentinel — it overwrites the inner typed code and hides the
`errors.Is`
chain. This is intentional (use it to *deliberately* downgrade), but it
was a
recurring conversion defect; the rule "add context with `merr.Wrap`,
downgrade
with `WrapErr*Err`" is enforced by convention and reviewed across the
diff.
2. **InputError × `retry.Do` blast radius.** Marking a sentinel
`InputError` makes
any `retry.Do(...)` without a `RetryErr` predicate stop retrying it.
Reviewers
should sanity-check that no transient use of the 19 newly-marked
sentinels
(especially `ErrParameterInvalid`) sits inside a retry loop that needed
to keep
   spinning. The known flusher cases were handled (see C2).
3. **The ~340 raw-error allowlist.** What remains as bare `errors.New`
is, by
design: package-level sentinels (caught by `errors.Is`), `//go:build
test`
sites, and out-of-band trees (`cmd/`, `tests/`, codegen, walimpls). The
linter
only bans the *direct-return* form; assignment-then-return escapes and
the full
no-exceptions ban are deferred to an AST-based linter (Tier 2,
documented).
4. **segcore C++ second step deferred.** This PR unifies classification
on the Go
side; splitting the dual-semantic C++ codes at the source is a
follow-up.

---

## Validation

- `make verifiers`: Go side clean (gofmt + static-check across modules,
including
  the new `rawmerrerror` rule with a 0-hit baseline repo-wide).
- `make test-go`: passing; the one real regression introduced (a
datanode
`invalid_task_type` assertion shifting `1` → `5` from a ParameterInvalid
  conversion) was fixed in-tree.
- go_client e2e CreateIndex assertions aligned to the new merr messages.

---------

Signed-off-by: zhenshan.cao <zhenshan.cao@zilliz.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

configs/milvus.yaml

@@ -359,15 +359,15 @@ proxy:
     remoteMaxTime: 0 # The time interval allowed for uploading access log files. If the upload time of a log file exceeds this interval, the file will be deleted. Setting the value to 0 disables this feature.
     formatters:
       base:
-        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost]"
+        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [error_type: $error_type]"
       query:
-        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [expr: $method_expr] [params: $query_params]"
+        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [expr: $method_expr] [params: $query_params] [error_type: $error_type]"
         methods: "Query, Delete, /query, /v2/vectordb/entities/query, /v2/vectordb/entities/get, /v2/vectordb/entities/delete"
       search:
-        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [expr: $method_expr] [nq: $nq] [params: $search_params]"
+        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [expr: $method_expr] [nq: $nq] [params: $search_params] [error_type: $error_type]"
         methods: "HybridSearch, Search, /search, /v2/vectordb/entities/search, /v2/vectordb/entities/advanced_search, /v2/vectordb/entities/hybrid_search"
       upsert:
-        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [partial_update: $partial_update]"
+        format: "[$time_now] [ACCESS] <$user_name: $user_addr> $method_name [status: $method_status] [code: $error_code] [sdk: $sdk_version] [msg: $error_msg] [traceID: $trace_id] [timeCost: $time_cost] [database: $database_name] [collection: $collection_name] [partitions: $partition_name] [partial_update: $partial_update] [error_type: $error_type]"
         methods: Upsert
     cacheSize: 0 # Size of log of write cache, in byte. (Close write cache if size was 0)
     cacheFlushInterval: 3 # time interval of auto flush write cache, in seconds. (Close auto flush if interval was 0)

deployments/monitor/grafana/milvus-dashboard.json

@@ -4794,7 +4794,7 @@
                 "uid": "${datasource}"
               },
               "exemplar": true,
-              "expr": "sum(increase(milvus_proxy_req_count{app_kubernetes_io_instance=~\"$instance\", app_kubernetes_io_name=\"$app_name\", namespace=\"$namespace\", status=\"fail\"}[2m])/120) by(function_name, pod, node_id)",
+              "expr": "sum(increase(milvus_proxy_req_count{app_kubernetes_io_instance=~\"$instance\", app_kubernetes_io_name=\"$app_name\", namespace=\"$namespace\", status=~\"fail_.*\"}[2m])/120) by(function_name, pod, node_id)",
               "interval": "",
               "legendFormat": "{{function_name}}-{{pod}}-{{node_id}}",
               "queryType": "randomWalk",

docs/dev/error_handling_guide.md

@@ -0,0 +1,314 @@
+# Error Handling Guide
+
+How to produce and return errors in Milvus server code — the day-to-day how-to.
+For the underlying rules, the sentinel naming convention, and the enforcement
+roadmap, see [error_sentinel_convention.md](./error_sentinel_convention.md). For
+the canonical numeric code list, see the sentinel definitions in
+[`pkg/util/merr/errors.go`](../../pkg/util/merr/errors.go). (The
+[appendix_d_error_code.md](../developer_guides/appendix_d_error_code.md)
+appendix predates merr and lists the **deprecated** `commonpb.ErrorCode` enum,
+not the merr codes.)
+
+## TL;DR
+
+1. **Never** `return errors.New(...)` or `return fmt.Errorf(...)`. A linter
+   rejects it (see [The one rule](#the-one-rule-never-return-a-bare-error)).
+2. An error that leaves your component must be a **typed** error carrying a code.
+   The lingua franca across Milvus components is **merr**.
+3. Pick one of three forms: originate a typed merr, add context with `merr.Wrap`,
+   or — only if a caller branches on it by identity — a package-level sentinel.
+
+## The mental model: three kinds of error, one boundary rule
+
+Milvus has three legitimate kinds of error, distinguished by **how far they
+travel**, not by syntax:
+
+| Kind | Scope | Carrier | Example |
+|---|---|---|---|
+| **① merr** | across Milvus components & to clients | numeric code on the main gRPC wire | `merr.ErrCollectionNotFound` |
+| **② component-internal dialect** | between sub-modules of one big component | that component's own typed error + its own wire | `streamingutil/status.StreamingError` |
+| **③ internal sentinel** | within a single Go process | `errors.New` pointer identity, caught by `errors.Is` | `errSessionVersionCheckFailure` |
+
+The single rule that ties them together:
+
+> **The error you must return follows the interface's promise** — it is decided
+> by *who is on the other side of the boundary you are crossing*, not by where
+> the error was born.
+
+- Crossing into **another Milvus component** (proxy → rootcoord), or returning to
+  a **client**: the promise is **merr**. Translate to a typed merr at that
+  boundary.
+- Staying **inside one component** (e.g. the streaming sub-modules talking to
+  each other): the component may speak its own typed dialect (`StreamingError`).
+  It is **not** required to be merr — but it must still be *typed*, and it gets
+  translated to merr at the component's outer edge.
+- On **no** wire, ever: a **bare** `errors.New` / `fmt.Errorf`. Internal
+  sentinels (kind ③) are bare, but they never reach a wire — they are caught by
+  `errors.Is` and translated first.
+
+### Why component-internal dialects are allowed (the StreamingError case)
+
+Streaming is one big component whose sub-modules (streamingnode, streamingcoord,
+the streaming client) talk to each other constantly. They use
+`streamingutil/status.StreamingError`, which has its **own** error codes and its
+**own** gRPC encoding. That is deliberate: it is a *bounded context* with its own
+vocabulary. There is intentionally **no** global "StreamingError → merr"
+auto-converter — that would erase the dialect. Instead the conversion happens
+**once, at the consumer's boundary**, and the consumer decides how, based on
+what its own interface promises:
+
+```go
+// rootcoord consumes a streaming service inside CreateCollection.
+err := s.streamingService.DoSomething(ctx, ...) // may return *StreamingError
+if err != nil {
+    // (a) You care about the code the client sees → translate explicitly:
+    if se := status.AsStreamingError(err); se != nil && se.IsRateLimitRejected() {
+        return merr.WrapErrServiceRateLimit("streaming backpressure")
+    }
+    // (b) You don't care about a precise code → let it fall back at the
+    //     boundary (see "The safety net"); the client gets a generic
+    //     internal-class error. Prefer being explicit, but this is allowed.
+    return err
+}
+```
+
+## The one rule: never return a bare error
+
+```go
+return errors.New("segment not loaded")           // ❌ linter rejects
+return fmt.Errorf("segment %d not loaded", id)     // ❌ linter rejects
+```
+
+Why it is banned, even though the boundary would "fix it up" anyway: a bare error
+that escapes to a gRPC boundary becomes `Code=65535 (Unexpected)` — visually
+indistinguishable from "the server hit an unhandled bug". A wall of
+`errors.New("reason 1")`, `errors.New("reason 2")` is a sign nobody planned the
+error taxonomy: the caller cannot program against it, and it all collapses into
+one opaque code on the wire. A typed error costs one extra word and makes the
+failure *addressable*. The linter exists to build the habit: **the thing I
+return is always typed.**
+
+## Decision tree: what should I return?
+
+```
+Am I crossing into another component / returning to a client?
+│
+├─ No (staying inside my own component)
+│   ├─ My component has its own typed dialect (e.g. streaming)?
+│   │     → use that dialect's factory (status.New*), not merr, not errors.New
+│   └─ Otherwise → a typed merr (rules below); or, only if a caller in THIS
+│         process branches on the outcome by identity, a package-level sentinel (§3.3)
+│
+└─ Yes → I must return a typed merr:
+    ├─ Brand-new failure, no underlying error worth carrying?
+    │      → merr.WrapErrXxxMsg("detail %s", v)             (§3.1 originate)
+    │        and pick Input vs System deliberately (next section)
+    ├─ I hold an underlying error and want to KEEP its code, just add context?
+    │      → merr.Wrap(err, "while doing X")                (§3.2 add context)
+    └─ I hold an underlying error and want to DOWNGRADE it to a generic class?
+           → merr.WrapErrServiceInternalErr(err, "...")     (§3.2 — deliberate override)
+```
+
+## Input vs System: who is to blame?
+
+Every merr is classified as **InputError** (the request author's fault) or
+**SystemError** (Milvus's fault, the default). Choosing a factory chooses the
+classification, so when you originate an error, ask one question first:
+
+> **Would a correctly implemented Milvus ever hit this branch, given this
+> request?** If the request content itself triggers it → InputError. If
+> reaching this branch means a Milvus bug or internal failure → SystemError.
+
+Quick rules for the cases that get misclassified in practice:
+
+- A **plan / task type / request produced by a coordinator** is not user input.
+  An unrecognized task type or malformed compaction plan is an internal
+  protocol violation (think mixed-version rolling upgrade) →
+  `WrapErrServiceInternalMsg`, even though the check looks like validation.
+- **Data produced by segcore or another internal component** is not user
+  input. A violated data-shape contract (ValidData length, truncated vectors)
+  is a Milvus bug → `WrapErrServiceInternalMsg`.
+- A **TOCTOU race** (state was valid at check time, changed by execution time)
+  is not user input → keep it a system error.
+
+### How classification is attached
+
+Two mechanisms, used in different situations:
+
+1. **Baked-in sentinels.** ~25 sentinels are declared
+   `WithErrorType(InputError)` in `errors.go` (`ErrParameterInvalid/Missing/
+   TooLarge`, `ErrPrivilegeNotPermitted`, `ErrDatabaseInvalidName`, ...).
+   Using their factory *is* the classification — which is exactly why reaching
+   for `WrapErrParameterInvalidMsg` to express an internal assertion is wrong.
+2. **Boundary marking** for dual-use sentinels. `ErrCollectionNotFound` stays
+   SystemError (internal refresh/retry paths depend on that), and the proxy
+   boundary stamps it InputError only where the name came from the user:
+
+   ```go
+   // proxy meta cache, the central chokepoint for user-supplied names:
+   return collection, merr.WrapErrAsInputErrorWhen(err,
+       merr.ErrCollectionNotFound, merr.ErrDatabaseNotFound)
+   ```
+
+   `WrapErrAsInputError(err)` marks unconditionally;
+   `WrapErrAsInputErrorWhen(err, targets...)` marks only if the error's code
+   matches a target. Both preserve `errors.Is` and the code — they relabel the
+   classification, nothing else.
+
+### What the classification drives
+
+| Surface | InputError behavior |
+|---|---|
+| `commonpb.Status` | `ExtraInfo["is_input_error"]="true"`, `Retriable` forced `false` |
+| Prometheus | request counted as `fail_input` / `rejected_user` (vs `fail_system` / `rejected_system`) |
+| Access log / failure log | `error_type` field set accordingly |
+| proxy lb_policy | **no cross-replica failover** — retrying a bad request elsewhere can't help |
+| `retry.Do` | aborts immediately instead of retrying |
+
+The last two rows are why misclassification is not cosmetic: marking an
+internal failure as InputError disables the retry/failover machinery that
+would have healed it, and a dashboard blames users for Milvus bugs.
+
+### Pitfalls (each of these happened)
+
+- **Don't mark a shared sentinel InputError globally** to fix one callsite —
+  every internal `retry.Do` loop waiting on that error stops retrying. Use
+  boundary marking instead.
+- **Don't classify in a helper** what only the boundary can know. The same
+  not-found is the user's fault when the name came from a request, and a
+  system fault when it came from internal state — stamp at the chokepoint
+  where the origin is known.
+- **"Looks like validation" is not the test.** Coordinator-to-node protocol
+  checks, segcore output checks, and cgo boundary checks all look like
+  validation; none of them are user input.
+
+## The three correct ways
+
+### 3.1 Originate a typed error — `WrapErrXxxMsg` / `WrapErrXxx`
+
+When the failure starts here and there is no inner error worth preserving:
+
+```go
+return merr.WrapErrParameterInvalidMsg("nq (%d) exceeds the limit (%d)", nq, max)
+return merr.WrapErrCollectionNotFound(collectionName)
+```
+
+Pick the sentinel whose **code** matches the failure's meaning (see the sentinel
+definitions in [`pkg/util/merr/errors.go`](../../pkg/util/merr/errors.go)). This is the common
+case: most of the time you only need to attach a message to a well-chosen code,
+and the framework does the rest.
+
+### 3.2 Add context but keep the code — `merr.Wrap`, never `WrapErr*Err`
+
+When you already hold a typed error and only want to add a breadcrumb:
+
+```go
+if err := s.loadSegment(ctx, id); err != nil {
+    return merr.Wrap(err, "while loading sealed segment") // ✅ keeps the inner code
+}
+```
+
+`merr.Wrap` / `merr.Wrapf` is a thin wrapper (like `errors.Wrap`): it prepends
+context and **preserves** the inner error's code and its `errors.Is` chain.
+
+⚠️ Do **not** reach for `merr.WrapErrServiceInternalErr(err, ...)` (or any
+`merr.WrapErrXxxErr`) just to add context. Those build a
+`wrappedMilvusError{sentinel: ErrServiceInternal}` whose `code()` returns the
+**outer** sentinel's code — they **overwrite** the inner typed code with
+ServiceInternal (5) and force it non-retriable (`As()` resolves to the outer
+sentinel). The `errors.Is` chain itself survives via `Unwrap()`, but the typed
+code and retriability are masked. `WrapErrXxxErr` is *only* for
+when you **deliberately** want to relabel the inner error to a new code (e.g.
+collapse a noisy internal failure into one ServiceInternal for the client).
+
+This split is intentional, and the framework deliberately does **not** try to be
+clever: *keep the code* versus *downgrade the code* is a decision you state, not
+one the framework guesses. `merr.Wrap` **always** keeps the inner code;
+`WrapErrXxxErr` **always** relabels to the outer one. A helper that "smartly"
+preserved the inner code whenever it recognized a typed merr would blur the
+intent — a reader could no longer tell from the call site whether the author
+meant to preserve or to downgrade. The choice of helper *is* the statement of
+intent.
+
+### 3.2.1 The base-package case: pass through, wrap, or relabel?
+
+Low-level packages (`pkg/...`, `internal/util/...`) sit under many callers and
+usually receive an error from something even lower — etcd, S3, a third-party
+library, another util. For every such error you must consciously pick one of
+three. Getting this wrong in a base package is expensive: it is multiplied across
+every caller.
+
+| Choice | Use it when | How |
+|---|---|---|
+| **Pass through** the original `err` | the inner err is already a typed error meaningful to your caller, or your package has no business classifying it — let the caller decide | `return err` |
+| **Wrap, keep the code** | you want to add a breadcrumb (which key/path/op failed) without changing what the error *means* | `merr.Wrapf(err, "etcd txn on key %s", k)` |
+| **Relabel / downgrade** | the inner err is a leaky implementation / third-party detail your caller should not see; you translate it into the typed merr your package's interface promises | `merr.WrapErrXxxErr(err, "...")` |
+
+Deciding questions, in order:
+
+1. **Is the inner err already typed and meaningful to my caller?** → pass through.
+2. **Does any caller `errors.Is` the inner err's identity?** → pass through or
+   `merr.Wrap` (both preserve the chain); **never relabel** — it hides the chain.
+3. **Is the inner err a third-party / implementation detail I promise to hide?**
+   → relabel to the typed merr my interface promises.
+
+A base package that relabels too eagerly destroys codes the upper layers needed;
+one that passes a raw third-party error straight through leaks an untyped error
+toward the boundary. Neither is acceptable — the choice must be deliberate, and it
+follows your package's interface promise, not convenience.
+
+### 3.3 Need identity branching or reuse — a package-level sentinel
+
+Define a sentinel when a caller **in the same process** branches on the outcome
+by identity, via `errors.Is`:
+
+```go
+// internal/util/sessionutil/session_util.go — caught by isNotSessionVersionCheckFailure
+// and used as a retry.Do predicate; the identity must survive, so it stays a
+// bare sentinel rather than a merr error.
+var errSessionVersionCheckFailure = errors.New("session version check failure")
+```
+
+Rules for sentinels (full version in
+[error_sentinel_convention.md](./error_sentinel_convention.md)):
+
+- **Package-level, never function-local.** A local `x := errors.New(...)` is a
+  refactor hazard — tomorrow it gets hoisted and silently crosses a boundary.
+  Lift it to a `var` at package scope.
+- **Lowercase / unexported** when it lives in `internal/...`. If a
+  *cross-package* caller needs the signal, do **not** export the sentinel —
+  redesign the API to carry the signal in a return value (e.g.
+  `(ignored bool, err error)`).
+- It must be `errors.Is`-caught and translated to a typed merr (or
+  `merr.Success()`) **before** crossing any gRPC boundary. A sentinel that
+  reaches the wire is just an opaque `Code=65535`.
+
+When unsure between 3.1 and 3.3: if nobody does `errors.Is` on it, you don't need
+a sentinel — just originate a typed merr with a message (3.1).
+
+## The safety net: boundary fallback (and why not panic)
+
+If a non-typed error does reach a gRPC handler, `merr.Status(err)` falls back to
+`Code=65535 (Unexpected)`. This is a **backstop, not a feature**: it keeps the
+server from leaking internals or crashing, but the client gets an opaque code.
+Treat any `Code=65535` in logs as "someone forgot to type their error".
+
+Why the boundary **falls back instead of panicking**: third-party libraries and
+deep call chains produce errors on paths that tests cannot fully cover. Panicking
+on "not a typed merr" would turn a stray untyped error into an outage. The
+contract is therefore: **fall back to a generic code, never panic.** The goal of
+the linter and this guide is to make that fallback path *empty in practice* — so
+that every error a client sees was deliberately typed at its source.
+
+## Enforcement
+
+- A `gocritic`/`ruleguard` rule (`rawmerrerror` in `rules.go`) rejects
+  `return errors.New / fmt.Errorf / errors.Errorf` from function bodies. It runs
+  under `make verifiers` (via `static-check`), so no extra command is needed.
+  Exempt paths (run outside the request path): `*_test.go`, `cmd/`, `tests/`,
+  codegen, the walimpls harness, and `/mocks/` (generated mock helpers are
+  test infrastructure even though some lack a "Code generated" header).
+- It catches the **direct-return** form — the one that lets a raw error escape to
+  a boundary. Assignment-then-return escapes (`e := errors.New(); return e`) and
+  the full no-exceptions ban require the AST-based linter described as "Tier 2"
+  in [error_sentinel_convention.md](./error_sentinel_convention.md).

docs/dev/error_sentinel_convention.md

@@ -0,0 +1,324 @@
+# Error Sentinel Convention
+
+Milvus error handling has two distinct layers. This document describes the rules
+that separate them, the rationale, and an audit of where the codebase currently
+violates them.
+
+## The two layers
+
+### 1. Typed merr (wire-protocol errors)
+
+Defined in `pkg/util/merr/errors.go` (`ErrCollectionNotFound`,
+`ErrParameterInvalid`, etc.). These carry a numeric error code that is
+serialized into `commonpb.Status{ErrorCode, Reason}` and shipped to the
+client over gRPC. They are the only thing a client (or another Milvus
+component on the receiving end of an RPC) sees.
+
+Creation: `merr.WrapErrXxxMsg(...)` / `merr.WrapErrXxxErr(cause, ...)`
+(origination only — never to add context to a typed merr that already exists,
+see `merr.Wrap`).
+
+### 2. Internal sentinels (single-process control flow)
+
+Created with `errors.New(...)` at package scope (e.g. `errIgnoredAlterAlias`,
+`errReleaseCollectionNotLoaded`, `errNodeNotEnough`). These are signaling
+vocabulary inside a single Go process: a callee tells its caller "this is an
+idempotent no-op" or "the queue is empty" so the caller can branch / retry /
+ignore. They are **not** part of the wire protocol.
+
+The catcher is always `errors.Is(err, sentinelX)` at some boundary in the
+calling stack, and the boundary either:
+
+- translates to `merr.Success()` (idempotency: e.g. drop something that
+  doesn't exist → success), or
+- translates to a typed merr `merr.WrapErrXxxMsg(...)` (e.g. user already
+  exists → `WrapErrParameterInvalid`).
+
+## The hard invariant
+
+> **Any internal sentinel must be `errors.Is`-caught and translated to a
+> typed merr (or `Success`) before crossing any gRPC handler boundary —
+> client-facing or component-to-component.**
+
+Why "any gRPC boundary, not just client-facing": gRPC serializes errors to
+`commonpb.Status{ErrorCode, Reason}`. The Go-level pointer identity that
+`errors.New` sentinels rely on does not survive the wire. The peer's
+`merr.Error(status)` reconstructs a typed merr from the numeric code; the
+sentinel chain is gone forever. So a sentinel that escapes an internal coord
+RPC is just as broken as one that escapes a user-facing RPC — only quieter,
+because no customer sees the resulting `Code=65535 (unexpected)`.
+
+### What breaks the invariant
+
+The `errors.Is` chain survives `return err`, `errors.Wrap(err, "...")` /
+`merr.Wrap(err, "...")` (cockroachdb thin wrap), **and**
+`merr.WrapErrServiceInternalErr(err, "...")` — `milvusError.Unwrap()`
+returns the inner error, so `errors.Is(outer, innerSentinel)` stays true through any
+of them. It is **destroyed** only by:
+
+- Putting the cause in a format argument instead of the chain:
+  `merr.WrapErrXxxMsg("...: %s", err)`, or the `%w` mistake (`WrapErr*Msg`
+  formats with `fmt.Sprintf`, which does **not** honor `%w` and renders
+  `%!w(...)`). The inner error reaches the message text but is unreachable via
+  `Unwrap()`, so `errors.Is(outer, innerSentinel)` returns false.
+- Any custom wrapper that doesn't implement `Unwrap()`.
+
+Do **not** confuse this with the separate code/retriability rule.
+`merr.Wrap(err, ...)` and `merr.WrapErrXxxErr(err, ...)` both keep `errors.Is`
+intact, but they differ in what the result reports at the boundary:
+
+- `merr.Wrap(err, ...)` preserves the inner's `Code()` and `IsRetryable` — the
+  chain still resolves to the inner's `*milvusError`.
+- `merr.WrapErrXxxErr(err, ...)` reports the **outer** sentinel's code and
+  retriability: `As()` resolves to `ErrServiceInternal` (Code 5,
+  non-retriable), **masking** the inner's classification.
+
+So there are two distinct rules, often conflated:
+
+1. To keep `errors.Is` working: never stuff the cause into a format string;
+   pass it as the error argument.
+2. To add context **without changing the classification** (preserve the inner
+   code + retriability): use `merr.Wrap`, not `merr.WrapErr*Err`. Reserve
+   `merr.WrapErrXxxErr` for when you intend to *assert* a new classification
+   (e.g. this genuinely is a service-internal error). (See also
+   [feedback rule](#related-rules) on `merr.Wrap` vs `merr.WrapErr*Err`.)
+
+## Naming convention
+
+The convention has two layers, matched to the two error categories:
+
+### Wire-protocol layer (typed merr) — uppercase `Err*` in `pkg/util/merr` only
+
+All errors that may cross any gRPC boundary (client-facing or
+component-to-component) must be `*merr.milvusError` defined in
+`pkg/util/merr/errors.go`. They have:
+
+- A numeric code passed to `newMilvusError(...)`; uniqueness is enforced by
+  the init-time code registry (defining a second sentinel on an occupied code
+  panics at package init, since `milvusError.Is` matches by code alone)
+- A `var ErrXxx = newMilvusError(...)` declaration in `pkg/util/merr/errors.go`
+- An exported `WrapErrXxxMsg` / `WrapErrXxxErr` helper
+
+If an error needs to be visible to the wire, it lives here. No exceptions.
+
+Every sentinel also carries an Input-vs-System classification (who is to
+blame) that drives `Status.Retriable`, the `fail_input`/`fail_system` metric
+labels, lb_policy failover and `retry.Do`; see "Input vs System: who is to
+blame?" in [error_handling_guide.md](error_handling_guide.md).
+
+### Internal-sentinel layer — lowercase `errXxx`, same-package only
+
+Internal sentinels live in `internal/...` packages, are created with
+`errors.New(...)`, and are **lowercase / unexported**. The rule:
+
+> **A `var err* = errors.New(...)` declared in `internal/...` may only be
+> referenced inside the same Go package. Cross-package consumers must not
+> see it.**
+
+This makes Go visibility do the enforcement: if you need a signal across
+package boundaries, you either (a) lift it into the wire layer as a typed
+merr, or (b) redesign the API so the signal flows via a return value
+(e.g. `(ignored bool, err error)`), not via the error type.
+
+Example (current code, after the 04-coord cleanup):
+
+```go
+// errFull / errNoSuchElement are INTERNAL sentinels: caught by errors.Is
+// inside the compaction inspector / scheduler loop and never serialized
+// across any gRPC boundary.
+var (
+    errFull          = errors.New("compaction queue is full")
+    errNoSuchElement = errors.New("compaction queue has no element")
+)
+```
+
+### Cross-package idempotency: use a return-value flag, not an exported sentinel
+
+The previous code exported `meta.ErrResourceGroupOperationIgnored` so that
+the parent package `querycoordv2` could catch it via `errors.Is` and
+translate to `merr.Success()`. This was an exported `Err*` in
+`internal/...` — visually indistinguishable from a `merr.ErrXxx` typed
+wire error, easy to misuse.
+
+The current code instead encodes the signal in the return value:
+
+```go
+// meta/resource_manager.go
+func (rm *ResourceManager) CheckIfResourceGroupAddable(...) (ignored bool, err error) {
+    if proto.Equal(rm.groups[rgName].GetConfig(), cfg) {
+        return true, nil   // idempotent no-op
+    }
+    ...
+}
+
+// querycoordv2/ddl_callbacks_alter_resource_group.go (broadcaster)
+func (s *Server) broadcastCreateResourceGroup(...) (ignored bool, err error) {
+    if ignored, err := s.meta.CheckIfResourceGroupAddable(...); err != nil || ignored {
+        return ignored, err
+    }
+    ...
+}
+
+// querycoordv2/services.go (RPC handler)
+ignored, err := s.broadcastCreateResourceGroup(ctx, req)
+if err != nil { return merr.Status(err), nil }
+if ignored { return merr.Success(), nil }
+```
+
+No sentinel crosses the package boundary; the signal travels via a
+structured return value. This is the preferred pattern for any new
+cross-package idempotency case.
+
+## Current state (audit done on err-std-04-coord branch, 2026-05-19)
+
+Across `internal/{datacoord,rootcoord,querycoordv2}` there are 28
+`errors.New(...)` sentinels. Their fates:
+
+| Kind | Count | Examples | Status |
+|---|---|---|---|
+| Idempotency: caught → `merr.Success()` | 13 catch sites, ~12 distinct sentinels | `errIgnoredAlterAlias`, `errIgnoredCreateCollection`, `errReleaseCollectionNotLoaded`, `errUserNotFound`, ... | ✅ compliant |
+| Caught → translated to typed merr | 3 catch sites (`errUserAlreadyExists`, `errRoleAlreadyExists`, `errRoleNotExists`) | client gets `WrapErrParameterInvalidMsg(...)` or `WrapErrServiceInternalMsg(...)` | ✅ compliant (1100 / 5) |
+| Background-only (never enter an RPC handler) | 5 (`errFull`, `errNoSuchElement`, `errNodeNotEnough`, `errDisposed`, `errTypeNotFound`) | compaction queue / resource observer / session lifecycle / checker registry | ✅ compliant |
+| Cross-package idempotency via `(ignored bool, err error)` signature | 1 (resource group create/drop) | meta layer returns `ignored=true`; querycoordv2 RPC handler translates to `merr.Success()` — no sentinel crosses package | ✅ compliant (refactored from exported `ErrResourceGroupOperationIgnored` in this branch) |
+| Dead code (function with 0 callers) | 3 (`errNilResponse`, `errNilStatusResponse`, `errUnknownResponseType` — all only used by `VerifyResponse` in `datacoord/util.go`) | safe to delete | 🪦 cleanup candidate |
+| ~~**Violation**: escapes RPC handler with no catch~~ (resolved) | 3 (`errEmptyUsername`, `errEmptyRoleName`, `errEmptyPrivilegeGroupName`) | origin sites in `meta_table.go` now emit `WrapErrParameterInvalidMsg` directly; the bare sentinels are gone | ✅ resolved |
+| ~~**Semantic miscategorization**: caught and wrapped with the wrong typed merr code~~ (resolved) | 1 (`errTypeNotFound` in `ops_services.go:87,101`) | invalid `CheckerID` from the client now wrapped as `WrapErrParameterInvalidMsg` (code 1100), was `WrapErrServiceInternal` (code 5) | ✅ resolved |
+
+### Cleanup status
+
+1. ✅ Done — `errEmptyUsername` / `errEmptyRoleName` / `errEmptyPrivilegeGroupName`
+   at the origin sites in `internal/rootcoord/meta_table.go` now emit
+   `merr.WrapErrParameterInvalidMsg("username is empty")` etc. directly; the bare
+   sentinels no longer exist.
+2. ✅ Done — `errTypeNotFound` at the catcher in
+   `internal/querycoordv2/ops_services.go:87,101` is now wrapped as
+   `merr.WrapErrParameterInvalidMsg("invalid checker type %d: %v", req.CheckerID, err)`.
+3. Not done — delete `VerifyResponse` and its three dead-code sentinels
+   (`errNilResponse`, `errNilStatusResponse`, `errUnknownResponseType`).
+
+## Future linter ideas
+
+Three candidates, in order of how cheap they are to implement and how
+hard the enforcement is. **Tier 1.5's return form is now implemented (see
+below); Tier 1 and Tier 2 remain a design queue.**
+
+### Tier 1 — exported-sentinel ban (1 hour to write)
+
+The simplest rule: **`internal/...` packages may not declare exported
+`var Err\w+ = errors.New(...)`.** Scan all `internal/...` *.go files,
+fail CI if any match. Two paths to fix a violation:
+
+1. Lowercase it (`var errXxx = errors.New(...)`) — only callable inside the
+   same package. If the lint fails because a cross-package caller needs the
+   signal, see fix 2.
+2. Refactor the API so the signal travels via a return value
+   (e.g. add `ignored bool` to the return tuple) and delete the sentinel.
+
+This makes Go visibility itself the enforcement mechanism: anything that
+needs to look like `merr.ErrXxx` to a reviewer can only exist in
+`pkg/util/merr`. Internal sentinels stay quietly lowercase in their owning
+package.
+
+Lowercase sentinels (`var errXxx = errors.New(...)`) inside `internal/...`
+are still encouraged to carry an `INTERNAL: ...` doc comment for reviewer
+context, but it's not enforced — the visibility rule already prevents the
+worst-case (`Err*` collision with `merr.ErrXxx`).
+
+### Tier 1.5 — bare-usage ban (1 hour grep, half-day AST for 100% precision)
+
+> **Status — return form implemented.** The **return** form of this ban is now
+> enforced by a `gocritic`/`ruleguard` rule (`rawmerrerror` in `rules.go`), run
+> under `make verifiers`: it rejects `return errors.New / fmt.Errorf /
+> errors.Errorf` from function bodies (package-level sentinels, `cmd/`, `tests/`,
+> codegen and the walimpls harness exempt). The day-to-day guide is
+> [error_handling_guide.md](./error_handling_guide.md). The **no-exceptions**
+> form below (local `:=`, `panic(...)`, function argument) is *not* covered:
+> ruleguard's DSL cannot match "a call anywhere in a function body but not in a
+> `ValueSpec`", so the full ban still needs the AST-based Tier 2 linter.
+
+**Hardest enforcement, no exceptions.** `internal/...` packages may not
+use `errors.New(...)` or `errors.Errorf(...)` **inline inside a function
+body**. The only legal site for these calls is a package-level
+`var <Name> = errors.New(...)` (sentinel declaration).
+
+Allow/deny matrix:
+
+| Form | Location | Verdict |
+|---|---|---|
+| `var errInvalid = errors.New("invalid")` | package-level (file top / `var` block) | ✅ allowed |
+| `var ( errA = ...; errB = ... )` | package-level `var` block | ✅ allowed |
+| `return errors.New(...)` | function body | ❌ banned |
+| `x := errors.New(...)` | function body local | ❌ banned |
+| `panic(errors.New(...))` | function body | ❌ banned |
+| `foo(errors.New(...))` | function body argument | ❌ banned |
+
+**Why no exceptions** (even for "local break signal" / "log-only" /
+"panic-bound" cases that look harmless today):
+
+- Today's local var can be hoisted to package-level by tomorrow's refactor
+  and silently start crossing boundaries.
+- A linter with exceptions needs AST-level wire-reachability analysis
+  (expensive); a no-exception linter is one grep.
+- Forces authors to use the right primitive instead of `errors.New` as
+  a universal escape hatch:
+  - **break signal from a callback** → define a `type doneSignal struct{}`
+    that implements `error` and use `errors.As`. Intent is now in the type,
+    not in a string-keyed sentinel.
+  - **"unreachable" assertion** → just `panic(...)`. If caller already
+    does `if err != nil { panic(err) }`, fold it into the callee.
+  - **input validation / config validation** → `merr.WrapErrParameterInvalidMsg(...)`
+    or `status.NewInvalidArgument(...)` depending on layer.
+
+Implementation: grep version covers ~95% true violations in ~1 hour.
+AST version (go/analysis) covers the edge cases (e.g. `init()` body
+assigning to a package var) but needs ~half a day. Start with grep,
+upgrade if false-positive rate exceeds 5%.
+
+```bash
+# grep skeleton
+grep -rnE 'errors\.(New|Errorf)\(' internal/ --include='*.go' \
+  | grep -v _test.go \
+  | grep -vE ':[0-9]+:\s*(var\s+)?[A-Za-z_]+\s*=\s*errors\.(New|Errorf)' \
+  | grep -vE ':[0-9]+:\s*[A-Za-z_]+\s+(\w+\s+)?=\s*errors\.(New|Errorf)'
+# Any remaining line = violation
+```
+
+A `//nolint:err-bare` escape valve with a required justification comment
+handles the genuine outliers (a few `init()` patterns, embedded `errors.Mark`
+usage, etc.).
+
+### Tier 2 — escape-path linter (~1 day, go/analysis based)
+
+For every gRPC handler method (anything matching the
+`internal/{rootcoord,datacoord,querycoordv2}/services.go,root_coord.go,*_handler.go`
+pattern, return type `(*proto.XxxResponse, error)` or `(*commonpb.Status,
+error)`), trace the err-return data-flow. Any error that:
+
+- transitively originates from an `INTERNAL:`-tagged sentinel, **and**
+- reaches a `return Status{Code: merr.Status(err)}` or `return err` without
+  passing through an `errors.Is(err, internalSentinelX) { ... }` branch,
+
+is a violation. Report file:line of the leak.
+
+This catches the actual invariant violation (the 3 RBAC empty sentinels
+would have been flagged), not just naming hygiene. Requires AST analysis;
+worth doing if the cost of one more silent `Code=1` to a client is high.
+
+### Tier 3 (no longer necessary if Tier 2 is in place) — wrap-rule linter
+
+Scan `merr\.WrapErr[A-Za-z]+Err\(err, ` (note: first arg `err`, not a
+fresh string-only origination) and require the cause `err` to not itself be
+a typed merr. This is what
+[`feedback_merr_wrap_rule`](../../) already enforces by convention; Tier 2
+would catch the symptom (sentinel escape) as a side effect.
+
+## Related rules
+
+- `feedback_merr_wrap_rule` (this repo's collaboration memory): "Add context
+  to an existing err with `merr.Wrap` / `merr.Wrapf`, never with
+  `merr.WrapErr*Err` — the latter masks the inner typed code and
+  retriability (the `errors.Is` chain itself is preserved via `Unwrap()`)."
+- `project_errstd_autogen_defects`: three systematic defects in the
+  auto-generated `errors.Wrap → merr` conversion in this branch series;
+  defects #2 and #3 are direct consequences of violating the rules in
+  this document.

docs/developer_guides/appendix_d_error_code.md

@@ -1,5 +1,14 @@
 ## Appendix D. Error Code
 
+> **⚠️ Deprecated.** This appendix documents the legacy `commonpb.ErrorCode`
+> enum (Milvus 2.0-era), which is no longer the source of truth for error codes.
+> Current error codes are the `merr` sentinel codes (e.g. `ParameterInvalid`
+> = 1100, `ServiceInternal` = 5, `FunctionFailed` = 2400) carried in
+> `commonpb.Status.Code`. The canonical list is the sentinel definitions in
+> [`pkg/util/merr/errors.go`](../../pkg/util/merr/errors.go); see also
+> [error_handling_guide.md](../dev/error_handling_guide.md) and
+> [error_sentinel_convention.md](../dev/error_sentinel_convention.md).
+
 **ErrorCode**
 
 ```protobuf

examples/telemetry_e2e_test/go.mod

@@ -2,9 +2,9 @@ module github.com/milvus-io/milvus/examples/telemetry_e2e_test
 go 1.26.4
 
 require (
-	github.com/milvus-io/milvus-proto/go-api/v2 v2.6.6-0.20260304062516-e7e54d966ef2
+	github.com/milvus-io/milvus-proto/go-api/v2 v2.6.6-0.20260323081523-53649783989c
 	github.com/milvus-io/milvus/client/v2 v2.6.4-0.20251104142533-a2ce70d25256
-	google.golang.org/grpc v1.71.0
+	google.golang.org/grpc v1.79.3
 )
 
 replace (

examples/telemetry_e2e_test/go.sum

@@ -1,2 +1,4 @@
 github.com/milvus-io/milvus-proto/go-api/v2 v2.6.6-0.20260304062516-e7e54d966ef2/go.mod h1:/6UT4zZl6awVeXLeE7UGDWZvXj3IWkRsh3mqsn0DiAs=
+github.com/milvus-io/milvus-proto/go-api/v2 v2.6.6-0.20260323081523-53649783989c/go.mod h1:/6UT4zZl6awVeXLeE7UGDWZvXj3IWkRsh3mqsn0DiAs=
 google.golang.org/grpc v1.71.0/go.mod h1:H0GRtasmQOh9LkFoCPDu3ZrwUtD1YGE+b2vYBYd/8Ec=
+google.golang.org/grpc v1.79.3/go.mod h1:KmT0Kjez+0dde/v2j9vzwoAScgEPx/Bw1CYChhHLrHQ=

internal/agg/aggregate.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/planpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 const (
@@ -57,7 +58,7 @@ func isSupportedAggregateName(aggregateName string) bool {
 
 func NewAggregate(aggregateName string, aggFieldID int64, originalName string, fieldType schemapb.DataType) ([]AggregateBase, error) {
 	if !isSupportedAggregateName(aggregateName) {
-		return nil, fmt.Errorf("invalid Aggregation operator %s", aggregateName)
+		return nil, merr.WrapErrParameterInvalidMsg("invalid Aggregation operator %s", aggregateName)
 	}
 
 	if err := ValidateAggFieldType(aggregateName, fieldType); err != nil {
@@ -78,7 +79,7 @@ func NewAggregate(aggregateName string, aggFieldID int64, originalName string, f
 		return []AggregateBase{&MaxAggregate{fieldID: aggFieldID, originalName: originalName}}, nil
 	default:
 		// should never happen due to isSupportedAggregateName check
-		return nil, fmt.Errorf("invalid Aggregation operator %s", aggregateName)
+		return nil, merr.WrapErrParameterInvalidMsg("invalid Aggregation operator %s", aggregateName)
 	}
 }
 
@@ -93,13 +94,13 @@ func FromPB(pb *planpb.Aggregate) (AggregateBase, error) {
 	case planpb.AggregateOp_max:
 		return &MaxAggregate{fieldID: pb.GetFieldId()}, nil
 	default:
-		return nil, fmt.Errorf("invalid Aggregation operator %d", pb.Op)
+		return nil, merr.WrapErrParameterInvalidMsg("invalid Aggregation operator %d", pb.Op)
 	}
 }
 
 func AccumulateFieldValue(target *FieldValue, new *FieldValue) error {
 	if target == nil || new == nil {
-		return fmt.Errorf("target or new field value is nil")
+		return merr.WrapErrServiceInternalMsg("target or new field value is nil")
 	}
 
 	// Skip null values during accumulation
@@ -121,7 +122,7 @@ func AccumulateFieldValue(target *FieldValue, new *FieldValue) error {
 	}
 	// Verify types match
 	if reflect.TypeOf(target.val) != reflect.TypeOf(new.val) {
-		return fmt.Errorf("type mismatch: target is %T, new is %T", target.val, new.val)
+		return merr.WrapErrParameterInvalidMsg("type mismatch: target is %T, new is %T", target.val, new.val)
 	}
 
 	switch target.val.(type) {
@@ -136,7 +137,7 @@ func AccumulateFieldValue(target *FieldValue, new *FieldValue) error {
 	case float64:
 		target.val = target.val.(float64) + new.val.(float64)
 	default:
-		return fmt.Errorf("unsupported type: %T", target.val)
+		return merr.WrapErrParameterInvalidMsg("unsupported type: %T", target.val)
 	}
 	return nil
 }
@@ -255,17 +256,17 @@ func (bucket *Bucket) RowCount() int {
 
 func (bucket *Bucket) Accumulate(row *Row, idx int, keyCount int, aggs []AggregateBase) error {
 	if idx >= len(bucket.rows) || idx < 0 {
-		return fmt.Errorf("wrong idx:%d for bucket", idx)
+		return merr.WrapErrParameterInvalidMsg("wrong idx:%d for bucket", idx)
 	}
 	targetRow := bucket.rows[idx]
 	if targetRow == nil {
-		return fmt.Errorf("nil row at the target idx:%d, cannot accumulate the row", idx)
+		return merr.WrapErrServiceInternalMsg("nil row at the target idx:%d, cannot accumulate the row", idx)
 	}
 	if row.ColumnCount() != targetRow.ColumnCount() {
-		return fmt.Errorf("column count:%d in the row must be equal to the target row:%d", row.ColumnCount(), bucket.rows[idx].ColumnCount())
+		return merr.WrapErrParameterInvalidMsg("column count:%d in the row must be equal to the target row:%d", row.ColumnCount(), bucket.rows[idx].ColumnCount())
 	}
 	if row.ColumnCount() != keyCount+len(aggs) {
-		return fmt.Errorf("column count:%d in the row must be sum of keyCount:%d and the number of aggs:%d", row.ColumnCount(), keyCount, len(aggs))
+		return merr.WrapErrParameterInvalidMsg("column count:%d in the row must be sum of keyCount:%d and the number of aggs:%d", row.ColumnCount(), keyCount, len(aggs))
 	}
 	for col := keyCount; col < row.ColumnCount(); col++ {
 		targetRow.UpdateFieldValue(row, col, aggs[col-keyCount])

internal/agg/aggregate_reducer.go

@@ -2,7 +2,6 @@ package agg
 
 import (
 	"context"
-	"fmt"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	typeutil2 "github.com/milvus-io/milvus/internal/util/typeutil"
@@ -88,15 +87,15 @@ func (reducer *GroupAggReducer) EmptyAggResult() (*AggregationResult, error) {
 		} else {
 			field, err := helper.GetFieldFromID(agg.GetFieldId())
 			if err != nil {
-				return nil, fmt.Errorf("failed to get field schema for aggregate fieldID %d: %w", agg.GetFieldId(), err)
+				return nil, merr.Wrapf(err, "failed to get field schema for aggregate fieldID %d", agg.GetFieldId())
 			}
 			resultType, err := getAggregateResultType(agg.GetOp(), field.GetDataType())
 			if err != nil {
-				return nil, fmt.Errorf("failed to get result type for aggregate fieldID %d: %w", agg.GetFieldId(), err)
+				return nil, merr.Wrapf(err, "failed to get result type for aggregate fieldID %d", agg.GetFieldId())
 			}
 			emptyFieldData, err := genEmptyFieldDataByType(resultType)
 			if err != nil {
-				return nil, fmt.Errorf("failed to generate empty field data for result type %s: %w", resultType.String(), err)
+				return nil, merr.Wrapf(err, "failed to generate empty field data for result type %s", resultType.String())
 			}
 			ret.fieldDatas = append(ret.fieldDatas, emptyFieldData)
 		}
@@ -160,7 +159,7 @@ func genEmptyFieldDataByType(dataType schemapb.DataType) (*schemapb.FieldData, e
 		return genEmptyLongFieldData(dataType, []int64{0}), nil
 	default:
 		// For other types, try to use the original field's GenEmptyFieldData
-		return nil, fmt.Errorf("unsupported data type for aggregate result: %s", dataType.String())
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported data type for aggregate result: %s", dataType.String())
 	}
 }
 
@@ -187,10 +186,10 @@ func getAggregateResultType(op planpb.AggregateOp, inputType schemapb.DataType)
 		case schemapb.DataType_Float, schemapb.DataType_Double:
 			return schemapb.DataType_Double, nil
 		default:
-			return schemapb.DataType_None, fmt.Errorf("unsupported input type %s for sum aggregation", inputType.String())
+			return schemapb.DataType_None, merr.WrapErrParameterInvalidMsg("unsupported input type %s for sum aggregation", inputType.String())
 		}
 	default:
-		return schemapb.DataType_None, fmt.Errorf("unknown aggregate operator: %d", op)
+		return schemapb.DataType_None, merr.WrapErrParameterInvalidMsg("unknown aggregate operator: %d", op)
 	}
 }
 
@@ -201,12 +200,12 @@ func getAggregateResultType(op planpb.AggregateOp, inputType schemapb.DataType)
 // 3. Each fieldData's Type matches the expected type from schema
 func (reducer *GroupAggReducer) validateAggregationResults(results []*AggregationResult) error {
 	if reducer.schema == nil {
-		return fmt.Errorf("schema is nil, cannot validate field types")
+		return merr.WrapErrParameterInvalidMsg("schema is nil, cannot validate field types")
 	}
 
 	helper, err := typeutil.CreateSchemaHelper(reducer.schema)
 	if err != nil {
-		return fmt.Errorf("failed to create schema helper: %w", err)
+		return merr.Wrap(err, "failed to create schema helper")
 	}
 
 	numGroupingKeys := len(reducer.groupByFieldIds)
@@ -220,7 +219,7 @@ func (reducer *GroupAggReducer) validateAggregationResults(results []*Aggregatio
 	for _, fieldID := range reducer.groupByFieldIds {
 		field, err := helper.GetFieldFromID(fieldID)
 		if err != nil {
-			return fmt.Errorf("failed to get field schema for groupBy fieldID %d: %w", fieldID, err)
+			return merr.Wrapf(err, "failed to get field schema for groupBy fieldID %d", fieldID)
 		}
 		expectedTypes = append(expectedTypes, field.GetDataType())
 	}
@@ -234,11 +233,11 @@ func (reducer *GroupAggReducer) validateAggregationResults(results []*Aggregatio
 		} else {
 			field, err := helper.GetFieldFromID(agg.GetFieldId())
 			if err != nil {
-				return fmt.Errorf("failed to get field schema for aggregate fieldID %d: %w", agg.GetFieldId(), err)
+				return merr.Wrapf(err, "failed to get field schema for aggregate fieldID %d", agg.GetFieldId())
 			}
 			expectedType, err = getAggregateResultType(agg.GetOp(), field.GetDataType())
 			if err != nil {
-				return fmt.Errorf("failed to get aggregate result type for aggregate fieldID %d: %w", agg.GetFieldId(), err)
+				return merr.Wrapf(err, "failed to get aggregate result type for aggregate fieldID %d", agg.GetFieldId())
 			}
 		}
 		expectedTypes = append(expectedTypes, expectedType)
@@ -247,27 +246,27 @@ func (reducer *GroupAggReducer) validateAggregationResults(results []*Aggregatio
 	// Validate each result
 	for resultIdx, result := range results {
 		if result == nil {
-			return fmt.Errorf("result at index %d is nil", resultIdx)
+			return merr.WrapErrServiceInternalMsg("result at index %d is nil", resultIdx)
 		}
 
 		fieldDatas := result.GetFieldDatas()
 
 		// Check 1: fieldDatas length
 		if len(fieldDatas) != expectedColumnCount {
-			return fmt.Errorf("result at index %d has fieldDatas length %d, expected %d (numGroupingKeys=%d, numAggs=%d)",
+			return merr.WrapErrServiceInternalMsg("result at index %d has fieldDatas length %d, expected %d (numGroupingKeys=%d, numAggs=%d)",
 				resultIdx, len(fieldDatas), expectedColumnCount, numGroupingKeys, numAggs)
 		}
 
 		// Check 2: no nil fieldData and Check 3: type matching
 		for colIdx, fieldData := range fieldDatas {
 			if fieldData == nil {
-				return fmt.Errorf("result at index %d has nil fieldData at column %d", resultIdx, colIdx)
+				return merr.WrapErrServiceInternalMsg("result at index %d has nil fieldData at column %d", resultIdx, colIdx)
 			}
 
 			expectedType := expectedTypes[colIdx]
 			actualType := fieldData.GetType()
 			if actualType != expectedType {
-				return fmt.Errorf("result at index %d, column %d has type %s, expected %s",
+				return merr.WrapErrServiceInternalMsg("result at index %d, column %d has type %s, expected %s",
 					resultIdx, colIdx, schemapb.DataType_name[int32(actualType)], schemapb.DataType_name[int32(expectedType)])
 			}
 		}
@@ -352,15 +351,15 @@ func (reducer *GroupAggReducer) Reduce(ctx context.Context, results []*Aggregati
 	limitReached := false
 	for _, result := range results {
 		if result == nil {
-			return nil, fmt.Errorf("input result from any sources cannot be nil")
+			return nil, merr.WrapErrServiceInternalMsg("input result from any sources cannot be nil")
 		}
 		reducedResult.allRetrieveCount += result.GetAllRetrieveCount()
 		fieldDatas := result.GetFieldDatas()
 		if outputColumnCount != len(fieldDatas) {
-			return nil, fmt.Errorf("retrieved results from different segments have different size of columns")
+			return nil, merr.WrapErrServiceInternalMsg("retrieved results from different segments have different size of columns")
 		}
 		if outputColumnCount == 0 {
-			return nil, fmt.Errorf("retrieved results have no column data")
+			return nil, merr.WrapErrServiceInternalMsg("retrieved results have no column data")
 		}
 		rowCount := -1
 		for i := 0; i < outputColumnCount; i++ {
@@ -374,11 +373,11 @@ func (reducer *GroupAggReducer) Reduce(ctx context.Context, results []*Aggregati
 				rowCount = hashers[i].RowCount()
 			} else if i < numGroupingKeys {
 				if rowCount != hashers[i].RowCount() {
-					return nil, fmt.Errorf("field data:%d for different columns have different row count, %d vs %d, wrong state",
+					return nil, merr.WrapErrServiceInternalMsg("field data:%d for different columns have different row count, %d vs %d, wrong state",
 						i, rowCount, hashers[i].RowCount())
 				}
 			} else if rowCount != accumulators[i-numGroupingKeys].RowCount() {
-				return nil, fmt.Errorf("field data:%d for different columns have different row count, %d vs %d, wrong state",
+				return nil, merr.WrapErrServiceInternalMsg("field data:%d for different columns have different row count, %d vs %d, wrong state",
 					i, rowCount, accumulators[i-numGroupingKeys].RowCount())
 			}
 		}
@@ -443,9 +442,9 @@ func (reducer *GroupAggReducer) Reduce(ctx context.Context, results []*Aggregati
 				}
 			}
 			if totalGroupCount > maxGroupByGroups {
-				return nil, merr.WrapErrServiceInternal(fmt.Sprintf("GROUP BY produced too many groups (%d). "+
+				return nil, merr.WrapErrParameterInvalidMsg("GROUP BY produced too many groups (%d). "+
 					"Add filters or increase common.groupBy.maxGroups (current: %d)",
-					totalGroupCount, maxGroupByGroups))
+					totalGroupCount, maxGroupByGroups)
 			}
 			// Don't guarantee specific groups to be returned before milvus support order by
 		}
@@ -478,7 +477,7 @@ func SegcoreResults2AggResult(results []*segcorepb.RetrieveResults) ([]*Aggregat
 	aggResults := make([]*AggregationResult, len(results))
 	for i := 0; i < len(results); i++ {
 		if results[i] == nil {
-			return nil, fmt.Errorf("input segcore query results from any sources cannot be nil")
+			return nil, merr.WrapErrServiceInternalMsg("input segcore query results from any sources cannot be nil")
 		}
 		fieldsData := results[i].GetFieldsData()
 		allRetrieveCount := results[i].GetAllRetrieveCount()

internal/agg/aggregate_util.go

@@ -9,6 +9,7 @@ import (
 	"unsafe"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 func NewFieldAccessor(fieldType schemapb.DataType) (FieldAccessor, error) {
@@ -28,7 +29,7 @@ func NewFieldAccessor(fieldType schemapb.DataType) (FieldAccessor, error) {
 	case schemapb.DataType_Double:
 		return newFloat64FieldAccessor(), nil
 	default:
-		return nil, fmt.Errorf("unsupported data type for hasher")
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported data type for hasher")
 	}
 }
 
@@ -410,7 +411,7 @@ func AssembleSingleValue(fv *FieldValue, fieldData *schemapb.FieldData) error {
 		case schemapb.DataType_VarChar, schemapb.DataType_String:
 			fieldData.GetScalars().GetStringData().Data = append(fieldData.GetScalars().GetStringData().GetData(), "")
 		default:
-			return fmt.Errorf("unsupported DataType:%d", fieldData.GetType())
+			return merr.WrapErrParameterInvalidMsg("unsupported DataType:%d", fieldData.GetType())
 		}
 		return nil
 	}
@@ -421,47 +422,47 @@ func AssembleSingleValue(fv *FieldValue, fieldData *schemapb.FieldData) error {
 	case schemapb.DataType_Bool:
 		boolVal, ok := val.(bool)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected bool, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected bool, got %T", val)
 		}
 		fieldData.GetScalars().GetBoolData().Data = append(fieldData.GetScalars().GetBoolData().GetData(), boolVal)
 	case schemapb.DataType_Int8, schemapb.DataType_Int16, schemapb.DataType_Int32:
 		intVal, ok := val.(int32)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected int32, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected int32, got %T", val)
 		}
 		fieldData.GetScalars().GetIntData().Data = append(fieldData.GetScalars().GetIntData().GetData(), intVal)
 	case schemapb.DataType_Int64:
 		int64Val, ok := val.(int64)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected int64, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected int64, got %T", val)
 		}
 		fieldData.GetScalars().GetLongData().Data = append(fieldData.GetScalars().GetLongData().GetData(), int64Val)
 	case schemapb.DataType_Timestamptz:
 		timestampVal, ok := val.(int64)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected int64 for Timestamptz, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected int64 for Timestamptz, got %T", val)
 		}
 		fieldData.GetScalars().GetTimestamptzData().Data = append(fieldData.GetScalars().GetTimestamptzData().GetData(), timestampVal)
 	case schemapb.DataType_Float:
 		floatVal, ok := val.(float32)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected float32, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected float32, got %T", val)
 		}
 		fieldData.GetScalars().GetFloatData().Data = append(fieldData.GetScalars().GetFloatData().GetData(), floatVal)
 	case schemapb.DataType_Double:
 		doubleVal, ok := val.(float64)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected float64, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected float64, got %T", val)
 		}
 		fieldData.GetScalars().GetDoubleData().Data = append(fieldData.GetScalars().GetDoubleData().GetData(), doubleVal)
 	case schemapb.DataType_VarChar, schemapb.DataType_String:
 		stringVal, ok := val.(string)
 		if !ok {
-			return fmt.Errorf("type assertion failed: expected string, got %T", val)
+			return merr.WrapErrServiceInternalMsg("type assertion failed: expected string, got %T", val)
 		}
 		fieldData.GetScalars().GetStringData().Data = append(fieldData.GetScalars().GetStringData().GetData(), stringVal)
 	default:
-		return fmt.Errorf("unsupported DataType:%d", fieldData.GetType())
+		return merr.WrapErrParameterInvalidMsg("unsupported DataType:%d", fieldData.GetType())
 	}
 	return nil
 }
@@ -544,13 +545,13 @@ func NewAggregationFieldMap(originalUserOutputFields []string, groupByFields []s
 			// 2. Global aggregation (no GROUP BY): output_fields can only contain aggregation expressions
 			//    (e.g., "SELECT count(*), int64 FROM t" is invalid SQL — cannot mix aggregates with raw columns)
 			if numGroupingKeys > 0 {
-				return nil, fmt.Errorf(
+				return nil, merr.WrapErrParameterInvalidMsg(
 					"output field '%s' is not allowed: when using GROUP BY, output_fields can only contain "+
 						"group_by fields (%v) or aggregation expressions",
 					outputField, groupByFields,
 				)
 			}
-			return nil, fmt.Errorf(
+			return nil, merr.WrapErrParameterInvalidMsg(
 				"output field '%s' is not allowed: when using aggregation functions (e.g., count(*)), "+
 					"output_fields can only contain aggregation expressions, not regular columns",
 				outputField,
@@ -566,14 +567,14 @@ func NewAggregationFieldMap(originalUserOutputFields []string, groupByFields []s
 // and returns a new Double FieldData containing the average values.
 func ComputeAvgFromSumAndCount(sumFieldData *schemapb.FieldData, countFieldData *schemapb.FieldData) (*schemapb.FieldData, error) {
 	if sumFieldData == nil || countFieldData == nil {
-		return nil, fmt.Errorf("sumFieldData and countFieldData cannot be nil")
+		return nil, merr.WrapErrServiceInternalMsg("sumFieldData and countFieldData cannot be nil")
 	}
 
 	sumType := sumFieldData.GetType()
 	countType := countFieldData.GetType()
 
 	if countType != schemapb.DataType_Int64 {
-		return nil, fmt.Errorf("count field must be Int64 type, got %s", countType.String())
+		return nil, merr.WrapErrParameterInvalidMsg("count field must be Int64 type, got %s", countType.String())
 	}
 
 	countData := countFieldData.GetScalars().GetLongData().GetData()
@@ -598,27 +599,27 @@ func ComputeAvgFromSumAndCount(sumFieldData *schemapb.FieldData, countFieldData
 	case schemapb.DataType_Int64:
 		sumData := sumFieldData.GetScalars().GetLongData().GetData()
 		if len(sumData) != rowCount {
-			return nil, fmt.Errorf("sum and count field data must have the same length, got sum:%d, count:%d", len(sumData), rowCount)
+			return nil, merr.WrapErrParameterInvalidMsg("sum and count field data must have the same length, got sum:%d, count:%d", len(sumData), rowCount)
 		}
 		for i := 0; i < rowCount; i++ {
 			if countData[i] == 0 {
-				return nil, fmt.Errorf("division by zero: count is 0 at row %d", i)
+				return nil, merr.WrapErrParameterInvalidMsg("division by zero: count is 0 at row %d", i)
 			}
 			resultData = append(resultData, float64(sumData[i])/float64(countData[i]))
 		}
 	case schemapb.DataType_Double:
 		sumData := sumFieldData.GetScalars().GetDoubleData().GetData()
 		if len(sumData) != rowCount {
-			return nil, fmt.Errorf("sum and count field data must have the same length, got sum:%d, count:%d", len(sumData), rowCount)
+			return nil, merr.WrapErrParameterInvalidMsg("sum and count field data must have the same length, got sum:%d, count:%d", len(sumData), rowCount)
 		}
 		for i := 0; i < rowCount; i++ {
 			if countData[i] == 0 {
-				return nil, fmt.Errorf("division by zero: count is 0 at row %d", i)
+				return nil, merr.WrapErrParameterInvalidMsg("division by zero: count is 0 at row %d", i)
 			}
 			resultData = append(resultData, sumData[i]/float64(countData[i]))
 		}
 	default:
-		return nil, fmt.Errorf("unsupported sum field type for avg computation: %s", sumType.String())
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported sum field type for avg computation: %s", sumType.String())
 	}
 
 	result.GetScalars().GetDoubleData().Data = resultData

internal/agg/operators.go

@@ -1,14 +1,13 @@
 package agg
 
 import (
-	"fmt"
-
 	"github.com/milvus-io/milvus/pkg/v3/proto/planpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 func terminateSingleSlot(slots []*FieldValue) (any, error) {
 	if len(slots) != 1 {
-		return nil, fmt.Errorf("aggregate expects 1 accumulator slot, got %d", len(slots))
+		return nil, merr.WrapErrParameterInvalidMsg("aggregate expects 1 accumulator slot, got %d", len(slots))
 	}
 	if slots[0] == nil || slots[0].IsNull() {
 		return nil, nil
@@ -18,7 +17,7 @@ func terminateSingleSlot(slots []*FieldValue) (any, error) {
 
 func terminateAvg(slots []*FieldValue) (any, error) {
 	if len(slots) != 2 {
-		return nil, fmt.Errorf("avg expects 2 accumulator slots, got %d", len(slots))
+		return nil, merr.WrapErrParameterInvalidMsg("avg expects 2 accumulator slots, got %d", len(slots))
 	}
 	sum := slots[0]
 	count := slots[1]
@@ -52,7 +51,7 @@ func fieldValueToFloat64(v any) (float64, error) {
 	case float64:
 		return value, nil
 	default:
-		return 0, fmt.Errorf("avg expects numeric accumulator, got %T", v)
+		return 0, merr.WrapErrParameterInvalidMsg("avg expects numeric accumulator, got %T", v)
 	}
 }
 
@@ -80,7 +79,7 @@ func (sum *SumAggregate) NewState() []*FieldValue {
 
 func (sum *SumAggregate) UpdateState(slots []*FieldValue, new *FieldValue) error {
 	if len(slots) != 1 {
-		return fmt.Errorf("aggregate expects 1 accumulator slot, got %d", len(slots))
+		return merr.WrapErrParameterInvalidMsg("aggregate expects 1 accumulator slot, got %d", len(slots))
 	}
 	return sum.Update(slots[0], new)
 }
@@ -121,7 +120,7 @@ func (count *CountAggregate) NewState() []*FieldValue {
 
 func (count *CountAggregate) UpdateState(slots []*FieldValue, new *FieldValue) error {
 	if len(slots) != 1 {
-		return fmt.Errorf("aggregate expects 1 accumulator slot, got %d", len(slots))
+		return merr.WrapErrParameterInvalidMsg("aggregate expects 1 accumulator slot, got %d", len(slots))
 	}
 	if new == nil || new.IsNull() {
 		return nil
@@ -166,7 +165,7 @@ func (avg *AvgAggregate) Name() string {
 }
 
 func (avg *AvgAggregate) Update(target *FieldValue, new *FieldValue) error {
-	return fmt.Errorf("avg aggregate updates require aggregate state slots")
+	return merr.WrapErrServiceInternalMsg("avg aggregate updates require aggregate state slots")
 }
 
 func (avg *AvgAggregate) NewState() []*FieldValue {
@@ -175,7 +174,7 @@ func (avg *AvgAggregate) NewState() []*FieldValue {
 
 func (avg *AvgAggregate) UpdateState(slots []*FieldValue, new *FieldValue) error {
 	if len(slots) != 2 {
-		return fmt.Errorf("avg expects 2 accumulator slots, got %d", len(slots))
+		return merr.WrapErrParameterInvalidMsg("avg expects 2 accumulator slots, got %d", len(slots))
 	}
 	if err := avg.sum.Update(slots[0], new); err != nil {
 		return err
@@ -204,7 +203,7 @@ func (avg *AvgAggregate) OriginalName() string {
 
 func updateOrderedState(target *FieldValue, new *FieldValue, op string) error {
 	if target == nil || new == nil {
-		return fmt.Errorf("target or new field value is nil")
+		return merr.WrapErrServiceInternalMsg("target or new field value is nil")
 	}
 	if new.IsNull() {
 		return nil
@@ -234,41 +233,41 @@ func shouldReplaceOrderedValue(target, new any, op string) (bool, error) {
 	case int:
 		newVal, ok := new.(int)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is int, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is int, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	case int32:
 		newVal, ok := new.(int32)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is int32, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is int32, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	case int64:
 		newVal, ok := new.(int64)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is int64, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is int64, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	case float32:
 		newVal, ok := new.(float32)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is float32, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is float32, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	case float64:
 		newVal, ok := new.(float64)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is float64, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is float64, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	case string:
 		newVal, ok := new.(string)
 		if !ok {
-			return false, fmt.Errorf("type mismatch: target is string, new is %T", new)
+			return false, merr.WrapErrParameterInvalidMsg("type mismatch: target is string, new is %T", new)
 		}
 		return shouldReplaceOrdered(newVal, targetVal, op), nil
 	default:
-		return false, fmt.Errorf("unsupported type for %s aggregation: %T", op, target)
+		return false, merr.WrapErrParameterInvalidMsg("unsupported type for %s aggregation: %T", op, target)
 	}
 }
 
@@ -299,7 +298,7 @@ func (min *MinAggregate) NewState() []*FieldValue {
 
 func (min *MinAggregate) UpdateState(slots []*FieldValue, new *FieldValue) error {
 	if len(slots) != 1 {
-		return fmt.Errorf("aggregate expects 1 accumulator slot, got %d", len(slots))
+		return merr.WrapErrParameterInvalidMsg("aggregate expects 1 accumulator slot, got %d", len(slots))
 	}
 	return min.Update(slots[0], new)
 }
@@ -339,7 +338,7 @@ func (max *MaxAggregate) NewState() []*FieldValue {
 
 func (max *MaxAggregate) UpdateState(slots []*FieldValue, new *FieldValue) error {
 	if len(slots) != 1 {
-		return fmt.Errorf("aggregate expects 1 accumulator slot, got %d", len(slots))
+		return merr.WrapErrParameterInvalidMsg("aggregate expects 1 accumulator slot, got %d", len(slots))
 	}
 	return max.Update(slots[0], new)
 }

internal/agg/type_check.go

@@ -1,9 +1,8 @@
 package agg
 
 import (
-	"fmt"
-
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 func isSupportedAggFieldType(aggregateName string, dt schemapb.DataType) bool {
@@ -51,7 +50,7 @@ func isSupportedAggFieldType(aggregateName string, dt schemapb.DataType) bool {
 // Note: operator name is expected to be normalized (lowercase) by caller.
 func ValidateAggFieldType(aggregateName string, dt schemapb.DataType) error {
 	if !isSupportedAggFieldType(aggregateName, dt) {
-		return fmt.Errorf("aggregation operator %s does not support data type %s", aggregateName, dt.String())
+		return merr.WrapErrParameterInvalidMsg("aggregation operator %s does not support data type %s", aggregateName, dt.String())
 	}
 	return nil
 }

internal/allocator/cached_allocator.go

@@ -18,14 +18,13 @@ package allocator
 
 import (
 	"context"
-	"fmt"
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 
 	"github.com/milvus-io/milvus/pkg/v3/log"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 const (
@@ -249,10 +248,9 @@ func (ta *CachedAllocator) finishSyncRequest() {
 func (ta *CachedAllocator) failRemainRequest() {
 	var err error
 	if ta.SyncErr != nil {
-		err = fmt.Errorf("%s failRemainRequest err:%w", ta.Role, ta.SyncErr)
+		err = merr.Wrapf(ta.SyncErr, "%s failRemainRequest", ta.Role)
 	} else {
-		errMsg := fmt.Sprintf("%s failRemainRequest unexpected error", ta.Role)
+		err = merr.WrapErrServiceInternalMsg("%s failRemainRequest unexpected error", ta.Role)
-		err = errors.New(errMsg)
 	}
 	if len(ta.ToDoReqs) > 0 {
 		log.Warn("Allocator has some reqs to fail",
@@ -290,8 +288,7 @@ func (ta *CachedAllocator) Close() {
 	ta.CancelFunc()
 	ta.wg.Wait()
 	ta.TChan.Close()
-	errMsg := fmt.Sprintf("%s is closing", ta.Role)
+	ta.revokeRequest(merr.WrapErrServiceInternalMsg("%s is closing", ta.Role))
-	ta.revokeRequest(errors.New(errMsg))
 }
 
 // CleanCache is used to force synchronize with global allocator.

internal/allocator/id_allocator.go

@@ -18,14 +18,12 @@ package allocator
 
 import (
 	"context"
-	"fmt"
 	"time"
 
-	"github.com/cockroachdb/errors"
-
 	"github.com/milvus-io/milvus-proto/go-api/v3/commonpb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/rootcoordpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/commonpbutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 const (
@@ -101,7 +99,7 @@ func (ia *IDAllocator) syncID() (bool, error) {
 
 	cancel()
 	if err != nil {
-		return false, fmt.Errorf("syncID Failed:%w", err)
+		return false, merr.Wrapf(err, "syncID failed")
 	}
 	ia.idStart = resp.GetID()
 	ia.idEnd = ia.idStart + int64(resp.GetCount())
@@ -148,7 +146,7 @@ func (ia *IDAllocator) AllocOne() (UniqueID, error) {
 // Alloc allocates the id of the count number.
 func (ia *IDAllocator) Alloc(count uint32) (UniqueID, UniqueID, error) {
 	if ia.closed() {
-		return 0, 0, errors.New("fail to allocate ID, closed allocator")
+		return 0, 0, merr.WrapErrServiceInternalMsg("fail to allocate ID, closed allocator")
 	}
 	req := &IDRequest{BaseRequest: BaseRequest{Done: make(chan error), Valid: false}}
 

internal/allocator/local_allocator.go

@@ -17,8 +17,9 @@
 package allocator
 
 import (
-	"fmt"
 	"sync"
+
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // localAllocator implements the Interface.
@@ -40,12 +41,12 @@ func NewLocalAllocator(start, end int64) Interface {
 func (a *localAllocator) Alloc(count uint32) (int64, int64, error) {
 	cnt := int64(count)
 	if cnt <= 0 {
-		return 0, 0, fmt.Errorf("non-positive count is not allowed, count=%d", cnt)
+		return 0, 0, merr.WrapErrParameterInvalidMsg("non-positive count is not allowed, count=%d", cnt)
 	}
 	a.mu.Lock()
 	defer a.mu.Unlock()
 	if a.idStart+cnt > a.idEnd {
-		return 0, 0, fmt.Errorf("ID is exhausted, start=%d, end=%d, count=%d", a.idStart, a.idEnd, cnt)
+		return 0, 0, merr.WrapErrServiceInternalMsg("ID is exhausted, start=%d, end=%d, count=%d", a.idStart, a.idEnd, cnt)
 	}
 	start := a.idStart
 	a.idStart += cnt

internal/cdc/cluster/milvus_client.go

@@ -20,7 +20,6 @@ import (
 	"context"
 	"crypto/tls"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 	"google.golang.org/grpc"
 
@@ -28,6 +27,7 @@ import (
 	"github.com/milvus-io/milvus-proto/go-api/v3/milvuspb"
 	"github.com/milvus-io/milvus/client/v2/milvusclient"
 	"github.com/milvus-io/milvus/pkg/v3/log"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 )
 
@@ -52,7 +52,7 @@ func NewMilvusClient(ctx context.Context, cluster *commonpb.MilvusCluster) (Milv
 	// Build TLS config from per-cluster paramtable config.
 	tlsConfig, err := buildCDCTLSConfig(cluster.GetClusterId())
 	if err != nil {
-		return nil, errors.Wrap(err, "failed to build CDC TLS config")
+		return nil, merr.Wrap(err, "failed to build CDC TLS config")
 	}
 	if tlsConfig != nil {
 		config.WithTLSConfig(tlsConfig)
@@ -67,7 +67,7 @@ func NewMilvusClient(ctx context.Context, cluster *commonpb.MilvusCluster) (Milv
 
 	cli, err := milvusclient.New(ctx, config)
 	if err != nil {
-		return nil, errors.Wrap(err, "failed to create milvus client")
+		return nil, merr.Wrap(err, "failed to create milvus client")
 	}
 	return cli, nil
 }

internal/cdc/meta/replicate_meta.go

@@ -5,11 +5,11 @@ import (
 	"fmt"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	clientv3 "go.etcd.io/etcd/client/v3"
 	"google.golang.org/protobuf/proto"
 
 	"github.com/milvus-io/milvus/pkg/v3/proto/streamingpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 const requestTimeout = 30 * time.Second
@@ -44,7 +44,7 @@ func ListReplicatePChannels(ctx context.Context, etcdCli *clientv3.Client, prefi
 		meta := &streamingpb.ReplicatePChannelMeta{}
 		err = proto.Unmarshal([]byte(value), meta)
 		if err != nil {
-			return nil, errors.Wrapf(err, "unmarshal replicate pchannel meta %s failed", keys[k])
+			return nil, merr.Wrapf(err, "unmarshal replicate pchannel meta %s failed", keys[k])
 		}
 		channel := &ReplicateChannel{
 			Key:         keys[k],

internal/cdc/replication/replicatemanager/channel_replicator.go

@@ -35,6 +35,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/message"
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/message/adaptor"
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/options"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/syncutil"
 )
@@ -202,7 +203,7 @@ func (r *channelReplicator) getReplicateCheckpoint() (*utility.ReplicateCheckpoi
 	}
 	replicateInfo, err := r.targetClient.GetReplicateInfo(ctx, req)
 	if err != nil {
-		return nil, errors.Wrap(err, "failed to get replicate info")
+		return nil, merr.Wrap(err, "failed to get replicate info")
 	}
 
 	checkpoint := replicateInfo.GetCheckpoint()

internal/cdc/replication/replicatestream/msg_queue.go

@@ -115,7 +115,7 @@ func (q *msgQueue) Enqueue(ctx context.Context, msg message.ImmutableMessage) er
 	// Optional runtime check: enforce non-decreasing timetick
 	// if n := len(q.buf); n > 0 {
 	//     if q.buf[n-1].Timetick() > msg.Timetick() {
-	//         return fmt.Errorf("enqueue timetick order violation: last=%d new=%d", q.buf[n-1].Timetick(), msg.Timetick())
+	//         return merr.WrapErrParameterInvalidMsg("enqueue timetick order violation: last=%d new=%d", q.buf[n-1].Timetick(), msg.Timetick())
 	//     }
 	// }
 

internal/coordinator/file_resource_observer.go

@@ -165,7 +165,7 @@ func (m *FileResourceObserver) CheckNodeSynced(nodeID int64) bool {
 func (m *FileResourceObserver) CheckAllQnReady() error {
 	// return error if meta is not ready
 	if m.meta == nil {
-		return merr.WrapErrParameterInvalidMsg("rootcoord meta is not ready")
+		return merr.WrapErrServiceUnavailable("rootcoord meta is not ready")
 	}
 
 	resources, version := m.meta.ListFileResource(m.ctx)
@@ -177,7 +177,7 @@ func (m *FileResourceObserver) CheckAllQnReady() error {
 	var err error
 	m.distribution.Range(func(nodeID int64, node *NodeInfo) bool {
 		if node.NodeType == QueryNode && node.Version < version {
-			err = merr.WrapErrParameterInvalidMsg("node %d file resource not synced", nodeID)
+			err = merr.WrapErrServiceUnavailableMsg("file resource not synced, node-%d", nodeID)
 			return false
 		}
 		return true

internal/coordinator/restful_mgr_routes.go

@@ -298,10 +298,10 @@ func (s *mixCoordImpl) getQueryNodes(ctx context.Context) (*querypb.ListQueryNod
 		Base: commonpbutil.NewMsgBase(),
 	})
 	if err != nil {
-		return nil, fmt.Errorf("failed to list query nodes: %w", err)
+		return nil, merr.Wrapf(err, "failed to list query nodes")
 	}
 	if !merr.Ok(resp.GetStatus()) {
-		return nil, fmt.Errorf("failed to list query nodes: %s", resp.GetStatus().GetReason())
+		return nil, merr.Wrapf(merr.Error(resp.GetStatus()), "failed to list query nodes")
 	}
 	return resp, nil
 }
@@ -514,7 +514,7 @@ func (s *mixCoordImpl) getQueryCoordChannelBalanceActive(ctx context.Context) (b
 	}
 
 	if !merr.Ok(resp.GetStatus()) {
-		return channelActivate, fmt.Errorf("%s", resp.GetStatus().GetReason())
+		return channelActivate, merr.Error(resp.GetStatus())
 	}
 
 	return channelActivate && resp.IsActive, nil
@@ -528,7 +528,7 @@ func (s *mixCoordImpl) getQueryCoordBalanceActive(ctx context.Context) (bool, er
 		return false, err
 	}
 	if !merr.Ok(resp.GetStatus()) {
-		return false, fmt.Errorf("%s", resp.GetStatus().GetReason())
+		return false, merr.Error(resp.GetStatus())
 	}
 	return resp.IsActive, nil
 }
@@ -569,7 +569,7 @@ func (s *mixCoordImpl) controlQueryCoordChannelBalanceStatus(ctx context.Context
 	default:
 		// If the status is not recognized, return an informative error immediately.
 		// This avoids proceeding with an invalid state.
-		err = fmt.Errorf("invalid status value: '%s'. Use 'suspended', 'resumed' or 'active'", status)
+		err = merr.WrapErrParameterInvalidMsg("invalid status value: '%s'. Use 'suspended', 'resumed' or 'active'", status)
 	}
 
 	// --- Unified Error Handling ---
@@ -578,7 +578,7 @@ func (s *mixCoordImpl) controlQueryCoordChannelBalanceStatus(ctx context.Context
 	// First, check if there was a low-level error during the method execution (e.g., network issue).
 	if err != nil {
 		// Wrap the original error with more context.
-		return fmt.Errorf("%s: %w", errMsg, err)
+		return merr.Wrap(err, errMsg)
 	}
 	// If no errors were encountered, the operation was successful. Return nil to indicate success.
 	return nil
@@ -958,11 +958,11 @@ func (s *mixCoordImpl) handleQueryNodeStatusUpdate(ctx context.Context, nodeID i
 		errMsg = "failed to resume node"
 	default:
 		errMsg = "invalid status value. Use 'suspended', 'resumed' or 'active'"
-		err = fmt.Errorf("%s", errMsg)
+		err = merr.WrapErrParameterInvalidMsg("%s", errMsg)
 	}
 	err = merr.CheckRPCCall(resp, err)
 	if err != nil {
-		err = fmt.Errorf("%s: %w", errMsg, err)
+		err = merr.Wrap(err, errMsg)
 	}
 	return err
 }

internal/coordinator/snmanager/streaming_node_manager.go

@@ -12,6 +12,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/types"
 	"github.com/milvus-io/milvus/pkg/v3/util/funcutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/syncutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -99,7 +100,7 @@ func (s *StreamingNodeManager) GetLatestWALLocated(ctx context.Context, vchannel
 	}
 	serverID, ok := balancer.GetLatestWALLocated(ctx, pchannel)
 	if !ok {
-		return -1, errors.Errorf("channel: %s not found", vchannel)
+		return -1, merr.WrapErrChannelNotFound(vchannel)
 	}
 	return serverID, nil
 }

internal/coordinator/wal_callbacks.go

@@ -19,12 +19,12 @@ package coordinator
 import (
 	"context"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 
 	"github.com/milvus-io/milvus/internal/streamingcoord/server/broadcaster/registry"
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/message"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 )
 
@@ -66,7 +66,7 @@ func (c *WALCallback) alterWALV2AckCallback(
 	targetWALName := result.Message.Header().TargetWalName
 	mqTypeValue := message.WALName(targetWALName).String()
 	if mqTypeValue == "unknown" {
-		return errors.Errorf("invalid target WAL name: %v", targetWALName)
+		return merr.WrapErrServiceInternalMsg("invalid target WAL name: %v", targetWALName)
 	}
 
 	// Get etcd source to update configuration
@@ -74,7 +74,7 @@ func (c *WALCallback) alterWALV2AckCallback(
 	etcdSource, ok := paramMgr.GetEtcdSource()
 	if !ok {
 		logger.Warn("failed to update mq.type config, etcd source not enabled")
-		return errors.New("etcd source is not enabled, cannot update mq.type configuration")
+		return merr.WrapErrServiceInternalMsg("etcd source is not enabled, cannot update mq.type configuration")
 	}
 
 	// Update mq.type configuration in etcd
@@ -84,7 +84,7 @@ func (c *WALCallback) alterWALV2AckCallback(
 			zap.String("configKey", configKey),
 			zap.String("mqTypeValue", mqTypeValue),
 			zap.Error(err))
-		return errors.Wrap(err, "failed to update mq.type configuration in etcd")
+		return merr.Wrap(err, "failed to update mq.type configuration in etcd")
 	}
 
 	logger.Info("successfully updated mq.type configuration in etcd",

internal/core/thirdparty/tantivy/tantivy-binding/src/analyzer/dict/lindera/fetch.rs

@@ -131,10 +131,13 @@ async fn download_with_retry(
         for url in urls {
             match client.get(url).send().await {
                 Ok(resp) if resp.status().is_success() => {
-                    let content = resp
+                    let content = match resp.bytes().await {
-                        .bytes()
+                        Ok(c) => c,
-                        .await
+                        Err(e) => {
-                        .map_err(|e| TantivyBindingError::InternalError(e.to_string()))?;
+                            warn!("Failed to read body from {}: {}", url, e);
+                            continue;
+                        }
+                    };
 
                     // Calculate MD5 hash
                     let mut context = Context::new();

internal/datacoord/analyze_meta.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"sync"
 
 	"go.uber.org/zap"
@@ -28,6 +27,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/workerpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/timerecord"
 )
 
@@ -117,7 +117,7 @@ func (m *analyzeMeta) UpdateVersion(taskID int64, nodeID int64) error {
 
 	t, ok := m.tasks[taskID]
 	if !ok {
-		return fmt.Errorf("there is no task with taskID: %d", taskID)
+		return merr.WrapErrServiceInternalMsg("there is no task with taskID: %d", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.AnalyzeTask)
@@ -134,7 +134,7 @@ func (m *analyzeMeta) BuildingTask(taskID int64) error {
 
 	t, ok := m.tasks[taskID]
 	if !ok {
-		return fmt.Errorf("there is no task with taskID: %d", taskID)
+		return merr.WrapErrServiceInternalMsg("there is no task with taskID: %d", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.AnalyzeTask)
@@ -150,7 +150,7 @@ func (m *analyzeMeta) UpdateState(taskID int64, state indexpb.JobState, failReas
 
 	t, ok := m.tasks[taskID]
 	if !ok {
-		return fmt.Errorf("there is no task with taskID: %d", taskID)
+		return merr.WrapErrServiceInternalMsg("there is no task with taskID: %d", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.AnalyzeTask)
@@ -168,7 +168,7 @@ func (m *analyzeMeta) FinishTask(taskID int64, result *workerpb.AnalyzeResult) e
 
 	t, ok := m.tasks[taskID]
 	if !ok {
-		return fmt.Errorf("there is no task with taskID: %d", taskID)
+		return merr.WrapErrServiceInternalMsg("there is no task with taskID: %d", taskID)
 	}
 
 	log.Info("finish task meta...", zap.Int64("taskID", taskID), zap.String("state", result.GetState().String()),

internal/datacoord/backfill_result.go

@@ -21,10 +21,9 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/cockroachdb/errors"
-
 	"github.com/milvus-io/milvus/internal/storage"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // BackfillResult is the decoded form of the JSON produced by the Spark
@@ -86,34 +85,34 @@ var knownObjectSchemes = map[string]struct{}{
 //  3. Unknown scheme -> return ErrUnsupportedScheme.
 func normalizeObjectKey(raw, expectedBucket string) (string, error) {
 	if raw == "" {
-		return "", errors.New("empty object path")
+		return "", merr.WrapErrServiceInternalMsg("empty object path")
 	}
 	if !strings.Contains(raw, "://") {
 		key := strings.TrimPrefix(raw, "/")
 		if key == "" {
-			return "", errors.Newf("empty object key in %q", raw)
+			return "", merr.WrapErrServiceInternalMsg("empty object key in %q", raw)
 		}
 		return key, nil
 	}
 	idx := strings.Index(raw, "://")
 	scheme := strings.ToLower(raw[:idx])
 	if _, ok := knownObjectSchemes[scheme]; !ok {
-		return "", errors.Newf("unsupported object URI scheme %q in %q", scheme, raw)
+		return "", merr.WrapErrServiceInternalMsg("unsupported object URI scheme %q in %q", scheme, raw)
 	}
 	rest := raw[idx+3:]
 	slash := strings.Index(rest, "/")
 	if slash < 0 {
-		return "", errors.Newf("malformed object URI %q: missing object key", raw)
+		return "", merr.WrapErrServiceInternalMsg("malformed object URI %q: missing object key", raw)
 	}
 	bucket := rest[:slash]
 	key := rest[slash+1:]
 	if expectedBucket != "" && bucket != expectedBucket {
-		return "", errors.Newf("object URI bucket %q differs from datacoord bucket %q (path=%s)", bucket, expectedBucket, raw)
+		return "", merr.WrapErrServiceInternalMsg("object URI bucket %q differs from datacoord bucket %q (path=%s)", bucket, expectedBucket, raw)
 	}
 	// Reject inputs like "s3a://bucket/" that parse to an empty key -- passing
 	// an empty key to chunkManager.Read has undefined behavior across SDKs.
 	if key == "" {
-		return "", errors.Newf("empty object key in %q", raw)
+		return "", merr.WrapErrServiceInternalMsg("empty object key in %q", raw)
 	}
 	return key, nil
 }
@@ -147,20 +146,20 @@ func buildV2Groups(bucket string, entry *BackfillSegment) (map[int64]*datapb.Fie
 	for i := range entry.ColumnGroups {
 		g := &entry.ColumnGroups[i]
 		if len(g.FieldIDs) != 1 {
-			return nil, errors.Newf("backfill invariant violated: column group has %d field_ids (expected 1)", len(g.FieldIDs))
+			return nil, merr.WrapErrServiceInternalMsg("backfill invariant violated: column group has %d field_ids (expected 1)", len(g.FieldIDs))
 		}
 		n := int64(len(g.BinlogFiles))
 		if n == 0 {
-			return nil, errors.Newf("column group for field %d has no binlog files", g.FieldIDs[0])
+			return nil, merr.WrapErrServiceInternalMsg("column group for field %d has no binlog files", g.FieldIDs[0])
 		}
 		fid := g.FieldIDs[0]
 		if _, dup := out[fid]; dup {
-			return nil, errors.Newf("duplicate column group for field %d", fid)
+			return nil, merr.WrapErrServiceInternalMsg("duplicate column group for field %d", fid)
 		}
 		// row_count flows into EntriesNum; non-positive values are undefined
 		// (zero collapses presence markers, negatives break accounting).
 		if g.RowCount <= 0 {
-			return nil, errors.Newf("column group for field %d has non-positive row_count %d", fid, g.RowCount)
+			return nil, merr.WrapErrServiceInternalMsg("column group for field %d has non-positive row_count %d", fid, g.RowCount)
 		}
 		avg := g.RowCount / n
 		rem := g.RowCount - avg*n
@@ -176,7 +175,7 @@ func buildV2Groups(bucket string, entry *BackfillSegment) (map[int64]*datapb.Fie
 			}
 			logID, ok := parseLogIDFromKey(key)
 			if !ok {
-				return nil, errors.Newf("column group for field %d has binlog file %q with non-numeric trailing segment", fid, p)
+				return nil, merr.WrapErrServiceInternalMsg("column group for field %d has binlog file %q with non-numeric trailing segment", fid, p)
 			}
 			// LogPath must be empty at persistence time -- catalog.checkLogID
 			// rejects any Binlog with LogPath != "" (see

internal/datacoord/compaction_policy_clustering.go

@@ -31,6 +31,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/common"
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 )
 
@@ -241,7 +242,7 @@ func calculateClusteringCompactionConfig(coll *collectionInfo, view CompactionVi
 
 func estimateRowsBySegmentSize(segments []*SegmentView, expectedSegmentSize int64) (int64, error) {
 	if expectedSegmentSize <= 0 {
-		return 0, fmt.Errorf("invalid expected segment size %d", expectedSegmentSize)
+		return 0, merr.WrapErrServiceInternalMsg("invalid expected segment size %d", expectedSegmentSize)
 	}
 
 	var totalSize float64
@@ -258,17 +259,17 @@ func estimateRowsBySegmentSize(segments []*SegmentView, expectedSegmentSize int6
 	}
 
 	if totalRows == 0 || totalSize == 0 {
-		return 0, fmt.Errorf("segment view does not contain size info to estimate row count")
+		return 0, merr.WrapErrServiceInternalMsg("segment view does not contain size info to estimate row count")
 	}
 
 	rowSize := totalSize / float64(totalRows)
 	if rowSize <= 0 {
-		return 0, fmt.Errorf("invalid row size calculated from segment view")
+		return 0, merr.WrapErrServiceInternalMsg("invalid row size calculated from segment view")
 	}
 
 	rows := float64(expectedSegmentSize) / rowSize
 	if rows <= 0 {
-		return 0, fmt.Errorf("estimated max row count is not positive")
+		return 0, merr.WrapErrServiceInternalMsg("estimated max row count is not positive")
 	}
 
 	return int64(rows), nil

internal/datacoord/compaction_policy_forcemerge.go

@@ -2,7 +2,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"math"
 
 	"github.com/samber/lo"
@@ -102,7 +101,7 @@ func (policy *forceMergeCompactionPolicy) triggerOneCollection(
 
 	configMaxSize := getExpectedSegmentSize(policy.meta, collectionID, collection.Schema)
 	if targetSizeBytes < configMaxSize {
-		return nil, 0, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("targetSize %d MB should be greater than or equal to configMaxSize %d MB", targetSize, configMaxSize/(1024*1024)))
+		return nil, 0, merr.WrapErrParameterInvalidMsg("targetSize %d MB should be greater than or equal to configMaxSize %d MB", targetSize, configMaxSize/(1024*1024))
 	}
 
 	segments := policy.meta.SelectSegments(ctx, WithCollection(collectionID), SegmentFilterFunc(func(segment *SegmentInfo) bool {
@@ -182,7 +181,7 @@ var _ CollectionTopologyQuerier = (*metricsNodeMemoryQuerier)(nil)
 func (q *metricsNodeMemoryQuerier) GetCollectionTopology(ctx context.Context, collectionID int64) (*CollectionTopology, error) {
 	log := log.Ctx(ctx).With(zap.Int64("collectionID", collectionID))
 	if q.mixCoord == nil {
-		return nil, fmt.Errorf("mixCoord not available for topology query")
+		return nil, merr.WrapErrServiceInternalMsg("mixCoord not available for topology query")
 	}
 
 	// 1. Get replica information

internal/datacoord/compaction_queue.go

@@ -74,9 +74,12 @@ func (pq *PriorityQueue[T]) Update(item *Item[T], value T, priority int) {
 	heap.Fix(pq, item.index)
 }
 
+// errFull / errNoSuchElement are INTERNAL sentinels: caught by errors.Is
+// inside the compaction inspector / scheduler loop and never serialized
+// across any gRPC boundary. See docs/dev/error_sentinel_convention.md.
 var (
-	ErrFull          = errors.New("compaction queue is full")
+	errFull          = errors.New("compaction queue is full")
-	ErrNoSuchElement = errors.New("compaction queue has no element")
+	errNoSuchElement = errors.New("compaction queue has no element")
 )
 
 type Prioritizer func(t CompactionTask) int
@@ -101,7 +104,7 @@ func (q *CompactionQueue) Enqueue(t CompactionTask) error {
 	q.lock.Lock()
 	defer q.lock.Unlock()
 	if q.capacity > 0 && len(q.pq) >= q.capacity {
-		return ErrFull
+		return errFull
 	}
 
 	heap.Push(&q.pq, &Item[CompactionTask]{value: t, priority: q.prioritizer(t)})
@@ -113,7 +116,7 @@ func (q *CompactionQueue) Dequeue() (CompactionTask, error) {
 	defer q.lock.Unlock()
 
 	if len(q.pq) == 0 {
-		return nil, ErrNoSuchElement
+		return nil, errNoSuchElement
 	}
 
 	item := heap.Pop(&q.pq).(*Item[CompactionTask])

internal/datacoord/compaction_task_clustering.go

@@ -23,7 +23,6 @@ import (
 	"strconv"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.uber.org/atomic"
 	"go.uber.org/zap"
@@ -629,7 +628,7 @@ func (t *clusteringCompactionTask) processAnalyzing() error {
 		}
 	case indexpb.JobState_JobStateFailed:
 		log.Warn("analyze task fail", zap.Int64("analyzeID", t.GetTaskProto().GetAnalyzeTaskID()))
-		return errors.New(analyzeTask.FailReason)
+		return merr.WrapErrServiceInternalMsg(analyzeTask.FailReason)
 	default:
 	}
 	return nil

internal/datacoord/compaction_task_l0.go

@@ -319,7 +319,7 @@ func (t *l0CompactionTask) selectFlushedSegment() ([]*SegmentInfo, []*datapb.Com
 	for _, info := range flushedSegments {
 		// Sealed is unexpected, fail fast
 		if info.GetState() == commonpb.SegmentState_Sealed {
-			return nil, nil, fmt.Errorf("L0 compaction selected invalid sealed segment %d", info.GetID())
+			return nil, nil, merr.WrapErrServiceInternalMsg("L0 compaction selected invalid sealed segment %d", info.GetID())
 		}
 
 		sealedSegBinlogs = append(sealedSegBinlogs, &datapb.CompactionSegmentBinlogs{
@@ -455,7 +455,7 @@ func buildL0V3DeltaLogEntries(segmentID int64, deltalogs []*datapb.FieldBinlog)
 		for _, binlog := range fieldBinlog.GetBinlogs() {
 			path := binlog.GetLogPath()
 			if path == "" {
-				return nil, merr.WrapErrServiceInternal(fmt.Sprintf("L0 V3 compaction result missing deltalog path for segment %d, logID %d", segmentID, binlog.GetLogID()))
+				return nil, merr.WrapErrServiceInternalMsg("L0 V3 compaction result missing deltalog path for segment %d, logID %d", segmentID, binlog.GetLogID())
 			}
 			entries = append(entries, packed.DeltaLogEntry{
 				Path:       path,

internal/datacoord/compaction_task_mix.go

@@ -398,7 +398,7 @@ func (t *mixCompactionTask) BuildCompactionRequest() (*datapb.CompactionPlan, er
 		resources, err := t.meta.GetFileResources(context.Background(), taskSchema.GetFileResourceIds()...)
 		if err != nil {
 			log.Warn("get file resources for collection failed", zap.Int64("collectionID", taskProto.GetCollectionID()), zap.Error(err))
-			return nil, errors.Errorf("get file resources for sort compaction failed: %v", err)
+			return nil, merr.Wrap(err, "get file resources for sort compaction failed")
 		}
 		plan.FileResources = resources
 	}
@@ -411,7 +411,7 @@ func (t *mixCompactionTask) BuildCompactionRequest() (*datapb.CompactionPlan, er
 			return nil, merr.WrapErrSegmentNotFound(segID)
 		}
 		if taskSchemaVersion < segInfo.GetSchemaVersion() {
-			return nil, merr.WrapErrIllegalCompactionPlan(fmt.Sprintf("compaction task schema version %d is older than input segment %d schema version %d", taskSchemaVersion, segInfo.GetID(), segInfo.GetSchemaVersion()))
+			return nil, merr.WrapErrIllegalCompactionPlanMsg("compaction task schema version %d is older than input segment %d schema version %d", taskSchemaVersion, segInfo.GetID(), segInfo.GetSchemaVersion())
 		}
 		plan.SegmentBinlogs = append(plan.SegmentBinlogs, &datapb.CompactionSegmentBinlogs{
 			SegmentID:           segID,

internal/datacoord/compaction_trigger.go

@@ -225,7 +225,7 @@ func (t *compactionTrigger) getCollection(collectionID UniqueID) (*collectionInf
 	defer cancel()
 	coll, err := t.handler.GetCollection(ctx, collectionID)
 	if err != nil {
-		return nil, fmt.Errorf("collection ID %d not found, err: %w", collectionID, err)
+		return nil, merr.Wrapf(err, "collection ID %d not found", collectionID)
 	}
 	return coll, nil
 }
@@ -588,7 +588,10 @@ func (t *compactionTrigger) getCandidates(signal *compactionSignal) ([]chanPartS
 	segments := t.meta.SelectSegments(context.TODO(), filters...)
 	// some criterion not met or conflicted
 	if len(signal.segmentIDs) > 0 && len(segments) != len(signal.segmentIDs) {
-		return nil, merr.WrapErrServiceInternal("not all segment ids provided could be compacted")
+		// SelectSegments also filters segments that are transiently mid-flush /
+		// compacting / just dropped, so a count mismatch is usually server-side
+		// state, not a bad id from the caller.
+		return nil, merr.WrapErrServiceInternalMsg("not all segment ids provided could be compacted")
 	}
 
 	type category struct {

internal/datacoord/copy_segment_task.go

@@ -583,7 +583,7 @@ func AssembleCopySegmentRequest(task CopySegmentTask, job CopySegmentJob) (*data
 			if _, exists := newBuildIDs[srcBuildID]; !exists {
 				newID, err := t.alloc.AllocID(ctx)
 				if err != nil {
-					return merr.WrapErrServiceInternal(fmt.Sprintf("failed to allocate new buildID for source buildID %d", srcBuildID), err.Error())
+					return merr.Wrapf(err, "failed to allocate new buildID for source buildID %d", srcBuildID)
 				}
 				newBuildIDs[srcBuildID] = newID
 			}

internal/datacoord/ddl_callbacks.go

@@ -27,6 +27,7 @@ import (
 	"github.com/milvus-io/milvus/internal/streamingcoord/server/broadcaster/registry"
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/streaming/util/message"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // RegisterDDLCallbacks registers the ddl callbacks.
@@ -89,7 +90,7 @@ func (s *Server) startBroadcastWithCollectionID(ctx context.Context, collectionI
 func (s *Server) startBroadcastForRestoreSnapshot(ctx context.Context, collectionID int64, snapshotName string) (broadcaster.BroadcastAPI, error) {
 	coll, err := s.broker.DescribeCollectionInternal(ctx, collectionID)
 	if err != nil {
-		return nil, fmt.Errorf("collection %d does not exist: %w", collectionID, err)
+		return nil, merr.Wrapf(err, "collection %d does not exist", collectionID)
 	}
 	dbName := coll.GetDbName()
 	collectionName := coll.GetCollectionName()
@@ -161,15 +162,15 @@ func (s *Server) validateRestoreSnapshotResources(ctx context.Context, collectio
 	sourceCollectionID := snapshotData.SnapshotInfo.GetCollectionId()
 	snapshot, err := s.meta.snapshotMeta.GetSnapshot(ctx, sourceCollectionID, snapshotData.SnapshotInfo.GetName())
 	if err != nil {
-		return fmt.Errorf("snapshot %s does not exist for collection %d: %w",
+		return merr.Wrapf(err, "snapshot %s does not exist for collection %d",
-			snapshotData.SnapshotInfo.GetName(), sourceCollectionID, err)
+			snapshotData.SnapshotInfo.GetName(), sourceCollectionID)
 	}
 	log.Info("snapshot validated", zap.String("snapshotName", snapshot.GetName()))
 
 	// ========== Validate Collection Exists ==========
 	coll, err := s.broker.DescribeCollectionInternal(ctx, collectionID)
 	if err != nil {
-		return fmt.Errorf("collection %d does not exist: %w", collectionID, err)
+		return merr.Wrapf(err, "collection %d does not exist", collectionID)
 	}
 	dbName := coll.GetDbName()
 	collectionName := coll.GetCollectionName()
@@ -180,7 +181,7 @@ func (s *Server) validateRestoreSnapshotResources(ctx context.Context, collectio
 	// ========== Validate Partitions Exist ==========
 	partitionsResp, err := s.broker.ShowPartitions(ctx, collectionID)
 	if err != nil {
-		return fmt.Errorf("failed to get partitions for collection %d: %w", collectionID, err)
+		return merr.Wrapf(err, "failed to get partitions for collection %d", collectionID)
 	}
 
 	// Build set of existing partition names
@@ -192,8 +193,7 @@ func (s *Server) validateRestoreSnapshotResources(ctx context.Context, collectio
 	// Check all snapshot partitions exist
 	for partName := range snapshotData.Collection.GetPartitions() {
 		if !existingPartitions[partName] {
-			return fmt.Errorf("partition %s does not exist in collection %d",
+			return merr.WrapErrPartitionNotFound(partName, fmt.Sprintf("partition does not exist in collection %d", collectionID))
-				partName, collectionID)
 		}
 	}
 	log.Info("partitions validated", zap.Int("count", len(existingPartitions)))
@@ -211,8 +211,7 @@ func (s *Server) validateRestoreSnapshotResources(ctx context.Context, collectio
 			}
 		}
 		if !indexFound {
-			return fmt.Errorf("index %s for field %d does not exist in collection %d",
+			return merr.WrapErrIndexNotFound(indexInfo.GetIndexName(), fmt.Sprintf("index for field %d does not exist in collection %d", indexInfo.GetFieldID(), collectionID))
-				indexInfo.GetIndexName(), indexInfo.GetFieldID(), collectionID)
 		}
 	}
 	log.Info("indexes validated", zap.Int("count", len(snapshotData.Indexes)))

internal/datacoord/ddl_callbacks_import.go

@@ -134,10 +134,10 @@ func (s *Server) validateImportReplication(ctx context.Context, options []*commo
 	}
 
 	if !paramtable.Get().DataCoordCfg.ImportInReplicatingCluster.GetAsBool() {
-		return merr.WrapErrImportFailed("import in replicating cluster is not supported yet")
+		return merr.WrapErrOperationNotSupportedMsg("import in replicating cluster is not supported yet")
 	}
 	if importutilv2.IsAutoCommit(options) {
-		return merr.WrapErrImportFailed("auto_commit=true import in replicating cluster is not supported")
+		return merr.WrapErrOperationNotSupportedMsg("auto_commit=true import in replicating cluster is not supported")
 	}
 	return nil
 }
@@ -168,14 +168,14 @@ func (s *Server) broadcastImport(ctx context.Context,
 
 	// Validate the request before broadcasting
 	if err := s.validateImportRequest(ctx, msgFiles, options); err != nil {
-		return errors.Wrap(err, "failed to validate import request")
+		return merr.Wrap(err, "failed to validate import request")
 	}
 
 	// Get database name from collection metadata via broker
 	// This is safer than extracting from schema which may be stale
 	broadcaster, err := s.startBroadcastWithCollectionID(ctx, collectionID)
 	if err != nil {
-		return errors.Wrap(err, "failed to start broadcast with collection id")
+		return merr.Wrap(err, "failed to start broadcast with collection id")
 	}
 	defer broadcaster.Close()
 

internal/datacoord/ddl_callbacks_import_test.go

@@ -107,8 +107,9 @@ func (s *ImportCallbacksSuite) TestValidateImportRequest_MaxJobsExceededReturnsE
 	err := server.validateImportRequest(ctx, files, options)
 
 	s.Error(err)
-	// ValidateMaxImportJobExceed returns WrapErrImportFailed, not ErrServiceQuotaExceeded
+	// Job-count backpressure is a server-side condition -> ErrImportSysFailed
-	s.True(errors.Is(err, merr.ErrImportFailed))
+	// (must not be bucketed as fail_input).
+	s.True(errors.Is(err, merr.ErrImportSysFailed))
 	s.Contains(err.Error(), "The number of jobs has reached the limit")
 }
 
@@ -185,7 +186,7 @@ func (s *ImportCallbacksSuite) TestValidateImportRequest_ReplicatingClusterRetur
 	err := server.validateImportRequest(ctx, files, options)
 
 	s.Error(err)
-	s.True(errors.Is(err, merr.ErrImportFailed))
+	s.True(errors.Is(err, merr.ErrOperationNotSupported))
 	s.Contains(err.Error(), "replicating cluster")
 }
 
@@ -230,7 +231,7 @@ func (s *ImportCallbacksSuite) TestValidateImportRequest_ReplicatingClusterEnabl
 		{Key: "timeout", Value: "300s"},
 	})
 	s.Error(err)
-	s.True(errors.Is(err, merr.ErrImportFailed))
+	s.True(errors.Is(err, merr.ErrOperationNotSupported))
 	s.Contains(err.Error(), "auto_commit=true")
 
 	err = server.validateImportRequest(ctx, files, []*commonpb.KeyValuePair{
@@ -1159,7 +1160,8 @@ func testBroadcastRequiresVchannels(t *testing.T, broadcastFn func(*Server, cont
 
 	err := broadcastFn(server, ctx, job)
 	assert.Error(t, err)
-	assert.True(t, errors.Is(err, merr.ErrImportFailed))
+	// Missing vchannels is internal broadcast state -> ErrImportSysFailed.
+	assert.True(t, errors.Is(err, merr.ErrImportSysFailed))
 	assert.Contains(t, err.Error(), "job 7 has no vchannels")
 }
 

internal/datacoord/ddl_callbacks_snapshot_test.go

@@ -636,7 +636,8 @@ func TestValidateRestoreSnapshotResources_IndexNotFound(t *testing.T) {
 	err := server.validateRestoreSnapshotResources(ctx, 100, buildBaseSnapshotData())
 
 	assert.Error(t, err)
-	assert.Contains(t, err.Error(), "index vec_idx for field 100 does not exist")
+	assert.Contains(t, err.Error(), "index for field 100 does not exist")
+	assert.Contains(t, err.Error(), "vec_idx")
 }
 
 func TestValidateRestoreSnapshotResources_IndexFieldMismatch(t *testing.T) {
@@ -656,7 +657,8 @@ func TestValidateRestoreSnapshotResources_IndexFieldMismatch(t *testing.T) {
 	err := server.validateRestoreSnapshotResources(ctx, 100, buildBaseSnapshotData())
 
 	assert.Error(t, err)
-	assert.Contains(t, err.Error(), "index vec_idx for field 100 does not exist")
+	assert.Contains(t, err.Error(), "index for field 100 does not exist")
+	assert.Contains(t, err.Error(), "vec_idx")
 }
 
 func TestValidateRestoreSnapshotResources_Success(t *testing.T) {

internal/datacoord/errors.go

@@ -20,6 +20,9 @@ import (
 	"fmt"
 
 	"github.com/cockroachdb/errors"
+
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
+	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
 
 // errors for VerifyResponse
@@ -34,7 +37,7 @@ func msgDataCoordIsUnhealthy(coordID UniqueID) string {
 }
 
 func errDataCoordIsUnhealthy(coordID UniqueID) error {
-	return errors.New(msgDataCoordIsUnhealthy(coordID))
+	return merr.WrapErrServiceNotReady(typeutil.DataCoordRole, coordID, "not ready")
 }
 
 func msgSegmentNotFound(segID UniqueID) string {

internal/datacoord/external_collection_refresh_manager.go

@@ -306,7 +306,7 @@ func (m *externalCollectionRefreshManager) cleanupExploreTempForJob(jobID int64)
 func (m *externalCollectionRefreshManager) applyFinishedJobSegments(ctx context.Context, job *datapb.ExternalCollectionRefreshJob) error {
 	tasks := m.refreshMeta.GetTasksByJobID(job.GetJobId())
 	if len(tasks) == 0 {
-		return fmt.Errorf("job %d has no tasks to apply", job.GetJobId())
+		return merr.WrapErrServiceInternalMsg("job %d has no tasks to apply", job.GetJobId())
 	}
 
 	keptSet := make(map[int64]struct{})
@@ -315,11 +315,11 @@ func (m *externalCollectionRefreshManager) applyFinishedJobSegments(ctx context.
 	updatedSegments := make([]*datapb.SegmentInfo, 0)
 	for _, task := range tasks {
 		if task.GetState() != indexpb.JobState_JobStateFinished {
-			return fmt.Errorf("job %d has non-finished task %d in state %s",
+			return merr.WrapErrServiceInternalMsg("job %d has non-finished task %d in state %s",
 				job.GetJobId(), task.GetTaskId(), task.GetState().String())
 		}
 		if !task.GetResultReady() {
-			return fmt.Errorf("job %d has finished task %d without persisted refresh result; please retry refresh",
+			return merr.WrapErrServiceInternalMsg("job %d has finished task %d without persisted refresh result; please retry refresh",
 				job.GetJobId(), task.GetTaskId())
 		}
 		for _, segmentID := range task.GetKeptSegments() {
@@ -334,7 +334,7 @@ func (m *externalCollectionRefreshManager) applyFinishedJobSegments(ctx context.
 				continue
 			}
 			if _, ok := updatedSet[segment.GetID()]; ok {
-				return fmt.Errorf("job %d has duplicate updated segment %d from task %d",
+				return merr.WrapErrServiceInternalMsg("job %d has duplicate updated segment %d from task %d",
 					job.GetJobId(), segment.GetID(), task.GetTaskId())
 			}
 			updatedSet[segment.GetID()] = struct{}{}
@@ -714,7 +714,7 @@ func (m *externalCollectionRefreshManager) createTasksForJob(
 		if errors.Is(err, packed.ErrLoonTransient) {
 			return nil, newNonRetriableJobError("explore external files failed: %v", err)
 		}
-		return nil, fmt.Errorf("failed to explore external files: %w", err)
+		return nil, merr.WrapErrServiceInternalErr(err, "failed to explore external files")
 	}
 	if len(allFiles) == 0 {
 		return nil, newNonRetriableJobError("no files found in external source: %s", job.GetExternalSource())
@@ -829,7 +829,7 @@ func normalizeRefreshJobProgress(job *datapb.ExternalCollectionRefreshJob, state
 func (m *externalCollectionRefreshManager) GetJobProgress(ctx context.Context, jobID int64) (*datapb.ExternalCollectionRefreshJob, error) {
 	job := m.refreshMeta.GetJob(jobID)
 	if job == nil {
-		return nil, fmt.Errorf("job %d not found", jobID)
+		return nil, merr.WrapErrParameterInvalidMsg("refresh job %d not found", jobID)
 	}
 
 	// Aggregate state and progress from tasks
@@ -877,17 +877,17 @@ func (m *externalCollectionRefreshManager) exploreExternalFiles(
 	// when both present.
 	if job.GetExternalSource() != "" {
 		if err := externalspec.ValidateSourceAndSpec(job.GetExternalSource(), job.GetExternalSpec()); err != nil {
-			return nil, "", fmt.Errorf("external source/spec failed revalidation: %w", err)
+			return nil, "", merr.Wrap(err, "external source/spec failed revalidation")
 		}
 	}
 	spec, err := externalspec.ParseExternalSpec(job.GetExternalSpec())
 	if err != nil {
-		return nil, "", fmt.Errorf("failed to parse external spec: %w", err)
+		return nil, "", merr.Wrap(err, "failed to parse external spec")
 	}
 
 	collInfo := m.mt.GetCollection(job.GetCollectionId())
 	if collInfo == nil {
-		return nil, "", fmt.Errorf("collection %d not found", job.GetCollectionId())
+		return nil, "", merr.WrapErrCollectionNotFound(job.GetCollectionId())
 	}
 
 	columns := packed.GetColumnNamesFromSchema(collInfo.Schema)
@@ -908,7 +908,7 @@ func (m *externalCollectionRefreshManager) exploreExternalFiles(
 		extfs,
 	)
 	if err != nil {
-		return nil, "", fmt.Errorf("ExploreFilesReturnManifestPath failed: %w", err)
+		return nil, "", merr.WrapErrServiceInternalErr(err, "failed to explore files returning manifest path")
 	}
 
 	// Convert to proto type

internal/datacoord/external_collection_refresh_meta.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"sort"
 	"time"
 
@@ -30,6 +29,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/lock"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/timerecord"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -295,7 +295,7 @@ func (m *externalCollectionRefreshMeta) mutateJob(
 ) (applied bool, err error) {
 	job, ok := m.jobs.Get(jobID)
 	if !ok {
-		return false, fmt.Errorf("job %d not found", jobID)
+		return false, merr.WrapErrServiceInternalMsg("job %d not found", jobID)
 	}
 
 	m.jobLock.Lock(job.GetCollectionId())
@@ -304,7 +304,7 @@ func (m *externalCollectionRefreshMeta) mutateJob(
 	// Re-fetch after lock
 	job, ok = m.jobs.Get(jobID)
 	if !ok {
-		return false, fmt.Errorf("job %d not found", jobID)
+		return false, merr.WrapErrServiceInternalMsg("job %d not found", jobID)
 	}
 
 	cloneJob := proto.Clone(job).(*datapb.ExternalCollectionRefreshJob)
@@ -385,7 +385,7 @@ func (m *externalCollectionRefreshMeta) UpdateJobStateWithPreApply(
 ) (bool, error) {
 	job, ok := m.jobs.Get(jobID)
 	if !ok {
-		return false, fmt.Errorf("job %d not found", jobID)
+		return false, merr.WrapErrServiceInternalMsg("job %d not found", jobID)
 	}
 
 	m.jobLock.Lock(job.GetCollectionId())
@@ -395,7 +395,7 @@ func (m *externalCollectionRefreshMeta) UpdateJobStateWithPreApply(
 	// terminal state owns the one-time side effects.
 	job, ok = m.jobs.Get(jobID)
 	if !ok {
-		return false, fmt.Errorf("job %d not found", jobID)
+		return false, merr.WrapErrServiceInternalMsg("job %d not found", jobID)
 	}
 	if job.GetState() == indexpb.JobState_JobStateFinished ||
 		job.GetState() == indexpb.JobState_JobStateFailed {
@@ -416,7 +416,7 @@ func (m *externalCollectionRefreshMeta) UpdateJobStateWithPreApply(
 				log.Warn("update job state after pre-apply failed",
 					zap.Int64("jobID", jobID),
 					zap.Error(saveErr))
-				return false, fmt.Errorf("pre-apply failed: %w; additionally failed to persist Failed job state: %v", err, saveErr)
+				return false, merr.Wrapf(err, "pre-apply failed; additionally failed to persist Failed job state: %v", saveErr)
 			}
 			m.jobs.Insert(jobID, cloneJob)
 			m.addToCollectionJobs(cloneJob)
@@ -613,7 +613,7 @@ func (m *externalCollectionRefreshMeta) mutateTask(
 ) (applied bool, cloned *datapb.ExternalCollectionRefreshTask, err error) {
 	task, ok := m.tasks.Get(taskID)
 	if !ok {
-		return false, nil, fmt.Errorf("task %d not found", taskID)
+		return false, nil, merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	m.taskLock.Lock(task.GetJobId())
@@ -622,7 +622,7 @@ func (m *externalCollectionRefreshMeta) mutateTask(
 	// Re-fetch after lock
 	task, ok = m.tasks.Get(taskID)
 	if !ok {
-		return false, nil, fmt.Errorf("task %d not found", taskID)
+		return false, nil, merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneTask := proto.Clone(task).(*datapb.ExternalCollectionRefreshTask)

internal/datacoord/garbage_collector.go

@@ -1067,13 +1067,13 @@ func (gc *garbageCollector) isExpire(dropts Timestamp) bool {
 // Returns segmentID or error if path doesn't match.
 func parseV3SegmentID(rootPath, filePath string) (int64, error) {
 	if !strings.HasPrefix(filePath, rootPath) {
-		return 0, fmt.Errorf("path %q does not contain rootPath %q", filePath, rootPath)
+		return 0, merr.WrapErrServiceInternalMsg("path %q does not contain rootPath %q", filePath, rootPath)
 	}
 	p := strings.TrimPrefix(filePath[len(rootPath):], "/")
 	parts := strings.Split(p, "/")
 	// Minimum: insert_log/coll/part/seg/something
 	if len(parts) < 5 || parts[0] != common.SegmentInsertLogPath {
-		return 0, fmt.Errorf("not a V3 insert_log path: %s", filePath)
+		return 0, merr.WrapErrServiceInternalMsg("not a V3 insert_log path: %s", filePath)
 	}
 	return strconv.ParseInt(parts[3], 10, 64)
 }

internal/datacoord/garbage_collector_lob.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"path"
 	"strings"
 	"sync"
@@ -34,6 +33,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -271,8 +271,7 @@ func (lobCtx *lobGCContext) collectLOBFilesFromSegment(ctx context.Context, segm
 
 	lobFiles, err := lobCtx.cache.Get(ctx, manifestPath, lobCtx.storageConfig)
 	if err != nil {
-		return fmt.Errorf("failed to get LOB files from manifest for segment %d (path=%s): %w",
+		return merr.WrapErrServiceInternalErr(err, "failed to get LOB files from manifest for segment %d (path=%s)", segment.GetID(), manifestPath)
-			segment.GetID(), manifestPath, err)
 	}
 
 	for _, lobFile := range lobFiles {

internal/datacoord/import_meta.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"time"
 
 	"github.com/hashicorp/golang-lru/v2/expirable"
@@ -345,7 +344,7 @@ func (m *importMeta) HandleCommitVchannel(ctx context.Context, jobID int64, vcha
 
 	job := m.jobs[jobID]
 	if job == nil {
-		return merr.WrapErrImportFailed(fmt.Sprintf("job %d not found", jobID))
+		return merr.WrapErrImportSysFailedMsg("job %d not found", jobID)
 	}
 	// Idempotency: if vchannel already committed, skip.
 	for _, c := range job.GetCommittedVchannels() {

internal/datacoord/import_services_test.go

@@ -104,7 +104,7 @@ func (s *ImportServicesSuite) TestImportV2_AllocatorNilReturnsError() {
 
 	s.NoError(err)
 	s.NotNil(resp)
-	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
 	s.Contains(resp.GetStatus().GetReason(), "allocator not initialized")
 }
 
@@ -114,7 +114,7 @@ func (s *ImportServicesSuite) TestImportV2_AllocatorFailsReturnsError() {
 	server.stateCode.Store(commonpb.StateCode_Healthy)
 
 	mockAllocator := allocator.NewMockAllocator(s.T())
-	mockAllocator.EXPECT().AllocN(mock.Anything).Return(int64(0), int64(0), errors.New("allocation failed"))
+	mockAllocator.EXPECT().AllocN(mock.Anything).Return(int64(0), int64(0), merr.WrapErrServiceUnavailable("allocation failed"))
 	server.allocator = mockAllocator
 
 	req := &internalpb.ImportRequestInternal{
@@ -127,7 +127,7 @@ func (s *ImportServicesSuite) TestImportV2_AllocatorFailsReturnsError() {
 
 	s.NoError(err)
 	s.NotNil(resp)
-	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
 	s.Contains(resp.GetStatus().GetReason(), "failed to allocate job ID")
 }
 
@@ -164,7 +164,7 @@ func (s *ImportServicesSuite) TestImportV2_BroadcastFailsReturnsError() {
 	// Mock StartBroadcastWithResourceKeys to fail
 	mockBroadcast := mockey.Mock(broadcast.StartBroadcastWithResourceKeys).To(
 		func(ctx context.Context, keys ...message.ResourceKey) (broadcaster.BroadcastAPI, error) {
-			return nil, errors.New("broadcast failed")
+			return nil, merr.WrapErrServiceUnavailable("broadcast failed")
 		}).Build()
 	defer mockBroadcast.UnPatch()
 
@@ -199,8 +199,8 @@ func (s *ImportServicesSuite) TestImportV2_BroadcastFailsReturnsError() {
 
 	s.NoError(err)
 	s.NotNil(resp)
-	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
-	s.Contains(resp.GetStatus().GetReason(), "failed to broadcast import")
+	s.Contains(resp.GetStatus().GetReason(), "broadcast")
 }
 
 func (s *ImportServicesSuite) TestImportV2_SuccessReturnsJobID() {
@@ -395,7 +395,7 @@ func (s *ImportServicesSuite) TestCreateImportJobFromAck_AllocatorFailsReturnsEr
 	server.stateCode.Store(commonpb.StateCode_Healthy)
 
 	mockAllocator := allocator.NewMockAllocator(s.T())
-	mockAllocator.EXPECT().AllocN(mock.Anything).Return(int64(0), int64(0), errors.New("allocation failed"))
+	mockAllocator.EXPECT().AllocN(mock.Anything).Return(int64(0), int64(0), merr.WrapErrServiceUnavailable("allocation failed"))
 	server.allocator = mockAllocator
 
 	req := &internalpb.ImportRequestInternal{
@@ -412,7 +412,7 @@ func (s *ImportServicesSuite) TestCreateImportJobFromAck_AllocatorFailsReturnsEr
 
 	s.NoError(err)
 	s.NotNil(resp)
-	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
 	s.Contains(resp.GetStatus().GetReason(), "alloc id failed")
 }
 
@@ -493,7 +493,7 @@ func (s *ImportServicesSuite) TestCreateImportJobFromAck_AddJobFailsReturnsError
 	catalog.EXPECT().ListImportJobs(mock.Anything).Return(nil, nil)
 	catalog.EXPECT().ListPreImportTasks(mock.Anything).Return(nil, nil)
 	catalog.EXPECT().ListImportTasks(mock.Anything).Return(nil, nil)
-	catalog.EXPECT().SaveImportJob(mock.Anything, mock.Anything).Return(errors.New("save job failed"))
+	catalog.EXPECT().SaveImportJob(mock.Anything, mock.Anything).Return(merr.WrapErrServiceUnavailable("save job failed"))
 
 	importMeta, err := NewImportMeta(context.TODO(), catalog, nil, nil)
 	s.NoError(err)
@@ -528,7 +528,7 @@ func (s *ImportServicesSuite) TestCreateImportJobFromAck_AddJobFailsReturnsError
 
 	s.NoError(err)
 	s.NotNil(resp)
-	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+	s.True(errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
 	s.Contains(resp.GetStatus().GetReason(), "add import job failed")
 }
 

internal/datacoord/import_util.go

@@ -670,8 +670,8 @@ func ListBinlogsAndGroupBySegment(ctx context.Context,
 		return nil, merr.WrapErrImportFailed("no insert binlogs to import")
 	}
 	if len(importFile.GetPaths()) > 2 {
-		return nil, merr.WrapErrImportFailed(fmt.Sprintf("too many input paths for binlog import. "+
+		return nil, merr.WrapErrImportFailedMsg("too many input paths for binlog import. "+
-			"Valid paths length should be one or two, but got paths:%s", importFile.GetPaths()))
+			"Valid paths length should be one or two, but got paths:%s", importFile.GetPaths())
 	}
 
 	insertPrefix := importFile.GetPaths()[0]
@@ -790,18 +790,18 @@ func ListBinlogImportRequestFiles(ctx context.Context, cm storage.ChunkManager,
 	}
 	err := conc.AwaitAll(futures...)
 	if err != nil {
-		return nil, merr.WrapErrImportFailed(fmt.Sprintf("list binlogs failed, err=%s", err))
+		return nil, merr.WrapErrServiceUnavailableMsg("list binlogs failed, err=%s", err)
 	}
 
 	resFiles = lo.Filter(resFiles, func(file *internalpb.ImportFile, _ int) bool {
 		return len(file.GetPaths()) > 0
 	})
 	if len(resFiles) == 0 {
-		return nil, merr.WrapErrImportFailed(fmt.Sprintf("no binlog to import, input=%s", reqFiles))
+		return nil, merr.WrapErrImportFailedMsg("no binlog to import, input=%s", reqFiles)
 	}
 	if len(resFiles) > paramtable.Get().DataCoordCfg.MaxFilesPerImportReq.GetAsInt() {
-		return nil, merr.WrapErrImportFailed(fmt.Sprintf("The max number of import files should not exceed %d, but got %d",
+		return nil, merr.WrapErrImportFailedMsg("The max number of import files should not exceed %d, but got %d",
-			paramtable.Get().DataCoordCfg.MaxFilesPerImportReq.GetAsInt(), len(resFiles)))
+			paramtable.Get().DataCoordCfg.MaxFilesPerImportReq.GetAsInt(), len(resFiles))
 	}
 	log.Info("list binlogs prefixes for import done", zap.Int("num", len(resFiles)), zap.Any("binlog_prefixes", resFiles))
 	return resFiles, nil
@@ -812,7 +812,7 @@ func ValidateMaxImportJobExceed(ctx context.Context, importMeta ImportMeta) erro
 	maxNum := paramtable.Get().DataCoordCfg.MaxImportJobNum.GetAsInt()
 	executingNum := importMeta.CountJobBy(ctx, WithoutJobStates(internalpb.ImportJobState_Completed, internalpb.ImportJobState_Failed))
 	if executingNum >= maxNum {
-		return merr.WrapErrImportFailed(
+		return merr.WrapErrImportSysFailed(
 			fmt.Sprintf("The number of jobs has reached the limit, please try again later. " +
 				"If your request is set to only import a single file, " +
 				"please consider importing multiple files in one request for better efficiency."))

internal/datacoord/index_meta.go

@@ -468,7 +468,7 @@ func (m *indexMeta) canCreateIndex(req *indexpb.CreateIndexRequest, isJSON bool)
 					index.IndexName, index.FieldID, index.IndexParams, index.UserIndexParams, index.TypeParams)),
 				zap.String("current index", fmt.Sprintf("{index_name: %s, field_id: %d, index_params: %v, user_params: %v, type_params: %v}",
 					req.GetIndexName(), req.GetFieldID(), req.GetIndexParams(), req.GetUserIndexParams(), req.GetTypeParams())))
-			return 0, fmt.Errorf("CreateIndex failed: %s", errMsg)
+			return 0, merr.WrapErrParameterInvalidMsg("%s", errMsg)
 		}
 		if req.FieldID == index.FieldID {
 			if isJSON {
@@ -484,7 +484,11 @@ func (m *indexMeta) canCreateIndex(req *indexpb.CreateIndexRequest, isJSON bool)
 			// creating multiple indexes on same field is not supported
 			errMsg := "CreateIndex failed: creating multiple indexes on same field is not supported"
 			log.Warn(errMsg)
-			return 0, errors.New(errMsg)
+			// Client-caused conflict, same family as the "one distinct index per
+			// field" case above: return an input-class error rather than
+			// ServiceInternal (code 5, "never return out of Milvus") so the
+			// caller is not blamed with a system error / counted as fail_system.
+			return 0, merr.WrapErrParameterInvalidMsg("%s", errMsg)
 		}
 	}
 	return 0, nil
@@ -1012,7 +1016,7 @@ func (m *indexMeta) UpdateVersion(buildID, nodeID UniqueID) error {
 	log.Ctx(m.ctx).Info("IndexCoord metaTable UpdateVersion receive", zap.Int64("buildID", buildID), zap.Int64("nodeID", nodeID))
 	segIdx, ok := m.segmentBuildInfo.Get(buildID)
 	if !ok {
-		return fmt.Errorf("there is no index with buildID: %d", buildID)
+		return merr.WrapErrServiceInternalMsg("there is no index with buildID: %d", buildID)
 	}
 
 	updateFunc := func(segIdx *model.SegmentIndex) error {
@@ -1031,7 +1035,7 @@ func (m *indexMeta) UpdateIndexState(buildID UniqueID, state commonpb.IndexState
 	segIdx, ok := m.segmentBuildInfo.Get(buildID)
 	if !ok {
 		log.Ctx(m.ctx).Warn("there is no index with buildID", zap.Int64("buildID", buildID))
-		return fmt.Errorf("there is no index with buildID: %d", buildID)
+		return merr.WrapErrServiceInternalMsg("there is no index with buildID: %d", buildID)
 	}
 
 	updateFunc := func(segIdx *model.SegmentIndex) error {
@@ -1141,7 +1145,7 @@ func (m *indexMeta) BuildIndex(buildID UniqueID) error {
 
 	segIdx, ok := m.segmentBuildInfo.Get(buildID)
 	if !ok {
-		return fmt.Errorf("there is no index with buildID: %d", buildID)
+		return merr.WrapErrServiceInternalMsg("there is no index with buildID: %d", buildID)
 	}
 
 	updateFunc := func(segIdx *model.SegmentIndex) error {

internal/datacoord/index_service.go

@@ -1052,7 +1052,7 @@ func (s *Server) DropIndex(ctx context.Context, req *indexpb.DropIndexRequest) (
 			}
 			if typeutil.IsVectorType(field.GetDataType()) {
 				log.Warn("vector index cannot be dropped on loaded collection", zap.String("indexName", req.IndexName), zap.Int64("collectionID", req.GetCollectionID()), zap.Int64("fieldID", index.FieldID))
-				return merr.Status(merr.WrapErrParameterInvalidMsg(fmt.Sprintf("vector index cannot be dropped on loaded collection: %d", req.GetCollectionID()))), nil
+				return merr.Status(merr.WrapErrParameterInvalidMsg("vector index cannot be dropped on loaded collection: %d", req.GetCollectionID())), nil
 			}
 		}
 	}

internal/datacoord/meta.go

@@ -816,7 +816,7 @@ func (m *meta) GetSegmentsChannels(segmentIDs []UniqueID) (map[int64]string, err
 	for _, segmentID := range segmentIDs {
 		segment := m.segments.GetSegment(segmentID)
 		if segment == nil {
-			return nil, errors.New(fmt.Sprintf("cannot find segment %d", segmentID))
+			return nil, merr.WrapErrServiceInternalMsg("cannot find segment %d", segmentID)
 		}
 		segChannels[segmentID] = segment.GetInsertChannel()
 	}
@@ -840,7 +840,7 @@ func (m *meta) SetState(ctx context.Context, segmentID UniqueID, targetState com
 		if targetState == commonpb.SegmentState_Dropped {
 			return nil
 		}
-		return fmt.Errorf("segment is not exist with ID = %d", segmentID)
+		return merr.WrapErrSegmentNotFound(segmentID)
 	}
 	// Persist segment updates first.
 	clonedSegment := curSegInfo.Clone()
@@ -941,12 +941,12 @@ func (p *updateSegmentPack) Validate() error {
 		segmentInMeta := p.meta.segments.GetSegment(segment.ID)
 		if segmentInMeta.State == commonpb.SegmentState_Flushed && segment.State != commonpb.SegmentState_Dropped {
 			// if the segment is flushed, we should not update the segment meta, ignore the operation directly.
-			return errors.Wrapf(ErrIgnoredSegmentMetaOperation,
+			return merr.Wrapf(errIgnoredSegmentMetaOperation,
 				"segment is flushed, segmentID: %d",
 				segment.ID)
 		}
 		if segment.GetDmlPosition().GetTimestamp() < segmentInMeta.GetDmlPosition().GetTimestamp() {
-			return errors.Wrapf(ErrIgnoredSegmentMetaOperation,
+			return merr.Wrapf(errIgnoredSegmentMetaOperation,
 				"dml time tick is less than the segment meta, segmentID: %d, new incoming time tick: %d, existing time tick: %d",
 				segment.ID,
 				segment.GetDmlPosition().GetTimestamp(),
@@ -1223,7 +1223,7 @@ func updateManifestPathIfNewer(segment *SegmentInfo, manifestPath string) error
 		return err
 	}
 	if currentBase != incomingBase {
-		return merr.WrapErrServiceInternal(fmt.Sprintf("manifest base path mismatch for segment %d: current %s, incoming %s", segment.GetID(), currentBase, incomingBase))
+		return merr.WrapErrServiceInternalMsg("manifest base path mismatch for segment %d: current %s, incoming %s", segment.GetID(), currentBase, incomingBase)
 	}
 	if incomingVersion > currentVersion {
 		segment.ManifestPath = manifestPath
@@ -1824,6 +1824,14 @@ func (m *meta) UpdateSegmentsInfo(ctx context.Context, operators ...UpdateOperat
 
 	// Validate the update pack.
 	if err := updatePack.Validate(); err != nil {
+		// A stale save-binlog-paths update (segment already flushed, or an
+		// outdated time tick) is a benign no-op: skip the meta write and
+		// report success so the caller does not retry. The signal stays
+		// inside this package on purpose; see errIgnoredSegmentMetaOperation.
+		if errors.Is(err, errIgnoredSegmentMetaOperation) {
+			log.Ctx(ctx).Info("meta update: ignored stale segment meta operation", zap.Error(err))
+			return nil
+		}
 		return err
 	}
 	updatePack.prepareSegmentMetricUpdates()
@@ -2100,7 +2108,7 @@ func (m *meta) AddAllocation(segmentID UniqueID, allocation *Allocation) error {
 	if curSegInfo == nil {
 		// TODO: Error handling.
 		log.Ctx(m.ctx).Error("meta update: add allocation failed - segment not found", zap.Int64("segmentID", segmentID))
-		return errors.New("meta update: add allocation failed - segment not found")
+		return merr.WrapErrSegmentNotFound(segmentID, "meta update: add allocation failed")
 	}
 	// As we use global segment lastExpire to guarantee data correctness after restart
 	// there is no need to persist allocation to meta store, only update allocation in-memory meta.
@@ -2287,7 +2295,7 @@ func validateCompactionFallbackStartPosition(compactFromSegInfos []*SegmentInfo,
 	if maxCommitTs == 0 || fallbackStart.GetTimestamp() >= maxCommitTs {
 		return nil
 	}
-	return errors.Errorf(
+	return merr.WrapErrServiceInternalMsg(
 		"compaction fallback start position timestamp %d is earlier than max input commit timestamp %d",
 		fallbackStart.GetTimestamp(),
 		maxCommitTs)
@@ -2661,7 +2669,7 @@ func (m *meta) HasSegments(segIDs []UniqueID) (bool, error) {
 
 	for _, segID := range segIDs {
 		if _, ok := m.segments.segments[segID]; !ok {
-			return false, fmt.Errorf("segment is not exist with ID = %d", segID)
+			return false, merr.WrapErrServiceInternalMsg("segment is not exist with ID = %d", segID)
 		}
 	}
 	return true, nil
@@ -2747,7 +2755,7 @@ func (m *meta) collectionHasTextFields(collectionID int64) bool {
 // UpdateChannelCheckpoint updates and saves channel checkpoint.
 func (m *meta) UpdateChannelCheckpoint(ctx context.Context, vChannel string, pos *msgpb.MsgPosition) error {
 	if pos == nil || pos.GetMsgID() == nil {
-		return fmt.Errorf("channelCP is nil, vChannel=%s", vChannel)
+		return merr.WrapErrServiceInternalMsg("channelCP is nil, vChannel=%s", vChannel)
 	}
 
 	// Clamp checkpoint to not advance beyond min growing segment checkpoint.
@@ -3331,7 +3339,7 @@ func (m *meta) completeBumpSchemaVersionCompactionMutation(
 	if currentManifest != resultManifest {
 		manifestCompare, err := packed.CompareManifestPath(resultManifest, currentManifest)
 		if err != nil {
-			return nil, nil, merr.WrapErrIllegalCompactionPlan(fmt.Sprintf("schema bump compaction result manifest is not comparable with current manifest: %v", err))
+			return nil, nil, merr.WrapErrIllegalCompactionPlanMsg("schema bump compaction result manifest is not comparable with current manifest: %v", err)
 		}
 		if manifestCompare <= 0 {
 			return nil, nil, merr.WrapErrIllegalCompactionPlan("schema bump compaction result manifest is not newer than current manifest")
@@ -3392,7 +3400,7 @@ func (m *meta) completeBumpSchemaVersionReplacementMutation(
 ) ([]*SegmentInfo, *segMetricMutation, error) {
 	idRange := t.GetPreAllocatedSegmentIDs()
 	if idRange == nil || idRange.GetBegin()+1 != idRange.GetEnd() || resultSegment.GetSegmentID() != idRange.GetBegin() {
-		return nil, nil, merr.WrapErrIllegalCompactionPlan(fmt.Sprintf("schema bump replacement result segment ID %d does not match the pre-allocated segment ID", resultSegment.GetSegmentID()))
+		return nil, nil, merr.WrapErrIllegalCompactionPlanMsg("schema bump replacement result segment ID %d does not match the pre-allocated segment ID", resultSegment.GetSegmentID())
 	}
 
 	dropped := oldSegment.Clone()
@@ -3555,7 +3563,7 @@ func (m *meta) GetFileResources(ctx context.Context, resourceIDs ...int64) ([]*i
 		if resource, ok := m.resourceIDMap[id]; ok {
 			resources = append(resources, resource)
 		} else {
-			return nil, errors.Errorf("file resource %d not found", id)
+			return nil, merr.WrapErrServiceInternalMsg("file resource %d not found", id)
 		}
 	}
 	return resources, nil

internal/datacoord/meta_test.go

@@ -3498,7 +3498,7 @@ func TestUpdateSegmentsInfo(t *testing.T) {
 			UpdateStartPosition([]*datapb.SegmentStartPosition{{SegmentID: 1, StartPosition: &msgpb.MsgPosition{MsgID: []byte{1, 2, 3}}}}),
 			UpdateCheckPointOperator(1, []*datapb.CheckPoint{{SegmentID: 1, NumOfRows: 10, Position: &msgpb.MsgPosition{MsgID: []byte{1, 2, 3}, Timestamp: 99}}}, true),
 		)
-		assert.True(t, errors.Is(err, ErrIgnoredSegmentMetaOperation))
+		assert.NoError(t, err) // stale update is swallowed as a benign no-op; segment must stay unchanged below
 
 		err = meta.UpdateSegmentsInfo(
 			context.TODO(),
@@ -3536,7 +3536,7 @@ func TestUpdateSegmentsInfo(t *testing.T) {
 				[]*datapb.FieldBinlog{}),
 			UpdateCheckPointOperator(1, []*datapb.CheckPoint{{SegmentID: 1, NumOfRows: 12, Position: &msgpb.MsgPosition{MsgID: []byte{1, 2, 3}, Timestamp: 101}}}, true),
 		)
-		assert.True(t, errors.Is(err, ErrIgnoredSegmentMetaOperation))
+		assert.NoError(t, err) // stale update is swallowed as a benign no-op; segment must stay unchanged below
 
 		updated = meta.GetHealthySegment(context.TODO(), 1)
 		assert.Equal(t, updated.NumOfRows, int64(20))

internal/datacoord/meta_util.go

@@ -22,7 +22,16 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 )
 
-var ErrIgnoredSegmentMetaOperation = errors.New("ignored segment meta operation")
+// errIgnoredSegmentMetaOperation is an internal control-flow signal: the
+// requested segment meta update is stale (segment already flushed, or an
+// outdated time tick) and is skipped as a benign no-op. It never crosses the
+// package boundary — UpdateSegmentsInfo swallows it and returns nil.
+//
+// It MUST stay a plain errors.New identity sentinel. A merr-typed value would
+// make errors.Is match by error code (milvusError.Is compares codes only), so
+// any error sharing the code — e.g. a real etcd write failure wrapped as
+// ServiceInternal — would be misread as ignorable and acknowledged as success.
+var errIgnoredSegmentMetaOperation = errors.New("ignored segment meta operation")
 
 // reviseVChannelInfo will revise the datapb.VchannelInfo for upgrade compatibility from 2.0.2
 func reviseVChannelInfo(vChannel *datapb.VchannelInfo) {

internal/datacoord/metrics_info.go

@@ -18,10 +18,8 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"sync"
 
-	"github.com/cockroachdb/errors"
 	"github.com/tidwall/gjson"
 	"go.uber.org/zap"
 	"golang.org/x/sync/errgroup"
@@ -138,7 +136,7 @@ func (s *Server) getSegmentsJSON(ctx context.Context, req *milvuspb.GetMetricsRe
 		}
 		return string(bs), nil
 	}
-	return "", fmt.Errorf("invalid param value in=[%s], it should be dc or dn", in)
+	return "", merr.WrapErrParameterInvalidMsg("invalid param value in=[%s], it should be dc or dn", in)
 }
 
 func (s *Server) getDistJSON(ctx context.Context, req *milvuspb.GetMetricsRequest) string {
@@ -308,7 +306,7 @@ func (s *Server) getIndexNodeMetrics(ctx context.Context, req *milvuspb.GetMetri
 		},
 	}
 	if node == nil {
-		return infos, errors.New("IndexNode is nil")
+		return infos, merr.WrapErrServiceInternalMsg("index node is nil")
 	}
 
 	metrics, err := node.GetMetrics(ctx, req)

internal/datacoord/partition_stats_meta.go

@@ -2,7 +2,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"sync"
 
 	"go.uber.org/zap"
@@ -183,11 +182,11 @@ func (psm *partitionStatsMeta) innerSaveCurrentPartitionStatsVersion(collectionI
 
 	if _, ok := psm.partitionStatsInfos[vChannel]; !ok {
 		return merr.WrapErrClusteringCompactionMetaError("SaveCurrentPartitionStatsVersion",
-			fmt.Errorf("update current partition stats version failed, there is no partition info exists with collID: %d, partID: %d, vChannel: %s", collectionID, partitionID, vChannel))
+			merr.WrapErrServiceInternalMsg("update current partition stats version failed, there is no partition info exists with collID: %d, partID: %d, vChannel: %s", collectionID, partitionID, vChannel))
 	}
 	if _, ok := psm.partitionStatsInfos[vChannel][partitionID]; !ok {
 		return merr.WrapErrClusteringCompactionMetaError("SaveCurrentPartitionStatsVersion",
-			fmt.Errorf("update current partition stats version failed, there is no partition info exists with collID: %d, partID: %d, vChannel: %s", collectionID, partitionID, vChannel))
+			merr.WrapErrServiceInternalMsg("update current partition stats version failed, there is no partition info exists with collID: %d, partID: %d, vChannel: %s", collectionID, partitionID, vChannel))
 	}
 
 	if err := psm.catalog.SaveCurrentPartitionStatsVersion(psm.ctx, collectionID, partitionID, vChannel, currentPartitionStatsVersion); err != nil {

internal/datacoord/segment_allocation_policy.go

@@ -23,12 +23,12 @@ import (
 	"sort"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/commonpb"
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/tsoutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
@@ -38,7 +38,7 @@ type calUpperLimitPolicy func(schema *schemapb.CollectionSchema) (int, error)
 
 func calBySchemaPolicy(schema *schemapb.CollectionSchema) (int, error) {
 	if schema == nil {
-		return -1, errors.New("nil schema")
+		return -1, merr.WrapErrServiceInternalMsg("nil schema")
 	}
 	sizePerRecord, err := typeutil.EstimateSizePerRecord(schema)
 	if err != nil {
@@ -46,7 +46,7 @@ func calBySchemaPolicy(schema *schemapb.CollectionSchema) (int, error) {
 	}
 	// check zero value, preventing panicking
 	if sizePerRecord == 0 {
-		return -1, errors.New("zero size record schema found")
+		return -1, merr.WrapErrServiceInternalMsg("zero size record schema found")
 	}
 	threshold := Params.DataCoordCfg.SegmentMaxSize.GetAsFloat() * 1024 * 1024
 	return int(threshold / float64(sizePerRecord)), nil

internal/datacoord/segment_manager.go

@@ -23,7 +23,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/zap"
@@ -36,6 +35,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/util/lock"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/retry"
@@ -299,7 +299,7 @@ func (s *SegmentManager) genLastExpireTsForSegments() (Timestamp, error) {
 	}, retry.Attempts(Params.DataCoordCfg.AllocLatestExpireAttempt.GetAsUint()), retry.Sleep(200*time.Millisecond))
 	if allocateErr != nil {
 		log.Warn("cannot allocate latest lastExpire from rootCoord", zap.Error(allocateErr))
-		return 0, errors.New("global max expire ts is unavailable for segment manager")
+		return 0, merr.WrapErrServiceInternalMsg("global max expire ts is unavailable for segment manager")
 	}
 	return latestTs, nil
 }
@@ -466,7 +466,7 @@ func (s *SegmentManager) estimateMaxNumOfRows(collectionID UniqueID) (int, error
 	// it's ok to use meta.GetCollection here, since collection meta is set before using segmentManager
 	collMeta := s.meta.GetCollection(collectionID)
 	if collMeta == nil {
-		return -1, fmt.Errorf("failed to get collection %d", collectionID)
+		return -1, merr.WrapErrServiceInternalMsg("failed to get collection %d", collectionID)
 	}
 	return s.estimatePolicy(collMeta.Schema)
 }

internal/datacoord/server.go

@@ -26,7 +26,6 @@ import (
 	"time"
 
 	"github.com/blang/semver/v4"
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"github.com/tidwall/gjson"
 	"github.com/tikv/client-go/v2/txnkv"
@@ -462,7 +461,7 @@ func (s *Server) SetSession(session sessionutil.SessionInterface) error {
 	s.session = session
 	s.icSession = session
 	if s.session == nil {
-		return errors.New("session is nil, the etcd client connection may have failed")
+		return merr.WrapErrServiceNotReadyMsg("session is nil, the etcd client connection may have failed")
 	}
 	return nil
 }
@@ -618,7 +617,7 @@ func (s *Server) initKV() error {
 		s.kv = etcdkv.NewEtcdKV(s.etcdCli, s.metaRootPath,
 			etcdkv.WithRequestTimeout(paramtable.Get().EtcdCfg.RequestTimeout.GetAsDuration(time.Millisecond)))
 	default:
-		return retry.Unrecoverable(fmt.Errorf("not supported meta store: %s", metaType))
+		return retry.Unrecoverable(merr.WrapErrServiceInternalMsg("unsupported meta store: %s", metaType))
 	}
 	log.Info("data coordinator successfully connected to metadata store", zap.String("metaType", metaType))
 	return nil
@@ -1027,7 +1026,7 @@ func (s *Server) flushFlushingSegment(ctx context.Context, segmentID UniqueID) e
 				return ctx.Err()
 			}
 			// underlying etcd may return context canceled, so we need to return a error to retry.
-			return errors.New("flush segment complete failed")
+			return merr.WrapErrServiceInternalMsg("flush segment complete failed")
 		}
 		return nil
 	}, retry.AttemptAlways())
@@ -1117,7 +1116,7 @@ func (s *Server) CleanMeta() error {
 	err2 := s.watchClient.RemoveWithPrefix(s.ctx, "")
 	if err2 != nil {
 		if err != nil {
-			err = fmt.Errorf("failed to CleanMeta[metadata cleanup error: %w][watchdata cleanup error: %v]", err, err2)
+			err = merr.Wrapf(err, "failed to clean meta (watchdata cleanup error: %v)", err2)
 		} else {
 			err = err2
 		}

internal/datacoord/services.go

@@ -137,7 +137,7 @@ func (s *Server) flushCollection(ctx context.Context, collectionID UniqueID, flu
 		for _, channel := range coll.VChannelNames {
 			sealedSegmentIDs, err := s.segmentManager.SealAllSegments(ctx, channel, toFlushSegments)
 			if err != nil {
-				return nil, errors.Wrapf(err, "failed to flush collection %d", collectionID)
+				return nil, merr.Wrapf(err, "failed to flush collection %d", collectionID)
 			}
 			for _, sealedSegmentID := range sealedSegmentIDs {
 				sealedSegmentsIDDict[sealedSegmentID] = true
@@ -691,13 +691,12 @@ func (s *Server) SaveBinlogPaths(ctx context.Context, req *datapb.SaveBinlogPath
 		UpdateAsDroppedIfEmptyWhenFlushing(req.GetSegmentID()),
 	)
 
-	// Update segment info in memory and meta.
+	// Update segment info in memory and meta. Stale updates (segment already
+	// flushed / outdated time tick) are swallowed inside UpdateSegmentsInfo as
+	// benign no-ops, so any error here is a real failure.
 	if err := s.meta.UpdateSegmentsInfo(ctx, operators...); err != nil {
-		if !errors.Is(err, ErrIgnoredSegmentMetaOperation) {
+		log.Error("save binlog and checkpoints failed", zap.Error(err))
-			log.Error("save binlog and checkpoints failed", zap.Error(err))
+		return merr.Status(err), nil
-			return merr.Status(err), nil
-		}
-		log.Info("save binlog and checkpoints failed with ignorable error", zap.Error(err))
 	}
 
 	s.meta.SetLastWrittenTime(req.GetSegmentID())
@@ -1292,7 +1291,7 @@ func (s *Server) GetMetrics(ctx context.Context, req *milvuspb.GetMetricsRequest
 		msg := "failed to get metrics"
 		log.Warn(msg, zap.Error(err))
 		return &milvuspb.GetMetricsResponse{
-			Status: merr.Status(errors.Wrap(err, msg)),
+			Status: merr.Status(merr.Wrap(err, msg)),
 		}, nil
 	}
 
@@ -1627,7 +1626,7 @@ OUTER:
 						break OUTER
 					}
 				} else {
-					resp.Status = merr.Status(merr.WrapErrParameterInvalidMsg("FlushAllTss or FlushAllTs is required"))
+					resp.Status = merr.Status(merr.WrapErrParameterMissingMsg("FlushAllTss or FlushAllTs is required"))
 					return resp, nil
 				}
 			}
@@ -1925,12 +1924,12 @@ func (s *Server) ImportV2(ctx context.Context, in *internalpb.ImportRequestInter
 	jobID := in.GetJobID()
 	if jobID == 0 {
 		if s.allocator == nil {
-			resp.Status = merr.Status(merr.WrapErrImportFailed("allocator not initialized"))
+			resp.Status = merr.Status(merr.WrapErrServiceUnavailable("allocator not initialized"))
 			return resp, nil
 		}
 		jobID, _, err = s.allocator.AllocN(1)
 		if err != nil {
-			resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("failed to allocate job ID: %v", err)))
+			resp.Status = merr.Status(merr.Wrap(err, "failed to allocate job ID"))
 			return resp, nil
 		}
 	}
@@ -1950,7 +1949,7 @@ func (s *Server) ImportV2(ctx context.Context, in *internalpb.ImportRequestInter
 	)
 	if err != nil {
 		log.Warn("failed to broadcast import message", zap.Error(err))
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("failed to broadcast import: %v", err)))
+		resp.Status = merr.Status(merr.Wrap(err, "failed to broadcast import"))
 		return resp, nil
 	}
 
@@ -2000,7 +1999,7 @@ func (s *Server) createImportJobFromAck(ctx context.Context, in *internalpb.Impo
 	// Allocate file ids.
 	idStart, _, err := s.allocator.AllocN(int64(len(files)) + 1)
 	if err != nil {
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("alloc id failed: %v", err)))
+		resp.Status = merr.Status(merr.Wrap(err, "alloc id failed"))
 		return resp, nil
 	}
 	files = lo.Map(files, func(importFile *internalpb.ImportFile, i int) *internalpb.ImportFile {
@@ -2013,7 +2012,7 @@ func (s *Server) createImportJobFromAck(ctx context.Context, in *internalpb.Impo
 		return resp, nil
 	}
 	if err != nil {
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("get collection failed: %v", err)))
+		resp.Status = merr.Status(merr.Wrap(err, "get collection failed"))
 		return resp, nil
 	}
 	if importCollectionInfo == nil {
@@ -2048,7 +2047,7 @@ func (s *Server) createImportJobFromAck(ctx context.Context, in *internalpb.Impo
 	}
 	err = s.importMeta.AddJob(ctx, job)
 	if err != nil {
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("add import job failed: %v", err)))
+		resp.Status = merr.Status(merr.Wrap(err, "add import job failed"))
 		return resp, nil
 	}
 
@@ -2075,13 +2074,13 @@ func (s *Server) GetImportProgress(ctx context.Context, in *internalpb.GetImport
 	}
 	jobID, err := strconv.ParseInt(in.GetJobID(), 10, 64)
 	if err != nil {
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("parse job id failed: %v", err)))
+		resp.Status = merr.Status(merr.WrapErrParameterInvalidMsg("parse job id failed: %v", err))
 		return resp, nil
 	}
 
 	job := s.importMeta.GetJob(ctx, jobID)
 	if job == nil {
-		resp.Status = merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("import job does not exist, jobID=%d", jobID)))
+		resp.Status = merr.Status(merr.WrapErrImportSysFailedMsg("import job does not exist, jobID=%d", jobID))
 		return resp, nil
 	}
 	progress, state, importedRows, totalRows, reason := GetJobProgress(ctx, jobID, s.importMeta, s.meta)
@@ -2481,14 +2480,14 @@ func (s *Server) RestoreSnapshot(ctx context.Context, req *datapb.RestoreSnapsho
 
 	// Validate parameters
 	if req.GetName() == "" {
-		err := merr.WrapErrParameterInvalidMsg("snapshot name is required")
+		err := merr.WrapErrParameterMissingMsg("snapshot name is required")
 		log.Warn("invalid request", zap.Error(err))
 		return &datapb.RestoreSnapshotResponse{
 			Status: merr.Status(err),
 		}, nil
 	}
 	if req.GetTargetCollectionName() == "" {
-		err := merr.WrapErrParameterInvalidMsg("target collection name is required")
+		err := merr.WrapErrParameterMissingMsg("target collection name is required")
 		log.Warn("invalid request", zap.Error(err))
 		return &datapb.RestoreSnapshotResponse{
 			Status: merr.Status(err),
@@ -2887,7 +2886,7 @@ func (s *Server) ListRefreshExternalCollectionJobs(ctx context.Context, req *dat
 func (s *Server) broadcastCommitImportMessage(ctx context.Context, job ImportJob) error {
 	vchannels := job.GetVchannels()
 	if len(vchannels) == 0 {
-		return merr.WrapErrImportFailed(fmt.Sprintf("job %d has no vchannels", job.GetJobID()))
+		return merr.WrapErrImportSysFailedMsg("job %d has no vchannels", job.GetJobID())
 	}
 
 	broadcaster, err := s.startBroadcastWithCollectionID(ctx, job.GetCollectionID())
@@ -2914,7 +2913,7 @@ func (s *Server) broadcastCommitImportMessage(ctx context.Context, job ImportJob
 func (s *Server) broadcastRollbackImportMessage(ctx context.Context, job ImportJob) error {
 	vchannels := job.GetVchannels()
 	if len(vchannels) == 0 {
-		return merr.WrapErrImportFailed(fmt.Sprintf("job %d has no vchannels", job.GetJobID()))
+		return merr.WrapErrImportSysFailedMsg("job %d has no vchannels", job.GetJobID())
 	}
 
 	broadcaster, err := s.startBroadcastWithCollectionID(ctx, job.GetCollectionID())
@@ -2949,7 +2948,7 @@ func (s *Server) validateAndExecuteImportAction(
 	}
 	job := s.importMeta.GetJob(ctx, jobID)
 	if job == nil {
-		return merr.Status(merr.WrapErrImportFailed(fmt.Sprintf("job %d not found", jobID))), nil
+		return merr.Status(merr.WrapErrImportSysFailedMsg("job %d not found", jobID)), nil
 	}
 	if st := validateState(job); st != nil {
 		return st, nil

internal/datacoord/services_commit_backfill.go

@@ -186,7 +186,7 @@ const maxBackfillResultBytes int64 = 64 * 1024 * 1024 // 64MiB
 // reject s3a://<other-bucket>/... paths early.
 func (s *Server) loadBackfillResult(ctx context.Context, rawPath string) (*BackfillResult, error) {
 	if rawPath == "" {
-		return nil, merr.WrapErrParameterInvalidMsg("result_path is required")
+		return nil, merr.WrapErrParameterMissingMsg("result_path is required")
 	}
 	bucket := bucketFromChunkManager(s.meta.chunkManager)
 	key, err := normalizeObjectKey(rawPath, bucket)

internal/datacoord/services_test.go

@@ -1645,7 +1645,6 @@ func TestGetRecoveryInfoV2(t *testing.T) {
 
 func TestImportV2(t *testing.T) {
 	ctx := context.Background()
-	mockErr := errors.New("mock err")
 
 	t.Run("ImportV2", func(t *testing.T) {
 		// server not healthy
@@ -1682,11 +1681,11 @@ func TestImportV2(t *testing.T) {
 		s.importMeta, err = NewImportMeta(context.TODO(), catalog, nil, nil)
 		assert.NoError(t, err)
 		alloc := allocator.NewMockAllocator(t)
-		alloc.EXPECT().AllocN(mock.Anything).Return(0, 0, mockErr)
+		alloc.EXPECT().AllocN(mock.Anything).Return(0, 0, merr.WrapErrServiceUnavailable("mock err"))
 		s.allocator = alloc
 		resp, err = s.ImportV2(ctx, &internalpb.ImportRequestInternal{})
 		assert.NoError(t, err)
-		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrServiceUnavailable))
 	})
 
 	t.Run("GetImportProgress", func(t *testing.T) {
@@ -1703,7 +1702,7 @@ func TestImportV2(t *testing.T) {
 			JobID: "@%$%$#%",
 		})
 		assert.NoError(t, err)
-		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrParameterInvalid))
 
 		// job does not exist
 		catalog := mocks.NewDataCoordCatalog(t)
@@ -1723,7 +1722,8 @@ func TestImportV2(t *testing.T) {
 			JobID: "-1",
 		})
 		assert.NoError(t, err)
-		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportFailed))
+		// job-not-found is a server-side orchestration issue, not malformed user data
+		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrImportSysFailed))
 
 		// normal case
 		var job ImportJob = &importJob{
@@ -3156,7 +3156,7 @@ func TestServer_RestoreSnapshot(t *testing.T) {
 
 		assert.NoError(t, err)
 		assert.Error(t, merr.Error(resp.GetStatus()))
-		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrParameterInvalid))
+		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrParameterMissing))
 	})
 
 	t.Run("missing_collection_name", func(t *testing.T) {
@@ -3173,7 +3173,7 @@ func TestServer_RestoreSnapshot(t *testing.T) {
 
 		assert.NoError(t, err)
 		assert.Error(t, merr.Error(resp.GetStatus()))
-		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrParameterInvalid))
+		assert.True(t, errors.Is(merr.Error(resp.GetStatus()), merr.ErrParameterMissing))
 	})
 
 	t.Run("snapshot_not_found", func(t *testing.T) {

internal/datacoord/session/session.go

@@ -18,12 +18,12 @@ package session
 
 import (
 	"context"
-	"fmt"
 
 	"github.com/cockroachdb/errors"
 
 	"github.com/milvus-io/milvus/internal/types"
 	"github.com/milvus-io/milvus/pkg/v3/util/lock"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 var errDisposed = errors.New("client is disposed")
@@ -84,7 +84,7 @@ func (n *Session) GetOrCreateClient(ctx context.Context) (types.DataNodeClient,
 	}
 
 	if n.clientCreator == nil {
-		return nil, fmt.Errorf("unable to create client for %s because of a nil client creator", n.info.Address)
+		return nil, merr.WrapErrServiceInternalMsg("unable to create client for %s because of a nil client creator", n.info.Address)
 	}
 
 	err := n.initClient(ctx)

internal/datacoord/snapshot.go

@@ -33,6 +33,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // S3 snapshot storage path constants.
@@ -136,7 +137,7 @@ func getManifestSchemaByVersion(version int) (avro.Schema, error) {
 	case 3:
 		return getManifestSchemaV3()
 	default:
-		return nil, fmt.Errorf("unsupported manifest schema version: %d", version)
+		return nil, merr.WrapErrServiceInternalMsg("unsupported manifest schema version: %d", version)
 	}
 }
 
@@ -508,22 +509,22 @@ func GetSegmentManifestPath(manifestDir string, segmentID int64) string {
 func (w *SnapshotWriter) Save(ctx context.Context, snapshot *SnapshotData) (string, error) {
 	// Validate input parameters
 	if snapshot == nil {
-		return "", fmt.Errorf("snapshot cannot be nil")
+		return "", merr.WrapErrServiceInternalMsg("snapshot cannot be nil")
 	}
 	if snapshot.SnapshotInfo == nil {
-		return "", fmt.Errorf("snapshot info cannot be nil")
+		return "", merr.WrapErrServiceInternalMsg("snapshot info cannot be nil")
 	}
 	collectionID := snapshot.SnapshotInfo.GetCollectionId()
 	if collectionID <= 0 {
-		return "", fmt.Errorf("invalid collection ID: %d", collectionID)
+		return "", merr.WrapErrServiceInternalMsg("invalid collection ID: %d", collectionID)
 	}
 	if snapshot.Collection == nil {
-		return "", fmt.Errorf("collection description cannot be nil")
+		return "", merr.WrapErrServiceInternalMsg("collection description cannot be nil")
 	}
 
 	snapshotID := snapshot.SnapshotInfo.GetId()
 	if snapshotID <= 0 {
-		return "", fmt.Errorf("invalid snapshot ID: %d", snapshotID)
+		return "", merr.WrapErrServiceInternalMsg("invalid snapshot ID: %d", snapshotID)
 	}
 	manifestDir, metadataPath := GetSnapshotPaths(w.chunkManager.RootPath(), collectionID, snapshotID)
 
@@ -533,7 +534,7 @@ func (w *SnapshotWriter) Save(ctx context.Context, snapshot *SnapshotData) (stri
 	for _, segment := range snapshot.Segments {
 		manifestPath := GetSegmentManifestPath(manifestDir, segment.GetSegmentId())
 		if err := w.writeSegmentManifest(ctx, manifestPath, segment); err != nil {
-			return "", fmt.Errorf("failed to write manifest for segment %d: %w", segment.GetSegmentId(), err)
+			return "", merr.Wrapf(err, "failed to write manifest for segment %d", segment.GetSegmentId())
 		}
 		manifestPaths = append(manifestPaths, manifestPath)
 	}
@@ -557,7 +558,7 @@ func (w *SnapshotWriter) Save(ctx context.Context, snapshot *SnapshotData) (stri
 	// Step 3: Write metadata file with all manifest paths
 	// This file is the entry point for reading the snapshot
 	if err := w.writeMetadataFile(ctx, metadataPath, snapshot, manifestPaths, storagev2Manifests); err != nil {
-		return "", fmt.Errorf("failed to write metadata file: %w", err)
+		return "", merr.Wrap(err, "failed to write metadata file")
 	}
 
 	log.Info("Successfully wrote metadata file",
@@ -584,12 +585,12 @@ func (w *SnapshotWriter) writeSegmentManifest(ctx context.Context, manifestPath
 	// Serialize to Avro binary format (single record per file)
 	avroSchema, err := getManifestSchema()
 	if err != nil {
-		return fmt.Errorf("failed to get manifest schema: %w", err)
+		return merr.WrapErrServiceInternalErr(err, "failed to get manifest schema")
 	}
 
 	binaryData, err := avro.Marshal(avroSchema, entry)
 	if err != nil {
-		return fmt.Errorf("failed to serialize entry to avro: %w", err)
+		return merr.WrapErrServiceInternalErr(err, "failed to serialize entry to avro")
 	}
 
 	// Write to object storage
@@ -634,7 +635,7 @@ func (w *SnapshotWriter) writeMetadataFile(ctx context.Context, metadataPath str
 	}
 	jsonData, err := opts.Marshal(metadata)
 	if err != nil {
-		return fmt.Errorf("failed to marshal metadata to JSON: %w", err)
+		return merr.WrapErrServiceInternalErr(err, "failed to marshal metadata to JSON")
 	}
 
 	// Write to object storage
@@ -663,13 +664,13 @@ func (w *SnapshotWriter) writeMetadataFile(ctx context.Context, metadataPath str
 func (w *SnapshotWriter) Drop(ctx context.Context, metadataFilePath string) error {
 	// Validate input parameter
 	if metadataFilePath == "" {
-		return fmt.Errorf("metadata file path cannot be empty")
+		return merr.WrapErrServiceInternalMsg("metadata file path cannot be empty")
 	}
 
 	// Step 1: Read metadata file to discover manifest file paths
 	metadata, err := w.readMetadataFile(ctx, metadataFilePath)
 	if err != nil {
-		return fmt.Errorf("failed to read metadata file: %w", err)
+		return merr.Wrap(err, "failed to read metadata file")
 	}
 
 	snapshotID := metadata.GetSnapshotInfo().GetId()
@@ -678,7 +679,7 @@ func (w *SnapshotWriter) Drop(ctx context.Context, metadataFilePath string) erro
 	manifestList := metadata.GetManifestList()
 	if len(manifestList) > 0 {
 		if err := w.chunkManager.MultiRemove(ctx, manifestList); err != nil {
-			return fmt.Errorf("failed to remove manifest files: %w", err)
+			return merr.Wrap(err, "failed to remove manifest files")
 		}
 		log.Info("Successfully removed manifest files",
 			zap.Int("count", len(manifestList)),
@@ -687,7 +688,7 @@ func (w *SnapshotWriter) Drop(ctx context.Context, metadataFilePath string) erro
 
 	// Step 3: Remove the metadata file (entry point)
 	if err := w.chunkManager.Remove(ctx, metadataFilePath); err != nil {
-		return fmt.Errorf("failed to remove metadata file: %w", err)
+		return merr.Wrap(err, "failed to remove metadata file")
 	}
 	log.Info("Successfully removed metadata file",
 		zap.String("metadataFilePath", metadataFilePath))
@@ -703,7 +704,7 @@ func (w *SnapshotWriter) readMetadataFile(ctx context.Context, filePath string)
 	// Read raw JSON content from object storage
 	data, err := w.chunkManager.Read(ctx, filePath)
 	if err != nil {
-		return nil, fmt.Errorf("failed to read metadata file: %w", err)
+		return nil, merr.Wrap(err, "failed to read metadata file")
 	}
 
 	// Use protojson for deserialization to correctly handle protobuf oneof fields
@@ -712,7 +713,7 @@ func (w *SnapshotWriter) readMetadataFile(ctx context.Context, filePath string)
 		DiscardUnknown: true,
 	}
 	if err := opts.Unmarshal(data, metadata); err != nil {
-		return nil, fmt.Errorf("failed to parse metadata JSON: %w", err)
+		return nil, merr.WrapErrServiceInternalErr(err, "failed to parse metadata JSON")
 	}
 
 	return metadata, nil
@@ -775,13 +776,13 @@ func NewSnapshotReader(cm storage.ChunkManager) *SnapshotReader {
 func (r *SnapshotReader) ReadSnapshot(ctx context.Context, metadataFilePath string, includeSegments bool) (*SnapshotData, error) {
 	// Validate input
 	if metadataFilePath == "" {
-		return nil, fmt.Errorf("metadata file path cannot be empty")
+		return nil, merr.WrapErrServiceInternalMsg("metadata file path cannot be empty")
 	}
 
 	// Step 1: Read metadata file (always required)
 	metadata, err := r.readMetadataFile(ctx, metadataFilePath)
 	if err != nil {
-		return nil, fmt.Errorf("failed to read metadata file: %w", err)
+		return nil, merr.Wrap(err, "failed to read metadata file")
 	}
 
 	// Step 2: Optionally read segment details from manifest files
@@ -791,7 +792,7 @@ func (r *SnapshotReader) ReadSnapshot(ctx context.Context, metadataFilePath stri
 		for _, manifestPath := range metadata.GetManifestList() {
 			segments, err := r.readManifestFile(ctx, manifestPath, int(metadata.GetFormatVersion()))
 			if err != nil {
-				return nil, fmt.Errorf("failed to read manifest file %s: %w", manifestPath, err)
+				return nil, merr.Wrapf(err, "failed to read manifest file %s", manifestPath)
 			}
 			allSegments = append(allSegments, segments...)
 		}
@@ -830,7 +831,7 @@ func (r *SnapshotReader) readMetadataFile(ctx context.Context, filePath string)
 	// Read raw JSON content from object storage
 	data, err := r.chunkManager.Read(ctx, filePath)
 	if err != nil {
-		return nil, fmt.Errorf("failed to read metadata file: %w", err)
+		return nil, merr.Wrap(err, "failed to read metadata file")
 	}
 
 	// Use protojson for deserialization to correctly handle protobuf oneof fields
@@ -839,12 +840,12 @@ func (r *SnapshotReader) readMetadataFile(ctx context.Context, filePath string)
 		DiscardUnknown: true,
 	}
 	if err := opts.Unmarshal(data, metadata); err != nil {
-		return nil, fmt.Errorf("failed to parse metadata JSON: %w", err)
+		return nil, merr.WrapErrServiceInternalErr(err, "failed to parse metadata JSON")
 	}
 
 	// Validate format version compatibility
 	if err := validateFormatVersion(int(metadata.GetFormatVersion())); err != nil {
-		return nil, fmt.Errorf("incompatible snapshot format: %w", err)
+		return nil, merr.WrapErrServiceInternalErr(err, "incompatible snapshot format")
 	}
 
 	return metadata, nil
@@ -865,7 +866,7 @@ func validateFormatVersion(version int) error {
 
 	// Check if version is too new for current code
 	if version > SnapshotFormatVersion {
-		return fmt.Errorf("snapshot format version %d is too new, current supported version: %d (please upgrade Milvus)",
+		return merr.WrapErrServiceInternalMsg("snapshot format version %d is too new, current supported version: %d (please upgrade Milvus)",
 			version, SnapshotFormatVersion)
 	}
 
@@ -888,20 +889,20 @@ func (r *SnapshotReader) readManifestFile(ctx context.Context, filePath string,
 	// Read raw Avro content from object storage
 	data, err := r.chunkManager.Read(ctx, filePath)
 	if err != nil {
-		return nil, fmt.Errorf("failed to read manifest file: %w", err)
+		return nil, merr.Wrap(err, "failed to read manifest file")
 	}
 
 	// Get Avro schema for the specified format version
 	avroSchema, err := getManifestSchemaByVersion(formatVersion)
 	if err != nil {
-		return nil, fmt.Errorf("failed to get manifest schema for version %d: %w", formatVersion, err)
+		return nil, merr.WrapErrServiceInternalErr(err, "failed to get manifest schema for version %d", formatVersion)
 	}
 
 	// Deserialize Avro binary data to single ManifestEntry record
 	var record ManifestEntry
 	err = avro.Unmarshal(avroSchema, data, &record)
 	if err != nil {
-		return nil, fmt.Errorf("failed to parse avro data: %w", err)
+		return nil, merr.WrapErrServiceInternalErr(err, "failed to parse avro data")
 	}
 
 	// Convert ManifestEntry record to protobuf SegmentDescription
@@ -974,7 +975,7 @@ func (r *SnapshotReader) readManifestFile(ctx context.Context, filePath string,
 func (r *SnapshotReader) ListSnapshots(ctx context.Context, collectionID int64) ([]*datapb.SnapshotInfo, error) {
 	// Validate input parameters
 	if collectionID <= 0 {
-		return nil, fmt.Errorf("invalid collection ID: %d", collectionID)
+		return nil, merr.WrapErrServiceInternalMsg("invalid collection ID: %d", collectionID)
 	}
 
 	// Construct metadata directory path
@@ -984,7 +985,7 @@ func (r *SnapshotReader) ListSnapshots(ctx context.Context, collectionID int64)
 	// List all files in the metadata directory
 	files, _, err := storage.ListAllChunkWithPrefix(ctx, r.chunkManager, metadataDir, false)
 	if err != nil {
-		return nil, fmt.Errorf("failed to list metadata files: %w", err)
+		return nil, merr.Wrap(storage.ToMilvusIoError(metadataDir, err), "failed to list metadata files")
 	}
 
 	// Read each metadata file and extract SnapshotInfo

internal/datacoord/snapshot_manager.go

@@ -562,7 +562,7 @@ func (sm *snapshotManager) validateCMEKCompatibility(
 	// Get target database properties first (needed for both encrypted and non-encrypted snapshots)
 	dbResp, err := sm.broker.DescribeDatabase(ctx, targetDbName)
 	if err != nil {
-		return fmt.Errorf("failed to describe target database %s: %w", targetDbName, err)
+		return merr.Wrapf(err, "failed to describe target database %s", targetDbName)
 	}
 	targetIsEncrypted := hookutil.IsDBEncrypted(dbResp.GetProperties())
 
@@ -640,7 +640,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 	// ========================================================================
 	phase0Lock, err := startRestoreLock(ctx, sourceCollectionID, snapshotName, targetDbName, targetCollectionName)
 	if err != nil {
-		return 0, fmt.Errorf("failed to acquire restore lock: %w", err)
+		return 0, merr.Wrap(err, "failed to acquire restore lock")
 	}
 
 	// Pin the source snapshot while holding the phase-0 lock. The pin is the
@@ -662,7 +662,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 	pinID, activePins, err := sm.snapshotMeta.PinSnapshot(ctx, sourceCollectionID, snapshotName, pinTTLSeconds)
 	if err != nil {
 		phase0Lock.Close()
-		return 0, fmt.Errorf("failed to pin source snapshot under restore lock: %w", err)
+		return 0, merr.Wrap(err, "failed to pin source snapshot under restore lock")
 	}
 	setSnapshotActivePinsGauge(sourceCollectionID, snapshotName, activePins)
 	phase0Lock.Close()
@@ -691,7 +691,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 	// Phase 1: Read snapshot data (now protected by the refcount guard)
 	snapshotData, err := sm.ReadSnapshotData(ctx, sourceCollectionID, snapshotName)
 	if err != nil {
-		return 0, fmt.Errorf("failed to read snapshot data: %w", err)
+		return 0, merr.Wrap(err, "failed to read snapshot data")
 	}
 	log.Info("snapshot data loaded",
 		zap.Int("segmentCount", len(snapshotData.Segments)),
@@ -707,7 +707,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 	// Phase 2: Restore collection and partitions
 	collectionID, err := sm.RestoreCollection(ctx, snapshotData, targetCollectionName, targetDbName)
 	if err != nil {
-		return 0, fmt.Errorf("failed to restore collection: %w", err)
+		return 0, merr.Wrap(err, "failed to restore collection")
 	}
 	log.Info("collection and partitions restored", zap.Int64("collectionID", collectionID))
 
@@ -718,7 +718,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 		if rollbackErr := rollback(ctx, targetDbName, targetCollectionName); rollbackErr != nil {
 			log.Error("rollback failed", zap.Error(rollbackErr))
 		}
-		return 0, fmt.Errorf("failed to restore indexes: %w", err)
+		return 0, merr.Wrap(err, "failed to restore indexes")
 	}
 	log.Info("indexes restored", zap.Int("indexCount", len(snapshotData.Indexes)))
 
@@ -730,7 +730,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 		if rollbackErr := rollback(ctx, targetDbName, targetCollectionName); rollbackErr != nil {
 			log.Error("rollback failed", zap.Error(rollbackErr))
 		}
-		return 0, fmt.Errorf("failed to allocate job ID: %w", err)
+		return 0, merr.Wrap(err, "failed to allocate job ID")
 	}
 	log.Info("pre-allocated job ID for restore", zap.Int64("jobID", jobID))
 
@@ -741,7 +741,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 		if rollbackErr := rollback(ctx, targetDbName, targetCollectionName); rollbackErr != nil {
 			log.Error("rollback failed", zap.Error(rollbackErr))
 		}
-		return 0, fmt.Errorf("failed to start broadcaster for restore message: %w", err)
+		return 0, merr.Wrap(err, "failed to start broadcaster for restore message")
 	}
 	defer func() {
 		if restoreBroadcaster != nil {
@@ -760,7 +760,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 		if rollbackErr := rollback(ctx, targetDbName, targetCollectionName); rollbackErr != nil {
 			log.Error("rollback failed", zap.Error(rollbackErr))
 		}
-		err = fmt.Errorf("resource validation failed: %w", valErr)
+		err = merr.Wrap(valErr, "resource validation failed")
 		return 0, err
 	}
 
@@ -785,7 +785,7 @@ func (sm *snapshotManager) RestoreSnapshot(
 		if rollbackErr := rollback(ctx, targetDbName, targetCollectionName); rollbackErr != nil {
 			log.Error("rollback failed", zap.Error(rollbackErr))
 		}
-		err = fmt.Errorf("failed to broadcast restore message: %w", bcErr)
+		err = merr.Wrap(bcErr, "failed to broadcast restore message")
 		return 0, err
 	}
 
@@ -872,14 +872,14 @@ func (sm *snapshotManager) RestoreIndexes(
 	// Get collection info for dbId
 	coll, err := sm.broker.DescribeCollectionInternal(ctx, collectionID)
 	if err != nil {
-		return fmt.Errorf("failed to describe collection %d: %w", collectionID, err)
+		return merr.Wrapf(err, "failed to describe collection %d", collectionID)
 	}
 
 	for _, indexInfo := range snapshotData.Indexes {
 		// Allocate new index ID
 		indexID, err := sm.allocator.AllocID(ctx)
 		if err != nil {
-			return fmt.Errorf("failed to allocate index ID: %w", err)
+			return merr.Wrap(err, "failed to allocate index ID")
 		}
 
 		// Build index model from snapshot data
@@ -898,19 +898,19 @@ func (sm *snapshotManager) RestoreIndexes(
 
 		// Validate the index params (basic validation without JSON path parsing)
 		if err := ValidateIndexParams(index); err != nil {
-			return fmt.Errorf("failed to validate index %s: %w", indexInfo.GetIndexName(), err)
+			return merr.Wrapf(err, "failed to validate index %s", indexInfo.GetIndexName())
 		}
 
 		// Check scalar index engine version for JSON path indexes with new types
 		if err := sm.checkJSONPathIndexVersion(index); err != nil {
-			return fmt.Errorf("failed to validate index %s: %w", indexInfo.GetIndexName(), err)
+			return merr.Wrapf(err, "failed to validate index %s", indexInfo.GetIndexName())
 		}
 
 		// Create a new broadcaster for each index
 		// (each broadcaster can only be used once due to resource key lock consumption)
 		b, err := startBroadcaster(ctx, collectionID, snapshotName)
 		if err != nil {
-			return fmt.Errorf("failed to start broadcaster for index %s: %w", indexInfo.GetIndexName(), err)
+			return merr.Wrapf(err, "failed to start broadcaster for index %s", indexInfo.GetIndexName())
 		}
 
 		// Broadcast CreateIndex message directly to DDL WAL
@@ -930,7 +930,7 @@ func (sm *snapshotManager) RestoreIndexes(
 		)
 		b.Close()
 		if err != nil {
-			return fmt.Errorf("failed to broadcast create index %s: %w", indexInfo.GetIndexName(), err)
+			return merr.Wrapf(err, "failed to broadcast create index %s", indexInfo.GetIndexName())
 		}
 
 		log.Ctx(ctx).Info("index restored via DDL WAL broadcast",
@@ -977,14 +977,14 @@ func (sm *snapshotManager) RestoreData(
 	snapshotData, err := sm.ReadSnapshotData(ctx, sourceCollectionID, snapshotName)
 	if err != nil {
 		log.Error("failed to read snapshot data", zap.Error(err))
-		return 0, fmt.Errorf("failed to read snapshot data: %w", err)
+		return 0, merr.Wrap(err, "failed to read snapshot data")
 	}
 
 	// ========== Phase 2: Build partition mapping ==========
 	partitionMapping, err := sm.buildPartitionMapping(ctx, snapshotData, collectionID)
 	if err != nil {
 		log.Error("failed to build partition mapping", zap.Error(err))
-		return 0, fmt.Errorf("partition mapping failed: %w", err)
+		return 0, merr.Wrap(err, "partition mapping failed")
 	}
 	log.Info("partition mapping built", zap.Any("partitionMapping", partitionMapping))
 
@@ -992,14 +992,14 @@ func (sm *snapshotManager) RestoreData(
 	channelMapping, err := sm.buildChannelMapping(ctx, snapshotData, collectionID)
 	if err != nil {
 		log.Error("failed to build channel mapping", zap.Error(err))
-		return 0, fmt.Errorf("channel mapping failed: %w", err)
+		return 0, merr.Wrap(err, "channel mapping failed")
 	}
 
 	// ========== Phase 4: Create copy segment job ==========
 	// Use the pre-allocated jobID from the WAL message
 	if err := sm.createRestoreJob(ctx, collectionID, channelMapping, partitionMapping, snapshotData, jobID, pinID); err != nil {
 		log.Error("failed to create restore job", zap.Error(err))
-		return 0, fmt.Errorf("restore job creation failed: %w", err)
+		return 0, merr.Wrap(err, "restore job creation failed")
 	}
 
 	log.Info("restore data completed successfully",
@@ -1047,7 +1047,7 @@ func (sm *snapshotManager) restoreUserPartitions(
 		}
 
 		if err := sm.broker.CreatePartition(ctx, req); err != nil {
-			return fmt.Errorf("failed to create partition %s: %w", partitionName, err)
+			return merr.Wrapf(err, "failed to create partition %s", partitionName)
 		}
 	}
 
@@ -1330,7 +1330,7 @@ func (sm *snapshotManager) GetRestoreState(ctx context.Context, jobID int64) (*d
 	// Get job
 	job := sm.copySegmentMeta.GetJob(ctx, jobID)
 	if job == nil {
-		err := merr.WrapErrImportFailed(fmt.Sprintf("restore job not found: jobID=%d", jobID))
+		err := merr.WrapErrImportSysFailedMsg("restore job not found: jobID=%d", jobID)
 		log.Warn("restore job not found")
 		return nil, err
 	}

internal/datacoord/snapshot_meta.go

@@ -572,7 +572,7 @@ func (sm *snapshotMeta) SaveSnapshot(ctx context.Context, snapshot *SnapshotData
 
 	if err := sm.catalog.SaveSnapshot(ctx, snapshot.SnapshotInfo); err != nil {
 		log.Error("failed to save pending snapshot to catalog", zap.Error(err))
-		return fmt.Errorf("failed to save pending snapshot to catalog: %w", err)
+		return merr.Wrap(err, "failed to save pending snapshot to catalog")
 	}
 	log.Info("saved pending snapshot to catalog")
 
@@ -582,7 +582,7 @@ func (sm *snapshotMeta) SaveSnapshot(ctx context.Context, snapshot *SnapshotData
 		// S3 write failed, leave PENDING record for GC to clean up
 		log.Error("failed to save snapshot to S3, pending record left for GC cleanup",
 			zap.Error(err))
-		return fmt.Errorf("failed to save snapshot to S3: %w", err)
+		return merr.Wrap(err, "failed to save snapshot to S3")
 	}
 	snapshot.SnapshotInfo.S3Location = metadataFilePath
 	log.Info("saved snapshot data to S3", zap.String("s3Location", metadataFilePath))
@@ -598,7 +598,7 @@ func (sm *snapshotMeta) SaveSnapshot(ctx context.Context, snapshot *SnapshotData
 		// GC will eventually cleanup via the PENDING marker.
 		log.Error("failed to update snapshot to committed state, pending record left for GC cleanup",
 			zap.Error(err))
-		return fmt.Errorf("failed to update snapshot to committed state: %w", err)
+		return merr.Wrap(err, "failed to update snapshot to committed state")
 	}
 
 	// Catalog committed — safe to insert into in-memory cache and register protection.
@@ -809,8 +809,7 @@ func (sm *snapshotMeta) DropSnapshotsByCollection(ctx context.Context, collectio
 	}
 
 	if len(errs) > 0 {
-		return dropped, fmt.Errorf("failed to drop %d/%d snapshots for collection %d: %w",
+		return dropped, merr.WrapErrServiceInternalErr(merr.Combine(errs...), "failed to drop %d/%d snapshots for collection %d", len(errs), len(snapshotNames), collectionID)
-			len(errs), len(snapshotNames), collectionID, merr.Combine(errs...))
 	}
 
 	return dropped, nil
@@ -1006,7 +1005,7 @@ func (sm *snapshotMeta) GetPendingSnapshots(ctx context.Context, pendingTimeout
 	// Get all snapshots from catalog (etcd)
 	snapshots, err := sm.catalog.ListSnapshots(ctx)
 	if err != nil {
-		return nil, fmt.Errorf("failed to list snapshots from catalog: %w", err)
+		return nil, merr.Wrap(err, "failed to list snapshots from catalog")
 	}
 
 	now := time.Now().UnixMilli()
@@ -1057,7 +1056,7 @@ func (sm *snapshotMeta) CleanupPendingSnapshot(ctx context.Context, snapshot *da
 func (sm *snapshotMeta) GetDeletingSnapshots(ctx context.Context) ([]*datapb.SnapshotInfo, error) {
 	snapshots, err := sm.catalog.ListSnapshots(ctx)
 	if err != nil {
-		return nil, fmt.Errorf("failed to list snapshots from catalog: %w", err)
+		return nil, merr.Wrap(err, "failed to list snapshots from catalog")
 	}
 
 	deletingSnapshots := make([]*datapb.SnapshotInfo, 0)
@@ -1774,7 +1773,7 @@ func (sm *snapshotMeta) generatePinID(snapshot *datapb.SnapshotInfo) (int64, err
 	for i := 0; i < 3; i++ {
 		var b [8]byte
 		if _, err := cryptorand.Read(b[:]); err != nil {
-			return 0, fmt.Errorf("failed to generate random pin ID: %w", err)
+			return 0, merr.WrapErrServiceInternalErr(err, "failed to generate random pin ID")
 		}
 		id := int64(binary.BigEndian.Uint64(b[:]) >> 1)
 		if id == 0 {
@@ -1784,5 +1783,5 @@ func (sm *snapshotMeta) generatePinID(snapshot *datapb.SnapshotInfo) (int64, err
 			return id, nil
 		}
 	}
-	return 0, fmt.Errorf("failed to generate unique pin ID after 3 attempts")
+	return 0, merr.WrapErrServiceInternalMsg("failed to generate unique pin ID after 3 attempts")
 }

internal/datacoord/stats_task_meta.go

@@ -182,7 +182,7 @@ func (stm *statsTaskMeta) UpdateVersion(taskID, nodeID int64) error {
 
 	t, ok := stm.tasks.Get(taskID)
 	if !ok {
-		return fmt.Errorf("task %d not found", taskID)
+		return merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.StatsTask)
@@ -212,7 +212,7 @@ func (stm *statsTaskMeta) UpdateTaskState(taskID int64, state indexpb.JobState,
 
 	t, ok := stm.tasks.Get(taskID)
 	if !ok {
-		return fmt.Errorf("task %d not found", taskID)
+		return merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.StatsTask)
@@ -239,7 +239,7 @@ func (stm *statsTaskMeta) UpdateBuildingTask(taskID int64) error {
 
 	t, ok := stm.tasks.Get(taskID)
 	if !ok {
-		return fmt.Errorf("task %d not found", taskID)
+		return merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.StatsTask)
@@ -267,7 +267,7 @@ func (stm *statsTaskMeta) FinishTask(taskID int64, result *workerpb.StatsResult)
 
 	t, ok := stm.tasks.Get(taskID)
 	if !ok {
-		return fmt.Errorf("task %d not found", taskID)
+		return merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.StatsTask)
@@ -358,7 +358,7 @@ func (stm *statsTaskMeta) MarkTaskCanRecycle(taskID int64) error {
 
 	t, ok := stm.tasks.Get(taskID)
 	if !ok {
-		return fmt.Errorf("task %d not found", taskID)
+		return merr.WrapErrServiceInternalMsg("task %d not found", taskID)
 	}
 
 	cloneT := proto.Clone(t).(*indexpb.StatsTask)

internal/datacoord/task_index.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"path"
 	"time"
 
@@ -241,7 +240,7 @@ func (it *indexBuildTask) prepareJobRequest(ctx context.Context, segment *Segmen
 		var err error
 		params, err = Params.KnowhereConfig.UpdateIndexParams(GetIndexType(params), paramtable.BuildStage, params)
 		if err != nil {
-			return nil, fmt.Errorf("failed to update index build params: %w", err)
+			return nil, merr.WrapErrServiceInternalErr(err, "failed to update index build params")
 		}
 	}
 
@@ -249,14 +248,14 @@ func (it *indexBuildTask) prepareJobRequest(ctx context.Context, segment *Segmen
 		var err error
 		params, err = indexparams.UpdateDiskIndexBuildParams(Params, params)
 		if err != nil {
-			return nil, fmt.Errorf("failed to append index build params: %w", err)
+			return nil, merr.WrapErrServiceInternalErr(err, "failed to append index build params")
 		}
 	}
 
 	// Get collection info and field
 	collectionInfo, err := it.handler.GetCollection(ctx, segment.GetCollectionID())
 	if err != nil {
-		return nil, fmt.Errorf("failed to get collection info: %w", err)
+		return nil, merr.Wrap(err, "failed to get collection info")
 	}
 
 	schema := collectionInfo.Schema
@@ -271,7 +270,7 @@ func (it *indexBuildTask) prepareJobRequest(ctx context.Context, segment *Segmen
 	}
 
 	if field == nil {
-		return nil, fmt.Errorf("field not found with ID %d", fieldID)
+		return nil, merr.WrapErrFieldNotFound(fieldID)
 	}
 
 	// Extract dim only for vector types to avoid unnecessary warnings

internal/datacoord/task_refresh_external_collection.go

@@ -34,6 +34,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/taskcommon"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 )
 
@@ -115,7 +116,7 @@ func (t *refreshExternalCollectionTask) validateSource() error {
 	// Validate against job-level snapshot to isolate in-flight tasks from schema changes.
 	job := t.refreshMeta.GetJob(t.GetJobId())
 	if job == nil {
-		return fmt.Errorf("job %d not found", t.GetJobId())
+		return merr.WrapErrServiceInternalMsg("job %d not found", t.GetJobId())
 	}
 
 	currentSource := job.GetExternalSource()
@@ -125,7 +126,7 @@ func (t *refreshExternalCollectionTask) validateSource() error {
 	taskSpec := t.GetExternalSpec()
 
 	if currentSource != taskSource || currentSpec != taskSpec {
-		return fmt.Errorf(
+		return merr.WrapErrServiceInternalMsg(
 			"task source mismatch: task source=%s/%s, job source=%s/%s (task belongs to a different refresh job)",
 			taskSource, taskSpec, currentSource, currentSpec,
 		)
@@ -215,7 +216,7 @@ func applyExternalCollectionSegmentUpdate(
 	logFields ...zap.Field,
 ) error {
 	if mt == nil {
-		return fmt.Errorf("meta is nil, cannot update segments")
+		return merr.WrapErrServiceInternalMsg("meta is nil, cannot update segments")
 	}
 	fields := append(logFields, zap.Int64("collectionID", collectionID))
 	log := log.Ctx(ctx).With(fields...)
@@ -228,14 +229,14 @@ func applyExternalCollectionSegmentUpdate(
 	for _, segID := range keptSegmentIDs {
 		segment := mt.segments.GetSegment(segID)
 		if segment == nil {
-			return fmt.Errorf("kept segment %d not found", segID)
+			return merr.WrapErrServiceInternalMsg("kept segment %d not found", segID)
 		}
 		if segment.GetCollectionID() != collectionID {
-			return fmt.Errorf("collection mismatch for kept segment %d: existing %d, want %d",
+			return merr.WrapErrServiceInternalMsg("collection mismatch for kept segment %d: existing %d, want %d",
 				segID, segment.GetCollectionID(), collectionID)
 		}
 		if segment.GetState() == commonpb.SegmentState_Dropped {
-			return fmt.Errorf("cannot keep dropped segment %d", segID)
+			return merr.WrapErrServiceInternalMsg("cannot keep dropped segment %d", segID)
 		}
 		keptSegmentMap[segID] = true
 	}
@@ -250,10 +251,10 @@ func applyExternalCollectionSegmentUpdate(
 			return err
 		}
 		if keptSegmentMap[seg.GetID()] {
-			return fmt.Errorf("segment %d cannot be both kept and updated", seg.GetID())
+			return merr.WrapErrServiceInternalMsg("segment %d cannot be both kept and updated", seg.GetID())
 		}
 		if _, ok := upsertSegmentMap[seg.GetID()]; ok {
-			return fmt.Errorf("duplicate updated segment %d", seg.GetID())
+			return merr.WrapErrServiceInternalMsg("duplicate updated segment %d", seg.GetID())
 		}
 		upsertSegmentMap[seg.GetID()] = seg
 		validUpdatedSegments = append(validUpdatedSegments, seg)
@@ -308,7 +309,7 @@ func applyExternalCollectionSegmentUpdate(
 			zap.Int("activeSegmentCount", activeSegmentCount),
 			zap.Int("keptSegments", len(keptSegmentMap)),
 			zap.Int("updatedSegments", len(upsertSegmentMap)))
-		return fmt.Errorf("safety check failed: refusing to drop all %d segments without replacement (keptSegments=%d, updatedSegments=%d)",
+		return merr.WrapErrServiceInternalMsg("safety check failed: refusing to drop all %d segments without replacement (keptSegments=%d, updatedSegments=%d)",
 			activeSegmentCount, len(keptSegmentMap), len(upsertSegmentMap))
 	}
 
@@ -330,15 +331,15 @@ func applyExternalCollectionSegmentUpdate(
 
 	collInfo := mt.GetCollection(collectionID)
 	if collInfo == nil {
-		return fmt.Errorf("collection %d not found in meta", collectionID)
+		return merr.WrapErrServiceInternalMsg("collection %d not found in meta", collectionID)
 	}
 	// External collections are single-shard, single-partition (enforced at creation).
 	// Assert exactly-one here to catch any invariant violation from data corruption or legacy data.
 	if len(collInfo.VChannelNames) != 1 {
-		return fmt.Errorf("external collection %d expected exactly 1 VChannel, got %d", collectionID, len(collInfo.VChannelNames))
+		return merr.WrapErrServiceInternalMsg("external collection %d expected exactly 1 VChannel, got %d", collectionID, len(collInfo.VChannelNames))
 	}
 	if len(collInfo.Partitions) != 1 {
-		return fmt.Errorf("external collection %d expected exactly 1 partition, got %d", collectionID, len(collInfo.Partitions))
+		return merr.WrapErrServiceInternalMsg("external collection %d expected exactly 1 partition, got %d", collectionID, len(collInfo.Partitions))
 	}
 	insertChannel := collInfo.VChannelNames[0]
 	partitionID := collInfo.Partitions[0]
@@ -477,14 +478,14 @@ func applyExternalCollectionSegmentUpdate(
 
 func validateExternalRefreshUpdatedSegment(incoming *datapb.SegmentInfo, collectionID int64) error {
 	if incoming.GetCollectionID() != 0 && incoming.GetCollectionID() != collectionID {
-		return fmt.Errorf("collection mismatch for segment %d: got %d, want %d",
+		return merr.WrapErrServiceInternalMsg("collection mismatch for segment %d: got %d, want %d",
 			incoming.GetID(), incoming.GetCollectionID(), collectionID)
 	}
 	if incoming.GetManifestPath() == "" {
-		return fmt.Errorf("updated segment %d has empty manifest path", incoming.GetID())
+		return merr.WrapErrServiceInternalMsg("updated segment %d has empty manifest path", incoming.GetID())
 	}
 	if len(incoming.GetBinlogs()) == 0 {
-		return fmt.Errorf("updated segment %d has empty fake binlogs", incoming.GetID())
+		return merr.WrapErrServiceInternalMsg("updated segment %d has empty fake binlogs", incoming.GetID())
 	}
 	return nil
 }
@@ -513,36 +514,36 @@ func validateExternalRefreshNewSegment(incoming *datapb.SegmentInfo) error {
 
 func validateExternalRefreshPatch(oldSeg *SegmentInfo, incoming *datapb.SegmentInfo, collectionID int64) error {
 	if oldSeg == nil {
-		return fmt.Errorf("existing segment is nil")
+		return merr.WrapErrServiceInternalMsg("existing segment is nil")
 	}
 	if oldSeg.GetCollectionID() != collectionID {
-		return fmt.Errorf("collection mismatch for segment %d: existing %d, want %d",
+		return merr.WrapErrServiceInternalMsg("collection mismatch for segment %d: existing %d, want %d",
 			oldSeg.GetID(), oldSeg.GetCollectionID(), collectionID)
 	}
 	if oldSeg.GetState() == commonpb.SegmentState_Dropped {
-		return fmt.Errorf("cannot patch dropped segment %d", oldSeg.GetID())
+		return merr.WrapErrServiceInternalMsg("cannot patch dropped segment %d", oldSeg.GetID())
 	}
 	if incoming.GetCollectionID() != 0 && incoming.GetCollectionID() != collectionID {
-		return fmt.Errorf("collection mismatch for segment %d: got %d, want %d",
+		return merr.WrapErrServiceInternalMsg("collection mismatch for segment %d: got %d, want %d",
 			incoming.GetID(), incoming.GetCollectionID(), collectionID)
 	}
 	if incoming.GetNumOfRows() != oldSeg.GetNumOfRows() {
-		return fmt.Errorf("row count changed for segment %d: got %d, want %d",
+		return merr.WrapErrServiceInternalMsg("row count changed for segment %d: got %d, want %d",
 			incoming.GetID(), incoming.GetNumOfRows(), oldSeg.GetNumOfRows())
 	}
 	if incoming.GetStorageVersion() != 0 && incoming.GetStorageVersion() != oldSeg.GetStorageVersion() {
-		return fmt.Errorf("storage version changed for segment %d: got %d, want %d",
+		return merr.WrapErrServiceInternalMsg("storage version changed for segment %d: got %d, want %d",
 			incoming.GetID(), incoming.GetStorageVersion(), oldSeg.GetStorageVersion())
 	}
 	if incoming.GetSchemaVersion() < oldSeg.GetSchemaVersion() {
-		return fmt.Errorf("schema version rollback for segment %d: got %d, want >= %d",
+		return merr.WrapErrServiceInternalMsg("schema version rollback for segment %d: got %d, want >= %d",
 			incoming.GetID(), incoming.GetSchemaVersion(), oldSeg.GetSchemaVersion())
 	}
 	if incoming.GetManifestPath() == "" {
-		return fmt.Errorf("patched segment %d has empty manifest path", incoming.GetID())
+		return merr.WrapErrServiceInternalMsg("patched segment %d has empty manifest path", incoming.GetID())
 	}
 	if len(incoming.GetBinlogs()) == 0 {
-		return fmt.Errorf("patched segment %d has empty fake binlogs", incoming.GetID())
+		return merr.WrapErrServiceInternalMsg("patched segment %d has empty fake binlogs", incoming.GetID())
 	}
 	if err := validateExternalRefreshBinlogRowCount(incoming, oldSeg.GetNumOfRows()); err != nil {
 		return err
@@ -553,14 +554,14 @@ func validateExternalRefreshPatch(oldSeg *SegmentInfo, incoming *datapb.SegmentI
 func validateExternalRefreshBinlogRowCount(segment *datapb.SegmentInfo, expectedRows int64) error {
 	binlogRows := segmentutil.CalcRowCountFromBinLog(segment)
 	if binlogRows == -1 {
-		return fmt.Errorf("invalid binlog row count for segment %d", segment.GetID())
+		return merr.WrapErrServiceInternalMsg("invalid binlog row count for segment %d", segment.GetID())
 	}
 	if expectedRows > 0 && binlogRows != expectedRows {
-		return fmt.Errorf("binlog row count mismatch for segment %d: got %d, want %d",
+		return merr.WrapErrServiceInternalMsg("binlog row count mismatch for segment %d: got %d, want %d",
 			segment.GetID(), binlogRows, expectedRows)
 	}
 	if binlogRows > 0 && binlogRows != segment.GetNumOfRows() {
-		return fmt.Errorf("binlog row count mismatch for segment %d: got %d, segment rows %d",
+		return merr.WrapErrServiceInternalMsg("binlog row count mismatch for segment %d: got %d, segment rows %d",
 			segment.GetID(), binlogRows, segment.GetNumOfRows())
 	}
 	return nil
@@ -612,7 +613,7 @@ func (t *refreshExternalCollectionTask) CreateTaskOnWorker(nodeID int64, cluster
 	log.Info("creating refresh task on worker")
 
 	if t.mt == nil {
-		err = fmt.Errorf("meta is nil, cannot create task on worker")
+		err = merr.WrapErrServiceInternalMsg("meta is nil, cannot create task on worker")
 		return
 	}
 
@@ -625,7 +626,7 @@ func (t *refreshExternalCollectionTask) CreateTaskOnWorker(nodeID int64, cluster
 	// Re-read task from meta to sync in-memory state (nodeID and version)
 	updatedTask := t.refreshMeta.GetTask(t.GetTaskId())
 	if updatedTask == nil {
-		err = fmt.Errorf("task %d not found after version update", t.GetTaskId())
+		err = merr.WrapErrServiceInternalMsg("task %d not found after version update", t.GetTaskId())
 		return
 	}
 	t.ExternalCollectionRefreshTask = updatedTask
@@ -668,11 +669,11 @@ func (t *refreshExternalCollectionTask) CreateTaskOnWorker(nodeID int64, cluster
 	// lock, before they are supported.
 	collInfo := t.mt.GetCollection(t.GetCollectionId())
 	if collInfo == nil {
-		err = fmt.Errorf("collection %d not found in meta", t.GetCollectionId())
+		err = merr.WrapErrServiceInternalMsg("collection %d not found in meta", t.GetCollectionId())
 		return
 	}
 	if len(collInfo.Partitions) != 1 {
-		err = fmt.Errorf("external collection %d expected exactly 1 partition, got %d", t.GetCollectionId(), len(collInfo.Partitions))
+		err = merr.WrapErrServiceInternalMsg("external collection %d expected exactly 1 partition, got %d", t.GetCollectionId(), len(collInfo.Partitions))
 		return
 	}
 	partitionID := collInfo.Partitions[0]

internal/datacoord/task_stats.go

@@ -18,7 +18,6 @@ package datacoord
 
 import (
 	"context"
-	"fmt"
 	"time"
 
 	"github.com/cockroachdb/errors"
@@ -299,11 +298,17 @@ func (st *statsTask) handleEmptySegment(ctx context.Context) error {
 // Prepare the stats request
 func (st *statsTask) prepareJobRequest(ctx context.Context, segment *SegmentInfo) (*workerpb.CreateStatsRequest, error) {
 	collInfo, err := st.handler.GetCollection(ctx, segment.GetCollectionID())
-	if err != nil || collInfo == nil {
+	if err != nil {
-		return nil, fmt.Errorf("failed to get collection info: %w", err)
+		return nil, merr.Wrap(err, "failed to get collection info")
+	}
+	// GetCollection can return (nil, nil) on a cache miss; merr.Wrap(nil) would
+	// be nil and silently submit a malformed request, so guard collInfo
+	// separately with a typed not-found.
+	if collInfo == nil {
+		return nil, merr.WrapErrCollectionNotFound(segment.GetCollectionID())
 	}
 	if collInfo.Schema == nil || len(collInfo.Schema.GetFields()) == 0 {
-		return nil, fmt.Errorf("collection schema is nil or has no fields, collectionID: %d", segment.GetCollectionID())
+		return nil, merr.WrapErrServiceInternalMsg("collection schema is nil or has no fields, collectionID: %d", segment.GetCollectionID())
 	}
 
 	// Calculate binlog allocation
@@ -314,7 +319,7 @@ func (st *statsTask) prepareJobRequest(ctx context.Context, segment *SegmentInfo
 	// Allocate IDs
 	start, end, err := st.allocator.AllocN(binlogNum + int64(len(collInfo.Schema.GetFunctions())) + 1)
 	if err != nil {
-		return nil, fmt.Errorf("failed to allocate log IDs: %w", err)
+		return nil, merr.Wrap(err, "failed to allocate log IDs")
 	}
 
 	// Create the request

internal/datanode/compactor/bump_schema_version_compactor.go

@@ -26,7 +26,6 @@ import (
 
 	"github.com/apache/arrow/go/v17/arrow"
 	"github.com/apache/arrow/go/v17/arrow/array"
-	"github.com/cockroachdb/errors"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/zap"
 
@@ -113,14 +112,16 @@ func (t *bumpSchemaVersionCompactionTask) preCompact() error {
 		return t.ctx.Err()
 	}
 
-	// Check segment binlogs: must have exactly one segment
+	// Check segment binlogs: must have exactly one segment.
+	// The plan is produced by datacoord, so a malformed plan is an internal
+	// protocol violation, not user input.
 	if len(t.plan.GetSegmentBinlogs()) != 1 {
-		return errors.Newf("schema bump compaction plan is illegal, must have exactly one segment, but got %d segments, planID = %d", len(t.plan.GetSegmentBinlogs()), t.GetPlanID())
+		return merr.WrapErrServiceInternalMsg("schema bump compaction plan is illegal, must have exactly one segment, but got %d segments, planID = %d", len(t.plan.GetSegmentBinlogs()), t.GetPlanID())
 	}
 
 	segment := t.plan.GetSegmentBinlogs()[0]
 	if segment.GetStorageVersion() < storage.StorageV3 || segment.GetManifest() == "" {
-		return errors.Newf("schema bump compaction requires a StorageV3 segment with manifest, planID = %d, segmentID = %d, storageVersion = %d", t.GetPlanID(), segment.GetSegmentID(), segment.GetStorageVersion())
+		return merr.WrapErrServiceInternalMsg("schema bump compaction requires a StorageV3 segment with manifest, planID = %d, segmentID = %d, storageVersion = %d", t.GetPlanID(), segment.GetSegmentID(), segment.GetStorageVersion())
 	}
 
 	return nil
@@ -147,7 +148,7 @@ func (t *bumpSchemaVersionCompactionTask) missingFunctionInputSchema(missingFunc
 		for _, inputFieldID := range functionSchema.GetInputFieldIds() {
 			inputField := typeutil.GetField(schema, inputFieldID)
 			if inputField == nil {
-				return nil, nil, errors.New("input field not found in schema")
+				return nil, nil, merr.WrapErrParameterInvalidMsg("input field not found in schema")
 			}
 			if err := validateMaterializationInputField(functionSchema, inputField); err != nil {
 				return nil, nil, err
@@ -196,11 +197,11 @@ func bm25AdditionalInputFields(schema *schemapb.CollectionSchema, inputField *sc
 		return nil, err
 	}
 	if multiAnalyzerParams.ByField == "" {
-		return nil, errors.New("multi_analyzer_params missing required 'by_field' key")
+		return nil, merr.WrapErrParameterInvalidMsg("multi_analyzer_params missing required 'by_field' key")
 	}
 	byField := typeutil.GetFieldByName(schema, multiAnalyzerParams.ByField)
 	if byField == nil {
-		return nil, errors.New("input field not found in schema")
+		return nil, merr.WrapErrParameterInvalidMsg("input field not found in schema")
 	}
 	return []*schemapb.FieldSchema{byField}, nil
 }
@@ -217,7 +218,7 @@ func (t *bumpSchemaVersionCompactionTask) missingFunctionOutputFields(missingFun
 			outputFieldID := functionSchema.GetOutputFieldIds()[outputIndex]
 			outputField := typeutil.GetField(schema, outputFieldID)
 			if outputField == nil {
-				return nil, nil, errors.New("output field not found in schema")
+				return nil, nil, merr.WrapErrParameterInvalidMsg("output field not found in schema")
 			}
 			if err := validateMaterializationOutputField(functionSchema, outputField); err != nil {
 				return nil, nil, err
@@ -227,7 +228,7 @@ func (t *bumpSchemaVersionCompactionTask) missingFunctionOutputFields(missingFun
 		}
 	}
 	if len(fields) == 0 {
-		return nil, nil, errors.New("no missing function output fields")
+		return nil, nil, merr.WrapErrParameterInvalidMsg("no missing function output fields")
 	}
 	return fields, fieldIDs, nil
 }
@@ -236,22 +237,22 @@ func validateSupportedMissingFunctionMaterialization(functionSchema *schemapb.Fu
 	switch functionSchema.GetType() {
 	case schemapb.FunctionType_BM25:
 		if len(functionSchema.GetInputFieldIds()) == 0 {
-			return errors.New("bm25 function should have input fields")
+			return merr.WrapErrParameterInvalidMsg("bm25 function should have input fields")
 		}
 		if len(functionSchema.GetOutputFieldIds()) == 0 {
-			return errors.New("bm25 function should have output fields")
+			return merr.WrapErrParameterInvalidMsg("bm25 function should have output fields")
 		}
 		return nil
 	case schemapb.FunctionType_MinHash:
 		if len(functionSchema.GetInputFieldIds()) == 0 {
-			return errors.New("minhash function should have input fields")
+			return merr.WrapErrParameterInvalidMsg("minhash function should have input fields")
 		}
 		if len(functionSchema.GetOutputFieldIds()) == 0 {
-			return errors.New("minhash function should have output fields")
+			return merr.WrapErrParameterInvalidMsg("minhash function should have output fields")
 		}
 		return nil
 	default:
-		return errors.New("unsupported function type")
+		return merr.WrapErrParameterInvalidMsg("unsupported function type")
 	}
 }
 
@@ -259,11 +260,11 @@ func validateMaterializationInputField(functionSchema *schemapb.FunctionSchema,
 	switch functionSchema.GetType() {
 	case schemapb.FunctionType_BM25:
 		if field.GetDataType() != schemapb.DataType_VarChar && field.GetDataType() != schemapb.DataType_Text {
-			return errors.New("input field data type must be varchar or text for bm25 materialization")
+			return merr.WrapErrParameterInvalidMsg("input field data type must be varchar or text for bm25 materialization")
 		}
 	case schemapb.FunctionType_MinHash:
 		if field.GetDataType() != schemapb.DataType_VarChar && field.GetDataType() != schemapb.DataType_Text {
-			return errors.New("input field data type must be varchar or text for minhash materialization")
+			return merr.WrapErrParameterInvalidMsg("input field data type must be varchar or text for minhash materialization")
 		}
 	}
 	return nil
@@ -273,11 +274,11 @@ func validateMaterializationOutputField(functionSchema *schemapb.FunctionSchema,
 	switch functionSchema.GetType() {
 	case schemapb.FunctionType_BM25:
 		if field.GetDataType() != schemapb.DataType_SparseFloatVector {
-			return errors.New("output field data type must be sparse float vector for bm25 materialization")
+			return merr.WrapErrParameterInvalidMsg("output field data type must be sparse float vector for bm25 materialization")
 		}
 	case schemapb.FunctionType_MinHash:
 		if field.GetDataType() != schemapb.DataType_BinaryVector {
-			return errors.New("output field data type must be binary vector for minhash materialization")
+			return merr.WrapErrParameterInvalidMsg("output field data type must be binary vector for minhash materialization")
 		}
 	}
 	return nil
@@ -546,7 +547,7 @@ func (t *bumpSchemaVersionCompactionTask) runFullSchemaRewrite(existingFields ma
 		}
 		manifestPath, err = packed.AddStatsToManifest(manifestPath, t.compactionParams.StorageConfig, statEntries)
 		if err != nil {
-			return nil, merr.WrapErrServiceInternal("failed to add writer stats to schema bump full rewrite manifest", err.Error())
+			return nil, merr.Wrap(err, "failed to add writer stats to schema bump full rewrite manifest")
 		}
 	}
 	sortedInsertLogs := storage.SortFieldBinlogs(insertLogs)
@@ -602,7 +603,7 @@ func (t *bumpSchemaVersionCompactionTask) runFullSchemaRewrite(existingFields ma
 func appendBM25StatsFromArrowArray(stats *storage.BM25Stats, arr arrow.Array) (int, error) {
 	binaryArray, ok := arr.(*array.Binary)
 	if !ok {
-		return 0, errors.Newf("bm25 output field must be arrow binary array, got %T", arr)
+		return 0, merr.WrapErrParameterInvalidMsg("bm25 output field must be arrow binary array, got %T", arr)
 	}
 	memorySize := 0
 	for i := 0; i < binaryArray.Len(); i++ {
@@ -667,7 +668,7 @@ func (t *bumpSchemaVersionCompactionTask) setupWriter(outputFields []*schemapb.F
 
 	basePath, existingVersion, err := packed.UnmarshalManifestPath(segment.GetManifest())
 	if err != nil {
-		return nil, merr.WrapErrServiceInternal("failed to parse existing manifest for schema bump", err.Error())
+		return nil, merr.WrapErrDataIntegrityMsg("failed to parse existing manifest for schema bump: %s", err.Error())
 	}
 	writerResult, err := t.newV3WriterResult(outputSchema, newColumnGroups, segment, collectionID, basePath, existingVersion)
 	if err != nil {
@@ -685,7 +686,7 @@ func (t *bumpSchemaVersionCompactionTask) setupWriter(outputFields []*schemapb.F
 
 func (t *bumpSchemaVersionCompactionTask) newV3WriterResult(schema *schemapb.CollectionSchema, columnGroups []storagecommon.ColumnGroup, segment *datapb.CompactionSegmentBinlogs, collectionID int64, basePath string, baseVersion int64) (*bumpSchemaVersionWriterResult, error) {
 	if segment.GetStorageVersion() < storage.StorageV3 || segment.GetManifest() == "" {
-		return nil, errors.New("schema bump compaction requires a StorageV3 segment with manifest")
+		return nil, merr.WrapErrParameterInvalidMsg("schema bump compaction requires a StorageV3 segment with manifest")
 	}
 
 	pluginContext, err := hookutil.GetCPluginContext(t.plan.GetPluginContext(), collectionID)
@@ -732,7 +733,7 @@ func (t *bumpSchemaVersionCompactionTask) addV3Stats(prefix string, fieldID int6
 	statsRelPath := fmt.Sprintf("_stats/%s.%d/%d", prefix, fieldID, statsID)
 	absStatsPath := path.Join(writerResult.basePath, statsRelPath)
 	if err := packed.WriteFile(t.compactionParams.StorageConfig, absStatsPath, bytes); err != nil {
-		return merr.WrapErrServiceInternal("failed to write V3 stats", err.Error())
+		return merr.Wrap(err, "failed to write V3 stats")
 	}
 	writerResult.v3Stats = append(writerResult.v3Stats, packed.FieldBinlogStatEntry(prefix, fieldID, &datapb.FieldBinlog{
 		FieldID: fieldID,
@@ -773,7 +774,7 @@ func (t *bumpSchemaVersionCompactionTask) buildMergedLogsV3(segment *datapb.Comp
 func (t *bumpSchemaVersionCompactionTask) preserveDeltaLogsV3(segment *datapb.CompactionSegmentBinlogs, manifestPath string) (string, error) {
 	deltaPaths, err := packed.GetDeltaLogPathsFromManifest(segment.GetManifest(), t.compactionParams.StorageConfig)
 	if err != nil {
-		return "", merr.WrapErrServiceInternal("failed to read V3 delta logs from existing manifest", err.Error())
+		return "", merr.Wrap(err, "failed to read V3 delta logs from existing manifest")
 	}
 	if len(deltaPaths) == 0 {
 		return manifestPath, nil
@@ -784,12 +785,12 @@ func (t *bumpSchemaVersionCompactionTask) preserveDeltaLogsV3(segment *datapb.Co
 		deltaSummaries = append(deltaSummaries, fieldBinlog.GetBinlogs()...)
 	}
 	if len(deltaSummaries) != len(deltaPaths) {
-		return "", merr.WrapErrServiceInternal(fmt.Sprintf("V3 delta manifest path count %d does not match segment delta summary count %d", len(deltaPaths), len(deltaSummaries)))
+		return "", merr.WrapErrServiceInternalMsg("V3 delta manifest path count %d does not match segment delta summary count %d", len(deltaPaths), len(deltaSummaries))
 	}
 
 	basePath, _, err := packed.UnmarshalManifestPath(manifestPath)
 	if err != nil {
-		return "", merr.WrapErrServiceInternal("failed to parse new V3 manifest for delta preservation", err.Error())
+		return "", merr.WrapErrDataIntegrityMsg("failed to parse new V3 manifest for delta preservation: %s", err.Error())
 	}
 
 	deltaEntries := make([]packed.DeltaLogEntry, 0, len(deltaPaths))
@@ -797,7 +798,7 @@ func (t *bumpSchemaVersionCompactionTask) preserveDeltaLogsV3(segment *datapb.Co
 		newDeltaPath := metautil.BuildDeltaLogPathV3(basePath, deltaSummaries[i].GetLogID())
 		if newDeltaPath != deltaPath {
 			if err := t.chunkManager.Copy(t.ctx, deltaPath, newDeltaPath); err != nil {
-				return "", merr.WrapErrServiceInternal("failed to copy V3 delta log for schema bump full rewrite", err.Error())
+				return "", merr.Wrap(err, "failed to copy V3 delta log for schema bump full rewrite")
 			}
 		}
 		deltaEntries = append(deltaEntries, packed.DeltaLogEntry{
@@ -807,7 +808,7 @@ func (t *bumpSchemaVersionCompactionTask) preserveDeltaLogsV3(segment *datapb.Co
 	}
 	newManifest, err := packed.AddDeltaLogsToManifest(manifestPath, t.compactionParams.StorageConfig, deltaEntries)
 	if err != nil {
-		return "", merr.WrapErrServiceInternal("failed to preserve V3 delta logs in full rewrite manifest", err.Error())
+		return "", merr.Wrap(err, "failed to preserve V3 delta logs in full rewrite manifest")
 	}
 	return newManifest, nil
 }
@@ -923,7 +924,7 @@ func (t *bumpSchemaVersionCompactionTask) runMissingFunctionMaterialization(ctx
 			if outputCol == nil {
 				releaseWrappedRecord(wrapped, record)
 				span.End()
-				return nil, merr.WrapErrServiceInternal(fmt.Sprintf("output field %d not found in materialized record", outputFieldID))
+				return nil, merr.WrapErrServiceInternalMsg("output field %d not found in materialized record", outputFieldID)
 			}
 			batchMemSize := arrowArrayMemorySize(outputCol)
 			if stats, ok := statsByField[outputFieldID]; ok {
@@ -997,7 +998,7 @@ func (t *bumpSchemaVersionCompactionTask) runMissingFunctionMaterialization(ctx
 		},
 	)
 	if err != nil {
-		return nil, merr.WrapErrServiceInternal("failed to commit schema bump V3 manifest", err.Error())
+		return nil, merr.Wrap(err, "failed to commit schema bump V3 manifest")
 	}
 	log.Info("[schema-bump-partial-writer] writer output and bm25 stats committed",
 		zap.String("manifestPath", manifestPath),

internal/datanode/compactor/clustering_compactor.go

@@ -28,7 +28,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/atomic"
@@ -663,7 +662,7 @@ func (t *clusteringCompactionTask) mappingSegment(
 			row, ok := v.Value.(map[typeutil.UniqueID]interface{})
 			if !ok {
 				log.Warn("convert interface to map wrong")
-				return errors.New("unexpected error")
+				return merr.WrapErrServiceInternalMsg("unexpected error")
 			}
 			expireTs := int64(-1)
 			if hasTTLField {
@@ -976,7 +975,7 @@ func (t *clusteringCompactionTask) iterAndGetScalarAnalyzeResult(pkIter *storage
 		// rowValue := vIter.GetData().(*iterators.InsertRow).GetValue()
 		row, ok := (*v).Value.(map[typeutil.UniqueID]interface{})
 		if !ok {
-			return nil, 0, errors.New("unexpected error")
+			return nil, 0, merr.WrapErrServiceInternalMsg("unexpected error")
 		}
 
 		expireTs := int64(-1)

internal/datanode/compactor/compactor_common.go

@@ -46,6 +46,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexcgopb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 	"github.com/milvus-io/milvus/pkg/v3/util/tsoutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
@@ -80,7 +81,7 @@ func createTextIndex(ctx context.Context,
 		}
 		binlogs, ok := fieldBinlogs[fieldID]
 		if !ok {
-			return nil, fmt.Errorf("field binlog not found for field %d", fieldID)
+			return nil, merr.WrapErrParameterInvalidMsg("field binlog not found for field %d", fieldID)
 		}
 		result := make([]string, 0, len(binlogs))
 		for _, binlog := range binlogs {
@@ -136,7 +137,7 @@ func createTextIndex(ctx context.Context,
 			if segment.GetManifest() != "" {
 				basePath, _, err := packed.UnmarshalManifestPath(segment.GetManifest())
 				if err != nil {
-					return fmt.Errorf("failed to unmarshal manifest path for text_index basePath: %w", err)
+					return merr.Wrap(err, "failed to unmarshal manifest path for text_index basePath")
 				}
 				statsBasePath = fmt.Sprintf("%s/_stats/text_index.%d", basePath, field.GetFieldID())
 			}

internal/datanode/compactor/l0_compactor.go

@@ -18,10 +18,10 @@ package compactor
 
 import (
 	"context"
+	"fmt"
 	"math"
 	"sync"
 
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/zap"
@@ -40,6 +40,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
 	"github.com/milvus-io/milvus/pkg/v3/util/funcutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/hardware"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/timerecord"
@@ -135,7 +136,9 @@ func (t *LevelZeroCompactionTask) Compact() (*datapb.CompactionPlanResult, error
 	})
 	if len(targetSegments) == 0 {
 		log.Warn("compact wrong, not target sealed segments")
-		return nil, errors.New("illegal compaction plan with empty target segments")
+		// The plan is produced by datacoord, so a malformed plan is an internal
+		// protocol violation, not user input.
+		return nil, merr.WrapErrServiceInternalMsg("illegal compaction plan with empty target segments")
 	}
 	err = binlog.DecompressCompactionBinlogsWithRootPath(t.compactionParams.StorageConfig.GetRootPath(), l0Segments)
 	if err != nil {
@@ -416,7 +419,7 @@ func (t *LevelZeroCompactionTask) process(ctx context.Context, l0MemSize int64,
 	ratio := paramtable.Get().DataNodeCfg.L0BatchMemoryRatio.GetAsFloat()
 	memLimit := float64(hardware.GetFreeMemoryCount()) * ratio
 	if float64(l0MemSize) > memLimit {
-		return nil, errors.Newf("L0 compaction failed, not enough memory, request memory size: %v, memory limit: %v", l0MemSize, memLimit)
+		return nil, merr.Wrap(merr.ErrServiceMemoryLimitExceeded, fmt.Sprintf("L0 compaction failed, not enough memory, request memory size: %v, memory limit: %v", l0MemSize, memLimit))
 	}
 
 	log.Info("L0 compaction process start")

internal/datanode/compactor/mix_compactor.go

@@ -25,7 +25,6 @@ import (
 	"time"
 
 	"github.com/apache/arrow/go/v17/arrow/array"
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/zap"
@@ -42,6 +41,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/funcutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/timerecord"
@@ -110,11 +110,13 @@ func (t *mixCompactionTask) preCompact() error {
 	}
 
 	if len(t.plan.GetSegmentBinlogs()) < 1 {
-		return errors.Newf("compaction plan is illegal, there's no segments in compaction plan, planID = %d", t.GetPlanID())
+		// The plan is produced by datacoord, so a malformed plan is an internal
+		// protocol violation, not user input.
+		return merr.WrapErrServiceInternalMsg("compaction plan is illegal, there's no segments in compaction plan, planID = %d", t.GetPlanID())
 	}
 
 	if t.plan.GetMaxSize() == 0 {
-		return errors.Newf("compaction plan is illegal, empty maxSize, planID = %d", t.GetPlanID())
+		return merr.WrapErrServiceInternalMsg("compaction plan is illegal, empty maxSize, planID = %d", t.GetPlanID())
 	}
 
 	t.collectionID = t.plan.GetSegmentBinlogs()[0].GetCollectionID()
@@ -440,7 +442,7 @@ func (t *mixCompactionTask) Compact() (*datapb.CompactionPlanResult, error) {
 
 	if isEmpty {
 		log.Warn("compact wrong, all segments' binlogs are empty")
-		return nil, errors.New("illegal compaction plan")
+		return nil, merr.WrapErrServiceInternalMsg("illegal compaction plan")
 	}
 
 	sortMergeAppicable := t.compactionParams.UseMergeSort

internal/datanode/compactor/record_materializer.go

@@ -17,12 +17,9 @@
 package compactor
 
 import (
-	"fmt"
-
 	"github.com/apache/arrow/go/v17/arrow"
 	"github.com/apache/arrow/go/v17/arrow/array"
 	"github.com/apache/arrow/go/v17/arrow/memory"
-	"github.com/cockroachdb/errors"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	"github.com/milvus-io/milvus/internal/storage"
@@ -79,7 +76,7 @@ func NewRecordMaterializer(schema *schemapb.CollectionSchema, functions []*schem
 		}
 		if runner == nil {
 			materializer.Close()
-			return nil, errors.Newf("failed to set up function runner for %s", functionSchema.GetName())
+			return nil, merr.WrapErrFunctionFailedMsg("failed to set up function runner for %s", functionSchema.GetName())
 		}
 		functionMaterializer, err := newFunctionMaterializer(schema, runner, outputIndexes, true)
 		if err != nil {
@@ -243,7 +240,7 @@ func (r *selectedRecord) Column(fieldID storage.FieldID) arrow.Array {
 	defer selected.Release()
 	col := selected.Column(fieldID)
 	if col == nil {
-		r.err = errors.Newf("selected record field %d not found", fieldID)
+		r.err = merr.WrapErrServiceInternalMsg("selected record field %d not found", fieldID)
 		return nil
 	}
 	col.Retain()
@@ -347,7 +344,7 @@ func newFunctionMaterializer(schema *schemapb.CollectionSchema, runner function.
 	case schemapb.FunctionType_MinHash:
 		return newMinHashFunctionMaterializer(schema, runner, missingOutputIndexes, ownRunner)
 	default:
-		return nil, errors.Newf("unsupported function type %s", functionSchema.GetType().String())
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported function type %s", functionSchema.GetType().String())
 	}
 }
 
@@ -355,35 +352,35 @@ func newMinHashFunctionMaterializer(schema *schemapb.CollectionSchema, runner fu
 	functionSchema := runner.GetSchema()
 	inputFields := runner.GetInputFields()
 	if len(inputFields) == 0 {
-		return nil, errors.New("minhash function should have input fields")
+		return nil, merr.WrapErrFunctionFailedMsg("minhash function should have input fields")
 	}
 	inputFieldIDs := make([]int64, 0, len(inputFields))
 	for _, inputField := range inputFields {
 		if inputField == nil || typeutil.GetField(schema, inputField.GetFieldID()) == nil {
-			return nil, errors.New("input field not found in schema")
+			return nil, merr.WrapErrFunctionFailedMsg("input field not found in schema")
 		}
 		if inputField.GetDataType() != schemapb.DataType_VarChar && inputField.GetDataType() != schemapb.DataType_Text {
-			return nil, errors.New("input field data type must be varchar or text for minhash function materialization")
+			return nil, merr.WrapErrFunctionFailedMsg("input field data type must be varchar or text for minhash function materialization")
 		}
 		inputFieldIDs = append(inputFieldIDs, inputField.GetFieldID())
 	}
 
 	outputFieldIDs := functionSchema.GetOutputFieldIds()
 	if len(outputFieldIDs) == 0 {
-		return nil, errors.New("minhash function should have output fields")
+		return nil, merr.WrapErrFunctionFailedMsg("minhash function should have output fields")
 	}
 
 	outputFields := make(map[int64]*schemapb.FieldSchema, len(outputFieldIDs))
 	for _, outputFieldID := range outputFieldIDs {
 		outputField := typeutil.GetField(schema, outputFieldID)
 		if outputField == nil {
-			return nil, errors.New("output field not found in schema")
+			return nil, merr.WrapErrFunctionFailedMsg("output field not found in schema")
 		}
 		if outputField.GetDataType() != schemapb.DataType_BinaryVector {
-			return nil, errors.New("output field data type must be binary vector for minhash function materialization")
+			return nil, merr.WrapErrFunctionFailedMsg("output field data type must be binary vector for minhash function materialization")
 		}
 		if outputField.GetNullable() {
-			return nil, errors.Newf("function output field cannot be nullable: function %s, field %s", functionSchema.GetName(), outputField.GetName())
+			return nil, merr.WrapErrFunctionFailedMsg("function output field cannot be nullable: function %s, field %s", functionSchema.GetName(), outputField.GetName())
 		}
 		outputFields[outputFieldID] = outputField
 	}
@@ -402,35 +399,35 @@ func newBM25FunctionMaterializer(schema *schemapb.CollectionSchema, runner funct
 	functionSchema := runner.GetSchema()
 	inputFields := runner.GetInputFields()
 	if len(inputFields) == 0 {
-		return nil, errors.New("bm25 function should have input fields")
+		return nil, merr.WrapErrParameterInvalidMsg("bm25 function should have input fields")
 	}
 	inputFieldIDs := make([]int64, 0, len(inputFields))
 	for _, inputField := range inputFields {
 		if inputField == nil || typeutil.GetField(schema, inputField.GetFieldID()) == nil {
-			return nil, errors.New("input field not found in schema")
+			return nil, merr.WrapErrParameterInvalidMsg("input field not found in schema")
 		}
 		if inputField.GetDataType() != schemapb.DataType_VarChar && inputField.GetDataType() != schemapb.DataType_Text {
-			return nil, errors.New("input field data type must be varchar or text for bm25 function materialization")
+			return nil, merr.WrapErrParameterInvalidMsg("input field data type must be varchar or text for bm25 function materialization")
 		}
 		inputFieldIDs = append(inputFieldIDs, inputField.GetFieldID())
 	}
 
 	outputFieldIDs := functionSchema.GetOutputFieldIds()
 	if len(outputFieldIDs) == 0 {
-		return nil, errors.New("bm25 function should have output fields")
+		return nil, merr.WrapErrParameterInvalidMsg("bm25 function should have output fields")
 	}
 
 	outputFields := make(map[int64]*schemapb.FieldSchema, len(outputFieldIDs))
 	for _, outputFieldID := range outputFieldIDs {
 		outputField := typeutil.GetField(schema, outputFieldID)
 		if outputField == nil {
-			return nil, errors.New("output field not found in schema")
+			return nil, merr.WrapErrParameterInvalidMsg("output field not found in schema")
 		}
 		if outputField.GetDataType() != schemapb.DataType_SparseFloatVector {
-			return nil, errors.New("output field data type must be sparse float vector for bm25 function materialization")
+			return nil, merr.WrapErrParameterInvalidMsg("output field data type must be sparse float vector for bm25 function materialization")
 		}
 		if outputField.GetNullable() {
-			return nil, errors.Newf("function output field cannot be nullable: function %s, field %s", functionSchema.GetName(), outputField.GetName())
+			return nil, merr.WrapErrParameterInvalidMsg("function output field cannot be nullable: function %s, field %s", functionSchema.GetName(), outputField.GetName())
 		}
 		outputFields[outputFieldID] = outputField
 	}
@@ -459,7 +456,7 @@ func (m *bm25FunctionMaterializer) Materialize(rec storage.Record) (map[int64]ar
 		return nil, err
 	}
 	if len(outputs) != len(m.outputFieldIDs) {
-		return nil, errors.Newf("bm25 function materialization expects %d outputs, got %d", len(m.outputFieldIDs), len(outputs))
+		return nil, merr.WrapErrFunctionFailedMsg("bm25 function materialization expects %d outputs, got %d", len(m.outputFieldIDs), len(outputs))
 	}
 
 	result := make(map[int64]arrow.Array, len(m.missingOutputIndexes))
@@ -468,7 +465,7 @@ func (m *bm25FunctionMaterializer) Materialize(rec storage.Record) (map[int64]ar
 		outputSparseArray, ok := outputs[outputIndex].(*schemapb.SparseFloatArray)
 		if !ok {
 			releaseArrowArrays(result)
-			return nil, errors.Newf("unexpected output type from BM25 function runner, expected SparseFloatArray, got %T", outputs[outputIndex])
+			return nil, merr.WrapErrFunctionFailedMsg("unexpected output type from BM25 function runner, expected SparseFloatArray, got %T", outputs[outputIndex])
 		}
 		arr, err := buildSparseFloatVectorArrowArray(m.outputFields[outputFieldID], outputSparseArray, rec.Len())
 		if err != nil {
@@ -500,7 +497,7 @@ func (m *minHashFunctionMaterializer) Materialize(rec storage.Record) (map[int64
 		return nil, err
 	}
 	if len(outputs) != len(m.outputFieldIDs) {
-		return nil, errors.Newf("minhash function materialization expects %d outputs, got %d", len(m.outputFieldIDs), len(outputs))
+		return nil, merr.WrapErrFunctionFailedMsg("minhash function materialization expects %d outputs, got %d", len(m.outputFieldIDs), len(outputs))
 	}
 
 	result := make(map[int64]arrow.Array, len(m.missingOutputIndexes))
@@ -509,12 +506,12 @@ func (m *minHashFunctionMaterializer) Materialize(rec storage.Record) (map[int64
 		outputFieldData, ok := outputs[outputIndex].(*schemapb.FieldData)
 		if !ok {
 			releaseArrowArrays(result)
-			return nil, errors.Newf("unexpected output type from MinHash function runner, expected FieldData, got %T", outputs[outputIndex])
+			return nil, merr.WrapErrFunctionFailedMsg("unexpected output type from MinHash function runner, expected FieldData, got %T", outputs[outputIndex])
 		}
 		vectorField := outputFieldData.GetVectors()
 		if vectorField == nil || vectorField.GetBinaryVector() == nil {
 			releaseArrowArrays(result)
-			return nil, errors.New("unexpected output from MinHash function runner, expected binary vector field data")
+			return nil, merr.WrapErrFunctionFailedMsg("unexpected output from MinHash function runner, expected binary vector field data")
 		}
 		fieldData := &storage.BinaryVectorFieldData{
 			Data: vectorField.GetBinaryVector(),
@@ -522,7 +519,7 @@ func (m *minHashFunctionMaterializer) Materialize(rec storage.Record) (map[int64
 		}
 		if fieldData.RowNum() != rec.Len() {
 			releaseArrowArrays(result)
-			return nil, errors.Newf("minhash function output row count mismatch, expected %d, got %d", rec.Len(), fieldData.RowNum())
+			return nil, merr.WrapErrFunctionFailedMsg("minhash function output row count mismatch, expected %d, got %d", rec.Len(), fieldData.RowNum())
 		}
 		arr, err := buildArrowArrayFromFieldData(m.outputFields[outputFieldID], fieldData, rec.Len())
 		if err != nil {
@@ -577,7 +574,7 @@ func missingNonMaterializedSchemaFields(schema *schemapb.CollectionSchema, exist
 func stringInputsFromRecord(rec storage.Record, fieldID int64) ([]string, error) {
 	col := rec.Column(fieldID)
 	if col == nil {
-		return nil, merr.WrapErrServiceInternal(fmt.Sprintf("input field %d not found in record", fieldID))
+		return nil, merr.WrapErrFunctionFailedMsg("input field %d not found in record", fieldID)
 	}
 	inputs := make([]string, rec.Len())
 	switch values := col.(type) {
@@ -588,16 +585,16 @@ func stringInputsFromRecord(rec storage.Record, fieldID int64) ([]string, error)
 			}
 		}
 	case *array.Binary:
-		return nil, merr.WrapErrServiceInternal("cannot materialize bm25 from text binary values without lob decoding")
+		return nil, merr.WrapErrFunctionFailedMsg("cannot materialize bm25 from text binary values without lob decoding")
 	default:
-		return nil, merr.WrapErrServiceInternal(fmt.Sprintf("input field %d data type must be varchar or text for bm25 function materialization, got %T", fieldID, col))
+		return nil, merr.WrapErrFunctionFailedMsg("input field %d data type must be varchar or text for bm25 function materialization, got %T", fieldID, col)
 	}
 	return inputs, nil
 }
 
 func buildSparseFloatVectorArrowArray(field *schemapb.FieldSchema, outputSparseArray *schemapb.SparseFloatArray, rowCount int) (arrow.Array, error) {
 	if len(outputSparseArray.GetContents()) != rowCount {
-		return nil, errors.Newf("bm25 function output row count mismatch, expected %d, got %d", rowCount, len(outputSparseArray.GetContents()))
+		return nil, merr.WrapErrFunctionFailedMsg("bm25 function output row count mismatch, expected %d, got %d", rowCount, len(outputSparseArray.GetContents()))
 	}
 
 	fieldData := &storage.SparseFloatVectorFieldData{
@@ -612,7 +609,7 @@ func buildSparseFloatVectorArrowArray(field *schemapb.FieldSchema, outputSparseA
 
 func buildArrowArrayFromFieldData(field *schemapb.FieldSchema, fieldData storage.FieldData, rowCount int) (arrow.Array, error) {
 	if fieldData.RowNum() != rowCount {
-		return nil, errors.Newf("function output row count mismatch for field %d, expected %d, got %d", field.GetFieldID(), rowCount, fieldData.RowNum())
+		return nil, merr.WrapErrFunctionFailedMsg("function output row count mismatch for field %d, expected %d, got %d", field.GetFieldID(), rowCount, fieldData.RowNum())
 	}
 
 	outputSchema := &schemapb.CollectionSchema{Fields: []*schemapb.FieldSchema{field}}

internal/datanode/compactor/segment_writer.go

@@ -22,7 +22,6 @@ import (
 	"math"
 
 	"github.com/apache/arrow/go/v17/arrow/array"
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.uber.org/atomic"
 	"go.uber.org/zap"
@@ -37,6 +36,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/etcdpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -318,7 +318,7 @@ func (w *SegmentWriter) WriteRecord(r storage.Record) error {
 		for fieldID, stats := range w.bm25Stats {
 			field, ok := r.Column(fieldID).(*array.Binary)
 			if !ok {
-				return errors.New("bm25 field value not found")
+				return merr.WrapErrServiceInternalMsg("bm25 field value not found")
 			}
 			stats.AppendBytes(field.Value(i))
 		}
@@ -341,12 +341,12 @@ func (w *SegmentWriter) Write(v *storage.Value) error {
 	for fieldID, stats := range w.bm25Stats {
 		data, ok := v.Value.(map[storage.FieldID]interface{})[fieldID]
 		if !ok {
-			return errors.New("bm25 field value not found")
+			return merr.WrapErrServiceInternalMsg("bm25 field value not found")
 		}
 
 		bytes, ok := data.([]byte)
 		if !ok {
-			return errors.New("bm25 field value not sparse bytes")
+			return merr.WrapErrServiceInternalMsg("bm25 field value not sparse bytes")
 		}
 		stats.AppendBytes(bytes)
 	}

internal/datanode/compactor/sort_compaction.go

@@ -23,7 +23,6 @@ import (
 	"time"
 
 	"github.com/apache/arrow/go/v17/arrow/array"
-	"github.com/cockroachdb/errors"
 	"github.com/samber/lo"
 	"go.opentelemetry.io/otel"
 	"go.uber.org/zap"
@@ -109,7 +108,9 @@ func (t *sortCompactionTask) preCompact() error {
 	}
 
 	if len(t.plan.GetSegmentBinlogs()) != 1 {
-		return errors.Newf("sort compaction should handle exactly one segment, but got %d segments, planID = %d",
+		// The plan is produced by datacoord, so a malformed plan is an internal
+		// protocol violation, not user input.
+		return merr.WrapErrServiceInternalMsg("sort compaction should handle exactly one segment, but got %d segments, planID = %d",
 			len(t.plan.GetSegmentBinlogs()), t.GetPlanID())
 	}
 

internal/datanode/data_node.go

@@ -27,7 +27,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/tidwall/gjson"
 	clientv3 "go.etcd.io/etcd/client/v3"
 	"go.uber.org/zap"
@@ -49,6 +48,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
 	"github.com/milvus-io/milvus/pkg/v3/util/expr"
 	"github.com/milvus-io/milvus/pkg/v3/util/lifetime"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metricsinfo"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
@@ -166,7 +166,7 @@ func (node *DataNode) Register() error {
 func (node *DataNode) initSession() error {
 	node.session = sessionutil.NewSession(node.ctx)
 	if node.session == nil {
-		return errors.New("failed to initialize session")
+		return merr.WrapErrServiceInternalMsg("failed to initialize session")
 	}
 	node.session.Init(typeutil.DataNodeRole, node.address, false)
 	sessionutil.SaveServerInfo(typeutil.DataNodeRole, node.session.ServerID)
@@ -282,7 +282,7 @@ func (node *DataNode) isHealthy() bool {
 // ReadyToFlush tells whether DataNode is ready for flushing
 func (node *DataNode) ReadyToFlush() error {
 	if !node.isHealthy() {
-		return errors.New("DataNode not in HEALTHY state")
+		return merr.Wrap(merr.ErrServiceNotReady, "DataNode not in HEALTHY state")
 	}
 	return nil
 }

internal/datanode/external/function_executor.go

@@ -20,6 +20,7 @@ import (
 	"github.com/milvus-io/milvus/internal/util/function/embedding"
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
 
@@ -65,11 +66,11 @@ func ExecuteFunctionsForSegment(
 	inputManifestPath, err := packed.CreateSegmentManifestWithBasePath(
 		ctx, basePath, format, sourceColumns, fragments, storageConfig)
 	if err != nil {
-		return "", fmt.Errorf("create input manifest: %w", err)
+		return "", merr.Wrap(err, "create input manifest")
 	}
 	_, inputVersion, err := packed.UnmarshalManifestPath(inputManifestPath)
 	if err != nil {
-		return "", fmt.Errorf("parse input manifest path: %w", err)
+		return "", merr.Wrap(err, "parse input manifest path")
 	}
 
 	outputFields, outputSchema, err := buildOutputSchema(schema)
@@ -94,7 +95,7 @@ func ExecuteFunctionsForSegment(
 	colGroups := []storagecommon.ColumnGroup{{Columns: lo.Range(len(outputFields))}}
 	writer, err := packed.NewFFIPackedWriter(basePath, outputArrow, colGroups, storageConfig, nil)
 	if err != nil {
-		return "", fmt.Errorf("open output writer: %w", err)
+		return "", merr.Wrap(err, "open output writer")
 	}
 	writer.AsNewColumnGroups()
 
@@ -108,7 +109,7 @@ func ExecuteFunctionsForSegment(
 
 	output, err := writer.Close()
 	if err != nil {
-		return "", fmt.Errorf("close output writer: %w", err)
+		return "", merr.Wrap(err, "close output writer")
 	}
 	if output != nil {
 		defer output.Destroy()
@@ -120,7 +121,7 @@ func ExecuteFunctionsForSegment(
 	}
 	manifestPath, err := packed.CommitManifestUpdates(basePath, inputVersion, storageConfig, updates)
 	if err != nil {
-		return "", fmt.Errorf("commit function output manifest: %w", err)
+		return "", merr.Wrap(err, "commit function output manifest")
 	}
 
 	log.Info("function execution completed",
@@ -139,7 +140,7 @@ func buildOutputSchema(schema *schemapb.CollectionSchema) ([]*schemapb.FieldSche
 		return nil, nil, err
 	}
 	if len(outputFields) == 0 {
-		return nil, nil, fmt.Errorf("no function output fields; executor should not have been invoked")
+		return nil, nil, merr.WrapErrServiceInternalMsg("no function output fields; executor should not have been invoked")
 	}
 	return outputFields, &schemapb.CollectionSchema{
 		Name:   schema.GetName(),
@@ -177,7 +178,7 @@ func functionOutputFields(schema *schemapb.CollectionSchema) ([]*schemapb.FieldS
 				continue
 			}
 			if _, ok := fieldsByID[fieldID]; !ok {
-				return nil, fmt.Errorf("function output field id %d not found in schema", fieldID)
+				return nil, merr.WrapErrParameterInvalidMsg("function output field id %d not found in schema", fieldID)
 			}
 			addOutputID(fieldID)
 		}
@@ -187,7 +188,7 @@ func functionOutputFields(schema *schemapb.CollectionSchema) ([]*schemapb.FieldS
 			}
 			field, ok := fieldsByName[fieldName]
 			if !ok {
-				return nil, fmt.Errorf("function output field %s not found in schema", fieldName)
+				return nil, merr.WrapErrParameterInvalidMsg("function output field %s not found in schema", fieldName)
 			}
 			addOutputID(field.GetFieldID())
 		}
@@ -222,7 +223,7 @@ func openInputReader(
 		}),
 	)
 	if err != nil {
-		return nil, fmt.Errorf("open input manifest: %w", err)
+		return nil, merr.Wrap(err, "open input manifest")
 	}
 	return reader, nil
 }
@@ -236,7 +237,7 @@ func buildFunctionExecutionSchema(
 	schema *schemapb.CollectionSchema,
 ) (*schemapb.CollectionSchema, *schemapb.CollectionSchema, typeutil.Set[int64], error) {
 	if schema == nil {
-		return nil, nil, nil, fmt.Errorf("collection schema is nil")
+		return nil, nil, nil, merr.WrapErrParameterInvalidMsg("collection schema is nil")
 	}
 	inputIDs := make(map[int64]struct{})
 	for _, fn := range schema.GetFunctions() {
@@ -291,24 +292,24 @@ func buildFunctionExecutionSchema(
 		inputSchema.Fields = append(inputSchema.Fields, f)
 
 		if schema.GetExternalSource() != "" && f.GetExternalField() == "" {
-			return nil, nil, nil, fmt.Errorf("function input field %s has no external_field", f.GetName())
+			return nil, nil, nil, merr.WrapErrParameterInvalidMsg("function input field %s has no external_field", f.GetName())
 		}
 	}
 
 	for inputID := range inputIDs {
 		if _, ok := seenInputFields[inputID]; !ok {
 			if _, ok := seenExecutionFields[inputID]; !ok {
-				return nil, nil, nil, fmt.Errorf("function input field id %d not found in schema", inputID)
+				return nil, nil, nil, merr.WrapErrParameterInvalidMsg("function input field id %d not found in schema", inputID)
 			}
 		}
 	}
 	for outputID := range outputIDs {
 		if _, ok := seenExecutionFields[outputID]; !ok {
-			return nil, nil, nil, fmt.Errorf("function output field id %d not found in schema", outputID)
+			return nil, nil, nil, merr.WrapErrParameterInvalidMsg("function output field id %d not found in schema", outputID)
 		}
 	}
 	if len(inputSchema.GetFields()) == 0 {
-		return nil, nil, nil, fmt.Errorf("no source input columns for function execution")
+		return nil, nil, nil, merr.WrapErrParameterInvalidMsg("no source input columns for function execution")
 	}
 	return inputSchema, executionSchema, requiredInputFields, nil
 }
@@ -332,7 +333,7 @@ func streamBatches(
 			break
 		}
 		if err != nil {
-			return totalRows, fmt.Errorf("read input batch: %w", err)
+			return totalRows, merr.Wrap(err, "read input batch")
 		}
 		if rec == nil {
 			break
@@ -341,7 +342,7 @@ func streamBatches(
 		batch, err := storage.RecordToInsertData(rec, executionSchema, requiredInputFields)
 		rec.Release()
 		if err != nil {
-			return totalRows, fmt.Errorf("record to InsertData: %w", err)
+			return totalRows, merr.Wrap(err, "record to InsertData")
 		}
 		if batch.GetRowNum() == 0 {
 			continue
@@ -351,7 +352,7 @@ func streamBatches(
 			ClusterID: clusterID,
 			DBName:    schema.GetDbName(),
 		}); err != nil {
-			return totalRows, fmt.Errorf("execute functions: %w", err)
+			return totalRows, merr.Wrap(err, "execute functions")
 		}
 
 		if err := accumulateBM25Stats(batch, bm25Acc); err != nil {
@@ -359,7 +360,7 @@ func streamBatches(
 		}
 
 		if err := writeOutputBatch(batch, outputSchema, outputArrow, writer); err != nil {
-			return totalRows, fmt.Errorf("write output batch: %w", err)
+			return totalRows, merr.Wrap(err, "write output batch")
 		}
 		totalRows += int64(batch.GetRowNum())
 	}
@@ -404,13 +405,13 @@ func accumulateBM25Stats(batch *storage.InsertData, acc map[int64]*storage.BM25S
 	for outID, stats := range acc {
 		raw, present := batch.Data[outID]
 		if !present || raw == nil {
-			return fmt.Errorf(
+			return merr.WrapErrFunctionFailedMsg(
 				"BM25 output field %d missing from batch; executeFunctions did not populate it",
 				outID)
 		}
 		fd, ok := raw.(*storage.SparseFloatVectorFieldData)
 		if !ok {
-			return fmt.Errorf(
+			return merr.WrapErrFunctionFailedMsg(
 				"BM25 output field %d has wrong type %T (want *SparseFloatVectorFieldData)",
 				outID, raw)
 		}
@@ -431,18 +432,18 @@ func appendBM25Stats(
 	}
 	log := log.Ctx(ctx)
 	if updates == nil {
-		return fmt.Errorf("manifest updates is nil")
+		return merr.WrapErrServiceInternalMsg("manifest updates is nil")
 	}
 
 	entries := make([]packed.StatEntry, 0, len(acc))
 	for outID, stats := range acc {
 		blob, err := stats.Serialize()
 		if err != nil {
-			return fmt.Errorf("serialize bm25 stats for field %d: %w", outID, err)
+			return merr.Wrapf(err, "serialize bm25 stats for field %d", outID)
 		}
 		fullPath := path.Join(basePath, fmt.Sprintf("_stats/bm25.%d/%d", outID, 0))
 		if err := packed.WriteFile(storageConfig, fullPath, blob); err != nil {
-			return fmt.Errorf("write bm25 stats file %s: %w", fullPath, err)
+			return merr.Wrapf(err, "write bm25 stats file %s", fullPath)
 		}
 		entries = append(entries, packed.StatEntry{
 			Key:   fmt.Sprintf("bm25.%d", outID),
@@ -481,7 +482,7 @@ func finalizeBM25Stats(
 
 	basePath, version, err := packed.UnmarshalManifestPath(manifestPath)
 	if err != nil {
-		return "", fmt.Errorf("parse manifest path: %w", err)
+		return "", merr.Wrap(err, "parse manifest path")
 	}
 
 	updates := &packed.ManifestUpdates{}

internal/datanode/external/manager.go

@@ -29,6 +29,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // TaskKey uniquely identifies an external collection task.
@@ -259,7 +260,8 @@ func (m *ExternalCollectionManager) SubmitTask(
 					zap.ByteString("stack", stack))
 				reason := fmt.Sprintf("task panicked: %v", r)
 				m.UpdateResult(clusterID, taskID, indexpb.JobState_JobStateFailed, reason, info.KeptSegments, nil)
-				retErr = fmt.Errorf("%s", reason)
+				// A recovered panic is a server-side failure, never caller input.
+				retErr = merr.WrapErrServiceInternalMsg("%s", reason)
 			}
 		}()
 		log.Info("executing external collection task in pool",

internal/datanode/external/task_update.go

@@ -164,22 +164,22 @@ func (t *RefreshExternalCollectionTask) PreExecute(ctx context.Context) error {
 		zap.Int64("collectionID", t.req.GetCollectionID()))
 
 	if t.req == nil {
-		return fmt.Errorf("request is nil")
+		return merr.WrapErrParameterInvalidMsg("request is nil")
 	}
 	if t.req.GetSchema() == nil {
-		return fmt.Errorf("schema is nil in request")
+		return merr.WrapErrParameterInvalidMsg("schema is nil in request")
 	}
 	if t.req.GetStorageConfig() == nil {
-		return fmt.Errorf("storage config is nil in request")
+		return merr.WrapErrParameterInvalidMsg("storage config is nil in request")
 	}
 	if t.req.GetExternalSource() == "" {
-		return fmt.Errorf("external source is empty in request")
+		return merr.WrapErrParameterInvalidMsg("external source is empty in request")
 	}
 
 	// Parse and cache external spec for reuse during Execute
 	spec, err := externalspec.ParseExternalSpec(t.req.GetExternalSpec())
 	if err != nil {
-		return fmt.Errorf("failed to parse external spec: %w", err)
+		return merr.Wrap(err, "failed to parse external spec")
 	}
 	t.parsedSpec = spec
 	t.columns = packed.GetColumnNamesFromSchema(t.req.GetSchema())
@@ -198,7 +198,7 @@ func (t *RefreshExternalCollectionTask) Execute(ctx context.Context) error {
 
 	// Initialize pre-allocated segment IDs from request
 	if t.req.GetPreAllocatedSegmentIds() == nil {
-		return fmt.Errorf("pre-allocated segment IDs not provided in request")
+		return merr.WrapErrParameterInvalidMsg("pre-allocated segment IDs not provided in request")
 	}
 
 	t.preallocatedIDRange = t.req.GetPreAllocatedSegmentIds()
@@ -212,13 +212,13 @@ func (t *RefreshExternalCollectionTask) Execute(ctx context.Context) error {
 	// Fetch fragments from external source
 	newFragments, err := t.fetchFragmentsFromExternalSource(ctx)
 	if err != nil {
-		return fmt.Errorf("failed to fetch fragments: %w", err)
+		return merr.Wrap(err, "failed to fetch fragments")
 	}
 
 	// Build current segment -> fragments mapping
 	currentSegmentFragments, err := t.buildCurrentSegmentFragments()
 	if err != nil {
-		return fmt.Errorf("failed to build current segment fragments: %w", err)
+		return merr.Wrap(err, "failed to build current segment fragments")
 	}
 
 	// Compare and organize segments
@@ -235,7 +235,7 @@ func (t *RefreshExternalCollectionTask) Execute(ctx context.Context) error {
 func (t *RefreshExternalCollectionTask) fetchFragmentsFromExternalSource(ctx context.Context) ([]packed.Fragment, error) {
 	manifestPath := t.req.GetExploreManifestPath()
 	if manifestPath == "" {
-		return nil, fmt.Errorf("explore manifest path is required but not provided")
+		return nil, merr.WrapErrParameterMissingMsg("explore manifest path is required but not provided")
 	}
 
 	log.Ctx(ctx).Info("reading file list from explore manifest",
@@ -329,7 +329,7 @@ func (t *RefreshExternalCollectionTask) organizeSegments(
 		var err error
 		outputColumns, err = functionOutputColumnNames(t.req.GetSchema())
 		if err != nil {
-			return nil, fmt.Errorf("resolve function output columns: %w", err)
+			return nil, merr.Wrap(err, "resolve function output columns")
 		}
 	}
 
@@ -455,7 +455,7 @@ func (t *RefreshExternalCollectionTask) segmentHasFunctionOutputColumns(seg *dat
 	}
 	hasColumns, err := packed.ManifestHasColumns(seg.GetManifestPath(), t.req.GetStorageConfig(), outputColumns)
 	if err != nil {
-		return false, fmt.Errorf("check function output columns for segment %d: %w", seg.GetID(), err)
+		return false, merr.Wrapf(err, "check function output columns for segment %d", seg.GetID())
 	}
 	return hasColumns, nil
 }
@@ -557,7 +557,7 @@ func (t *RefreshExternalCollectionTask) patchSegmentForMissingColumns(
 		t.req.GetStorageConfig(),
 	)
 	if err != nil {
-		return nil, fmt.Errorf("failed to append manifest columns for segment %d: %w", seg.GetID(), err)
+		return nil, merr.Wrapf(err, "failed to append manifest columns for segment %d", seg.GetID())
 	}
 
 	sampleRows := paramtable.Get().QueryNodeCfg.ExternalCollectionSampleRows.GetAsInt()
@@ -574,7 +574,7 @@ func (t *RefreshExternalCollectionTask) patchSegmentForMissingColumns(
 		t.req.GetStorageConfig(),
 	)
 	if err != nil {
-		return nil, fmt.Errorf("failed to sample external field sizes for segment %d: %w", seg.GetID(), err)
+		return nil, merr.Wrapf(err, "failed to sample external field sizes for segment %d", seg.GetID())
 	}
 	externalAvgBytes := sumFieldSizes(fieldSizes, schema)
 	if externalAvgBytes <= 0 {
@@ -690,7 +690,7 @@ func (t *RefreshExternalCollectionTask) balanceFragmentsToSegments(ctx context.C
 		}
 		// Each segment needs 2 IDs: one for segment, one for fake binlog logID
 		if t.nextAllocID+1 >= t.preallocatedIDRange.End {
-			return nil, fmt.Errorf("insufficient pre-allocated IDs: need 2 more but only have %d IDs in range [%d, %d)",
+			return nil, merr.WrapErrParameterInvalidMsg("insufficient pre-allocated IDs: need 2 more but only have %d IDs in range [%d, %d)",
 				t.preallocatedIDRange.End-t.nextAllocID,
 				t.preallocatedIDRange.Begin,
 				t.preallocatedIDRange.End)
@@ -739,7 +739,7 @@ func (t *RefreshExternalCollectionTask) balanceFragmentsToSegments(ctx context.C
 				manifestPath, err = t.createManifestForSegment(ctx, work.segmentID, work.fragments)
 			}
 			if err != nil {
-				return "", fmt.Errorf("failed to create manifest for segment %d: %w", work.segmentID, err)
+				return "", merr.Wrapf(err, "failed to create manifest for segment %d", work.segmentID)
 			}
 			return manifestPath, nil
 		})
@@ -834,7 +834,7 @@ func (t *RefreshExternalCollectionTask) balanceFragmentsToSegments(ctx context.C
 			log.Warn("external field size sample produced non-positive total",
 				zap.String("manifestPath", manifestPath),
 				zap.Int64("total", total))
-			recordErr(fmt.Errorf("sampled field sizes sum to %d (schema may have no external_field mappings, or sampled rows are empty)", total))
+			recordErr(merr.WrapErrParameterInvalidMsg("sampled field sizes sum to %d (schema may have no external_field mappings, or sampled rows are empty)", total))
 			return 0, false
 		}
 		return total, true
@@ -882,9 +882,9 @@ func (t *RefreshExternalCollectionTask) balanceFragmentsToSegments(ctx context.C
 			if strings.Contains(rootCause, "not found in schema") {
 				hint = "; check external_field mappings in collection schema against actual parquet columns"
 			}
-			return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf(
+			return nil, merr.WrapErrParameterInvalidMsg(
 				"external field size sampling failed for all %d segment(s): %s%s",
-				len(manifestPaths), rootCause, hint))
+				len(manifestPaths), rootCause, hint)
 		}
 		// Fill any zero slots (sampling failed mid-loop) with the first
 		// successful average so every segment gets a non-zero MemorySize.
@@ -1032,7 +1032,7 @@ func estimateFunctionOutputBytesPerRow(schema *schemapb.CollectionSchema) (int64
 			Fields: []*schemapb.FieldSchema{field},
 		})
 		if err != nil {
-			return 0, fmt.Errorf("estimate function output field %s: %w", field.GetName(), err)
+			return 0, merr.Wrapf(err, "estimate function output field %s", field.GetName())
 		}
 		total += int64(size)
 	}

internal/datanode/importv2/copy_segment_utils.go

@@ -18,7 +18,6 @@ package importv2
 
 import (
 	"context"
-	"fmt"
 	"path"
 	"strconv"
 	"strings"
@@ -34,6 +33,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 )
 
@@ -114,12 +114,12 @@ func transformManifestPath(
 ) (string, error) {
 	basePath, version, err := packed.UnmarshalManifestPath(manifestPath)
 	if err != nil {
-		return "", fmt.Errorf("failed to unmarshal manifest path: %w", err)
+		return "", merr.Wrap(err, "failed to unmarshal manifest path")
 	}
 
 	targetBasePath, err := generateTargetPath(basePath, source, target)
 	if err != nil {
-		return "", fmt.Errorf("failed to generate target base path: %w", err)
+		return "", merr.Wrap(err, "failed to generate target base path")
 	}
 
 	targetManifestPath := packed.MarshalManifestPath(targetBasePath, version)
@@ -227,18 +227,18 @@ func collectSegmentFiles(
 		// StorageV3+: binlog paths MUST come from manifest
 		manifestPath := source.GetManifestPath()
 		if manifestPath == "" {
-			return nil, fmt.Errorf("storage_version=%d requires manifest_path but it is empty (segmentID=%d)",
+			return nil, merr.WrapErrParameterInvalidMsg("storage_version=%d requires manifest_path but it is empty (segmentID=%d)",
 				source.GetStorageVersion(), source.GetSegmentId())
 		}
 
 		basePath, _, err := packed.UnmarshalManifestPath(manifestPath)
 		if err != nil {
-			return nil, fmt.Errorf("failed to unmarshal manifest path %q for segment %d: %w", manifestPath, source.GetSegmentId(), err)
+			return nil, merr.Wrapf(err, "failed to unmarshal manifest path %q for segment %d", manifestPath, source.GetSegmentId())
 		}
 
 		allFiles, listErr := listAllFiles(ctx, cm, basePath)
 		if listErr != nil {
-			return nil, fmt.Errorf("failed to list files from manifest base path %q for segment %d: %w", basePath, source.GetSegmentId(), listErr)
+			return nil, merr.Wrapf(listErr, "failed to list files from manifest base path %q for segment %d", basePath, source.GetSegmentId())
 		}
 
 		// Empty file list is OK for V3 — segment may have only deltas and no insert binlogs
@@ -319,7 +319,7 @@ func generateMappingsFromFiles(
 			}
 
 			if err != nil {
-				return fmt.Errorf("failed to generate target path for %s file %s: %w", fileType, srcPath, err)
+				return merr.Wrapf(err, "failed to generate target path for %s file %s", fileType, srcPath)
 			}
 			mappings[srcPath] = dstPath
 		}
@@ -379,13 +379,13 @@ func CopySegmentAndIndexFiles(
 	// Step 1: Collect all files to copy
 	files, err := collectSegmentFiles(ctx, cm, source)
 	if err != nil {
-		return nil, nil, fmt.Errorf("failed to collect segment files: %w", err)
+		return nil, nil, merr.Wrap(err, "failed to collect segment files")
 	}
 
 	// Step 2: Generate src->dst mappings for file copying
 	mappings, err := generateMappingsFromFiles(files, source, target)
 	if err != nil {
-		return nil, nil, fmt.Errorf("failed to generate file mappings: %w", err)
+		return nil, nil, merr.Wrap(err, "failed to generate file mappings")
 	}
 
 	// Step 3: Execute all copy operations
@@ -400,7 +400,7 @@ func CopySegmentAndIndexFiles(
 			fields = append(fields, logFields...)
 			fields = append(fields, zap.String("src", src), zap.String("dst", dst), zap.Error(err))
 			log.Warn("failed to copy file", fields...)
-			return nil, copiedFiles, fmt.Errorf("failed to copy file from %s to %s: %w", src, dst, err)
+			return nil, copiedFiles, merr.Wrapf(err, "failed to copy file from %s to %s", src, dst)
 		}
 		copiedFiles = append(copiedFiles, dst)
 	}
@@ -418,7 +418,7 @@ func CopySegmentAndIndexFiles(
 			if _, exists := mappings[srcPath]; !exists {
 				dstPath, pathErr := generateTargetPath(srcPath, source, target)
 				if pathErr != nil {
-					return nil, copiedFiles, fmt.Errorf("failed to generate target path for pb insert binlog %s: %w", srcPath, pathErr)
+					return nil, copiedFiles, merr.Wrapf(pathErr, "failed to generate target path for pb insert binlog %s", srcPath)
 				}
 				mappings[srcPath] = dstPath
 			}
@@ -430,20 +430,20 @@ func CopySegmentAndIndexFiles(
 	// Step 4: Build index metadata from source
 	indexInfos, textIndexInfos, jsonKeyIndexInfos, err := buildIndexInfoFromSource(source, target, mappings)
 	if err != nil {
-		return nil, copiedFiles, fmt.Errorf("failed to build index info: %w", err)
+		return nil, copiedFiles, merr.Wrap(err, "failed to build index info")
 	}
 
 	// Step 5: Generate segment metadata with path mappings
 	segmentInfo, err := generateSegmentInfoFromSource(source, target, mappings)
 	if err != nil {
-		return nil, copiedFiles, fmt.Errorf("failed to generate segment info: %w", err)
+		return nil, copiedFiles, merr.Wrap(err, "failed to generate segment info")
 	}
 
 	// Step 6: Compress paths
 	err = binlog.CompressBinLogs(segmentInfo.GetBinlogs(), segmentInfo.GetStatslogs(),
 		segmentInfo.GetDeltalogs(), segmentInfo.GetBm25Logs())
 	if err != nil {
-		return nil, copiedFiles, fmt.Errorf("failed to compress binlog paths: %w", err)
+		return nil, copiedFiles, merr.Wrap(err, "failed to compress binlog paths")
 	}
 
 	for _, indexInfo := range indexInfos {
@@ -474,7 +474,7 @@ func CopySegmentAndIndexFiles(
 	if useManifest {
 		targetManifestPath, err := transformManifestPath(source.GetManifestPath(), source, target)
 		if err != nil {
-			return nil, copiedFiles, fmt.Errorf("failed to transform manifest path: %w", err)
+			return nil, copiedFiles, merr.Wrap(err, "failed to transform manifest path")
 		}
 		result.ManifestPath = targetManifestPath
 	}
@@ -526,7 +526,7 @@ func transformFieldBinlogs(
 				}
 				dstPath, ok := mappings[srcPath]
 				if !ok {
-					return nil, 0, fmt.Errorf("no mapping found for source path: %s", srcPath)
+					return nil, 0, merr.WrapErrServiceInternalMsg("no mapping found for source path: %s", srcPath)
 				}
 				dstBinlog.LogPath = dstPath
 			}
@@ -583,7 +583,7 @@ func generateSegmentInfoFromSource(
 	// Process insert binlogs (count rows)
 	binlogs, totalRows, err := transformFieldBinlogs(source.GetInsertBinlogs(), mappings, true, source.GetIsExternalCollection())
 	if err != nil {
-		return nil, fmt.Errorf("failed to transform insert binlogs: %w", err)
+		return nil, merr.Wrap(err, "failed to transform insert binlogs")
 	}
 	segmentInfo.Binlogs = binlogs
 	segmentInfo.ImportedRows = totalRows
@@ -591,21 +591,21 @@ func generateSegmentInfoFromSource(
 	// Process stats binlogs (no row counting)
 	statslogs, _, err := transformFieldBinlogs(source.GetStatsBinlogs(), mappings, false, false)
 	if err != nil {
-		return nil, fmt.Errorf("failed to transform stats binlogs: %w", err)
+		return nil, merr.Wrap(err, "failed to transform stats binlogs")
 	}
 	segmentInfo.Statslogs = statslogs
 
 	// Process delta binlogs (no row counting)
 	deltalogs, _, err := transformFieldBinlogs(source.GetDeltaBinlogs(), mappings, false, false)
 	if err != nil {
-		return nil, fmt.Errorf("failed to transform delta binlogs: %w", err)
+		return nil, merr.Wrap(err, "failed to transform delta binlogs")
 	}
 	segmentInfo.Deltalogs = deltalogs
 
 	// Process BM25 binlogs (no row counting)
 	bm25logs, _, err := transformFieldBinlogs(source.GetBm25Binlogs(), mappings, false, false)
 	if err != nil {
-		return nil, fmt.Errorf("failed to transform BM25 binlogs: %w", err)
+		return nil, merr.Wrap(err, "failed to transform BM25 binlogs")
 	}
 	segmentInfo.Bm25Logs = bm25logs
 
@@ -635,7 +635,7 @@ func generateTargetPath(sourcePath string, source *datapb.CopySegmentSource, tar
 	}
 
 	if logTypeIndex == -1 || logTypeIndex+3 >= len(parts) {
-		return "", fmt.Errorf("invalid binlog path structure: %s (expected log_type at a valid position)", sourcePath)
+		return "", merr.WrapErrParameterInvalidMsg("invalid binlog path structure: %s (expected log_type at a valid position)", sourcePath)
 	}
 
 	// Replace IDs in order: collectionID, partitionID, segmentID
@@ -667,7 +667,7 @@ func generateTargetLOBPath(sourcePath string, source *datapb.CopySegmentSource,
 	// Path: .../{insert_log}/{coll}/{part}/lobs/...
 	// Need at least logTypeIndex + 2 (coll and part) after insert_log
 	if logTypeIndex == -1 || logTypeIndex+2 >= len(parts) {
-		return "", fmt.Errorf("invalid LOB path structure: %s", sourcePath)
+		return "", merr.WrapErrParameterInvalidMsg("invalid LOB path structure: %s", sourcePath)
 	}
 
 	parts[logTypeIndex+1] = strconv.FormatInt(target.GetCollectionId(), 10)
@@ -709,7 +709,7 @@ func buildIndexInfoFromSource(
 		for _, srcPath := range srcIndex.GetIndexFilePaths() {
 			targetPath, ok := mappings[srcPath]
 			if !ok {
-				return nil, nil, nil, fmt.Errorf("no mapping found for index file: %s", srcPath)
+				return nil, nil, nil, merr.WrapErrServiceInternalMsg("no mapping found for index file: %s", srcPath)
 			}
 			targetPaths = append(targetPaths, targetPath)
 		}
@@ -753,7 +753,7 @@ func buildIndexInfoFromSource(
 			for _, srcFile := range srcText.GetFiles() {
 				targetFile, ok := mappings[srcFile]
 				if !ok {
-					return nil, nil, nil, fmt.Errorf("no mapping found for text index file: %s", srcFile)
+					return nil, nil, nil, merr.WrapErrServiceInternalMsg("no mapping found for text index file: %s", srcFile)
 				}
 				targetFiles = append(targetFiles, targetFile)
 			}
@@ -785,7 +785,7 @@ func buildIndexInfoFromSource(
 			for _, srcFile := range srcJSON.GetFiles() {
 				targetFile, ok := mappings[srcFile]
 				if !ok {
-					return nil, nil, nil, fmt.Errorf("no mapping found for JSON index file: %s", srcFile)
+					return nil, nil, nil, merr.WrapErrServiceInternalMsg("no mapping found for JSON index file: %s", srcFile)
 				}
 				targetFiles = append(targetFiles, targetFile)
 			}
@@ -896,7 +896,7 @@ func generateTargetIndexPath(
 	}
 
 	if keywordIdx == -1 {
-		return "", fmt.Errorf("keyword '%s' not found in path: %s", keyword, sourcePath)
+		return "", merr.WrapErrServiceInternalMsg("keyword '%s' not found in path: %s", keyword, sourcePath)
 	}
 
 	// Set offsets based on index type
@@ -928,13 +928,13 @@ func generateTargetIndexPath(
 		segmentOffset = 6
 		buildIDOffset = 2
 	default:
-		return "", fmt.Errorf("unsupported index type: %s (expected '%s', '%s', '%s', or '%s')",
+		return "", merr.WrapErrParameterInvalidMsg("unsupported index type: %s (expected '%s', '%s', '%s', or '%s')",
 			indexType, IndexTypeVectorScalarV0, IndexTypeText, IndexTypeJSONKey, IndexTypeJSONStats)
 	}
 
 	// Validate path structure has enough components
 	if keywordIdx+segmentOffset >= len(parts) {
-		return "", fmt.Errorf("invalid %s path structure: %s (expected '%s' with at least %d components after it)",
+		return "", merr.WrapErrParameterInvalidMsg("invalid %s path structure: %s (expected '%s' with at least %d components after it)",
 			indexType, sourcePath, indexType, segmentOffset+1)
 	}
 

internal/datanode/importv2/task_copy_segment.go

@@ -22,7 +22,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
@@ -30,6 +29,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/log"
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
 
@@ -382,7 +382,7 @@ func (t *CopySegmentTask) copySingleSegment(source *datapb.CopySegmentSource, ta
 		reason := "no insert/delete binlogs for segment"
 		log.Error(reason, logFields...)
 		t.manager.Update(t.GetTaskID(), UpdateState(datapb.ImportTaskStateV2_Failed), UpdateReason(reason))
-		return nil, errors.New(reason)
+		return nil, merr.WrapErrParameterInvalidMsg(reason)
 	}
 
 	// Step 2: Copy all segment files (binlogs + indexes) together

internal/datanode/importv2/task_preimport.go

@@ -35,6 +35,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/datapb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/internalpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/conc"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -192,9 +193,9 @@ func (t *PreImportTask) readFileStat(reader importutilv2.Reader, fileIdx int) er
 	}
 	maxSize := paramtable.Get().DataNodeCfg.MaxImportFileSizeInGB.GetAsFloat() * 1024 * 1024 * 1024
 	if fileSize > int64(maxSize) {
-		return errors.New(fmt.Sprintf(
+		return merr.WrapErrParameterInvalidMsg(
 			"The import file size has reached the maximum limit allowed for importing, "+
-				"fileSize=%d, maxSize=%d", fileSize, int64(maxSize)))
+				"fileSize=%d, maxSize=%d", fileSize, int64(maxSize))
 	}
 
 	totalRows := 0

internal/datanode/importv2/util.go

@@ -50,7 +50,7 @@ import (
 )
 
 func WrapTaskNotFoundError(taskID int64) error {
-	return merr.WrapErrImportFailed(fmt.Sprintf("cannot find import task with id %d", taskID))
+	return merr.WrapErrImportSysFailedMsg("cannot find import task with id %d", taskID)
 }
 
 func NewSyncTask(ctx context.Context,
@@ -151,7 +151,7 @@ func PickSegment(segments []*datapb.ImportRequestSegment, vchannel string, parti
 	})
 
 	if len(candidates) == 0 {
-		return 0, fmt.Errorf("no candidate segments found for channel %s and partition %d",
+		return 0, merr.WrapErrServiceInternalMsg("no candidate segments found for channel %s and partition %d",
 			vchannel, partitionID)
 	}
 
@@ -392,7 +392,7 @@ func AppendNullableDefaultFieldsData(schema *schemapb.CollectionSchema, data *st
 				}
 			}
 		default:
-			return fmt.Errorf("unexpected data type: %d, cannot be filled with default value", dataType)
+			return merr.WrapErrServiceInternalMsg("unexpected data type: %d, cannot be filled with default value", dataType)
 		}
 
 		if err != nil {
@@ -409,7 +409,7 @@ func FillDynamicData(schema *schemapb.CollectionSchema, data *storage.InsertData
 	}
 	dynamicField := typeutil.GetDynamicField(schema)
 	if dynamicField == nil {
-		return merr.WrapErrImportFailed("collection schema is illegal, enable_dynamic_field is true but the dynamic field doesn't exist")
+		return merr.WrapErrImportSysFailed("collection schema is illegal, enable_dynamic_field is true but the dynamic field doesn't exist")
 	}
 
 	tempData, ok := data.Data[dynamicField.GetFieldID()]

internal/datanode/index/scheduler.go

@@ -80,7 +80,7 @@ func (queue *IndexTaskQueue) addUnissuedTask(t Task) error {
 	defer queue.utLock.Unlock()
 
 	if queue.utFull() {
-		return errors.New("index task queue is full")
+		return merr.Wrap(merr.ErrServiceResourceInsufficient, "index task queue is full")
 	}
 	queue.unissuedTasks.PushBack(t)
 	select {

internal/datanode/index/task_stats.go

@@ -52,6 +52,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/indexpb"
 	"github.com/milvus-io/milvus/pkg/v3/proto/workerpb"
 	_ "github.com/milvus-io/milvus/pkg/v3/util/funcutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/metautil"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/timerecord"
@@ -292,7 +293,7 @@ func (st *statsTask) sort(ctx context.Context) ([]*datapb.FieldBinlog, error) {
 			newManifest, err := packed.AddStatsToManifest(
 				st.manifestPath, st.req.GetStorageConfig(), statEntries)
 			if err != nil {
-				return nil, fmt.Errorf("failed to add stats to manifest: %w", err)
+				return nil, merr.Wrap(err, "failed to add stats to manifest")
 			}
 			st.manifestPath = newManifest
 		}
@@ -496,7 +497,7 @@ func (st *statsTask) createTextIndex(ctx context.Context,
 		}
 		binlogs, ok := fieldBinlogs[fieldID]
 		if !ok && !enableNull {
-			return nil, fmt.Errorf("field binlog not found for field %d", fieldID)
+			return nil, merr.WrapErrServiceInternalMsg("field binlog not found for field %d", fieldID)
 		}
 		result := make([]string, 0, len(binlogs))
 		for _, binlog := range binlogs {
@@ -618,7 +619,7 @@ func (st *statsTask) createTextIndex(ctx context.Context,
 		newManifest, err := packed.AddStatsToManifest(
 			st.manifestPath, st.req.GetStorageConfig(), statEntries)
 		if err != nil {
-			return fmt.Errorf("failed to add text index stats to manifest: %w", err)
+			return merr.Wrap(err, "failed to add text index stats to manifest")
 		}
 		st.manifestPath = newManifest
 		// Dual-write: keep textIndexLogs populated so it is stored in segment metadata as a
@@ -685,7 +686,7 @@ func (st *statsTask) createJSONKeyStats(ctx context.Context,
 		}
 		binlogs, ok := fieldBinlogs[fieldID]
 		if !ok && !enableNull {
-			return nil, fmt.Errorf("field binlog not found for field %d", fieldID)
+			return nil, merr.WrapErrServiceInternalMsg("field binlog not found for field %d", fieldID)
 		}
 		result := make([]string, 0, len(binlogs))
 		for _, binlog := range binlogs {
@@ -796,7 +797,7 @@ func (st *statsTask) createJSONKeyStats(ctx context.Context,
 		newManifest, err := packed.AddStatsToManifest(
 			st.manifestPath, st.req.GetStorageConfig(), statEntries)
 		if err != nil {
-			return fmt.Errorf("failed to add JSON key stats to manifest: %w", err)
+			return merr.Wrap(err, "failed to add JSON key stats to manifest")
 		}
 		st.manifestPath = newManifest
 		// Dual-write: keep jsonKeyIndexStats populated so it is stored in segment metadata as a
@@ -829,7 +830,7 @@ func computeStatsBasePath(req *workerpb.CreateStatsRequest, manifestPath string,
 	if req.GetStorageVersion() == storage.StorageV3 {
 		basePath, _, err := packed.UnmarshalManifestPath(manifestPath)
 		if err != nil {
-			return "", fmt.Errorf("failed to unmarshal manifest path for %s basePath: %w", statsType, err)
+			return "", merr.Wrapf(err, "failed to unmarshal manifest path for %s basePath", statsType)
 		}
 		return fmt.Sprintf("%s/_stats/%s.%d", basePath, statsType, fieldID), nil
 	}
@@ -845,7 +846,7 @@ func computeStatsBasePath(req *workerpb.CreateStatsRequest, manifestPath string,
 			req.GetTaskID(), req.GetTaskVersion(),
 			req.GetCollectionID(), req.GetPartitionID(), req.GetTargetSegmentID(), fieldID), nil
 	}
-	return "", fmt.Errorf("unknown stats type: %s", statsType)
+	return "", merr.WrapErrParameterInvalidMsg("unknown stats type: %s", statsType)
 }
 
 func buildIndexParams(

internal/datanode/index_services.go

@@ -21,7 +21,6 @@ import (
 	"fmt"
 	"strconv"
 
-	"github.com/cockroachdb/errors"
 	"go.opentelemetry.io/otel"
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/trace"
@@ -82,7 +81,9 @@ func (node *DataNode) CreateJob(ctx context.Context, req *workerpb.CreateJobRequ
 		Cancel: taskCancel,
 		State:  commonpb.IndexState_InProgress,
 	}); oldInfo != nil {
-		err := merr.WrapErrIndexDuplicate(req.GetIndexName(), "building index task existed")
+		// Node-internal dedup of a coordinator-dispatched build task: a duplicate
+		// is a scheduling race, not the user re-creating an index.
+		err := merr.WrapErrServiceInternalMsg("building index task existed, index=%s, buildID=%d", req.GetIndexName(), req.GetBuildID())
 		log.Warn("duplicated index build task", zap.Error(err))
 		metrics.DataNodeBuildIndexTaskCounter.WithLabelValues(paramtable.GetStringNodeID(), metrics.FailLabel).Inc()
 		return merr.Status(err), nil
@@ -254,7 +255,7 @@ func (node *DataNode) CreateJobV2(ctx context.Context, req *workerpb.CreateJobV2
 		return node.createStatsTask(ctx, statsRequest)
 	default:
 		log.Warn("DataNode receive unknown type job")
-		return merr.Status(fmt.Errorf("DataNode receive unknown type job with TaskID: %d", req.GetTaskID())), nil
+		return merr.Status(merr.WrapErrServiceInternalMsg("DataNode receive unknown type job with TaskID: %d", req.GetTaskID())), nil
 	}
 }
 
@@ -454,7 +455,7 @@ func (node *DataNode) QueryJobsV2(ctx context.Context, req *workerpb.QueryJobsV2
 	default:
 		log.Warn("DataNode receive querying unknown type jobs")
 		return &workerpb.QueryJobsV2Response{
-			Status: merr.Status(errors.New("DataNode receive querying unknown type jobs")),
+			Status: merr.Status(merr.WrapErrServiceInternalMsg("DataNode receive querying unknown type jobs")),
 		}, nil
 	}
 }
@@ -479,7 +480,7 @@ func (node *DataNode) queryIndexTask(ctx context.Context, req *workerpb.QueryJob
 	log.Debug("query index jobs result success", zap.Any("results", results))
 	if len(results) == 0 {
 		return &workerpb.QueryJobsV2Response{
-			Status: merr.Status(fmt.Errorf("tasks '%v' not found", req.GetTaskIDs())),
+			Status: merr.Status(merr.WrapErrServiceInternalMsg("tasks '%v' not found", req.GetTaskIDs())),
 		}, nil
 	}
 	return &workerpb.QueryJobsV2Response{
@@ -508,7 +509,7 @@ func (node *DataNode) queryStatsTask(ctx context.Context, req *workerpb.QueryJob
 	log.Debug("query stats job result success", zap.Any("results", results))
 	if len(results) == 0 {
 		return &workerpb.QueryJobsV2Response{
-			Status: merr.Status(fmt.Errorf("tasks '%v' not found", req.GetTaskIDs())),
+			Status: merr.Status(merr.WrapErrServiceInternalMsg("tasks '%v' not found", req.GetTaskIDs())),
 		}, nil
 	}
 	return &workerpb.QueryJobsV2Response{
@@ -542,7 +543,7 @@ func (node *DataNode) queryAnalyzeTask(ctx context.Context, req *workerpb.QueryJ
 	log.Debug("query analyze jobs result success", zap.Any("results", results))
 	if len(results) == 0 {
 		return &workerpb.QueryJobsV2Response{
-			Status: merr.Status(fmt.Errorf("tasks '%v' not found", req.GetTaskIDs())),
+			Status: merr.Status(merr.WrapErrServiceInternalMsg("tasks '%v' not found", req.GetTaskIDs())),
 		}, nil
 	}
 	return &workerpb.QueryJobsV2Response{
@@ -613,6 +614,6 @@ func (node *DataNode) DropJobsV2(ctx context.Context, req *workerpb.DropJobsV2Re
 		return merr.Success(), nil
 	default:
 		log.Warn("DataNode receive dropping unknown type jobs")
-		return merr.Status(errors.New("DataNode receive dropping unknown type jobs")), nil
+		return merr.Status(merr.WrapErrServiceInternalMsg("DataNode receive dropping unknown type jobs")), nil
 	}
 }

internal/datanode/services.go

@@ -194,11 +194,11 @@ func (node *DataNode) CompactionV2(ctx context.Context, req *datapb.CompactionPl
 	}
 
 	if req.GetBeginLogID() == 0 {
-		return merr.Status(merr.WrapErrParameterInvalidMsg("invalid beginLogID")), nil
+		return merr.Status(merr.WrapErrServiceInternalMsg("invalid beginLogID")), nil
 	}
 
 	if req.GetPreAllocatedLogIDs().GetBegin() == 0 || req.GetPreAllocatedLogIDs().GetEnd() == 0 {
-		return merr.Status(merr.WrapErrParameterInvalidMsg(fmt.Sprintf("invalid beginID %d or invalid endID %d", req.GetPreAllocatedLogIDs().GetBegin(), req.GetPreAllocatedLogIDs().GetEnd()))), nil
+		return merr.Status(merr.WrapErrServiceInternalMsg(fmt.Sprintf("invalid beginID %d or invalid endID %d", req.GetPreAllocatedLogIDs().GetBegin(), req.GetPreAllocatedLogIDs().GetEnd()))), nil
 	}
 
 	/*
@@ -233,7 +233,7 @@ func (node *DataNode) CompactionV2(ctx context.Context, req *datapb.CompactionPl
 		)
 	case datapb.CompactionType_MixCompaction:
 		if req.GetPreAllocatedSegmentIDs() == nil || req.GetPreAllocatedSegmentIDs().GetBegin() == 0 {
-			return merr.Status(merr.WrapErrParameterInvalidMsg("invalid pre-allocated segmentID range")), nil
+			return merr.Status(merr.WrapErrServiceInternalMsg("invalid pre-allocated segmentID range")), nil
 		}
 		pk, err := typeutil.GetPrimaryFieldSchema(req.GetSchema())
 		if err != nil {
@@ -256,7 +256,7 @@ func (node *DataNode) CompactionV2(ctx context.Context, req *datapb.CompactionPl
 		)
 	case datapb.CompactionType_ClusteringCompaction:
 		if req.GetPreAllocatedSegmentIDs() == nil || req.GetPreAllocatedSegmentIDs().GetBegin() == 0 {
-			return merr.Status(merr.WrapErrParameterInvalidMsg("invalid pre-allocated segmentID range")), nil
+			return merr.Status(merr.WrapErrServiceInternalMsg("invalid pre-allocated segmentID range")), nil
 		}
 		if namespaceEnabled {
 			var sortFields []int64
@@ -281,7 +281,7 @@ func (node *DataNode) CompactionV2(ctx context.Context, req *datapb.CompactionPl
 		}
 	case datapb.CompactionType_SortCompaction:
 		if req.GetPreAllocatedSegmentIDs() == nil || req.GetPreAllocatedSegmentIDs().GetBegin() == 0 {
-			return merr.Status(merr.WrapErrParameterInvalidMsg("invalid pre-allocated segmentID range")), nil
+			return merr.Status(merr.WrapErrServiceInternalMsg("invalid pre-allocated segmentID range")), nil
 		}
 		pk, err := typeutil.GetPrimaryFieldSchema(req.GetSchema())
 		if err != nil {
@@ -306,7 +306,7 @@ func (node *DataNode) CompactionV2(ctx context.Context, req *datapb.CompactionPl
 		task = compactor.NewBumpSchemaVersionCompactionTask(taskCtx, cm, req, compactionParams)
 	default:
 		log.Warn("Unknown compaction type", zap.String("type", req.GetType().String()))
-		return merr.Status(merr.WrapErrParameterInvalidMsg("Unknown compaction type: %v", req.GetType().String())), nil
+		return merr.Status(merr.WrapErrServiceInternalMsg("Unknown compaction type: %v", req.GetType().String())), nil
 	}
 
 	succeed, err := node.compactionExecutor.Enqueue(task)
@@ -790,7 +790,7 @@ func (node *DataNode) CreateTask(ctx context.Context, request *workerpb.CreateTa
 		}
 		return node.CopySegment(ctx, req)
 	default:
-		err := fmt.Errorf("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
+		err := merr.WrapErrServiceInternalMsg("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
 		log.Ctx(ctx).Warn("CreateTask failed", zap.Error(err))
 		return merr.Status(err), nil
 	}
@@ -807,7 +807,7 @@ func wrapQueryTaskResult[Resp proto.Message](resp Resp, properties taskcommon.Pr
 	}
 	statusResp, ok := any(resp).(ResponseWithStatus)
 	if !ok {
-		return &workerpb.QueryTaskResponse{Status: merr.Status(fmt.Errorf("response does not implement GetStatus"))}, nil
+		return &workerpb.QueryTaskResponse{Status: merr.Status(merr.WrapErrServiceInternalMsg("response does not implement GetStatus"))}, nil
 	}
 	return &workerpb.QueryTaskResponse{
 		Status:     statusResp.GetStatus(),
@@ -937,7 +937,7 @@ func (node *DataNode) QueryTask(ctx context.Context, request *workerpb.QueryTask
 		resProperties.AppendReason(resp.GetReason())
 		return wrapQueryTaskResult(resp, resProperties)
 	default:
-		err := fmt.Errorf("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
+		err := merr.WrapErrServiceInternalMsg("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
 		log.Ctx(ctx).Warn("QueryTask failed", zap.Error(err))
 		return &workerpb.QueryTaskResponse{
 			Status: merr.Status(err),
@@ -997,7 +997,7 @@ func (node *DataNode) DropTask(ctx context.Context, request *workerpb.DropTaskRe
 			zap.String("clusterID", clusterID))
 		return merr.Success(), nil
 	default:
-		err := fmt.Errorf("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
+		err := merr.WrapErrServiceInternalMsg("unrecognized task type '%s', properties=%v", taskType, request.GetProperties())
 		log.Ctx(ctx).Warn("DropTask failed", zap.Error(err))
 		return merr.Status(err), nil
 	}

internal/datanode/services_test.go

@@ -594,7 +594,10 @@ func (s *DataNodeServicesSuite) TestCreateTask() {
 		}
 		status, err := s.node.CreateTask(s.ctx, req)
 		s.NoError(err)
-		s.Equal(commonpb.ErrorCode_UnexpectedError, status.GetErrorCode())
+		// taskcommon.GetTaskType classifies an unrecognized task type as
+		// ServiceInternal: task types are coordinator-assigned, so a mismatch is
+		// an internal protocol violation, not user input.
+		s.Equal(merr.Code(merr.ErrServiceInternal), status.GetCode())
 	})
 }
 
@@ -685,7 +688,10 @@ func (s *DataNodeServicesSuite) TestQueryTask() {
 		}
 		resp, err := s.node.QueryTask(s.ctx, req)
 		s.NoError(err)
-		s.Equal(commonpb.ErrorCode_UnexpectedError, resp.GetStatus().GetErrorCode())
+		// taskcommon.GetTaskType classifies an unrecognized task type as
+		// ServiceInternal: task types are coordinator-assigned, so a mismatch is
+		// an internal protocol violation, not user input.
+		s.Equal(merr.Code(merr.ErrServiceInternal), resp.GetStatus().GetCode())
 	})
 }
 
@@ -771,7 +777,10 @@ func (s *DataNodeServicesSuite) TestDropTask() {
 		}
 		status, err := s.node.DropTask(s.ctx, req)
 		s.NoError(err)
-		s.Equal(commonpb.ErrorCode_UnexpectedError, status.GetErrorCode())
+		// taskcommon.GetTaskType classifies an unrecognized task type as
+		// ServiceInternal: task types are coordinator-assigned, so a mismatch is
+		// an internal protocol violation, not user input.
+		s.Equal(merr.Code(merr.ErrServiceInternal), status.GetCode())
 	})
 }
 

internal/distributed/datanode/client/client.go

@@ -20,7 +20,6 @@ import (
 	"context"
 	"fmt"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 	"google.golang.org/grpc"
 
@@ -36,6 +35,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/workerpb"
 	"github.com/milvus-io/milvus/pkg/v3/util/commonpbutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/funcutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -57,11 +57,11 @@ type Client struct {
 // NewClient creates a client for DataNode.
 func NewClient(ctx context.Context, addr string, serverID int64, encryption bool) (types.DataNodeClient, error) {
 	if addr == "" {
-		return nil, errors.New("address is empty")
+		return nil, merr.WrapErrParameterInvalidMsg("address is empty")
 	}
 	sess := sessionutil.NewSession(context.Background())
 	if sess == nil {
-		err := errors.New("new session error, maybe can not connect to etcd")
+		err := merr.WrapErrServiceUnavailable("new session error, maybe can not connect to etcd")
 		log.Ctx(ctx).Debug("DataNodeClient New Etcd Session failed", zap.Error(err))
 		return nil, err
 	}

internal/distributed/mixcoord/client/client.go

@@ -69,7 +69,7 @@ type Client struct {
 func NewClient(ctx context.Context) (types.MixCoordClient, error) {
 	sess := sessionutil.NewSession(context.Background())
 	if sess == nil {
-		err := errors.New("new session error, maybe can not connect to etcd")
+		err := merr.WrapErrServiceUnavailable("new session error, maybe can not connect to etcd")
 		log.Ctx(ctx).Debug("New MixCoord Client failed", zap.Error(err))
 		return nil, err
 	}
@@ -120,7 +120,7 @@ func (c *Client) getMixCoordAddr() (string, error) {
 			return c.getCompatibleMixCoordAddr()
 		} else {
 			log.Warn("MixCoordClient mess key not exist", zap.Any("key", key))
-			return "", errors.New("find no available mixcoord, check mixcoord state")
+			return "", merr.WrapErrNodeNotFound(0, "find no available mixcoord, check mixcoord state")
 		}
 	}
 	log.Debug("MixCoordClient GetSessions success",
@@ -137,12 +137,12 @@ func (c *Client) getCompatibleMixCoordAddr() (string, error) {
 	msess, _, err := c.sess.GetSessions(c.ctx, typeutil.RootCoordRole)
 	if err != nil {
 		log.Debug("mixCoordClient getSessions failed", zap.Any("key", typeutil.RootCoordRole), zap.Error(err))
-		return "", errors.New("find no available mixcoord, check mixcoord state")
+		return "", merr.WrapErrNodeNotFound(0, "find no available mixcoord, check mixcoord state")
 	}
 	ms, ok := msess[typeutil.RootCoordRole]
 	if !ok {
 		log.Warn("MixCoordClient mess key not exist", zap.Any("key", typeutil.RootCoordRole))
-		return "", errors.New("find no available mixcoord, check mixcoord state")
+		return "", merr.WrapErrNodeNotFound(0, "find no available mixcoord, check mixcoord state")
 	}
 	log.Debug("MixCoordClient GetSessions use rootCoord", zap.Any("key", typeutil.RootCoordRole))
 	c.grpcClient.SetNodeID(ms.ServerID)

internal/distributed/proxy/client/client.go

@@ -20,7 +20,6 @@ import (
 	"context"
 	"fmt"
 
-	"github.com/cockroachdb/errors"
 	"go.uber.org/zap"
 	"google.golang.org/grpc"
 
@@ -35,6 +34,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/v3/proto/proxypb"
 	"github.com/milvus-io/milvus/pkg/v3/util/commonpbutil"
 	"github.com/milvus-io/milvus/pkg/v3/util/funcutil"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/paramtable"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
@@ -50,11 +50,11 @@ type Client struct {
 // NewClient creates a new client instance
 func NewClient(ctx context.Context, addr string, nodeID int64) (types.ProxyClient, error) {
 	if addr == "" {
-		return nil, errors.New("address is empty")
+		return nil, merr.WrapErrParameterInvalidMsg("address is empty")
 	}
 	sess := sessionutil.NewSession(context.Background())
 	if sess == nil {
-		err := errors.New("new session error, maybe can not connect to etcd")
+		err := merr.WrapErrServiceUnavailable("new session error, maybe can not connect to etcd")
 		log.Ctx(ctx).Debug("Proxy client new session failed", zap.Error(err))
 		return nil, err
 	}

internal/distributed/proxy/httpserver/handler.go

@@ -17,13 +17,12 @@
 package httpserver
 
 import (
-	"fmt"
-
 	"github.com/gin-gonic/gin"
 	"google.golang.org/protobuf/proto"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/milvuspb"
 	"github.com/milvus-io/milvus/internal/types"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 // Handlers handles http requests
@@ -107,7 +106,7 @@ func (h *Handlers) handleDummy(c *gin.Context) (interface{}, error) {
 	// use ShouldBind to supports binding JSON, XML, YAML, and protobuf.
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.Dummy(c, &req)
 }
@@ -116,11 +115,11 @@ func (h *Handlers) handleCreateCollection(c *gin.Context) (interface{}, error) {
 	wrappedReq := WrappedCreateCollectionRequest{}
 	err := shouldBind(c, &wrappedReq)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	schemaProto, err := proto.Marshal(&wrappedReq.Schema)
 	if err != nil {
-		return nil, fmt.Errorf("%w: marshal schema failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "marshal schema failed")
 	}
 	req := &milvuspb.CreateCollectionRequest{
 		Base:             wrappedReq.Base,
@@ -138,7 +137,7 @@ func (h *Handlers) handleDropCollection(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DropCollectionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DropCollection(c, &req)
 }
@@ -147,7 +146,7 @@ func (h *Handlers) handleHasCollection(c *gin.Context) (interface{}, error) {
 	req := milvuspb.HasCollectionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.HasCollection(c, &req)
 }
@@ -156,7 +155,7 @@ func (h *Handlers) handleDescribeCollection(c *gin.Context) (interface{}, error)
 	req := milvuspb.DescribeCollectionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DescribeCollection(c, &req)
 }
@@ -165,7 +164,7 @@ func (h *Handlers) handleLoadCollection(c *gin.Context) (interface{}, error) {
 	req := milvuspb.LoadCollectionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.LoadCollection(c, &req)
 }
@@ -174,7 +173,7 @@ func (h *Handlers) handleReleaseCollection(c *gin.Context) (interface{}, error)
 	req := milvuspb.ReleaseCollectionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ReleaseCollection(c, &req)
 }
@@ -183,7 +182,7 @@ func (h *Handlers) handleGetCollectionStatistics(c *gin.Context) (interface{}, e
 	req := milvuspb.GetCollectionStatisticsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetCollectionStatistics(c, &req)
 }
@@ -192,7 +191,7 @@ func (h *Handlers) handleShowCollections(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ShowCollectionsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ShowCollections(c, &req)
 }
@@ -201,7 +200,7 @@ func (h *Handlers) handleCreatePartition(c *gin.Context) (interface{}, error) {
 	req := milvuspb.CreatePartitionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.CreatePartition(c, &req)
 }
@@ -210,7 +209,7 @@ func (h *Handlers) handleDropPartition(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DropPartitionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DropPartition(c, &req)
 }
@@ -219,7 +218,7 @@ func (h *Handlers) handleHasPartition(c *gin.Context) (interface{}, error) {
 	req := milvuspb.HasPartitionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.HasPartition(c, &req)
 }
@@ -228,7 +227,7 @@ func (h *Handlers) handleLoadPartitions(c *gin.Context) (interface{}, error) {
 	req := milvuspb.LoadPartitionsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.LoadPartitions(c, &req)
 }
@@ -237,7 +236,7 @@ func (h *Handlers) handleReleasePartitions(c *gin.Context) (interface{}, error)
 	req := milvuspb.ReleasePartitionsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ReleasePartitions(c, &req)
 }
@@ -246,7 +245,7 @@ func (h *Handlers) handleGetPartitionStatistics(c *gin.Context) (interface{}, er
 	req := milvuspb.GetPartitionStatisticsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetPartitionStatistics(c, &req)
 }
@@ -255,7 +254,7 @@ func (h *Handlers) handleShowPartitions(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ShowPartitionsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ShowPartitions(c, &req)
 }
@@ -264,7 +263,7 @@ func (h *Handlers) handleCreateAlias(c *gin.Context) (interface{}, error) {
 	req := milvuspb.CreateAliasRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.CreateAlias(c, &req)
 }
@@ -273,7 +272,7 @@ func (h *Handlers) handleDropAlias(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DropAliasRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DropAlias(c, &req)
 }
@@ -282,7 +281,7 @@ func (h *Handlers) handleAlterAlias(c *gin.Context) (interface{}, error) {
 	req := milvuspb.AlterAliasRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.AlterAlias(c, &req)
 }
@@ -291,7 +290,7 @@ func (h *Handlers) handleCreateIndex(c *gin.Context) (interface{}, error) {
 	req := milvuspb.CreateIndexRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.CreateIndex(c, &req)
 }
@@ -300,7 +299,7 @@ func (h *Handlers) handleDescribeIndex(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DescribeIndexRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DescribeIndex(c, &req)
 }
@@ -309,7 +308,7 @@ func (h *Handlers) handleGetIndexState(c *gin.Context) (interface{}, error) {
 	req := milvuspb.GetIndexStateRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetIndexState(c, &req)
 }
@@ -318,7 +317,7 @@ func (h *Handlers) handleGetIndexBuildProgress(c *gin.Context) (interface{}, err
 	req := milvuspb.GetIndexBuildProgressRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetIndexBuildProgress(c, &req)
 }
@@ -327,7 +326,7 @@ func (h *Handlers) handleDropIndex(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DropIndexRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DropIndex(c, &req)
 }
@@ -336,11 +335,11 @@ func (h *Handlers) handleInsert(c *gin.Context) (interface{}, error) {
 	wrappedReq := WrappedInsertRequest{}
 	err := shouldBind(c, &wrappedReq)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	req, err := wrappedReq.AsInsertRequest()
 	if err != nil {
-		return nil, fmt.Errorf("%w: convert body to pb failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "convert body to pb failed")
 	}
 	return h.proxy.Insert(c, req)
 }
@@ -349,7 +348,7 @@ func (h *Handlers) handleDelete(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DeleteRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.Delete(c, &req)
 }
@@ -358,10 +357,10 @@ func (h *Handlers) handleSearch(c *gin.Context) (interface{}, error) {
 	wrappedReq := SearchRequest{}
 	err := shouldBind(c, &wrappedReq)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	if wrappedReq.HasSearchAggregation() {
-		return nil, fmt.Errorf("%w: searchAggregation is not supported for low-level REST search", errBadRequest)
+		return nil, badRequestf(merr.WrapErrParameterInvalidMsg("searchAggregation is not supported for low-level REST search"), "invalid request")
 	}
 	req := milvuspb.SearchRequest{
 		Base:               wrappedReq.Base,
@@ -392,7 +391,7 @@ func (h *Handlers) handleQuery(c *gin.Context) (interface{}, error) {
 	req := milvuspb.QueryRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.Query(c, &req)
 }
@@ -401,7 +400,7 @@ func (h *Handlers) handleFlush(c *gin.Context) (interface{}, error) {
 	req := milvuspb.FlushRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.Flush(c, &req)
 }
@@ -410,7 +409,7 @@ func (h *Handlers) handleCalcDistance(c *gin.Context) (interface{}, error) {
 	wrappedReq := WrappedCalcDistanceRequest{}
 	err := shouldBind(c, &wrappedReq)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 
 	req := milvuspb.CalcDistanceRequest{
@@ -426,7 +425,7 @@ func (h *Handlers) handleGetFlushState(c *gin.Context) (interface{}, error) {
 	req := milvuspb.GetFlushStateRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetFlushState(c, &req)
 }
@@ -435,7 +434,7 @@ func (h *Handlers) handleGetPersistentSegmentInfo(c *gin.Context) (interface{},
 	req := milvuspb.GetPersistentSegmentInfoRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetPersistentSegmentInfo(c, &req)
 }
@@ -444,7 +443,7 @@ func (h *Handlers) handleGetQuerySegmentInfo(c *gin.Context) (interface{}, error
 	req := milvuspb.GetQuerySegmentInfoRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetQuerySegmentInfo(c, &req)
 }
@@ -453,7 +452,7 @@ func (h *Handlers) handleGetReplicas(c *gin.Context) (interface{}, error) {
 	req := milvuspb.GetReplicasRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetReplicas(c, &req)
 }
@@ -462,7 +461,7 @@ func (h *Handlers) handleGetMetrics(c *gin.Context) (interface{}, error) {
 	req := milvuspb.GetMetricsRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetMetrics(c, &req)
 }
@@ -471,7 +470,7 @@ func (h *Handlers) handleLoadBalance(c *gin.Context) (interface{}, error) {
 	req := milvuspb.LoadBalanceRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.LoadBalance(c, &req)
 }
@@ -480,7 +479,7 @@ func (h *Handlers) handleGetCompactionState(c *gin.Context) (interface{}, error)
 	req := milvuspb.GetCompactionStateRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetCompactionState(c, &req)
 }
@@ -489,7 +488,7 @@ func (h *Handlers) handleGetCompactionStateWithPlans(c *gin.Context) (interface{
 	req := milvuspb.GetCompactionPlansRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetCompactionStateWithPlans(c, &req)
 }
@@ -498,7 +497,7 @@ func (h *Handlers) handleManualCompaction(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ManualCompactionRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ManualCompaction(c, &req)
 }
@@ -507,7 +506,7 @@ func (h *Handlers) handleImport(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ImportRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.Import(c, &req)
 }
@@ -516,7 +515,7 @@ func (h *Handlers) handleGetImportState(c *gin.Context) (interface{}, error) {
 	req := milvuspb.GetImportStateRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.GetImportState(c, &req)
 }
@@ -525,7 +524,7 @@ func (h *Handlers) handleListImportTasks(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ListImportTasksRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ListImportTasks(c, &req)
 }
@@ -534,7 +533,7 @@ func (h *Handlers) handleCreateCredential(c *gin.Context) (interface{}, error) {
 	req := milvuspb.CreateCredentialRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.CreateCredential(c, &req)
 }
@@ -543,7 +542,7 @@ func (h *Handlers) handleUpdateCredential(c *gin.Context) (interface{}, error) {
 	req := milvuspb.UpdateCredentialRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.UpdateCredential(c, &req)
 }
@@ -552,7 +551,7 @@ func (h *Handlers) handleDeleteCredential(c *gin.Context) (interface{}, error) {
 	req := milvuspb.DeleteCredentialRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.DeleteCredential(c, &req)
 }
@@ -561,7 +560,7 @@ func (h *Handlers) handleListCredUsers(c *gin.Context) (interface{}, error) {
 	req := milvuspb.ListCredUsersRequest{}
 	err := shouldBind(c, &req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: parse body failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "parse body failed")
 	}
 	return h.proxy.ListCredUsers(c, &req)
 }

internal/distributed/proxy/httpserver/handler_v1_test.go

@@ -1434,7 +1434,7 @@ func TestFp16Bf16VectorsV1(t *testing.T) {
 			assert.Nil(t, err, "case %d: ", i)
 			assert.Equal(t, testcase.errCode, returnBody.Code, "case %d: ", i, string(testcase.requestBody))
 			if testcase.errCode != 0 {
-				assert.Equal(t, testcase.errMsg, returnBody.Message, "case %d: ", i, string(testcase.requestBody))
+				assert.Contains(t, returnBody.Message, testcase.errMsg, "case %d: ", i, string(testcase.requestBody))
 			}
 			fmt.Println(w.Body.String())
 		})

internal/distributed/proxy/httpserver/handler_v2.go

@@ -410,6 +410,35 @@ func wrapperPost(newReq newReqFunc, v2 handlerFuncV2) gin.HandlerFunc {
 			dbName,
 			collectionName,
 		).Inc()
+
+		// Mirror the fail_input/fail_system metric split into the logs so a
+		// failed REST request can be filtered by error_type. System failures are
+		// logged at Warn (actionable); input failures at Info (expected user
+		// mistakes — keeping them at Warn would spam the logs).
+		if label == metrics.FailSystemLabel || label == metrics.FailInputLabel {
+			var status *commonpb.Status
+			switch r := resp.(type) {
+			case interface{ GetStatus() *commonpb.Status }:
+				status = r.GetStatus()
+			case *commonpb.Status:
+				status = r
+			}
+			errType := merr.SystemError
+			if label == metrics.FailInputLabel {
+				errType = merr.InputError
+			}
+			logger := log.Ctx(ctx).With(
+				zap.String("method", methodTag),
+				zap.String("error_type", errType.String()),
+				zap.Int32("code", status.GetCode()),
+				zap.String("reason", status.GetReason()),
+			)
+			if errType == merr.InputError {
+				logger.Info("restful request returned an input error")
+			} else {
+				logger.Warn("restful request returned a system error")
+			}
+		}
 	}
 }
 
@@ -543,6 +572,9 @@ func wrapperProxyWithLimit(ctx context.Context, ginCtx *gin.Context, req any, ch
 	}
 
 	if err != nil {
+		// Expose the exact classification (incl. boundary InputError marks) to
+		// the REST access log; key must match accesslog/info.ContextErrorType.
+		ginCtx.Set("error_type", merr.GetErrorType(err).String())
 		log.Ctx(ctx).Warn("high level restful api, grpc call failed", zap.Error(err))
 		if !ignoreErr {
 			HTTPAbortReturn(ginCtx, http.StatusOK, gin.H{HTTPReturnCode: merr.Code(err), HTTPReturnMessage: err.Error()})
@@ -1025,7 +1057,7 @@ func (h *HandlersV2) waitForFlush(ctx context.Context, dbName string, collection
 	segmentIDs := flushResp.GetCollSegIDs()[collectionName].GetData()
 	flushTs, ok := flushResp.GetCollFlushTs()[collectionName]
 	if !ok {
-		return merr.WrapErrServiceInternal(fmt.Sprintf("failed to get flush timestamp for collection %s", collectionName))
+		return merr.WrapErrServiceInternalMsg("failed to get flush timestamp for collection %s", collectionName)
 	}
 
 	ticker := time.NewTicker(500 * time.Millisecond)
@@ -1521,7 +1553,7 @@ func generatePlaceholderGroup(ctx context.Context, body string, collSchema *sche
 					fieldName = field.Name
 					vectorField = field
 				} else {
-					return nil, errors.New("search without annsField, but already found multiple vector fields: [" + fieldName + ", " + field.Name + ",,,]")
+					return nil, merr.WrapErrParameterInvalidMsg("search without annsField, but already found multiple vector fields: [%s, %s,,,]", fieldName, field.Name)
 				}
 			}
 		}
@@ -1547,7 +1579,7 @@ func generatePlaceholderGroup(ctx context.Context, body string, collSchema *sche
 		}
 	}
 	if vectorField == nil {
-		return nil, errors.New("cannot find a vector field named: " + fieldName)
+		return nil, merr.WrapErrFieldNotFound(fieldName, "cannot find a vector field")
 	}
 	dim := int64(0)
 	if !typeutil.IsSparseFloatVectorType(vectorField.DataType) {

internal/distributed/proxy/httpserver/handler_v2_test.go

@@ -3352,13 +3352,13 @@ func TestSearchV2(t *testing.T) {
 	queryTestCases = append(queryTestCases, requestBodyTestCase{
 		path:        SearchAction,
 		requestBody: []byte(`{"collectionName": "book", "data": [[0.1, 0.2]], "annsField": "word_count", "filter": "book_id in [2, 4, 6, 8]", "limit": 4, "outputFields": ["word_count"], "params": {"radius":0.9, "range_filter": 0.1}, "groupingField": "test"}`),
-		errMsg:      "can only accept json format request, error: cannot find a vector field named: word_count",
+		errMsg:      "cannot find a vector field",
 		errCode:     1801,
 	})
 	queryTestCases = append(queryTestCases, requestBodyTestCase{
 		path:        AdvancedSearchAction,
 		requestBody: []byte(`{"collectionName": "hello_milvus", "search": [{"data": [[0.1, 0.2]], "annsField": "float_vector1", "metricType": "L2", "limit": 3}, {"data": [[0.1, 0.2]], "annsField": "float_vector2", "metricType": "L2", "limit": 3}], "rerank": {"strategy": "rrf", "params": {"k":  1}}}`),
-		errMsg:      "can only accept json format request, error: cannot find a vector field named: float_vector1",
+		errMsg:      "cannot find a vector field",
 		errCode:     1801,
 	})
 	// multiple annsFields

internal/distributed/proxy/httpserver/timeout_middleware.go

@@ -27,7 +27,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/gin-gonic/gin"
 	"go.uber.org/zap"
 
@@ -93,7 +92,7 @@ func (w *timeoutResponseRecorder) Write(data []byte) (int, error) {
 	w.mu.Lock()
 	defer w.mu.Unlock()
 	if w.closed || w.body == nil {
-		return 0, errors.New("response writer closed")
+		return 0, merr.WrapErrServiceInternalMsg("response writer closed")
 	}
 	if !w.written() {
 		w.size = 0
@@ -107,7 +106,7 @@ func (w *timeoutResponseRecorder) WriteString(s string) (int, error) {
 	w.mu.Lock()
 	defer w.mu.Unlock()
 	if w.closed || w.body == nil {
-		return 0, errors.New("response writer closed")
+		return 0, merr.WrapErrServiceInternalMsg("response writer closed")
 	}
 	if !w.written() {
 		w.size = 0
@@ -161,7 +160,7 @@ func (w *timeoutResponseRecorder) Written() bool {
 func (w *timeoutResponseRecorder) Flush() {}
 
 func (w *timeoutResponseRecorder) Hijack() (net.Conn, *bufio.ReadWriter, error) {
-	return nil, nil, errors.New("response writer does not support hijack")
+	return nil, nil, merr.WrapErrServiceInternalMsg("response writer does not support hijack")
 }
 
 func (w *timeoutResponseRecorder) CloseNotify() <-chan bool {
@@ -176,7 +175,7 @@ func (w *timeoutResponseRecorder) CommitTo(realWriter gin.ResponseWriter) error
 	w.mu.Lock()
 	defer w.mu.Unlock()
 	if w.closed || w.body == nil {
-		return errors.New("response writer closed")
+		return merr.WrapErrServiceInternalMsg("response writer closed")
 	}
 
 	dst := realWriter.Header()

internal/distributed/proxy/httpserver/utils.go

@@ -27,7 +27,6 @@ import (
 	"strings"
 	"time"
 
-	"github.com/cockroachdb/errors"
 	"github.com/gin-gonic/gin"
 	"github.com/spf13/cast"
 	"github.com/tidwall/gjson"
@@ -239,7 +238,7 @@ func convertRange(field *schemapb.FieldSchema, result gjson.Result) (string, err
 func checkGetPrimaryKey(coll *schemapb.CollectionSchema, idResult gjson.Result) (string, error) {
 	primaryField, ok := getPrimaryField(coll)
 	if !ok {
-		return "", fmt.Errorf("collection: %s has no primary field", coll.Name)
+		return "", merr.WrapErrParameterInvalidMsg("collection: %s has no primary field", coll.Name)
 	}
 	resultStr, err := convertRange(primaryField, idResult)
 	if err != nil {
@@ -253,7 +252,7 @@ func checkGetPrimaryKey(coll *schemapb.CollectionSchema, idResult gjson.Result)
 // based on the primary key field type
 func convertIDsToSchemapbIDs(ids []interface{}, pkField *schemapb.FieldSchema) (*schemapb.IDs, error) {
 	if len(ids) == 0 {
-		return nil, errors.New("ids array cannot be empty")
+		return nil, merr.WrapErrParameterMissingMsg("ids array cannot be empty")
 	}
 
 	switch pkField.DataType {
@@ -270,18 +269,18 @@ func convertIDsToSchemapbIDs(ids []interface{}, pkField *schemapb.FieldSchema) (
 				// JSON numbers are decoded as float64
 				// Check if the float has a fractional part
 				if v != math.Trunc(v) {
-					return nil, fmt.Errorf("invalid int64 id at index %d: %v has fractional part", i, v)
+					return nil, merr.WrapErrParameterInvalidMsg("invalid int64 id at index %d: %v has fractional part", i, v)
 				}
 				int64ID = int64(v)
 			case string:
 				// Try to parse string as int64
 				parsed, err := strconv.ParseInt(v, 10, 64)
 				if err != nil {
-					return nil, fmt.Errorf("invalid int64 id at index %d: %v, error: %v", i, id, err)
+					return nil, merr.WrapErrParameterInvalidErr(err, "invalid int64 id at index %d: %v", i, id)
 				}
 				int64ID = parsed
 			default:
-				return nil, fmt.Errorf("invalid id type at index %d: expected int64, got %T", i, id)
+				return nil, merr.WrapErrParameterInvalidMsg("invalid id type at index %d: expected int64, got %T", i, id)
 			}
 			int64IDs = append(int64IDs, int64ID)
 		}
@@ -304,10 +303,10 @@ func convertIDsToSchemapbIDs(ids []interface{}, pkField *schemapb.FieldSchema) (
 				// Convert number to string
 				stringID = fmt.Sprintf("%v", v)
 			default:
-				return nil, fmt.Errorf("invalid id type at index %d: expected string, got %T", i, id)
+				return nil, merr.WrapErrParameterInvalidMsg("invalid id type at index %d: expected string, got %T", i, id)
 			}
 			if stringID == "" {
-				return nil, fmt.Errorf("empty string id at index %d", i)
+				return nil, merr.WrapErrParameterInvalidMsg("empty string id at index %d", i)
 			}
 			stringIDs = append(stringIDs, stringID)
 		}
@@ -320,7 +319,7 @@ func convertIDsToSchemapbIDs(ids []interface{}, pkField *schemapb.FieldSchema) (
 		}, nil
 
 	default:
-		return nil, fmt.Errorf("unsupported primary key type: %s", pkField.DataType.String())
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported primary key type: %s", pkField.DataType.String())
 	}
 }
 
@@ -784,7 +783,7 @@ func checkAndSetData(body []byte, collSchema *schemapb.CollectionSchema, partial
 				if !containsString(fieldNames, mapKey) {
 					if collSchema.EnableDynamicField {
 						if mapKey == common.MetaFieldName {
-							return nil, nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("use the invalid field name(%s) when enable dynamicField", mapKey))
+							return nil, nil, merr.WrapErrParameterInvalidMsg("use the invalid field name(%s) when enable dynamicField", mapKey)
 						}
 						mapValueStr := mapValue.String()
 						switch mapValue.Type {
@@ -1106,7 +1105,7 @@ func decodeByteVectorElement(v gjson.Result, dim, bytesPerVec int64, isFloat16 b
 			return nil, err
 		}
 		if int64(len(row)) != dim {
-			return nil, fmt.Errorf("vector dim mismatch: expect %d, got %d", dim, len(row))
+			return nil, merr.WrapErrParameterInvalidMsg("vector dim mismatch: expect %d, got %d", dim, len(row))
 		}
 		if isFloat16 {
 			return typeutil.Float32ArrayToFloat16Bytes(row), nil
@@ -1114,14 +1113,14 @@ func decodeByteVectorElement(v gjson.Result, dim, bytesPerVec int64, isFloat16 b
 		return typeutil.Float32ArrayToBFloat16Bytes(row), nil
 	}
 	if v.Type != gjson.String {
-		return nil, fmt.Errorf("expect float vector array or base64 string")
+		return nil, merr.WrapErrParameterInvalidMsg("expect float vector array or base64 string")
 	}
 	var row []byte
 	if err := json.Unmarshal([]byte(v.Raw), &row); err != nil {
 		return nil, err
 	}
 	if int64(len(row)) != bytesPerVec {
-		return nil, fmt.Errorf("byte length mismatch: expect %d, got %d", bytesPerVec, len(row))
+		return nil, merr.WrapErrParameterInvalidMsg("byte length mismatch: expect %d, got %d", bytesPerVec, len(row))
 	}
 	return row, nil
 }
@@ -1293,7 +1292,7 @@ func encodeEmbListQuery(vecs []gjson.Result, elemType schemapb.DataType, dim int
 
 func buildStructArrayFieldData(structSchema *schemapb.StructArrayFieldSchema, perRow []structArrayRow) (*schemapb.FieldData, error) {
 	if len(perRow) == 0 {
-		return nil, fmt.Errorf("struct array field %s has no rows", structSchema.GetName())
+		return nil, merr.WrapErrParameterInvalidMsg("struct array field %s has no rows", structSchema.GetName())
 	}
 	subs := structSchema.GetFields()
 	subFieldData := make([]*schemapb.FieldData, 0, len(subs))
@@ -1308,12 +1307,12 @@ func buildStructArrayFieldData(structSchema *schemapb.StructArrayFieldSchema, pe
 			for rowIdx, row := range perRow {
 				val, ok := row[short]
 				if !ok {
-					return nil, fmt.Errorf("struct %s row %d missing sub-field %s",
+					return nil, merr.WrapErrParameterInvalidMsg("struct %s row %d missing sub-field %s",
 						structSchema.GetName(), rowIdx, short)
 				}
 				scalar, ok := val.(*schemapb.ScalarField)
 				if !ok {
-					return nil, fmt.Errorf("struct %s sub-field %s row %d: unexpected payload type %T",
+					return nil, merr.WrapErrParameterInvalidMsg("struct %s sub-field %s row %d: unexpected payload type %T",
 						structSchema.GetName(), short, rowIdx, val)
 				}
 				arrayArray.Data = append(arrayArray.Data, scalar)
@@ -1341,12 +1340,12 @@ func buildStructArrayFieldData(structSchema *schemapb.StructArrayFieldSchema, pe
 			for rowIdx, row := range perRow {
 				val, ok := row[short]
 				if !ok {
-					return nil, fmt.Errorf("struct %s row %d missing sub-field %s",
+					return nil, merr.WrapErrParameterInvalidMsg("struct %s row %d missing sub-field %s",
 						structSchema.GetName(), rowIdx, short)
 				}
 				vf, ok := val.(*schemapb.VectorField)
 				if !ok {
-					return nil, fmt.Errorf("struct %s sub-field %s row %d: unexpected payload type %T",
+					return nil, merr.WrapErrParameterInvalidMsg("struct %s sub-field %s row %d: unexpected payload type %T",
 						structSchema.GetName(), short, rowIdx, val)
 				}
 				vecArray.Data = append(vecArray.Data, vf)
@@ -1365,7 +1364,7 @@ func buildStructArrayFieldData(structSchema *schemapb.StructArrayFieldSchema, pe
 				},
 			})
 		default:
-			return nil, fmt.Errorf("unsupported struct sub-field data type: %s", sub.GetDataType())
+			return nil, merr.WrapErrParameterInvalidMsg("unsupported struct sub-field data type: %s", sub.GetDataType())
 		}
 	}
 	return &schemapb.FieldData{
@@ -1403,11 +1402,11 @@ func extractStructArrayRow(fd *schemapb.FieldData, rowIdx int, schema *schemapb.
 		case schemapb.DataType_Array:
 			rowData := sub.GetScalars().GetArrayData().GetData()
 			if rowIdx >= len(rowData) {
-				return nil, fmt.Errorf("struct sub-field %s missing row %d", short, rowIdx)
+				return nil, merr.WrapErrParameterInvalidMsg("struct sub-field %s missing row %d", short, rowIdx)
 			}
 			values := scalarArrayToInterfaces(rowData[rowIdx])
 			if len(values) != elemCount {
-				return nil, fmt.Errorf("struct sub-field %s element count mismatch: expect %d got %d",
+				return nil, merr.WrapErrParameterInvalidMsg("struct sub-field %s element count mismatch: expect %d got %d",
 					short, elemCount, len(values))
 			}
 			for i, v := range values {
@@ -1416,28 +1415,28 @@ func extractStructArrayRow(fd *schemapb.FieldData, rowIdx int, schema *schemapb.
 		case schemapb.DataType_ArrayOfVector:
 			va := sub.GetVectors().GetVectorArray()
 			if va == nil {
-				return nil, fmt.Errorf("struct sub-field %s has no vector array", short)
+				return nil, merr.WrapErrParameterInvalidMsg("struct sub-field %s has no vector array", short)
 			}
 			if rowIdx >= len(va.GetData()) {
-				return nil, fmt.Errorf("struct sub-field %s missing row %d", short, rowIdx)
+				return nil, merr.WrapErrParameterInvalidMsg("struct sub-field %s missing row %d", short, rowIdx)
 			}
 			dim, ok := subDims[short]
 			if !ok || dim <= 0 {
-				return nil, fmt.Errorf("schema missing dim for struct sub-field %s", short)
+				return nil, merr.WrapErrParameterInvalidMsg("schema missing dim for struct sub-field %s", short)
 			}
 			values, err := vectorFieldToInterfaces(va.GetData()[rowIdx], va.GetElementType(), dim)
 			if err != nil {
 				return nil, err
 			}
 			if len(values) != elemCount {
-				return nil, fmt.Errorf("struct sub-field %s vector element count mismatch: expect %d got %d",
+				return nil, merr.WrapErrParameterInvalidMsg("struct sub-field %s vector element count mismatch: expect %d got %d",
 					short, elemCount, len(values))
 			}
 			for i, v := range values {
 				out[i][short] = v
 			}
 		default:
-			return nil, fmt.Errorf("unsupported struct sub-field type %s", sub.GetType())
+			return nil, merr.WrapErrParameterInvalidMsg("unsupported struct sub-field type %s", sub.GetType())
 		}
 	}
 	return out, nil
@@ -1455,7 +1454,7 @@ func structArraySubDims(fieldName string, schema *schemapb.CollectionSchema) (ma
 			}
 			dim, err := getDim(sub)
 			if err != nil {
-				return nil, fmt.Errorf("schema sub-field %s has no dim: %w", sub.GetName(), err)
+				return nil, merr.WrapErrParameterInvalidErr(err, "schema sub-field %s has no dim", sub.GetName())
 			}
 			subDims[subShortName(sub)] = dim
 		}
@@ -1469,22 +1468,22 @@ func structSubElemCount(sub *schemapb.FieldData, rowIdx int, subDims map[string]
 	case schemapb.DataType_Array:
 		rowData := sub.GetScalars().GetArrayData().GetData()
 		if rowIdx >= len(rowData) {
-			return 0, fmt.Errorf("struct sub-field %s row %d out of range", sub.GetFieldName(), rowIdx)
+			return 0, merr.WrapErrParameterInvalidMsg("struct sub-field %s row %d out of range", sub.GetFieldName(), rowIdx)
 		}
 		return len(scalarArrayToInterfaces(rowData[rowIdx])), nil
 	case schemapb.DataType_ArrayOfVector:
 		va := sub.GetVectors().GetVectorArray()
 		if va == nil || rowIdx >= len(va.GetData()) {
-			return 0, fmt.Errorf("struct sub-field %s row %d out of range", sub.GetFieldName(), rowIdx)
+			return 0, merr.WrapErrParameterInvalidMsg("struct sub-field %s row %d out of range", sub.GetFieldName(), rowIdx)
 		}
 		short := structFieldShortName(sub.GetFieldName())
 		dim, ok := subDims[short]
 		if !ok || dim <= 0 {
-			return 0, fmt.Errorf("schema missing dim for struct sub-field %s", short)
+			return 0, merr.WrapErrParameterInvalidMsg("schema missing dim for struct sub-field %s", short)
 		}
 		return vectorFieldElemCount(va.GetData()[rowIdx], va.GetElementType(), dim)
 	default:
-		return 0, fmt.Errorf("unsupported struct sub-field type %s", sub.GetType())
+		return 0, merr.WrapErrParameterInvalidMsg("unsupported struct sub-field type %s", sub.GetType())
 	}
 }
 
@@ -1539,7 +1538,7 @@ func scalarArrayToInterfaces(sf *schemapb.ScalarField) []interface{} {
 
 func vectorFieldElemCount(vf *schemapb.VectorField, elemType schemapb.DataType, dim int64) (int, error) {
 	if dim <= 0 {
-		return 0, fmt.Errorf("invalid dim %d", dim)
+		return 0, merr.WrapErrParameterInvalidMsg("invalid dim %d", dim)
 	}
 	switch elemType {
 	case schemapb.DataType_FloatVector:
@@ -1553,13 +1552,13 @@ func vectorFieldElemCount(vf *schemapb.VectorField, elemType schemapb.DataType,
 	case schemapb.DataType_Int8Vector:
 		return len(vf.GetInt8Vector()) / int(dim), nil
 	default:
-		return 0, fmt.Errorf("unsupported vector element type %s", elemType)
+		return 0, merr.WrapErrParameterInvalidMsg("unsupported vector element type %s", elemType)
 	}
 }
 
 func vectorFieldToInterfaces(vf *schemapb.VectorField, elemType schemapb.DataType, dim int64) ([]interface{}, error) {
 	if dim <= 0 {
-		return nil, fmt.Errorf("invalid dim %d", dim)
+		return nil, merr.WrapErrParameterInvalidMsg("invalid dim %d", dim)
 	}
 	switch elemType {
 	case schemapb.DataType_FloatVector:
@@ -1612,7 +1611,7 @@ func vectorFieldToInterfaces(vf *schemapb.VectorField, elemType schemapb.DataTyp
 		}
 		return out, nil
 	default:
-		return nil, fmt.Errorf("unsupported vector element type %s", elemType)
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported vector element type %s", elemType)
 	}
 }
 
@@ -1620,7 +1619,7 @@ func convertFloatVectorToArray(vector [][]float32, dim int64) ([]float32, error)
 	floatArray := make([]float32, 0)
 	for _, arr := range vector {
 		if int64(len(arr)) != dim {
-			return nil, fmt.Errorf("[]float32 size %d doesn't equal to vector dimension %d of %s",
+			return nil, merr.WrapErrParameterInvalidMsg("[]float32 size %d doesn't equal to vector dimension %d of %s",
 				len(arr), dim, schemapb.DataType_name[int32(schemapb.DataType_FloatVector)])
 		}
 		for i := int64(0); i < dim; i++ {
@@ -1643,7 +1642,7 @@ func convertBinaryVectorToArray(vector [][]byte, dim int64, dataType schemapb.Da
 	binaryArray := make([]byte, 0, len(vector)*int(bytesLen))
 	for _, arr := range vector {
 		if int64(len(arr)) != bytesLen {
-			return nil, fmt.Errorf("[]byte size %d doesn't equal to vector dimension %d of %s",
+			return nil, merr.WrapErrParameterInvalidMsg("[]byte size %d doesn't equal to vector dimension %d of %s",
 				len(arr), dim, schemapb.DataType_name[int32(dataType)])
 		}
 		for i := int64(0); i < bytesLen; i++ {
@@ -1657,7 +1656,7 @@ func convertInt8VectorToArray(vector [][]int8, dim int64) ([]byte, error) {
 	byteArray := make([]byte, 0)
 	for _, arr := range vector {
 		if int64(len(arr)) != dim {
-			return nil, fmt.Errorf("[]int8 size %d doesn't equal to vector dimension %d of %s",
+			return nil, merr.WrapErrParameterInvalidMsg("[]int8 size %d doesn't equal to vector dimension %d of %s",
 				len(arr), dim, schemapb.DataType_name[int32(schemapb.DataType_Int8Vector)])
 		}
 		for i := int64(0); i < dim; i++ {
@@ -1691,7 +1690,7 @@ func reflectValueCandi(v reflect.Value) (map[string]fieldCandi, error) {
 		}
 		return result, nil
 	default:
-		return nil, fmt.Errorf("unsupport row type: %s", v.Kind().String())
+		return nil, merr.WrapErrParameterInvalidMsg("unsupport row type: %s", v.Kind().String())
 	}
 }
 
@@ -1713,7 +1712,7 @@ func convertToIntArray(dataType schemapb.DataType, arr interface{}) []int32 {
 func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool, sch *schemapb.CollectionSchema, inInsert bool, partialUpdate bool) ([]*schemapb.FieldData, error) {
 	rowsLen := len(rows)
 	if rowsLen == 0 {
-		return []*schemapb.FieldData{}, errors.New("no row need to be convert to columns")
+		return []*schemapb.FieldData{}, merr.WrapErrParameterInvalidMsg("no row need to be convert to columns")
 	}
 
 	isDynamic := sch.EnableDynamicField
@@ -1802,7 +1801,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 			dim, _ := getDim(field)
 			nameDims[field.Name] = dim
 		default:
-			return nil, fmt.Errorf("the type(%v) of field(%v) is not supported, use other sdk please", field.DataType, field.Name)
+			return nil, merr.WrapErrParameterInvalidMsg("the type(%v) of field(%v) is not supported, use other sdk please", field.DataType, field.Name)
 		}
 		nameColumns[field.Name] = data
 		fieldData[field.Name] = &schemapb.FieldData{
@@ -1813,7 +1812,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 		}
 	}
 	if len(nameDims) == 0 && len(sch.Functions) == 0 && !partialUpdate {
-		return nil, fmt.Errorf("collection: %s has no vector field or functions", sch.Name)
+		return nil, merr.WrapErrParameterInvalidMsg("collection: %s has no vector field or functions", sch.Name)
 	}
 
 	dynamicCol := make([][]byte, 0, rowsLen)
@@ -1836,7 +1835,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 					continue
 				}
 				if !allowInsertAutoID {
-					return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("no need to pass pk field(%s) when autoid==true in insert", field.Name))
+					return nil, merr.WrapErrParameterInvalidMsg("no need to pass pk field(%s) when autoid==true in insert", field.Name)
 				}
 			}
 			if (field.Nullable || field.DefaultValue != nil) && !ok {
@@ -1851,7 +1850,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 				if partialUpdate {
 					continue
 				}
-				return nil, fmt.Errorf("row %d does not has field %s", idx, field.Name)
+				return nil, merr.WrapErrParameterInvalidMsg("row %d does not has field %s", idx, field.Name)
 			}
 			fieldLen[field.Name] += 1
 			switch field.DataType {
@@ -1893,7 +1892,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 					vec := typeutil.Float32ArrayToFloat16Bytes(candi.v.Interface().([]float32))
 					nameColumns[field.Name] = append(nameColumns[field.Name].([][]byte), vec)
 				default:
-					return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("invalid type(%v) of field(%v) ", field.DataType, field.Name))
+					return nil, merr.WrapErrParameterInvalidMsg("invalid type(%v) of field(%v) ", field.DataType, field.Name)
 				}
 			case schemapb.DataType_BFloat16Vector:
 				switch candi.v.Interface().(type) {
@@ -1903,7 +1902,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 					vec := typeutil.Float32ArrayToBFloat16Bytes(candi.v.Interface().([]float32))
 					nameColumns[field.Name] = append(nameColumns[field.Name].([][]byte), vec)
 				default:
-					return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("invalid type(%v) of field(%v) ", field.DataType, field.Name))
+					return nil, merr.WrapErrParameterInvalidMsg("invalid type(%v) of field(%v) ", field.DataType, field.Name)
 				}
 			case schemapb.DataType_SparseFloatVector:
 				content := candi.v.Interface().([]byte)
@@ -1915,7 +1914,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 			case schemapb.DataType_Int8Vector:
 				nameColumns[field.Name] = append(nameColumns[field.Name].([][]int8), candi.v.Interface().([]int8))
 			default:
-				return nil, fmt.Errorf("the type(%v) of field(%v) is not supported, use other sdk please", field.DataType, field.Name)
+				return nil, merr.WrapErrParameterInvalidMsg("the type(%v) of field(%v) is not supported, use other sdk please", field.DataType, field.Name)
 			}
 
 			delete(set, field.Name)
@@ -1931,7 +1930,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 			}
 			bs, err := json.Marshal(m)
 			if err != nil {
-				return nil, fmt.Errorf("failed to marshal dynamic field %w", err)
+				return nil, merr.WrapErrParameterInvalidErr(err, "failed to marshal dynamic field")
 			}
 			dynamicCol = append(dynamicCol, bs)
 		}
@@ -1949,7 +1948,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 						zap.String("fieldName", name),
 						zap.Int("fieldLen", len(validData)),
 						zap.Int("rowsLen", rowsLen))
-					return nil, fmt.Errorf("column %s has length %d, expected %d", name, len(validData), rowsLen)
+					return nil, merr.WrapErrParameterInvalidMsg("column %s has length %d, expected %d", name, len(validData), rowsLen)
 				}
 			} else {
 				log.Info("skip empty field for partial update",
@@ -1963,7 +1962,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 				zap.String("fieldName", name),
 				zap.Int("fieldLen", fieldLen[name]),
 				zap.Int("rowsLen", rowsLen))
-			return nil, fmt.Errorf("column %s has length %d, expected %d", name, fieldLen[name], rowsLen)
+			return nil, merr.WrapErrParameterInvalidMsg("column %s has length %d, expected %d", name, fieldLen[name], rowsLen)
 		}
 
 		colData := fieldData[name]
@@ -2183,7 +2182,7 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 				},
 			}
 		default:
-			return nil, fmt.Errorf("the type(%v) of field(%v) is not supported, use other sdk please", colData.Type, name)
+			return nil, merr.WrapErrParameterInvalidMsg("the type(%v) of field(%v) is not supported, use other sdk please", colData.Type, name)
 		}
 		colData.ValidData = validDataMap[name]
 		columns = append(columns, colData)
@@ -2212,11 +2211,11 @@ func anyToColumns(rows []map[string]interface{}, validDataMap map[string][]bool,
 				if partialUpdate {
 					continue
 				}
-				return nil, fmt.Errorf("row %d does not has struct field %s", rowIdx, structField.GetName())
+				return nil, merr.WrapErrParameterInvalidMsg("row %d does not has struct field %s", rowIdx, structField.GetName())
 			}
 			sr, ok := val.(structArrayRow)
 			if !ok {
-				return nil, fmt.Errorf("row %d struct field %s has unexpected payload type %T",
+				return nil, merr.WrapErrParameterInvalidMsg("row %d struct field %s has unexpected payload type %T",
 					rowIdx, structField.GetName(), val)
 			}
 			perRow = append(perRow, sr)
@@ -2437,27 +2436,27 @@ func fieldDataValueCount(fieldData *schemapb.FieldData) (int64, error) {
 		dim := fieldData.GetVectors().GetDim()
 		bytesPerRow := dim / 8
 		if bytesPerRow <= 0 {
-			return 0, fmt.Errorf("invalid binary vector dimension %d for field %s", dim, fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("invalid binary vector dimension %d for field %s", dim, fieldData.GetFieldName())
 		}
 		return int64(len(fieldData.GetVectors().GetBinaryVector())) / bytesPerRow, nil
 	case schemapb.DataType_FloatVector:
 		dim := fieldData.GetVectors().GetDim()
 		if dim <= 0 {
-			return 0, fmt.Errorf("invalid float vector dimension %d for field %s", dim, fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("invalid float vector dimension %d for field %s", dim, fieldData.GetFieldName())
 		}
 		return int64(len(fieldData.GetVectors().GetFloatVector().GetData())) / dim, nil
 	case schemapb.DataType_Float16Vector:
 		dim := fieldData.GetVectors().GetDim()
 		bytesPerRow := dim * 2
 		if bytesPerRow <= 0 {
-			return 0, fmt.Errorf("invalid float16 vector dimension %d for field %s", dim, fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("invalid float16 vector dimension %d for field %s", dim, fieldData.GetFieldName())
 		}
 		return int64(len(fieldData.GetVectors().GetFloat16Vector())) / bytesPerRow, nil
 	case schemapb.DataType_BFloat16Vector:
 		dim := fieldData.GetVectors().GetDim()
 		bytesPerRow := dim * 2
 		if bytesPerRow <= 0 {
-			return 0, fmt.Errorf("invalid bfloat16 vector dimension %d for field %s", dim, fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("invalid bfloat16 vector dimension %d for field %s", dim, fieldData.GetFieldName())
 		}
 		return int64(len(fieldData.GetVectors().GetBfloat16Vector())) / bytesPerRow, nil
 	case schemapb.DataType_SparseFloatVector:
@@ -2465,7 +2464,7 @@ func fieldDataValueCount(fieldData *schemapb.FieldData) (int64, error) {
 	case schemapb.DataType_Int8Vector:
 		dim := fieldData.GetVectors().GetDim()
 		if dim <= 0 {
-			return 0, fmt.Errorf("invalid int8 vector dimension %d for field %s", dim, fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("invalid int8 vector dimension %d for field %s", dim, fieldData.GetFieldName())
 		}
 		return int64(len(fieldData.GetVectors().GetInt8Vector())) / dim, nil
 	case schemapb.DataType_ArrayOfStruct:
@@ -2479,10 +2478,10 @@ func fieldDataValueCount(fieldData *schemapb.FieldData) (int64, error) {
 		case schemapb.DataType_ArrayOfVector:
 			return int64(len(subs[0].GetVectors().GetVectorArray().GetData())), nil
 		default:
-			return 0, fmt.Errorf("unsupported struct sub-field type %s for field %s", subs[0].GetType(), fieldData.GetFieldName())
+			return 0, merr.WrapErrParameterInvalidMsg("unsupported struct sub-field type %s for field %s", subs[0].GetType(), fieldData.GetFieldName())
 		}
 	default:
-		return 0, fmt.Errorf("the type(%v) of field(%v) is not supported, use other sdk please", fieldData.GetType(), fieldData.GetFieldName())
+		return 0, merr.WrapErrParameterInvalidMsg("the type(%v) of field(%v) is not supported, use other sdk please", fieldData.GetType(), fieldData.GetFieldName())
 	}
 }
 
@@ -2532,7 +2531,7 @@ func newFieldDataRowAccessor(fieldData *schemapb.FieldData) (*fieldDataRowAccess
 		return accessor, nil
 	}
 	if valueCount != validCount {
-		return nil, fmt.Errorf("field %s has %d valid rows, but data length is %d", fieldData.GetFieldName(), validCount, valueCount)
+		return nil, merr.WrapErrParameterInvalidMsg("field %s has %d valid rows, but data length is %d", fieldData.GetFieldName(), validCount, valueCount)
 	}
 	accessor.compactIndices = compactIndices
 	return accessor, nil
@@ -2543,7 +2542,7 @@ func (accessor *fieldDataRowAccessor) rowIndex(rowIdx int64) (int64, bool, error
 		return rowIdx, true, nil
 	}
 	if rowIdx >= int64(len(accessor.validData)) {
-		return 0, false, fmt.Errorf("row index %d out of range for field %s valid data length %d", rowIdx, accessor.fieldData.GetFieldName(), len(accessor.validData))
+		return 0, false, merr.WrapErrParameterInvalidMsg("row index %d out of range for field %s valid data length %d", rowIdx, accessor.fieldData.GetFieldName(), len(accessor.validData))
 	}
 	if !accessor.validData[rowIdx] {
 		return 0, false, nil
@@ -2575,7 +2574,7 @@ func buildQueryResp(rowsNum int64, needFields []string, fieldDataList []*schemap
 				stringPks := ids.GetStrId().GetData()
 				rowsNum = int64(len(stringPks))
 			default:
-				return nil, errors.New("the type of primary key(id) is not supported, use other sdk please")
+				return nil, merr.WrapErrParameterInvalidMsg("the type of primary key(id) is not supported, use other sdk please")
 			}
 		}
 	}
@@ -2716,7 +2715,7 @@ func buildQueryResp(rowsNum int64, needFields []string, fieldDataList []*schemap
 				stringPks := ids.GetStrId().GetData()
 				row[pkFieldName] = stringPks[i]
 			default:
-				return nil, errors.New("the type of primary key(id) is not supported, use other sdk please")
+				return nil, merr.WrapErrParameterInvalidMsg("the type of primary key(id) is not supported, use other sdk please")
 			}
 		}
 		if scores != nil && int64(len(scores)) > i {
@@ -2734,29 +2733,31 @@ func hasSearchAggregationResult(results *schemapb.SearchResultData) bool {
 
 func buildSearchAggregationResp(results *schemapb.SearchResultData, enableInt64 bool, collectionSchema *schemapb.CollectionSchema) ([]gin.H, error) {
 	if results == nil {
-		return nil, errors.New("search_aggregation result is nil")
+		// The aggregation payload is produced by the server-side reduce, never
+		// by the request: a malformed shape is an internal contract violation.
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation result is nil")
 	}
 	aggTopks := results.GetAggTopks()
 	pbBuckets := results.GetAggBuckets()
 	if len(aggTopks) == 0 {
-		return nil, errors.New("search_aggregation response missing agg_topks")
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation response missing agg_topks")
 	}
 	if results.GetNumQueries() <= 0 {
-		return nil, errors.New("search_aggregation response missing nq")
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation response missing nq")
 	}
 	if len(aggTopks) != int(results.GetNumQueries()) {
-		return nil, fmt.Errorf("search_aggregation agg_topks length %d does not match nq %d", len(aggTopks), results.GetNumQueries())
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation agg_topks length %d does not match nq %d", len(aggTopks), results.GetNumQueries())
 	}
 
 	total := int64(0)
 	for _, topk := range aggTopks {
 		if topk < 0 {
-			return nil, errors.New("search_aggregation agg_topks cannot contain negative values")
+			return nil, merr.WrapErrServiceInternalMsg("search_aggregation agg_topks cannot contain negative values")
 		}
 		total += topk
 	}
 	if total != int64(len(pbBuckets)) {
-		return nil, fmt.Errorf("search_aggregation agg_topks sum %d does not match bucket count %d", total, len(pbBuckets))
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation agg_topks sum %d does not match bucket count %d", total, len(pbBuckets))
 	}
 
 	output := make([]gin.H, 0, len(aggTopks))
@@ -2778,7 +2779,7 @@ func buildSearchAggregationResp(results *schemapb.SearchResultData, enableInt64
 
 func buildAggBucketResp(pb *schemapb.AggBucket, enableInt64 bool, collectionSchema *schemapb.CollectionSchema) (gin.H, error) {
 	if pb == nil {
-		return nil, errors.New("search_aggregation bucket is nil")
+		return nil, merr.WrapErrServiceInternalMsg("search_aggregation bucket is nil")
 	}
 	bucket := gin.H{
 		"key":       buildAggBucketKeyResp(pb.GetKey(), enableInt64),
@@ -2986,7 +2987,7 @@ func convertConsistencyLevel(reqConsistencyLevel string) (commonpb.ConsistencyLe
 	if reqConsistencyLevel != "" {
 		level, ok := commonpb.ConsistencyLevel_value[reqConsistencyLevel]
 		if !ok {
-			return 0, false, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("parameter:'%s' is incorrect, please check it", reqConsistencyLevel))
+			return 0, false, merr.WrapErrParameterInvalidMsg("parameter:'%s' is incorrect, please check it", reqConsistencyLevel)
 		}
 		return commonpb.ConsistencyLevel(level), false, nil
 	}
@@ -3096,7 +3097,7 @@ func convertDefaultValue(value interface{}, dataType schemapb.DataType) (*schema
 		}
 		return data, nil
 	default:
-		return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("Unexpected default value type: %s", dataType.String()))
+		return nil, merr.WrapErrParameterInvalidMsg("Unexpected default value type: %s", dataType.String())
 	}
 }
 
@@ -3491,7 +3492,7 @@ func generateSearchParams(reqSearchParams map[string]interface{}) ([]*commonpb.K
 	for key, value := range reqSearchParams {
 		if val, ok := paramsMap[key]; ok {
 			if !deepEqual(val, value) {
-				return nil, merr.WrapErrParameterInvalidMsg(fmt.Sprintf("ambiguous parameter: %s, in search_param: %v, in search_param.params: %v", key, value, val))
+				return nil, merr.WrapErrParameterInvalidMsg("ambiguous parameter: %s, in search_param: %v, in search_param.params: %v", key, value, val)
 			}
 		} else if key != Params {
 			paramsMap[key] = value

internal/distributed/proxy/httpserver/utils_test.go

@@ -25,6 +25,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/cockroachdb/errors"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -36,6 +37,7 @@ import (
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	"github.com/milvus-io/milvus/internal/json"
 	"github.com/milvus-io/milvus/pkg/v3/common"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
 
@@ -3005,7 +3007,8 @@ func TestBuildQueryResps(t *testing.T) {
 	}
 
 	_, err := buildQueryResp(int64(0), outputFields, newFieldData([]*schemapb.FieldData{}, 1000), generateIDs(schemapb.DataType_Int64, 3), DefaultScores, true, nil)
-	assert.Equal(t, "the type(1000) of field(wrong-field-type) is not supported, use other sdk please", err.Error())
+	assert.Contains(t, err.Error(), "the type(1000) of field(wrong-field-type) is not supported, use other sdk please")
+	assert.True(t, errors.Is(err, merr.ErrParameterInvalid))
 
 	res, err := buildQueryResp(int64(0), outputFields, []*schemapb.FieldData{}, generateIDs(schemapb.DataType_Int64, 3), DefaultScores, true, nil)
 	assert.Equal(t, 3, len(res))

internal/distributed/proxy/httpserver/wrap_request.go

@@ -17,15 +17,13 @@
 package httpserver
 
 import (
-	"fmt"
-
-	"github.com/cockroachdb/errors"
 	"google.golang.org/protobuf/proto"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/commonpb"
 	"github.com/milvus-io/milvus-proto/go-api/v3/milvuspb"
 	"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
 	"github.com/milvus-io/milvus/internal/json"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 	"github.com/milvus-io/milvus/pkg/v3/util/typeutil"
 )
 
@@ -70,7 +68,7 @@ type WrappedInsertRequest struct {
 func (w *WrappedInsertRequest) AsInsertRequest() (*milvuspb.InsertRequest, error) {
 	fieldData, err := convertFieldDataArray(w.FieldsData)
 	if err != nil {
-		return nil, fmt.Errorf("%w: convert field data failed: %v", errBadRequest, err)
+		return nil, badRequestf(err, "convert field data failed")
 	}
 	return &milvuspb.InsertRequest{
 		Base:           w.Base,
@@ -98,12 +96,12 @@ func (f *FieldData) makePbFloat16OrBfloat16Array(raw json.RawMessage, serializeF
 		return nil, 0, newFieldDataError(f.FieldName, err)
 	}
 	if len(wrappedData) < 1 {
-		return nil, 0, errors.New("at least one row for insert")
+		return nil, 0, merr.WrapErrParameterInvalidMsg("at least one row for insert")
 	}
 	array0 := wrappedData[0]
 	dim := len(array0)
 	if dim < 1 {
-		return nil, 0, errors.New("dim must >= 1")
+		return nil, 0, merr.WrapErrParameterInvalidMsg("dim must >= 1")
 	}
 	data := make([]byte, 0, len(wrappedData)*dim*2)
 	for _, fp32Array := range wrappedData {
@@ -238,12 +236,12 @@ func (f *FieldData) AsSchemapb() (*schemapb.FieldData, error) {
 			return nil, newFieldDataError(f.FieldName, err)
 		}
 		if len(wrappedData) < 1 {
-			return nil, errors.New("at least one row for insert")
+			return nil, merr.WrapErrParameterInvalidMsg("at least one row for insert")
 		}
 		array0 := wrappedData[0]
 		dim := len(array0)
 		if dim < 1 {
-			return nil, errors.New("dim must >= 1")
+			return nil, merr.WrapErrParameterInvalidMsg("dim must >= 1")
 		}
 		data := make([]float32, len(wrappedData)*dim)
 
@@ -299,7 +297,7 @@ func (f *FieldData) AsSchemapb() (*schemapb.FieldData, error) {
 			return nil, newFieldDataError(f.FieldName, err)
 		}
 		if len(wrappedData) < 1 {
-			return nil, errors.New("at least one row for insert")
+			return nil, merr.WrapErrParameterInvalidMsg("at least one row for insert")
 		}
 		data := make([][]byte, len(wrappedData))
 		dim := int64(0)
@@ -333,12 +331,12 @@ func (f *FieldData) AsSchemapb() (*schemapb.FieldData, error) {
 			return nil, newFieldDataError(f.FieldName, err)
 		}
 		if len(wrappedData) < 1 {
-			return nil, errors.New("at least one row for insert")
+			return nil, merr.WrapErrParameterInvalidMsg("at least one row for insert")
 		}
 		array0 := wrappedData[0]
 		dim := len(array0)
 		if dim < 1 {
-			return nil, errors.New("dim must >= 1")
+			return nil, merr.WrapErrParameterInvalidMsg("dim must >= 1")
 		}
 		data := make([]byte, len(wrappedData)*dim)
 
@@ -358,13 +356,13 @@ func (f *FieldData) AsSchemapb() (*schemapb.FieldData, error) {
 			},
 		}
 	default:
-		return nil, errors.New("unsupported data type")
+		return nil, merr.WrapErrParameterInvalidMsg("unsupported data type")
 	}
 	return &ret, nil
 }
 
 func newFieldDataError(field string, err error) error {
-	return fmt.Errorf("parse field[%s]: %s", field, err.Error())
+	return merr.WrapErrParameterInvalidErr(err, "parse field[%s]", field)
 }
 
 func convertFieldDataArray(input []*FieldData) ([]*schemapb.FieldData, error) {

internal/distributed/proxy/httpserver/wrapper.go

@@ -25,10 +25,18 @@ import (
 	"github.com/gin-gonic/gin/binding"
 
 	"github.com/milvus-io/milvus-proto/go-api/v3/commonpb"
+	"github.com/milvus-io/milvus/pkg/v3/util/merr"
 )
 
 var errBadRequest = errors.New("bad request")
 
+// badRequestf wraps err with the package-internal errBadRequest sentinel and
+// a contextual message, preserving the inner error chain. wrapHandler maps
+// any error that matches errBadRequest to HTTP 400.
+func badRequestf(err error, format string, args ...any) error {
+	return merr.Mark(merr.Wrapf(err, format, args...), errBadRequest)
+}
+
 // handlerFunc handles http request with gin context
 type handlerFunc func(c *gin.Context) (interface{}, error)