Files
sijie-ni-0214andGitHub a7db8873e1 fix: prevent schema updates from racing insert payload conversion (#51483)
## Problem

During schema evolution, SQN can convert an insert payload with the old
schema and then race with a schema update before ProcessInsert creates
and writes the growing segment. If the old payload still contains a
dropped field, the native collection may already use the new schema and
segcore rejects the write with unexpected new field.

## Solution

- Add a collection-local schema-transition RW lock.
- Insert holds the reader lock from schema snapshot use through payload
conversion and delegator.ProcessInsert, so a growing segment is created
and written in the same schema epoch as its payload.
- Schema update and existing-collection load refresh paths take the
writer lock around native schema mutation and Go schema snapshot
publication.
- Acquire a temporary collection lease before releasing
collectionManager.mut. This keeps the collection alive while the writer
waits, without serializing unrelated collection-manager operations
behind a schema transition.
- Preserve dropped-field replay handling: when DDL has already
completed, payload conversion filters fields absent from the current
schema before inserting into a new-schema growing segment.

## Tests

- Add a deterministic native regression: pause after a v950 payload
containing field 580 is converted, queue the v951 drop-field update,
then resume a real NewSegment plus growing.Insert and verify both insert
and DDL succeed.
- Assert the transition reader covers both payload conversion and
ProcessInsert.
- Cover the DDL-first replay path, where field 580 is filtered before
native insertion.
- Cover both schema writer entry points: UpdateSchema and
existing-collection PutOrRef.

## Validation

- make static-check
- make build-go
- Native pipeline regression tests repeated 20 times with dynamic,test
- Collection schema-transition and lease tests repeated 20 times
- internal/querynodev2/pipeline and internal/querynodev2/segments
package tests
- gofumpt -l and git diff --check

Fixes #51436

Signed-off-by: sijie-ni-0214 <sijie.ni@zilliz.com>
2026-07-17 17:26:39 +08:00

248 lines
8.8 KiB
Go

// Licensed to the LF AI & Data foundation under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package pipeline
import (
"fmt"
"sort"
"github.com/milvus-io/milvus-proto/go-api/v3/msgpb"
"github.com/milvus-io/milvus-proto/go-api/v3/schemapb"
"github.com/milvus-io/milvus/internal/querynodev2/delegator"
"github.com/milvus-io/milvus/internal/querynodev2/segments"
"github.com/milvus-io/milvus/internal/storage"
"github.com/milvus-io/milvus/internal/util/function"
base "github.com/milvus-io/milvus/internal/util/pipeline"
"github.com/milvus-io/milvus/pkg/v3/metrics"
"github.com/milvus-io/milvus/pkg/v3/mlog"
"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"
)
type insertNode struct {
*BaseNode
collectionID int64
channel string
manager *DataManager
delegator delegator.ShardDelegator
functionStore *function.FunctionRunnerLocalStore
}
func (iNode *insertNode) addInsertData(insertDatas map[UniqueID]*delegator.InsertData, msg *InsertMsg, schema *schemapb.CollectionSchema) {
ctx := msg.TraceCtx()
insertRecord, skippedFields, err := storage.TransferInsertMsgToInsertRecord(schema, msg)
if err != nil {
err = merr.Wrap(err, "failed to get primary keys")
mlog.Error(ctx, err.Error(), mlog.Int64("collectionID", iNode.collectionID), mlog.String("channel", iNode.channel))
panic(err)
}
if len(skippedFields) > 0 {
// Attributing the skip to dropped fields is safe only because a SchemaChange
// message never shares a pack with inserts: the WAL adaptor emits every V1/V2
// message as its own pack, msgdispatcher never merges packs, and the DML
// micro-batcher refuses to merge non-Insert/Delete messages. If that
// pack-granularity invariant is ever relaxed, filtering against the current
// schema could silently drop fields that exist in a newer schema.
mlog.Warn(ctx, "skip insert payload fields absent from current schema, fields are dropped since the message was written",
mlog.FieldCollectionID(iNode.collectionID),
mlog.FieldSegmentID(msg.SegmentID),
mlog.String("channel", iNode.channel),
mlog.Int32("schemaVersion", schema.GetVersion()),
mlog.Int64s("skippedFieldIDs", skippedFields))
metrics.QueryNodeSkippedInsertFieldCount.
WithLabelValues(paramtable.GetStringNodeID(), fmt.Sprint(iNode.collectionID)).
Add(float64(len(skippedFields)))
}
iData, ok := insertDatas[msg.SegmentID]
if !ok {
iData = &delegator.InsertData{
PartitionID: msg.PartitionID,
InsertRecord: insertRecord,
StartPosition: &msgpb.MsgPosition{
Timestamp: msg.BeginTs(),
ChannelName: msg.GetShardName(),
},
}
insertDatas[msg.SegmentID] = iData
} else {
err := typeutil.MergeFieldData(iData.InsertRecord.FieldsData, insertRecord.FieldsData)
if err != nil {
mlog.Error(ctx, "failed to merge field data", mlog.String("channel", iNode.channel), mlog.Err(err))
panic(err)
}
iData.InsertRecord.NumRows += insertRecord.NumRows
}
if err := iNode.appendBM25Stats(iData, msg, schema); err != nil {
mlog.Error(ctx, "failed to append BM25 stats from insert message", mlog.String("channel", iNode.channel), mlog.Err(err))
panic(err)
}
pks, err := segments.GetPrimaryKeys(msg, schema)
if err != nil {
mlog.Error(ctx, "failed to get primary keys from insert message", mlog.Err(err))
panic(err)
}
iData.PrimaryKeys = append(iData.PrimaryKeys, pks...)
iData.RowIDs = append(iData.RowIDs, msg.RowIDs...)
iData.Timestamps = append(iData.Timestamps, msg.Timestamps...)
mlog.Debug(ctx, "pipeline fetch insert msg",
mlog.Int64("collectionID", iNode.collectionID),
mlog.Int64("segmentID", msg.SegmentID),
mlog.Int("insertRowNum", len(pks)),
mlog.Uint64("timestampMin", msg.BeginTimestamp),
mlog.Uint64("timestampMax", msg.EndTimestamp))
}
func (iNode *insertNode) Close() {
iNode.functionStore.Close()
}
// Insert task
func (iNode *insertNode) Operate(in Msg) Msg {
metrics.QueryNodeWaitProcessingMsgCount.WithLabelValues(paramtable.GetStringNodeID(), metrics.InsertLabel).Dec()
nodeMsg := in.(*insertNodeMsg)
if len(nodeMsg.insertMsgs) > 0 {
sort.Slice(nodeMsg.insertMsgs, func(i, j int) bool {
return nodeMsg.insertMsgs[i].BeginTs() < nodeMsg.insertMsgs[j].BeginTs()
})
ctx := nodeMsg.insertMsgs[0].TraceCtx()
collection := iNode.manager.Collection.Get(iNode.collectionID)
if collection == nil {
mlog.Error(ctx, "insertNode with collection not exist", mlog.Int64("collection", iNode.collectionID))
panic("insertNode with collection not exist")
}
collection.WithInsertSchemaTransition(func(schema *schemapb.CollectionSchema) {
functionOutputFieldIDs, err := iNode.functionStore.OutputFieldIDs(schema)
if err != nil {
mlog.Error(ctx, "failed to get embedding output fields", mlog.String("channel", iNode.channel), mlog.Err(err))
panic(err)
}
insertDatas := make(map[UniqueID]*delegator.InsertData)
for _, msg := range nodeMsg.insertMsgs {
if len(functionOutputFieldIDs) > 0 && !function.HasAllFieldDataByID(msg.GetFieldsData(), functionOutputFieldIDs) {
if err := iNode.functionStore.FillEmbeddingData(iNode.collectionID, schema, msg.InsertRequest); err != nil {
mlog.Error(msg.TraceCtx(), "failed to fill embedding data for insert message", mlog.String("channel", iNode.channel), mlog.Err(err))
panic(err)
}
}
iNode.addInsertData(insertDatas, msg, schema)
}
iNode.delegator.ProcessInsert(insertDatas)
})
}
metrics.QueryNodeWaitProcessingMsgCount.WithLabelValues(paramtable.GetStringNodeID(), metrics.DeleteLabel).Inc()
return &deleteNodeMsg{
deleteMsgs: nodeMsg.deleteMsgs,
timeRange: nodeMsg.timeRange,
schema: nodeMsg.schema,
schemaBarrierTs: nodeMsg.schemaBarrierTs,
}
}
func newInsertNode(
collectionID UniqueID,
channel string,
manager *DataManager,
delegator delegator.ShardDelegator,
schema *schemapb.CollectionSchema,
maxQueueLength int32,
) (*insertNode, error) {
iNode := &insertNode{
BaseNode: base.NewBaseNode(fmt.Sprintf("InsertNode-%s", channel), maxQueueLength),
collectionID: collectionID,
channel: channel,
manager: manager,
delegator: delegator,
functionStore: function.NewFunctionRunnerLocalStore(),
}
if _, err := iNode.functionStore.OutputFieldIDs(schema); err != nil {
return nil, err
}
return iNode, nil
}
func (iNode *insertNode) appendBM25Stats(iData *delegator.InsertData, msg *InsertMsg, schema *schemapb.CollectionSchema) error {
outputFieldIDs, err := getBM25OutputFieldIDs(schema)
if err != nil {
return err
}
if len(outputFieldIDs) == 0 {
return nil
}
if iData.BM25Stats == nil {
iData.BM25Stats = make(map[int64]*storage.BM25Stats)
}
for _, outputFieldID := range outputFieldIDs {
outputData := getFieldData(msg.FieldsData, outputFieldID)
if outputData == nil {
return merr.WrapErrFunctionFailedMsg("BM25 output field %d not found in insert message", outputFieldID)
}
if err := appendBM25StatsFromFieldData(iData.BM25Stats, outputFieldID, outputData); err != nil {
return err
}
}
return nil
}
func getBM25OutputFieldIDs(schema *schemapb.CollectionSchema) ([]int64, error) {
outputFieldIDs := make([]int64, 0)
for _, fn := range schema.GetFunctions() {
if fn.GetType() != schemapb.FunctionType_BM25 {
continue
}
outputField := typeutil.GetFunctionOutputField(schema, fn)
if outputField == nil {
return nil, merr.WrapErrFunctionFailedMsg("function %s output field not found", fn.GetName())
}
outputFieldIDs = append(outputFieldIDs, outputField.GetFieldID())
}
return outputFieldIDs, nil
}
func appendBM25StatsFromFieldData(stats map[int64]*storage.BM25Stats, outputFieldID int64, fieldData *schemapb.FieldData) error {
sparseArray := fieldData.GetVectors().GetSparseFloatVector()
if sparseArray == nil {
return merr.WrapErrFunctionFailedMsg("BM25 output field %d is not sparse float vector data", outputFieldID)
}
if _, ok := stats[outputFieldID]; !ok {
stats[outputFieldID] = storage.NewBM25Stats()
}
stats[outputFieldID].AppendBytes(sparseArray.GetContents()...)
return nil
}
func getFieldData(datas []*schemapb.FieldData, fieldID int64) *schemapb.FieldData {
for _, data := range datas {
if data.GetFieldId() == fieldID {
return data
}
}
return nil
}