mirror of
https://github.com/milvus-io/milvus.git
synced 2026-07-21 10:15:43 +00:00
yihao.daiandGitHub
96f05a0dd8
## Summary Fixes #49645 Bulk import can hit `unexpected EOF` when a remote object stream stays idle during function execution and the object-store reader returns a premature EOF. This PR makes sequential import readers recover by reopening the object reader at the last successful offset when EOF happens before the known object size. ## Changes - Add reopen-and-retry support to `RetryableReader` for retryable read errors and premature EOF. - Use `ChunkManager.Size` as a fallback object-size source for readers whose `Size()` is lazy or unreliable before reading. - Return unexpected EOF, not clean EOF, when premature EOF retries are exhausted. - Enable the reopen path for JSON, CSV, and Numpy import readers. - Keep Parquet unchanged because Arrow uses `ReadAt`/`Seek`, which this sequential wrapper does not cover. ## Test Plan - [x] `make static-check` - [x] `git diff --check` - [x] `go test ./internal/util/importutilv2/common ./internal/util/importutilv2/json ./internal/util/importutilv2/csv ./internal/util/importutilv2/numpy ./internal/util/importutilv2/parquet -count=1 -ldflags="-r ${RPATH}"` --------- Signed-off-by: Yihao Dai <yihao.dai@zilliz.com>