zhaoyudi-creator commented on issue #19907: URL: https://github.com/apache/hudi/issues/19907#issuecomment-5660555841
> Fixes that might be related: > > [#5185](https://github.com/apache/hudi/pull/5185) [#5763](https://github.com/apache/hudi/pull/5763) [#7433](https://github.com/apache/hudi/pull/7433) [#8263](https://github.com/apache/hudi/pull/8263) Thanks for the reply. I've reproduced this on **1.1** and can confirm the race still exists — it doesn't appear to be covered by any existing mechanism, so this looks like a genuine bug rather than a known/handled case. ## Reproduction Steps 1. Perform the **first write to an empty partition**, and stop the job after the checkpoint succeeds but before the instant finishes committing (this creates an inflight instant that needs recommit). 2. Restart the job. The **JM-side recommit completes *after* the TM-side `bootstrapIndexIfNeed`**, so the write task mints a brand-new fileId while the committed view is still empty. ## Key Logs ``` 2026-09-14 14:47:57.461 INFO [Sink: bucket_write(t:p) (2/2)#0] BucketStreamWriteFunction - Loading Hoodie Table test_bucket_field_dep, with path .../test_bucket_field_dep/2026-09-13 # TM bootstrapIndexIfNeed starts 2026-09-14 14:47:57.463 INFO [meta-event-handle] StreamWriteOperatorCoordinator - Recommit instant 20260914144335797 # JM starts recommitting the old instant 2026-09-14 14:49:58.131 INFO [meta-event-handle] BaseHoodieWriteClient - Committed 20260914144335797 # old instant (fileId-A) committed, ~2 min AFTER bootstrap 2026-09-14 14:50:01.584 INFO [meta-event-handle] BaseHoodieWriteClient - Committed 20260914144958570 # new instant (fileId-B) committed ``` Meanwhile, the following two log lines **never appear**: ``` Should load this partition bucket %s with fileId %s Adding fileId %s to the bucket %s of partition ``` These are only printed when `bootstrapIndexIfNeed` reads a fileId for the bucket from the committed view. Their absence is direct proof that **the committed view was empty at bootstrap time and fileId-A was invisible**, so the write task minted fileId-B in `defineRecordLocation`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
