[spark] Clean up orphan files left by Spark speculative execution - #8927
Open
hbgstc123 wants to merge 12 commits into
Open
[spark] Clean up orphan files left by Spark speculative execution#8927hbgstc123 wants to merge 12 commits into
hbgstc123 wants to merge 12 commits into
Conversation
Contributor
|
cc @Zouxxyy |
Contributor
|
The speculative loser cleanup only deletes the main data files, not |
added 8 commits
August 10, 2026 01:41
Use explicit java.util.function.Consumer instead of Scala lambdas when calling TableWriteImpl.prepareCommit so Scala 2.12/2.13 can resolve the overload correctly.
Use DataFileMeta.collectFiles in FileStoreCommitImpl.abort so blobref, vector, and row sidecar files are cleaned up alongside the main data file.
Register partial commit messages before maintainer work that can fail, so task-side cleanup can abort already-drained data files.
Use DataFileMeta.collectFiles in AppendOnlyWriter.close and extend tests to cover bloom-filter sidecar cleanup for flushed but uncommitted files.
hbgstc123
force-pushed
the
spark-speculative-writer-cleanup
branch
from
August 9, 2026 17:43
d5621a0 to
f7517be
Compare
added 4 commits
August 10, 2026 02:20
Convert prepareCommit results to immutable Seq before registerPrepared and return, fixing Scala 2.13 compile errors in paimon-spark-common_2.13.
…leanup Split prepareCommit paths so Spark onPrepared runs before maintainers while Flink and other 2-arg callers still materialize CommitMessage after maintainer work, preserving idle writer cleanup in WriterOperatorTest.
TableWriteImpl routes Flink prepareCommit through the 3-arg overload with a null callback, so idle writer cleanup must not use the Spark maintainer path unless onPrepared is non-null.
When onPrepared is null, call FileStoreWrite.prepareCommit directly so Flink idle writer cleanup keeps the legacy maintainer ordering instead of the Spark 3-arg path.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Problem
When Spark speculative execution is enabled, Paimon Spark writers use
useCommitCoordinator=false. Multiple task attempts can all finish writing and callprepareCommit(), but Spark only commits the winner'sWriterCommitMessage. Loser attempts leave data/index files on disk that are never referenced by any snapshot — orphan files.Why it happens
Spark kills the loser attempt by interrupting the task thread. Paimon writers do not abort prepared files on interruption today, and core abort/delete paths can fail with
InterruptedIOExceptionwhile the thread is interrupted, so loser files are often left behind until orphan cleanup runs (if ever).What this PR does
SparkAttemptCleanupto detect interruption and abort unprepared/prepared files before the attempt returns successfully.FileIO.deleteQuietlyIgnoringInterrupt,FileStoreCommitImpl.abortindex cleanup, writer abort executors).prepareCommitso a mid-attempt kill can still abort already-finished files.Tests
SparkAttemptCleanupTest,DataEvolutionTableDataWriteTestFileStoreCommitAbortTest,FileIOTest,DeletionVectorsIndexFileTestSparkSpeculativeWriteITCase(DSv2/V1 insert with speculation; loser cleanup and orphan-clean fallback)