smaheshwar-pltr opened a new pull request, #16825: URL: https://github.com/apache/iceberg/pull/16825
## Core: Serialize commits in `BaseCommitService` Relates to #9521. ### Problem With `partial-progress.enabled=true` and a high `max-concurrent-file-group-rewrites`, rewrites hit many `CommitFailedException` retries on the table. Setting concurrency to `1`, or disabling partial progress, avoids it. ### Cause The single-threaded `committerService` exists to serialize table commits, but commits weren't running on it during rewrites. `offer()` called `commitReadyCommitGroups()` directly, so `commitOrClean()` executed on the calling rewrite thread — and the `synchronized` block guarded only batch *selection*, not the commit itself. Many concurrent rewrites therefore committed to the table in parallel and conflicted. (During the active phase the committer thread only committed once `running` was false, so it sat idle.) This is the same single-committer invariant the Flink sink already enforces, where `FlinkSink` pins `IcebergFilesCommitter` to `.setParallelism(1).setMaxParallelism(1)` — concurrency belongs in the writers, not the committer. Beyond retries, the conflicts make `succeededCommits()` fall short, inflating the `failedCommits` count in `RewriteDataFilesSparkAction`, which can exceed `partial-progress.max-failed-commits` and fail the whole rewrite. ### Fix Commit only on the single committer thread: `offer()` just enqueues, and the committer loop commits ready batches throughout (not just at drain). Commits to the table are now serialized. Batching, partial-progress behavior, internal bookkeeping, and `close()` drain/abort/timeout are unchanged. The public API and the `commitOrClean` / `abortFileGroup` contract are untouched. ### Testing Adds `testCommitsAreSerialized`, which asserts `commitOrClean` is never invoked by more than one thread at once. It fails on the pre-fix code (observes up to N concurrent commits) and passes after. Existing `TestCommitService` cases pass unchanged. 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
