Repository navigation
DEV-1230: Fix TDB1 write-back race that loses committed data - #5
Merged
Merged
Conversation
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
beginInternal() called begin$() (synchronized) and then noteTxnStart() after the monitor had been released. A transaction finishing in that gap (notifyCommit/notifyAbort are synchronized) observed activeReaders == 0 && activeWriters == 0 and ran processDelayedReplayQueue, which clears the queued transactions' BlockMgrJournal write sets and then replays the journal into the base dataset. The new transaction, built on the last queued transaction's view (determineBaseDataset), meanwhile read blocks that were in neither place and saw stale, pre-commit content. For a writer this surfaces as "Secondary index duplicate: GSPO->..." on a later add, and once such a writer commits its stale blocks the loss is durable: committed quads vanish from some indexes, or node table lookups fail. Move noteTxnStart() under the same monitor as begin$() (beginSync$), and make promoteExec$() synchronized, since READ_COMMITTED promotion reaches it with the same gap. Nothing in the moved code can block: the only lock taken is the base dataset's MRSW lock in READ mode, which no code path in jena-tdb1 takes in WRITE mode. Reproduced with a stress harness on unmodified jena-tdb1 6.2.0 (4 of 4 on-disk runs of 90-150 s corrupted committed data) and deterministically with fault-injection hooks. With this change: jena-tdb1 test suite passes (1005 tests), the deterministic scenario no longer loses data, and the stress harness runs clean, including with promoted transactions. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KmWBMLwuUqWg8pQtHHe8x
cygri
approved these changes
Sep 21, 2026
This was referenced Sep 23, 2026
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.
Internal patched build of Apache Jena 6.2.0, published as
6.2.0-tq-1.Two commits: the POM version bump across all modules, then Richard's fix cherry-picked unchanged from
tdb1-txn-start-race(4634b5c). It applied to the 6.2.0 release tag without conflicts.The fix records a transaction's start under the same monitor that created it, so a reader finishing in that window can no longer replay the journal underneath a starting transaction and cause committed quads to be lost.
Validation on this branch: jena-tdb1 suite 1005 tests, 0 failures, matching the baseline on the ticket. A deterministic regression test in TopBraid-Suite fails 20/20 against stock 6.2.0 (180 of 240 committed quads lost) and passes 20/20 against this build.
Not for upstream yet — per DEV-1230 we use it internally first.
Intending to merge this into
base-jena-6.2.0rather than close it and reference the tag, following the 5.6.0-tq-1 precedent. That way a later-tq-Nbranched off the base inherits the fix instead of silently losing it. Thejena-6.2.0-tq-1tag gets created either way.🤖 Generated with Claude Code