Fix: Disable partial retries + correct RpcContext handling to resolve inconsistent reads when using statement.getMoreResults() (#29451) #29455
+95
−102
Conversation
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
📌 Summary This PR fixes a heap-use-after-free detected by ASAN in yb::YBThreadPool::Impl::NotifyWorker, where a Worker could be freed while another thread was concurrently accessing it through waiting_workers.Pop(). 🐞 Root Cause Multiple threads could concurrently pop from waiting_workers. A Worker in IdleStop state could be erased and deleted while another thread was still reading it. This resulted in nondeterministic crashes under ASAN and caused the following test to fail: CDCSDKConsumptionConsistentChangesTest. TestLSNDeterminismWithSpecialRecordOnRestartWithPartialAck 🔧 Fix Added waiting_workers_active_pops counter in ThreadPoolShare to track active Pop() operations. Deferred deletion of Worker objects if pops are active, using a new deferred_deletes_ list. Ensured Shutdown() waits for all pops to finish before freeing deferred workers. ✅ Validation Ran ybd asan --cxx-test integration-tests_cdcsdk_consumption_consistent_changes-test — no more heap-use-after-free crashes. Verified that normal enqueue/dequeue behavior is unaffected. Checked that no leaks remain (all deferred deletes are flushed in shutdown). 📊 Impact Fixes flaky ASAN test failures in DocDB thread pool. Minimal overhead: adds two atomic ops per worker notification. No API changes. 🔗 References Fixes: yugabyte#28297 Jira: DB-17979
Adityakk9031
changed the title
✅ Deploy Preview for infallible-bardeen-164bc9 ready!Built without sensitive environment variables
To edit notification comments on pull requests, go to your Netlify project configuration. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Jira Issue: DB-19247
Issue Type: Bug
Area: YSQL
🧩 Issue Summary (from #29451)
This PR fixes inconsistent reads caused when using the JDBC
getMoreResults()API in YSQL workloads. The issue resulted in missing or duplicated row results during multi-resultset queries inside the consistency validation workload.The same workload passes in PostgreSQL but fails in YugabyteDB with off-by-one mismatches such as:
This happens because:
RpcContextwas passed incorrectly by value, causing dropped results during multi-result RPC execution.🔧 Root Cause
getMoreResults()expects all resultsets to remain intact.PgClientServiceImpl::Perform()incorrectly passedRpcContextby value, causing the context to go out of scope while async work was still using it.✅ Fix Summary
This PR contains two critical fixes:
1. Disable Partial-Restart Retries for Multi-Result or Read Batches
Added logic in
PgSession::Perform():disable_retries_on_restarts = true.This prevents tserver from replaying only a subset of the statements, ensuring all resultsets remain consistent across retries.
2. Correct
RpcContextPassing inPgClientServiceImpl::Perform()Updated function signature to pass
rpc::RpcContext*instead of a temporary stack value.This ensures the RPC context remains valid for the duration of async execution.
📝 Files Changed
src/yb/yql/pggate/pg_session.cc— Added detection logic + new Perform option flag.src/yb/yql/pggate/pg_client_service.cc— ChangedPerform()to correctly passRpcContext.🧪 Testing
getMoreResults().