Wait for MARKER_RECORDED to fire version callback on replay#2821
Merged
Conversation
eamsden
changed the title
eamsden
force-pushed
the
eamsden/interleaved-UpdateCompleted-getversion
branch
from
ff59a05 to
43ab9df
Compare
mjameswh
reviewed
Mar 31, 2026
mjameswh
reviewed
Mar 31, 2026
eamsden
force-pushed
the
eamsden/interleaved-UpdateCompleted-getversion
branch
2 times, most recently
from
31e4195 to
316a40b
Compare
Contributor
Author
|
Converting to draft until at least the shut-off of |
Add a Java replay test that mirrors the Kotlin GreetingWorkflow sample and replays the provided workflow history fixture. Assert that replay fails with the embedded TMPRL1100 NonDeterministicException message so the reproducer stays pinned to the reported failure mode.
Run a constrained experiment for the interleaved update replay bug by changing VersionStateMachine replay timing only for histories with SKIP_YIELD_ON_VERSION set. In that path, getVersion still returns synchronously, but the replay callback is no longer fired at fake RECORD_MARKER command creation and is instead delayed until the real MARKER_RECORDED event is matched. The goal of the experiment was to verify that flagged histories do not depend on the current early replay callback or its extra eventLoop scheduling. The legacy interleaved update repro history does not have SKIP_YIELD_ON_VERSION, so it continues to fail unchanged and serves as the control case. Verified with: ./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingRemoveTest" --tests "io.temporal.workflow.versionTests.GetVersionRemovedInReplayTest" --tests "io.temporal.workflow.versionTests.GetVersionWithoutCommandEventTest" --tests "io.temporal.workflow.versionTests.GetVersionAndTimerTest" --tests "io.temporal.workflow.versionTests.GetVersionMultipleCallsTest" --tests "io.temporal.workflow.versionTests.GetVersionInSignalTest" --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingTest" --tests "io.temporal.workflow.versionTests.GetVersionInterleavedUpdateReplayTest" --tests "io.temporal.internal.replay.GetVersionInterleavedUpdateReplayTaskHandlerTest" ./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionRemovedInReplayTest" --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingRemoveTest" --tests "io.temporal.workflow.versionTests.GetVersionMultipleCallsTest" --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingTest"
Change VersionStateMachine replay semantics so getVersion no longer resumes workflow code when the fake RECORD_MARKER command is created. Replay now waits until the real MARKER_RECORDED event is matched before firing the version callback, which makes version-marker ordering consistent with replayed side effects. This fixes the interleaved update replay bug reproduced by testGetVersionInterleavedUpdateReplayHistory.json. That history previously failed replay with [TMPRL1100] because the second getVersion callback ran ahead of update completion protocol handling. After this change, the same recorded history replays successfully through both WorkflowReplayer and the lower-level direct-query replay task handler. The earlier flag-gated experiment showed that delaying the callback was already safe for histories with SKIP_YIELD_ON_VERSION. This commit removes that temporary gating and applies the same replay ordering to all histories. Verified with: ./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingRemoveTest" --tests "io.temporal.workflow.versionTests.GetVersionRemovedInReplayTest" --tests "io.temporal.workflow.versionTests.GetVersionWithoutCommandEventTest" --tests "io.temporal.workflow.versionTests.GetVersionAndTimerTest" --tests "io.temporal.workflow.versionTests.GetVersionMultipleCallsTest" --tests "io.temporal.workflow.versionTests.GetVersionInSignalTest" --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingTest" --tests "io.temporal.workflow.versionTests.GetVersionInterleavedUpdateReplayTest" --tests "io.temporal.internal.replay.GetVersionInterleavedUpdateReplayTaskHandlerTest" ./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionInterleavedUpdateReplayTest" --tests "io.temporal.internal.replay.GetVersionInterleavedUpdateReplayTaskHandlerTest"
`VERSION_WAIT_FOR_MARKER`
have elected not to fix that specific history. Make sure there is a run-then-replay regression test which confirms new histories won't be broken that way.
`sdkMetadata.langUsedFlags`
eamsden
force-pushed
the
eamsden/interleaved-UpdateCompleted-getversion
branch
from
316a40b to
648d583
Compare
current deployment version
mjameswh
reviewed
Jun 8, 2026
| @@ -1,5 +1,6 @@ | |||
| name: Continuous Integration | |||
| permissions: | |||
| checks: write | |||
Contributor
Author
There was a problem hiding this comment.
It was part of fixing CI flakes. I think in this case It was for
to reliably succeed, but I'm checking that recall.
Contributor
Author
There was a problem hiding this comment.
It turns out not to be necessary, sorry this slipped through.
mjameswh
reviewed
Jun 8, 2026
Co-authored-by: James Watkins-Harvey <mjameswh@users.noreply.github.com>
mjameswh
approved these changes
Jun 8, 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.
What was changed
VersionStateMachinereplay behavior behind a new SDK flagVERSION_WAIT_FOR_MARKERsogetVersion()no longer resumes workflow code when the fakeRECORD_MARKERcommand is created during replay.MARKER_RECORDEDevent is matched before firing the version callback.Why?
Replay was resuming
getVersion()too early relative to update completion protocol handling. In the interleaved update history from issue#2796, that allowed replay to queue the second version marker before the update completion work had been matched, which produced a replay-only nondeterminism error:[TMPRL1100] getVersion call before the existing version marker eventDelaying the replay callback until the actual marker event is matched makes version replay ordering consistent with replayed side effects and fixes replay for the recorded history.
Checklist
Closes Replay fails: UpdateCompleted events between version markers breaks replay #2796 but does not fix replay of existing histories.
How was this tested:
./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingRemoveTest" --tests "io.temporal.workflow.versionTests.GetVersionRemovedInReplayTest" --tests "io.temporal.workflow.versionTests.GetVersionWithoutCommandEventTest" --tests "io.temporal.workflow.versionTests.GetVersionAndTimerTest" --tests "io.temporal.workflow.versionTests.GetVersionMultipleCallsTest" --tests "io.temporal.workflow.versionTests.GetVersionInSignalTest" --tests "io.temporal.workflow.versionTests.GetVersionMultithreadingTest" --tests "io.temporal.workflow.versionTests.GetVersionInterleavedUpdateReplayTest" --tests "io.temporal.internal.replay.GetVersionInterleavedUpdateReplayTaskHandlerTest"./gradlew --offline :temporal-sdk:test --tests "io.temporal.workflow.versionTests.GetVersionInterleavedUpdateReplayTest" --tests "io.temporal.internal.replay.GetVersionInterleavedUpdateReplayTaskHandlerTest"WorkflowReplayerand the lower-level direct-query replay task handler.