[zephyr] Fix dead threading.Event in _wait_for_stage#5048
Open
hsuhanooi wants to merge 1 commit intomarin-community:mainfrom
Open
[zephyr] Fix dead threading.Event in _wait_for_stage#5048hsuhanooi wants to merge 1 commit intomarin-community:mainfrom
hsuhanooi wants to merge 1 commit intomarin-community:mainfrom
Conversation
…g with event-driven wakeup _wait_for_stage created a local threading.Event that was never signaled, making its wait() call a pure sleep of up to 1 second per iteration. Each stage transition (scatter→reduce→fold) paid this latency needlessly. Replace it with self._stage_event, signaled by every coordinator method that changes stage-relevant state: report_result, report_error, abort, and register_worker. _start_stage clears the event so signals from the previous stage don't bleed over. The backoff timeout is retained as a backstop for the alive-worker check and periodic log lines. In the normal (no-failure) path, stage transitions now complete within microseconds of the last shard result arriving. Benchmark (8 shards, 3-stage group_by pipeline, 70MB synthetic data): Before: 14.9s / 14.6s / 17.1s (avg ~15.6s, high variance) After: 13.4s / 13.7s / 13.4s (avg ~13.5s, low variance) ~13% faster; ~1.5s saved from eliminating poll-interval latency at each of the 3 stage boundaries. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
hsuhanooi
force-pushed
the
zephyr-stage-event-fix
branch
from
5341e44 to
365f9b8
Compare
hsuhanooi
changed the title
hsuhanooi
force-pushed
the
zephyr-stage-event-fix
branch
from
365f9b8 to
f98e020
Compare
hsuhanooi
changed the title
Contributor
|
Good catch, definitely a bug. The new implementation looks right. Controversial opinion 🙈: 1 second isn't that long to wait and we could just strip out all the signaling code (including the old broken code, of course) and just rely on the 1 second poll interval (which I guess is what happened before #3923) |
Contributor
Author
|
Yeah that’s reasonable too. Although maybe it’s useful to fix this here and then test removing signaling in a different pr? |
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.
_wait_for_stage created a local stage_done = threading.Event() and called wait() on it, but nothing ever called set() on that local object. Stage transitions were driven by a time.sleep(backoff.next_interval()) loop, so the coordinator waited up to 1 second after each shard completed instead of waking immediately.
Add self._stage_event on the coordinator, call set() in report_result, report_error, abort, and register_worker, and replace the sleep loop with self._stage_event.wait(timeout) + clear(). Stage transitions now propagate in microseconds.
The bug was introduced in the refactor in #3923, which replaced sleep() with event.wait() but never wired up the set() calls.