Make result queue poll for shutdown, and tidy up at shutdown #3709
Conversation
This file contains 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
This poll happens at the configured htex poll period, which defaults to 10ms. Under heavy result load, this shoudn't result in much additional load: the poll loop will already be looping a lot to process the results. Under lower result load, there is a slight observable increase in CPU usage: a 30second sleep task shows this before this PR: before: real 0m37.451s user 0m2.160s sys 0m0.376s run 2, user 2.160s run 3, user 2.116s and this after this PR: real 0m37.473s user 0m2.400s sys 0m0.557s Run 2, 2.457s Run 3, 2.452s At shutdown, the ZMQ socket for incoming results is closed. This reduces both the number of threads and number of file descriptors left behind by the `--config local` tests. For example: $ pytest parsl/tests/test_monitoring/ --config local Before this PR, at end of test: 32 threads, 451 fds open. After this PR, at end of test: 1 thread, 48 fds open. This is part of PR #3397 shutdown tidyup.
khk-globus
approved these changes
Dec 2, 2024
Comment on lines
-460
to
+465
| msgs = self.incoming_q.get() | ||
| msgs = self.incoming_q.get(timeout_ms=self.poll_period) | ||
| if msgs is None: # timeout | ||
| continue |
There was a problem hiding this comment.
Oh good. This exact procedure has been on my mind for quite awhile, so thank you for this!
I'm thinking at some point we can go farther and simply block completely, but I haven't fully gotten through the shutdown logic for this route yet. But "on my mind," at some point.
benclifford
added a commit
that referenced
this pull request
Jan 14, 2025
This was introduced in PR #2965 to help debug ZMQ hangs. PR #3709 changed this loop to run every poll period (by default, every 10ms) rather than every result message. That turns this log line into a huge source of log noise/volume. This PR removes that log line - ZMQ debugging must proceed without it. parsl.log size from: pytest parsl/tests/ --config parsl/tests/configs/htex_local.py before this PR: 7 megabytes after this PR: 2 megabytes
github-merge-queue bot
pushed a commit
that referenced
this pull request
Jan 14, 2025
This was introduced in PR #2965 to help debug ZMQ hangs. PR #3709 changed this loop to run every poll period (by default, every 10ms) rather than every result message. That turns this log line into a huge source of log noise/volume. This PR removes that log line - ZMQ debugging must proceed without it. # Changed Behaviour parsl.log size from: pytest parsl/tests/ --config parsl/tests/configs/htex_local.py before this PR: 7 megabytes after this PR: 2 megabytes ## Type of change - Bug fix
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This poll happens at the configured htex poll period, which defaults to 10ms.
Under heavy result load, this shoudn't result in much additional load: the poll loop will already be looping a lot to process the results.
Under lower result load, there is a slight observable increase in CPU usage: a 30second sleep task shows this before this PR:
before:
real 0m37.451s
user 0m2.160s
sys 0m0.376s
run 2, user 2.160s
run 3, user 2.116s
and this after this PR:
real 0m37.473s
user 0m2.400s
sys 0m0.557s
Run 2, 2.457s
Run 3, 2.452s
At shutdown, the ZMQ socket for incoming results is closed.
This reduces both the number of threads and number of file descriptors left behind by the
--config localtests. For example:$ pytest parsl/tests/test_monitoring/ --config local
Before this PR, at end of test: 32 threads, 451 fds open.
After this PR, at end of test: 1 thread, 48 fds open.
This is part of PR #3397 shutdown tidyup.
Description
Please include a summary of the change and (optionally) which issue is fixed. Please also include
relevant motivation and context.
Changed Behaviour
nothing should be really visible to normal users. Increased CPU usage in the above documented situations.
Type of change