Switch the unbounded thread pool to use a latch instead of parkers #2846
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
stevenengler
force-pushed
the
simplify-scheduler
branch
3 times, most recently
from
1b5b8f0
to
e1d57a1
Compare
stevenengler
force-pushed
the
simplify-scheduler
branch
3 times, most recently
from
eaa3d9e
to
5204be9
Compare
stevenengler
changed the title
sporksmith
approved these changes
Apr 11, 2023
stevenengler
force-pushed
the
simplify-scheduler
branch
from
5204be9
to
fd3ca74
Compare
stevenengler
added a commit
that referenced
this pull request
Apr 12, 2023
The latest nightly tor benchmark showed a huge runtime performance change:  @sporksmith found that we were performing futex operations on the `Arc` address rather than its inner `AtomicI32` address, causing the waiter to spin. This was caused by a bug in #2846. We expect that this PR will increase the runtime back to the original runtime, and we can investigate spinning properly later.
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Outdated suggestions cannot be applied.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.
I'm not sure that the
ThreadUnparkerworks correctly when using rust's parking API, so I've updated a comment in it and replaced it in the thread-per-core scheduler (the unbounded thread pool) with a simple futex-based latch instead. The thread-per-host scheduler continues to use the (possibly broken)ThreadUnparker.Another advantage is that it can futex-wake all threads with one syscall, rather than unparking each thread individually.
I ran a benchmark on an earlier version of this PR and there was no performance improvement.