[ALT] core: use a separate syscall thread for aio fallbacks - #299
Open
ballard26 wants to merge 1 commit into
Open
[ALT] core: use a separate syscall thread for aio fallbacks#299ballard26 wants to merge 1 commit into
ballard26 wants to merge 1 commit into
Conversation
There was a problem hiding this comment.
Pull request overview
This PR reduces latency interference between unrelated blocking syscalls and linux-aio io_submit() fallback submissions by splitting the reactor’s syscall offload work into two thread pools when the linux-aio backend is selected.
Changes:
- Create a dedicated
_aio_thread_poolfor linux-aio to serviceaio_fallbacksubmissions separately from other blocking syscalls. - Route
io_threaded_fallbacksmetrics to the correct thread pool depending on the submit reason. - Extend the syscall poller to complete/interrupt-manage both thread pools when present.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
src/core/reactor.cc |
Conditionally creates and services a second thread pool for AIO fallbacks; updates metrics routing and polling/interrupt handling. |
src/core/reactor_backend.cc |
Routes AIO retry submissions (io_submit() fallbacks) to the AIO-dedicated syscall thread for linux-aio. |
include/seastar/core/reactor.hh |
Adds the _aio_thread_pool member with clarifying comments. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
ballard26
force-pushed
the
aio-syscall-thread-alt
branch
from
July 15, 2026 16:56
44971ee to
f505665
Compare
The reactor offloads blocking syscalls onto a single syscall thread. Disk aio submissions can fall back to a blocking io_submit() (in aio_storage_context, used by both the linux-aio and epoll backends), so a slow or stalled file or process syscall (e.g. open, stat, fsync, waitpid) can delay the aio retries queued behind it, and vice versa. Give aio_storage_context its own syscall thread for that io_submit() fallback, so aio submissions no longer share a queue with file and process operations. The pool lives wherever an aio_storage_context does (the linux-aio and epoll backends) and is exposed to the reactor's syscall poller via reactor_backend::aio_thread_pool(); backends without disk aio (io_uring) have none.
ballard26
force-pushed
the
aio-syscall-thread-alt
branch
from
July 16, 2026 02:19
f505665 to
97a443c
Compare
travisdowns
reviewed
Jul 20, 2026
| if (auto* aio_pool = _r._backend->aio_thread_pool()) { | ||
| n += aio_pool->complete(); | ||
| } | ||
| return n; |
Member
There was a problem hiding this comment.
shouldn't this be its own poller added by the aio backend?
travisdowns
reviewed
Jul 20, 2026
travisdowns
left a comment
travisdowns
left a comment
Member
There was a problem hiding this comment.
Seems like maybe we should still bit it "more" into the aio backend? E.g., now we expand the existing poller in the reactor to know about the aio poller, which is backend specific, but maybe it should be a new poller which is registered by the aio backend.
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
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.
The reactor offloads blocking syscalls onto a single syscall thread. On the linux-aio backend this thread also serves aio read/write submissions that fall back to a blocking io_submit(), so a slow or stalled file or process syscall (e.g. open, stat, fsync, waitpid) can delay aio retries queued behind it, and vice versa.
Split the work when the linux-aio backend is in use. A dedicated syscall thread and queue now serve aio submission fallbacks, while the existing thread keeps handling every other blocking syscall. Submissions are routed by their thread_pool_submit_reason, so aio_fallback work no longer shares a queue with file and process operations. Other backends, which have no aio fallback path, keep using a single worker as before.