Epoll

[sgerbino]

Summary by CodeRabbit

• Bug Fixes

  • Improved handling of non-blocking socket readiness and cancellation to reduce missed or delayed I/O completions and ensure more reliable error propagation.

• Chores

  • Consolidated internal event-processing and operation-routing logic to improve stability, predictability, and efficiency of asynchronous socket operations.

[coderabbitai]

📝 Walkthrough

Walkthrough

Consolidates epoll readiness and op handling: descriptor mutex is acquired unconditionally and connect/read/write ops are processed or reassigned under the guard; EAGAIN/EWOULDBLOCK is normalized, cancellations post/finish deterministically, and completed ops are moved to a local queue for scheduling.

Changes

Cohort / File(s)	Summary
Epoll socket I/O src/corosio/src/detail/epoll/sockets.cpp	Replaced per-operation perform_now checks with a single mutex-guarded path that handles readiness, EAGAIN/EWOULDBLOCK normalization, cancellation, and reassignment of connect/read/write ops in-place.
Epoll acceptor src/corosio/src/detail/epoll/acceptors.cpp	Removed split perform_now logic: acquire descriptor mutex, inline perform_io when read_ready, handle EAGAIN/cancellation explicitly, and store or post read_op under the guard.
Epoll scheduler / descriptor state src/corosio/src/detail/epoll/scheduler.cpp	Consolidated EPOLLIN/EPOLLOUT/error paths to process read_op, connect_op, and write_op under the mutex; normalize retries, drain/complete ops on error, and move completed ops to a local queue via pointer exchanges.

Sequence Diagram(s)

mermaid
sequenceDiagram
participant Client as Client / Socket
participant Descriptor as descriptor_state
participant Scheduler as Scheduler / Reactor
participant LocalOps as LocalOps Queue
Client->>Descriptor: epoll event (EPOLLIN / EPOLLOUT / ERR)
Descriptor->>Descriptor: lock mutex; inspect read_op/write_op/connect_op and ready flags
alt op exists and ready
Descriptor->>Client: call op.perform_io()
Client-->>Descriptor: result (ok / EAGAIN / error / cancelled)
alt EAGAIN
Descriptor->>Descriptor: normalize errno; set *_ready or reassign op pointer
else cancelled or error
Descriptor->>LocalOps: exchange pointer → enqueue (post/finish)
else ok
Descriptor->>LocalOps: move completed op to queue
end
else no op ready
Descriptor->>Descriptor: set read_ready/write_ready
end
Descriptor->>Scheduler: unlock mutex
Scheduler->>LocalOps: schedule/dispatch enqueued ops

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• Epoll optimizations #106 — Refactors epoll descriptor/state and op lifecycle; touches the same readiness, locking, and op reassignment paths.
• Add portable select() reactor backend for POSIX platforms #87 — Alters epoll sockets/acceptor/scheduler interactions and readiness/cancellation semantics overlapping these changes.
• Add speculative completion fast path for socket I/O #113 — Changes socket I/O fast-paths for read/write/connect that closely relate to the updated control flow here.

Poem

🐰 In a burrow of sockets I hop and spin,
I grab the mutex, tidy ops within,
If EAGAIN nudges, I pause and then queue,
Cancel or finish — I post it through,
Hopping events forward, neat and trim 🥕

🚥 Pre-merge checks | ✅ 1 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Title check	❓ Inconclusive	The pull request title 'Epoll' is too vague and does not clearly convey the main objective of the changes, which is to fix a lost-wakeup window in epoll event handling.	Consider a more descriptive title such as 'Fix lost-wakeup window in epoll edge-triggered mode' or 'Add edge-triggered retry loop to epoll perform_now paths' to better reflect the actual changes.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

No actionable comments were generated in the recent review. 🎉

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[cppalliance-bot]

An automated preview of the documentation is available at https://116.corosio.prtest3.cppalliance.org/index.html

If more commits are pushed to the pull request, the docs will rebuild at the same URL.

2026-02-10 01:51:17 UTC

[coderabbitai]

Actionable comments posted: 3

🤖 Fix all issues with AI agents

In `@src/corosio/src/detail/epoll/sockets.cpp`:
- Around line 376-395: The do_write_io retry loop misses a cancellation check
after it re-parks the operation into desc_state_.write_op: after the block that
sets desc_state_.write_ready = false / desc_state_.write_op = &op and breaks,
add the same cancellation-check performed elsewhere (check op.cancelled or
whatever cancellation flag the code uses, and if cancelled call svc_.post(&op)
and svc_.work_finished() then return) so we don't skip handling a cancellation
when we re-park the write; locate the logic around op.perform_io(), op.errn,
svc_.post, svc_.work_finished, and desc_state_.write_op in do_write_io and
insert the check immediately after re-parking/breaking from the retry loop.
- Around line 287-306: The read retry loop that calls op.perform_io() and then
parks the operation by setting desc_state_.read_op = &op lacks the same
cancellation check used in the non-perform_now path; before returning after
breaking out of that loop (i.e., after desc_state_.read_op is set and control
would hit the return), check op.cancelled and if true, clear
desc_state_.read_op, call svc_.post(&op) and svc_.work_finished() (matching the
cancellation handling used elsewhere), then return; update the logic around the
loop that sets desc_state_.read_op to mirror the cancel-handling in the other
path so cancelled ops are completed immediately.
- Around line 187-206: After the loop that sets desc_state_.connect_op = &op and
breaks, add the same cancellation re-check used in the non-perform_now path:
after parking the op (when op.errn was EAGAIN/EWOULDBLOCK and
desc_state_.write_ready was false) check op.cancelled under the same
locking/claiming logic and, if cancelled, clear desc_state_.connect_op, post the
op to svc_ and call svc_.work_finished() (or otherwise reclaim/complete the
operation) before returning std::noop_coroutine(); this mirrors the cancel()
claim semantics and prevents stuck ops—look for the connect handling around
op.perform_io(), desc_state_.connect_op and the cancel()/op.cancelled logic to
copy the behavior exactly.

🧹 Nitpick comments (1)

src/corosio/src/detail/epoll/sockets.cpp (1)

187-205: Consider extracting the duplicated retry-loop logic into a helper.

All three retry loops (connect, read, write) follow an identical structure: for(;;) → perform_io() → check errn → post/finish or re-check readiness → park or continue. The only differences are the readiness flag name (write_ready/read_ready) and the op slot (connect_op/read_op/write_op). A small template or lambda helper taking the flag and slot as parameters would eliminate the triplication and reduce the risk of one path diverging from the others (as demonstrated by the cancellation gap above).

Also applies to: 287-305, 376-394

[cppalliance-bot]

GCOVR code coverage report https://116.corosio.prtest3.cppalliance.org/gcovr/index.html
LCOV code coverage report https://116.corosio.prtest3.cppalliance.org/genhtml/index.html
Coverage Diff Report https://116.corosio.prtest3.cppalliance.org/diff-report/index.html

Build time: 2026-02-10 01:54:13 UTC

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In `@src/corosio/src/detail/epoll/scheduler.cpp`:
- Around line 238-255: The error-only drain currently runs only when err && !(ev
& (EPOLLIN | EPOLLOUT)) which can leave write/connect ops stranded if EPOLLERR
arrives with partial direction flags; change the logic that handles err to
unconditionally drain and complete any pending read_op, write_op, and connect_op
when err is set (i.e., check if (err) and for each of read_op, write_op,
connect_op call complete(err, 0) and push via std::exchange as done today), so
all directions are cleared defensively while preserving the existing completion
semantics and use of local_ops.

[codecov]

Codecov Report

❌ Patch coverage is 45.00000% with 44 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.14%. Comparing base (b058adb) to head (6922bdb).
⚠️ Report is 7 commits behind head on develop.

Files with missing lines	Patch %	Lines
src/corosio/src/detail/epoll/sockets.cpp	33.33%	22 Missing ⚠️
src/corosio/src/detail/epoll/scheduler.cpp	58.33%	15 Missing ⚠️
src/corosio/src/detail/epoll/acceptors.cpp	36.36%	7 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #116      +/-   ##
===========================================
+ Coverage    79.70%   80.14%   +0.43%     
===========================================
  Files           65       65              
  Lines         5661     5640      -21     
===========================================
+ Hits          4512     4520       +8     
+ Misses        1149     1120      -29     

Files with missing lines	Coverage Δ
src/corosio/src/detail/epoll/acceptors.cpp	81.85% <36.36%> (+2.18%)	⬆️
src/corosio/src/detail/epoll/scheduler.cpp	80.86% <58.33%> (+0.66%)	⬆️
src/corosio/src/detail/epoll/sockets.cpp	71.46% <33.33%> (+2.60%)	⬆️

... and 3 files with indirect coverage changes

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b058adb...6922bdb. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.