Skip to content

fix: let poll() complete ready I/O and fix related run-loop bugs - #423

Open
sgerbino wants to merge 7 commits into
cppalliance:developfrom
sgerbino:pr/poll-reaps-io
Open

sgerbino wants to merge 7 commits into
cppalliance:developfrom
sgerbino:pr/poll-reaps-io

Conversation

@sgerbino

@sgerbino sgerbino commented Nov 16, 2026 •

Copy link
Copy Markdown
Collaborator

On epoll, select and kqueue, poll() and poll_one() returned 0 when only the reactor sentinel was queued. They never ran the reactor, so ready I/O and expired timers waited for a later run(). They now make one non-blocking pass. A pass that queues nothing wakes a sleeping peer before returning, so the sentinel is not stranded. On epoll, a non-blocking pass also checks the clock for due timers, because the timerfd can fire a little after the deadline.

Fixing that turned up several related run-loop bugs, each in its own commit:

  • Bounded runs (run_for, run_one_for and the _until forms) now block at most once per call, then finish as a poll does. Before, a bounded pass handed back the reactor without waking a sleeping run() thread. A follower woken to take the reactor blocked for a second full slice, so run_for(200ms) could take 350 ms. A handler the pass found was also left for the next call. io_uring had the same overrun.
  • A run nested in a handler now sees the work that handler posted. Posts from a running thread go to its private queue, which a nested poll() could not see, and a nested run_one() could hang. Each run function now takes over the enclosing frame's queue and work count on entry.
  • A handler that throws on io_uring or IOCP now still counts its work as finished. Before, the work count leaked and run() never returned.
  • Docs: on Windows, queued work may still run after stop(), because each run() sees the stop only when it reaches the stop's wakeup. This matches asio's IOCP backend and is now documented rather than changed.

Behaviour change: a bounded run given a zero duration or a deadline that has already passed now returns 0 at once, without running work or stopping the context. Before, it ran one non-blocking pass, so run_for(0ms) worked like poll(). The new behaviour matches asio. Code that used a zero bound as a poll should call poll().

Tests:

  • test/unit/io_context.cpp adds a matrix that runs all 8 run functions through 10 situations. These cover no work, posted and chained work, a parked read (idle and ready), expired and future timers, stopping with work queued, stopping inside a handler, and a throwing handler. Each row asserts every field of the result. Backend differences are gated explicitly, such as the IOCP stop semantics and io_uring consuming the data before a readiness check.
  • bounded_run.cpp covers poll on ready reads and expired timers, and a bounded pass that runs what it found.
  • The nested-run and passed-deadline cases have their own tests.
  • Each new test failed on develop.

Verified locally on Linux: each commit builds and passes the unit suites (220/0), and TSan is clean. A differential harness of 81 scenarios matches asio row for row on epoll, select and io_uring. Develop differed on 7 rows and hung on a nested run_one. The cost is one extra non-blocking syscall per poll() call. run() is unchanged.

The timing races (the lost wakeup and the deadline overrun) have no unit tests, because they depend on timing. Scratch reproducers confirmed both fixes.

Resolves #418

Summary by CodeRabbit

  • Bug Fixes
    • Improved timed-run behavior: calls with expired deadlines or nonpositive durations return immediately without running work or stopping the context.
    • Improved handling of queued work, timers, and bounded waits across supported platforms.
    • Ensured work is marked complete even when a handler or coroutine throws.
  • Documentation
    • Clarified that on Windows, queued work may execute after stop() until each run() call receives the stop wakeup.

…ends

When the reactor's queue entry was the only thing queued, a zero-timeout
call returned without running the reactor. A read parked on a socket that
already had data never completed through poll(), however often it was
called, and a timer past its deadline never fired; the reactor ran only
if some other handler happened to be queued.

A zero timeout now runs a non-blocking reactor pass in that case too. The
pass that ends a poll is the one that queues nothing, so poll() still
returns once no work is ready. The pass runs without the lock, so another
thread may find the queue empty and go to sleep meanwhile; an empty pass
wakes one before returning, so the reactor isn't left with nobody to run
it.

The epoll timerfd is armed relative to when it was set, so it can fire a
little after a deadline. A pass that doesn't block now also checks the
clock, so a timer that is already due fires on that pass.
run_for() and run_one_for() hold the reactor for a bounded pass, then
return without waking anyone. A thread that called run() meanwhile found
the queue empty and went to sleep, so once the bounded call left, nobody
ran the reactor: ready I/O and expired timers waited until some other
thread happened to post work.

A bounded pass now wakes one sleeping thread before returning, as an
empty poll does, so the reactor is taken over. That makes a bounded call
more likely to find the reactor held by another thread, and its wait for
work then started over each time it timed out, never giving the caller a
chance to see its deadline pass.

A bounded call now blocks at most once, in its wait or in a reactor
pass, and then finishes as a poll does. A wake that hands it the reactor
no longer starts a second full block, and a handler its pass found runs
in that call instead of the next one. The io_uring scheduler's bounded
wait blocks once in the same way.
Work a handler posts goes to its thread's private queue and reaches the
shared queue when the handler returns. A poll(), poll_one() or run_one()
called from inside that handler pushed a fresh frame and never saw it:
poll() returned without running it, and run_one() blocked with the work
sitting in the frame below.

Each run and poll entry now moves the enclosing frame's private queue
and its work count to the shared queue first, so the nested call runs
what was posted before it.
run_until() and run_one_until(), and run_for() and run_one_for() through
them, always made one scheduler pass, even when the deadline had already
passed. A late call therefore acted as poll() or poll_one(): it ran
whatever was ready, and stopped the context if no work was left.

A deadline that has passed now returns 0 straight away, leaving queued
work and the context's state alone. Tests that relied on that one pass
now get there another way: those expecting a bounded call to run work or
stop the context use a long bound, which returns as soon as it does, and
those that only start coroutines before a bounded wait start them with
poll(). Preemption past a short deadline can no longer skip the step.
These schedulers marked a handler's work finished only after the handler
returned. A handler that threw left its work counted, so the context
never again saw it had nothing to do: it never stopped by itself, and a
later run() blocked with no work to wait for. On IOCP a signal handler
resumed inline left the signal service's own work counted the same way.

The work is now finished as the handler exits, however it exits.
IOCP learns of a stop only when a thread reaches the stop's wakeup in
the completion port. Queued work can still run after stop() is called.
The other backends check for the stop before running anything further.
run(), run_one(), poll(), poll_one() and the bounded forms are each
called with no work, with posted work, with work a handler posts, with a
parked read that has data and one that has none, with a timer past its
deadline and one far from it, on a stopped context, with stop() called
from a handler, and with a handler that throws. Each check compares one
line naming the function and what it saw, so a failure on any backend
shows the whole result.

A situation a backend can't make deterministic is skipped there: only
the reactors leave readiness for the test to wait on, and IOCP reports
timer expiry from a thread of its own. On IOCP, work queued ahead of a
stop still runs, so the stop checks expect that there.
@coderabbitai

coderabbitai Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

Explain Change →

No actionable comments were generated in the recent review.

Walkthrough

Timed run methods now use deadline-guarded waits. The reactor adopts pending work from enclosing contexts and changes its polling behavior. Backend schedulers update timer checks, wait transitions, and completion work accounting.

Changes: Scheduler runtime

Layer / File(s) Summary
Timed run semantics
include/boost/corosio/io_context.hpp, include/boost/corosio/native/native_io_context.hpp
Timed run documentation describes immediate returns for expired or nonpositive limits. Both run_one_until implementations use deadline-guarded loops with waits capped at one second.
Reactor work adoption and polling
include/boost/corosio/native/detail/reactor/reactor_scheduler.hpp
The reactor adopts pending work from enclosing contexts in its run and poll entry points. Its zero-timeout and bounded-wait paths update when they return or continue.
Backend polling and timer checks
include/boost/corosio/native/detail/epoll/epoll_scheduler.hpp, include/boost/corosio/native/detail/uring/uring_scheduler.hpp
The epoll scheduler checks due timers after a nonblocking pass. The io_uring scheduler clears positive timeouts after a ring pass or timed wait, then polls without blocking.
Completion work accounting
include/boost/corosio/native/detail/iocp/win_scheduler.hpp, include/boost/corosio/native/detail/iocp/win_signals.hpp, include/boost/corosio/native/detail/uring/uring_scheduler.hpp
IOCP, signal, and io_uring completion paths use scope guards to finish work when handler execution or coroutine resumption exits, including by exception.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Change: Bug fix · Severity of issue fixed: Medium

Suggested reviewers: mvandeberg, vinniefalco

Merge Risk: Minimal · up to 8bfee

This change lets poll() complete ready I/O and expired timers, bounds timed runs, and finishes work accounting when handlers throw. No concrete merge-blocking risk was identified.

Pre-merge checks: Passed 4

Check name Status Explanation
Description Check Passed Check skipped - CodeRabbit’s high-level summary is enabled.
Title check Passed The title names the primary change: poll() now completes ready I/O on reactor backends, plus related run-loop fixes. The raw summary shows poll/poll_one now run a non-blocking reactor pass, bounded ru…
Linked Issues check Passed [#418] The issue requires that a zero-timeout poll on reactor backends run one non-blocking reactor pass when only the reactor sentinel is queued, then process the completions it produces. The change …
Out of Scope Changes check Passed Beyond #418, the change set includes bounded-run fixes (run_for, run_one_for and _until forms now block at most once; io_uring positive-timeout clearing; run_one_until deadline loop in native_io_conte…
Full details: Title check

Explanation

The title names the primary change: poll() now completes ready I/O on reactor backends, plus related run-loop fixes. The raw summary shows poll/poll_one now run a non-blocking reactor pass, bounded runs block at most once, and the run-loop and IOCP/io_uring work accounting changes. The title is specific and matches the main objective of PR #423 and linked issue #418. The title does not mention the documented zero-duration run behavior change, but it does not need to cover every detail.

Full details: Linked Issues check

Explanation

[#418] The issue requires that a zero-timeout poll on reactor backends run one non-blocking reactor pass when only the reactor sentinel is queued, then process the completions it produces. The change set addresses this. In reactor_scheduler.hpp, do_one no longer returns early on a zero timeout while work remains. A zero-timeout call returns only when the pass leaves the queue idle. The poll and poll_one entry points adopt enclosing work before the pass. The PR states that this applies to epoll, select and kqueue, which share this scheduler. On epoll, epoll_scheduler.hpp also checks the nearest timer expiry after a nonblocking pass, so due timers run. The issue's acceptance example (a parked socket read completing within a bounded number of poll() calls) is covered by the reported reproducers and by tests in test/unit/bounded_run.cpp and test/unit/io_context.cpp. Those test files are excluded from this review (ignored_files), so the test results are not independently verified here. The implementation evidence is sufficient for the core requirement. IOCP and io_uring are outside the issue's stated scope.

Full details: Out of Scope Changes check

Explanation

Beyond #418, the change set includes bounded-run fixes (run_for, run_one_for and _until forms now block at most once; io_uring positive-timeout clearing; run_one_until deadline loop in native_io_context.hpp), nested-run work adoption (adopt_enclosing_work in reactor_scheduler.hpp), RAII work-finished guards on throwing handlers (win_scheduler.hpp, win_signals.hpp, uring_scheduler.hpp), and documentation of IOCP stop semantics in io_context.hpp. These changes are not required by #418. However, the PR objectives list them as intended run-loop fixes of this PR. Each is tied to the same poll and run-loop code path that #418 changes, and the zero-duration behaviour change is documented. The changes are therefore not unrelated. A reviewer may still choose to split them, but no demonstrated unrelated change exists.

Finishing Touches

Generate unit tests (beta)

  • Create a new PR
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

I’m a rabbit; timers tick,
I hop through polls, both slow and quick.
A ready task can join the queue,
And guards finish the work it’s due.
I nibble clover, then I’m through!

Recent review info

Run configuration

Commits

Reviewing files that changed from the base of the PR and between 4fde70c and 8bfeef5.

Files ignored due to path filters: 5

  • doc/modules/ROOT/pages/4.guide/4c.io-context.adoc is excluded by !**/doc/**
  • test/unit/bounded_run.cpp is excluded by !**/test/**
  • test/unit/io_context.cpp is excluded by !**/test/**
  • test/unit/native/native_io_context.cpp is excluded by !**/test/**
  • test/unit/reactor_teardown_inflight.cpp is excluded by !**/test/**
Files selected for processing: 7

  • include/boost/corosio/io_context.hpp
  • include/boost/corosio/native/detail/epoll/epoll_scheduler.hpp
  • include/boost/corosio/native/detail/iocp/win_scheduler.hpp
  • include/boost/corosio/native/detail/iocp/win_signals.hpp
  • include/boost/corosio/native/detail/reactor/reactor_scheduler.hpp
  • include/boost/corosio/native/detail/uring/uring_scheduler.hpp
  • include/boost/corosio/native/native_io_context.hpp

Included review availability: This review used your included allowance. Your plan provides up to 1 included review per hour; 0 remain after this review.

@cppalliance-bot

Copy link
Copy Markdown

An automated preview of the documentation is available at https://423.corosio.prtest3.cppalliance.org/index.html

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

2026-11-16 20:41:38 UTC

@greptile-apps

greptile-apps Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

RetriggerView in GreptileConfidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable issues were found.

Summary

Fixes poll() and poll_one() so reactor backends check ready I/O and due timers.

  • Poll calls now complete ready I/O and expired timers.
  • Bounded runs wait once, then finish work their pass found.
  • Nested reactor runs can see work their handler posted.
  • Throwing handlers still finish their counted work.
  • The stop documentation explains when Windows may still run queued work.

Intentional behavior documented by sgerbino: zero or expired bounds return without running work; Windows may run queued work after stop(). Timing-race tests remain deferred because they depend on timing.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Enter a run function] --> B[Publish enclosing handler's queued work]
    B --> C{Take reactor sentinel?}
    C -->|Yes| D[Check I/O and timers]
    D --> E[Return sentinel to queue]
    E --> F{Bounded call?}
    F -->|Yes| G[Use zero timeout from now on]
    F -->|No| H[Continue]
    G --> H
    H --> I{Ready work remains?}
    I -->|Yes| J[Run a handler]
    I -->|No, polling| K[Wake a waiting peer and return]
    C -->|Ready handler| J
    J --> L[Finish work count, even on exception]
Loading

Reviews (1) · Last reviewed commit: "test: check every run function in the sa..." · Reviewed by Greptile

@cppalliance-bot

Copy link
Copy Markdown

GCOVR code coverage report https://423.corosio.prtest3.cppalliance.org/gcovr/index.html
LCOV code coverage report https://423.corosio.prtest3.cppalliance.org/genhtml/index.html
Coverage Diff Report https://423.corosio.prtest3.cppalliance.org/diff-report/index.html

Build time: 2026-11-16 20:58:38 UTC

@github-actions

github-actions Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

Benchmark report

4fde70cb8ba5 (base) vs 8bfeef54ef89 (head)

  • ✅ linux — clean (epoll, uring)
  • ✅ windows — clean (iocp)
  • ✅ macos — clean (kqueue)
linux — full results (226 benchmarks, 5 iterations, 1.5s each)

accept_churn

Benchmark Backend Metric Base Head Δ Noise
burst/10 epoll ops_per_sec 2.02K ops/s 2.02K ops/s 🟢▲ +0.23% 1.27%
burst/10 uring ops_per_sec 2.39K ops/s 2.43K ops/s 🟢▲ +1.44% 1.09%
burst/100 epoll ops_per_sec 205.0 ops/s 205.4 ops/s 🟢▲ +0.67% 1.29%
burst/100 uring ops_per_sec 247.4 ops/s 249.5 ops/s 🟢▲ +1.07% 0.77%
burst_lockless/10 epoll ops_per_sec 2.04K ops/s 2.03K ops/s 🟢▲ +0.07% 1.24%
burst_lockless/10 uring ops_per_sec 2.40K ops/s 2.41K ops/s 🟢▲ +0.20% 0.16%
burst_lockless/100 epoll ops_per_sec 206.3 ops/s 206.6 ops/s 🟢▲ +0.06% 1.01%
burst_lockless/100 uring ops_per_sec 248.7 ops/s 249.4 ops/s 🟢▲ +0.12% 0.60%
concurrent/1 epoll ops_per_sec 15.93K ops/s 16.01K ops/s 🟢▲ +0.11% 1.25%
concurrent/1 uring ops_per_sec 17.15K ops/s 17.24K ops/s 🟢▲ +0.90% 1.18%
concurrent/16 epoll ops_per_sec 16.00K ops/s 16.09K ops/s 🟢▲ +0.86% 1.26%
concurrent/16 uring ops_per_sec 18.79K ops/s 18.96K ops/s 🟢▲ +0.60% 1.21%
concurrent/4 epoll ops_per_sec 15.99K ops/s 16.05K ops/s 🟢▲ +0.02% 1.46%
concurrent/4 uring ops_per_sec 17.82K ops/s 17.86K ops/s 🟢▲ +0.25% 1.14%
sequential epoll ops_per_sec 15.85K ops/s 16.04K ops/s 🟢▲ +0.84% 1.26%
sequential uring ops_per_sec 17.21K ops/s 17.33K ops/s 🟢▲ +0.68% 1.27%
sequential_lockless epoll ops_per_sec 16.10K ops/s 16.20K ops/s 🟢▲ +0.49% 1.25%
sequential_lockless uring ops_per_sec 17.36K ops/s 17.48K ops/s 🟢▲ +0.14% 1.08%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 epoll ops_per_sec 3.89K ops/s 3.86K ops/s 🔴▼ -0.10% 1.26%
concurrent_parents/1 uring ops_per_sec 4.00K ops/s 4.01K ops/s 🟢▲ +0.10% 0.92%
concurrent_parents/16 epoll ops_per_sec 3.81K ops/s 3.77K ops/s 🔴▼ -1.85% 1.24%
concurrent_parents/16 uring ops_per_sec 3.92K ops/s 3.94K ops/s 🟢▲ +0.20% 0.80%
concurrent_parents/4 epoll ops_per_sec 3.82K ops/s 3.78K ops/s 🔴▼ -1.18% 1.18%
concurrent_parents/4 uring ops_per_sec 3.96K ops/s 3.97K ops/s 🟢▲ +0.18% 0.49%
concurrent_parents_lockless/1 epoll ops_per_sec 3.94K ops/s 3.88K ops/s 🔴▼ -2.19% 1.27%
concurrent_parents_lockless/1 uring ops_per_sec 4.04K ops/s 4.07K ops/s 🟢▲ +0.19% 1.24%
concurrent_parents_lockless/16 epoll ops_per_sec 3.84K ops/s 3.80K ops/s 🔴▼ -1.23% 1.07%
concurrent_parents_lockless/16 uring ops_per_sec 3.97K ops/s 3.98K ops/s 🟢▲ +0.56% 0.95%
concurrent_parents_lockless/4 epoll ops_per_sec 3.86K ops/s 3.81K ops/s 🔴▼ -2.20% 1.10%
concurrent_parents_lockless/4 uring ops_per_sec 4.00K ops/s 4.03K ops/s 🟢▲ +0.36% 1.19%
fork_join/1 epoll ops_per_sec 56.60K ops/s 56.33K ops/s 🔴▼ -0.18% 0.53%
fork_join/1 uring ops_per_sec 54.07K ops/s 53.97K ops/s 🔴▼ -0.23% 0.33%
fork_join/16 epoll ops_per_sec 3.86K ops/s 3.84K ops/s 🔴▼ -0.04% 0.92%
fork_join/16 uring ops_per_sec 4.00K ops/s 4.01K ops/s 🟢▲ +0.24% 0.89%
fork_join/4 epoll ops_per_sec 15.24K ops/s 15.18K ops/s 🔴▼ -0.12% 1.01%
fork_join/4 uring ops_per_sec 14.94K ops/s 14.98K ops/s 🟢▲ +0.16% 0.50%
fork_join/64 epoll ops_per_sec 948.2 ops/s 946.9 ops/s 🟢▲ +0.06% 0.57%
fork_join/64 uring ops_per_sec 994.4 ops/s 995.0 ops/s 🟢▲ +0.07% 1.03%
fork_join_lockless/1 epoll ops_per_sec 57.24K ops/s 57.30K ops/s 🔴▼ -0.34% 0.46%
fork_join_lockless/1 uring ops_per_sec 55.93K ops/s 55.82K ops/s 🔴▼ -0.12% 0.90%
fork_join_lockless/16 epoll ops_per_sec 3.90K ops/s 3.90K ops/s 🟢▲ +0.32% 1.20%
fork_join_lockless/16 uring ops_per_sec 4.05K ops/s 4.05K ops/s 🟢▲ +0.13% 0.99%
fork_join_lockless/4 epoll ops_per_sec 15.38K ops/s 15.38K ops/s 🔴▼ -0.47% 1.01%
fork_join_lockless/4 uring ops_per_sec 15.28K ops/s 15.27K ops/s 🟢▲ +0.34% 1.16%
fork_join_lockless/64 epoll ops_per_sec 956.6 ops/s 961.9 ops/s 🟢▲ +0.62% 1.19%
fork_join_lockless/64 uring ops_per_sec 1.00K ops/s 1.01K ops/s 🟢▲ +0.08% 1.15%
nested/16 epoll ops_per_sec 943.6 ops/s 949.6 ops/s 🟢▲ +0.65% 1.30%
nested/16 uring ops_per_sec 988.4 ops/s 990.7 ops/s 🔴▼ -0.17% 1.02%
nested/4 epoll ops_per_sec 3.83K ops/s 3.85K ops/s 🟢▲ +0.18% 1.24%
nested/4 uring ops_per_sec 3.99K ops/s 3.99K ops/s 🟢▲ +0.03% 0.91%
nested_lockless/16 epoll ops_per_sec 956.9 ops/s 955.4 ops/s 🟢▲ +0.29% 1.21%
nested_lockless/16 uring ops_per_sec 998.5 ops/s 1.00K ops/s 🔴▼ -0.07% 0.96%
nested_lockless/4 epoll ops_per_sec 3.88K ops/s 3.90K ops/s 🟢▲ +0.39% 1.21%
nested_lockless/4 uring ops_per_sec 4.03K ops/s 4.04K ops/s 🔴▼ -0.13% 1.15%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll ops_per_sec 56.71K ops/s 56.85K ops/s 🟢▲ +0.14% 0.28%
concurrent/1 uring ops_per_sec 53.09K ops/s 53.16K ops/s 🔴▼ -0.05% 1.07%
concurrent/16 epoll ops_per_sec 59.78K ops/s 59.96K ops/s 🟢▲ +0.11% 0.27%
concurrent/16 uring ops_per_sec 63.42K ops/s 63.33K ops/s 🔴▼ -0.41% 0.38%
concurrent/32 epoll ops_per_sec 59.41K ops/s 59.56K ops/s 🟢▲ +0.17% 0.38%
concurrent/32 uring ops_per_sec 63.43K ops/s 63.38K ops/s 🔴▼ -0.02% 0.33%
concurrent/4 epoll ops_per_sec 59.03K ops/s 59.25K ops/s 🟢▲ +0.43% 0.27%
concurrent/4 uring ops_per_sec 60.91K ops/s 60.70K ops/s 🔴▼ -0.33% 0.32%
multithread/1 epoll ops_per_sec 59.23K ops/s 59.56K ops/s 🟢▲ +0.49% 0.34%
multithread/1 uring ops_per_sec 63.33K ops/s 63.66K ops/s 🔴▼ -0.17% 1.10%
multithread/16 epoll ops_per_sec 164.1K ops/s 164.8K ops/s 🟢▲ +0.22% 0.35%
multithread/16 uring ops_per_sec 100.9K ops/s 100.6K ops/s 🔴▼ -0.35% 0.36%
multithread/2 epoll ops_per_sec 87.61K ops/s 88.32K ops/s 🟢▲ +0.91% 0.47%
multithread/2 uring ops_per_sec 75.29K ops/s 75.27K ops/s 🟢▲ +0.54% 0.63%
multithread/4 epoll ops_per_sec 161.0K ops/s 160.2K ops/s 🔴▼ -0.55% 0.70%
multithread/4 uring ops_per_sec 120.4K ops/s 120.3K ops/s 🟢▲ +0.13% 0.30%
multithread/8 epoll ops_per_sec 163.0K ops/s 163.3K ops/s 🔴▼ -0.12% 0.64%
multithread/8 uring ops_per_sec 112.2K ops/s 111.9K ops/s 🔴▼ -0.44% 0.26%
single_conn epoll ops_per_sec 56.86K ops/s 56.95K ops/s 🟢▲ +0.42% 0.80%
single_conn uring ops_per_sec 53.41K ops/s 53.16K ops/s 🔴▼ -0.10% 1.20%
single_conn_lockless epoll ops_per_sec 57.36K ops/s 57.64K ops/s 🟢▲ +0.45% 0.32%
single_conn_lockless uring ops_per_sec 55.06K ops/s 55.12K ops/s 🟢▲ +0.03% 1.11%

local_socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll latency_mean_ns 7.25 µs 7.09 µs 🟢▲ +0.94% 4.30%
concurrent/1 uring latency_mean_ns 5.66 µs 5.78 µs 🔴▼ -0.17% 5.47%
concurrent/16 epoll latency_mean_ns 109.84 µs 109.34 µs 🟢▲ +0.36% 1.01%
concurrent/16 uring latency_mean_ns 93.33 µs 92.36 µs 🟢▲ +0.88% 1.13%
concurrent/4 epoll latency_mean_ns 27.60 µs 27.62 µs 🟢▲ +0.93% 1.79%
concurrent/4 uring latency_mean_ns 23.02 µs 23.24 µs 🔴▼ -0.98% 1.71%
concurrent_lockless/1 epoll latency_mean_ns 7.27 µs 6.97 µs 🟢▲ +2.23% 4.59%
concurrent_lockless/1 uring latency_mean_ns 5.78 µs 5.73 µs 🟢▲ +0.57% 5.83%
concurrent_lockless/16 epoll latency_mean_ns 108.04 µs 108.42 µs 🔴▼ -0.27% 0.74%
concurrent_lockless/16 uring latency_mean_ns 92.65 µs 92.41 µs 🔴▼ -0.16% 0.93%
concurrent_lockless/4 epoll latency_mean_ns 27.52 µs 27.28 µs 🟢▲ +0.23% 2.09%
concurrent_lockless/4 uring latency_mean_ns 23.32 µs 23.18 µs 🟢▲ +0.09% 1.16%
pingpong/1 epoll latency_mean_ns 7.17 µs 7.33 µs 🔴▼ -2.23% 5.35%
pingpong/1 uring latency_mean_ns 5.75 µs 5.72 µs 🟢▲ +0.16% 5.37%
pingpong/1024 epoll latency_mean_ns 7.80 µs 7.61 µs 🟢▲ +1.88% 4.59%
pingpong/1024 uring latency_mean_ns 6.15 µs 6.42 µs 🔴▼ -0.11% 6.25%
pingpong/64 epoll latency_mean_ns 7.35 µs 7.24 µs 🟢▲ +0.17% 4.88%
pingpong/64 uring latency_mean_ns 5.79 µs 5.78 µs 🟢▲ +0.19% 5.53%
pingpong_lockless/1 epoll latency_mean_ns 6.96 µs 6.95 µs 🟢▲ +0.03% 0.80%
pingpong_lockless/1 uring latency_mean_ns 5.85 µs 5.71 µs 🟢▲ +0.56% 6.59%
pingpong_lockless/1024 epoll latency_mean_ns 7.73 µs 7.65 µs 🟢▲ +0.48% 4.82%
pingpong_lockless/1024 uring latency_mean_ns 6.39 µs 6.40 µs 🔴▼ -0.57% 4.62%
pingpong_lockless/64 epoll latency_mean_ns 7.10 µs 7.12 µs 🟢▲ +0.15% 4.42%
pingpong_lockless/64 uring latency_mean_ns 5.62 µs 5.90 µs 🔴▼ -0.26% 6.70%

local_socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 epoll bytes_per_sec 309.2 MB/s 304.5 MB/s 🔴▼ -2.12% 1.02%
bidirectional/1024 uring bytes_per_sec 354.4 MB/s 354.7 MB/s 🟢▲ +0.02% 1.01%
bidirectional/1048576 epoll bytes_per_sec 3.22 GB/s 3.24 GB/s 🟢▲ +0.42% 1.86%
bidirectional/1048576 uring bytes_per_sec 3.57 GB/s 3.57 GB/s 🔴▼ -1.11% 2.00%
bidirectional/16384 epoll bytes_per_sec 3.11 GB/s 3.05 GB/s 🔴▼ -1.34% 1.57%
bidirectional/16384 uring bytes_per_sec 2.96 GB/s 2.99 GB/s 🟢▲ +1.32% 3.01%
bidirectional/262144 epoll bytes_per_sec 3.19 GB/s 3.24 GB/s 🟢▲ +3.21% 3.29%
bidirectional/262144 uring bytes_per_sec 3.55 GB/s 3.57 GB/s 🔴▼ -0.24% 2.93%
bidirectional/4096 epoll bytes_per_sec 1.01 GB/s 1.00 GB/s 🔴▼ -0.17% 1.82%
bidirectional/4096 uring bytes_per_sec 1.13 GB/s 1.13 GB/s 🟢▲ +0.34% 1.11%
bidirectional/65536 epoll bytes_per_sec 2.99 GB/s 2.95 GB/s 🔴▼ -0.26% 2.34%
bidirectional/65536 uring bytes_per_sec 3.42 GB/s 3.36 GB/s 🟢▲ +0.66% 4.95%
bidirectional_lockless/1024 epoll bytes_per_sec 307.5 MB/s 309.4 MB/s 🟢▲ +0.13% 1.33%
bidirectional_lockless/1024 uring bytes_per_sec 353.8 MB/s 359.3 MB/s 🟢▲ +1.87% 1.21%
bidirectional_lockless/1048576 epoll bytes_per_sec 3.21 GB/s 3.25 GB/s 🟢▲ +0.69% 2.15%
bidirectional_lockless/1048576 uring bytes_per_sec 3.58 GB/s 3.55 GB/s 🔴▼ -1.01% 2.74%
bidirectional_lockless/16384 epoll bytes_per_sec 3.11 GB/s 3.08 GB/s 🔴▼ -0.19% 3.71%
bidirectional_lockless/16384 uring bytes_per_sec 2.95 GB/s 3.02 GB/s 🟢▲ +3.03% 3.83%
bidirectional_lockless/262144 epoll bytes_per_sec 3.20 GB/s 3.24 GB/s 🟢▲ +1.97% 2.76%
bidirectional_lockless/262144 uring bytes_per_sec 3.58 GB/s 3.58 GB/s 🔴▼ -1.10% 2.83%
bidirectional_lockless/4096 epoll bytes_per_sec 1.01 GB/s 1.01 GB/s 🟢▲ +0.05% 1.83%
bidirectional_lockless/4096 uring bytes_per_sec 1.13 GB/s 1.14 GB/s 🟢▲ +0.98% 1.16%
bidirectional_lockless/65536 epoll bytes_per_sec 3.01 GB/s 2.97 GB/s 🔴▼ -1.14% 1.90%
bidirectional_lockless/65536 uring bytes_per_sec 3.45 GB/s 3.41 GB/s 🟢▲ +0.71% 4.47%
unidirectional/1024 epoll bytes_per_sec 299.9 MB/s 295.5 MB/s 🔴▼ -1.96% 0.86%
unidirectional/1024 uring bytes_per_sec 354.8 MB/s 352.7 MB/s 🔴▼ -0.44% 0.85%
unidirectional/1048576 epoll bytes_per_sec 3.31 GB/s 3.31 GB/s 🟢▲ +0.51% 2.91%
unidirectional/1048576 uring bytes_per_sec 3.57 GB/s 3.54 GB/s 🔴▼ -2.15% 3.62%
unidirectional/16384 epoll bytes_per_sec 3.00 GB/s 3.07 GB/s 🟢▲ +2.32% 4.30%
unidirectional/16384 uring bytes_per_sec 3.01 GB/s 3.03 GB/s 🟢▲ +0.97% 1.38%
unidirectional/262144 epoll bytes_per_sec 3.29 GB/s 3.34 GB/s 🟢▲ +3.13% 2.75%
unidirectional/262144 uring bytes_per_sec 3.50 GB/s 3.52 GB/s 🔴▼ -0.28% 3.93%
unidirectional/4096 epoll bytes_per_sec 982.7 MB/s 980.2 MB/s 🔴▼ -0.15% 0.29%
unidirectional/4096 uring bytes_per_sec 1.13 GB/s 1.12 GB/s 🟢▲ +0.20% 0.96%
unidirectional/65536 epoll bytes_per_sec 3.25 GB/s 3.23 GB/s 🟢▲ +0.79% 3.14%
unidirectional/65536 uring bytes_per_sec 3.46 GB/s 3.45 GB/s 🔴▼ -0.44% 3.22%
unidirectional_lockless/1024 epoll bytes_per_sec 304.2 MB/s 301.6 MB/s 🔴▼ -0.94% 0.85%
unidirectional_lockless/1024 uring bytes_per_sec 355.6 MB/s 357.3 MB/s ⚪ 0.00% 1.50%
unidirectional_lockless/1048576 epoll bytes_per_sec 3.34 GB/s 3.33 GB/s 🟢▲ +1.05% 2.38%
unidirectional_lockless/1048576 uring bytes_per_sec 3.60 GB/s 3.56 GB/s 🔴▼ -1.44% 2.86%
unidirectional_lockless/16384 epoll bytes_per_sec 3.08 GB/s 3.10 GB/s 🟢▲ +1.26% 2.51%
unidirectional_lockless/16384 uring bytes_per_sec 3.04 GB/s 3.04 GB/s 🔴▼ -0.28% 2.08%
unidirectional_lockless/262144 epoll bytes_per_sec 3.29 GB/s 3.34 GB/s 🟢▲ +3.70% 2.67%
unidirectional_lockless/262144 uring bytes_per_sec 3.57 GB/s 3.55 GB/s 🔴▼ -1.86% 3.75%
unidirectional_lockless/4096 epoll bytes_per_sec 994.0 MB/s 998.3 MB/s 🔴▼ -0.21% 1.70%
unidirectional_lockless/4096 uring bytes_per_sec 1.13 GB/s 1.14 GB/s 🟢▲ +1.59% 1.32%
unidirectional_lockless/65536 epoll bytes_per_sec 3.31 GB/s 3.25 GB/s 🔴▼ -1.55% 2.97%
unidirectional_lockless/65536 uring bytes_per_sec 3.46 GB/s 3.51 GB/s 🟢▲ +0.14% 2.62%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll latency_mean_ns 14.77 µs 14.81 µs 🟢▲ +0.04% 0.94%
concurrent/1 uring latency_mean_ns 13.70 µs 13.46 µs 🟢▲ +1.76% 1.21%
concurrent/16 epoll latency_mean_ns 232.48 µs 233.00 µs 🟢▲ +0.23% 1.16%
concurrent/16 uring latency_mean_ns 220.70 µs 217.93 µs 🟢▲ +0.84% 1.29%
concurrent/4 epoll latency_mean_ns 58.06 µs 58.23 µs 🟢▲ +0.29% 1.17%
concurrent/4 uring latency_mean_ns 54.92 µs 54.21 µs 🟢▲ +0.58% 1.41%
concurrent_lockless/1 epoll latency_mean_ns 14.65 µs 14.71 µs 🔴▼ -0.06% 1.04%
concurrent_lockless/1 uring latency_mean_ns 13.61 µs 13.43 µs 🟢▲ +0.81% 1.43%
concurrent_lockless/16 epoll latency_mean_ns 231.15 µs 231.71 µs 🟢▲ +0.21% 1.13%
concurrent_lockless/16 uring latency_mean_ns 220.01 µs 218.30 µs 🟢▲ +0.27% 1.20%
concurrent_lockless/4 epoll latency_mean_ns 57.76 µs 57.94 µs 🟢▲ +0.11% 1.06%
concurrent_lockless/4 uring latency_mean_ns 54.58 µs 53.96 µs 🟢▲ +0.51% 1.50%
pingpong/1 epoll latency_mean_ns 14.68 µs 14.60 µs 🟢▲ +0.16% 1.41%
pingpong/1 uring latency_mean_ns 13.51 µs 13.49 µs 🔴▼ -0.16% 1.27%
pingpong/1024 epoll latency_mean_ns 15.04 µs 14.97 µs 🔴▼ -0.30% 1.33%
pingpong/1024 uring latency_mean_ns 13.87 µs 13.86 µs 🔴▼ -0.02% 1.20%
pingpong/64 epoll latency_mean_ns 14.76 µs 14.68 µs 🔴▼ -0.27% 1.63%
pingpong/64 uring latency_mean_ns 13.60 µs 13.57 µs 🔴▼ -0.18% 1.18%
pingpong_lockless/1 epoll latency_mean_ns 14.57 µs 14.54 µs 🔴▼ -0.11% 1.12%
pingpong_lockless/1 uring latency_mean_ns 13.54 µs 13.48 µs 🟢▲ +0.42% 1.12%
pingpong_lockless/1024 epoll latency_mean_ns 14.98 µs 14.97 µs 🟢▲ +0.15% 0.30%
pingpong_lockless/1024 uring latency_mean_ns 13.90 µs 13.69 µs 🟢▲ +2.25% 1.12%
pingpong_lockless/64 epoll latency_mean_ns 14.67 µs 14.62 µs 🟢▲ +0.12% 1.18%
pingpong_lockless/64 uring latency_mean_ns 13.61 µs 13.53 µs 🟢▲ +0.60% 1.15%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 epoll bytes_per_sec 133.4 MB/s 134.6 MB/s 🟢▲ +0.15% 1.40%
bidirectional/1024 uring bytes_per_sec 150.0 MB/s 152.2 MB/s 🟢▲ +2.40% 1.40%
bidirectional/1048576 epoll bytes_per_sec 3.29 GB/s 3.28 GB/s 🟢▲ +0.19% 2.86%
bidirectional/1048576 uring bytes_per_sec 3.34 GB/s 3.33 GB/s 🔴▼ -0.45% 0.97%
bidirectional/16384 epoll bytes_per_sec 1.61 GB/s 1.64 GB/s 🟢▲ +2.16% 1.10%
bidirectional/16384 uring bytes_per_sec 1.61 GB/s 1.62 GB/s 🟢▲ +0.41% 0.99%
bidirectional/262144 epoll bytes_per_sec 3.22 GB/s 3.24 GB/s 🟢▲ +0.60% 1.14%
bidirectional/262144 uring bytes_per_sec 3.29 GB/s 3.27 GB/s 🔴▼ -0.86% 1.18%
bidirectional/4096 epoll bytes_per_sec 504.8 MB/s 508.6 MB/s 🟢▲ +0.69% 1.13%
bidirectional/4096 uring bytes_per_sec 521.1 MB/s 526.2 MB/s 🟢▲ +0.78% 1.36%
bidirectional/65536 epoll bytes_per_sec 2.64 GB/s 2.59 GB/s 🔴▼ -3.26% 2.23%
bidirectional/65536 uring bytes_per_sec 2.87 GB/s 2.85 GB/s 🔴▼ -1.01% 2.18%
bidirectional_lockless/1024 epoll bytes_per_sec 136.1 MB/s 135.0 MB/s 🔴▼ -0.32% 1.55%
bidirectional_lockless/1024 uring bytes_per_sec 151.5 MB/s 151.0 MB/s 🔴▼ -0.65% 1.04%
bidirectional_lockless/1048576 epoll bytes_per_sec 3.36 GB/s 3.31 GB/s 🔴▼ -3.43% 2.99%
bidirectional_lockless/1048576 uring bytes_per_sec 3.32 GB/s 3.33 GB/s 🟢▲ +0.66% 0.85%
bidirectional_lockless/16384 epoll bytes_per_sec 1.64 GB/s 1.62 GB/s 🔴▼ -0.29% 1.80%
bidirectional_lockless/16384 uring bytes_per_sec 1.61 GB/s 1.63 GB/s 🟢▲ +1.02% 0.84%
bidirectional_lockless/262144 epoll bytes_per_sec 3.26 GB/s 3.25 GB/s 🔴▼ -0.56% 1.68%
bidirectional_lockless/262144 uring bytes_per_sec 3.24 GB/s 3.25 GB/s 🟢▲ +0.57% 0.80%
bidirectional_lockless/4096 epoll bytes_per_sec 513.4 MB/s 508.7 MB/s 🔴▼ -0.09% 1.59%
bidirectional_lockless/4096 uring bytes_per_sec 524.9 MB/s 525.5 MB/s 🔴▼ -0.42% 1.06%
bidirectional_lockless/65536 epoll bytes_per_sec 2.62 GB/s 2.64 GB/s 🟢▲ +0.76% 2.30%
bidirectional_lockless/65536 uring bytes_per_sec 2.86 GB/s 2.88 GB/s 🟢▲ +0.15% 2.40%
multithread/2 epoll bytes_per_sec 3.65 GB/s 3.67 GB/s 🟢▲ +1.45% 1.53%
multithread/2 uring bytes_per_sec 4.22 GB/s 4.22 GB/s 🔴▼ -0.02% 0.48%
multithread/4 epoll bytes_per_sec 5.77 GB/s 5.73 GB/s 🔴▼ -1.20% 1.43%
multithread/4 uring bytes_per_sec 6.74 GB/s 6.73 GB/s 🔴▼ -0.16% 0.34%
multithread/8 epoll bytes_per_sec 6.35 GB/s 6.34 GB/s 🔴▼ -0.16% 0.95%
multithread/8 uring bytes_per_sec 6.85 GB/s 6.89 GB/s 🟢▲ +0.93% 0.56%
unidirectional/1024 epoll bytes_per_sec 125.3 MB/s 125.4 MB/s 🔴▼ -0.07% 1.01%
unidirectional/1024 uring bytes_per_sec 148.5 MB/s 149.0 MB/s 🟢▲ +0.18% 1.06%
unidirectional/1048576 epoll bytes_per_sec 3.38 GB/s 3.37 GB/s 🔴▼ -0.14% 0.32%
unidirectional/1048576 uring bytes_per_sec 3.42 GB/s 3.38 GB/s 🔴▼ -0.78% 1.30%
unidirectional/16384 epoll bytes_per_sec 1.53 GB/s 1.52 GB/s 🔴▼ -1.15% 1.65%
unidirectional/16384 uring bytes_per_sec 1.64 GB/s 1.69 GB/s 🟢▲ +3.06% 3.11%
unidirectional/262144 epoll bytes_per_sec 3.34 GB/s 3.35 GB/s 🔴▼ -0.11% 0.97%
unidirectional/262144 uring bytes_per_sec 3.24 GB/s 3.26 GB/s 🟢▲ +0.02% 0.96%
unidirectional/4096 epoll bytes_per_sec 471.9 MB/s 471.0 MB/s 🔴▼ -0.40% 1.24%
unidirectional/4096 uring bytes_per_sec 521.2 MB/s 520.7 MB/s 🟢▲ +0.61% 2.23%
unidirectional/65536 epoll bytes_per_sec 2.72 GB/s 2.79 GB/s 🟢▲ +0.67% 4.28%
unidirectional/65536 uring bytes_per_sec 2.85 GB/s 2.80 GB/s 🔴▼ -0.65% 2.11%
unidirectional_lockless/1024 epoll bytes_per_sec 127.1 MB/s 126.6 MB/s 🔴▼ -0.02% 1.40%
unidirectional_lockless/1024 uring bytes_per_sec 148.4 MB/s 148.8 MB/s 🟢▲ +0.46% 0.38%
unidirectional_lockless/1048576 epoll bytes_per_sec 3.30 GB/s 3.33 GB/s 🟢▲ +0.39% 1.29%
unidirectional_lockless/1048576 uring bytes_per_sec 3.38 GB/s 3.37 GB/s 🔴▼ -0.17% 0.71%
unidirectional_lockless/16384 epoll bytes_per_sec 1.56 GB/s 1.55 GB/s 🔴▼ -0.76% 1.65%
unidirectional_lockless/16384 uring bytes_per_sec 1.67 GB/s 1.71 GB/s 🟢▲ +1.46% 3.28%
unidirectional_lockless/262144 epoll bytes_per_sec 3.32 GB/s 3.35 GB/s 🟢▲ +1.50% 1.32%
unidirectional_lockless/262144 uring bytes_per_sec 3.26 GB/s 3.30 GB/s 🟢▲ +1.35% 1.12%
unidirectional_lockless/4096 epoll bytes_per_sec 475.3 MB/s 471.8 MB/s 🔴▼ -0.65% 1.31%
unidirectional_lockless/4096 uring bytes_per_sec 520.1 MB/s 520.1 MB/s 🔴▼ -0.15% 1.21%
unidirectional_lockless/65536 epoll bytes_per_sec 2.80 GB/s 2.82 GB/s 🟢▲ +0.55% 0.77%
unidirectional_lockless/65536 uring bytes_per_sec 2.84 GB/s 2.84 GB/s 🟢▲ +0.54% 2.65%
windows — full results (77 benchmarks, 5 iterations, 1.5s each)

accept_churn

Benchmark Backend Metric Base Head Δ Noise
burst/10 iocp ops_per_sec 1.56K ops/s 1.56K ops/s 🔴▼ -0.05% 0.66%
burst/100 iocp ops_per_sec 156.8 ops/s 156.9 ops/s 🟢▲ +0.50% 1.10%
burst_lockless/10 iocp ops_per_sec 1.57K ops/s 1.57K ops/s 🟢▲ +0.07% 0.18%
burst_lockless/100 iocp ops_per_sec 160.0 ops/s 159.7 ops/s 🟢▲ +0.09% 0.65%
concurrent/1 iocp ops_per_sec 12.56K ops/s 12.61K ops/s 🟢▲ +0.28% 0.42%
concurrent/16 iocp ops_per_sec 13.47K ops/s 13.49K ops/s 🟢▲ +0.04% 0.44%
concurrent/4 iocp ops_per_sec 13.03K ops/s 13.21K ops/s 🔴▼ -0.26% 3.21%
sequential iocp ops_per_sec 12.52K ops/s 12.55K ops/s 🟢▲ +0.25% 0.25%
sequential_lockless iocp ops_per_sec 12.57K ops/s 12.60K ops/s 🟢▲ +0.19% 0.25%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 iocp ops_per_sec 7.37K ops/s 7.34K ops/s 🔴▼ -0.56% 0.37%
concurrent_parents/16 iocp ops_per_sec 6.81K ops/s 6.79K ops/s 🔴▼ -0.35% 1.00%
concurrent_parents/4 iocp ops_per_sec 7.28K ops/s 7.27K ops/s 🔴▼ -0.21% 0.64%
concurrent_parents_lockless/1 iocp ops_per_sec 7.39K ops/s 7.35K ops/s 🔴▼ -0.32% 0.28%
concurrent_parents_lockless/16 iocp ops_per_sec 6.82K ops/s 6.81K ops/s 🔴▼ -0.45% 1.14%
concurrent_parents_lockless/4 iocp ops_per_sec 7.29K ops/s 7.28K ops/s 🔴▼ -0.32% 0.75%
fork_join/1 iocp ops_per_sec 112.2K ops/s 112.4K ops/s 🟢▲ +0.11% 0.50%
fork_join/16 iocp ops_per_sec 7.35K ops/s 7.34K ops/s 🔴▼ -0.06% 0.19%
fork_join/4 iocp ops_per_sec 29.21K ops/s 29.10K ops/s 🔴▼ -0.44% 0.34%
fork_join/64 iocp ops_per_sec 1.81K ops/s 1.82K ops/s 🔴▼ -0.24% 1.86%
fork_join_lockless/1 iocp ops_per_sec 112.5K ops/s 112.3K ops/s 🔴▼ -0.08% 0.37%
fork_join_lockless/16 iocp ops_per_sec 7.36K ops/s 7.34K ops/s 🔴▼ -0.14% 0.36%
fork_join_lockless/4 iocp ops_per_sec 29.29K ops/s 29.16K ops/s 🔴▼ -0.37% 0.32%
fork_join_lockless/64 iocp ops_per_sec 1.82K ops/s 1.82K ops/s 🔴▼ -0.13% 1.17%
nested/16 iocp ops_per_sec 1.79K ops/s 1.79K ops/s 🔴▼ -0.37% 0.76%
nested/4 iocp ops_per_sec 7.22K ops/s 7.21K ops/s 🔴▼ -0.09% 0.26%
nested_lockless/16 iocp ops_per_sec 1.80K ops/s 1.79K ops/s 🔴▼ -0.22% 0.73%
nested_lockless/4 iocp ops_per_sec 7.23K ops/s 7.17K ops/s 🔴▼ -0.74% 0.57%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 iocp ops_per_sec 111.3K ops/s 111.3K ops/s 🟢▲ +0.01% 0.60%
concurrent/16 iocp ops_per_sec 108.2K ops/s 108.8K ops/s 🟢▲ +0.69% 0.34%
concurrent/32 iocp ops_per_sec 107.5K ops/s 108.0K ops/s 🟢▲ +0.22% 0.52%
concurrent/4 iocp ops_per_sec 108.9K ops/s 109.1K ops/s 🟢▲ +0.62% 0.60%
multithread/1 iocp ops_per_sec 107.3K ops/s 108.1K ops/s 🟢▲ +0.50% 0.47%
multithread/16 iocp ops_per_sec 263.0K ops/s 264.5K ops/s 🟢▲ +0.39% 0.29%
multithread/2 iocp ops_per_sec 180.8K ops/s 180.9K ops/s 🟢▲ +0.01% 0.52%
multithread/4 iocp ops_per_sec 223.9K ops/s 224.5K ops/s 🟢▲ +0.67% 0.73%
multithread/8 iocp ops_per_sec 263.4K ops/s 264.8K ops/s 🟢▲ +0.54% 0.40%
single_conn iocp ops_per_sec 111.2K ops/s 110.5K ops/s 🔴▼ -0.30% 1.35%
single_conn_lockless iocp ops_per_sec 111.3K ops/s 111.1K ops/s 🔴▼ -0.17% 0.93%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 iocp latency_mean_ns 7.61 µs 7.56 µs 🟢▲ +0.76% 0.38%
concurrent/16 iocp latency_mean_ns 132.77 µs 127.35 µs 🟢▲ +3.94% 4.30%
concurrent/4 iocp latency_mean_ns 30.50 µs 30.47 µs 🟢▲ +0.44% 0.39%
concurrent_lockless/1 iocp latency_mean_ns 7.63 µs 7.58 µs 🟢▲ +0.79% 0.32%
concurrent_lockless/16 iocp latency_mean_ns 122.06 µs 124.84 µs 🔴▼ -1.28% 4.49%
concurrent_lockless/4 iocp latency_mean_ns 30.51 µs 30.39 µs 🟢▲ +0.28% 0.51%
pingpong/1 iocp latency_mean_ns 7.59 µs 7.59 µs 🟢▲ +0.15% 0.53%
pingpong/1024 iocp latency_mean_ns 7.71 µs 7.69 µs 🟢▲ +0.04% 0.47%
pingpong/64 iocp latency_mean_ns 7.64 µs 7.59 µs 🟢▲ +0.83% 0.34%
pingpong_lockless/1 iocp latency_mean_ns 7.59 µs 7.55 µs 🟢▲ +0.53% 0.42%
pingpong_lockless/1024 iocp latency_mean_ns 7.72 µs 7.70 µs 🟢▲ +0.17% 0.60%
pingpong_lockless/64 iocp latency_mean_ns 7.61 µs 7.57 µs 🟢▲ +0.36% 0.25%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 iocp bytes_per_sec 280.9 MB/s 280.7 MB/s 🔴▼ -0.03% 0.52%
bidirectional/1048576 iocp bytes_per_sec 4.59 GB/s 5.04 GB/s 🟢▲ +16.05% 7.97%
bidirectional/16384 iocp bytes_per_sec 3.73 GB/s 3.75 GB/s 🟢▲ +0.28% 1.02%
bidirectional/262144 iocp bytes_per_sec 9.55 GB/s 9.56 GB/s 🔴▼ -0.66% 2.80%
bidirectional/4096 iocp bytes_per_sec 1.08 GB/s 1.08 GB/s 🔴▼ -0.50% 0.46%
bidirectional/65536 iocp bytes_per_sec 8.89 GB/s 9.04 GB/s 🟢▲ +0.49% 3.04%
bidirectional_lockless/1024 iocp bytes_per_sec 281.5 MB/s 280.6 MB/s 🔴▼ -0.34% 0.64%
bidirectional_lockless/1048576 iocp bytes_per_sec 4.91 GB/s 5.15 GB/s 🔴▼ -1.03% 9.83%
bidirectional_lockless/16384 iocp bytes_per_sec 3.73 GB/s 3.72 GB/s 🔴▼ -0.57% 0.87%
bidirectional_lockless/262144 iocp bytes_per_sec 9.64 GB/s 9.67 GB/s 🟢▲ +0.53% 1.80%
bidirectional_lockless/4096 iocp bytes_per_sec 1.08 GB/s 1.08 GB/s 🔴▼ -0.18% 0.53%
bidirectional_lockless/65536 iocp bytes_per_sec 8.93 GB/s 8.91 GB/s 🔴▼ -0.26% 1.17%
multithread/2 iocp bytes_per_sec 12.64 GB/s 12.59 GB/s 🔴▼ -0.29% 0.49%
multithread/4 iocp bytes_per_sec 16.80 GB/s 16.77 GB/s 🔴▼ -0.62% 1.08%
multithread/8 iocp bytes_per_sec 18.42 GB/s 18.43 GB/s 🔴▼ -0.01% 0.34%
unidirectional/1024 iocp bytes_per_sec 284.7 MB/s 283.9 MB/s 🔴▼ -0.07% 0.46%
unidirectional/1048576 iocp bytes_per_sec 4.21 GB/s 4.08 GB/s 🔴▼ -2.78% 4.77%
unidirectional/16384 iocp bytes_per_sec 3.79 GB/s 3.78 GB/s 🔴▼ -1.10% 1.04%
unidirectional/262144 iocp bytes_per_sec 9.06 GB/s 8.30 GB/s 🔴▼ -5.23% 10.54%
unidirectional/4096 iocp bytes_per_sec 1.09 GB/s 1.09 GB/s 🔴▼ -0.44% 0.31%
unidirectional/65536 iocp bytes_per_sec 9.26 GB/s 8.78 GB/s 🟢▲ +0.01% 11.41%
unidirectional_lockless/1024 iocp bytes_per_sec 284.3 MB/s 282.9 MB/s 🔴▼ -0.82% 0.65%
unidirectional_lockless/1048576 iocp bytes_per_sec 4.52 GB/s 4.80 GB/s 🟢▲ +7.76% 5.44%
unidirectional_lockless/16384 iocp bytes_per_sec 3.79 GB/s 3.78 GB/s 🔴▼ -0.25% 0.68%
unidirectional_lockless/262144 iocp bytes_per_sec 9.59 GB/s 9.76 GB/s 🟢▲ +0.81% 3.33%
unidirectional_lockless/4096 iocp bytes_per_sec 1.09 GB/s 1.09 GB/s 🔴▼ -0.53% 0.32%
unidirectional_lockless/65536 iocp bytes_per_sec 9.30 GB/s 9.27 GB/s 🔴▼ -0.36% 1.66%
macos — full results (113 benchmarks, 5 iterations, 1.5s each)

accept_churn

Benchmark Backend Metric Base Head Δ Noise
burst/10 kqueue ops_per_sec 4.59K ops/s 4.58K ops/s 🟢▲ +0.05% 0.91%
burst/100 kqueue ops_per_sec 471.4 ops/s 467.3 ops/s 🔴▼ -0.50% 1.08%
burst_lockless/10 kqueue ops_per_sec 4.65K ops/s 4.63K ops/s 🔴▼ -0.35% 0.63%
burst_lockless/100 kqueue ops_per_sec 469.6 ops/s 474.0 ops/s 🟢▲ +0.97% 1.30%
concurrent/1 kqueue ops_per_sec 20.66K ops/s 20.68K ops/s 🔴▼ -0.21% 0.50%
concurrent/16 kqueue ops_per_sec 37.74K ops/s 37.92K ops/s 🟢▲ +0.56% 0.95%
concurrent/4 kqueue ops_per_sec 29.31K ops/s 29.40K ops/s 🔴▼ -0.16% 0.98%
sequential kqueue ops_per_sec 21.04K ops/s 21.03K ops/s 🔴▼ -0.38% 0.98%
sequential_lockless kqueue ops_per_sec 20.90K ops/s 20.88K ops/s 🟢▲ +0.08% 1.26%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 kqueue ops_per_sec 8.14K ops/s 8.18K ops/s 🟢▲ +1.00% 1.27%
concurrent_parents/16 kqueue ops_per_sec 7.71K ops/s 7.28K ops/s 🟢▲ +0.42% 19.21%
concurrent_parents/4 kqueue ops_per_sec 8.34K ops/s 8.39K ops/s 🔴▼ -0.02% 1.70%
concurrent_parents_lockless/1 kqueue ops_per_sec 8.23K ops/s 8.26K ops/s 🟢▲ +1.02% 1.48%
concurrent_parents_lockless/16 kqueue ops_per_sec 7.79K ops/s 7.19K ops/s 🔴▼ -0.53% 21.27%
concurrent_parents_lockless/4 kqueue ops_per_sec 8.42K ops/s 8.46K ops/s 🟢▲ +0.33% 1.65%
fork_join/1 kqueue ops_per_sec 49.02K ops/s 49.22K ops/s 🟢▲ +0.89% 0.87%
fork_join/16 kqueue ops_per_sec 8.14K ops/s 8.20K ops/s 🟢▲ +1.12% 1.26%
fork_join/4 kqueue ops_per_sec 23.61K ops/s 23.64K ops/s 🟢▲ +0.41% 1.02%
fork_join/64 kqueue ops_per_sec 2.10K ops/s 2.12K ops/s 🟢▲ +0.32% 2.51%
fork_join_lockless/1 kqueue ops_per_sec 50.39K ops/s 50.35K ops/s 🔴▼ -0.01% 0.24%
fork_join_lockless/16 kqueue ops_per_sec 8.24K ops/s 8.28K ops/s 🟢▲ +0.99% 1.43%
fork_join_lockless/4 kqueue ops_per_sec 23.65K ops/s 23.71K ops/s 🔴▼ -0.02% 0.67%
fork_join_lockless/64 kqueue ops_per_sec 2.12K ops/s 2.14K ops/s 🟢▲ +0.49% 2.56%
nested/16 kqueue ops_per_sec 2.09K ops/s 2.10K ops/s 🟢▲ +0.50% 2.63%
nested/4 kqueue ops_per_sec 8.06K ops/s 8.10K ops/s 🟢▲ +0.70% 0.96%
nested_lockless/16 kqueue ops_per_sec 2.10K ops/s 2.13K ops/s 🟢▲ +0.77% 2.46%
nested_lockless/4 kqueue ops_per_sec 8.14K ops/s 8.21K ops/s 🟢▲ +1.25% 1.36%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue ops_per_sec 49.71K ops/s 49.70K ops/s 🔴▼ -0.18% 0.41%
concurrent/16 kqueue ops_per_sec 130.7K ops/s 129.6K ops/s 🔴▼ -0.96% 0.33%
concurrent/32 kqueue ops_per_sec 128.7K ops/s 128.0K ops/s 🔴▼ -0.35% 0.58%
concurrent/4 kqueue ops_per_sec 90.17K ops/s 90.45K ops/s 🟢▲ +0.60% 1.00%
multithread/1 kqueue ops_per_sec 128.8K ops/s 128.0K ops/s 🔴▼ -0.33% 0.85%
multithread/16 kqueue ops_per_sec 133.5K ops/s 132.9K ops/s 🔴▼ -0.18% 1.05%
multithread/2 kqueue ops_per_sec 185.5K ops/s 186.0K ops/s 🟢▲ +0.11% 0.48%
multithread/4 kqueue ops_per_sec 161.8K ops/s 161.4K ops/s 🔴▼ -0.24% 0.64%
multithread/8 kqueue ops_per_sec 136.6K ops/s 136.6K ops/s 🔴▼ -0.01% 1.30%
single_conn kqueue ops_per_sec 49.56K ops/s 49.70K ops/s 🟢▲ +0.34% 0.20%
single_conn_lockless kqueue ops_per_sec 49.90K ops/s 49.99K ops/s 🟢▲ +0.13% 0.44%

local_socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue latency_mean_ns 1.83 µs 1.83 µs 🔴▼ -0.64% 0.56%
concurrent/16 kqueue latency_mean_ns 27.23 µs 27.21 µs 🟢▲ +0.33% 0.52%
concurrent/4 kqueue latency_mean_ns 6.84 µs 6.84 µs 🟢▲ +0.19% 0.57%
concurrent_lockless/1 kqueue latency_mean_ns 1.81 µs 1.82 µs 🔴▼ -0.55% 0.58%
concurrent_lockless/16 kqueue latency_mean_ns 27.03 µs 26.94 µs 🟢▲ +0.32% 0.51%
concurrent_lockless/4 kqueue latency_mean_ns 6.78 µs 6.77 µs 🔴▼ -0.06% 0.45%
pingpong/1 kqueue latency_mean_ns 1.82 µs 1.82 µs 🟢▲ +0.25% 0.68%
pingpong/1024 kqueue latency_mean_ns 1.92 µs 1.92 µs 🟢▲ +0.52% 0.50%
pingpong/64 kqueue latency_mean_ns 1.83 µs 1.82 µs 🟢▲ +0.90% 0.46%
pingpong_lockless/1 kqueue latency_mean_ns 1.81 µs 1.81 µs 🔴▼ -0.48% 0.71%
pingpong_lockless/1024 kqueue latency_mean_ns 1.91 µs 1.92 µs 🔴▼ -0.37% 0.52%
pingpong_lockless/64 kqueue latency_mean_ns 1.81 µs 1.81 µs 🔴▼ -0.23% 0.48%

local_socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 kqueue bytes_per_sec 1.30 GB/s 1.30 GB/s 🔴▼ -0.45% 0.35%
bidirectional/1048576 kqueue bytes_per_sec 3.88 GB/s 3.88 GB/s 🔴▼ -0.25% 1.04%
bidirectional/16384 kqueue bytes_per_sec 3.88 GB/s 3.88 GB/s 🔴▼ -0.04% 0.91%
bidirectional/262144 kqueue bytes_per_sec 3.89 GB/s 3.88 GB/s 🔴▼ -0.30% 1.03%
bidirectional/4096 kqueue bytes_per_sec 2.94 GB/s 2.92 GB/s 🔴▼ -0.77% 0.89%
bidirectional/65536 kqueue bytes_per_sec 3.87 GB/s 3.89 GB/s 🟢▲ +0.19% 1.55%
bidirectional_lockless/1024 kqueue bytes_per_sec 1.30 GB/s 1.30 GB/s 🔴▼ -0.11% 0.66%
bidirectional_lockless/1048576 kqueue bytes_per_sec 3.98 GB/s 3.97 GB/s 🔴▼ -0.50% 1.11%
bidirectional_lockless/16384 kqueue bytes_per_sec 3.97 GB/s 3.98 GB/s 🔴▼ -0.17% 1.41%
bidirectional_lockless/262144 kqueue bytes_per_sec 3.97 GB/s 3.98 GB/s 🔴▼ -0.37% 1.20%
bidirectional_lockless/4096 kqueue bytes_per_sec 2.97 GB/s 2.98 GB/s 🟢▲ +0.39% 0.27%
bidirectional_lockless/65536 kqueue bytes_per_sec 3.98 GB/s 3.99 GB/s 🟢▲ +0.33% 0.73%
unidirectional/1024 kqueue bytes_per_sec 1.28 GB/s 1.27 GB/s 🔴▼ -0.33% 0.33%
unidirectional/1048576 kqueue bytes_per_sec 3.55 GB/s 3.52 GB/s 🔴▼ -1.11% 0.47%
unidirectional/16384 kqueue bytes_per_sec 3.53 GB/s 3.53 GB/s 🔴▼ -0.08% 0.54%
unidirectional/262144 kqueue bytes_per_sec 3.54 GB/s 3.52 GB/s 🔴▼ -0.81% 0.53%
unidirectional/4096 kqueue bytes_per_sec 2.73 GB/s 2.71 GB/s 🔴▼ -0.42% 0.52%
unidirectional/65536 kqueue bytes_per_sec 3.54 GB/s 3.51 GB/s 🔴▼ -1.08% 0.79%
unidirectional_lockless/1024 kqueue bytes_per_sec 1.29 GB/s 1.28 GB/s 🔴▼ -0.25% 0.63%
unidirectional_lockless/1048576 kqueue bytes_per_sec 3.61 GB/s 3.61 GB/s 🔴▼ -0.13% 0.83%
unidirectional_lockless/16384 kqueue bytes_per_sec 3.61 GB/s 3.61 GB/s 🔴▼ -0.07% 0.67%
unidirectional_lockless/262144 kqueue bytes_per_sec 3.61 GB/s 3.60 GB/s 🔴▼ -0.56% 1.05%
unidirectional_lockless/4096 kqueue bytes_per_sec 2.77 GB/s 2.76 GB/s 🔴▼ -0.30% 0.43%
unidirectional_lockless/65536 kqueue bytes_per_sec 3.61 GB/s 3.61 GB/s 🔴▼ -0.19% 0.84%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue latency_mean_ns 18.71 µs 18.72 µs 🔴▼ -0.19% 0.35%
concurrent/16 kqueue latency_mean_ns 120.49 µs 119.90 µs 🟢▲ +0.16% 1.01%
concurrent/4 kqueue latency_mean_ns 43.02 µs 43.05 µs 🔴▼ -0.16% 1.16%
concurrent_lockless/1 kqueue latency_mean_ns 18.64 µs 18.65 µs 🔴▼ -0.14% 0.13%
concurrent_lockless/16 kqueue latency_mean_ns 120.10 µs 119.69 µs 🟢▲ +0.11% 1.16%
concurrent_lockless/4 kqueue latency_mean_ns 42.47 µs 42.38 µs 🟢▲ +0.36% 0.65%
pingpong/1 kqueue latency_mean_ns 15.50 µs 15.52 µs ⚪ 0.00% 1.27%
pingpong/1024 kqueue latency_mean_ns 19.03 µs 19.02 µs 🔴▼ -0.03% 0.37%
pingpong/64 kqueue latency_mean_ns 18.69 µs 18.69 µs 🔴▼ -0.16% 0.26%
pingpong_lockless/1 kqueue latency_mean_ns 15.35 µs 15.34 µs 🔴▼ -0.08% 0.24%
pingpong_lockless/1024 kqueue latency_mean_ns 19.01 µs 19.02 µs 🔴▼ -0.09% 0.26%
pingpong_lockless/64 kqueue latency_mean_ns 18.64 µs 18.64 µs 🔴▼ -0.11% 0.19%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 kqueue bytes_per_sec 474.0 MB/s 472.6 MB/s 🔴▼ -0.01% 1.52%
bidirectional/1048576 kqueue bytes_per_sec 7.79 GB/s 7.57 GB/s 🟢▲ +0.58% 4.68%
bidirectional/16384 kqueue bytes_per_sec 4.58 GB/s 4.55 GB/s 🔴▼ -0.49% 1.05%
bidirectional/262144 kqueue bytes_per_sec 8.74 GB/s 8.79 GB/s 🔴▼ -0.26% 2.59%
bidirectional/4096 kqueue bytes_per_sec 1.82 GB/s 1.81 GB/s 🔴▼ -1.12% 0.86%
bidirectional/65536 kqueue bytes_per_sec 7.55 GB/s 7.55 GB/s 🔴▼ -0.30% 1.76%
bidirectional_lockless/1024 kqueue bytes_per_sec 477.9 MB/s 472.9 MB/s 🔴▼ -0.76% 0.85%
bidirectional_lockless/1048576 kqueue bytes_per_sec 7.85 GB/s 7.73 GB/s 🟢▲ +0.25% 4.57%
bidirectional_lockless/16384 kqueue bytes_per_sec 4.60 GB/s 4.58 GB/s 🔴▼ -0.09% 1.65%
bidirectional_lockless/262144 kqueue bytes_per_sec 8.82 GB/s 8.75 GB/s 🔴▼ -0.48% 2.05%
bidirectional_lockless/4096 kqueue bytes_per_sec 1.83 GB/s 1.83 GB/s 🟢▲ +0.28% 0.51%
bidirectional_lockless/65536 kqueue bytes_per_sec 7.60 GB/s 7.56 GB/s 🔴▼ -0.97% 1.43%
multithread/2 kqueue bytes_per_sec 7.69 GB/s 7.43 GB/s 🔴▼ -0.91% 3.49%
multithread/4 kqueue bytes_per_sec 3.59 GB/s 3.71 GB/s 🟢▲ +2.06% 6.38%
multithread/8 kqueue bytes_per_sec 3.14 GB/s 3.20 GB/s 🟢▲ +2.01% 9.35%
unidirectional/1024 kqueue bytes_per_sec 460.5 MB/s 458.7 MB/s 🟢▲ +0.56% 2.13%
unidirectional/1048576 kqueue bytes_per_sec 8.98 GB/s 8.79 GB/s 🔴▼ -1.09% 5.74%
unidirectional/16384 kqueue bytes_per_sec 4.50 GB/s 4.44 GB/s 🔴▼ -0.16% 2.75%
unidirectional/262144 kqueue bytes_per_sec 8.89 GB/s 8.91 GB/s 🟢▲ +0.49% 2.85%
unidirectional/4096 kqueue bytes_per_sec 1.81 GB/s 1.80 GB/s 🔴▼ -0.28% 2.24%
unidirectional/65536 kqueue bytes_per_sec 7.40 GB/s 7.34 GB/s 🔴▼ -1.20% 2.90%
unidirectional_lockless/1024 kqueue bytes_per_sec 481.3 MB/s 474.9 MB/s 🔴▼ -0.74% 2.29%
unidirectional_lockless/1048576 kqueue bytes_per_sec 9.06 GB/s 8.82 GB/s 🔴▼ -0.65% 4.88%
unidirectional_lockless/16384 kqueue bytes_per_sec 4.53 GB/s 4.54 GB/s 🟢▲ +0.82% 1.17%
unidirectional_lockless/262144 kqueue bytes_per_sec 9.02 GB/s 9.14 GB/s 🟢▲ +0.47% 4.57%
unidirectional_lockless/4096 kqueue bytes_per_sec 1.81 GB/s 1.81 GB/s 🟢▲ +0.80% 2.30%
unidirectional_lockless/65536 kqueue bytes_per_sec 7.35 GB/s 7.44 GB/s 🟢▲ +2.39% 1.79%

Run & raw JSON artifacts · flag rule: |median Δ| > max(2%, 3×CV of the noisier side) · advisory only

@sgerbino sgerbino added the benchmark Triggers benchmark suite label Nov 16, 2026
@github-actions github-actions Bot removed the benchmark Triggers benchmark suite label Nov 16, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Backlog

Development

Successfully merging this pull request may close these issues.

poll() and poll_one() never complete ready I/O on the reactor backends

2 participants