Skip to content

perf: run handlers queued behind the reactor first while that pays off - #416

Draft
sgerbino wants to merge 1 commit into
developfrom
pr/recycling-free-list
Draft

sgerbino wants to merge 1 commit into
developfrom
pr/recycling-free-list

Conversation

@sgerbino

@sgerbino sgerbino commented Nov 12, 2026 •

Copy link
Copy Markdown
Collaborator

When the reactor sentinel reached the front of the queue with handlers still behind it, the scheduler ran a non-blocking reactor poll before running them. Those handlers usually start the next I/O themselves, such as a coroutine an I/O completion woke, so running them first can save a reactor pass.

The sentinel now moves behind the queued handlers once, then polls on its next turn regardless, so handlers that keep reposting still cannot starve I/O. poll() and poll_one() are unchanged.

Waiting doesn't always pay. On macOS, loopback data arrives a little later than on Linux. A fixed deferral there let the handlers drain the queue, their speculative reads and writes hit EAGAIN about 8× as often and parked, and kqueue bidirectional/1048576 lost 13–14% in every run. reactor_defer_policy scores each deferral by what its turn found:

  • handlers still queued: a hit, score −1
  • queue empty: a miss, score +8, capped at 64
  • the sentinel defers only while the score is below 8; each turn that polls instead lowers it by 1, so a workload that changes gets tried again

restart() resets the score along with any pending deferral.

Tests:

  • reactor_defer_policy.cpp checks the scoring deterministically, with no timing.
  • testPostedWakeRunsBeforeNextPoll in the reactor fault suite counts the backend's wait calls to check that a posted wake runs before the next poll.

The recycling frame allocator change this used to pair with merged as cppalliance/capy#485, so both sides of the benchmark A/B build against capy develop.

macOS ARM64 CI, kqueue, 7 interleaved reps, vs deferral off:

Row fixed deferral adaptive
bidirectional/1048576 −14.4% +1.3%
bidirectional_lockless/1048576 −13.5% +1.2%
bidirectional/262144 −0.3% −0.7%
unidirectional/1048576 −0.1% +0.4%
pingpong/64 ≈0 ≈0

Local epoll, one idle machine, 5 interleaved runs:

Row off fixed adaptive
pingpong/64 306.8K 328.4K 327.8K
concurrent_parents_lockless/16 15,430 15,698 15,474
fork_join_lockless/64 4586 4584 4586
bidirectional/1048576 8.54 8.54 8.44 ±1.9%

The adaptive rule keeps the latency gain and recovers the macOS bulk rows. It gives up fixed deferral's small concurrent_parents_lockless/16 gain: that workload's retries land at the end of a lap, where the queue empties.

@cppalliance-bot

cppalliance-bot commented Nov 12, 2026 •

Copy link
Copy Markdown

GCOVR code coverage report https://416.corosio.prtest3.cppalliance.org/gcovr/index.html
LCOV code coverage report https://416.corosio.prtest3.cppalliance.org/genhtml/index.html
Coverage Diff Report https://416.corosio.prtest3.cppalliance.org/diff-report/index.html

Build time: 2026-11-17 23:22:51 UTC

@sgerbino
sgerbino force-pushed the pr/recycling-free-list branch from 7c6bd6d to 1e59c16 Compare November 12, 2026 15:07
@cppalliance-bot

cppalliance-bot commented Nov 12, 2026 •

Copy link
Copy Markdown

An automated preview of the documentation is available at https://416.corosio.prtest3.cppalliance.org/index.html

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

2026-11-17 23:10:39 UTC

@github-actions

github-actions Bot commented Nov 12, 2026 •

Copy link
Copy Markdown

Benchmark report

7cb81d1a76c6 (base) vs cf9d8598ef8e (head)

  • ⚠️ linux — 7 flagged (epoll, uring)
  • ✅ windows — clean (iocp)
  • ⚠️ macos — 13 flagged (kqueue)

⚠️ Flagged

Platform Backend Benchmark Metric Base Head Δ Noise
linux epoll local_socket_latency/concurrent/1 latency_mean_ns 6.83 µs 6.32 µs 🟢▲ +6.90% 2.23%
linux epoll local_socket_latency/concurrent/16 latency_mean_ns 104.28 µs 101.45 µs 🟢▲ +2.71% 0.46%
linux epoll local_socket_latency/concurrent_lockless/16 latency_mean_ns 103.32 µs 100.64 µs 🟢▲ +2.57% 0.31%
linux epoll local_socket_latency/pingpong/1 latency_mean_ns 6.76 µs 6.28 µs 🟢▲ +6.50% 1.61%
linux epoll local_socket_latency/pingpong_lockless/64 latency_mean_ns 6.74 µs 6.30 µs 🟢▲ +6.35% 0.72%
linux epoll socket_latency/concurrent/1 latency_mean_ns 14.45 µs 14.10 µs 🟢▲ +2.59% 0.52%
linux epoll socket_latency/concurrent_lockless/1 latency_mean_ns 14.43 µs 14.06 µs 🟢▲ +2.52% 0.17%
macos kqueue accept_churn/burst_lockless/10 ops_per_sec 4.62K ops/s 4.78K ops/s 🟢▲ +3.26% 0.63%
macos kqueue local_socket_latency/concurrent/1 latency_mean_ns 1.82 µs 1.71 µs 🟢▲ +6.68% 0.58%
macos kqueue local_socket_latency/concurrent/4 latency_mean_ns 6.82 µs 6.66 µs 🟢▲ +2.38% 0.44%
macos kqueue local_socket_latency/concurrent_lockless/1 latency_mean_ns 1.80 µs 1.69 µs 🟢▲ +6.70% 0.36%
macos kqueue local_socket_latency/concurrent_lockless/4 latency_mean_ns 6.74 µs 6.57 µs 🟢▲ +2.68% 0.40%
macos kqueue local_socket_latency/pingpong/1 latency_mean_ns 1.82 µs 1.70 µs 🟢▲ +6.22% 0.41%
macos kqueue local_socket_latency/pingpong/1024 latency_mean_ns 1.92 µs 1.80 µs 🟢▲ +6.24% 0.51%
macos kqueue local_socket_latency/pingpong/64 latency_mean_ns 1.82 µs 1.70 µs 🟢▲ +6.36% 0.33%
macos kqueue local_socket_latency/pingpong_lockless/1 latency_mean_ns 1.80 µs 1.68 µs 🟢▲ +6.69% 0.40%
macos kqueue local_socket_latency/pingpong_lockless/1024 latency_mean_ns 1.91 µs 1.78 µs 🟢▲ +6.68% 0.47%
macos kqueue local_socket_latency/pingpong_lockless/64 latency_mean_ns 1.81 µs 1.69 µs 🟢▲ +6.34% 0.51%
macos kqueue local_socket_throughput/unidirectional/1024 bytes_per_sec 1.28 GB/s 1.32 GB/s 🟢▲ +3.06% 0.37%
macos kqueue local_socket_throughput/unidirectional_lockless/1024 bytes_per_sec 1.28 GB/s 1.32 GB/s 🟢▲ +2.91% 0.34%
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 2.03K ops/s 2.03K ops/s 🟢▲ +0.19% 1.03%
burst/10 uring ops_per_sec 2.42K ops/s 2.43K ops/s 🔴▼ -0.03% 1.38%
burst/100 epoll ops_per_sec 206.2 ops/s 205.0 ops/s 🔴▼ -0.32% 1.05%
burst/100 uring ops_per_sec 248.5 ops/s 249.0 ops/s 🟢▲ +0.52% 0.59%
burst_lockless/10 epoll ops_per_sec 2.04K ops/s 2.04K ops/s 🟢▲ +0.31% 0.92%
burst_lockless/10 uring ops_per_sec 2.43K ops/s 2.43K ops/s ⚪ 0.00% 1.20%
burst_lockless/100 epoll ops_per_sec 207.5 ops/s 206.9 ops/s 🟢▲ +0.22% 1.06%
burst_lockless/100 uring ops_per_sec 249.2 ops/s 249.8 ops/s 🟢▲ +0.26% 0.63%
concurrent/1 epoll ops_per_sec 16.15K ops/s 16.37K ops/s 🟢▲ +1.45% 1.17%
concurrent/1 uring ops_per_sec 17.45K ops/s 17.58K ops/s ⚪ 0.00% 1.33%
concurrent/16 epoll ops_per_sec 16.21K ops/s 16.29K ops/s 🟢▲ +0.99% 1.39%
concurrent/16 uring ops_per_sec 18.93K ops/s 19.09K ops/s 🔴▼ -0.03% 1.27%
concurrent/4 epoll ops_per_sec 16.18K ops/s 16.35K ops/s 🟢▲ +0.95% 1.26%
concurrent/4 uring ops_per_sec 18.02K ops/s 18.14K ops/s 🔴▼ -0.33% 1.27%
sequential epoll ops_per_sec 16.17K ops/s 16.44K ops/s 🟢▲ +1.46% 0.88%
sequential uring ops_per_sec 17.44K ops/s 17.51K ops/s 🟢▲ +0.02% 0.77%
sequential_lockless epoll ops_per_sec 16.18K ops/s 16.45K ops/s 🟢▲ +1.20% 1.36%
sequential_lockless uring ops_per_sec 17.53K ops/s 17.60K ops/s 🟢▲ +0.19% 0.84%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 epoll ops_per_sec 3.94K ops/s 3.95K ops/s 🟢▲ +0.25% 0.32%
concurrent_parents/1 uring ops_per_sec 4.08K ops/s 4.06K ops/s 🔴▼ -0.32% 0.98%
concurrent_parents/16 epoll ops_per_sec 3.86K ops/s 3.85K ops/s 🔴▼ -0.16% 0.32%
concurrent_parents/16 uring ops_per_sec 4.00K ops/s 4.01K ops/s 🟢▲ +0.24% 1.23%
concurrent_parents/4 epoll ops_per_sec 3.86K ops/s 3.86K ops/s 🔴▼ -0.11% 0.27%
concurrent_parents/4 uring ops_per_sec 4.04K ops/s 4.03K ops/s 🔴▼ -0.11% 1.02%
concurrent_parents_lockless/1 epoll ops_per_sec 3.96K ops/s 3.97K ops/s 🟢▲ +0.24% 0.33%
concurrent_parents_lockless/1 uring ops_per_sec 4.14K ops/s 4.11K ops/s 🔴▼ -0.48% 1.20%
concurrent_parents_lockless/16 epoll ops_per_sec 3.87K ops/s 3.88K ops/s 🟢▲ +0.28% 0.37%
concurrent_parents_lockless/16 uring ops_per_sec 4.02K ops/s 4.03K ops/s 🟢▲ +0.17% 0.96%
concurrent_parents_lockless/4 epoll ops_per_sec 3.88K ops/s 3.89K ops/s 🟢▲ +0.18% 0.35%
concurrent_parents_lockless/4 uring ops_per_sec 4.07K ops/s 4.07K ops/s 🔴▼ -0.08% 1.10%
fork_join/1 epoll ops_per_sec 57.99K ops/s 57.94K ops/s 🟢▲ +0.32% 0.99%
fork_join/1 uring ops_per_sec 55.90K ops/s 55.98K ops/s 🔴▼ -0.02% 1.22%
fork_join/16 epoll ops_per_sec 3.94K ops/s 3.95K ops/s 🟢▲ +0.26% 0.33%
fork_join/16 uring ops_per_sec 4.10K ops/s 4.10K ops/s 🟢▲ +0.07% 1.26%
fork_join/4 epoll ops_per_sec 15.54K ops/s 15.57K ops/s 🟢▲ +0.12% 0.20%
fork_join/4 uring ops_per_sec 15.44K ops/s 15.38K ops/s 🔴▼ -0.05% 1.35%
fork_join/64 epoll ops_per_sec 964.6 ops/s 967.0 ops/s 🟢▲ +0.17% 0.18%
fork_join/64 uring ops_per_sec 1.01K ops/s 1.02K ops/s 🟢▲ +0.57% 1.26%
fork_join_lockless/1 epoll ops_per_sec 58.35K ops/s 58.57K ops/s 🟢▲ +0.29% 0.20%
fork_join_lockless/1 uring ops_per_sec 56.75K ops/s 57.12K ops/s 🟢▲ +0.19% 1.44%
fork_join_lockless/16 epoll ops_per_sec 3.95K ops/s 3.97K ops/s 🟢▲ +0.45% 0.18%
fork_join_lockless/16 uring ops_per_sec 4.11K ops/s 4.13K ops/s 🟢▲ +0.59% 1.10%
fork_join_lockless/4 epoll ops_per_sec 15.64K ops/s 15.70K ops/s 🟢▲ +0.45% 0.30%
fork_join_lockless/4 uring ops_per_sec 15.53K ops/s 15.58K ops/s 🔴▼ -0.33% 1.38%
fork_join_lockless/64 epoll ops_per_sec 969.7 ops/s 972.9 ops/s 🟢▲ +0.35% 0.33%
fork_join_lockless/64 uring ops_per_sec 1.02K ops/s 1.02K ops/s 🔴▼ -0.12% 1.02%
nested/16 epoll ops_per_sec 961.3 ops/s 961.4 ops/s ⚪ 0.00% 0.25%
nested/16 uring ops_per_sec 1.00K ops/s 1.00K ops/s 🔴▼ -0.02% 0.35%
nested/4 epoll ops_per_sec 3.92K ops/s 3.92K ops/s 🟢▲ +0.13% 0.32%
nested/4 uring ops_per_sec 4.05K ops/s 4.05K ops/s 🔴▼ -0.07% 0.24%
nested_lockless/16 epoll ops_per_sec 966.0 ops/s 966.3 ops/s 🔴▼ -0.10% 0.29%
nested_lockless/16 uring ops_per_sec 1.01K ops/s 1.01K ops/s 🔴▼ -0.03% 1.02%
nested_lockless/4 epoll ops_per_sec 3.93K ops/s 3.94K ops/s 🟢▲ +0.44% 0.41%
nested_lockless/4 uring ops_per_sec 4.11K ops/s 4.08K ops/s 🔴▼ -0.41% 1.03%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll ops_per_sec 59.38K ops/s 59.49K ops/s 🟢▲ +0.18% 1.12%
concurrent/1 uring ops_per_sec 55.23K ops/s 55.67K ops/s 🟢▲ +0.07% 1.24%
concurrent/16 epoll ops_per_sec 62.38K ops/s 62.61K ops/s 🟢▲ +0.47% 1.02%
concurrent/16 uring ops_per_sec 65.54K ops/s 65.92K ops/s 🟢▲ +1.42% 1.25%
concurrent/32 epoll ops_per_sec 61.89K ops/s 62.03K ops/s 🟢▲ +0.26% 1.05%
concurrent/32 uring ops_per_sec 65.30K ops/s 65.97K ops/s 🟢▲ +1.88% 1.12%
concurrent/4 epoll ops_per_sec 62.10K ops/s 62.25K ops/s 🟢▲ +0.10% 1.02%
concurrent/4 uring ops_per_sec 63.44K ops/s 63.55K ops/s 🔴▼ -0.09% 1.14%
multithread/1 epoll ops_per_sec 61.93K ops/s 61.90K ops/s 🟢▲ +0.11% 0.96%
multithread/1 uring ops_per_sec 65.36K ops/s 66.01K ops/s 🟢▲ +0.84% 0.96%
multithread/16 epoll ops_per_sec 172.6K ops/s 173.7K ops/s 🟢▲ +0.35% 0.45%
multithread/16 uring ops_per_sec 104.1K ops/s 104.5K ops/s 🟢▲ +0.45% 0.34%
multithread/2 epoll ops_per_sec 93.98K ops/s 94.68K ops/s 🟢▲ +0.55% 0.28%
multithread/2 uring ops_per_sec 79.22K ops/s 79.41K ops/s 🟢▲ +0.27% 0.59%
multithread/4 epoll ops_per_sec 170.0K ops/s 171.3K ops/s 🟢▲ +1.25% 0.86%
multithread/4 uring ops_per_sec 125.2K ops/s 125.4K ops/s 🟢▲ +0.23% 0.79%
multithread/8 epoll ops_per_sec 172.4K ops/s 173.2K ops/s 🟢▲ +0.43% 0.56%
multithread/8 uring ops_per_sec 116.7K ops/s 116.8K ops/s 🟢▲ +0.04% 0.63%
single_conn epoll ops_per_sec 59.04K ops/s 59.52K ops/s 🟢▲ +0.52% 1.06%
single_conn uring ops_per_sec 55.32K ops/s 55.74K ops/s 🔴▼ -0.02% 1.24%
single_conn_lockless epoll ops_per_sec 59.72K ops/s 59.97K ops/s 🟢▲ +0.41% 1.02%
single_conn_lockless uring ops_per_sec 56.31K ops/s 56.88K ops/s 🟢▲ +0.61% 1.21%

local_socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll latency_mean_ns 6.83 µs 6.32 µs 🟢▲ +6.90% 2.23%
concurrent/1 uring latency_mean_ns 5.48 µs 5.68 µs 🔴▼ -0.12% 6.15%
concurrent/16 epoll latency_mean_ns 104.28 µs 101.45 µs 🟢▲ +2.71% 0.46%
concurrent/16 uring latency_mean_ns 86.69 µs 87.15 µs 🔴▼ -0.12% 0.75%
concurrent/4 epoll latency_mean_ns 26.52 µs 25.39 µs 🟢▲ +3.80% 1.40%
concurrent/4 uring latency_mean_ns 21.57 µs 21.92 µs 🔴▼ -1.72% 1.73%
concurrent_lockless/1 epoll latency_mean_ns 6.85 µs 6.30 µs 🟢▲ +6.12% 4.49%
concurrent_lockless/1 uring latency_mean_ns 5.36 µs 5.28 µs 🔴▼ -0.51% 5.07%
concurrent_lockless/16 epoll latency_mean_ns 103.32 µs 100.64 µs 🟢▲ +2.57% 0.31%
concurrent_lockless/16 uring latency_mean_ns 86.42 µs 87.06 µs 🔴▼ -0.77% 0.64%
concurrent_lockless/4 epoll latency_mean_ns 26.17 µs 25.16 µs 🟢▲ +3.57% 1.49%
concurrent_lockless/4 uring latency_mean_ns 21.65 µs 21.72 µs 🔴▼ -0.01% 1.28%
pingpong/1 epoll latency_mean_ns 6.76 µs 6.28 µs 🟢▲ +6.50% 1.61%
pingpong/1 uring latency_mean_ns 5.21 µs 5.58 µs 🔴▼ -8.47% 5.52%
pingpong/1024 epoll latency_mean_ns 7.27 µs 7.11 µs 🔴▼ -3.32% 5.28%
pingpong/1024 uring latency_mean_ns 5.64 µs 5.89 µs 🔴▼ -1.78% 5.53%
pingpong/64 epoll latency_mean_ns 6.91 µs 6.56 µs 🟢▲ +6.12% 4.70%
pingpong/64 uring latency_mean_ns 5.37 µs 5.39 µs 🟢▲ +1.45% 5.85%
pingpong_lockless/1 epoll latency_mean_ns 6.94 µs 6.26 µs 🟢▲ +6.75% 5.31%
pingpong_lockless/1 uring latency_mean_ns 5.23 µs 5.21 µs 🟢▲ +0.03% 1.04%
pingpong_lockless/1024 epoll latency_mean_ns 7.35 µs 6.94 µs 🟢▲ +5.63% 5.34%
pingpong_lockless/1024 uring latency_mean_ns 5.90 µs 5.64 µs 🟢▲ +1.23% 6.35%
pingpong_lockless/64 epoll latency_mean_ns 6.74 µs 6.30 µs 🟢▲ +6.35% 0.72%
pingpong_lockless/64 uring latency_mean_ns 5.40 µs 5.63 µs 🔴▼ -0.34% 5.74%

local_socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 epoll bytes_per_sec 314.0 MB/s 317.7 MB/s 🟢▲ +1.64% 1.06%
bidirectional/1024 uring bytes_per_sec 360.3 MB/s 363.4 MB/s 🟢▲ +0.22% 1.25%
bidirectional/1048576 epoll bytes_per_sec 3.26 GB/s 3.22 GB/s 🔴▼ -1.79% 2.55%
bidirectional/1048576 uring bytes_per_sec 3.58 GB/s 3.60 GB/s 🟢▲ +1.75% 2.03%
bidirectional/16384 epoll bytes_per_sec 3.15 GB/s 3.14 GB/s 🟢▲ +1.02% 3.92%
bidirectional/16384 uring bytes_per_sec 3.00 GB/s 3.01 GB/s 🟢▲ +1.24% 1.88%
bidirectional/262144 epoll bytes_per_sec 3.24 GB/s 3.25 GB/s 🔴▼ -1.66% 3.32%
bidirectional/262144 uring bytes_per_sec 3.57 GB/s 3.61 GB/s 🟢▲ +3.62% 2.56%
bidirectional/4096 epoll bytes_per_sec 1.03 GB/s 1.03 GB/s 🟢▲ +0.59% 0.98%
bidirectional/4096 uring bytes_per_sec 1.15 GB/s 1.15 GB/s 🔴▼ -0.11% 0.97%
bidirectional/65536 epoll bytes_per_sec 3.02 GB/s 3.06 GB/s 🟢▲ +1.27% 1.01%
bidirectional/65536 uring bytes_per_sec 3.49 GB/s 3.47 GB/s 🟢▲ +0.74% 3.16%
bidirectional_lockless/1024 epoll bytes_per_sec 314.8 MB/s 320.3 MB/s 🟢▲ +1.69% 1.19%
bidirectional_lockless/1024 uring bytes_per_sec 363.4 MB/s 365.6 MB/s 🟢▲ +0.68% 0.68%
bidirectional_lockless/1048576 epoll bytes_per_sec 3.27 GB/s 3.20 GB/s 🔴▼ -1.83% 2.69%
bidirectional_lockless/1048576 uring bytes_per_sec 3.58 GB/s 3.61 GB/s 🟢▲ +1.99% 1.46%
bidirectional_lockless/16384 epoll bytes_per_sec 3.08 GB/s 3.18 GB/s 🟢▲ +1.70% 3.62%
bidirectional_lockless/16384 uring bytes_per_sec 3.06 GB/s 3.05 GB/s 🔴▼ -1.26% 2.37%
bidirectional_lockless/262144 epoll bytes_per_sec 3.25 GB/s 3.24 GB/s 🔴▼ -2.07% 2.75%
bidirectional_lockless/262144 uring bytes_per_sec 3.57 GB/s 3.60 GB/s 🟢▲ +0.66% 1.79%
bidirectional_lockless/4096 epoll bytes_per_sec 1.03 GB/s 1.03 GB/s 🟢▲ +1.02% 1.28%
bidirectional_lockless/4096 uring bytes_per_sec 1.15 GB/s 1.15 GB/s 🔴▼ -0.37% 0.53%
bidirectional_lockless/65536 epoll bytes_per_sec 3.01 GB/s 3.05 GB/s 🟢▲ +1.14% 0.64%
bidirectional_lockless/65536 uring bytes_per_sec 3.53 GB/s 3.50 GB/s 🔴▼ -0.69% 1.98%
unidirectional/1024 epoll bytes_per_sec 304.0 MB/s 314.2 MB/s 🟢▲ +3.04% 1.13%
unidirectional/1024 uring bytes_per_sec 362.1 MB/s 364.4 MB/s 🟢▲ +1.25% 0.86%
unidirectional/1048576 epoll bytes_per_sec 3.31 GB/s 3.28 GB/s 🟢▲ +1.10% 3.72%
unidirectional/1048576 uring bytes_per_sec 3.59 GB/s 3.62 GB/s 🟢▲ +1.60% 2.43%
unidirectional/16384 epoll bytes_per_sec 3.07 GB/s 3.17 GB/s 🟢▲ +2.40% 2.04%
unidirectional/16384 uring bytes_per_sec 3.06 GB/s 3.11 GB/s 🟢▲ +1.95% 2.77%
unidirectional/262144 epoll bytes_per_sec 3.30 GB/s 3.31 GB/s 🔴▼ -1.19% 2.93%
unidirectional/262144 uring bytes_per_sec 3.56 GB/s 3.60 GB/s 🟢▲ +2.63% 1.43%
unidirectional/4096 epoll bytes_per_sec 1.00 GB/s 1.02 GB/s 🟢▲ +2.33% 0.84%
unidirectional/4096 uring bytes_per_sec 1.15 GB/s 1.16 GB/s 🟢▲ +0.82% 1.28%
unidirectional/65536 epoll bytes_per_sec 3.28 GB/s 3.35 GB/s 🟢▲ +1.65% 1.30%
unidirectional/65536 uring bytes_per_sec 3.55 GB/s 3.49 GB/s 🔴▼ -2.10% 1.59%
unidirectional_lockless/1024 epoll bytes_per_sec 307.2 MB/s 317.1 MB/s 🟢▲ +3.04% 1.18%
unidirectional_lockless/1024 uring bytes_per_sec 364.6 MB/s 363.9 MB/s 🟢▲ +0.37% 0.93%
unidirectional_lockless/1048576 epoll bytes_per_sec 3.35 GB/s 3.26 GB/s 🔴▼ -1.25% 5.41%
unidirectional_lockless/1048576 uring bytes_per_sec 3.58 GB/s 3.63 GB/s 🟢▲ +1.84% 2.11%
unidirectional_lockless/16384 epoll bytes_per_sec 3.11 GB/s 3.20 GB/s 🟢▲ +3.31% 2.01%
unidirectional_lockless/16384 uring bytes_per_sec 3.12 GB/s 3.03 GB/s 🔴▼ -3.96% 3.49%
unidirectional_lockless/262144 epoll bytes_per_sec 3.33 GB/s 3.32 GB/s 🔴▼ -0.73% 2.84%
unidirectional_lockless/262144 uring bytes_per_sec 3.58 GB/s 3.61 GB/s 🟢▲ +2.58% 1.98%
unidirectional_lockless/4096 epoll bytes_per_sec 1.01 GB/s 1.03 GB/s 🟢▲ +1.93% 1.17%
unidirectional_lockless/4096 uring bytes_per_sec 1.16 GB/s 1.15 GB/s 🔴▼ -0.48% 0.45%
unidirectional_lockless/65536 epoll bytes_per_sec 3.30 GB/s 3.36 GB/s 🟢▲ +0.87% 1.29%
unidirectional_lockless/65536 uring bytes_per_sec 3.57 GB/s 3.55 GB/s 🔴▼ -0.57% 2.62%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 epoll latency_mean_ns 14.45 µs 14.10 µs 🟢▲ +2.59% 0.52%
concurrent/1 uring latency_mean_ns 13.28 µs 13.24 µs 🟢▲ +0.23% 1.25%
concurrent/16 epoll latency_mean_ns 228.86 µs 225.46 µs 🟢▲ +1.59% 0.26%
concurrent/16 uring latency_mean_ns 215.14 µs 213.97 µs 🟢▲ +0.12% 1.14%
concurrent/4 epoll latency_mean_ns 57.11 µs 56.13 µs 🟢▲ +1.82% 0.29%
concurrent/4 uring latency_mean_ns 53.27 µs 53.13 µs 🟢▲ +0.17% 1.17%
concurrent_lockless/1 epoll latency_mean_ns 14.43 µs 14.06 µs 🟢▲ +2.52% 0.17%
concurrent_lockless/1 uring latency_mean_ns 13.30 µs 13.20 µs 🟢▲ +0.63% 0.67%
concurrent_lockless/16 epoll latency_mean_ns 228.14 µs 224.66 µs 🟢▲ +1.42% 0.21%
concurrent_lockless/16 uring latency_mean_ns 215.15 µs 212.47 µs 🟢▲ +0.85% 1.31%
concurrent_lockless/4 epoll latency_mean_ns 56.88 µs 56.00 µs 🟢▲ +1.54% 0.18%
concurrent_lockless/4 uring latency_mean_ns 53.39 µs 53.06 µs 🟢▲ +0.11% 1.04%
pingpong/1 epoll latency_mean_ns 14.39 µs 13.96 µs 🟢▲ +2.84% 1.08%
pingpong/1 uring latency_mean_ns 13.24 µs 13.13 µs 🟢▲ +0.30% 1.43%
pingpong/1024 epoll latency_mean_ns 14.69 µs 14.29 µs 🟢▲ +2.58% 1.10%
pingpong/1024 uring latency_mean_ns 13.53 µs 13.43 µs 🟢▲ +0.09% 1.31%
pingpong/64 epoll latency_mean_ns 14.43 µs 14.02 µs 🟢▲ +2.49% 1.25%
pingpong/64 uring latency_mean_ns 13.33 µs 13.18 µs 🟢▲ +0.42% 1.35%
pingpong_lockless/1 epoll latency_mean_ns 14.28 µs 13.91 µs 🟢▲ +2.69% 1.20%
pingpong_lockless/1 uring latency_mean_ns 13.20 µs 13.02 µs 🟢▲ +0.55% 1.41%
pingpong_lockless/1024 epoll latency_mean_ns 14.60 µs 14.29 µs 🟢▲ +2.31% 0.99%
pingpong_lockless/1024 uring latency_mean_ns 13.49 µs 13.46 µs 🟢▲ +0.04% 1.20%
pingpong_lockless/64 epoll latency_mean_ns 14.35 µs 13.98 µs 🟢▲ +2.60% 1.08%
pingpong_lockless/64 uring latency_mean_ns 13.25 µs 13.14 µs 🟢▲ +0.07% 1.30%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 epoll bytes_per_sec 135.2 MB/s 137.1 MB/s 🟢▲ +0.55% 1.34%
bidirectional/1024 uring bytes_per_sec 152.4 MB/s 151.8 MB/s 🔴▼ -0.17% 1.02%
bidirectional/1048576 epoll bytes_per_sec 3.33 GB/s 3.35 GB/s 🟢▲ +0.34% 2.54%
bidirectional/1048576 uring bytes_per_sec 3.35 GB/s 3.38 GB/s 🟢▲ +0.97% 0.42%
bidirectional/16384 epoll bytes_per_sec 1.65 GB/s 1.66 GB/s 🟢▲ +1.49% 1.61%
bidirectional/16384 uring bytes_per_sec 1.63 GB/s 1.64 GB/s 🟢▲ +0.51% 1.19%
bidirectional/262144 epoll bytes_per_sec 3.25 GB/s 3.24 GB/s 🔴▼ -0.53% 1.13%
bidirectional/262144 uring bytes_per_sec 3.25 GB/s 3.29 GB/s 🟢▲ +1.34% 0.70%
bidirectional/4096 epoll bytes_per_sec 509.3 MB/s 516.8 MB/s 🟢▲ +0.94% 1.24%
bidirectional/4096 uring bytes_per_sec 526.3 MB/s 528.8 MB/s 🟢▲ +0.24% 0.85%
bidirectional/65536 epoll bytes_per_sec 2.66 GB/s 2.63 GB/s 🔴▼ -1.64% 3.96%
bidirectional/65536 uring bytes_per_sec 2.88 GB/s 2.90 GB/s 🟢▲ +0.94% 1.55%
bidirectional_lockless/1024 epoll bytes_per_sec 135.4 MB/s 135.9 MB/s 🟢▲ +0.35% 0.19%
bidirectional_lockless/1024 uring bytes_per_sec 152.6 MB/s 152.6 MB/s 🟢▲ +0.02% 1.13%
bidirectional_lockless/1048576 epoll bytes_per_sec 3.23 GB/s 3.31 GB/s 🟢▲ +3.50% 2.84%
bidirectional_lockless/1048576 uring bytes_per_sec 3.30 GB/s 3.30 GB/s 🟢▲ +0.02% 0.95%
bidirectional_lockless/16384 epoll bytes_per_sec 1.64 GB/s 1.65 GB/s 🟢▲ +0.45% 1.31%
bidirectional_lockless/16384 uring bytes_per_sec 1.58 GB/s 1.59 GB/s 🟢▲ +0.29% 0.85%
bidirectional_lockless/262144 epoll bytes_per_sec 3.22 GB/s 3.24 GB/s 🟢▲ +1.28% 0.80%
bidirectional_lockless/262144 uring bytes_per_sec 3.22 GB/s 3.24 GB/s 🟢▲ +0.56% 0.88%
bidirectional_lockless/4096 epoll bytes_per_sec 510.7 MB/s 513.6 MB/s 🟢▲ +0.69% 0.35%
bidirectional_lockless/4096 uring bytes_per_sec 527.9 MB/s 529.4 MB/s 🟢▲ +0.08% 1.81%
bidirectional_lockless/65536 epoll bytes_per_sec 2.66 GB/s 2.65 GB/s 🔴▼ -1.21% 3.78%
bidirectional_lockless/65536 uring bytes_per_sec 2.78 GB/s 2.78 GB/s 🟢▲ +0.53% 0.98%
multithread/2 epoll bytes_per_sec 3.66 GB/s 3.70 GB/s 🟢▲ +1.90% 2.10%
multithread/2 uring bytes_per_sec 4.24 GB/s 4.24 GB/s 🔴▼ -0.42% 0.52%
multithread/4 epoll bytes_per_sec 5.78 GB/s 6.01 GB/s 🟢▲ +3.92% 1.89%
multithread/4 uring bytes_per_sec 6.75 GB/s 6.77 GB/s 🔴▼ -0.07% 0.75%
multithread/8 epoll bytes_per_sec 6.33 GB/s 6.42 GB/s 🟢▲ +1.57% 0.92%
multithread/8 uring bytes_per_sec 6.89 GB/s 6.87 GB/s 🔴▼ -0.40% 0.70%
unidirectional/1024 epoll bytes_per_sec 127.1 MB/s 128.9 MB/s 🟢▲ +0.78% 1.47%
unidirectional/1024 uring bytes_per_sec 150.4 MB/s 150.8 MB/s 🔴▼ -0.11% 1.21%
unidirectional/1048576 epoll bytes_per_sec 3.37 GB/s 3.37 GB/s 🔴▼ -0.23% 0.97%
unidirectional/1048576 uring bytes_per_sec 3.40 GB/s 3.39 GB/s 🔴▼ -0.07% 1.13%
unidirectional/16384 epoll bytes_per_sec 1.56 GB/s 1.59 GB/s 🟢▲ +1.06% 1.29%
unidirectional/16384 uring bytes_per_sec 1.69 GB/s 1.66 GB/s 🔴▼ -0.54% 3.08%
unidirectional/262144 epoll bytes_per_sec 3.34 GB/s 3.32 GB/s 🔴▼ -0.19% 1.24%
unidirectional/262144 uring bytes_per_sec 3.27 GB/s 3.24 GB/s 🔴▼ -1.05% 0.75%
unidirectional/4096 epoll bytes_per_sec 477.8 MB/s 482.0 MB/s 🟢▲ +0.72% 1.35%
unidirectional/4096 uring bytes_per_sec 525.8 MB/s 523.7 MB/s 🟢▲ +0.14% 1.60%
unidirectional/65536 epoll bytes_per_sec 2.76 GB/s 2.80 GB/s 🟢▲ +0.48% 5.13%
unidirectional/65536 uring bytes_per_sec 2.86 GB/s 2.84 GB/s 🟢▲ +0.79% 3.01%
unidirectional_lockless/1024 epoll bytes_per_sec 126.8 MB/s 127.9 MB/s 🟢▲ +0.84% 0.18%
unidirectional_lockless/1024 uring bytes_per_sec 151.5 MB/s 150.4 MB/s 🔴▼ -0.09% 1.01%
unidirectional_lockless/1048576 epoll bytes_per_sec 3.31 GB/s 3.38 GB/s 🟢▲ +2.20% 0.87%
unidirectional_lockless/1048576 uring bytes_per_sec 3.37 GB/s 3.36 GB/s 🔴▼ -0.35% 1.82%
unidirectional_lockless/16384 epoll bytes_per_sec 1.56 GB/s 1.59 GB/s 🟢▲ +1.53% 1.85%
unidirectional_lockless/16384 uring bytes_per_sec 1.73 GB/s 1.71 GB/s 🔴▼ -0.70% 0.86%
unidirectional_lockless/262144 epoll bytes_per_sec 3.33 GB/s 3.36 GB/s 🟢▲ +0.77% 1.09%
unidirectional_lockless/262144 uring bytes_per_sec 3.29 GB/s 3.28 GB/s 🔴▼ -0.36% 1.26%
unidirectional_lockless/4096 epoll bytes_per_sec 476.8 MB/s 479.6 MB/s 🟢▲ +0.47% 0.36%
unidirectional_lockless/4096 uring bytes_per_sec 532.3 MB/s 520.8 MB/s 🔴▼ -2.23% 0.87%
unidirectional_lockless/65536 epoll bytes_per_sec 2.77 GB/s 2.83 GB/s 🟢▲ +1.02% 2.02%
unidirectional_lockless/65536 uring bytes_per_sec 2.83 GB/s 2.81 GB/s 🔴▼ -0.99% 0.84%
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 🟢▲ +0.36% 0.31%
burst/100 iocp ops_per_sec 156.1 ops/s 155.3 ops/s 🔴▼ -0.57% 0.97%
burst_lockless/10 iocp ops_per_sec 1.56K ops/s 1.57K ops/s 🟢▲ +0.32% 0.27%
burst_lockless/100 iocp ops_per_sec 158.8 ops/s 158.9 ops/s 🟢▲ +0.01% 0.90%
concurrent/1 iocp ops_per_sec 12.45K ops/s 12.57K ops/s 🟢▲ +1.05% 0.37%
concurrent/16 iocp ops_per_sec 13.40K ops/s 13.47K ops/s 🟢▲ +0.62% 0.38%
concurrent/4 iocp ops_per_sec 13.07K ops/s 13.17K ops/s 🟢▲ +0.77% 0.28%
sequential iocp ops_per_sec 12.46K ops/s 12.52K ops/s 🟢▲ +0.62% 0.42%
sequential_lockless iocp ops_per_sec 12.50K ops/s 12.59K ops/s 🟢▲ +0.73% 0.28%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 iocp ops_per_sec 7.37K ops/s 7.36K ops/s 🔴▼ -0.16% 0.23%
concurrent_parents/16 iocp ops_per_sec 6.83K ops/s 6.85K ops/s 🟢▲ +0.17% 1.13%
concurrent_parents/4 iocp ops_per_sec 7.28K ops/s 7.28K ops/s 🔴▼ -0.14% 0.78%
concurrent_parents_lockless/1 iocp ops_per_sec 7.38K ops/s 7.39K ops/s 🟢▲ +0.28% 0.25%
concurrent_parents_lockless/16 iocp ops_per_sec 6.84K ops/s 6.85K ops/s 🔴▼ -0.08% 0.94%
concurrent_parents_lockless/4 iocp ops_per_sec 7.28K ops/s 7.30K ops/s 🟢▲ +0.03% 0.51%
fork_join/1 iocp ops_per_sec 112.9K ops/s 113.0K ops/s 🟢▲ +0.24% 0.43%
fork_join/16 iocp ops_per_sec 7.34K ops/s 7.36K ops/s 🟢▲ +0.15% 0.35%
fork_join/4 iocp ops_per_sec 29.18K ops/s 29.21K ops/s 🟢▲ +0.06% 0.40%
fork_join/64 iocp ops_per_sec 1.81K ops/s 1.83K ops/s 🟢▲ +0.25% 2.10%
fork_join_lockless/1 iocp ops_per_sec 113.2K ops/s 112.6K ops/s 🔴▼ -0.35% 0.42%
fork_join_lockless/16 iocp ops_per_sec 7.35K ops/s 7.36K ops/s 🟢▲ +0.09% 0.33%
fork_join_lockless/4 iocp ops_per_sec 29.15K ops/s 29.19K ops/s 🔴▼ -0.02% 0.43%
fork_join_lockless/64 iocp ops_per_sec 1.81K ops/s 1.83K ops/s 🟢▲ +0.16% 2.12%
nested/16 iocp ops_per_sec 1.79K ops/s 1.80K ops/s 🟢▲ +0.30% 1.59%
nested/4 iocp ops_per_sec 7.22K ops/s 7.24K ops/s 🟢▲ +0.41% 0.21%
nested_lockless/16 iocp ops_per_sec 1.80K ops/s 1.80K ops/s 🟢▲ +0.04% 0.89%
nested_lockless/4 iocp ops_per_sec 7.22K ops/s 7.25K ops/s 🟢▲ +0.29% 0.21%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 iocp ops_per_sec 111.6K ops/s 111.5K ops/s 🟢▲ +0.09% 0.37%
concurrent/16 iocp ops_per_sec 108.8K ops/s 108.8K ops/s 🟢▲ +0.01% 0.33%
concurrent/32 iocp ops_per_sec 108.3K ops/s 108.2K ops/s 🔴▼ -0.04% 0.33%
concurrent/4 iocp ops_per_sec 109.2K ops/s 109.2K ops/s 🟢▲ +0.10% 0.42%
multithread/1 iocp ops_per_sec 108.3K ops/s 108.1K ops/s 🔴▼ -0.27% 0.40%
multithread/16 iocp ops_per_sec 264.6K ops/s 264.3K ops/s 🔴▼ -0.24% 0.45%
multithread/2 iocp ops_per_sec 179.9K ops/s 181.3K ops/s 🟢▲ +0.39% 0.62%
multithread/4 iocp ops_per_sec 224.4K ops/s 223.7K ops/s 🔴▼ -0.31% 0.81%
multithread/8 iocp ops_per_sec 264.3K ops/s 264.4K ops/s 🟢▲ +0.01% 0.34%
single_conn iocp ops_per_sec 111.6K ops/s 111.3K ops/s 🔴▼ -0.33% 0.58%
single_conn_lockless iocp ops_per_sec 111.8K ops/s 111.3K ops/s 🔴▼ -0.52% 0.37%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 iocp latency_mean_ns 7.57 µs 7.61 µs 🔴▼ -0.42% 0.31%
concurrent/16 iocp latency_mean_ns 131.61 µs 133.86 µs 🔴▼ -2.07% 6.25%
concurrent/4 iocp latency_mean_ns 30.37 µs 30.46 µs 🔴▼ -0.38% 0.38%
concurrent_lockless/1 iocp latency_mean_ns 7.57 µs 7.59 µs 🔴▼ -0.48% 0.39%
concurrent_lockless/16 iocp latency_mean_ns 121.31 µs 122.08 µs 🔴▼ -0.62% 0.33%
concurrent_lockless/4 iocp latency_mean_ns 30.41 µs 30.44 µs 🔴▼ -0.03% 0.33%
pingpong/1 iocp latency_mean_ns 7.54 µs 7.59 µs 🔴▼ -1.07% 0.52%
pingpong/1024 iocp latency_mean_ns 7.66 µs 7.72 µs 🔴▼ -0.78% 0.23%
pingpong/64 iocp latency_mean_ns 7.57 µs 7.61 µs 🔴▼ -0.54% 0.38%
pingpong_lockless/1 iocp latency_mean_ns 7.54 µs 7.59 µs 🔴▼ -0.50% 0.46%
pingpong_lockless/1024 iocp latency_mean_ns 7.67 µs 7.71 µs 🔴▼ -0.58% 0.54%
pingpong_lockless/64 iocp latency_mean_ns 7.60 µs 7.60 µs 🔴▼ -0.04% 0.32%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 iocp bytes_per_sec 282.2 MB/s 280.1 MB/s 🔴▼ -0.67% 0.41%
bidirectional/1048576 iocp bytes_per_sec 4.81 GB/s 4.40 GB/s 🔴▼ -10.28% 5.91%
bidirectional/16384 iocp bytes_per_sec 3.72 GB/s 3.73 GB/s 🟢▲ +0.43% 0.94%
bidirectional/262144 iocp bytes_per_sec 9.26 GB/s 9.58 GB/s 🔴▼ -0.10% 7.70%
bidirectional/4096 iocp bytes_per_sec 1.08 GB/s 1.08 GB/s 🔴▼ -0.11% 0.47%
bidirectional/65536 iocp bytes_per_sec 8.69 GB/s 9.01 GB/s 🟢▲ +1.13% 8.10%
bidirectional_lockless/1024 iocp bytes_per_sec 281.3 MB/s 280.7 MB/s 🔴▼ -0.08% 0.33%
bidirectional_lockless/1048576 iocp bytes_per_sec 5.02 GB/s 4.66 GB/s 🔴▼ -3.68% 9.14%
bidirectional_lockless/16384 iocp bytes_per_sec 3.72 GB/s 3.72 GB/s 🟢▲ +0.18% 0.55%
bidirectional_lockless/262144 iocp bytes_per_sec 9.55 GB/s 9.64 GB/s 🔴▼ -0.19% 2.76%
bidirectional_lockless/4096 iocp bytes_per_sec 1.08 GB/s 1.08 GB/s 🟢▲ +0.29% 0.40%
bidirectional_lockless/65536 iocp bytes_per_sec 8.94 GB/s 8.97 GB/s 🟢▲ +0.32% 0.51%
multithread/2 iocp bytes_per_sec 12.60 GB/s 12.56 GB/s 🔴▼ -0.51% 0.82%
multithread/4 iocp bytes_per_sec 16.73 GB/s 16.79 GB/s 🟢▲ +0.44% 0.98%
multithread/8 iocp bytes_per_sec 18.40 GB/s 18.41 GB/s 🟢▲ +0.18% 0.95%
unidirectional/1024 iocp bytes_per_sec 285.0 MB/s 284.5 MB/s 🔴▼ -0.22% 0.60%
unidirectional/1048576 iocp bytes_per_sec 4.07 GB/s 4.08 GB/s 🟢▲ +3.90% 4.56%
unidirectional/16384 iocp bytes_per_sec 3.78 GB/s 3.79 GB/s 🟢▲ +0.19% 0.62%
unidirectional/262144 iocp bytes_per_sec 8.75 GB/s 9.19 GB/s 🟢▲ +3.33% 4.88%
unidirectional/4096 iocp bytes_per_sec 1.09 GB/s 1.09 GB/s 🔴▼ -0.59% 0.42%
unidirectional/65536 iocp bytes_per_sec 9.29 GB/s 8.84 GB/s 🔴▼ -0.45% 9.67%
unidirectional_lockless/1024 iocp bytes_per_sec 284.3 MB/s 283.8 MB/s 🔴▼ -0.09% 0.56%
unidirectional_lockless/1048576 iocp bytes_per_sec 4.44 GB/s 4.35 GB/s 🟢▲ +1.15% 3.18%
unidirectional_lockless/16384 iocp bytes_per_sec 3.77 GB/s 3.71 GB/s 🔴▼ -0.12% 4.44%
unidirectional_lockless/262144 iocp bytes_per_sec 9.51 GB/s 9.70 GB/s 🟢▲ +1.77% 4.27%
unidirectional_lockless/4096 iocp bytes_per_sec 1.09 GB/s 1.08 GB/s 🟢▲ +0.01% 1.34%
unidirectional_lockless/65536 iocp bytes_per_sec 8.98 GB/s 8.85 GB/s 🟢▲ +3.55% 12.69%
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 4.61K ops/s 4.72K ops/s 🟢▲ +1.99% 0.51%
burst/100 kqueue ops_per_sec 469.5 ops/s 466.9 ops/s 🔴▼ -0.88% 1.81%
burst_lockless/10 kqueue ops_per_sec 4.62K ops/s 4.78K ops/s 🟢▲ +3.26% 0.63%
burst_lockless/100 kqueue ops_per_sec 476.5 ops/s 472.6 ops/s 🔴▼ -0.89% 1.74%
concurrent/1 kqueue ops_per_sec 20.71K ops/s 20.58K ops/s 🔴▼ -0.06% 0.62%
concurrent/16 kqueue ops_per_sec 37.55K ops/s 37.61K ops/s 🔴▼ -0.07% 0.69%
concurrent/4 kqueue ops_per_sec 29.30K ops/s 29.35K ops/s 🟢▲ +0.04% 0.22%
sequential kqueue ops_per_sec 20.97K ops/s 21.01K ops/s 🔴▼ -0.47% 0.91%
sequential_lockless kqueue ops_per_sec 20.91K ops/s 20.84K ops/s 🟢▲ +0.63% 1.03%

fan_out

Benchmark Backend Metric Base Head Δ Noise
concurrent_parents/1 kqueue ops_per_sec 8.12K ops/s 8.20K ops/s 🟢▲ +1.11% 0.57%
concurrent_parents/16 kqueue ops_per_sec 8.23K ops/s 8.19K ops/s 🟢▲ +0.95% 17.06%
concurrent_parents/4 kqueue ops_per_sec 8.33K ops/s 8.26K ops/s 🔴▼ -0.43% 1.42%
concurrent_parents_lockless/1 kqueue ops_per_sec 8.20K ops/s 8.30K ops/s 🟢▲ +1.84% 0.83%
concurrent_parents_lockless/16 kqueue ops_per_sec 8.27K ops/s 8.22K ops/s 🔴▼ -0.02% 16.45%
concurrent_parents_lockless/4 kqueue ops_per_sec 8.35K ops/s 8.34K ops/s 🔴▼ -0.35% 1.65%
fork_join/1 kqueue ops_per_sec 48.92K ops/s 49.14K ops/s 🟢▲ +0.50% 0.87%
fork_join/16 kqueue ops_per_sec 8.14K ops/s 8.18K ops/s 🟢▲ +0.70% 0.55%
fork_join/4 kqueue ops_per_sec 23.65K ops/s 23.65K ops/s 🔴▼ -0.30% 0.63%
fork_join/64 kqueue ops_per_sec 2.12K ops/s 2.14K ops/s 🟢▲ +0.69% 2.26%
fork_join_lockless/1 kqueue ops_per_sec 50.21K ops/s 50.44K ops/s 🟢▲ +0.56% 0.31%
fork_join_lockless/16 kqueue ops_per_sec 8.26K ops/s 8.29K ops/s 🟢▲ +0.02% 1.04%
fork_join_lockless/4 kqueue ops_per_sec 23.65K ops/s 23.76K ops/s 🟢▲ +0.48% 0.35%
fork_join_lockless/64 kqueue ops_per_sec 2.13K ops/s 2.13K ops/s 🔴▼ -0.32% 2.51%
nested/16 kqueue ops_per_sec 2.08K ops/s 2.07K ops/s 🔴▼ -0.35% 2.21%
nested/4 kqueue ops_per_sec 8.05K ops/s 8.10K ops/s 🟢▲ +0.09% 0.65%
nested_lockless/16 kqueue ops_per_sec 2.11K ops/s 2.09K ops/s 🟢▲ +0.22% 2.30%
nested_lockless/4 kqueue ops_per_sec 8.15K ops/s 8.16K ops/s 🟢▲ +0.21% 0.78%

http_server

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue ops_per_sec 49.68K ops/s 49.50K ops/s 🔴▼ -0.42% 0.40%
concurrent/16 kqueue ops_per_sec 130.0K ops/s 130.8K ops/s 🟢▲ +1.12% 0.58%
concurrent/32 kqueue ops_per_sec 128.7K ops/s 129.5K ops/s 🟢▲ +1.06% 1.06%
concurrent/4 kqueue ops_per_sec 90.51K ops/s 90.83K ops/s 🟢▲ +0.18% 0.47%
multithread/1 kqueue ops_per_sec 127.9K ops/s 129.5K ops/s 🟢▲ +1.00% 1.20%
multithread/16 kqueue ops_per_sec 133.0K ops/s 132.8K ops/s 🔴▼ -0.45% 0.62%
multithread/2 kqueue ops_per_sec 185.5K ops/s 185.5K ops/s 🔴▼ -0.08% 0.50%
multithread/4 kqueue ops_per_sec 161.0K ops/s 162.9K ops/s 🟢▲ +1.72% 0.77%
multithread/8 kqueue ops_per_sec 136.9K ops/s 138.0K ops/s 🟢▲ +0.22% 1.93%
single_conn kqueue ops_per_sec 49.60K ops/s 49.53K ops/s 🔴▼ -0.38% 0.52%
single_conn_lockless kqueue ops_per_sec 49.87K ops/s 49.87K ops/s 🟢▲ +0.27% 0.39%

local_socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue latency_mean_ns 1.82 µs 1.71 µs 🟢▲ +6.68% 0.58%
concurrent/16 kqueue latency_mean_ns 27.13 µs 26.77 µs 🟢▲ +1.25% 0.23%
concurrent/4 kqueue latency_mean_ns 6.82 µs 6.66 µs 🟢▲ +2.38% 0.44%
concurrent_lockless/1 kqueue latency_mean_ns 1.80 µs 1.69 µs 🟢▲ +6.70% 0.36%
concurrent_lockless/16 kqueue latency_mean_ns 26.90 µs 26.57 µs 🟢▲ +1.35% 0.21%
concurrent_lockless/4 kqueue latency_mean_ns 6.74 µs 6.57 µs 🟢▲ +2.68% 0.40%
pingpong/1 kqueue latency_mean_ns 1.82 µs 1.70 µs 🟢▲ +6.22% 0.41%
pingpong/1024 kqueue latency_mean_ns 1.92 µs 1.80 µs 🟢▲ +6.24% 0.51%
pingpong/64 kqueue latency_mean_ns 1.82 µs 1.70 µs 🟢▲ +6.36% 0.33%
pingpong_lockless/1 kqueue latency_mean_ns 1.80 µs 1.68 µs 🟢▲ +6.69% 0.40%
pingpong_lockless/1024 kqueue latency_mean_ns 1.91 µs 1.78 µs 🟢▲ +6.68% 0.47%
pingpong_lockless/64 kqueue latency_mean_ns 1.81 µs 1.69 µs 🟢▲ +6.34% 0.51%

local_socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 kqueue bytes_per_sec 1.30 GB/s 1.32 GB/s 🟢▲ +1.68% 0.32%
bidirectional/1048576 kqueue bytes_per_sec 3.88 GB/s 3.87 GB/s 🔴▼ -0.28% 0.74%
bidirectional/16384 kqueue bytes_per_sec 3.89 GB/s 3.90 GB/s 🟢▲ +0.82% 1.06%
bidirectional/262144 kqueue bytes_per_sec 3.88 GB/s 3.87 GB/s 🔴▼ -0.69% 1.19%
bidirectional/4096 kqueue bytes_per_sec 2.91 GB/s 2.94 GB/s 🟢▲ +0.70% 0.82%
bidirectional/65536 kqueue bytes_per_sec 3.88 GB/s 3.88 GB/s 🟢▲ +0.15% 0.94%
bidirectional_lockless/1024 kqueue bytes_per_sec 1.31 GB/s 1.33 GB/s 🟢▲ +1.69% 0.23%
bidirectional_lockless/1048576 kqueue bytes_per_sec 3.98 GB/s 3.96 GB/s 🔴▼ -0.41% 0.58%
bidirectional_lockless/16384 kqueue bytes_per_sec 3.98 GB/s 3.99 GB/s 🟢▲ +0.33% 1.05%
bidirectional_lockless/262144 kqueue bytes_per_sec 3.98 GB/s 3.95 GB/s 🔴▼ -0.97% 0.84%
bidirectional_lockless/4096 kqueue bytes_per_sec 3.00 GB/s 2.99 GB/s 🔴▼ -0.35% 0.42%
bidirectional_lockless/65536 kqueue bytes_per_sec 3.99 GB/s 3.97 GB/s 🔴▼ -0.95% 0.77%
unidirectional/1024 kqueue bytes_per_sec 1.28 GB/s 1.32 GB/s 🟢▲ +3.06% 0.37%
unidirectional/1048576 kqueue bytes_per_sec 3.53 GB/s 3.54 GB/s 🔴▼ -0.13% 0.66%
unidirectional/16384 kqueue bytes_per_sec 3.54 GB/s 3.55 GB/s 🟢▲ +0.55% 0.60%
unidirectional/262144 kqueue bytes_per_sec 3.53 GB/s 3.54 GB/s 🟢▲ +0.45% 0.44%
unidirectional/4096 kqueue bytes_per_sec 2.72 GB/s 2.72 GB/s 🔴▼ -0.11% 0.50%
unidirectional/65536 kqueue bytes_per_sec 3.53 GB/s 3.54 GB/s 🟢▲ +0.35% 0.50%
unidirectional_lockless/1024 kqueue bytes_per_sec 1.28 GB/s 1.32 GB/s 🟢▲ +2.91% 0.34%
unidirectional_lockless/1048576 kqueue bytes_per_sec 3.61 GB/s 3.62 GB/s 🔴▼ -0.03% 0.74%
unidirectional_lockless/16384 kqueue bytes_per_sec 3.61 GB/s 3.62 GB/s 🔴▼ -0.01% 0.85%
unidirectional_lockless/262144 kqueue bytes_per_sec 3.61 GB/s 3.62 GB/s 🟢▲ +0.24% 0.74%
unidirectional_lockless/4096 kqueue bytes_per_sec 2.76 GB/s 2.77 GB/s 🟢▲ +0.12% 0.72%
unidirectional_lockless/65536 kqueue bytes_per_sec 3.61 GB/s 3.63 GB/s 🔴▼ -0.18% 0.90%

socket_latency

Benchmark Backend Metric Base Head Δ Noise
concurrent/1 kqueue latency_mean_ns 18.75 µs 18.67 µs 🟢▲ +0.39% 0.22%
concurrent/16 kqueue latency_mean_ns 120.31 µs 120.22 µs 🟢▲ +0.30% 0.89%
concurrent/4 kqueue latency_mean_ns 43.31 µs 42.81 µs 🟢▲ +0.67% 0.99%
concurrent_lockless/1 kqueue latency_mean_ns 18.67 µs 18.62 µs 🟢▲ +0.25% 0.10%
concurrent_lockless/16 kqueue latency_mean_ns 120.05 µs 119.97 µs 🟢▲ +0.30% 1.14%
concurrent_lockless/4 kqueue latency_mean_ns 42.87 µs 42.30 µs 🟢▲ +1.20% 1.05%
pingpong/1 kqueue latency_mean_ns 15.37 µs 15.55 µs 🔴▼ -0.69% 1.48%
pingpong/1024 kqueue latency_mean_ns 19.04 µs 19.00 µs 🟢▲ +0.06% 0.24%
pingpong/64 kqueue latency_mean_ns 18.73 µs 18.69 µs 🟢▲ +0.39% 0.28%
pingpong_lockless/1 kqueue latency_mean_ns 15.34 µs 15.33 µs 🟢▲ +0.02% 0.14%
pingpong_lockless/1024 kqueue latency_mean_ns 18.98 µs 19.05 µs 🔴▼ -0.19% 0.47%
pingpong_lockless/64 kqueue latency_mean_ns 18.64 µs 18.62 µs 🟢▲ +0.18% 0.06%

socket_throughput

Benchmark Backend Metric Base Head Δ Noise
bidirectional/1024 kqueue bytes_per_sec 468.7 MB/s 479.0 MB/s 🟢▲ +1.26% 1.77%
bidirectional/1048576 kqueue bytes_per_sec 7.62 GB/s 7.73 GB/s 🟢▲ +1.89% 0.92%
bidirectional/16384 kqueue bytes_per_sec 4.49 GB/s 4.54 GB/s 🟢▲ +1.04% 1.37%
bidirectional/262144 kqueue bytes_per_sec 8.77 GB/s 8.76 GB/s 🟢▲ +2.15% 2.33%
bidirectional/4096 kqueue bytes_per_sec 1.80 GB/s 1.83 GB/s 🟢▲ +0.82% 1.90%
bidirectional/65536 kqueue bytes_per_sec 7.42 GB/s 7.59 GB/s 🟢▲ +2.02% 1.44%
bidirectional_lockless/1024 kqueue bytes_per_sec 466.9 MB/s 478.2 MB/s 🟢▲ +2.47% 1.36%
bidirectional_lockless/1048576 kqueue bytes_per_sec 7.53 GB/s 7.77 GB/s 🟢▲ +1.48% 3.59%
bidirectional_lockless/16384 kqueue bytes_per_sec 4.54 GB/s 4.62 GB/s 🟢▲ +1.93% 1.79%
bidirectional_lockless/262144 kqueue bytes_per_sec 8.91 GB/s 8.94 GB/s 🟢▲ +2.22% 4.30%
bidirectional_lockless/4096 kqueue bytes_per_sec 1.80 GB/s 1.84 GB/s 🟢▲ +1.56% 2.31%
bidirectional_lockless/65536 kqueue bytes_per_sec 7.52 GB/s 7.69 GB/s 🟢▲ +3.66% 1.72%
multithread/2 kqueue bytes_per_sec 7.42 GB/s 7.50 GB/s 🔴▼ -0.68% 3.03%
multithread/4 kqueue bytes_per_sec 3.59 GB/s 3.61 GB/s 🔴▼ -3.56% 10.65%
multithread/8 kqueue bytes_per_sec 3.23 GB/s 3.09 GB/s 🔴▼ -5.53% 7.78%
unidirectional/1024 kqueue bytes_per_sec 454.3 MB/s 461.3 MB/s 🟢▲ +0.88% 1.10%
unidirectional/1048576 kqueue bytes_per_sec 8.76 GB/s 8.87 GB/s 🟢▲ +1.11% 1.53%
unidirectional/16384 kqueue bytes_per_sec 4.41 GB/s 4.51 GB/s 🟢▲ +2.01% 1.81%
unidirectional/262144 kqueue bytes_per_sec 8.81 GB/s 8.82 GB/s 🟢▲ +0.38% 4.42%
unidirectional/4096 kqueue bytes_per_sec 1.78 GB/s 1.83 GB/s 🟢▲ +2.18% 2.09%
unidirectional/65536 kqueue bytes_per_sec 7.35 GB/s 7.38 GB/s 🟢▲ +0.56% 1.29%
unidirectional_lockless/1024 kqueue bytes_per_sec 469.2 MB/s 484.9 MB/s 🟢▲ +3.42% 1.35%
unidirectional_lockless/1048576 kqueue bytes_per_sec 8.82 GB/s 8.89 GB/s 🟢▲ +0.28% 1.55%
unidirectional_lockless/16384 kqueue bytes_per_sec 4.43 GB/s 4.59 GB/s 🟢▲ +4.01% 2.68%
unidirectional_lockless/262144 kqueue bytes_per_sec 9.00 GB/s 9.07 GB/s 🟢▲ +1.41% 5.72%
unidirectional_lockless/4096 kqueue bytes_per_sec 1.79 GB/s 1.84 GB/s 🟢▲ +2.11% 1.74%
unidirectional_lockless/65536 kqueue bytes_per_sec 7.34 GB/s 7.38 GB/s 🟢▲ +0.19% 1.19%

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

@sgerbino
sgerbino force-pushed the pr/recycling-free-list branch from 1e59c16 to f4b0d0d Compare November 12, 2026 19:16
@coderabbitai

coderabbitai Bot commented Nov 12, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

@sgerbino
sgerbino force-pushed the pr/recycling-free-list branch from f4b0d0d to 97fbd80 Compare November 13, 2026 05:57
@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
@sgerbino
sgerbino force-pushed the pr/recycling-free-list branch from 97fbd80 to 54d7f4b Compare November 17, 2026 22:48
@sgerbino sgerbino changed the title perf: run handlers queued behind the reactor before polling again perf: run handlers queued behind the reactor first while that pays off Nov 17, 2026
When the reactor sentinel reached the front with handlers still queued,
the scheduler polled the reactor without blocking before running them.
Those handlers usually start the next I/O themselves, such as a
coroutine an I/O completion woke, so running them first can save a
reactor pass.

The sentinel now moves behind the queued handlers once and polls on its
next turn regardless, so handlers that keep reposting still cannot
starve I/O. Waiting does not always pay: when the handlers drain the
queue, the deferred turn has nothing left to overlap with, and where
loopback data arrives later their I/O parks instead of completing.
reactor_defer_policy scores each deferral by whether its turn still
found handlers queued, and stops deferring while deferrals that empty
the queue are frequent. Turns that poll instead let it try again.

restart() forgets a pending deferral and the score, so a run stopped
with the sentinel moved back doesn't cost the next run its own. poll()
and poll_one() are unchanged.
@sgerbino
sgerbino force-pushed the pr/recycling-free-list branch from 54d7f4b to cf9d859 Compare November 17, 2026 23:02

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.

2 participants