Repository navigation
Spin through the tick where sleeps wake late - #53
Merged
Merged
Conversation
The loop slept to 1 ms before each deadline and spun the rest. On an oversubscribed VM (GitHub's macOS runners: 3 cores, load near 20) a 10 ms sleep woke 30-90 ms late, so the 100 Hz loop ran near 40 Hz with status and acks a tick or more behind. A thread that spins is not preempted there, one that sleeps is. The timer now estimates how late its sleeps wake and widens the busy window to match, up to the whole tick; on a machine whose sleep is accurate the window stays at the configured 1 ms. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GUfKe4qYwmMZJgoch8LDT6
… clock The path tests' sampler polled status every 20 ms with a sleep; where sleeps wake late it caught a fast path too sparsely to see its corners, and with the loop now on time the arm moves fast enough there to miss them. It records every published frame as it arrives instead. The loop-timer tests run on a virtual clock, so a preempted CI runner cannot fake a late sleep where the test means an accurate one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GUfKe4qYwmMZJgoch8LDT6
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
LoopTimerslept until 1 ms before each deadline and spun the rest. On GitHub's macOS runners (3 vCPUs, load average near 20) a 10 mssleepwakes 30–90 ms late, so the 100 Hz loop ran at ~40 Hz: status went out at ~8 Hz instead of 20, and acks slipped a tick or more. That's the "macOS loopback runs late" pattern the harness tests worked around. Measured on CI: a spinning thread there sees no stalls, a sleeping one wakes late. A latency-critical QoS class made no difference.The timer now keeps an estimate of how late its sleeps wake and widens the busy window to
threshold + 2 × lateness, capped at the whole tick. On a machine whose sleep is accurate (Linux, Windows) the window stays at the configured 1 ms. While spinning, the estimate fades, so the loop retries sleeping and narrows again if sleep recovers. Float arithmetic only; nothing allocates in the tick.tests/unit/test_loop_timer_late_sleep.pysimulates sleeps that wake 20 ms late. Before: 100 overruns in 200 ticks; after: at most a few. With exact sleep the window stays near 1 ms.🤖 Generated with Claude Code
https://claude.ai/code/session_01GUfKe4qYwmMZJgoch8LDT6