fix(core): stop session work barrier from pinning a core at 100% CPU - #1084
Merged
Conversation
`SessionWorkBarrier.waitForSettled` re-samples the agent event queue in a `while (true)` loop that only ever awaits promises. When a session's subscriber re-chains `_agentEventQueue` with already-resolved promises, the two samples in a round never agree and the loop never leaves the microtask queue: timers and IO are starved and the process pins one core at 100% CPU. Measured at 25.6M iterations/sec, and reproduced here as a test file that hangs hard enough that vitest cannot even fire its own 30s timeout — matching the intermittent spin seen in `omo --mode rpc --multi-session` session children. The first rounds still re-sample immediately, because that tight re-check is what observes a continuation scheduled during the same turn and keeps queued work from being reported as settled. Only once the queue proves it is not converging does the loop yield a real event-loop turn, which bounds CPU without changing when a converging session settles. The yield uses `MessageChannel` rather than `setTimeout`/`setImmediate`: mocked timers replace both, which would stall every caller that drives retry and compaction recovery on a mocked clock, while `process.nextTick` runs ahead of the microtask drain and would not relieve the starvation at all. Verified: full `npm test` for @code-yeongyu/senpi green (8662 passed).
code-yeongyu
enabled auto-merge
August 22, 2026 13:29
…ier-spin # Conflicts: # packages/coding-agent/CHANGELOG.md
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
SessionWorkBarrier.waitForSettled ran a while(true) loop awaiting only already-resolved promises — never leaving the microtask queue (measured 25.6M iterations/sec at 100% CPU; a reproduction test hung past vitest's own 30s timeout with the process pinned at exactly 100.0%). This starves the event loop so RPC/socket IO (client acks) can't be serviced — the mechanism behind the desktop server's ack-timeout bursts under load.
Fix: keep the tight microtask re-sampling for the first 16 rounds (load-bearing for compaction recovery ordering — verified by test), and only a non-converging queue yields a real event-loop turn via MessageChannel (setTimeout/setImmediate deadlock under vi.useFakeTimers; process.nextTick runs ahead of the drain and never relieves starvation).
Verification: failing-first reproduction (hang/100% CPU) → fixed (0.4% CPU, 138ms); full @code-yeongyu/senpi suite 1059 files / 8662 tests green.
Summary by cubic
Stop
SessionWorkBarrier.waitForSettledfrom pinning a core at 100% CPU by yielding a real event-loop turn when the event queue does not converge; previously it re-sampled only already-resolved promises in a tight loop that never left the microtask queue.MessageChannelto yield, keeping timers and IO responsive and working under mocked timers (e.g.,vi.useFakeTimers()).packages/coding-agent/CHANGELOG.md; full@code-yeongyu/senpitest suite remains green.Written for commit b856e6d. Summary will update on new commits.