You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
CI: issue411-abort-usage 'client abort keeps sniffed usage' timed out once on windows-latest Node 22; cause not established #52
Category K - launcher, self-update and CI. Build: PR #51 head 538b3e36e (based on master at f11665bd2; the PR changes tests/bounded-maintenance.test.ts only). Screened open and closed issues and PRs on this fork: #6 (closed) is a different client-abort test, tests/wire-exit-matrix.test.ts, on the same job cell; #3 and #11 are unrelated. Nothing covers this test.
Observation
tests/issue411-abort-usage.test.ts, test #411 plugin stream: client abort keeps sniffed usage, logs info (not error), clears the upstream timer, failed once in CI:
condition not met within timeout, thrown by waitFor (line 139) called from line 162; test duration 3450 ms. It was the only failure in the job (1918 pass, 1 fail, 13 skipped). The two later #411 tests in the same file passed.
Which wait timed out. Three consecutive 3-second waitFor calls follow the abort (usage recorded as 1234 input tokens; live upstream timers back to 0; the info log client aborted mid-stream). The reported line 162 is the first of them in the checked-out file, but line mapping under tsx was not verified, and the 3450 ms duration fits any single one of them.
The cause. The job log has no proxy log lines for this test. Not reproduced locally; no local run was attempted.
Do not assume a slow runner or change runtime behaviour without reproducing. A comment in the test records an earlier flake of the same test on a loaded windows-latest runner (ranxianglei#2305 review upstream), which was addressed by adding the third wait.
Suggested next step
Loop the file on Windows with Node 22 under CPU load, with the failing waitFor reporting which condition it was waiting on, before deciding between a longer wait and a product-side ordering problem.
Reproduction attempt, 2026-11-14: tests/issue411-abort-usage.test.ts on Windows 11 with Node v22.23.4 (official zip, not installed), 10 rounds of 4 copies in parallel using the documented test command. Result: clean 40 of 40 (exit 0, 0 failed, 0 cancelled). Not reproduced; cause still not established.
PR #65 makes the timeout error name which wait expired. It changes no wait or assertion.
Fork CI has run Windows / Node 24 only since #54, so the cell where this was seen no longer exists. Recommendation for the owner: close this as not reproducible on a retired cell once #65 is in, and reopen if the test fails on Node 24.
Activity
anuzzolilli-oss commentedon Nov 14, 2026
(ai: anthropic/claude-fable-5-2)
Reproduction attempt, 2026-11-14:
tests/issue411-abort-usage.test.tson Windows 11 with Node v22.23.4 (official zip, not installed), 10 rounds of 4 copies in parallel using the documented test command. Result: clean 40 of 40 (exit 0, 0 failed, 0 cancelled). Not reproduced; cause still not established.PR #65 makes the timeout error name which wait expired. It changes no wait or assertion.
Fork CI has run Windows / Node 24 only since #54, so the cell where this was seen no longer exists. Recommendation for the owner: close this as not reproducible on a retired cell once #65 is in, and reopen if the test fails on Node 24.