SendQueueBehaviourTest.restart_resumes_the_attachment_chain starved in CI #340 (60s runTest budget, phase-1 load) #122
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
CI #340 (commit
846b1e1): UncompletedCoroutinesError after 1m in :katrix-testkit:allTests during build phase 1 — the #74/D33 starvation shape, on a test deflaked once before (165b510). Green 3/3 locally at the same commit; nothing in846b1e1(CSPRNG seam, D71) plausibly blocks — JVM SecureRandom nextBytes is /dev/urandom, non-blocking. Watch: if it recurs under CI contention, either move testkit's jvmTest to the phase-2 quiet-runner list (D33 pattern) or raise this test's runTest budget.Likely root cause found: the D71 CSPRNG seam's JVM actual used new SecureRandom(), whose FIRST seed can come from /dev/random — blocking in an entropy-starved DinD container. #122's first occurrence was on that very commit (CI #340), and CI #348/#352 wedged in the session-heavy store/appservice suites (many pickle-key draws). Fixed by NativePRNGNonBlocking (urandom for seed and draws) with fallback; if the next runs stay clean this closes.
Correction: the CI #348/#352 wedges are NOT the /dev/random block — 352's final log shows the DinD daemon died mid-run (docker socket unreachable at kill; no test ever failed), tracked as the new runner-health issue. The NativePRNGNonBlocking change stands as correct hygiene, and this issue's original phase-1 starvation (CI #340) remains plausibly entropy/contention; reopen only if it recurs.