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
run-slot-gauges-lock.test.ts's dead-lock takeover case fails now and then on ubuntu-26.04 #1077
The gauges-lock test "eight looks at a dead lock together each read the stale or the new reading, and leave one well-formed reading and no lock" (tests/scripts/ops/run-slot-gauges-lock.test.ts) fails now and then on Linux: one of the eight looks prints a slot count other than 1 (the stale reading) or 4 (the new one).
Another uutils tool, at least in isolation: on ubuntu-26.04, mv is GNU 9.7, and rmdir and stat are uutils 0.10.0. A probe ran this case alone 60 times with only mkdir swapped and 60 times with all 104 of the image's gnu* tools first on PATH: 0 bad rounds each time. So it hasn't been reproduced outside the whole file.
What we don't know yet
Which value the bad look printed. The assertion reports only false, so a run that fails says nothing about the look: an empty line (the look died), a shut count, or something else.
Whether it needs the rest of the file running with it (a leftover look from an earlier case, the load from the one-second fake tops) or just more rounds than 120.
Done when
The failing value and its stderr are known. A first step is for the case to report each look's slots and stderr when it fails (toEqual on the list, or a message carrying them).
The cause is shown red on the current code and green after the fix, with 20 green runs in a row of the file in ubuntu:26.04.
Activity
fix(PG-956): the run-slot tests take a lock as the Mac does on CI's u…
fix(PG-1092): run-slot.sh clears a LANES from the caller's env at its…
fix(PG-1077): a look counts run-slot's slots 2 to 4 on one gauges rea…