update-test: the lock assertion flakes under system load #126

New issue

Open

opened 2026-08-23 22:28:43 +01:00 by joshquinlan · 0 comments

joshquinlan commented

2026-08-23 22:28:43 +01:00

Owner

"the lock the update takes can be held against it" fails intermittently when the machine is busy (roughly 4 to 7 in 10 under six concurrent workers, 0 in 10 idle). The background flock -n LOCK -c sleep holder sometimes exits at once, meaning something held the lock at that instant, yet fuser at the failure point names no holder and lock_free was awaited first. A run_detached fd leak was already ruled out and fixed (511d466). Next lead: instrument which flock invocation fails and whether the runtime-dir tmpfs is involved; or hold the lock from the test's own shell with exec {fd}<> instead of a child flock.

"the lock the update takes can be held against it" fails intermittently when the machine is busy (roughly 4 to 7 in 10 under six concurrent workers, 0 in 10 idle). The background `flock -n LOCK -c sleep` holder sometimes exits at once, meaning something held the lock at that instant, yet fuser at the failure point names no holder and lock_free was awaited first. A run_detached fd leak was already ruled out and fixed (511d466). Next lead: instrument which flock invocation fails and whether the runtime-dir tmpfs is involved; or hold the lock from the test's own shell with exec {fd}<> instead of a child flock.

joshquinlan added this to the 7 Update and polish milestone

2026-08-23 22:28:43 +01:00

joshquinlan added the

bug

status/backlog

repo/core

labels

2026-08-23 22:28:43 +01:00