* The suite watchdogs give up the first time a poll tick cannot start a process
ci_suite_heartbeat read its clock through a command substitution, so the
watchdog needed a fork to tell the time, and a sleep that returns without
waiting either killed it under the caller's errexit or turned
test-timeout.sh's wait loop into a fork storm. Both fail exactly when the
box is short of processes, which is when the watchdog is the only thing
still reporting.
The clock now assigns hb_time instead of printing, the heartbeat's tick is
guarded, and the wait loop gives up after ten ticks that never waited,
naming the test in the progress log the off-box watchdog reads.
References #1038.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Xavier Roche <roche@httrack.com>
* Review fixes: dump the crawl logs before TMPDIR goes, and pin the guards with mutants
The nofork exit skipped dump_crawl_logs, so the EXIT trap deleted the very
logs that would explain the hang. reap_bounded stays out: it polls on the
same broken tick.
250 now asserts the tenth tick by count, that the wedged child was killed,
and that a tick failing every other call is never read as a wedge. 171's
sleep stub returns 126 once, which is what catches the heartbeat guard
being removed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Xavier Roche <roche@httrack.com>
---------
Signed-off-by: Xavier Roche <roche@httrack.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
f5fbd1a82e