kevinveenbirkenbachandClaude Opus 5.5 c323288170 fix(swarm): end the whole step when the watchdog aborts it ...

translate#1 (CI run 40625679270) hit the disk floor and the watchdog
returned rc 75, but its proc.terminate() reached only the step's python
wrapper, which dies on SIGTERM without passing it on. ansible-playbook was
re-parented to PID 1 (rescue processes.txt) and kept deploying and testing
for another eleven minutes, ate the rescue headroom, and was SIGKILLed by
the teardown, which surfaced as a misleading rc 137.

The step now runs in its own process group; an abort signals the group
(SIGTERM, 30 s, SIGKILL). SIGINT and SIGTERM sent to the watchdog itself
are forwarded to that group while the step runs, so the outer `timeout`
in 00_one.sh (no --foreground) still stops the step.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-28 12:33:51 +01:00

..

swarm

fix(swarm): end the whole step when the watchdog aborts it

2026-10-28 12:33:51 +01:00

__init__.py

…