mirror of
https://github.com/alexbelgium/hassio-addons.git
synced 2026-09-20 00:34:00 +02:00
The add-on could not be stopped: Home Assistant showed an Error status after a few seconds and the container kept running and still served the web UI (#3049). cont-init.d/99-run.sh started nginx in the foreground -- '&>' is a redirect, not a background operator -- and ha_entrypoint.sh runs every cont-init.d script sequentially in the foreground. That script therefore never returned, so the entrypoint never reached the code that installs the terminate() handler forwarding SIGTERM to the application. The reporter's log shows both halves of this: it prints 'Starting custom scripts' and never reaches 'Everything started!'. The add-on also shipped no 'init:' key, so Supervisor's default of true made Docker inject its own init as PID 1 and left ha_entrypoint.sh as PID 2, where the 'if $PID1' block holding the trap is skipped outright. Background the launch and set init: false. The script then returns, the entrypoint installs its trap, and nginx -- orphaned by the exiting script -- is reparented to the entrypoint as PID 1, where terminate()'s 'pgrep -P $$' finds it and signals it directly. Backgrounding from cont-init.d is what 24 other add-ons here already do (autobrr runs a bare 'nginx &'). Moving the launch to services.d was considered and rejected: ha_entrypoint.sh runs each services.d/*/run inside a restart subshell, so the application ends up a grandchild of PID 1 while terminate() enumerates direct children only. Reproduced -- the app survives that path unsignalled -- and it is the larger change. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>