Signal-cli got stuck and health check failed

[marksftw]

Sage stopped responding in Signal. Claude debugged and got it working again by restarting the signal-cli service. It then made this recommendation:

One thing worth noting: the signal-cli health check in docker-compose.yml spawns bash/grep processes that become zombies when the check times out (because signal-cli's Java process is PID 1 and doesn't reap children). Over time this could exhaust the PID space. If this failure mode recurs, it might be worth either switching to a simpler health check or adding tini as an init process in that container.

[AnthonyRonning]

Is this on the new jre version?

[marksftw]

Yes, this is on the JRE version signal-cli-jre:latest. The switch from native/GraalVM was already done in #4. The theory is that the health check spawns bash/grep child processes, and since Java (PID 1) doesn't reap zombies, they accumulate over time and could eventually cause issues. Adding tini as init or simplifying the health check might help.

[AnthonyRonning]

I’ve tried a bunch of things to make it more reliable and nothing’s worked so far. If you want, you can go ahead and make that health check change on your own docker compose and see if it works and make a PR if it doesnt require a restart after 2 days.