sigtrap only triggers on INT

[miekg]

killing caddy (and coredns) should trigger OnFinalShutdown callbacks, while looking at the code (on why this isn't happening), I see https://github.com/mholt/caddy/blob/master/sigtrap.go#L40 is only trapping on SIGINT.

Is this intentional? Add SIGTERM here would make a lot of sense of kill sends this by default.

[mholt]

Yeah, this is intentional, and is as-documented: https://caddyserver.com/docs/cli#signals -- SIGINT and SIGQUIT execute the shutdown hooks. I suppose this could be changed if necessary, but I just haven't had a good reason to (yet?).

[miekg]

fair enough
(/me should RTFM)

Thanks!

[mholt]

No problem! Let me know if it causes any trouble as-is.

[miekg]

thinking some more. SIGTERM should do this as well, can prolly find in the posix spec. But graceful shutdown usually is done via TERM (and then a KILL).

Kubernetes uses TERM as well, not sure I can find a way to change this for a specific deployment.

[miekg]

Yes, please change; as SIGTERM and then a SIGKILL is pretty normal.

Alternatively I could prepare (or move invasive PR) to makes this configurable, so CoreDNS can behave differerently than Caddy in this regard.

[miekg]

looked at the wrong file. I (only?) need the sigtrap_posix.go on. This is trivial if you want to make TERM call the callback as well.
If not a caddy.<Variable> could be exported that when set will just call fallthrough from the syscall.SIGTERM case.