Fix: SIGINT was ignored by the cln docker container

[NicolasDorier]

Changelog-Fixed: Fix: SIGINT was ignored by the cln docker container

This was reported by one of our user.

I could replicate with

docker kill --signal=SIGINT <container>

I confirm that now both, with and without --signal=SIGINT works properly.

The file to fix is tools/docker-entrypoint.sh (the Dockerfile copies it to /entrypoint.sh at lines 245 and 278 for the two image stages — the Dockerfile itself is fine).

The bug lives in three lines of that script:

• Lines 64–65: set -m + lightningd "$@" & — lightningd is started as a background job in its own process group, so it will never see signals sent to PID 1.
• Line 122 (last line): fg %- — bash re-foregrounds lightningd and blocks there as PID 1 with no trap. A PID 1 with no handler installed never receives SIGTERM at all, so docker stop does nothing until the grace-period expires and the container is SIGKILLed — exit 137 on every shutdown, and lightningd never closes its database cleanly.

The fix — replace line 122's fg %- with:

trap 'kill -TERM "$LIGHTNINGD_PID" 2>/dev/null' TERM INT
wait "$LIGHTNINGD_PID"
trap - TERM INT
wait "$LIGHTNINGD_PID"
exit $?

Why this shape and not just adding a trap above fg: bash defers trap execution while a foreground job is running, so a trap combined with fg would only fire after lightningd already exited. wait, by contrast, is interrupted by trapped signals — the first wait returns when SIGTERM arrives and the trap forwards it to lightningd, the second wait collects lightningd's actual exit status.