SIGINT signals through crun fail

[baude]

When sending a SIGINT to a crun container, the signal seems to get eaten or ignored. If replace crun with runc, the it seems to work. What I don't understand is why this seems to be only observed using the remote podman client.

The following reproducer gets to the heart of the matter reasonably quick.

set -e
#signals=(1 2 3 4 5 6 8 10 12 13 14 15 16 20 21 22 23 24 25 26 64)
signals=(15 2)
SOCK_FILE=$(mktemp --dry-run --tmpdir io.podman.XXXXXX)
VARLINK_LOG=$(mktemp --dry-run --tmpdir varlink.log.XXXXXX)
export PODMAN_VARLINK_ADDRESS=unix:$SOCK_FILE
PODMAN="podman --runtime=crun"
PODMAN_REMOTE="podman-remote"

cleanup() {
    set +e
    echo "Cleaning up"
    $PODMAN_REMOTE stop --timeout 0 foobar
    $PODMAN_REMOTE rm -f foobar
    kill -9 %1
    rm -f $SOCK_FILE
    echo "Varlinkg log file: $VARLINK_LOG"
}

trap cleanup EXIT
echo "Starting background varlink server"
sudo rm -rf ${HOME}/.local/share/containers ${HOME}/.config/containers
$PODMAN varlink --timeout=0 $PODMAN_VARLINK_ADDRESS &> $VARLINK_LOG &
sleep 5s
echo "Starting background signal-recipient container"
$PODMAN_REMOTE run -d --name=foobar fedora-minimal sh -c "for i in ${signals[*]}; do trap \"echo got: \$i\" \$i; done; echo READY; while ! test -e /stop; do sleep 0.05; done;echo DONE"

for s in $(fmt --width=2 <<< "${signals[*]}" | sort -r); do
    echo "Sending signal $s"
    $PODMAN_REMOTE kill -s $s foobar &> /dev/null
    sleep 0.1s
    $PODMAN_REMOTE logs --tail 1 foobar
done
sleep 3s
echo "Final, complete ourput from foobar container:"
$PODMAN_REMOTE logs foobar

[baude]

@giuseppe can you take a look at this? we need this to get F31 CI testing working for podman. many thanks!

[giuseppe]

fixed here: #233

AFAICS, crun doesn't change the signals, so it might be podman-remote to do it.

It is possible to reset the handlers to the default with env --default-signal sh.. I suggest to leave the env --default-signal block anyway, as this behaviour is not specified in the OCI specs.

[baude]

@giuseppe the change of signal was a misinterpretation of the things initially. It looks like from our side the signal is just eaten.

[baude]

confirmed this fixes things