snapcraft/commands/daemon.start: always remount /bin/kmod

[simondeziel]

Upon update, the daemon.start would keep any existing mountbind of /bin/kmod rather than remount a fresh one. This lead to such situation:

root@sdeziel-lemur:~# nsenter -m -t "$(lxc info | awk '/server_pid:/ {print $2}')"
root@sdeziel-lemur:/# mount | grep kmod
/var/lib/snapd/snaps/lxd_26615.snap (deleted) on /usr/bin/kmod type squashfs (ro,relatime,errors=continue,threads=single

Always remounting that script should avoid holding onto old snap data.

[tomponline]

Does LXD also need to umount this on LXD stop?

[tomponline]

@simondeziel I had a quick search in lxd-pkg-snap for /bin/kmod and noticed there are some hardcoded references to "core22", given we are planning to change to core24 soon, I suspect this may become an issue. Please could you see if there is a way to make that dynamic somehow to refer to the correct core?

[tomponline]

@simondeziel ah looks like you had the same though as me here #262 thanks!

[simondeziel]

Does LXD also need to umount this on LXD stop?

daemon.stop does surprisingly little umount'ing, should we also umount other bits? @mihalicyn I'd like your input on this, please.

[tomponline]

Does LXD also need to umount this on LXD stop?

daemon.stop does surprisingly little umount'ing, should we also umount other bits? @mihalicyn I'd like your input on this, please.

If daemon.stop is called during systemctl reload snap.lxd.daemon then we probably should unmount anything as it could be in use by running instances.

[simondeziel]

While daemon.stop has code meant to handle reloads:

https://github.com/canonical/lxd-pkg-snap/blob/latest-edge/snapcraft/commands/daemon.stop#L32-L42:

echo "=> Stop reason is: ${reason}"
...
# Handle reloads and crashes
if [ "${reason}" = "reload" ] || [ "${reason}" = "crashed" ]; then
    exit 0
fi

It is not invoked by either systemctl reload snap.lxd.daemon or the heavier wait snap restart lxd --reload as I'm not seeing the => Stop reason is: message in the logs.

This is probably fine because reloads have a dedicated script (daemon.reload, https://github.com/canonical/lxd-pkg-snap/blob/latest-edge/snapcraft.yaml#L81) which essentially just signals the existing LXD that a reload is needed an then kill it.

[tomponline]

Well thats odd that snap restart lxd doesnt invoke it, id have thought it would invoke stop and then start hooks (as that does stop instances).

There are clearly differences though as there is this open snapd bug https://bugs.launchpad.net/snapd/+bug/2028141

[simondeziel]

Well thats odd that snap restart lxd doesnt invoke it, id have thought it would invoke stop and then start hooks (as that does stop instances).

Did you miss the --reload in snap restart lxd --reload?

[tomponline]

Well thats odd that snap restart lxd doesnt invoke it, id have thought it would invoke stop and then start hooks (as that does stop instances).

Did you miss the --reload in snap restart lxd --reload?

i did :)