Update containerd charm "detects" proxy change, restarts service and breaks connections

After containerd charm was upgraded from 61 to later version it starts to break long running connections via kubectl port-forward (logs, exex sessions are broken too). It exits with an error

...
Handling connection for 6666
E0812 11:08:41.769848   30805 portforward.go:233] lost connection to pod

At the same containerd unit log shows:

2020-08-12 11:08:47 INFO juju-log Reactive main running for hook update-status
2020-08-12 11:08:47 INFO juju-log Invoking reactive handler: reactive/containerd.py:172:update_status
2020-08-12 11:08:47 INFO juju-log Invoking reactive handler: reactive/containerd.py:449:proxy_changed
2020-08-12 11:08:48 INFO juju-log Proxy changed, writing new file to /etc/systemd/system/containerd.service.d/proxy.conf
2020-08-12 11:08:48 INFO juju-log Invoking reactive handler: reactive/containerd.py:506:publish_config
2020-08-12 11:08:48 INFO juju-log Invoking reactive handler: reactive/container_runtime_common.py:15:enable_grub_cgroups
2020-08-12 11:08:48 INFO juju-log Invoking reactive handler: hooks/relations/docker-registry/requires.py:58:remove_ready:docker-registry
2020-08-12 11:08:48 INFO juju-log Invoking reactive handler: hooks/relations/untrusted-container-runtime/provides.py:18:broken:untrusted
2020-08-12 11:08:48 INFO juju-log Invoking reactive handler: reactive/containerd.py:490:restart_containerd
2020-08-12 11:08:48 INFO juju-log status-set: maintenance: Restarting containerd
2020-08-12 11:08:48 INFO juju-log status-set: active: Container runtime available
2020-08-12 11:08:48 INFO juju.worker.uniter.operation runhook.go:142 ran "update-status" hook (via explicit, bespoke hook script)

This behavior observed on charm version 80, but if my memory does not fail some earlier versions showed same behavior. No such behavior is observed on version 61, where there are less stuff installed. I have tried to compare charms, but was not able to relate changes with such behavior.

There are no changes in the proxy configuration. File content remains the same. Cluster is installed and nothing being done. What can trigger this reactive hook?

Hey @elvinas,

I’ve not seen that issue before. Could you please log it as a bug (Bugs : Charmed Kubernetes) against “Containerd Subordinate Charm”?

Please also attach the tarball generated by juju-crashdump -a debug-layer -a config. You may, first, have to run sudo snap install juju-crashdump --classic.

Thanks,
Joe

It seems was already was a known issue and it is fixed in release 88.

So case can be closed.