Flannel network lost, requiring service restart

[Alex Pearce]

Kubernetes pods regularly (several times per week) lose their ability to communicate over the Flannel network. I have noticed these patterns:

1. The Ubuntu 20.04 nodes have never exhibited this problem, only Ubuntu 22.04 nodes.
2. Restarting the Flannel service usually restores Flannel network connectivity (`systemctl restart flannel`).

My best guess is that this is due to https://github.com/flannel-io/flannel/issues/1474, however I have not been able to pin down exactly why. In particular I do not understand why my 20.04 nodes do not exhibit the problem.

All nodes are running revision 52 of the Canal charm, version 0.11.0+ck1/3.10.1.

The issue linked above was addressed in Flannel v0.15.1. The Flannel charm pins v0.11.0:

https://github.com/charmed-kubernetes/charm-flannel/blob/523336db25001088a7336dffb2616c5be23a46fb/build-flannel-resources.sh#L4

I am happy to submit a PR which bumps the version. Ideally I would test this but am unsure how.

[Alex Pearce (alexpearce)]

https://github.com/charmed-kubernetes/charm-flannel/pull/85

[Adam Dyess (addyess)]

Updated the PR:
* includes changes to flannel version
* includes changed to etcd version
* updates etcdctl commands to use etcd v3
* support for building charm resources with github actions

https://github.com/charmed-kubernetes/charm-flannel/pull/86

I confirmed that an upgrade from the previous flannel updated to etcd v3 without issue.

[Kevin W Monroe (kwmonroe)]

We'll need a new pin so canal picks up the flannel build changes -- similar to what we did here:

https://github.com/charmed-kubernetes/layer-canal/pull/73/files

[Alex Pearce (alexpearce)]

This is super, thanks very much for the help.

I've opened a layer-canal PR to follow the Flannel change:

https://github.com/charmed-kubernetes/layer-canal/pull/76

[Kevin W Monroe (kwmonroe)]

Followup PR to sync the charm-flannel etcd3 and resource-building changes to canal:

https://github.com/charmed-kubernetes/layer-canal/pull/77