Improve CI reliability and speed

[keyvaann]

At the moment the CI jobs frequently fail or take a long time to run due to various reasons, this should be fixed in order to in order to make contributions easier. List of issues and their proposed solution:

• Being forced to run install helm installation at time because of --concurrency 1, merging Use helmfile dependency management #241 should solve this issue.
• Pulling from Dockerhub on each install. This is a failing point since we often run into Dockerhub quota limitation and need to retry. Solution would be to have a pull through docker registry as a cache and change the configuration to pull from there. Following guides can be used:
  • https://docs.k3s.io/installation/private-registry
  • https://thomasdegraaff.eu/notes/2021/12/04/k3s_docker_pull_through_registry/
• Use merge queue
  • https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/configuring-pull-request-merges/managing-a-merge-queue#triggering-merge-group-checks-with-github-actions

[pvannierop]

@keyvaann So, if I understand well, we reach the limit during CI because every time we destroy and reinstall the K3S cluster, right? Would it be a solution to not destroy the cluster, but run helmfile destroy? Would that keep all the docker images in the K3S cache?

[keyvaann]

I don't know if helmfile destroy will also remove the CRDs, if it does then we can use it. If it doesn't, we can copy the docker image cache somewhere else before removing the cluster and then restore it after installing the K3S again. That would be quite a fast and easy fix.

[keyvaann]

Related to #298 and #299