Fix failing traefik install #34
Conversation
consideRatio
force-pushed
the
pr/fix-broken-traefik
branch
from
327eac0
to
3bfe212
Compare
|
@manics I rebased this on main after the calico bump was merged. |
The action fail using some k8s versions without this in current latest release. I think it is related to how they install Traefik v2 and traefik in general now via a HelmChart CRD. To handle this, we can add these wait steps and allow them to fail.
consideRatio
force-pushed
the
pr/fix-broken-traefik
branch
from
3bfe212
to
92bcc5c
Compare
| @@ -170,6 +170,11 @@ runs: | |||
| kubectl rollout status --watch --timeout 300s deployment/metrics-server -n kube-system | |||
| fi | |||
| if [[ "${{ inputs.traefik-enabled }}" == true ]]; then | |||
| # NOTE: Different versions of k3s install traefik in different ways, | |||
| # by waiting for these jobs if they exist, we will be fine no | |||
| # matter what. | |||
There was a problem hiding this comment.
Do you know which versions use which method? If so it'd be helpful to include a comment here in case someone needs to debug other problems, and also to ease cleanup when we next drop an old K3s version.
There was a problem hiding this comment.
@manics I don't, I think its different patch versions of k3s releases rather than specific k8s versions etc.
I saw k8s 1.17 and 1.19 fail but 1.18 succeed if i recall correctly etc, and some installed Traefik v1, other v2, some installed via a HelmChart resource, others didn't.
There was a problem hiding this comment.
There was no clear information when they changed HOW they install traefik, only that they install v2 in some later versions, but they started adopting a different way to install traefik in earlier versions as well so I don't see myself able to answer this so well.
|
If the |
Ah yepp! |
This action has started failing with modern k3s versions. I verified this by re-running our tests manually on current HEAD. I think it is related to how they install Traefik v2 and how they install traefik in general now - via a HelmChart CRD and a HelmChart controller that starts up Job resources to install the chart.
This is a quite simple and robust workaround I think.
Closes #32.
I have three PRs open that will cause merge conflicts with each other at the moment 😱
@manics I suggest this action plan:
What do you think?