search

kcp-dev / helm-charts

Helm chart repo for KCP
Apache License 2.0
5 stars 21 forks source link

Update included etcd StatefulSet version #72

Closed embik closed 1 month ago

embik commented 10 months ago

The etcd StatefulSet is on a 3.5 version that is quite old. We should update it to a more recent 3.5.x patch release due to some important fixes in etcd.

However, this will require reworking the command because it uses /bin/sh, and etcd switched to distroless as base image in one of the 3.5 patch releases. /bin/sh does not exist in distroless and therefore we need to change the StatefulSet to not do any shell "scripting".

kcp-ci-bot commented 5 months ago

Issues go stale after 90d of inactivity. After a furter 30 days, they will turn rotten. Mark the issue as fresh with /remove-lifecycle stale.

If this issue is safe to close now please do so with /close.

/lifecycle stale

embik commented 5 months ago

/remove-lifecycle stale

kcp-ci-bot commented 2 months ago

Issues go stale after 90d of inactivity. After a furter 30 days, they will turn rotten. Mark the issue as fresh with /remove-lifecycle stale.

If this issue is safe to close now please do so with /close.

/lifecycle stale

embik commented 2 months ago

/remove-lifecycle stale

luxas commented 2 months ago

Do you think the solution proposed in https://github.com/kcp-dev/helm-charts/issues/94#issue-2448460334 would work, i.e. just use containerspec env instead of the CLI args?

How do I verify stuff keeps working? Are there e2e jobs in CI covering this codepath?

embik commented 2 months ago

just use containerspec env instead of the CLI args?

I think that should work. You might even get away with doing the templating of the peer list directly in --initial-cluster, but an environment variable should do the trick as well and would keep things more readable.

How do I verify stuff keeps working? Are there e2e jobs in CI covering this codepath?

There currently is a job that checks if the Kubernetes resources generated by Helm in the default configuration is valid, but no e2e test of the setup. If you install your version of the Helm chart onto a cluster and etcds start and kcp runs without crashing, things should be okay.

That being said, with this specific change I'd be curious if previous installations can be upgraded, but I'm happy to assist there and also run a little upgrade test if needed.

luxas commented 2 months ago

Thanks, I'll send a small PR for that (probably today then), and see if things keep working 👍

luxas commented 1 month ago

Sent PR now, sorry, got side-tracked with other things end of last week 😅