search

vexxhost / magnum-cluster-api

Cluster API driver for OpenStack Magnum
Apache License 2.0
41 stars 17 forks source link

fix: stop unexpected rollouts #324

Closed mnaser closed 3 months ago

mnaser commented 3 months ago

When making any node group changes, we were previously making an entire change of all the cluster resources. With this change, we're instead simply making the small changes in the node group which should avoid unexpected full rollouts of node groups when changing min/max for autoscaling or resizing a cluster.

okozachenko1203 commented 3 months ago

As long as the coe cluster upgrade with the same coe clustertemplate is supported, I think there is no downside in the approach that will trigger capi clusterclass update only when coe cluster upgrade requested. If coe cluster upgrade requires a different coe clustertemplate from the current one, users have to create a twin coe cluster template for clusterclass update. If so, we will have more coe clustertemplates, and will end up with having the same number of coe clsutertemplates as the number of coe clusters in the worst case even though we delete the old coe clustertemplates which no coe cluster refers to. But fortunately, magnum api allows cluster upgrade with the current coe clustertemplate like heat does with --existing flag. So we can do capi clusterclass update without k8s version change or even without any changes on openstack coe cluster resource in this way.

mnaser commented 3 months ago

Nice, I didn't know that @okozachenko1203 so this is the best of both worlds.

I will also try and document this a little bit deeper as well in the documentation as well.