Michael McCune comments

Results 369 comments of


                                            Michael McCune

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

not sure where this is at, but i do think it's important /remove-lifecycle stale

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

thank you for the comment and link @songminglong !

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

i believe this is still happening, but i have not had time to investigate further. /remove-lifecycle stale

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

@mattpopa thanks for the extra details, i'm curious which provider you are using? (is this problem more than just the clusterapi provider)

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

> it used to work ok before using mixed policy instances on the AWS ASG (used to only use spot for CI purposes, but due to increasing AWS spot capacity...

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

i think this is worth keeping open, but i have a feeling that the solution is going to be complicated. if i understand correctly, what we need to do is...

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

i'm not sure that we have anyone to champion this forward. it seems like an issue worth investigating further, but i do not have the time currently.

clusterapi: Cluster Autoscaler destroys healthy nodes when processing old unregistered nodes when multiple machinesets exist

/remove-lifecycle rotten i have a feeling this is directly related to #8494

[Flaking Test] Restart [Disruptive] [KubeUp] should restart all nodes and ensure all nodes and pods recover

we are discussing this at the sig cloud provider office hours and while it is a gcp cloud provider issue, we think that we might need involvement from sig node...

SIG Cloud Provider update

adding some links to help with info needed by https://github.com/kubernetes/community/pull/7815