Hi, At this version, master restarts shouldn't be quite as frequent (especially due to autoscaling changes), although they may still happen when clusters size exceeds some threshold.
It's hard to guess what may be causing it in this case without having a look at the cluster - can you open the ticket with GCP support, providing cluster details? Thanks, Aleksandra On Friday, September 7, 2018 at 6:49:57 PM UTC+2, Ludovic Havet wrote: > > Hi, > > Actually it happens on all versions since we started developement. > > We are currently on 1.10.6-gke.1, but it has always been like that. > > I guess we can see easily in the system logs if kubernetes is being > restarted? > > Thanks > Ludovic > > Le vendredi 7 septembre 2018 13:47:47 UTC+2, Aleksandra Malinowska a > écrit : >> >> Hi, >> >> What is the Kubernetes master version of the affected cluster? Does this >> happen every time when adding a new node, or just for a first node in the >> new node pool? >> >> One reason Kubernetes API server may be temporarily unavailable is if >> it's being restarted (this can happen e.g. due to configuration update, >> auto-upgrade or resizing). Exact behavior depends on version and cluster >> size (number of nodes), so it's hard to pinpoint the reason without this >> information. >> >> Thanks, >> Aleksandra >> >> On Thursday, September 6, 2018 at 5:30:17 PM UTC+2, Ludovic Havet wrote: >>> >>> Here is more details on the problem. >>> >>> The idea behind it is to manually manage pods into the not default >>> nodepool. >>> >>> The nodepool is created in autoscaling. It is not the default nodepool >>> but an additional one created manually through gcloud api. >>> >>> When a pod is created in the nodepool and that there is no remaining >>> ressource, autoscaling is acting and a new node is being created. >>> >>> The problem is that the whole cluster hangs and its api is not >>> accessible as long as the new node is being created for the nodepool. This >>> means during about 1-2 minutes. >>> >>> It is as if kubernetes was not responding on the api, during one of its >>> not default nodepool autoscaling. >>> >>> This does not happen when autoscaling is happening on the default >>> nodepool. >>> >>> I think it is an issue with GKE. >>> >>> Any help would be appreciated :) >>> >>> Thanks! >>> >>> >>> >>> -- You received this message because you are subscribed to the Google Groups "Kubernetes user discussion and Q&A" group. To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-users+unsubscr...@googlegroups.com. To post to this group, send email to kubernetes-users@googlegroups.com. Visit this group at https://groups.google.com/group/kubernetes-users. For more options, visit https://groups.google.com/d/optout.
