Open sjenning opened 5 years ago
I'm not sure what "fixed" is for this issue, but if an investigation could be done on where the activity is coming from and if it is justified, that would be good.
/assign @mrogers950
Most of all controllers should have 0 write rate on the api if no one is modifying their core objects. no one should be writing except on external edges. no one should have writes that only change a timestamp, etc.
@sjenning @smarterclayton we've got some issues with our operator sync loop causing this, that @enj and I are trying to iron out.
@ericavonb opening this to track the improvement effort
Operator is making 13-14 requests/s against the API server. 3-4 of those are PUTs i.e. mutating which can cause a lot potentially unneeded etcd and watcher activity.
query
sort_desc(sum by (client) (rate(apiserver_request_count[10m])))
query
sort_desc(sum without (instance,code,type,contentType,job,endpoint,scope,service) (rate(apiserver_request_count{client="service-serving-cert-signer/v0.0.0 (linux/amd64) kubernetes/$Format",verb=~"GET|PUT|POST|UPDATE"}[10m])))