-
Bug
-
Resolution: Unresolved
-
Undefined
-
None
-
4.18.0
-
None
-
Important
-
None
-
Rejected
-
False
-
-
Description of problem:
Telco KPI testing of rc.4+ shows that oopenshift-etcd _ etcd-<host> container count increased from 4 -> 5
Version-Release number of selected component (if applicable):
4.18.0-rc.4+
How reproducible:
100%
Steps to Reproduce:
1. Measure baseline at 4.17.0
2. Deploy 4.18.0-rc.4
3. Check container count
Actual results:
Container count is 5
Expected results:
Container count remains at 4
Additional info:
I'm trying to understand the reason for a container count increase and whether we can avoid it in a RDS deployment as we're trying to keep any changes to load or container counts to a minimum.
[OCPBUGS-49711] openshift-etcd container count increased from 4 -> 5
rhn-support-imiller there is no concerning (as in violating our 10mc/20%) cpu utilization increase related to this change, I can try to get raw numbers now. API utilization metric is something I'm still trying to make sense and there is an unaccounted increase.
tjungblu@redhat.com it was suggested in a conversation that kubeapi requests to etcd are tracked in prometheus together, I'm personally not 100% confident in this, so I'm still exploring
Is there any data on how active (cpu use) this container is? Can the container's cpu use be pulled from a test cluster?
I don't think so, this container goes straight to etcd. Check your audit logs.
tjungblu@redhat.com quick question, I've also noticed increased api/verb utilization for kubeapi server, can this be caused by said extra container in etcd?
rhn-support-imiller I don't have a per-container breakdown, but the whole pod is within 2-3% of CPU increase in worst case scenario (2 out of 7 runs), other 5 runs are within 1% difference. Do we continue as "noted" and just move on in this case? This is stats from the rc.9 builds btw.