Description of problem:
After upgrade from 3.10 to 3.11, found that infrastructure pods have no expected priorityclass.
Means should follow this PR or later PR: https://github.com/openshift/openshift-ansible/pull/9166
Version-Release number of the following components:
rpm -q openshift-ansible
rpm -q ansible
Steps to Reproduce:
1. Setup 3.10 cluster
2. Upgrade to 3.11 version
3. Check the priorityclass of infrastructure pods
oc get pods --all-namespaces -o yaml |grep -i priority
Please include the entire output from the last TASK line through the end of output if an error is generated
3. Some infrastructure pods are not system-cluster/node-critical priorityclass after upgrade
3. All infrastructure pods system-cluster/node-critical priorityclass after upgrade
Please attach logs from ansible-playbook with the -vvv flag
And also this PR: "Control plane static pods (apiserver, etcd, controller-manager) must be assigned highest priority class system-node-critical."[https://github.com/openshift/openshift-ansible/pull/9801]
(In reply to weiwei jiang from comment #1)
> And also this PR: "Control plane static pods (apiserver, etcd,
> controller-manager) must be assigned highest priority class
Weiwei, did you test with the above PR? My guess is that above PR should be enough. I am also going to try upgrade by myself now as I am not very clear how upgrade from 3.10 to 3.11 is working exactly.
I think i misread the issue. So what pods are not getting right priority? could you list them?
I can reproduce that there are several pods in various namespaces like openshift-monitoring, openshift-console and openshift-web-console are not assigned right priority class. I am working on a PR that I will send after some testing. There are some operators (monitoring and prometheus) also involved so need to check what pods they are starting and how they pass configuration to their pods.
Here are PRs:
When above PRs are merged, there will be more follow-up PRs too.
All pods except Prometheus and Alertmanager have the priority class set now, Jessica and I decided this is not a release blocker and therefore moving to 3.11.z.
To fully resolve this, we need a newer version of the Prometheus Operator that introduces a lot of changes compared to what we're shipping in 3.11 right now, so we're only fixing this in 4.1.
The PR to fix the final pieces in 4.1: https://github.com/openshift/cluster-monitoring-operator/pull/311
https://github.com/openshift/cluster-monitoring-operator/pull/311 just landed so all monitoring components should now have the appropriate priority class.
There is not available payload which packages the fix to test, so postpone the testing until we have available payload
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.