Hello!

I'm seeing an unexpected label value assignment happening, and I'm not sure
how it's happening.  It is possible it is in my own helm charts and
templates somewhere, but I'm not seeing it, so I'm beginning to think this
is happening in the FlinkDeployment CRD in the operator code somewhere.

I'm using FlinkDeployment podTemplate
<https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-main/docs/custom-resource/pod-template/>
to add an 'app' label:

 podTemplate:
    apiVersion: v1
    kind: Pod
    metadata:
      labels:
        app: flink-app
        release: flink-example
        ...

I also have this app label set in the FlinkDeployment labels:

kind: FlinkDeployment
metadata:
  name: flink-app-flink-example
  labels:
    app: flink-app
    chart: flink-app-0.1.1
    release: flink-example

Since I've set app: flink-app in the podTemplate, I would expect all pods
to get this label.  The FlinkDeployment resource has this label value as
expected.  However, I see that in the pods, as well as the Deployment that
are created by FlinkDeployment:

*$ kubectl -n flink-app0 describe deployments flink-app-flink-example*
...
Name:                   flink-app-flink-example
Namespace:              flink-app0
CreationTimestamp:      Thu, 19 Jan 2023 12:42:05 -0500
Labels:                 app=flink-app-flink-example
                        component=jobmanager
...

Pod Template:
  Labels:           app=flink-app-flink-example
                    component=jobmanager
                    release=flink-example
                    ...


*$ kubectl -n flink-app0 describe pod
flink-app-flink-example-d974cb595-788ch*
...
Labels:       app=flink-app-flink-example
              component=jobmanager
              pod-template-hash=d974cb595
              release=flink-example
...


I'd expect the app label to be 'flink-app' for at least the Deployment
PodTemplate and the Pod, if not the Deployment itself too.

Something is overriding the app label in podTemplate, and I don't think
it's my chart or installation.  I looked in flink-kubernetes-operator code
and I didn't find where this was happening either.  I am not setting e.g.
kubernetes.jobmanager.labels
<https://nightlies.apache.org/flink/flink-docs-release-1.15/docs/deployment/config/#kubernetes-jobmanager-labels>
.

Is this expected?

Thank you!

-Andrew Otto
 Wikimedia Foundation

Reply via email to