Hi, We found that jobmanager annotations defined by pod template does not work. However, annotations defined by kubernetes.jobmanager.annotations [0] This behavior is different from document [1]. [0] https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/config/#kubernetes-jobmanager-annotations [1] https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/resource-providers/native_kubernetes/#pod-template -- ChangZhuo Chen (陳昌倬) czchen@{czchen,debian}.org http://czchen.info/ Key fingerprint = BA04 346D C2E1 FE63 C790 8793 CC65 B0CD EC27 5D5B signature.asc (849 bytes) Download Attachment |
Could you please share your pod template and the value of kubernetes.jobmanager.annotations? Usually the annotations of pod template and flink config options will be merged. And the flink config options has higher priority if you are specifying same name annotation. I have verified in minikube and it could take effect as expected. Best, Yang ChangZhuo Chen (陳昌倬) <[hidden email]> 于2021年6月11日周五 下午4:52写道:
|
On Fri, Jun 11, 2021 at 11:19:09PM +0800, Yang Wang wrote:
> Could you please share your pod template and the value of > kubernetes.jobmanager.annotations? > > Usually the annotations of pod template and flink config options will be > merged. And the flink config > options has higher priority if you are specifying same name annotation. > > I have verified in minikube and it could take effect as expected. Hi, There are other finding for this issue: * For jobanager: * annotations, and labels in pod template do not work. * annotations, and labels in -Dkubernetes.jobmanager.* work. * For taskmanager: * annotations, and labels in pod template work The following is jobmanager pod template: apiVersion: batch/v1 kind: Pod metadata: labels: app: jobmanager helm.sh/chart: <censored> app.kubernetes.io/name: <censored> app.kubernetes.io/instance: <censored> app.kubernetes.io/version: <censored> app.kubernetes.io/managed-by: Helm annotations: rt.prometheus.io/scrape: 'true' rt.prometheus.io/path: '/' rt.prometheus.io/port: '9249' The following is pod created as jobmanager: Name: <censored> Namespace: <censored> Priority: 2000000 Priority Class Name: medium Node: <censored> Start Time: Fri, 11 Jun 2021 23:38:52 +0800 Labels: app=<censored> component=jobmanager pod-template-hash=55846fd8f7 type=flink-native-kubernetes Annotations: <none> -- ChangZhuo Chen (陳昌倬) czchen@{czchen,debian}.org http://czchen.info/ Key fingerprint = BA04 346D C2E1 FE63 C790 8793 CC65 B0CD EC27 5D5B signature.asc (849 bytes) Download Attachment |
The annotations, and labels in the pod template will only apply to the JobManager pod, not the JobManager deployment. Best, Yang ChangZhuo Chen (陳昌倬) <[hidden email]> 于2021年6月11日周五 下午11:44写道: On Fri, Jun 11, 2021 at 11:19:09PM +0800, Yang Wang wrote: |
On Tue, Jun 15, 2021 at 04:22:07PM +0800, Yang Wang wrote:
> The annotations, and labels in the pod template will only apply to the > JobManager pod, not the JobManager deployment. Thanks for the information. Is this behavior by design? In document, it looks like there is no different between jobmanager and taskmanager when handling annotations and labels [0]. [0] https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/deployment/resource-providers/native_kubernetes/#pod-template -- ChangZhuo Chen (陳昌倬) czchen@{czchen,debian}.org http://czchen.info/ Key fingerprint = BA04 346D C2E1 FE63 C790 8793 CC65 B0CD EC27 5D5B signature.asc (849 bytes) Download Attachment |
Yes. It is the by-design behavior. Because the pod template is only applicable to the "pod", not other resources(e.g. deployment, configmap). Currently, the JobManager pod is managed by deployment and the naked TaskManager pods are managed by Flink ResourceManager. This is the root cause which makes the difference. Best, Yang ChangZhuo Chen (陳昌倬) <[hidden email]> 于2021年6月15日周二 下午4:27写道: On Tue, Jun 15, 2021 at 04:22:07PM +0800, Yang Wang wrote: |
On Tue, Jun 15, 2021 at 04:40:00PM +0800, Yang Wang wrote:
> Yes. It is the by-design behavior. Because the pod template is only > applicable to the "pod", not other resources(e.g. deployment, configmap). > > Currently, the JobManager pod is managed by deployment and the naked > TaskManager pods are managed by Flink ResourceManager. > This is the root cause which makes the difference. Thanks for the clarification. -- ChangZhuo Chen (陳昌倬) czchen@{czchen,debian}.org http://czchen.info/ Key fingerprint = BA04 346D C2E1 FE63 C790 8793 CC65 B0CD EC27 5D5B signature.asc (849 bytes) Download Attachment |
Free forum by Nabble | Edit this page |