How deleting a deployment in fabric8 kubernetes-client works?

[wangyang0918]

In Flink project, we are using the following code to delete a deployment. We start a Flink master deployment in the K8s cluster and it will call the stopAndCleanupCluster when all jobs finished. However, we find that it could not work and have a high probability the deployment could not be deleted, instead the replication has decreased to 0.

So when delete a deployment via fabric8 kubernetes-client, does it decrease the replication to 0 first? And then delete the deployment and replicationset. Right?

I really appreciate for your help and any suggestion.

	public void stopAndCleanupCluster(String clusterId) {
		this.internalClient
			.apps()
			.deployments()
			.inNamespace(this.nameSpace)
			.withName(KubernetesUtils.getDeploymentName(clusterId))
			.cascading(true)
			.delete();
	}

[rohanKanojia]

cacading() has been marked as deprecated. Please use withPropagationPolicy() instead

[wangyang0918]

@rohanKanojia Thanks a lot for your quick response. Do you mean withPropagationPolicy set to Foreground could solve this problem?

If the deleting deployment is not a one-step operation, then it might be not be executed completely, especially we do this in the deployment itself.

[rohanKanojia]

I think cascading(true) should also be able to delete Deployment and its dependents. It sets orphanDependents fields to false in DeleteOptions.

But I would suggest you should set using withPropagationPolicy set to Foreground or Background for deletion operations

[wangyang0918]

@rohanKanojia So do you mean the deleting deployment only send one http request to K8s apiserver? If not, it might not be executed completely.

[rohanKanojia]

Yes, deletion is just an API request to Kubernetes APIServer

[wangyang0918]

Thanks a lot. It is really strange that the deployment is not deleted. And the replication has been set to 0 instead.

[rohanKanojia]

Which version are you using?

[zhengcanbin]

Yes, deletion is just an API request to Kubernetes APIServer

The Flink community is using V4.5.2, that is not a one step request

[rohanKanojia]

Umm, that's pretty old version. Could you please try on v4.9.0 or v4.9.1 instead?

[rohanKanojia]

@wangyang0918: I'm not sure if LeaderElection support is there in 4.5.2. I remember you once requested it and our team added it. Are you not using it?

[zhengcanbin]

Umm, that's pretty old version. Could you please try on v4.9.0 or v4.9.1 instead?

@rohanKanojia I notice the latest version is v4.10.1, is it safe that we use it in production?

[rohanKanojia]

ah, We recently split the Kubernetes Model into smaller jars per apiVersion in #2137 . But it has introduced some regressions #2195, #2201 and #2205

If you're not affected by these bugs, you can upgrade. We have plans on fixing them and then cutting a stable release.

[zhengcanbin]

ah, We recently split the Kubernetes Model into smaller jars per apiVersion in #2137 . But it has introduced some regressions #2195, #2201 and #2205

If you're not affected by these bugs, you can upgrade. We have plans on fixing them and then cutting a stable release.

Got it. Thank you @rohanKanojia!

[wangyang0918]

Bumping the version to 4.9.1 could fix our problem. @rohanKanojia Thanks a lot for you help.