[BUG] Failed to start kruise-manager when CR has been created before webhook injection

[ntboot]

What happened:

Error deploying kruise-controller-manager

Environment:

• Kruise version:
  

[FillZpp]

Em...Let me guess your operations:

1. First, you applied the CRDs into your cluster.
2. Then, you created a StatefulSet.apps.kruise.io resource.
3. After that, you tried to install the kruise-controller-manager, and unfortunately you got an error.

Am i right?

[ntboot]

1. extend CRD
2. then deploy kruise-manager

Reference document: https://github.com/openkruise/kruise/blob/master/docs/tutorial/kruise-install.md

[FillZpp]

@ntboot Could you show me kubectl get statefulset.apps.kruise.io

[ntboot]

I don't understand why kruise-controller-manager will miss restart once my guest book is installed.

[FillZpp]

@jian-he @resouer I think this is a bug of guestbook-kruise chart.

We can not put both kruise-manager and StatefulSet.apps.kruise.io resources in chart, because the manager must be started before users create the AdvancedStatefulSet. But we have no way to define the order that yamls in chart which should be applied first and others shouldn't be applied util the first one has been ready.

[ntboot]

I know what you mean, you have to deploy kruise-controller-manager correctly first, and then you can't deploy the related resources of astatefulset controller correctly. But I also follow the document step by step, but still encountered the above problems.

[ntboot]

I like this project very much, and I hope that the elastic expansion behind you will open up as soon as possible. So now I want to solve this problem

[resouer]

@FillZpp If I'm understanding it correctly, you are suggesting we should very kruise-manager is running before deploy any applications?

We can not put both kruise-manager and StatefulSet.apps.kruise.io resources in chart

kruise-manager is not part of guestbook-kruise charts, it's started separately by following https://github.com/openkruise/kruise/blob/master/docs/tutorial/kruise-install.md

[FillZpp]

@resouer Got it.

@ntboot So i think the problem is, after kruise-manager installed, you went to install guestbook chart immediately, but the kruise-manager had not been ready yet.

This could be a regular problem, i have to think of how to avoid it.

[ntboot]

I guess so. I did a test waiting for kruise-manger to start for 8 minutes. Then I started an astatefuleset controller, and finally kruise-manger was restart.

[ntboot]

@FillZpp

[FillZpp]

#50 will fix this problem.

After deployed kruise into a cluster, users can not create resources defined in kruise util kurise-manager has been started.

[ntboot]

I changed a new cluster operation and succeeded! Thank you

However, the cluster that had problems before cleaned up the relevant resources, and then try again is not good. My understanding is that the order of deleting resources should be:

1. sts. apps. kruise. IO

2. kruise-manger

Am I right? @FillZpp

[FillZpp]

@ntboot Delete all sts.apps.kruise.io resources, then delete the pod kubectl delete pod -n kruise-system kruise-controller-manager-0.
After that, the StatefulSet named kruise-controller-manager will create a new Pod which should be Ready.

[ntboot]

got it