k8s-handle fails on serialized deploy (CRD -> Object)

[ghost]

CRD defined object deploy fails even in right sequence serialized case:

1. Deploy CRD
2. Deploy custom object

config.yaml:

  - template: "zalando-postgres-operator/OperatorConfigurationCustomResourceDefinition.yaml.j2"
  - template: "zalando-postgres-operator/OperatorConfiguration.yaml.j2"

Log:

2020-01-09 07:24:52 ERROR:k8s_handle:RuntimeError: No valid plural name of resource definition discovered

[rvadim]

Hi, can you provide full example here, so we can reproduce problem?

[furiousassault]

@rvadim, I think we see how the issue can be reproduced.

The problem is, when we have a CRD and the objects of that CRD in the same deployment, it's highly likely will be failed due to some time gap between CRD accept by the Apiserver and the moment when it can handle the objects of that CRD.

Sort of my bad, when I implemented this functionality, I haven't checked and handled this case: deployed CRD and objects in different deployments and didn't want to make logic more complex.
We hadn't used deployments described by @kichay at that moment, so it wasn't a high priority issue.

To avoid that, we should implement a mechanism of CRD deployment which listens the events until the event that confirms CRD readiness appears (or timeout expires), and continue deployment or fail after that.

I can take this issue when I'll have some free time, or you can take it, if you want.