wait until deploy completes

[jrynyt]

Ideally kubectl apply would have something like a --wait option, so you know the actual rollout of Pods (etc) completed successfully, but it doesn't. So the success of a deploy here doesn't actually indicate the deploy succeeded, as the rollout happens asynchronously.

We could use kubectl rollout status after the kubectl apply, which blocks on deployment completion. It looks like you can even use -f to track all the resources in the config file.

[tonglil]

This should be satisfied by #24 as you can run that command after apply, as well as run additional commands like kubectl get ....

[tonglil]

It might also be a good idea to re-open #25 to use the -R option (and watch all resources in the folder).

[balser]

I understand #24 is very flexible, but if there are a some common workflows/tasks related to the deploy, it might be better to be implemented once by the plugin vs implemented by each app owner.

[jrynyt]

I would consider this a bug, not an optional thing that devs might choose to do. The drone deploy should succeed/fail based on the success/failure of the k8s deploy.

Is #25 necessary for this to work? i.e. it would run in a separate container than the apply?

[tonglil]

I can see this being a separate feature, will mark it as so.

It isn't a bug because this isn't a feature of kubectl apply, and is part of a larger feature request for multiple resource types to signal "ready".

The plugin will currently fail when the resource config spec fails, the same conditions as doing it via the command line.

[jrynyt]

I don't agree: the drone-gke plugin claims to "deploy container images to Kubernetes on Google Container Engine" -- not "run kubectl apply, which starts the deploy but doesn't wait for it to succeed or fail." kubectl apply isn't even mentioned in the doc, so a successful plugin execution but failed deploy looks like a bug.

See the conversation in #gke from 2/24.

"I know I’ve personally misconfigured my own app and only found out because I checked the logs."

[tonglil]

The rollout command does not work for any other Kubernetes resources specified in the .kube.yml, which is why I am hesitant to add it as a parameter of the plugin since the template can contain any K8s resource.

The behavior that person described is the same as when using kubectl from the command line. Kubernetes is based on the model of convergence, and in certain scenarios things will never converge (out of resources, errors, resource limitations external to the cluster, etc).

I do not see the intersection between the plugin's claim and the behavior of Kubernetes. Maybe it should be "configure Kubernetes resources on Google Container Engine" instead to clarify?

[jrynyt]

I'll test how rollout behaves when other resources are in the kube files. I assumed it would know what to do (ignore) since it has -f and -R. I'm happy to do this work so I can learn more about drone and go.

The plugin does not say anywhere that it is a wrapper for kubectl, so I'm not sure how a user would intuit this behavior. My point is that if convergence fails, the deployment tool should not indicate a successful deploy.