After Flagger rollback, I am not able to redeploy the same version of container image again

[praseedasathaye]

Here are the steps that followed:

1. There is version 1 of my application running which uses container image version 1.
2. Then I do canary deployment using Flagger by setting the image to version 2 and the canary analysis starts.
3. I introduce errors in my application and the canary deployment rollbacks to version 1 which is good so far.
4. Now I want to redeploy version 2, this time just allow it to go through with no error injected.
5. But when I set the container image to version 2, Flagger does not pick it up.
6. As per documentation below is mentioned:
   A canary analysis is triggered by changes in any of the following objects:
   Deployment PodSpec (container image, command, ports, env, resources, etc)
   ConfigMaps mounted as volumes or mapped to environment variables
   Secrets mounted as volumes or mapped to environment variables

[yebyen]

Please check the FAQ, how to retry a failed release:

https://docs.flagger.app/faq#how-to-retry-a-failed-release

You can set an update to an annotation so that flagger knows to retry the release. The canary has already been updated to the new image version (that update which failed, and hasn't been reverted somehow else except in Flagger), which is why setting the image version to the same value a second time has no effect. So update something else, which will trigger a new canary progressing.

[praseedasathaye]

Thank you. That helps!

[praseedasathaye]

Only adding to deployment spec and applying that, works. Below does not work.

kubectl -n mynamespace annotate deployment.v1.apps/myapp timestamp="2021-03-22T1:50"

[kingdonb]

From the FAQ, the recommended pattern is (anticipating that you're going to be using Flux, so as a patch that makes) an update to the pod template spec:

apiVersion: apps/v1
kind: Deployment
spec:
  template:
    metadata:
      annotations:
        timestamp: "2020-03-10T14:24:48+0000"

Updating an annotation on the deployment itself will not work because it does not update the pod template spec, as you correctly noted. I think that updating annotations on the pod template spec through kubectl annotate might be a bit more complicated, or even impossible.

(Edit: according to kubernetes/kubernetes#37666, it is impossible to handle this imperatively with kubectl annotate.)

[praseedasathaye]

Thanks @kingdonb for your quick response on this.