Testing pod fails

[AndreasBergmeier6176]

Expected behavior

Testing should just succeed

Actual behavior

Testing fails

Information

• Skaffold version: 2.7.0
• Operating system: Ubuntu 22.04
• Installed via: skaffold run
• Contents of skaffold.yaml:

apiVersion: skaffold/v4beta6
kind: Config
metadata:
  name: grafana
manifests:
  helm:
    releases:
      - name: grafana
        remoteChart: grafana
        namespace: monitoring
        version: 6.60.1
deploy:
  kubectl:
    defaultNamespace: monitoring

Grafana has a test pod. The pod succeeds but is then not marked as ready once successful.
Thus Skaffold prints:

pods: pods not ready: [grafana-test]

Steps to reproduce the behavior

1. skaffold run

[pydolan]

As noted by #8515 (comment), remote chart releases should be in deploy instead of manifests. With this, you can also remove the kubctl section; for example:

apiVersion: skaffold/v4beta7
kind: Config
metadata:
  name: grafana
deploy:
  helm:
    releases:
      - name: grafana
        repo: https://grafana.github.io/helm-charts
        remoteChart: grafana
        version: 6.60.1
        namespace: monitoring
        createNamespace: true

Side note to skaffold team: please add a note about this to https://skaffold.dev/docs/references/yaml/ (I also struggled with this for far too long!)

[AndreasBergmeier6176]

As noted by #8515 (comment), remote chart releases should be in deploy instead of manifests. With this, you can also remove the kubctl section; for example:

Only if you REALLY want to deploy with helm - there are enough reasons not to.
That being said - if it is supported to have Helm in manifests, then it would be good to have some kind of "compat" behavior.

[pydolan]

I think the issue is that some Charts require Helm for installing because of features like Helm Tests and Hooks. In the case of Grafana in your original config, we're probably seeing something like this:

1. skaffold runs helm template to collect all k8s manifests
2. skaffold applies these without any ordering
3. the grafana-test pod runs and fails because the main graphana pods/etc were still building/deploying

When using the deploy section, helm install is used by skaffold, and we see the grafana-test pod waits until everything is finished deploying