Log pods for which we deployed the owner

[dgageot]

This PR is an experiment. Many things are broken.

This is how application logs work in Skaffold:

• A logger watches for all the pod event in a set of namespaces
• For each pod event, we look for the images used in the pod.
• If any container in that pod uses an image that Skaffold built, we output the logs for that container

This is tricky because:

• We might end up logging a pod that was deployed beforehand and uses the same image tags.
• We miss the pods that Skaffold deployed but that use pre-built images

We tried to rely on the runID label to solve that issue but it also has issues:

• The runID mechanism break idempotency: a skaffold run just after a skaffold run should do nothing. With the runID, every object is updated.
• If we switch from a runID label to a runID annotation, it will not cascade down to the pods.

This PR introduces a new mechanism:

• kubectl/kustomize deployer, returns the list of resources it applied, with their uid. We keep that.
• helm deployer used to add a label to the resources it deployed. Instead, we'll collect all the uids.
• the pod watcher, when it sees a pod event looks for the top level parent and checks that this is in the set of uids we deployed.

Signed-off-by: David Gageot [email protected]

[gsquared94]

Since this loops []Artifacts which only contains skaffold-built images, this still won't select containers for pre-built images which belong to a different Deployment altogether; it only improves upon init containers and multi-container pods, right?

[dgageot]

exactly. Not yet.