Surface pod scheduling errors in service status

[rgregg]

In what area(s)?

When using resource requests to specify requirements that are unobtainable in the cluster, no indication of why the service is not working is available on the service object. The only visibility to an error is that ConfigurationsReady and Ready are in the Unknown status.

What version of Knative?

0.5.2

Expected Behavior

Knative would provide a failure indication with enough detail to debug what's happening from the ksvc object.

Actual Behavior

No indication that an error has occurred is visible on the ksvc.

Steps to Reproduce the Problem

Deploy a new service (say, hello world) with the following resource requests:

resources:
  limits:
    cpu: "80"
  requests:
    cpu: "80"

[rgregg]

As an example, if you create a ksvc with a bad container image, you'll get an RevisionFailed message with the error text indicating that the image couldn't be fetched (not that the text of that error is great).

[shashwathi]

/assign

[jonjohnsonjr]

I hit a similar issue with ErrImagePull. If you deploy an invalid image, you can see that there is an ErrImagePull/ImagePullBackoff status on the pods, but that doesn't get propagated up through the deployment :/

[shashwathi]

Looking at the source code it appears that specific error case i.e., container execution has been terminated with non zero code is caught but no other error/scheduling issue is bubbled up. I will work on this and see if I can make this better.