Surface image pull errors in service status #4192
Description
In what area(s)?
/area API
What version of Knative?
HEAD
Expected Behavior
We should provide some indication that ImagePullBackoff is happening when trying to deploy a service.
Actual Behavior
In the revision, everything is Unknown because we are deploying.
status:
conditions:
- lastTransitionTime: "2019-05-29T22:48:43Z"
message: Requests to the target are being buffered as resources are provisioned.
reason: Queued
severity: Info
status: Unknown
type: Active
- lastTransitionTime: "2019-05-29T22:48:43Z"
status: "True"
type: BuildSucceeded
- lastTransitionTime: "2019-05-29T22:48:43Z"
reason: Deploying
status: Unknown
type: ContainerHealthy
- lastTransitionTime: "2019-05-29T22:48:43Z"
reason: Deploying
status: Unknown
type: Ready
- lastTransitionTime: "2019-05-29T22:48:43Z"
reason: Deploying
status: Unknown
type: ResourcesAvailableIn the service, we are just waiting for the revision to become Ready.
Interestingly, we know that the deployment is not Progressing:
deployment:
status:
conditions:
- lastTransitionTime: "2019-05-29T22:48:43Z"
lastUpdateTime: "2019-05-29T22:48:43Z"
message: Deployment does not have minimum availability.
reason: MinimumReplicasUnavailable
status: "False"
type: Available
- lastTransitionTime: "2019-05-29T22:50:44Z"
lastUpdateTime: "2019-05-29T22:50:44Z"
message: ReplicaSet "autoscale-go-d8flr-deployment-7bc5c75ff4" has timed out progressing.
reason: ProgressDeadlineExceeded
status: "False"
type: Progressing
observedGeneration: 1
replicas: 1
unavailableReplicas: 1
updatedReplicas: 1
But because the revision's Active condition is Unknown:
serving/pkg/apis/serving/v1alpha1/revision_lifecycle.go
Lines 123 to 127 in 298aeea
We assume activation is required, so we don't propagate the deployment ProgressDeadlineExceeded condition:
That seems like it might be a separate bug?
Regardless, in this case, we can check to see if the pod's user-container is in state waiting and surface an error if the deployment has also timed out with ProgressDeadlineExceeded:
pod:
status:
containerStatuses:
- image: gcr.io/jonjohnson-test/autoscale-go@sha256:e5e89c5fd57c717b49d41be89faebc526bdcda017e898ae86c2bf20f5cd339b5
imageID: ""
lastState: {}
name: user-container
ready: false
restartCount: 0
state:
waiting:
message: Back-off pulling image "gcr.io/jonjohnson-test/autoscale-go@sha256:e5e89c5fd57c717b49d41be89faebc526bdcda017e898ae86c2bf20f5cd339b5"
reason: ImagePullBackOff
hostIP: 10.128.0.50
phase: Pending
podIP: 10.60.0.9
qosClass: Burstable
startTime: "2019-05-29T22:48:43Z"Steps to Reproduce the Problem
Deploy an ksvc with a non-existent image by (valid) digest (so tag -> digest resolution will skip over it). The pod will never become ready because it can't pull the image.