Expose last warnings from eventlog for Pod+PVC+SVC

[radeksimko]

Examples

Service

* kubernetes_service.test: Waiting for service "default/radek-blablah" to assign IP/hostname for a load balancer
   * CreatingLoadBalancerFailed: Error creating load balancer (will retry): Failed to create load balancer for service default/radek-blablah: requested ip 10.0.0.1 is neither static nor assigned to LB a049988bc615511e781c642010a8a005(default/radek-blablah): <nil>

PVC

* kubernetes_persistent_volume_claim.example: timeout while waiting for state to become 'Bound' (last state: 'Pending', timeout: 5m0s)
   * ProvisioningFailed: Failed to provision volume with StorageClass "standard": claim.Spec.Selector is not supported for dynamic provisioning on GCE

Pod

* kubernetes_pod.test: timeout while waiting for state to become 'Running' (last state: 'Pending', timeout: 5m0s)
   * FailedSync: Error syncing pod, skipping: failed to "StartContainer" for "example" with ImagePullBackOff: "Back-off pulling image \"nginx:notexists\""

   * Failed: Failed to pull image "nginx:notexists": rpc error: code = 2 desc = Tag notexists not found in repository docker.io/library/nginx
   * FailedSync: Error syncing pod, skipping: failed to "StartContainer" for "example" with ErrImagePull: "rpc error: code = 2 desc = Tag notexists not found in repository docker.io/library/nginx"

[tombuildsstuff]

otherwise LGTM :)

[tombuildsstuff]

given it's being used in multiple resources, any reason to not pull this out into a separate method?

[radeksimko]

As discussed, we can do this later if the error handling sprawls through more resources. I think there's still benefit in not decoupling it as we're making an API call here and it should be more obvious from these extra LOC that something "heavy-weighted" is happening here. 🙂

[tombuildsstuff]

I'd argue this would read cleaner if only returning the combined err + warns object within this if statement?

[radeksimko]

Sure, I'll change that. 👍