Error reconciling ClusterChannelProvisioner when dispatcher service is updated

[neosab]

Expected Behavior

Channel controllers (Kafka & in-memory) reconcile ClusterChannelProvisioner when the dispatcher service (owned by them) is updated.

Actual Behavior

They run into an error Could not find ClusterChannelProvisioner during reconciliation

in-memory-controller logs:

{"level":"info","ts":"2018-11-30T00:50:33.704Z","logger":"provisioner","caller":"clusterchannelprovisioner/reconcile.go:70","msg":"Could not find ClusterChannelProvisioner","eventing.knative.dev/clusterChannelProvisioner":"in-memory-channel","eventing.knative.dev/clusterChannelProvisionerComponent":"Controller","controller":"in-memory-channel-controller","request":"knative-eventing/in-memory-channel","error":"ClusterChannelProvisioner.eventing.knative.dev \"in-memory-channel\" not found"}

kafka controller logs

{"level":"info","ts":"2018-11-29T19:06:41.687Z","logger":"provisioner","caller":"controller/reconcile.go:44","msg":"reconciling ClusterChannelProvisioner","request":"knative-eventing/kafka"}
{"level":"info","ts":"2018-11-29T19:06:41.687Z","logger":"provisioner","caller":"controller/reconcile.go:49","msg":"could not find ClusterChannelProvisioner","request":"knative-eventing/kafka"}

Steps to Reproduce the Problem

1. Start the in-memory or kafka controllers
2. Manually update the dispatcher svc spec say kafka-dispatcher or in-memory-channel-dispatcher
3. Check the controlller logs. The reconcilliation would have failed.

Note: You can also notice this error when you just start the controller as a we receive two reconciliation request. First one is for the watched CCP object and that succeeds. The second one is triggered because of the owned dispatcher svc and this one would fail.

Additional Info

The root cause is the behavior with controller runtime'sEnqueueRequestForOwner. The reconciliation request triggered because of a change to the owned object includes the namespace of the owned object i.e the namespace of the dispatcher service. So we get "request":"knative-eventing/kafka" instead of "request":"/kafka". The same for in-memory channel controller.

[grantr]

Thanks for debugging this @neosab! Seems like a bug in controller-runtime.

[grantr]

Here's an issue describing the CR bug: kubernetes-sigs/controller-runtime#214

[neosab]

Ah, I did see that issue in controller-runtime but I somehow thought it was closed as a non-issue. Good to know.