Cached API discovery information can lead to failures to retrieve infrastructure and bootstrap references

[ncdc]

/kind bug

What steps did you take and what happened:

1. Deploy CAPI
2. Wait for CAPI pod to be 100% up and running
3. Deploy an infra or bootstrap provider
4. Create a Cluster and e.g. an AWSCluster
5. CAPI pod fails to set the ownerReference on the AWSCluster because it doesn't know about the AWSCluster resource

What did you expect to happen:

1. It sets the ownerReference correctly

Anything else you would like to add:
In #1280 we added a RESTMapper that is able to flush its cache when it encounters an unknown kind, but it's only used by clusterctl. We need to do the same thing for our controllers.

Environment:

• Cluster-api version: v0.2.1

[ncdc]

/assign
/lifecycle active