regression on k8s v1.5.1

[hongchaodeng]

Today I upgraded cluster to v1.5.1 and master e2e test failed:

https://jenkins-etcd-public.prod.coreos.systems/job/etcd-operator-master/173/console

[hongchaodeng]

logs:

20:22:48 --- FAIL: TestCreateSelfHostedClusterWithBootMember (120.45s)
20:22:48 	self_hosted_test.go:80: etcdserver is ready
20:22:48 	self_hosted_test.go:112: failed to create 3 members etcd cluster: timed out waiting for the condition
...
20:20:48 --- FAIL: TestCreateSelfHostedCluster (240.03s)
20:20:48 	self_hosted_test.go:48: failed to create 3 members self-hosted etcd cluster: timed out waiting for the condition

[hongchaodeng]

Seems like defaulting has regression:

20:20:48 time="2016-12-22T20:20:30Z" level=error msg="fail to reconcile: Pod \"test-etcd-3x3vp-0044\" is invalid: spec.initContainers[0].env[0].valueFrom.fieldRef.apiVersion: Required value" cluster-name=test-etcd-3x3vp pkg=cluster 

[hongchaodeng]

Another problem, namespace getting stuck:

[hongchaodeng]

jenkins cluster downgraded to 1.4.7

[hongchaodeng]

Defaulting issue tracked on upstream: kubernetes/kubernetes#39189

[hongchaodeng]

Just tried on latest master (v1.5.2-beta). Another issue:

$ kubectl logs -f etcd-operator                
E0104 03:16:45.980462       5 leaderelection.go:252] error retrieving endpoint: the server does not allow access to the requested resource (get endpoints etcd-operator)
...

[hongchaodeng]

another regression on GC unable to recognize kind in REST mapping: kubernetes/kubernetes#39816

[xiang90]

@hongchaodeng Are these all "resolved" on upstream?

[xiang90]

If all of them are tracked on upstream, and we have work around in our code, we suggest to close this umbrella issue.

[hongchaodeng]

Right.

• defaulting has been worked around
• GC is worked around on upstream
• RBAC is tracked on another issue