https://k8s-testgrid.appspot.com/sig-cluster-lifecycle-all#kubeadm-gce-cni-calico is and have been broken for a long time. We should fix it ASAP or garbage-collect the job.
The source PR (https://github.com/kubernetes/kubernetes-anywhere/pull/405) merged long time ago wasn't enough apparently. We need Calico folks to help maintain this stuff, otherwise we have to remove it IMO. It has been broken all the time I think :/
cc @caseydavenport @cmluciano @tmjd @ozdanborne
FYI @timothysc
luxas
All 6 comments
/cc @detiber
This is weird, did something else change, b/c we qualify every time we build our quickstart.
timothysc
on 11 May 2018
Something with the manifests in the deployment is just wrong, not Calico itself...
luxas
on 11 May 2018
I'm leaning towards garbage collecting this test, since:
- Nobody seems to want to claim ownership for the failures
- The Calico code and install manifests are already thoroughly tested through e2e jobs owned by the Tigera team, both on kubeadm and other environments.
- The Calico manifests themselves in the kubernetes-anywhere repo aren't the ones that users actually deploy.
Given the above I don't think this job adds a lot of value. I think it was originally intended as an "early warning" for when kubeadm changes broke CNI providers, but given kubernetes-anywhere doesn't really deploy what users do...
The only thing might be if we could update it to use the official manifests from docs.projectcalico.org then I think it would be worth keeping, since it could give early warning for when changes to kubeadm itself impact the Calico integration.
caseydavenport
on 12 May 2018
Okay I'm in favor of removing this then. As Tim said, Heptio also run Calico+kubeadm testing internally so if something in that integration breaks we'll know about it from you guys.
luxas
on 12 May 2018
sgtm.
timothysc
on 15 May 2018
@hh will work on this :tada:
luxas
on 28 May 2018
Related issues
chuckha
路
3Comments
mlevesquedion
路
3Comments
andersla
路
4Comments
cnmade
路
4Comments
jessfraz
路
3Comments
Most helpful comment
I'm leaning towards garbage collecting this test, since:
Given the above I don't think this job adds a lot of value. I think it was originally intended as an "early warning" for when kubeadm changes broke CNI providers, but given kubernetes-anywhere doesn't really deploy what users do...
The only thing might be if we could update it to use the official manifests from
docs.projectcalico.orgthen I think it would be worth keeping, since it could give early warning for when changes to kubeadm itself impact the Calico integration.