error: group map[...] is already registered

[janwillies]

When trying to install a chart, tiller dies with the following error:

$ helm install --namespace=infra infra/charts/ingress/ --name nginx-ingress

Tiller is listening on :44134
Probes server is listening on :44135
Storage driver is ConfigMap
2016-11-25 13:04:46.387816 I | storage.go:94: Listing all releases with filter
2016-11-25 13:04:54.636425 I | storage.go:133: Getting release history for 'nginx-ingress'
2016-11-25 13:04:54.646510 I | release_server.go:862: Executing pre-install hooks for nginx-ingress
2016-11-25 13:04:54.646533 I | release_server.go:889: Hooks complete for pre-install nginx-ingress
2016-11-25 13:04:54.646539 I | storage.go:133: Getting release history for 'nginx-ingress'
error: group map[certificates.k8s.io:0xc420448380 rbac.authorization.k8s.io:0xc4203aa150 storage.k8s.io:0xc4203aa1c0 federation:0xc4203aa380 apps:0xc4203abf10 autoscaling:0xc420448150 batch:0xc420448310 componentconfig:0xc4204483f0 extensions:0xc4203aa070 policy:0xc4203aa0e0 :0xc4203abea0 authentication.k8s.io:0xc4203abf80 authorization.k8s.io:0xc4204480e0] is already registered

helm ls works, but returns nothing obviously. helm version also works.

I'm running kubernetes-1.5.0-beta.2 in ABAC mode. The kube-system namespace has the following permissions:

{"apiVersion": "abac.authorization.kubernetes.io/v1beta1", "kind": "Policy", "spec": {"user":"system:serviceaccount:kube-system:default","namespace":"*","resource":"*","apiGroup":"*"}}

[intlabs]

I'm also seeing the same issue with kubernetes-1.5.0-beta.1

[technosophos]

@intlabs are you also using ABAC?

[janwillies]

I've been using the same ABAC setup with kube-1.3 and tiller worked fine. Just not on 1.5

[intlabs]

@technosophos, actually no sorry - I should have been clearer. We are deploying kube via kubeadm with the standard configuration.

It would be great to get to to the bottom of this, though, as we are building the development environment for Kolla-Kubernetes's Helm implementation and we need to use k8s 1.5 for dynamic Ceph PVC's and StatefullSet support. As a consequence, we are currently having to work around this by deploying k8s 1.4.6 and then bringing in the 1.5 controller-manager to provide a short-term fix, though this is obviously far from ideal.

Cheers

Pete

[janwillies]

@technosophos any pointers where to start looking?

[v1k0d3n]

@technosophos this relates to the issue i brought up over slack. if it helps, the repo we've been using (that exposed this issue for us) is https://github.com/att-comdev/halcyon-vagrant-kubernetes.

you can use that as a testbed, since all the versioning is optionally configurable. i know you mentioned that you couldn't reproduce...maybe this would help?

[technosophos]

I think we're going to have to test out switching the Kubernetes library we use from 1.4.x o 1.5.0-beta and try it. We seem to be hitting some sort of version mismatch.

[enisoc]

This looks similar to kubernetes/kubernetes#35791.

[technosophos]

We are now working through several issues with the K8s 1.5 release, but I think that switch will fix this error.

[intlabs]

@technosophos, brilliant - thanks for keeping us updated with your progress :)

[adamreese]

1.5 support in now merged