After deploying a Kubernetes 1.5 cluster at SDSC, we noticed that the addons were not all running as they should be. Performing a kubectl create manually on the .yml file showed the services would run properly for several minutes before terminating themselves for seemingly no reason.

Step to recreate:

1. Deploy a Kubernetes 1.5 cluster via deploy-tools
2. Create grafana via kubectl

   kubectl create -f /tmp/grafana.yml

3. Watch kube get pods for changes (~5 min)

   watch kubectl get pods

   • You should see the grafana / heapster / influx pods all go into a "Terminating" state.

This may be related to the addons and their associated update script, but more investigation will certainly be needed.

This ticket is complete when the addons (grafana and kibana, at minimum) continue to run for longer than 10 minutes on a multi-node Kubernetes 1.5 cluster without terminating themselves.