In order to test migrating the wikikube control planes to hardware nodes and colocate with etcd we should migrate the wikikube clusters from the current setup with 2 control planes + 3 etcd VMs to 3 VMs that co-locate kube-apiserver and etcd.
This will be useful for testing new puppet roles as well as the procedure and will ensure that staging and prod clusters are set up the same way as prod (which is required so we can catch errors early during future k8s upgrades etc.).
- Request a new VM (4 CPU/5 GB) in codfw (currently etcd runs on 2 CPU/3 GB and apiserver on 1 CPU/3 GB) without DRBD
- Extend puppet to allow stacked installs
- Extend prometheus puppet to automatically scrape all etcd instances
- Write down what needs to be done to migrate a cluster to stacked masters https://wikitech.wikimedia.org/wiki/Kubernetes/Clusters/Add_or_remove_control-planes
- Add two additional new stacked masters to staging-codfw
- Remove the old control-planes and etcd nodes (delete VMs etc)
- Repeat the process for staging-eqiad