Deploy etcd nodes on OpenStack Controller nodes

[jeremyeder]

Due to storage latency requirements for etcd, we found in 3.4 testing that putting etcd inside a VM backed by rotational disks (via ceph) was unstable. We moved etcd to the OpenStack controllers, which use local disk, was sufficient to stabilize etcd.

Ceph also takes this approach by colocating it's Monitor nodes on OpenStack controllers. Feedback on this colocation approach internally has been positive.

[ioggstream]

Hi @jeremyeder Can you please share more details on that?

How can I reproduce the issue and check if it affects my installation too?

Thx++, R.

[jeremyeder]

You just need to provide etcd with sufficiently performant storage. If you're affected by this, you will see errors and warnings in the etcd logs.