Running ETCD in k8s without persistent storage?

[Richard Cunningham]

Hi,

We have started doing some testing with the bitnami helm chart for etcd.  We are effectively using etcd for a data cache so, for the data that we are storing in etcd we don't necessarily need high data durability but we do need the cluster to be highly available so multiple nodes in a cluster.  Our k8s cluster does not provide Persistent Volume Claims so we wanted to try and run it with `emptyDir`.

However we are finding the simple things like pod restart is causing the cluster to break with a cluster of 3 pods.

Is it possible to run a HA cluster without a persistent storage layer and rely solely on network replication for pod recovery?

In the event of total cluster lost it would be acceptable to just redeploy the cluster but for basic maintenance operations we would need etcd to remain stable.