For my use case, I would like to avoid multiple Elasticsearch clusters performing rolling updates at the same time as the result of an ECK operator upgrade.
It seems that the restricted installation of ECK facilitates this when independent ECK operators manage separate Elasticsearch clusters.
However, there remains a coupling through CRD definitions.
Assuming my preamble is correct, I hade two questions (for a single GKE cluster):
- Can updating the CRDs (helm release) result in a rollout restart of the Elasticsearch clusters? My expectation is that the CRD releases remain backwards compatible and thereby ensure that upgrading does not affect any actively running clusters.
- Once CRDs (helm release) have been upgraded, what are the compatibility guarantees with operators of an older version (helm release)? For instance, can I upgrade the CRD helm release and upgrade an ECK operator helm release but leave another ECK operator on its former version (or is that unsafe)?