Issues #581 and #582 call our use of etcd into question, if it's still in use in some form, we should reconsider the value of the etcd operator.
It's:
Complicated internally, which is a bad sign.
Prone to mysterious behaviour, like when the production cluster lots all its etcd nodes (somehow) and the operator wouldn't start new ones until we deleted and re-created the EtcdCluster object. All our etcd data got lost in the process.
Really slow to start (it's a big delay in k8s test cluster setup) for some reason.
So perhaps we can just embed etcd in the dotmesh server, or start a cluster up from the dotmesh operator, or something else. Assuming we still use etcd at all.
Issues #581 and #582 call our use of etcd into question, if it's still in use in some form, we should reconsider the value of the etcd operator.
It's:
EtcdCluster
object. All our etcd data got lost in the process.So perhaps we can just embed etcd in the dotmesh server, or start a cluster up from the dotmesh operator, or something else. Assuming we still use etcd at all.