We do have reports of sporadic readonly cluster with CS in Sky. I suggest we review startup

1. for docker-compose
docker-compose up -d && docker exec -it mcs1 provision
here we wait for all nodes to start and then execute cluster demo on mcs1
this is pretty simple https://github.com/mariadb-corporation/mariadb-skysql-columnstore-docker/blob/master/scripts/provision

2. for sky this is very different . look at the logic https://github.com/mariadb-corporation/mariadb-skysql-columnstore-docker/blob/12de7a737b47dc14c09321401c996fdce3c01ebe/scripts/skysql-specific-startup.sh#L80
it seems that SKY is using quite a complex logic and the timing is not clear.

Could we switch to the provision model where provision script allows to pass in the array of hosts

-operator start N nodes
-each node completes starting server and CMAPI server
-operator calls cluster_demo on node 1 and passes array of nodes . The cluster demo completes the startup

the pod shutdown should be handled unirformly. Ideally a new deprovision script that Todd provides and operator invokes at end