Shutting down a node
It’s possible to shutdown server on fly, but that would lead to failure of some queries.
More safer way:
Remove server (which is going to be disabled) from remote_server section of config.xml on all servers.
Remove server from load balancer, so new queries wouldn’t hit it.
Wait until all already running queries would finish execution on it. It’s possible to check it via query:
Run sync replica query in related shard replicas via query:
SYSTEM SYNC REPLICA db.table;
SYSTEM SHUTDOWN query doesn’t wait until query completion and tries to kill all queries immediately after receiving signal, even if there is setting