Replicated database supports some replicas as observer role to get metadata and apply log

Use case The sql to create Replicated database is:

CREATE DATABASE test_db ENGINE = Replicated('zoo_path', '{shard_name}', '{replica_name}') [SETTINGS ...]

In cluster mode, we should use on cluster default to make all server to create the database test_db. But when the cluster has new server joined, the server haven't the test_db database. In some case, some server quit from the cluster, then all dll in the Replicated database will be unfinished.

Describe the solution you'd like

I suggest we should provide a server setting default_replicated_database_path='/clickhouse/databases', it's a zk path and the Replicated engine use default argument Replicated('default_replicated_database_path/{db_name}', '{shard_name}', '{replica_name}').

If a server is run as observer by server setting observe_replicated_database=true, then it will create all databases under path /clickhouse/databases and only get metadata and apply log in every replicated database, just like a learner in raft.

Additional context After implement this feature, the cluster of compute group will be more elastic.

ClickHouse / ClickHouse

Replicated database supports some replicas as observer role to get metadata and apply log #50824