It's very helpful, to add/remove nodes at runtime and rebalance the content with existing datanodes.
Otherwise a system interruption is necessary to create a dump and restart the cluster.
There are some fundamentals necessary to accomplish that:
- initdb a node with GTM to avoid global transaction ID conflict.
- dump a node with no-data
- restore a node with pg_restore
- lock DDL from being run on cluster nodes during a node addition. This insure that the new node catalog is in sync with the others.
Nothing has been implemented yet as this project lacks developers and community members.