Adding new data nodes to a cluster

[Pierre Sion]

Hello,

The MySQL cluster documentation states, in chapter 7.13, that upon adding new data nodes to the cluster, to repartition existing data across the new node group, an "ALTER ONLINE TABLE XXX REORGANIZE PARTITION" must be issued for each table.

I must admit I am a little bit taken aback by this approach;

1- what happens if we handle a "complex" schema with maybe 40 tables, with foreign-key relations? Is there a recommended stategy to order the tables to alter, such as trying to represent the schema as some kind of graph, and start issuing the ALTER TABLE statements on the leaves first, then up to the top?

2- what happens if we have high-volume tables, say 10 millions rows per table? How long will one ALTER TABLE last? Will it severely affect read/write performance of applications still accessing the cluster?

3- can the cluster be reliably accessed while the tables are altered ? The documentation mentions an "internal commit point", upon which failures are rollbacked or "rolled forward". Does someone know what kind of threshold this "commit point" represents? Is it configurable?

4- last but not least, could someone share their experience of adding new data nodes in production? Is it something that goes generally smoothly?

Thank you in advance for your feedback,
Best regards,
Pierre Sion