Re: NDB Nodegroup questions

[Mikael Ronström]

Think of a Nodegroup as a shard, within a node group all data is
replicated. Another node group contains completely different data
and thus data is partitioned amongst the node groups.

So in your example node 1 and node 2 replicates to each other,
so thus node 3 and node 4, node 5 and node 6 and node 7 and node 8.

If you lose all nodes within one node group the cluster goes down
and you have to restart the entire cluster. But you can lose one
node in each node group without the cluster crashing and you can
bring those nodes back as online operations.

When you added the nodes you created the opportunity to use more nodes
to store your data. New tables will immediately take this opportunity.

For already existing tables you need to execute
ALTER TABLE table_name REORGANIZE PARTITION algorithm=inplace;
on all tables, it is also a good idea to follow this up with
OPTIMIZE TABLE to reclaim some of the memory used in the tables
in node 1 through 4.

So think of your add node operations as a three step approach:
1) Add new node groups (shards)
2) Reorganisze your data to use the new node groups
3) Reclaim your memory space in the old node groups