We have a three node cluster on PXC 5.5. Writes happen on only one of the nodes. Whenever we alter a table (specifically very large tables), the entire cluster hangs. Process list shows many connections in state ‘wsrep in pre-commit stage’ on the main write node. These are queries pointing to tables not related to the alter table.
Is this simply a case of one of the nodes causing flow control? We have plans to replace our slowest node very soon, so I’m hoping it’s that. If we were less concerned about a single node being slightly behind, would increasing fc_limit help the cluster from getting hung up during alters?