key collision

I am a newbie to the Percona cluster, please bear with me. I created a 3 node cluster over 2 geo-locations. Location A has 2 nodes with HAProxy load balancer, where Loc B has just 1 node. Everything seems fine. The web-app that makes use of it is a PHP, OLTP-type home grown application with off-course the typical PK-FK relations. Under a single MySQL box on the Innodb engine everything runs fine.

My question: If the network connection between Loc A and B is severed for some time and on both Loc A and B independant data inserts take place via the web-app, how does the cluster when the network connection re-establishes prevent/resolve key collision and subsequent bad pk-fk relations?

Thank you for your insight

This looks like more of a cluster and replication management.


have u read …?

Mainly the GTID would take care the collision stuff, these GTID will be shared across all nodes and will be executed when the cluster becomes stable.
The auto increment value also plays important role to avoid collisions.

In my point of view
For best results keep the auto increment value default on all nodes and the PXC should be of updated to latest version on all nodes!.

Note: Now the new 5.6 version Fully supports GTID replication

Phuiberts, if you split network between location A (2 nodes) and location B (1 node) - the quorum will be maintained only in A, hence B will stop accepting writes or reads to avoid split brain situation.
See also here:…in#split-brain