Bootstrapping galera problem with quorum

[Ari Maniatis]

I'm trying to migrate an existing mariadb 10.1 server to a three node cluster with the minimum downtime. We have a few problems:

1. On our first attempt we put all three nodes into the existing server, and restarted with the --wsrep-new-cluster command. That was fine, but when we joined the next node, the original server decided it was now in split brain and stopped responding to write queries.
2. So we tried again and tried to join only one node (even though this meant an even number of nodes). We tried to set pc.weight on the original server so that it would continue to operate even if it felt it was a minority. However setting that value, also set the value on the second node (as seen in the logs). Which seems a bit silly.
3. When we persevered, the original server became so overloaded it stopped responding to queries. We only have 40Gb of data, but that took a little over 30 minutes to rsync during which time the original server was almost unresponsive (simple selects took 60 seconds and more).

During migration should we be turning off the quorum? Is there a way to slow down the rsync to reduce load?

Ari