Does Percona MySQL support arbitrator‑node capability

MySQL & MariaDB Percona Distribution for MySQL
1

I would like to check if there is any roadmap or plan for Percona Server for MySQL to support the group_replication_arbitrator feature for MySQL Group Replication.

Currently, the upstream Oracle MySQL Group Replication only supports full‑data members; lightweight arbitrator nodes that participate in quorum voting without storing user data are not available. The group_replication_arbitrator enhancement exists in GreatSQL. Are there any plans to implement similar arbitrator‑node capability in Percona’s Group Replication?

2

Hi Leo,

Thanks for reaching out. I explored the gitrepo and documentation for GreatSQL and reviewed the arbitrator specific information. Haven’t tested it though. I do not see any such requirement / feature request in Percona Server but you can surely raise a detailed feature request here: https://perconadev.atlassian.net/jira/software/c/projects/PS/list.

We have detailed information for contributing.

Thanks,
K

3

Hello @Leo_Ma,
lightweight arbitrator functionality already exists in Percona XtraDB Cluster. Since Group Replication is an upstream project, I doubt we would be adding new functionality to a “competing” clustering solution. If arbitrator nodes are something you need in your architecture, I strongly recommend you look at the more mature XtraDB Cluster solution.

4

I try to use Percona XtraDB Cluster in my product ，but full recovery and ddl is not easy to control

5

Maybe instead of giving it up, you can check the possible options for DDL by following Online schema upgrade - Percona XtraDB Cluster as well as our pt-online-schema-change tool, along with specific PXC options pt-online-schema-change - Percona Toolkit Documentation

For the crash recovery, the following would help: Crash recovery - Percona XtraDB Cluster

We are always available to help if anything is needed from the PXC side.

6

Neither of these require the use of an arbitrator. Full recovery is achieved by restoring a backup to node1, bootstrap node1, then start node2, and node3. This is almost the same as with Group Replication (ie: init the first node, then init nodes 2 and 3) but you must restore the backup to every node with GR. With PXC, you only restore to the first node and the other nodes manage sync automatically.

I’m not sure what issue you are having with DDL, maybe you can explain the issue, but DDL is quite simple in PXC. You either have the DDL execute on every node simultaneously in a non-blocking way (NBO), or you perform a rolling schema upgrade (RSU).