ProxySQL keeps sending SELECT traffic to a PXC node in Donor/Desynced

MySQL & MariaDB Percona Operator for MySQL
1

Hello!
ProxySQL keeps sending SELECT traffic to a PXC node in Donor/Desynced state during backups — how to exclude it?

Description:

I’m running Percona XtraDB Cluster Operator v1.17.0 (3 nodes) on Kubernetes with ProxySQL + pxc-backup.
During our scheduled backup/dump job, one of the nodes frequently goes into:

SHOW GLOBAL STATUS LIKE ‘wsrep_local_state_comment’; – Donor/Desynced
SHOW GLOBAL STATUS LIKE ‘wsrep_cluster_status’; – Primary
SHOW GLOBAL STATUS LIKE ‘wsrep_ready’; – ON

While in Donor/Desynced, the node still accepts SELECT queries, and ProxySQL continues to route read traffic to it. This sometimes causes clients to read stale data (the node is temporarily behind / applying queue grows).

Steps to Reproduce:

–

Version:

v1.17.0

Logs:

–

Expected Result:

ProxySQL should not route reads to nodes in Donor/Desynced; only Synced nodes should receive SELECT traffic, preventing stale reads during backups.

Actual Result:

ProxySQL still keeps this node in the reader pool and continues routing SELECT queries to it. As a result, clients sometimes read stale/outdated data from that node until it returns to Synced state.

Thank you!