Best practices for max_connections and threadsConnected monitoring in PXC 8.0

1

Hi all,

I would like to discuss the impact of high mysql connection counts (threadsConnected) on Percona XtraDB Cluster (PXC) 8.0 health and performance. Let’s assume 5 node cluster (PXC 8.0) including one garbd deamon with 2 load balancers (with single-writer setup)

In general, is my assumption correct that even if threadsConnected is relatively high (cca 110 connections), the cluster’s replication health remains unaffected as long as Flow Control is not triggered?

In other words, is connection count a secondary metric compared to wsrep_flow_control_paused, or can a high number of open connections (even if many are in a Sleep state) negatively impact the cluster’s health and performance?

Your input will be much appreciated!

2

Yes, your assumption is basically correct.

Flow Control is the signal to watch, not raw connection count.

PXC Primary health signals.

These directly describe cluster health:

  • wsrep_flow_control_paused
  • wsrep_flow_control_active
  • wsrep_local_recv_queue
  • wsrep_local_send_queue
  • wsrep_cert_deps_distance
  • wsrep_local_cert_failures

If these are calm, the cluster is replicating cleanly.

1 Like