renecannao
commented
1 day ago First hypothesis: You probably have available only 50% of the CPU core you think you have available. If you are running in a cloud environment, the guest OS might think it has access to all N vCPUs, but the underlying virtualization layer and resource management mechanisms enforced by the cloud provider are ultimately controlling the actual CPU allocation and utilization. The 50% limit is almost certainly a deliberate constraint set at the VM or account level.
Another hypothesis: Assuming that ProxySQL is mostly spending time in performing network activities, it is also possible that you are hitting a bottleneck that is not CPU, but network. It could be hitting bandwidth limits, packets limits, or a variety of limits in handling network related interrupts.
My first recommendation is to profile CPU utilization, and then investigate possible bottleneck.
But this issue is unlikely related to ProxySQL's mysql-threads configuration
I have six ProxySQL instances running with 8 CPUs each and mysql-threads=24. However, under high traffic, the proxies are still only using a maximum of 50% CPU, which causes latency on the application side. Adding more proxies stabilizes the application and improves performance. I tried increasing mysql-threads beyond 24, but it didn’t make any noticeable impact.