I am noticing in benchmark evaluations for RHEL AI products that the PCP OpenMetrics PMDA takes alot longer than all other PMDAs I have running.

For example on my laptop with just Grafana metrics, compared to some other metric domains ...

nathans@fedora:~$ time pminfo -v kernel.all

real    0m0.026s
user    0m0.008s
sys    0m0.009s
nathans@fedora:~$ time pminfo -v simple

real    0m0.027s
user    0m0.011s
sys    0m0.012s
nathans@fedora:~$ time pminfo -v openmetrics

real    0m2.641s
user    0m0.006s
sys    0m0.014s
nathans@fedora:~$ 

Linux kernel PMDA is written in C, while simple PMDA here is the python version - but just has a handful of metrics.  The OpenMetrics PMDA is pretty close to the timeout even with just the one metric source, running with vLLM and other metrics exporters I have seen it time out during benchmark runs.