[Bug] incorrect CPU usage collected by pulsar

[thetumbled]

Search before asking

• I searched in the issues and found nothing similar.

Version

pulsar 2.9.3
linux 4.19.16

Minimal reproduce step

1. enable CGROUP
2. assign 23 CPUs to broker process,while there are 40 CPUs in the system
   taskset fffffe0000 bin/***

What did you expect to see?

1. cpu usage is smaller than 100%

What did you see instead?

1. cpu usage is greater than 100%,and the cpu usage reported by pulsar need to be fixed by multiply 23/40.
2. for example,pulsar could report 150%,but the real cpu usage is 150*23/40=86.25%

Anything else?

this is because Runtime.getRuntime().availableProcessors() method return 23, which is the number of processors can be utilized by this process, but what we need is 40.
we could retrieve metadata from ZK, and verify that the limit of cpu is 2300, not 4000.

Are you willing to submit a PR?

• I'm willing to submit a PR!

[mattisonchao]

Thanks for your contribution!

[lhotari]

this is because Runtime.getRuntime().availableProcessors() method return 23, which is the number of processors can be utilized by this process, but what we need is 40.

One workaround for the above issue is to add JVM option -XX:ActiveProcessorCount=n where n=the total available CPU cores. In the example case, adding -XX:ActiveProcessorCount=n to Pulsar's JVM options would be a workaround.

[thetumbled]

actually,we restrict pulsar broker to use 23 CPU cores on purpose. i wonder whether there are any problem to set a dummy value 40.

This HotSpot option is recognized by OpenJ9 for compatibility. Use this option to override the number of CPUs that the VM automatically detects and uses when creating threads for various subsystems.
https://www.eclipse.org/openj9/docs/xxactiveprocessorcount/

[github-actions]

The issue had no activity for 30 days, mark with Stale label.

[lhotari]

Addressed by #16832