GitHub user lhotari edited a comment on the discussion: Pulsar upgrade to 3.0.5 causes prometheus metrics timeouts on brokers
> I check on that cluster's info out of curiosity and its RedHat 8, with still > cgroups v1. yes, RHEL 8 defaults to cgroups v1, but [it's possible to switch to cgroups v2](https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/8/html-single/managing_monitoring_and_updating_the_kernel/index#mounting-cgroups-v2_using-cgroups-v2-to-control-distribution-of-cpu-time-for-applications). I don't think that there's much advantage of switching. What seems to matter more with Kubernetes and Java is having a fairly recent kernel version. Btw. It's recommended to configured Kubernetes notes with THP setting set to madvise. This is what you get also on Cloud provider managed Kubernetes nodes. Azul has a good guide [how to configure THP this for RHEL](https://docs.azul.com/prime/Enable-Huge-Pages#permanent-thp-setting-on-rhel-centos-oracle-linux-amazon-linux) and others. It's recommended to configure THP this way also when huge pages aren't used. The default `/sys/kernel/mm/transparent_hugepage/enabled` setting of `always` is bad for running Java unless `-XX:+AlwaysPreTouch` and `MALLOC_ARENA_MAX=2` is used. GitHub link: https://github.com/apache/pulsar/discussions/22897#discussioncomment-9808668 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
