OpenJDK 11 with cgroups v2 in OKD 4.9

[disposab1e]

The switch to cgroup v2 is a huge problem for all using older JDK's cause they depend on the cgroups v1 subsystem.

Is there a chance to install OKD 4.9 with cgroups v1?

[vrutkovs]

Is there a chance to install OKD 4.9 with cgroups v1?

Yes, add systemd.unified_cgroup_hierarchy=0 kernel argument via MC

[disposab1e]

@vrutkovs Thx for quick reponse! (as always!).

Can this be done via Adding day-1 kernel arguments like....

cat << EOF > 99-openshift-machineconfig-master-kargs.yaml
apiVersion: machineconfiguration.openshift.io/v1
kind: MachineConfig
metadata:
  labels:
    machineconfiguration.openshift.io/role: master
  name: 99-openshift-machineconfig-master-kargs
spec:
  kernelArguments:
    - systemd.unified_cgroup_hierarchy=0
EOF

[vrutkovs]

yes

[disposab1e]

It works! Brilliant!

[travier]

For reference: https://bugs.openjdk.java.net/browse/JDK-8230305
Any details about what exactly is broken and how would help here to make sure we don't break all Java based containers when moving to v2 by default.

[disposab1e]

I experienced a huge increase of memory usage. This starts with memory consumption increase in openshift-logging > 1.2Gi. JBoss EAP > 500Mi and in our maven builds > 2Gi.

I work with OCP 4.9 and OKD 4.9 and run all apps and services in both clusters the same way. After switching to OKD 4.9 a lot of Java services, happy in OCP 4.9, fail in OKD 4.9.

Yes I saw this referenced Ticket and realized this is fixed in 15 and this is no option for us. I'm sure many people who run < JDK 15 will have problems vith JVMs in OKD 4.9.

You stated in Road to OKD 4.9 that you have to add a note that cgroupsv2 is enabled on new installs. To be honest I think this really should be a warning!

[vrutkovs]

To be honest I think this really should be a warning!

How would the warning look like? "Cluster is installed, but please be aware that some Java apps may start consuming more memory?"

[disposab1e]

The change to cgroup v2 may effect your running containers. Please check your containers (e.g. Java JDK before version 15) before you upgrade to OKD 4.9. If you're not sure please consider using cgroup v1 by adding a day-1 kernel argument with:

cat << EOF > 99-openshift-machineconfig-worker-kargs.yaml
apiVersion: machineconfiguration.openshift.io/v1
kind: MachineConfig
metadata:
  labels:
    machineconfiguration.openshift.io/role: worker
  name: 99-openshift-machineconfig-worker-kargs
spec:
  kernelArguments:
    - systemd.unified_cgroup_hierarchy=0
EOF

[vrutkovs]

We're using https://github.com/openshift/installer to build OKD installer, so sure, if it gets accepted there it can be done. IMO its not a great idea - users are expected to read the documentation before installing and be aware of gotchas of their own workloads - instead of being warned at the end of the install

[disposab1e]

However... just let the people know... there might be a good reason why OCP is NOT switching to cgroup v2 ;-)