Alexandre Dutra

Related: https://github.com/k8ssandra/k8ssandra-operator/issues/490. I'd also note that currently the mgmt-api images we ship have only the following 2 combinations: * Cassandra 3.11 + OpenJDK 8 * Cassandra 4.0 + OpenJDK 11

I have some bad news. I installed K8ssandra with the above values. Generated cass-config-builder configuration: ```json { "cassandra-yaml": { ... }, "cluster-info": { "name": "k8ssandra", "seeds": "k8ssandra-seed-service,k8ssandra-dc1-additional-seed-service" }, "datacenter-info": {...

So it seems CMS is just unusable. I installed k8ssandra with the below values: ```yaml cassandra: version: 4.0.0 heap: size: 512M datacenters: - name: dc1 size: 1 gc: cms: enabled:...

Here are my suggestions for a correct cass-config builder config: Cassandra 3.11: heap sizes and GC settings should all go under `jvm-options` since this template takes care of everything in...

I just tried installing the same k8ssandra values above but with Cassandra 3.11. The generated cass-config builder config is perfect: ```json "jvm-options": { "additional-jvm-opts": [ "-Dcassandra.system_distributed_replication_dc_names=dc1", "-Dcassandra.system_distributed_replication_per_dc=1", "-Dcassandra.allow_alter_rf_during_range_movement=true" ], "cms_initiating_occupancy_fraction":...

> > So it seems CMS is just unusable. > > That was my conclusion as well when looking at this the other day. I think the [jvm11-server-options-cassandra-4.0.0.edn](https://github.com/datastax/cass-config-definitions/blob/master/resources/jvm11-server-options/cassandra/jvm11-server-options-cassandra-4.0.0.edn) file in...

CMS is present in `jvm8-server-options-cassandra-4.0.0.template`: this makes me think that CMS is not recommended indeed for C* 4.0.

@jsanda do we need to support jvm8-server.options in the operator or can I assume that it's always going to be jvm11-server.options?

> Java11 introduces new GC algorithms (Epsilon and ZGC), though I don't really think Epsilon is suitable for C*-type workloads, ZGC might well have some promise. Java12 adds a variant...

Note that #403 is going to change `startAllNodes` significantly, so we might want to investigate if this is still happening after that change is merged.