Comment 2 for bug 1934143

100% agreed. The default 4GB is fine, but we definitely need a way to bump it particularly for dedicated storage nodes instead of hyper-converged.

As stated in the bug description, cephadm has a logic trying to use 70% of host memory with OSDs. We could have a similar option and suggest setting 70% (but with an empty value for backward compatibility).
https://docs.ceph.com/en/latest/cephadm/services/osd/#automatically-tuning-osd-memory
https://github.com/ceph/ceph/blob/08718ae283f6a62612c74b10d5da7efd2ea5f896/src/pybind/mgr/cephadm/serve.py#L172