octopus: cephadm bootstrap --container-init broken in Octopus
In Octopus, when the user provides the "--container-init" option to "cephadm bootstrap", all containerized daemons in the cluster are supposed to get the "--init" option in their unit.run files, to ensure that an init process runs as PID 1 in the container. (Without this, the containerized daemons fail to generate coredumps when they crash.)
However, due to a bug in Octopus, when the "--container-init" option is provided to "cephadm bootstrap", only the daemons actually created by the bootstrap operation (which in some cases might be just 1 MON and 1 MGR) get the "--init" in their unit.run. Daemons created post-bootstrap using "ceph orch apply" do not get the "--init" goodness, giving rise to an undesirable situation where 1 MON and 1 MGR generate coredumps when they crash, but all other daemons in the cluster fail to do so.
Since "--container-init" is already deprecated in Octopus, I'm not sure if newer versions of Ceph (pacific, quincy, etc.) suffer from this bug.