Ceph » Orchestrator

On a newly deployed cephadm cluster Quincy/17.2.5 many daemons/services on the first machine (that was used for bootstrap) bind to "any address" while on other machines the same daemons are bound to a specific IP address:

• Services bound to "any address" on first node but to a specific IP addresses on other nodes: alertmanager, grafana, node-exporter, prometheus.
• Services bound similarly "any address" on all nodes: haproxy.ingress, mgr, rgw. These services seem normal.

Even after redeploying a service, for example grafana, still keeps the listening socket on the first/bootstrap node bound to "any address" and on other nodes bound to a specific IP address.

Example:

# ceph orch ps
NAME                               HOST    PORTS                 STATUS         REFRESHED  AGE  MEM USE  MEM LIM  VERSION         IMAGE ID
alertmanager.cep1                  cep1    *:9093,9094           running (59m)    84s ago   4w    24.8M        -                  ba2b418f427c
alertmanager.cep2                  cep2    192.0.2.12:9093,9094  running (59m)    85s ago   4w    23.7M        -                  ba2b418f427c
alertmanager.cep3                  cep3    192.0.2.13:9093,9094  running (59m)    20s ago  60m    26.6M        -                  ba2b418f427c
...

Looking at the individual nodes manually confirms that the listen socket is indeed bound as listed by ceph orch ps. This system is EL9 and podman 4.2.0.

The first node was bootstrapped alone without applying a spec file because there was persistent error trying to apply a spec file at bootstrap. The rest of the nodes and all services were deployed at one go in a single spec file that lists all hosts the exact same way.