Does this work right when ganesha is running on a different port?

Currently cephadm does not deploy ganesha on non-standard port.

Let's think about this. Cephadm doesn't (and hopefully never will) need the raw IP address. On the other hand. in case we want to do enhance cephadm, there is another option:

1. Add a new property ip or something similar to

2. Enhance

to run also getaddrinfo for the host. (Do it here, to be able to return the current IP of a host)

4. update the IP in the inventory, similar to

5. let ceph orch host ls also return this attribute

6. change this code here to also call self.mgr.get_hosts()

This way, everyone would benefit from this information.

But I'm perfectly happy with keeping this code here in show_nfs_cluster_info

Let's think about this. Cephadm doesn't (and hopefully never will) need the raw IP address.

I imagine what cephadm has for a hostname is just whatever gets passed to getaddrinfo (in ssh or elsewhere). Whether that "hostname" is also just a numeric one (i.e. an IP address) is irrelevant to it. (Edit:) Actually, I think it did matter because what I ended up doing in ceph-linode was lookup the FQDN for each VM:

https://github.com/batrick/ceph-linode/blob/cephadm/cephadm.yml#L63-L69

Which does not thrill me. I do think cephadm should be willing to take an IP address for the host so it works with private LANs. IMO, there really should not be any requirement that the cluster have hostnames which can be looked up via DNS.

Let's think about this. Cephadm doesn't (and hopefully never will) need the raw IP address.

I imagine what cephadm has for a hostname is just whatever gets passed to getaddrinfo (in ssh or elsewhere). Whether that "hostname" is also just a numeric one (i.e. an IP address) is irrelevant to it. (Edit:) Actually, I think it did matter because what I ended up doing in ceph-linode was lookup the FQDN for each VM:

https://github.com/batrick/ceph-linode/blob/cephadm/cephadm.yml#L63-L69

Which does not thrill me. I do think cephadm should be willing to take an IP address for the host so it works with private LANs.

you can do that already:

ceph orch host add <name> <addr>
# e.g.
ceph orch host add myhostname 1.2.3.4
# or
ceph orch host add myhostname myhostname.domain.tld

cephadm will then use the addr to connect to the host.

IMO, there really should not be any requirement that the cluster have hostnames which can be looked up via DNS.

yup. that was the reason we added the addr field

I would prefer moving this to cephadm and do name resolution only if ip is not specified.

Return an empty dictionary if there are no NFS clusters?

{}

and if the NFS cluster is defined, but no daemons are running?

{
    "ganesha-foo": {}
}

Return an empty dictionary if there are no NFS clusters?

{}

Done.

and if the NFS cluster is defined, but no daemons are running?

{
    "ganesha-foo": {}
}

I prefer not listing it if no daemons are present.

maybe still report the daemon exists, but set the IP to a NoneType?

Let's not add daemon if we cannot resolve its hostname.

on second thought, is the daemon_id useful in this context? maybe return a List instead?

Right daemon_id is not instead I have added cluster_id

{
    "vstart": [
        {
            "hostname": "varsha",
            "IP": [
                "::1",
                "127.0.0.1"
            ],
            "Port": 2049
        }
    ]
}