[UNANSWERED] PM 5.* - question regarding corosync.conf and node names

[mailinglists]

Hi guys.

to a cluster of 4 nodes running pve-manager/5.2-1/0fcd7879 I added a node p28 running:
pve-manager/5.3-7/e8ed1e22. (Yeah I know, I will update all once the secure systemd journald hits the repos.)

Now I see that new node has IP instead of hostname in ring 0 as shown below. I first noticed it in PM HTTP GUI.

Membership information
----------------------
    Nodeid      Votes Name
         1          1 p24 (local)
         2          1 p25
         3          1 p26
         4          1 p27
         5          1 10.31.1.28

I see that these names come from corosync.conf.

I do not remember adding other nodes by their names, so I wonder what has changed for it to be registered by IP instead by name. So how does "pvecm add" decide to add a node by IP or name?

I went to check DNS and PTR records thinking p28 might be missing and noticed that there were none, so I added them for all just in case. It seems, that DNS working or not, does not matter to corosync in this case.

I guess the cluster works because corosync is using brodcast packages and not direct IPs or node names listed in corosync.conf, but I still wonder do the node names in corosync play any role.

Can / should I just change corosync.conf on one node and restart corosync (systemctl restart corosync) on all nodes, or should I restart some other services too?
I do not use HA, so all VMs should stay online, but i won't be able to do any operations that require locking until quorum is established again, right?

So my final qustion is should I change them to IPs or to node names in corosync? I think I should change all to IPs, it should be safer.

 

[mailinglists]

Also if I change the node names to IPs in corosync.conf, I guess I also need to increase config version, right?

 

[mailinglists]

Can anyone answer the questions asked? (Bumbing the thread.

 

[mailinglists]

I read some more about corosync here: https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_corosync_configuration, but my questions about node names vs IPs still stand:

How does "pvecm add" decide to add a node by IP or name?
Should I change them to IPs or to node names in corosync?

@Chris @fabian , anyone?

 

[udo]

Hi,
I've used first nodenames (which are defined in each /etc/hosts to use the right network (cluster-network) and don't depend on dns).
But then I switched to IP-Adresses because of using ring0 + ring1.

Udo

 

[mailinglists]

Hi Udo and tnx for taking the time to answer one of my questions.

I guess I will change all nodes to their IPs (on one node, increment config number and then restart corosync just in case on that and all other nodes).
It will also help me with redundant rings for cluster communication if I decide I need them later. I don't do HA, so I don't need them here.

I wonder where do you define, when adding a new node, if IP or name will be used in corosync config, when using pvecm add?
I could swear that I always added by IP (I did not even have DNS records) but somehow node names ended up in corosync.conf with PVE 5.2 and IPs with PVE 5.3.