[pve-devel] [RFC pve-cluster] prefer 'name' subkey over 'ring0_addr' for nodename

[Thomas Lamprecht]

Ok, have some objections myself, the test i did were one an one node 
cluster (yeah, poor I know), on a three node cluster i have problems with:

> logging {
>   debug: off
>   to_syslog: yes
> }
>
> nodelist {
>   node {
>     nodeid: 2
>     quorum_votes: 1
>     ring0_addr: two-coro
>     name: two
>   }
>
>   node {
>     nodeid: 1
>     quorum_votes: 1
>     ring0_addr: one-coro
>     name: one
>   }
>
>   node {
>     nodeid: 3
>     quorum_votes: 1
>     ring0_addr: three-coro
>     name: three
>   }
>
> }
>
> quorum {
>   provider: corosync_votequorum
> }
>
> totem {
>   cluster_name: teachings
>   config_version: 11
>   ip_version: ipv4
>   secauth: on
>   version: 2
>   interface {
>     bindnetaddr: 10.10.1.151
>     ringnumber: 0
>   }
>
> }
I get the
> Oct 05 17:23:26 a corosync[1722]: [VOTEQ ] configuration error: 
> nodelist or quorum.expected_votes must be configured!
> Oct 05 17:23:26 a corosync[1722]: [VOTEQ ] will continue with current 
> runtime data
Error, which is a bit strange, my hosts file has

> # corosync
> 10.10.1.151 one-coro.proxmox.com one-coro
> 10.10.1.152 two-coro.proxmox.com two-coro
> 10.10.1.153 three-coro.proxmox.com three-coro

In there. I better recheck all.


On 10/05/2015 04:58 PM, Thomas Lamprecht wrote:
> I ran into trouble when switching the corosync communication to an own network, under the condition that I didn't want to move the webinterface address/communication with it.
> As a reference I followed roughly this procedure:
> I configured my new NIC/interface, added a new /etc/hosts entries for my static addresses where corosync should run, then changed the 'ring0_addr' from my nodes and the bindnetaddr from my totem respectively, followed by a reboot.
> It worked besides that pmxcfs uses ring0_addr for the nodenames, so .members had "wrong" nodename entries and the ha-manager (for example, there are more problems) couldn't find the lrm_status'es anymore (as it searched for the wrong /etc/pve/nodes/<...>) directory.
>
> As corosync.conf allows a node entry 'name' this should be used if defined, else we keep current behaviour and fall back to 'ring0_addr'.
> FYI, pacemacer also uses the name key, but only when ring0_addr is an IP address and not a hostname.
>
> Any objections? Else I will integrate those changes in pvecm, to ease up configuration for separated cluster communications and redundant ring protocol use.
>
> Thomas Lamprecht (1):
>    Prefer name key from corosync config for nodename
>
>   data/src/confdb.c | 6 ++++++
>   1 file changed, 6 insertions(+)
>