Problem while discovering the gluster nodes
When I tried to install the gluster nodes using gluster-colonizer, I got some issues with the discovery of the nodes. It was reporting that 2 nodes were found, then 1 and then zero and failing then the installation. While troubleshooting I noticed that gluster-discovery reports different ip each time:
$ gluster-discovery -v discover discovered host: 192.168.100.12 discovered host: 10.10.0.199 discovered host: 10.10.0.190
$ gluster-discovery -v discover discovered host: 192.168.100.13 discovered host: 10.10.0.199 discovered host: 192.168.100.12
$ gluster-discovery -v discover
discovered host: 192.168.100.13 discovered host: 10.10.0.191 discovered host: 10.10.0.199
As a workaround I disabled one interface on each host and this worked.
This is an effect of the @gluster/gluster-zeroconf tool. Right now, this indeterminately returns any provisioned IP address from the discovered node. I think it would be good for gluster-zeroconf to return all IPs, and then we can simply filter on the subnet we are looking for.
https://github.com/gluster/gluster-zeroconf/issues/8
I actually went the opposite way. As gluster-colonizer anyway asks (or self configure, don't remember right now) for a mgmt network, IMHO it would be good to limit the search on this specific subnet. That's why I did push some changes to gluster-zeroconf
We don't exactly search for the IPs in that way. The gluster-zeroconf tool advertises, and we detect that advertisement. We do then filter on the subnet we are looking for, but we have to wait until we get lucky and zeroconf returns only IPs on our subnet. Thus, we need to change what zeroconf advertises and cannot directly fix this in the colonizer.