[Gluster-users] changing IPs on existing gluster servers

[Brock Nanson]

Carlos,

Not sure how you went about this but I can tell you that I was successful in a simple configuration with two bricks. If I recall correctly, I first made sure that the peer configuration files referenced by machine name instead of IP. Then I changed the IP in each hosts file to suit new scenario. Finally I manually changed the static IP and when it came back up if all just worked.

Very simple test bed configuration though with only one NIC in each machine. 

Something to watch for... when I first did my original peer probe, the probed machine used the IP of the querying machine in the config file. The first machine used the name used in the probe command. I changed the IP to a name right at the start to allow the hosts file to determine the actual IP on both ends. 

Brock Nanson
  Original Message  
From: Carlos Capriotti
Sent: Tuesday, March 11, 2014 4:15 AM
To: gluster-users
Subject: [Gluster-users] changing IPs on existing gluster servers

Hello all.

Recently, playing around with gluster, trying to stress test it on several
scenarios, I ran into an unexpected problem:

gluster has a severe allergic reaction to changing IP address.


This happened on a server that has a single NIC, but I have cases with
multiple NICs. Once I changed the IP on a sever, gluster went babanas.
Trying to restart the service returned "error" (I admit the logs were not
very clear about what was going on), and that was that.

I changed the IP back, carefully rebooted the server just in case, checked
that gluster was started, stopped the only existing volume, deleted the
volume, stopped gluster, changed the IP, rebooted the server and....

Gluster would not start anymore. Kaput. I tried looking for config files
that would contain ip information, etc, in vain.

It might not seem like a big deal, but it may be: Right now I have my
environment running on bonded 1 Gbps NICs, but if I get the budget
approved, I am moving to 10 GB and will be KEEPING the 1 GB link for the
clients.

Final config (rather paranoid, I admit) will be one IP/nic for management
(I still have 10 Mpbs cards, believe me !), 1 bonded ip for clients
(redundancy), and the 10 GB for gluster data replication.

In that scenario, gluster WIL stop working almost for sure.

Any hints about this case ?

Cheers,

Carlos