Rookie question. I've been tinkering with a 10-node distributed-replicated setup and I wanted to test what would happen if 1 machine died and I had to rebuild it. gluster> volume info all Volume Name: data Type: Distributed-Replicate Status: Started Number of Bricks: 5 x 2 = 10 Transport-type: tcp Bricks: Brick1: dl180-101:/data Brick2: dl180-102:/data Brick3: dl180-103:/data Brick4: dl180-104:/data Brick5: dl180-105:/data Brick6: dl180-106:/data Brick7: dl180-107:/data Brick8: dl180-108:/data Brick9: dl180-109:/data Brick10: dl180-110:/data I took down dl180-102 (dl180-101 is its replicate buddy) and reinstalled the machine, as if we had some horrible failure and just had to start over again. What would be the best method to get the new 102 back in the cluster without data loss? I tried to remove the 101 and 102 bricks thinking it would migrate the data (on 101) to other nodes but it didn't do that. Do I manually have to copy data from 101:/data onto the glusterfs and then add the 101/102 bricks and rebalance? Could I have used replace-brick to move the data to other existing bricks? Thanks, Graeme
Hi, I use only a replicated setup (two server), so it might be slightly different, but I've done something similar recently. I rebuilt the server with the same hostname / IP as the previous system and then: 1. peer probe <server that is up> # (get trusted by others) 2. volume sync <server that is up> # (get the volume configuration) 3. ls -laR /gluster-mount-point # On a client with the volume mounted, watch /var/log/gluster/<mount-path>.log You could try (if you have not reused the same hostname / IP): 1. peer probe <server that is up> 2. volume sync <server that is up> 3. volume remove-brick <volume> <brick which is missing> 4. volume add-brick <volume> <brick which is replacing> 5. volume rebalance <volume> start OR/AND maybe ls -laR just in case... I don't think you could use 'replace-brick' if the brick you are removing is already down. As stated at the beginning, I only run a two-node replicated set up but I hope this helps! Regards, Andrew On Feb 2, 2011 5:16 PM, "Graeme Davis" <graeme at graeme.org> wrote: