6:23 p.m.
More questions on this -- since I have 5 servers . Could the following work ? Each
server has (1) 3TB RAID 6 partition that I want to use for contiguous storage.
Mountpoint for RAID 6 partition (3TB) /brick
Server A: VOL1 - Brick 1 directory /brick/brick1 (VOL1
Data brick)
Server B: VOL1 - Brick 2 + VOL2 - Brick 3 directory /brick/brick2 (VOL1 Data brick)
/brick/brick3 (VOL2 Arbitrator brick)
Server C: VOL1 - Brick 3 directory /brick/brick3 (VOL1
Data brick)
Server D: VOL2 - Brick 1 directory /brick/brick1 (VOL2
Data brick)
Server E VOL2 - Brick 2 directory /brick/brick2 (VOL2
Data brick)
Questions about this configuration
1. Is it safe to use a mount point 2 times ?
https://access.redhat.com/documentation/en-us/red_hat_gluster_storage/3.3/html/administration_guide/formatting_and_mounting_bricks#idm139907083110432
says "Ensure that no more than one brick is created from a single mount." In my
example I have VOL1 and VOL2 sharing the mountpoint /brick on Server B
As long as you keep the arbiter brick (VOL 2) separate from the data brick (VOL 1) , it
will be fine. A brick is a unique combination of Gluster TSP node + mount point. You can
use /brick/brick1 on all your nodes and the volume will be fine. Using
"/brick/brick1" for data brick (VOL1) and "/brick/brick1" for arbiter
brick (VOL2) on the same host IS NOT ACCEPTABLE. So just keep the brick names more unique
and everything will be fine. Maybe something like this will be easier to work with, but
you can set it to anything you want as long as you don't use same brick in 2 volumes:
serverA:/gluster_bricks/VOL1/brick1
serverB:/gluster_bricks/VOL1/brick2
serverB:/gluster_bricks/VOL2/arbiter
serverC:/gluster_bricks/VOL1/brick3
serverD:/gluster_bricks/VOL2/brick1
serverE:/gluster_bricks/VOL2/brick2
2. Could I start a standard replica3 (VOL1) with 3 data bricks and add 2 additional data
bricks plus 1 arbitrator brick (VOL2) to create a distributed-replicate cluster
providing ~6TB of contiguous storage ? .
By contiguous storage I mean that df -h would show ~6 TB disk space.
No, you either use 6 data bricks (subvol1 -> 3 data disks, subvol2 -> 3 data disks),
or you use 4 data + 2 arbiter bricks (subvol 1 -> 2 data + 1 arbiter, subvol 2 -> 2
data + 1 arbiter). The good thing is that you can reshape the volume once you have more
disks.
If you have only Linux VMs , you can follow point 1 and create 2 volumes which will be 2
storage domains in Ovirt. Then you can stripe (software raid 0 via mdadm or lvm native)
your VMs with 1 disk from the first volume and 1 disk from the second volume.
Actually , I'm using 4 gluster volumes for my NVMe as my network is too slow. My VMs
have 4 disks in a raid0 (boot) and striped LV (for "/").
Best Regards,
Strahil Nikolov