[Gluster-users] DRBD with gluser

Ran smtp.test61 at gmail.com
Mon Aug 31 07:35:18 UTC 2009

Hello ,
Im tring to set up gluster volume but to use drbd to replicate 2
vloumes insted of using the AFR translator , it took me 1 hour to
realize that the AFR is not stable for the following resones :

in the example i tested 2 servers in AFR and distribute only 1 AFR
(possible add more but just test with 1 AFR volume)

volume vo1
  type cluster/afr
#  option data-self-heal off
#  option   metadata-self-heal off
#  option   entry-self-heal  off
  subvolumes storage1 storage2

volume cluster
 type cluster/distribute
 option lookup-unhashed yes
 subvolumes vo1

- all workimg ok , then if node 1 is rebooed node 2 takes its place
and start serving KVM images ok also
but when node 1 comes back from reboot the KVM images is locked by server 2
and the data is not replicated back to server 1
i read the gluster papers regarding this issue with opened files that
were opened when the otehr member of the AFR is not online and its a
big issue seens
if server 1 just reboot for a minut then all replication is no good anymore

So i guues in this stage and limitation its better be called 1 way
replication with limitations and not proof for production use .

In any case i decided to do the redundency of server 1 and 2 with DRBD
wich is very stable and use gluster for its better abilities such as
option of growing the file system in the future if needed and the
client fail safe to other member node if 1 member is not online insted
of using linux HA(heatbeat)

I tested few client VOL configs and noticed that its not quite simple

If i ommit the AFR translator and use only

volume cluster
 type cluster/distribute
 option lookup-unhashed yes
 subvolumes storage1 storage2

with or without option lookup-unhashed yes
then after a while every file show up 2 times in the mount point
and its also not clear how the distribute will work on 2 DRBD(same volumes)

Is there any way to implement this aka AFR translator that doesnt
replicate at all e.g just bond the 2 nodes to 1 volume for load
balanced and distribute perposes .

I realy like gluster for its simplicity and if will be implemnted
right and will work ok too then its much better then other file
systems in that category like lustre for my needs anyway .

sorry for my english .

Rob .

More information about the Gluster-users mailing list