0

I have a weird issue regarding glusterfs,

So ive setup a very simple replication volume accross two nodes using this command

sudo gluster  volume create cloudStorage replica 2 transport tcp  machine1:/data  machine2:/data force

Ive then started the volume and mounted it on my two hosts (machine3, machine4) in /etc/fstab, this also works fine.

If I then create a folder on machine3 it will appear instantly on machine4 as available and is in the filesystem of machine2

If I then create a folder on machine4 it will appear on machine3's mount. But if i check machine1 the newly created folder doesn't appear. It eventually will appear on machine1's filesystem but is already available to machine3 to use and browse ? (machine1 can't browse to the newly created folder)

I plan to implement CTDB and im worried about a recovery file lock issue if gluster isn't replicating as it should be.

Im aware that this question may be abit vague so what I think im asking is

A) Why is replicating a single folder creation taking so long ?
B) Why does it take so long in one direction only ?
C) How can I improve the performance

A little background on the setup

The servers are

  • 2 ESXi
  • 2 Ubuntu 14.04 (glusterfs nodes)
  • 2 Ubuntu 16.04 (gluster client nodes)

They are linked between two ports in a draytek 2860N they are in the same building and the must be about 2 meters of cabling between the two ESXi hosts if that.

This image may help describe the problem

This May Help

1 Answer 1

0

Gluster doesn't replicate so much as allow each client to write to every relevant gluster node at the same time. This fan-out method of connectivity is what makes it so fast, but it is imperative that every client be able to connect to every gluster node.

There is a method of server-to-server replication, but it's for back-end processes and for correcting replication problems caused by clients that weren't connected to all gluster nodes.

Make sure that all clients have a stable connection to all storage nodes. If you have problems still, ort that doesn't help, please elaborate and I'll provide additional help.

1
  • Not sure if you're still paying attention here, but, if you are, can you add information about the server-to-server replication method to your answer? (I have a replicated pair where one server was down for four days and am having a devil of a time trying to find out how to force server-to-server replication and catch it up. I assumed heal or rebalance would do it, but no such luck.) Feb 15, 2018 at 12:47

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .