[Gluster-users] Missing files and folders

Niclas Hughes Niclas.Hughes at securetrading.com
Thu Feb 10 15:58:36 UTC 2011

Hi Jacob,

Thanks for your quick reply.

I understand that, but if I setup a 2 machine setup making the type: replicate and then add bricks it forces it to be Type: Distributed-Replicate.

Is there any way of having the setup stay as replicate so that it does not matter what machines go down.


-----Original Message-----
From: Jacob Shucart [mailto:jacob at gluster.com] 
Sent: 10 February 2011 15:53
To: Niclas Hughes
Cc: gluster-users at gluster.org
Subject: RE: [Gluster-users] Missing files and folders


Distribute-Replicate means that you have 2 sets of mirrors instead of just
one.  Gluster will distribute files evenly between these 2 mirror sets.
In the configuration below, Brick1 and Brick2 is 1 mirror set, and Brick3
and Brick4 is your second mirror set.  In this configuration approximately
50% of the files will go on the first mirror set and 50% will go on the
second mirror set.  This means that if you shut down a single server, all
of your files will be available.  If you were to shut down two servers
that were not both a part of the same mirror set like Brick1 and Brick3,
then all of your files would still be available.  If you shut down 2
servers that are part of the same mirror set, then the files on that set
will not be available until one of those servers comes back up.  This is
what happened below when you shut down Brick3 and Brick4.


-----Original Message-----
From: gluster-users-bounces at gluster.org
[mailto:gluster-users-bounces at gluster.org] On Behalf Of Niclas Hughes
Sent: Thursday, February 10, 2011 6:11 AM
To: gluster-users at gluster.org
Subject: [Gluster-users] Missing files and folders

version 3.1.2 latest Build Jan 18 2011

Volume Name: brick2
Type: Distributed-Replicate
Status: Started
Number of Bricks: 2 x 2 = 4
Transport-type: tcp
Brick1: linguest2:/data/brick2
Brick2: linguest3:/data/brick2
Brick3: linguest4:/data/brick2
Brick4: linguest5:/data/brick2

If you read a write files to each of these machines, then make sure that
these files are accessible on all 4 machines.
Now shutdown linguest4 and linguest5.

Originally the 2 machine setup was type replicate but when you add two
further machines it changes the type to be "Type: Distributed-Replicate"
as above.
Now that the two machines have been halted the other two machines now have
files missing.
As the documentation states: "Distributed replicated volumes replicate
(mirror) data across two or more nodes in the cluster" but if it was a
mirror then I would have expected the 2 machines to still have access to
all the files.

Does anyone know if this is what is meant to happen or is this a bug.
Also this has been tested many times and this is the result every time.


More information about the Gluster-users mailing list