[Gluster-users] Gluster volume synchronization issue with concurent access

Dmitry Kuznetsov dmitry.kuznetsov at ubisoft.com
Thu Feb 21 15:51:04 UTC 2013


We are experiencing an issue when trying to write to gluster volume concurrently from several servers.
Our gluster consists of 8 servers, each having 2 bricks. The system is designed to be used for very fast file access (write/read).
We have a php web servers running on each of the servers which will be writing to a gluster volume.
What is happening right now is that php fails (throws an exception) when it tries to create a directory (mkdir).
Actually it can't properly determine if a directory in question exists or not.
By the time that it makes a decision if(!is_dir($directory)){ mkdir($directory, 0755, true);  } , the directory in question is already created by
another server and mkdir crashes. After that the volume becomes corrupted and unusable.
Does anyone know how this synchronization issue can be dealt with?
Does configuring bricks in Array0 (Hardware) with SSDs will be enough to speed up the IO for Gluster to synchronise volume data?

Thank you.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130221/5fce21e5/attachment.html>

More information about the Gluster-users mailing list