Further to my last email, I've been trying to find out why GlusterFS is favouring one brick over another. In pretty much all of my tests gluster is favouring the MOST full brick to write to. This is not a good thing when the most full brick has less than 200GB free and I need to write a huge file to it. I've set cluster.min-free-disk on the volume, and it doesn't seem to have an effect. At all. I've tried setting it to 25%, and 400GB. When I run tests from an NFS client, they get written to the most full brick. There is less than 10% free on that brick so it should be ignore with the defaults anyway. Any ideas? JK
