Gluster users - Nov 2012 - why is a partly available volume still writeable?

Hi,

I'm doing some tests with glusterfs 3.3 and I noticed something weird. I
have a distributed volume consisting of a number of bricks across several
nodes. I access this volume from a linux client using the glusterfs mount
type. So far so good. Now, when I fail one or more bricks, to my surprise,
the volume does not go offline. I can still write to the partially online
volume. If I do a listing of the volume I can see some files gone, some
still there. Now, if I write a file of the same name and location as one of
the disappeared files, it will just get created. If I then bring the
missing bricks back online, I see this file twice! All metadata is the
same, including the inode number. How can this be? Is this expected
behaviour? Is there a way to prevent this?

Fred
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20121115/b8fab7f1/attachment.html>