Here is another scenario that I'm worried about.
If there is a file1 on brick1, and file2 on brick2.
and then brick1 goes down.
If I do a "mv file2 file1" or even up to the directory a "rename dir2 dir1",
will AFR properly handle these cases?
dd
From: "Daniel van Ham Colchete" <daniel.colchete@xxxxxxxxx>
To: gluster-devel <gluster-devel@xxxxxxxxxx>
Subject: Files erased while one brick was down in
AFRreturns from the afterlife
Date: Mon, 2 Jul 2007 14:30:40 -0300
MIME-Version: 1.0
People,
Gerry Reno raised a very interesting question about AFR.
Let's imagine the following case: two bricks AFRed. I create a few files
(say mail files inside a Maildir). The second brick goes down. I erase a
few
e-mails (checking my POP3 account). Them I bring back the second brick. The
files created/changed during this time will be copied to the second as soon
as I open() them. The problem is: if I "ls -la" I'll see the old erased
files. If I try to open them GlusterFS will even replicate it. Why this is
happening: when I do a "ls -la" Gluster will check both bricks and give the
list of files inside both servers.
I'm using only AFR, without unify. I tested it here and got the expected
result.
I can propose one solution for this problem. If you apply the same
versioning algorithm used with files to the directories themselves the
self-heal could detect witch dir has the newest file list and erase what's
necessary.
What do you think?
Best regards,
Daniel Colchete
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
_________________________________________________________________
http://newlivehotmail.com
