[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Gluster-devel] request for comments
From: |
John Rowe |
Subject: |
[Gluster-devel] request for comments |
Date: |
Wed, 02 May 2007 19:42:59 +0100 |
Go for it. Having the two bricks knowing about each other must make
sense. It could also make reconstruction a lot quicker if the working
brick knew that the failed brick was working until time T then when the
failed brick comes up it's probably safe to say that files with an mtime
less than T-120 were OK. And maybe in the long term the client-side stat
buffer could be updated via multicast..
On the other hand, when the master brick goes down it will be tricky to
make sure that the effective file-system is unchanged with the
switchover to the former slave. Several ways to think of doing that but
the devil is in the details.
John