[Gluster-users] Coherency problem with file buffer cache

[Brian Candler]

On Wed, Jun 13, 2012 at 11:27:12AM +0530, Sabyasachi Ruj wrote:
> I tried. The problem persists.
> 
> I think the elegant way to solve this is we should block fsync calls
> in a client until the glusterd server has notified all the clients
> that has the same file open.

I think the pattern we are concerned about is lock-read-write-unlock,
do you agree? (Since an unlocked read-write sequence would be fundamentally
subject to races anyway)

So maybe the time when you gain an exclusive lock on a file is the
appropriate time to flush your local cache? But that's just my thought.
Really the target would be whatever POSIX specifies, or failing that, how
NFS behaves.