[Gluster-users] GlusterFS running, but not syncing is done
krishna at zresearch.com
Thu Mar 12 14:08:19 UTC 2009
On Tue, Mar 10, 2009 at 2:23 AM, Keith Freedman <freedman at freeformit.com> wrote:
> At 05:34 AM 3/9/2009, Krishna Srinivas wrote:
>> Do not use single process as both server and client as we saw issues
>> related to locking. Can you see if using different processes for
>> server and client works fine w.r.t replication?
> this is news to me. when will this be fixed?
> It used to be that single process was recommended for performance reasons?
Yes it was recommended but later one of the users reported a bug in
file record locking.
To explain technically, fuse and protocol/server both maintain inode
table (hashed inode structure list). Hence when client and server are
used in same process a file will have two disparate inode structures -
one in fuse's inode table and one in protocol/client's inode table.
When two processes - one on the same machine and one on a different
machine acquire a lock on a file both act on different inode
structures and hence both lock calls succeeds - which is a problem.
Hence it is not advisable to use a single process for client and
We have no plans of fixing this now as things work fine when they are
separate processes and technically it is difficult to fix.
More information about the Gluster-users