When coherent byte range locks is attempted in afr setup, the clients hang and become unresponsive. Ping Pong (http://wiki.samba.org/index.php/Ping_pong) is being used to attempt coherent byte range locks.

The client and server vol files are attached. 

Thanks to John Simmonds (john2.simmonds.uk) for pointing this bug.

replicate/afr_lk_cbk used the struct flock returned by a previous STACK_WIND to do STACK_WIND on the next subvolume, instead of using the struct flock supplied by the user.

After applying http://patches.gluster.com/patch/813/ the ping pong tool behaves as expected in http://wiki.samba.org/index.php/Ping_pong

PATCH: http://patches.gluster.com/patch/813 in master (cluster/afr: Use user-supplied struct flock to STACK_WIND)

PATCH: http://patches.gluster.com/patch/813 in release-2.0 (cluster/afr: Use user-supplied struct flock to STACK_WIND)

I was the gentlemen who originally pointed out this bug and I am very happy to confirm that the patch fixed my/this problem.

I have tested this on the setup I have here that uses both the distribute and replicate translators. 

Thank you very much GlusterFS support team.

Regards,

John Simmonds

(In reply to comment #5)
> I was the gentlemen who originally pointed out this bug and I am very happy to
> confirm that the patch fixed my/this problem.
> 
> I have tested this on the setup I have here that uses both the distribute and
> replicate translators. 
> 

Thanks for your confirmation!