Bug 1324004

REVIEW: http://review.gluster.org/13906 (arbiter: write performance improvement) posted (#1) for review on master by Ravishankar N (ravishankar)

Note: With the patch applied, here is the throughput I get:

Arbiter:
0:root@vm2 ~$ gluster v create testvol replica 3  arbiter 1 127.0.0.2:/bricks/brick{1..3} forcevolume create: testvol: success: please start the volume to access data
0:root@vm2 ~$ gluster v start testvol
volume start: testvol: success
0:root@vm2 ~$ mount -t glusterfs 127.0.0.2:testvol /mnt/fuse_mnt
0:root@vm2 ~$ dd if=/dev/zero of=/mnt/fuse_mnt/file bs=1M count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 1.25445 s, 83.6 MB/s

REVIEW: http://review.gluster.org/13906 (arbiter: write performance improvement) posted (#2) for review on master by Ravishankar N (ravishankar)

REVIEW: http://review.gluster.org/13906 (arbiter: write performance improvement) posted (#3) for review on master by Ravishankar N (ravishankar)

REVIEW: http://review.gluster.org/13906 (arbiter: write performance improvement) posted (#4) for review on master by Ravishankar N (ravishankar)

COMMIT: http://review.gluster.org/13906 committed in master by Pranith Kumar Karampuri (pkarampu) 
------
commit e1004679563ef17c460f83098983baf105655712
Author: Ravishankar N <ravishankar>
Date:   Tue Apr 5 15:16:52 2016 +0530

    arbiter: write performance improvement
    
    Problem: The throughput for a 'dd' workload was much less for arbiter
    configuration when compared to normal replica-3 volume. There were 2
    issues:
    
    i)arbiter_writev was using the request dict as response dict while
    unwinding, leading to incorect GLUSTERFS_WRITE_IS_APPEND and
    GLUSTERFS_OPEN_FD_COUNT values (=4), leading to immediate post-ops
    because is_afr_delayed_changelog_post_op_needed() failed due to
    afr_are_multiple_fds_opened() check.
    
    ii) The arbiter code in afr was setting local->transaction.{start and len} =0
    to take full file locks. What this meant was even for simultaenous but
    non-overlapping writevs, afr_transaction_eager_lock_init() was not
    happening because afr_locals_overlap() always stays true. Consequently
    is_afr_delayed_changelog_post_op_needed() failed due to
    local->delayed_post_op not being set.
    
    Fix:
    i) Send appropriate response dict values in arbiter_writev.
    ii) Modify flock params instead of local->transaction.{start and len} to
    take full file locks in the transaction.
    
    Also changed _fill_writev_xdata() in posix to fill rsp_xdata for
    whatever key is requested for.
    
    Change-Id: I1c5fc5e98aba49ade540bb441a022e65b753432a
    BUG: 1324004
    Signed-off-by: Ravishankar N <ravishankar>
    Reported-by: Robert Rauch <robert.rauch>
    Reported-by: Russel Purinton <russell.purinton>
    Reviewed-on: http://review.gluster.org/13906
    Reviewed-by: Pranith Kumar Karampuri <pkarampu>
    Smoke: Gluster Build System <jenkins.com>
    CentOS-regression: Gluster Build System <jenkins.com>
    NetBSD-regression: NetBSD Build System <jenkins.org>

This bug is getting closed because a release has been made available that should address the reported issue. In case the problem is still not fixed with glusterfs-3.8.0, please open a new bug report.

glusterfs-3.8.0 has been announced on the Gluster mailinglists [1], packages for several distributions should become available in the near future. Keep an eye on the Gluster Users mailinglist [2] and the update infrastructure for your distribution.

[1] http://blog.gluster.org/2016/06/glusterfs-3-8-released/
[2] http://thread.gmane.org/gmane.comp.file-systems.gluster.user