Description of problem:
In a replica 2 volume, if one of the bricks is down and it reports its state before the online one, AFR tries to find another online brick in find_best_down_child(). Since priv->child_up array has been initialized with -1 and this function only checks if it's 0, it considers that the other brick is alive and sends a CHILD_UP notification.
At this point the other xlators start sending requests, which fail with ENOTCONN when they reach afr. This can cause several unexpected errors.
Version-Release number of selected component (if applicable): mainline
It happens randomly, depending on the order in which bricks are started.
Steps to Reproduce:
REVIEW: https://review.gluster.org/19440 (cluster/afr: remove unnecessary child_up initialization) posted (#1) for review on master by Xavier Hernandez
COMMIT: https://review.gluster.org/19440 committed in master by "Pranith Kumar Karampuri" <email@example.com> with a commit message- cluster/afr: remove unnecessary child_up initialization
The child_up array was initialized with all elements being -1 to
allow afr_notify() to differentiate down bricks from bricks that
haven't reported yet. With current implementation this is not needed
anymore and it was causing unexpected results when other parts of
the code considered that if child_up[i] != 0, it meant that it was up.
Signed-off-by: Xavier Hernandez <firstname.lastname@example.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-v4.1.0, please open a new bug report.
glusterfs-v4.1.0 has been announced on the Gluster mailinglists , packages for several distributions should become available in the near future. Keep an eye on the Gluster Users mailinglist  and the update infrastructure for your distribution.