This bug was reported from Eric's testing environment, it looks like a
live-lock problem, since the case have lasted for two days, but the process
stack always changed between three nodes.

[INFO]@{ocfs2dev1} Start multiple-node discontig block group testing...
[LOG]@{ocfs2dev1} discontig_runner.sh -m ocfs2dev1,ocfs2dev2,ocfs2dev3 -d
/dev/disk/by-path/ip-147.2.207.231:3260-iscsi-iqn.2016-09.example.com:ocfs2dev-lun-0
-s pcmk -n cluster /mnt/ocfs2
=====================Discontiguous block group test starts:  Wed Jan 11
01:37:27 CST 2017=====================
<- Running test with 512 bs and 4096 cs ->
[*] Multi-nodes Inodes Block Group Test:[80G[[1;32m PASS [0;39m]
[*] Multi-nodes Extents Block Group Test:Killed by signal 15.
Build was aborted