summaryrefslogtreecommitdiffstats
path: root/tools
diff options
context:
space:
mode:
authorAtin Mukherjee <amukherj@redhat.com>2016-10-21 18:27:45 +0530
committerKaushal M <kaushal@redhat.com>2016-10-25 04:28:10 -0700
commit054c294936845b1818d6e9b66514a832904e592e (patch)
treeac8a504efcdaa8c3324f9956c9087fd7fa0b86ea /tools
parentf4efbf0e3092bd2181f62be1e1f30f202678c866 (diff)
glusterd: use GF_BRICK_STOPPING as intermediate brickinfo->status state
On a volume stop trigger glusterd issues a brick-op to terminate the brick process during brick-op phase , however in the commit-op glusterd once again tries to kill the same process if it exists and then mark the brickinfo->status flag to GF_BRICK_STOPPED. In the former case, if brick is successfully killed there is a possibility that GlusterD will receive RPC_CLNT_DISCONNECT from the said brick process before even the commit op phase is executed and hence by that time brickinfo->status will still be set to GF_BRICK_STARTED. BRICK_DISCONNECT event should be only sent if a brick has been killed and not through a volume stop/remove brick trigger, however due to this trace, this event is also sent out on a volume stop. Fix is to introduce an intermediate state GF_BRICK_STOPPING which can be used to mark the brick status at brick op phase of volume stop/remove brick to avoid sending spurious BRICK_DISCONNECT events on a volume stop trigger. Change-Id: Ieed4450e1c988715e0f9958be44faa6b14be81e1 BUG: 1387652 Signed-off-by: Atin Mukherjee <amukherj@redhat.com> Reviewed-on: http://review.gluster.org/15699 Smoke: Gluster Build System <jenkins@build.gluster.org> NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org> CentOS-regression: Gluster Build System <jenkins@build.gluster.org> Reviewed-by: Kaushal M <kaushal@redhat.com>
Diffstat (limited to 'tools')
0 files changed, 0 insertions, 0 deletions