Red Hat Bugzilla – Bug 763294
insufficient consistency check on start of glusterd
Last modified: 2013-12-08 20:19:54 EST
If the gluster instance has terminated its course in a dirty way -- eg. we do
# pgrep gluster | xargs kill -9
(IRL assume a system crash, an OOM situation or a segfault), then when glusterd is started next time, it still shows the gluster volumes "started" -- albeit the gluserfs servers are not running.
Any inconstency in permanently stored state should be detected at startup time, so that if things gets cluttered, then, in lack of a better idea, as a last resort we can go for a full restart of the gluster ecosystem.
In this case, a check of pid files against volume info file seems to be sufficient.
PATCH: http://patches.gluster.com/patch/4927 in master (Restart all bricks which are down when glusterd comes up)