Ceph

when stoping ceph, multiple kill commands may be sended to one osd or monitor every one second at /etc/init.d/ceph. However, ceph iteself can only handle one kill command at its lifecycle. if there is more kill commands, the system will take over the rest kill command by terminating the osd or monitor imediatelly. This situation is easy to appear if ceph cluster currently has huge loads, where the osd or monitor can't exit successfully imediatelly at first kill command, due to jobs,like sync disk. After one second, another kill command will send to the same osd or monitor, which will destory the normal exit procedure causing data lose, or at least pg data disturbance.

at /etc/init.d/ceph
stop_daemon() {
......
while [ -e /proc/\$pid ] && grep -q $daemon /proc/\$pid/cmdline ; do
cmd=\"kill $signal \$pid\"
echo -n \$cmd...
\$cmd
sleep 1
continue
done
......
}