Hide Forgot
Description of problem: gluster 3.3 beta2 "/etc/init.d/glusterd restart" kills glusterfsd, which Joe Julian says is not supposed to. It should just restart the console management daemon. Looks like 3.2.4 has the same issue too. This causes active reads/writes to fail. /etc/init.d/glusterd: 11 GLUSTERFSD=glusterfsd 28 # Stop the service $BASE 29 stop() 30 { 31 echo -n $"Stopping $BASE:" 32 killproc $BASE 33 echo 34 pidof -c -o %PPID -x $GLUSTERFSD &> /dev/null 35 [ $? -eq 0 ] && killproc $GLUSTERFSD &> /dev/null Version-Release number of selected component (if applicable): 3.3 beta2 How reproducible: always Steps to Reproduce: 1. /etc/init.d/glusterd restart 2. active reads and writes stop / crash since glusterfsd is killed Actual results: active reads and writes stop / crash since glusterfsd is killed Expected results: active reads and writes should not stop since only the glusterd management process should be restarted Additional info:
patch sent : http://review.gluster.com/2919
CHANGE: http://review.gluster.com/2919 (init.d: stop only 'glusterd' process on '/etc/init.d/glusterd stop') merged in master by Anand Avati (avati)
Not a blocker. Patch already pushed on upstream, after more baking time, lets pull it in. Also it has some implication on upgrade process, hence not treating it as blocker.
not planing to have it in RHS 2.0 right away, closed upstream.
the bug fix is only in upstream, not in release-3.3. Hence moving it out of the ON_QA, and setting MODIFIED (as a standard practice @ Red Hat)
It looks like this has affected the stop functionality of the init script as it leaves glusterfsd running when service glusterd stop is ran. This jumped out at me in self heal testing, when running: service glusterd stop on node 2 create a bunch of files on client check the brick on node 2, files exist When stop is executed, should glusterfsd still be running? If it should be stopped maybe we could do something like: if [ $1 = "stop" ]; then <kill glusterfsd> fi
Ben, IMO, if command is "service glusterd stop", it should stop only glusterd, and not glusterfsd. That is because glusterd is a management daemon, and we should not be stopping data path process glusterfsd for the same. If one is doing some script to kill the brick, then please use 'kill glusterfsd' in the script itself.
Verified on glusterfs-3.4.0qa5-1.
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. http://rhn.redhat.com/errata/RHBA-2013-1262.html