On Mon, 2007-12-10 at 11:12 +0100, Alain Moulle wrote: > Hi > > .... > <quorumd interval="1" tko="10" votes="2" log_level="9" log_facility="local4" > status_file="/tmp/qdisk_status" label="QUORUMDISK"> * votes="1" * Get rid of status_file after you're done debugging. > </quorumd> > ... > <cman expected_votes="2" two_node="0"/> expected_votes="3" > And in all cases, I can't launch cman on the first node without avoiding an > automatic fence of second node after a while, and then cman accepts to run > on the first node (at least in both first cases 2/3 and 1/3, for 1/2 and 2/2 Qdiskd does not and can not prevent CMAN from fencing the dead portion of the cluster during startup. > I've stop cman before it fences the secund node as I saw in syslog that > it was about to fence in a while ...) > > Note that I do "service qdiskd start" after the "service cman start" and when > this last one is stalled a moment in fencing phase, because the qdisk man gives > in Limitations chapter : "cman must be running before the qdisk program can start" Not true anymore as of 5.1, but changing the order will not prevent cman from fencing the dead node. <fenced clean_start="1" /> will prevent fencing dead nodes on startup, but is dangerous to use. -- Lon -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
