Le 19/10/2016 à 08:53, Ulrich Windl a écrit : >>>> Ken Gaillot <[email protected]> schrieb am 18.10.2016 um 17:07 in >>>> Nachricht > <[email protected]>: >> On 10/14/2016 03:22 PM, Anne Nicolas wrote: > > [...] >>> cluster logs are flooded by : >>> Oct 14 17:42:28 [3445] bzvairsvr attrd: notice: >>> attrd_trigger_update: Sending flush op to all hosts for: >>> master-drbdserv (10000) >>> Oct 14 17:42:28 [3445] bzvairsvr attrd: notice: >>> attrd_perform_update: Sent update master-drbdserv=10000 failed: >>> Transport endpoint is not connected >> >> This is strange, and the cause of the problem. A master/slave resource >> agent will try to set node attributes indicating which node should >> become the master. Here, we see that this is failing -- it appears attrd >> (Pacemaker's node attribute daemon) is unable to talk to any other daemons. >> >> I'm not sure why this would happen, especially if the rest of the >> daemons do not have a problem talking to each other. But that's where >> you need to investigate. > > From my little experience it's a bad idea to route I/O traffic and cluster > communication over the same link: We had cases where cluster communication > (especially when using SCTP) showed errors when traffic was high. Maybe that > applies... > >> >> One thing I would say is that 1.1.8 is really old at this point, which >> means you're using the "legacy" attrd, which I'm not very familiar with. > > I agree: Even SLES11 SP4 uses old software, but it's at > "pacemaker-1.1.12-13.1" at least. Things _really_ got better with later > releases. >
I finally updated Pacemaker package ti the last version. Things are much more reactive and all my problems are gone. Thanks a lot for your advice. Just need now to propose some backport packages to my distribution :) > -- Anne Nicolas http://mageia.org _______________________________________________ Users mailing list: [email protected] http://clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
