[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[Linux-cluster] Cluster "averts" fencing


I have a four node cluster using fence-scsi and when I purposely fail any node but the primary (node 1) it seems to always “avert” fencing.  If I fail the primary node, the cluster correctly fences the node.


From node1’s ‘group_tool  dump fence’ when I fail node4:

1255903014 start default 103 members 2 3 1 
1255903014 do_recovery stop 98 start 103 finish 98
1255903014 add node 4 to list 1
1255903014 averting fence of node
1255903014 finish default 103


The node doesn’t get fenced and it retains it’s scsi registrations.


From the source: http://git.fedorahosted.org/git/fence.git?p=fence.git;a=blob;f=fence/fenced/recover.c


It looks like the conditions for failing are:
cpg_member = is_clean_daemon_member(node->nodeid);
ext = is_fenced_external(fd, node->nodeid);
if ((cluster_member && cpg_member) || ext) {
log_debug("averting fence of node %s "
  "cluster member %d cpg member %d external %d",
  node->name, cluster_member, cpg_member, ext);


I don’t think either  “is_clean_daemon_member” or “is_fenced”external” should be true.   Fenced isn’t started as a clean daemon and is_fenced_external (I believe) means that the node is fenced externally by another fenced daemon which shouldn’t be true either.


Any ideas what could be going on here?  Help or suggestions would be appreciated!



David Challoner


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]