Is it possible to prevent pcs/dlm/crm from fencing the node when it fails?
I use cluster only to be able to build GFS2 storage, and there's my own way of avoiding split-brain. So I do not need things like fence and dlm-locks.
I tried to config the on-fail operation, in CentOS7, in the following ways:
use pcs resource update dlm/clvmd/gfs2_res op monitor onfail=ignore
and edit the file /etc/corosync/dlm_cfg , edit the on-fail="ignore"
and reboot all the nodes.
After the reboot, I can check from pcs cluster cib, all the "fence" have become "ignore"
But when I kill the corosync in one node, the other still tries to fence it, as seen from the message log.
Mar 5 10:37:10 node0 stonith-ng[1733]: notice: remote_op_done: Operation reboot of node3 by node3 for stonith_admin.controld.24862@node3.a7be5018: No such device
Mar 5 10:37:10 node0 crmd[1739]: notice: tengine_stonith_notify: Peer node3 was not terminated (reboot) by node3 for node3: No such device (ref=a7be5018-4894-4de2-9273-8867d034d7af) by client stonith_admin.controld.24862
I just wonder if there's some way to totally avoid the fence? It just brings lots of trouble in my system.
