14.07.2020 13:19, Rohit Saini пишет: > Also, " Keep in mind that neither qdevice nor booth is "replacement" for > stonith. " > > Why not? qdevice/booth are handling the split-brain scenario, keeping one > master only even in case of local/geo network disjoints. Can you please > clarify more on this. >
I cannot comment on booth as I do not have experience with pacemaker geo clusters, but for qnetd ... - from general point of view solution that depends on quorum relies on well behaving nodes - that nodes that lost quorum do what they are expected to do. Solution that depends on fencing actively ensures that nodes are in defined state where they cannot conflict with remaining nodes. It is simply more robust. - you can partially mitigate it using watchdog, but then it is trivially to lose your cluster - qnetd is not available for whatever reasons (may be temporary network outage) and you need to reboot one node - and second node immediately kills itself. There is no integration between various parts, pacemaker does not tell SBD that another node cleanly left cluster and there is no split brain. For me it is unacceptable. _______________________________________________ Manage your subscription: https://lists.clusterlabs.org/mailman/listinfo/users ClusterLabs home: https://www.clusterlabs.org/