Good Evening,
we're currently evaluating proxmox for a new deployment. I was unable to find extensive documentation about HA / Cluster implementation in 4.0, what I learned so far is that the only "standard" component the new stack is actually using is corosync for distribute messaging and membership and rgmanager has been dropped in favor of custom tools. So far so good, but I've got some questions:
1)Fencing: I read that the new cluster infrastructure uses watchdog based fencing. Does this means that in the event of a split brain, quorum-less nodes will shoot themselves ?
2)How does the cluster knows ? It simply waits for a timeout ? What if watchdog fails and machine keeps running?
3)If 1) is true, in case of a two-node cluster (for management purposes only, not for HA) what will happen? Does the fencing get activated only if HA is ?
I tried searching on the forum and on the wiki, but found nothing about how the cluster actually works.
S.
we're currently evaluating proxmox for a new deployment. I was unable to find extensive documentation about HA / Cluster implementation in 4.0, what I learned so far is that the only "standard" component the new stack is actually using is corosync for distribute messaging and membership and rgmanager has been dropped in favor of custom tools. So far so good, but I've got some questions:
1)Fencing: I read that the new cluster infrastructure uses watchdog based fencing. Does this means that in the event of a split brain, quorum-less nodes will shoot themselves ?
2)How does the cluster knows ? It simply waits for a timeout ? What if watchdog fails and machine keeps running?
3)If 1) is true, in case of a two-node cluster (for management purposes only, not for HA) what will happen? Does the fencing get activated only if HA is ?
I tried searching on the forum and on the wiki, but found nothing about how the cluster actually works.
S.
