PVE Cluster fencing with IPMI

[stefws]

Wondering if there are good info around on how to setup IPMI fencing f.ex. with iLom/IPMI on HP Proliant boxes to do power fencing for HA clustering?

I have bought and read the Proxmox High Availability book from PackT Publishing, but that merely mentions IPMI as an option and then in details talks of snmp network switch fencing.

 

[adamb]

Wondering if there are good info around on how to setup IPMI fencing f.ex. with iLom/IPMI on HP Proliant boxes to do power fencing for HA clustering?

I have bought and read the Proxmox High Availability book from PackT Publishing, but that merely mentions IPMI as an option and then in details talks of snmp network switch fencing.

Its pretty straight forward. Configure the iLO/ipmi port with an IP/user/password. Then configure fencing within cluster.conf.

https://pve.proxmox.com/wiki/Fencing

If you need a more detailed example of cluster.conf, I can provide that to.

Keep in mind ipmi/iLO based fencing isn't perfect. If the server was to loose complete power, or both power supplies, the fence device would not be accessible. This doesn't happen often but its good to be aware about. We use a APC PDU which allows us to fence at the power level, then use ipmi as a secondary fence. Good luck!

 

[stefws]

Right, thanks! No Wiki URL shows good examples.

U mean if a server reboots after a power failure the cluster doesn't have control of such a restarted server?
But surely another cluster node should have grassed the VM ie. unassigned VMs from restarting node and thus they shouldn't be attempted started twice or...

 

[adamb]

Right, thanks! No Wiki URL shows good examples.

U mean if a server reboots after a power failure the cluster doesn't have control of such a restarted server?
But surely another cluster node should have grassed the VM ie. unassigned VMs from restarting node and thus they shouldn't be attempted started twice or...

A VM will not migrate until a successful fence action (so if the fence fails, no migration will take place). If a server goes down do to a power loss, and a fence action takes place during the power loss, nothing would failover because the fence action would fail. That wiki provides a good example for ipmi, they are't perfect but they are more than enough to get you going. Im not sure what you mean it shows no good example, it has a section for ipmi, follow that.

 

[stefws]

Okay makes sense then, thanks for high lighting!

PS! I said and mean it do shows good examples

 

[adamb]

Okay makes sense then, thanks for high lighting!

PS! I said and mean it do shows good examples

Roger, if you get stuck I can provide some other examples. Just remember, in order for a failover to take place, a successful fence action must be performed first.