HA oddity...

[christophe]

Hi all,

Testing HA is not trivial...

Context :
HP servers (DL 380G7), fencing devices : iLO3.
Proxmox v2.1 cluster.

Shortly :
- stop rgmanager. A VM managed by HA stops, then migrates, then restart. A live migration would be better, but OK.
- unplug Ethernet. A VM managed by HA migrates, then restart. OK.
- unplug power cord. A VM managed by HA STAYS on his node (cf clustat)!!! No migration, nothing...
But, when power is back, and while node is booting (BIOS screen visible), VM managed by HA migrates, and restart!!!
- idem when powering off the node via iLO3 (fence_ipmilan).

Is it a correct behavior?

Is iLO3 a valid fencing device in a proxmox cluster environment?

Thx,

Christophe.

 

[tom]

Hi all,

Testing HA is not trivial...

Context :
HP servers (DL 380G7), fencing devices : iLO3.
Proxmox v2.1 cluster.

Shortly :
- stop rgmanager. A VM managed by HA stops, then migrates, then restart. A live migration would be better, but OK.

if you do not want to stop the VM, why do you stop rgmanager? just live migrate to move it to another node.

- unplug Ethernet. A VM managed by HA migrates, then restart. OK.
- unplug power cord. A VM managed by HA STAYS on his node (cf clustat)!!! No migration, nothing...

if you have no fencing device (as you unplug the cord) it cannot work.

But, when power is back, and while node is booting (BIOS screen visible), VM managed by HA migrates, and restart!!!
- idem when powering off the node via iLO3 (fence_ipmilan).

Is it a correct behavior?

Is iLO3 a valid fencing device in a proxmox cluster environment?

Thx,

Christophe.

fencing is essential for reliable HA. if you want that fencing is working in all your tests consider a independent fencing device like APC power switches.

how many node do you have in your HA cluster?

 

[christophe]

if you do not want to stop the VM, why do you stop rgmanager? just live migrate to move it to another node.

Testing HA.

if you have no fencing device (as you unplug the cord) it cannot work.

Yes, now I can confirm that. I didn't knew before.

fencing is essential for reliable HA. if you want that fencing is working in all your tests consider a independent fencing device like APC power switches.

Yes, I will. It was so easy to use integrated iLO / iDRAC...

how many node do you have in your HA cluster?

Five nodes.

Thanks for your time and answer.

Christophe.