We implemented multiple working HA-Scenarios, and we do it in every of our proxmox advanced trainings. One productive used cluster now works for over an year without problems concerning HA at a telcommunication service provider's data center.
Proxmox High Availability (HA) cluster - is it working at all for anyone?
- Thread starter newacecorp
- Start date
We implemented multiple working HA-Scenarios, and we do it in every of our proxmox advanced trainings. One productive used cluster now works for over an year without problems concerning HA at a telcommunication service provider's data center.
That's seem to be ok for me to use DRAC as fencing device, if you have dual power supply, on 2 electrical circuits
What happens if you loose power to an APC PDU or fence capable switch? Exactly the same thing. This is not an argument for not using DRAC.
I discovered the problem with our cluster setup. Actually, there were two problems. First, the DRAC cards had the MTU set to 1500. Problem is that I have the internal network configured to use an MTU of 9000 to improve throughput with the storage SAN (NFS). This explains the hit-or-miss with fencing of the nodes. Second problem, one of the physical nodes was still using an MTU of 1500. This is what caused the cluster instabilities after one of the nodes was shutdown.
I've configured everything for an MTU of 1500 for the moment and have been pleasantly surprised that the node is fenced (turned off) when I yank the network on it. Also, the HA VMs are migrated to one of the remaining active nodes. I haven't figured out what the criteria is for a node to receive a particular HA VM, but it does work. I should mention that the node is restarted on another node, not live migrated. This makes some sense as this would require all nodes to have live copies of the active state (memory) of the HA VM, which I presume isn't (yet) possible with QEMU.
I've configured everything for an MTU of 1500 for the moment and have been pleasantly surprised that the node is fenced (turned off) when I yank the network on it. Also, the HA VMs are migrated to one of the remaining active nodes. I haven't figured out what the criteria is for a node to receive a particular HA VM, but it does work. I should mention that the node is restarted on another node, not live migrated. This makes some sense as this would require all nodes to have live copies of the active state (memory) of the HA VM, which I presume isn't (yet) possible with QEMU.
"I haven't figured out what the criteria is for a node to receive a particular HA VM" There is no criteria, except that the destination must be running;-)
I've been testing Proxmox 3.4 with pve-no-subscription repository. Does anyone know what happens with the installed modules when we switch to the enterprise (subscription) repository? Are these components immediately replaced when an "aptitude full-upgrade" is done the first time?
Enterprise repository does not contain newer packages, so nothing is replaced first time.
They are some work on qemu actually about it
http://wiki.qemu.org/Features/COLO
So, maybe it'll be ready for 2015
