How to set up poor man's HA

[GomoX]

Hey everyone,

I'd like to set up my Proxmox cluster so that in the event of a node failure I am able to restart dead guests on another node. I don't need this to happen automatically nor "ASAP". I just need to have a fail over mechanism that doesn't involve restoring from back up, so powering up manually when someone notices the service went away is perfectly OK.

¿Is this set up possible? Assume for this purpose that shared storage remains available (I can deal with that) and that the guest has its image in it. All I would need is a way to get ahold of the machine while it's offline and "move it" ("moving it" meaning just the configuration info) to the other node where it can be rebooted.

If I'm not mistaken this should be simple (at least in theory). Can it be done?

Thanks!

 

[dietmar]

Our distributed redundant configuration file system makes that simple:

node2# mv /etc/pve/nodes/<node1>/qemu-server/<vmid>.conf /etc/pve/nodes/<node2>/qemu-server/

Quite simple ;-)

 

[GomoX]

Oh, cool, can this be done through the GUI? Or does the node 1 need to be up for that?

 

[dietmar]

Oh, cool, can this be done through the GUI? Or does the node 1 need to be up for that?

No, that can't be done from GUI, because we want that users makes sure that node1 is 'down' when they do such things.

You should really make sure that node1 is not running, because else you run into the danger that the VM can be started on 2 nodes, which is usually dangerous.

 

[GomoX]

Oh, got it.

When the dead node comes back online, does it resync its DRBD before starting any VMs? Meaning, if the node is really down, there won't be an issue when it comes back up, right?

 

[dietmar]

Meaning, if the node is really down, there won't be an issue when it comes back up, right?

When the node comes back, it re-syncs the configuration file system /etc/pve/ before starting VMs. After that the node knows that you moved the VM and does not start it.

 

[GomoX]

Awesome, thanks for the explanations.