[sheepdog-users] (no subject)

[Valerio Pachera]

2013/5/28 xiao li <nudt.xiaoli at gmail.com>:
> and my VM will die when the node
> (10.2.7.12) is switched off.
...
> why do not the other nodes  take over the VM?

Sheepdog handles the storage redundancy "only".

To achieve something like you ask about, you need HA on the qemu processes.
In other words, something that notice when a qemu proccess is missing
and runs it on another node.
This requires obviously a shared storage. Sheepdog acts like it.
The vm is anyway not going to be in the same state when the node has die.
It will boot normally.

To be able to preserve the state of the vm even if the host dies, you
need processes redundancy:
the same vm running on two hosts at the same time.
This is something I only heard speaking about.
It requires the same (and expensive hardware) on both hosts, and I
don't know even it realy works.

Shortly, when a host dies on a sheepdog cluster, simply restart its
guests on the other hosts.
Mind the cache.
Better if you don't use to avoid data loss in such case.

If you simply plan to shutdown the host, you can live migrate the guest before.
This way guest will continue to run with no service interruption.