[Haizea] Problems with VM lifecycle on OpenNebula+Haizea

[Borja Sotomayor]

Hi,

> PROBLEM 1
> --------------
> When we deploy a VM  with "onevm submit <path to the template>" and then
> during its work shut it down or delete it, Haizea is not aware of this
> change. The lease for that machine remains active and may prevent other VM
> deployments.

My reply to your previous message on the Haizea mailing list should
address this problem. For those on the OpenNebula list, this is the
message I'm referring to:

https://mailman.cs.uchicago.edu/pipermail/haizea/2009-November/000084.html


> PROBLEM 2
> --------------
> When we deploy a machine as described above, and then do "stop" on it, via
> XML-RPC (we have our own web GUI application), it gets saved. When we try to
> resume that machine, it stays stuck in state "pending" and there is no
> action that can revive it.

How exactly are you trying to resume it? If you're using "onevm
resume" and the machine still stays in a "pending" state, that sounds
like an OpenNebula-only problem. For the OpenNebula developers to help
out, I suggest you send this as a separate message and include the
sequence of commands you used and the VM log (in
$ONE_LOCATION/var/$VM_ID/vm.log)


> But
> take a look at Amazon, they will allow you to deploy a machine, run it at
> certain times, turn it off, turn it on again and so forth. How does that fit
> into Haizea's view of the world?

In Haizea's view of the world, these would simply be separate leases,
not a single lease that gets turned on and off. Reusing an existing
image (instead of starting from a fresh one every time) should be
possible, as long the OpenNebula template in subsequent submits refers
to the saved copy of the image (although I've never tried this with
OpenNebula+Haizea). This is not to say that it wouldn't be more
natural to have a model where a lease can be "restarted" in the
future, without having to specify that a saved disk image must be used
(instead, this would be implicit in the "restart" operation). However,
doing this is not currently on our roadmap.

Cheers!
--Borja