Open Stack

Excerpts from Steven Hardy's message of 2015-01-16 01:11:44 -0800:
> On Fri, Jan 16, 2015 at 08:34:47AM +0000, Duan, Li-Gong (Gary at HPServers-Core-OE-PSC) wrote:
> >    Does Heat support executing an OpenStack operation, such as migrating an
> >    Nova instance, powering off a Nova instance?
> 
> We call state changes which don't affect the definition of the stack an
> "action", and we only support suspend and resume at present (e.g heat
> action-suspend <stack name>)
> 
> It may be possible to add support for more actions (until now nobody has
> asked for them), but note it only really makes sense to drive such actions
> via heat when dependencies/ordering are important.
> 
> For example, when suspending then resuming a stack containing a WebServer
> instance and a DatabaseServer instance, you want DatabaseServer to be
> resumed before WebServer (same order as on stack create)
> 
> So, it might make sense to have an action which can power-off a whole
> stack, turning off each nova node in the right order (you could write a
> little script which does the same thing quite easily though..).
> 
> It probably doesn't make much sense for heat to support things like
> migrating an instance, since it's an operation which isn't scoped to the
> stack and it's dependencies, it's likely an operator wants to migrate a
> workload off a specific compute node, which is something Heat has no
> visibility of at all.
> 
> >    I know currently Heat does a good job on launching cloud cluster or
> >    application, such as deploying a Nova instance with specified network
> >    configuration, but not sure how to control(not launch or delete) a Nova
> >    instance or cinder volume.
> 
> Right now, the easiest answer is write a little script which uses
> information from heat (e.g to get the UUID's for the resources you want to
> control) then e.g calls nova.
> 
> >    If Heat does not support these OpenStack operation, what is the best Heat
> >    way if we want to execute some operations, such as powering off a Nova
> >    instance, in Heat template?
> 
> As mentioned above, these lifecycle operations affect the stack state, but
> not it's definition, so it probably doesn't make sense to expose actions
> like powering off an instance in the heat template.

We have discussed in the past adding a target state as a property for
servers. Right now we only target ACTIVE, but we could certainly target
STOPPED and issue stops. It makes sense in certain instances.

That said, I think most of the time you want to use a workflow tool
(like Ansible or Mistral) to orchestrate actions. Heat can't possibly
know all the sequencing all services may require before and after a
server is stopped. So in these instances you either want the workflow
tool to drive Heat, or to have Heat defer to the workflow tool using
software-config/software-deployment.