openstack team mailing list archive

[Marek Denis <marek@xxxxxxxxxxx>]

W dniu 02.04.2011 22:44, Jesse Andrews pisze:
> What to do when an instance dies is application specific.
>
> Some applications may not care, autoscaling back to the proper size by
> themselves.  Other applications may need the resources to be returned in
> the most recent state.
>
> Currently openstack requires the user to handle recovery.

Is there any API/modules/scripts for monitoring instances from the 
OpenStack level?
I am indeed looking for the possibility to have the simple HA from the 
OpenStack level. Suppose, monitor all the instances within single 
project, and when the instance dies, try automatically running the same 
instance on different server. The data is now lost, but at least it will 
serve it's functions (let assume this is a big calculator, that doesn't 
have to keep any session or user data). When the old instance is 
available again (maybe the server had some problems with the network) 
close the new instance down and release resources.

> I expect there will be an option to mark a virtual machine as "HA" which
> would attempt to relaunch on crashes.  How exactly to implement it will
> depend on the cloud requirements. If the instances are backed to a SAN
> or distributed filesystem (like ceph) the VM can be relaunched with all
> state that has been flushed to disk.  On a cloud with local disk only,
> the scheduled snapshotting could allow the vm to a recent state.

nova objectstore doesn't do any data replication? Or it does but only 
raw system images?

> Or HA could mean relaunch the image (ala autoscale)
Does OpenStack itself provide such features or admins of particular VM 
instances should install and configure HA apps?

> What use case are you thinking about?

I assume this is cloude, so when I (let's say I am customer) have my VM 
with some unix flavour system I don't care whether the server that hosts 
my VM is down or not. I should be able to connect to the  VM that's 
running in the cloud, not the specific hardware server, am I right?

--
MD