Cool, I think you all understand the concerns here:

1. We can't treat the InstanceID as a ReservationID since they do two different 
things. InstanceID's are unique per instance and ReservationID's might span N 
instances. I don't like the idea of overloading these concepts. How is the 
caller supposed to know if they're getting back a ReservationID or an 
InstanceID? How to they ask for updates for each (one returns a single value, 
one returns a list?).

2. We need to handle "provision N instances" so the scheduler can effectively 
load balance the requests by looking at the current state of the system in a 
single view. Concurrent single-shot requests would be picked up by many 
different schedulers in many different zones and give an erratic distribution.

3. As Soren pointed out, we may want certain semantics around failure such as 
"all or nothing"

4. Other Nova users have mentioned a desire for instance requests such as "has 
GPU, is in North America and has a blue sticker on the box". If we try to do 
that with Flavors we need to clutter the Flavor table with 
most-common-denominator fields. We can handle this now with Zone/Host 
Capabilities and not have to extend the table at all. If you look at 
nova/tests/scheduler/test_host_filter.py you'll see an example of this in 
action. To Soren's point about "losing the ability to rely on a fixed set of 
topics in the message queue for doing scheduling" this is not the case, there 
are no new topics introduced. Instead there are simply extra arguments passed 
into the run_instance() method of the scheduler that understands these more 
complex instance requests.

That said, I was thinking of adding a POST /zone/server command to support 
these extended operations. It wouldn't affect anything currently in place and 
makes it clear that this is a zone-specific operation. Existing EC2 and core OS 
API operations are performed as usual.

Likewise, we need a way to query the results of a Reservation ID request 
without busting GET /servers/detail ... perhaps GET /zones/servers could do 
that?

The downside is that now we have two ways to create an instance that needs to 
be tested, etc.

-S


Confidentiality Notice: This e-mail message (including any attached or
embedded documents) is intended for the exclusive and confidential use of the
individual or entity to which this message is addressed, and unless otherwise
expressly indicated, is confidential and privileged information of Rackspace.
Any dissemination, distribution or copying of the enclosed material is 
prohibited.
If you receive this transmission in error, please notify us immediately by 
e-mail
at [email protected], and delete the original message.
Your cooperation is appreciated.

_______________________________________________
Mailing list: https://launchpad.net/~openstack
Post to     : [email protected]
Unsubscribe : https://launchpad.net/~openstack
More help   : https://help.launchpad.net/ListHelp

Reply via email to