[openstack-dev] [nova] Hard fail if you try to rename an AZ with instances in it?

Jay Pipes jaypipes at gmail.com
Tue Mar 27 15:37:34 UTC 2018


On 03/27/2018 10:40 AM, Matt Riedemann wrote:
> Sylvain has had a spec up for awhile [1] about solving an old issue 
> where admins can rename an AZ (via host aggregate metadata changes) 
> while it has instances in it, which likely results in at least user 
> confusion, but probably other issues later if you try to move those 
> instances, e.g. the request spec probably points at the original AZ name 
> and if that's gone (renamed) the scheduler probably pukes (would need to 
> test this).
> 
> Anyway, I'm wondering if anyone relies on this behavior, or if they 
> consider it a bug that the API allows admins to do this? I tend to 
> consider this a bug in the API, and should just be fixed without a 
> microversion. In other words, you shouldn't have to opt out of broken 
> behavior using microversions.
> 
> [1] https://review.openstack.org/#/c/446446/

Yet another flaw in the "design" of availability zones being metadata 
key/values on nova host aggregates.

If we want to actually fix the issue once and for all, we need to make 
availability zones a real thing that has a permanent identifier (UUID) 
and store that permanent identifier in the instance (not the instance 
metadata).

Or we can continue to paper over major architectural weaknesses like this.

-jay



More information about the OpenStack-dev mailing list