Availability zone support

[Andrew Wilkins]

On Sat, Jun 14, 2014 at 4:38 PM, John Meinel <john at arbash-meinel.com> wrote:

> ...
>
>
>>  2) If you have 2 services, you likely would rather them on the same AZ
>>> rather than spread across different AZ because of the different cost of
>>> network bandwidth. How do we manage multiple services? Do we just use a
>>> strictly deterministic ordering? (always sort the AZ names, round robin
>>> starting with the first one?)
>>>
>>
>> It's non-deterministic. Pseudo-random choice of the least populous AZs
>> for the distribution group.
>>
>> As for cost: this is why we have the placement directives. We default to
>> spreading, but enable users to control placement if necessary.
>>
>
> So default behavior matters. I feel that sharing AZ across services by
> default would probably give a better user experience. (yes units of a given
> service need to be spread for proper HA, but between them it probably
> doesn't matter because if one is dead the group is dead.)
>

Round-robin on a sorted list of AZs sounds fine. I've created a backlog
card to do that, though it may be simple enough to do it while I'm fixing
the ec2 issue.


> I'm curious how the behavior is different from existing use. If we have
> always just had each service randomly distributed it probably doesn't
> matter.
>

AFAIK, it's not defined anywhere. The AWS docs just say "if you don't
specify, we'll choose one".
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju-dev/attachments/20140616/360e7b21/attachment.html>