Thursday evening handover, 7/6/2012

[Dave Cheney]

> The reason that Environ.Instances is the way it is currently is that you aren't
> guaranteed to get all machines even if you have just started them,
> due to the eventual consistency semantics of ec2. (The original semantics
> of Instances was as you describe).
> 
> One possibility would be to interpret a nil slice argument to Instances
> as asking for all instances.

That would be good enough™. 

> Out of interest, when do you see this situation (extra instances not
> mentioned in the state) happening in practice?

The problem I'm trying to solve is.

Provisioning Agent is running

Provisioning Agent restarts as it looses connection to the state (this may actually be several seconds once we add in the retry delay). 

At the same time, someone issues juju remove-machine, removing it from the topology.

Without a way for the provisioning agent to say, Environ, please tell me your best guess at which instances you have running, the PA will never be able to action the removal of that instance.

Hopefully running the 'tell me which instances are running' action periodically (like the current pythonic PA does) will get close enough to consistency.

Cheers

Dave