On 07/24/2012 12:42 PM, Jason Guiditta wrote:
On 24/07/12 15:28 +0200, Jan Provaznik wrote:
- Show a deployment and its instances
Each instance is listed with name and a link to the real object 3. Index of deployments 4. Delete a deployment 5. Edit a deployment - not available
What about other deployment actions? 'stop', 'start' (once we can re-start stopped ones), 'reboot', etc.
My thought on this was that they should all just be treated as updates, iow, PUT the desired changes/fields in the request to initiate the desired action. Having a separate methods doesnt feel like it fits into REST properly to me. If there is disagreement here, I think this is a good place to iron it out, as the pattern needs to be established and made consistent now, before things get even more complex (like adding other layers of api, etc).
Hmm. The thing is it's not really an update. To 'stop' an instance isn't an update to the 'state' -- the state won't change until the action is done. We're explicitly requesting that an action occur on an instance ("please stop this now", "please start this tomorrow morning"), which, REST or not, doesn't feel like an update -- it's a request to do something that might (eventually) result in an update. Surely there's a REST-appropriate way of doing this that doesn' obscure the fact that it's not really an update.
Scott
There might be a (nested) resource for these actions, e.g. instanceaction. This resource would have only create action:
So requests would look like this: POST instance/id/instanceaction?type=restart POST instance/id/instanceaction?type=start POST instance/id/instanceaction?type=stop
Jan
Yes, I think something more along these lines makes sense, I was thinking something a little more generic, but we can work through impl details as needed. My thought was something like curl -X POST --header "Accept: application/xml" --header \ "Content-Type: application/xml" http://localhost:3000/action --data \ "<action> <target_type>instance</> <target_id>1</> <perform>start</> </action>"
In this way, you can initiate an action on any object that supports actions, and the system should perform the desired work. Of course, this also implies that the user can request a list of valid actions for a given object type, but that should be pretty easily accomplished with something like: curl --header "Accept: application/xml" http://localhost:3000/action \ --data-urlencode "target_type=instance"
possibly with a parameter like 'valid=options' or something? Scott, do these 2 variations of the same idea make more sense to you?
-j
I think this makes sense, yes. I know at one point we had a model method that returned a list of valid actions which, in a much earlier iteration, the UI actually used. As we're improving the UI again (and adding more valid actions now), we'll need this 'valid action list' in the UI at some point as well, so having it in the API here makes some sense.
Scott