On Aug 4, 2016 1:36 AM, "Matthew Williams" <<a href="mailto:matthew.williams@canonical.com">matthew.williams@canonical.com</a>> wrote: > > Those are all good points - but they also seem like implementation details. For example if an action relied on coordination around the leader then the action should be written to protect against that - since it may still break if the user was to run juju run-action myapp/0 && juju run-action myapp/1. > > My question is - if I have a model with an application scaled to 10 units, and I know the action is safe to run all the units why do I have to call run-action 10 times, shouldn't I just be able to call it once on the application? Is there a genuine reason we're not supporting this? > > Cheers > > Matty Having it follow the logic of 'juju run' and just trigger the action on all units (who could then coordinate what should actually be run with eg an "is-leader" check seems reasonable). Otherwise the user has to run is-leader to figure out where it should be run, and there will be an inherent race that the leader could change before they run the action. Please raise a bug for this and we'll try to see how we can schedule the work. John =:->