Comment 2 for bug 1394603

Running
     juju run --format json --all uname
will return json like this
   [{"MachineId":"0","Stdout":"Linux\n"},{"MachineId":"1","Stdout":"Linux\n"},{"MachineId":"2","Stdout":"Linux\n"}]

when there is an error, the machine's dict will contain keys for "ReturnCode" and "Stderr". We can check for ReturnCode, but in case it is changed in the future to be included when the value is zero, we might want to do
    for machine in response:
         if machine.get('ReturnCode', 0) != 0:
            raise Exception('juju run on machine %s returned %d: %s' (
                machine['MachineId'], machine['ReturnCode'], machine['Stderr'])

As we want to call run before and after the call to upgrade in the complex nesting of deploy_stack._deploy_stack(), we will want a helper function that calls env.juju('run', [ '--format', 'json', '--all', 'uname']), parses the output and checks the content. We can write a test for the helper function.