juju run-action --wait return code should (optionally) reflect the success/failure of the action

run-action currently creates an "operation" which is the parent for the individual tasks, each task being an invocation of an action on a unit.

We could special case the situation where there's only one target and hence one task.

I think that it would make sense to exit 0 if all the subtasks exited 0, and pick one of the failure codes in some consistent way when one or more actions in a batch failed.

I think the use case here is "automatically proceed with this thing I want to do if everything is okay; otherwise, stop, and tell me what broke so that I can intervene." In that use case, I think that we're okay proceeding if everything worked, and exiting with a non zero int if it didn't.

This is covered by roadmap work this cycle. Adding to 3.0.0 milestone.

In my ceph osd conversion task for one of our customer, I used juju run-action in my script.
I have `set -xue` at the top of script, and was expecting that if any cmd failed, exit and don't run next ones, which is important for ceph/osd/bacahe/disk operations.
However, that didn't work as expected, and I figured out that juju run-action still return 0 even the action failed, which was a surprise to me.

So I strongly agree with the issue mentioned in this case, juju run-action should care about the actual retcode of underneath task and let the caller know.

I will move this to the wishlist.