Comment 1 for bug 1870013

If something is going away, I would certainly expect there could be a race condition between one 'juju run' and the next. From the sound of this, the issue is more that it happens within a single hook execution, eg:

juju run --unit=postgresql/0 'echo $(relation-ids db); network-get db --format=yaml -r db:2'

If it was something like a deferred event that fires in another hook, that is the sort of thing that I don't think Juju can give guarantees on (I would expect that if something is removed between hook A and hook B, you may get errors if you try to access it in hook B).

It is plausible that this is a case of Dying vs Dead, and 'relation-ids' is including relations that are in Dying but network-get is treating them as gone.

It is also fairly plausible that network-get isn't properly leveraging the hook context and it is then running into problems where at the start of the hook it exists, but at the end of the hook it is gone. We would want to give a stable view of the world during the execution of the hook.