get_endpoint is trigger-happy

[PietroPasotti]

Bug Description

When you use charm-tracing, you typically do:

class MyCharm(...):
    def __init__(...):
        self.tracing = TracingRequirer(endpoints=['otlp_http'])
    def otlp_http_endpoint(self):
        return self.tracing.get_endpoint('otlp_http')

self.tracing.get_endpoint raises an exception if the endpoint hasn't been requested (yet).

Since the requested endpoints are app-level, TracingRequirer only requests them if the owner is the leader and silently does nothing if otherwise. When you're leader, there is no problem because on __init__, the requirer lib will request the endpoint and so by the time otlp_http_endpoint is accessed, endpoints have certainly been requested (even if there is no response yet).

When you're not leader, there are two issues, one verified and reproducible, and one potential:

• if you write a unit test for the non-leader path, the charm_tracing lib will attempt to obtain otlp_http_endpoint and blow up.
• in a scenario when you deploy multiple units at once, it may well be that the follower units execute their first hooks before the leader unit has had a chance to request the protocols (or even be related to tempo!)

To Reproduce

n/a

Environment

n/a

Relevant log output

n/a

Additional context

No response