Don't throw on send-only endpoints

[timbussmann]

Based on Particular/docs.particular.net#4264, is there a specific reason the metrics plugin throws at startup for send-only endpoints?

I understand that as of now, the captured metrics aren't that interesting for send-only endpoints however, throwing exceptions seems to be an unexpected and unusual behavior at fist glance.

[mikeminutillo]

This is actually the root Metrics Plugin that does this. It was assumed to be a mistake to turn Metrics on for an endpoint that literally will report no metrics. If there's a good business reason to have this switched on then we can change it pretty easily.

[timbussmann]

It was assumed to be a mistake to turn Metrics on for an endpoint that literally will report no metrics.

I'm not sure this is safe assumption since there is a good chance that users will reuse their configuration across endpoints, enabling monitoring by default. I think it would be better to not crash the endpoint but rather log a warning instead.

[mikeminutillo]

@Particular/metrics-maintainers thoughts on this?

And can you move it to the Metrics Plugin repo?

[dvdstelt]

I'm not sure this is safe assumption since there is a good chance that users will reuse their configuration across endpoints, enabling monitoring by default. I think it would be better to not crash the endpoint but rather log a warning instead.

Which users will not read and then contact us why they're not getting metrics on their sendonly endpoint.
If they reuse config across endpoints, they should make it so that they can alter the config before the endpoint is initialized/started.
If I can remember correctly this was thought through and the decision was to throw, to clearly notify users that this is not possible and at the same time reduce the number of support tickets.

[timbussmann]

still feels that behavior is too much driven by the actual metrics currently available. What if we add a metric which makes sense on send-only endpoints?
How many other features which don't make sense in a specific context currently throw an exception at startup?

[tmasternak]

I also think that we should consider the principle of least surprise. With the current feature set if we enabled monitoring for send-only endpoints we would most likely have to add some explanation in SP to indicate that there are no metrics for send-only endpoints and this is fine.

In other words, enabling metrics for send-only endpoints is not only a plugin change.

[Scooletz]

@tmasternak Do you mean something like showing an information "this is a send-only endpoint" or something similar?

I do agree, that it's not only a plugin change.