Some of the Prometheus metrics seem to be unused

[reivilibre]

Hi there, me again :).

I decided to set up Prometheus monitoring because I was hoping to get alerted when/if I stop receiving e-mails frequently (as that has usually shown to be a problem in the past).

I was quite confused by two points:

• the maddy_* prometheus metrics don't seem to appear in the /metrics output until they actually get incremented (maybe this is intentional? But as a Prometheus noob I would have liked to know on the documentation page before wondering if I missed something)
• some of the metrics seem to be unused — for example completedSMTPTransactions which doesn't seem to have a corresponding increment line

❯ rg completedSMTPTransactions
internal/endpoint/smtp/metrics.go
33:     completedSMTPTransactions = prometheus.NewCounterVec(
83:     prometheus.MustRegister(completedSMTPTransactions)

I can have a crack at these problems if you don't think I'm wrong.

I would also like to add some more metrics — for example on the number of e-mails that get successfully sent over LMTP (perhaps excluding ones received from local users ... hmm), so that I can have a very clear number of 'received e-mails' that actually make it to Dovecot rather than the queues getting deadlocked (which seems to have bitten me every few months on older versions). Does that sound reasonable?

[foxcpp]

the maddy_* prometheus metrics don't seem to appear in the /metrics output until they actually get incremented (maybe this is intentional? But as a Prometheus noob I would have liked to know on the documentation page before wondering if I missed something)

Metrics are parametrized by module name so before sending actually happens we don't know what to put in the module name.

[foxcpp]

I can have a crack at these problems if you don't think I'm wrong.

Oh, I must have missed completedTransactions.

excluding ones received from local users

I guess a bunch of metrics covering outbound delivery would be definitely useful. (ref #264)

rather than the queues getting deadlocked

maddy_queue_length probably can be used for this but I am open to adding more metrics that you think would be useful.