What are some examples of other database drivers that have adopted these logging conventions?

@sfackler I did some digging, here is a list at theOTEL registry.

A specific example in code in the first item of using the OTEL semantic conventions,go-pg.

This looks fantastic! I was actually just looking at implementing this exact same thing, and you've done it better than I would have. I will apply this patch and try it out in production for an app that I work on.

This works great, I ended up making two changes:

1. I set all of the spans to INFO because that is the level that we send to Honeycomb. I think at the very least prepare should be bumped to DEBUG, as trace is almost always too noisy to be useful and prepare spans are useful to me (I try to have a span for every external network call).
2. I added a span for connect, which delivered value as soon as I deployed it. Feel free to pull in this commit and/or modify it. I have already found the sub-span timings that I added to be useful as well.titanous@7e2e3a2

This works great, I ended up making two changes:

1. I set all of the spans to INFO because that is the level that we send to Honeycomb. I think at the very least prepare should be bumped to DEBUG, as trace is almost always too noisy to be useful and prepare spans are useful to me (I try to have a span for every external network call).
2. I added a span for connect, which delivered value as soon as I deployed it. Feel free to pull in this commit and/or modify it. I have already found the sub-span timings that I added to be useful as well.titanous@7e2e3a2

I merged in that commit and bumped all the trace levels as well. Thanks!

From what I can tell most or all of the libraries at that OTEL reference are external libraires that you can use with a database driver to apply tracing conventions.

I am hesitant to land this for a couple of reasons:

1. People can be using this library as a tracing appender, and this change can potentially cause some undesired infinite logging recursion. cfFeature Request: Optionally disable use of the log crate #215
2. This seems like a convention that your particular organization wants to use, e.g. "I set all of the spans to INFO because that is the level that we send to Honeycomb", not something that is uniformly applicable.

I think you'd probably be better served by following what those other libraries do and making a separate library that can wrap tokio-postgres and apply the appropriate tracing logic.

I think a separate library would be fine, though it seems like a hook system would be necessary to ergonomically log everything we have in this PR. For example, there is currently no good way for a separate library to properly trace connection attempts with all of the relevant metadata, especially with timings for each phase.

That approach seems like it could work in principle - I'd be interested in what the hook trait would look like.