Hi Sean, thanks for the code review!

Would GETINFO/GETCONF cache hit logging[1] benefit from log_once? Or even something between log and log_once (e.g. rate_limited_log)?

Interesting thought, but I don't think so. When I look for cache hits it's usually to answer the question of 'did doing activity X result in requesting parameter Y'. Having stem log once or drop logging when it exceeds a certain rate would make that more confusing.

Is there any particular reason for moving away from readthedocs.org? I'm curious if there was more than a desire to self-host as much as possible.

I talked a bit about the reason for moving on 'https://trac.torproject.org/7324'. Mostly the reasons were (in order by how much I cared about them)...

1. A shorter and more official sounding tpo domain.

2. The daily autobuilds on read-the-docs were always stuck, so the documentation was only updated when I manually built it. This meant that we usually had stale docs. Now that we're on our own hardware I also bumped the rate of checking for new content so the site updates within five minutes of a push.

3. Self-hosting provides better security and visitor privacy. I didn't experiment to see how good the sandboxing of read-the-docs builds were, but a community system that runs arbitrary code is a little spooky.

The heartbeat time tracking is a good idea[3]. But, I wonder, should this bother with an accessor method? Could this work as just well as an (non-callable) attribute?

I'd be fine with that. The only advantage I can think for having it as a method is if we want custom handling in certain situations (for instance, returning None if we're disconnected). However, we don't presently do that so I could go either way.