Tracking open/close connections efficiently?

[cnicodeme]

Hi!

(This is not an issue, sorry if it's not i the right place)

I'm wondering if there is a way to track when a user is connecting, but also when disconnecting, with guarantee about the disconnect (connection closed, lost, exception, error, etc).

The idea is to implement a counter to track the number of active connections at a specific time to limit the abusers.

Thanks in advance :)

[waynew]

We didn't have discussions enabled before, but I figure they would be a good choice for our project... and this issue is perfect for a discussion 😂

---

I don't know for sure if the functionality is built-in in a friendly way, but I know that I'm tracking it via session.host_name

I'm pretty sure that you can get the client connection from somewhere. You could check the aiosmtpd code to see how the hostname gets passed - if that's coming directly from the client socket then it's at least as trustworthy as any other mechanism. But if you have a proxy in front of your server then it might report the proxy as the client, rather than the ultimate end client, so you may need to do some poking around to figure it out. But generally speaking I think that the session object should, ya know, contain each session.

HTH!

[cnicodeme]

Wonderful :D !

I can easily track the "connections" part with the handle_HELO / handle_EHLO part, but it's more related to the disconnect where I'm having troubles.

Not every server will do a proper QUIT, and even if they would, any arising exceptions on my end could close the connection prematurely and cause the counter to not decrease.

One thing I thought about was overriding the connection_lost() from smtpd.py, but I'm not entirely sure it would be called every time with certainty. Do you know if that is the case ?

You mention tracking the session object. I agree that is an interesting idea, but the same remain : how to track that the session object is "closed" (set to none) ?

Another way, maybe (again, not sure), would be to somehow track the self._handler_coroutine event and add a callback on cancel?

[cnicodeme]

Trying to find the best solution, I came up with something that seems robust: I add a "done callback" to the handler_coroutine :

server._handler_coroutine.add_done_callback(some_func)

From my tests, it seems to handle all the cases (normal quit, disconnect from server, from client, exception, hard close).

Of course, I'm open to a better and more robust solution if any !

[pepoluan]

Seems to be a good feature to be added publicly (no leading underscore) in 1.5.0 :)