Matt Robenolt

> When you say client, you are specifically speaking of any arbitrary client making the HTTP request, correct? No, we mean instances of the `Raven` client.

That's fine with me. Like I said, was just curious if there was interest in having support for this in core. :) Thanks for the attention @sigmavirus24!

From my perspective, people are more frequently doing this by writing an adapter for `requests`, which, is obviously using urllib3. I personally prefer to just use urllib3 when I can....

And lastly for reference, we are going to be shipping this and adopting it internally for a while in production. https://github.com/getsentry/sentry/pull/10295

Maybe I’m missing some context here. But if I initialize the connection pool explicitly, where does URI parsing come into play? The usage here is something like: ```python UnixHTTPConnectionPool("/var/run/something.sock").urlopen("GET", "/")...

Oh, I see. Yeah, if we wanted this, we'd have to invent some URI pattern like: `unix:/var/run/docker.sock/foo` or something. I have no real interest in doing this, and I don't...

PoolManager only supports http and https. There are a few other connectionpool types under urllib3.contrib that all exist without going through PoolManager. I don’t think this would be a barrier...

Either way, I think working with PoolManager would be nice, but agree that opens up a can of worms for URIs.

Re: PoolManager If we did want to support some URI scheme, nginx seems to have a solution that works: http://nginx.org/en/docs/http/ngx_http_proxy_module.html#proxy_pass ``` http://unix:/foo.sock:/some/path ``` It's not pretty, but it works. I...

+1 to inventing a new method. :)