Notes on the rproxy program

Martin Pool


>>> HTTP accelerator mode

All we have to do here is treat the origin server as if it were our
upstream proxy: it can work out for itself how to cope with requests
with no hostname.

>>> Cache directory management

There's no need to lock the cache directory: it's meant to be run in
such a way that multiple processes can cooperate in using it.  At the
moment, all this means is that we do atomic-replace of new files
written into it.

>>> Direct connection

We're being a bit optimistic by sending the whole request -- HTTP/1.1
requires all servers to be able to handle full URIs even when not
proxying, but we may encounter older servers that don't cope.  Let's
see how we go. 

I tried this, and it turns out about half-and-half.  Current Apache
versions can cope, but anything worse tends to give a confusing
file-not-found message.

So, we will have to convert the URL to an absolute path if we're
making a direct connection.