On 03/27/2013 09:23 AM, Oscar Koeroo wrote:
I've scanned the pull request but couldn't directly see/understand what
will happen if I'm using two different ports for HTTP traffic to the
same server in the same curl run.
As I understood the implementation, big disclaimer, the pipelining works
such that you setup the TCP/IP connection once and push multiple
requests over it. When you do this, the HTTP requests are selected per
hostname.
Hence my thought: shouldn't this be selected per hostname+port number
instead?
The connection reuse (and the pipelining) is more specific than just
hostname, so two different sessions on the same server but different
ports should work just fine.
However, the MAX_HOST_CONNECTIONS limit will apply to all connections to
the same host, regardless of port.
This also has the effect that the server_supports_pipeline flag applies
to the host and not the server instance on the host. Room for
improvement, I guess.
Linus
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
