Re: [squid-users] Asyncronous squid ?

On Sun, 20 May 2001, Adrian de los Santos wrote:

> > The protocol that defines most Squid operations on Web objects
> > is HTTP (see RFC 2616). HTTP is connection oriented (like TCP), not
> > message oriented (like UDP). There are no unique request IDs that can
> > be shared with other caches/servers. An HTTP transaction is identified
> > by connection:sequential_id "pair" (sort of).
> >
> > Thus, in general, you cannot send a request using one
> > connection and receive a response on another connection. In theory,
> > you can fake source addresses of the connection from B to C so that C
> > thinks it is getting a request from A. DSR origin server load
> > balancers work that way, but you do not want to do that in your
> > environment.
>
> Why not ?

DSR load balancers use tricks that you cannot afford. For example,
load balancer and each origin server have the same IP address (among
other, different IP addresses). This makes traffic forwarding easy.
 
> Thats what i will like to do.. in fact i can put a couple of SNAT
> rules on B so the packets arrive to C as if they came from A...
>
> But The Squid process on the A machine dont expect any reply from
> C it expects a reply from B... (or since you tell me there its nothing
> on the TCP data which relates to the original connection, i mean there
> its no ID on the tcp data, its on the http packet itself) there its any
> workaround for this or im stuck ?

As far as I can tell, you are stuck unless you can make B appear as C
from A's point of view (that's what DSR load balancers do). And you
cannot do that trick without modifying Squid and relying on certain
non-general traffic topologies.

Perhaps you should re-think your design based on what exactly you want
to achieve rather than a particular method of doing that.

Alex.
Received on Mon May 21 2001 - 17:22:58 MDT