www-modproxy-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Graham Leggett <minf...@sharp.fm>
Subject Re: proxy-http problem (was 2.0.28 -- dumping core)
Date Fri, 16 Nov 2001 21:27:59 GMT
Ian Holsman wrote:

> I've had a better look into this problem and I think I figured what is
> causing it.
> We store state info in the connection record which holds the connection
> handle and other stuff.

This is correct - it's done so keepalive can work.

> The problem seems to occur when we request a page which has virtual
> includes which request another page on the same server.
> what happens is the 2nd request grabs the connection (which is still
> midway through the 1st request) and issues another request on the same
> connection.
> now this may be fine for things which understand pipelining. but
> I think it is screwing up 1.0 servers.

Even with pipelining it won't work - we're in the middle of a request
trying to make a new request, not at the end of the request :(

> the only way I can think of getting around this is to store state at the
> request level instead of the connection level.

This will break keepalives.

Rather fix it so that when proxy tries to write state to the connection
record, it checks first whether a state already exists. If a state does
exist then we have triggered our bug - get proxy to *not* write the
state and just serve a "Connection: close" on the subrequest.

The consequences of this are that keepalives cannot be supported for
subrequests within requests, however considering that practically a
page-within-a-page cannot logically follow the original page request
this should be fine.

minfrin@sharp.fm		"There's a moon
					over Bourbon Street
View raw message