More on the Akamai weirdness

Hi all,

I've been doing some sniffing and tracing on the Akamai issue, but have come no closer to understanding what the problem is.

The problem does in fact happen with both prefork and threaded MPMs, so threadsafe issues don't seem to be the problem.

The problem is very weird. On a non-Akamai site, an strace shows the process accept the connection, then send the headers, the receive the headers, then receive the response, writing it away to the browser.

On an Akamai site, the proxy suddenly hangs on one of the network reads of the response. It sometimes happens right after the headers have been read, sometimes after some of the body has been read. There is no segfault reported, and sniffing the network shows that the traffic from Akamai suddenly stops dead in it's tracks, probably because incoming data is not being acknowledged. Usually this happens all at once - as if all the processes have broken at once. (This is seen with Netscape v4.7, which does not support HTTP/1.1 keepalives and therefore fetches 4 http objects simultaneously).

I need to get some sleep at this point - if anyone else can look at this problem and get a fresh perspective on it it would be great...

Regards,
Graham

-- 
-----------------------------------------
minfrin@sharp.fm		"There's a moon
					over Bourbon Street
						tonight..."