[ic] SocketReadTimeout question

[John1]

On Tuesday, February 05, 2008 8:11 PM Grant wrote:

> Does anyone have any more info on why SocketReadTimeout came into
> being?  Is it designed to solve some broken pipe problems?
>
Yes, we noticed that the callback POST request from the banking systems' 3D 
secure authentication pages (credit card Verified by Visa and Mastercard 
Securecode authentication pages), was resulting in an Internal Server Error 
for some users.

After much head scratching a correlation with dial-up users and those behind 
web proxy servers was noted.  This lead to the discovery that 1 second (the 
default timeout) was never sufficient time for some dial-up connections. 
This included dial-up users from the following ISPs (though not exclusively, 
and possibly not all users):  Orange, Wanadoo, Freeserve, NTL dialup, and 
AOL.

The problem was replicated using an Orange dial-up connection, and upping 
the time limit to 5 seconds was found to be sufficient to solve the problem.

The problem appeared to be caused by an apparent time lapse in excess of 4 
seconds between the arrival of the POST request header and the eventual 
arrival of the POST request body.

Apache error log entry:
[Fri Jan 25 14:15:20 2008] [error] (32)Broken pipe: access to the_page.html 
failed for xxx.xxx.xxx.xxx, reason: error writing client block to 
Interchange

Interchange global error log entry:
- - - [25/January/2008:14:15:19 +0000] - - Died in server spawn: read: 
closed at /opt/interchange/lib/Vend/Server.pm line 657.

We have decided to run with a SocketReadTimeout of 10 seconds.