connection reset mystery

My fellow squid'ers,

I am trying to track down/fix a subtle problem I'm having - I'm
wondering if anyone out there has any suggestions. A configuration
setting I might tweak? A way to figure out what is going wrong?

THE PROBLEM:

 Clients with low bandwitdh sporadically get this message:

 "The connection with the server was reset"

 when POST'ing to one particular cgi script on our site.

THE NAUSEATING DETAILS:

 I'm running Version 2.2.STABLE5 on Debian GNU/Linux potato
 (x86). Squid is running in acceleration mode. The box it runs on
 sits on a firewall, pulling content from a cluster of identical web
 servers on the inside network. I've hardly touched the default squid
 configuration at all - I've just modified the port it runs on, set
 the appropriate httpd_accel_host, httpd_accel_port values, and opened
 it up to outside access.

 I tried playing with the various timeout settings, making them even
 more generous. But this seemed to have no effect, and I set them
 back to the way they were. I also increased the request_size to twice
 as much as I'd ever expect in a POST.

 The cgi is a kind of "save file" script. The greater the size of the
 file being POSTed, the higher the chance of the connection reset
 problem. The save cgi _always_ get called, the file always gets saved
 (the data gets posted fine), but the html output of the cgi never
 makes it back to the client, disturbing the user experience. POSTs
 to all the other cgi scripts work fine, but a comparably tiny amount
 of data is sent.
 
 The squid and apache log files _always_ record a normal (200) status for
 the call to the save cgi.

 And squid does a beautiful job of getting everything else into and
 out of the site. I'm rather smitten.

 I am confident this is a squid related problem, because if I send the
 POST directly to an apache server, it always works. This is what I
 am doing as a stopgap measure.

thank you for any insights,

/iain lamb, halfbrain.com
Received on Tue Nov 30 1999 - 13:44:01 MST