Received: (qmail 21426 invoked from network); 23 Feb 1999 00:24:29 -0000
Message-Id: <36D1F3D1.E9F15B93@uswest.com>
Date: Mon, 22 Feb 1999 16:18:25 -0800
From: Derek Andree <dxandr2@uswest.com>
Sender: dxandr2@uswest.com
To: apbugs@hyperreal.org
Cc: apbugs@Apache.Org
Subject: Servers hang on either Keepalive or on Read (if keepalive is turned off)

>Number:         3948
>Category:       pending
>Synopsis:       Servers hang on either Keepalive or on Read (if keepalive is turned off)
>Confidential:   yes
>Severity:       serious
>Priority:       medium
>Responsible:    gnats-admin
>State:          closed
>Class:          duplicate
>Submitter-Id:   unknown
>Arrival-Date:   Mon Feb 22 16:30:02 PST 1999
>Last-Modified:  Tue Apr 20 16:17:11 PDT 1999
>Originator:     
>Organization:
>Release:        
>Environment:
>Description:
>How-To-Repeat:
>Fix:
>Audit-Trail:
State-Changed-From-To: open-closed
State-Changed-By: dgaudet
State-Changed-When: Tue Apr 20 16:17:10 PDT 1999
State-Changed-Why:
[This is a standard response.]
This issue has been reported before;
please search the FAQ and the bug database.
Thanks for using Apache!
Class-Changed-From-To: sw-bug-duplicate
Class-Changed-By: dgaudet
Class-Changed-When: Tue Apr 20 16:17:10 PDT 1999
>Unformatted:
I have found a few other interesting items here (as well as other
reports of the same problem, none with a solution yet.)

I have turned off keepalive.  I can telnet to port 80 and let the
connection hang (not type anything), and it will... indefinitely.  On
another (working, Linux) machine, the connection is appropriately closed
upon reaching the timeout of 300 seconds.  For some reason, Solaris
appears to be ignoring the timeout value.  This is what I believe to be
the problem, and it appears to be replicable at least on my Sun machine
via telenet session.

My theory is that connections are made to httpd, yet they get hosed (net
congestion, browser timeout, etc) before the client can make a request
(like GET /).  Solaris/Apache will then just listen indefinitely without
closing the connection upon reaching the timeout value.  On some of
these "labotomized" children, I can do a kill -PIPE <PID> and get
success, others are less cooperative and require a kill -9 <PID>   I
have noted that these more stubborn children are the ones that have been
around for quite some time (well over 5000 seconds).

Hope this info helps.

--Derek