Re: FIN_WAIT_2 connections in Apache 1.2 beta (fwd)

[Marc Slemko <ma...@znep.com>]

One report of NO_LINGCLOSE helping.

---------- Forwarded message ----------
Date: Sun, 12 Jan 1997 13:45:32 -0500
From: Bill Nesbitt <bn...@hway.net>
To: Marc Slemko <ma...@znep.com>
Subject: Re: FIN_WAIT_2 connections in Apache 1.2 beta

At 11:33 AM 1/12/97 -0700, Marc Slemko wrote:

>I have gathered the names of everyone I could find who was having trouble
>with connections hanging in FIN_WAIT_2 with Apache 1.2 betas.  The exact
>problem is not known, but if you could answer the below questions we can
>hopefully figure out the problem and fix it.
>

>What OS and version are you using?
>

	Irix 6.2


>Does the problem happen with Apache 1.1.1?  Please check carefully to
>be sure of your answer; some number of connections in FIN_WAIT_2 are
>normal and are nothing new.  It is when large numbers start appearing that
>they being to cause trouble.
>

	Definately not.

>What negative results do the connections in FIN_WAIT_2 have on your box? 
>Are they just something you see hanging around, or do they have a serious
>impact (like crashing your box)?
>

	1) Gather large amounts of CPU time in interrupt mode, 2) runs low on memory, starts to swap, 3) crashes box due to lack of resources.


>What sort of traffic does the server handle?
>

	Anything and everything.  Nothing in particular.

>Try adding "-DNO_LINGCLOSE" to the EXTRA_CFLAGS in the Makefile and
>rebuilding.  Does this help?
>

	Yes.  I have found that it does help.  I think this might do the trick.


>Do you have a SendBufferSize directive in your config file?
>

	No.


>Does disabling keepalives ('KeepAlive 0' in the config file) help?
>

	No.


>Do you have any idea what OS and browser the remote systems involved with
>the FIN_WAIT_2s are running?
>

	Can really tell if it is a specific OS & browser.


>Does replacing the http_main.c from a recent 1.2 beta with the one at
>http://www.worldgate.com/~marcs/ap/http_main.c help?  (This version is not
>really suitable or desirable for running long term, but for testing
>purposes it should work fine and has many of the changes between 1.1.1 and
>1.2 hacked out.) 

>

	Have not tried this as the NOLING_CLOSE seems to be working for me.



Hope this helps. 

------------------------

<bold>Bill Nesbitt

</bold>bn@hway.net

Hiway Technologies, Inc.

http://www.hway.net