Sockets left in TIME_WAIT
(no name) ((no email))
9 Oct 96 7:13:14 EDT
Message-Id: <199610091111.HAA05710@vgi3mail.vanguard.com>
To: fastcgi-developers <fastcgi-developers@openmarket.com>
From: Alan Yagoda/VGI
Date: 9 Oct 96 7:13:14 EDT
Subject: Sockets left in TIME_WAIT
We are using a remote managed fastcgi application started with cgi-fcgi.
It appears that every time a request is issued from the OM Web Server to the
remote fastcgi application, the socket descriptor used by the fastcgi
application is left in a TIME_WAIT state (using 'netstat -a' on Solaris to view
descriptor state). Normally setting sockopt() with SO_REUSEADDR on the
listen() socket descriptor alleviates this problem. We are using a version of
cgi-fcgi that sets SO_REUSEADDR but we are still seeing the connections being
left in TIME_WAIT until the OS cleans up. We have gotten to the point where
we can't connect to the fastcgi application because it can't allocate any more
descriptors.
We had to reconfigure kernal paramteres using 'ndd' to clean up open
descriptors more frequently. I suspect that this is not the correct solution,
but it has worked so far.
Has anyone seen this problem? Any suggestions on how to close the
connections cleanly?
Thanks.
-Alan
*************************************************************************
Alan Yagoda Phone: 610 669 1743
Sr. Information Systems Engineer Fax: 610 669 1714
The Vanguard Group E-Mail: Alan_Yagoda@vanguard.com