WLS 5.1 hangs due to many sockets in CLOSE_WAIT state

Question

Hi,
we are having some serious problems with Weblogic Server 5.1 (SP 6) with
several of our customer installations (Solaris 5.8 and HP UX). The problem
relates to the WLS http server not dealing with socket connections in
CLOSE_WAIT state correctly. We have observered that these can be caused by
Netscape browsers when the end user clicks on the reload button. This causes
the socket connection to be placed in CLOSE_WAIT state, it stays like this
for ever and eventually when there are many sockets in this state WLS runs
out of file descriptors and hangs. We have tried using WLS Apache plugin but
Apache merely passes the CLOSE_WAIT's onto WLS web server. Any help
appreciated on this one,.....
regards Donal.

Answer 1

The problem is that your file descriptor table is filling up with sockets in the
CLOSE_WAIT state so there is no more room to put new file descriptors. Any
machine that speaks HTTP is subject to this so you need to apply these changes
both to the web server machine(s) and the application server machine(s).
The way to handle this on Solaris is to decrease the close wait interval:
ndd -set /dev/tcp tcp_close_wait_interval 60000 (Solaris 2.6 and below)
ndd -set /dev/tcp tcp_time_wait_interval 60000 (Solaris 2.7 and above)
(NOTE: It is recommended that you not go below 60 seconds for this, as per
Adrian Cockroft of SUN).
and increase the size of the file descriptor table by adding the following in
/etc/system:
rlim_fd_cur 8192
rlim_fd_max 8192
(NOTE: The close wait interval can also be set in /etc/system, though I don't
remember the name of the parameter off the top of my head.)
Also, make sure that you change the startWebLogic.sh script to grab more file
descriptors with ulimit (or simply remove the ulimit section of the script)...
Hope this helps,
Robert
Donal Mc Ateer wrote:
We have done this. We have made the following 'tweaks' to the Solaris
installation
ndd -set /dev/tcp tcp_keepalive_interval 300000
ndd -set /dev/tcp tcp_ip_abort_interval 2000
ndd -set /dev/tcp tcp_rexmit_interval_max 10000
ndd -set /dev/tcp tcp_fin_wait_2_flush_interval 67500
ndd -set /dev/tcp tcp_time_wait_interval 30000
is there anything else we can do? Is there any way of influencing the WLS
http server handling of these CLOSE_WAIT sockets? Is this being addressed in
WLS 6.0?
-Donal.
"shivu" <[email protected]> wrote in message
news:[email protected]...
try tuning the tcp parameters at solaris level.
(i think tcp_conn_wait)
"Donal Mc Ateer" <[email protected]> wrote:
Hi,
we are having some serious problems with Weblogic Server 5.1 (SP 6) with
several of our customer installations (Solaris 5.8 and HP UX). The
problem
relates to the WLS http server not dealing with socket connections in
CLOSE_WAIT state correctly. We have observered that these can be causedby
Netscape browsers when the end user clicks on the reload button. Thiscauses
the socket connection to be placed in CLOSE_WAIT state, it stays likethis
for ever and eventually when there are many sockets in this state WLSruns
out of file descriptors and hangs. We have tried using WLS Apache pluginbut
Apache merely passes the CLOSE_WAIT's onto WLS web server. Any help
appreciated on this one,.....
regards Donal.