URGENT - WL server crash in production - Native performance pack ??

Question

Hi,
We are experiencing periodic crashes in our production environment of the
nature where WLS just stops running and exits out, i.e. the JVM exits
abruptly. Many times we have seen http tunneling / socket related exception
just before the crash occurs.
Following is our configuration -
WL 4.5.1 service pack 13
JDK 1.1.7B
OS: Windows NT 4.0, sp 5
Oracle 8.1.5 with Oracle Thin drivers for JDK 1.1.7
Native IO : true i.e we are using the NT performance pack.
The latest exception that we have seen is as under -
Wed Jan 24 11:01:46 EST 2001:<E> <HTTP> Servlet request terminiated with
RuntimeException:
Wed Jan 24 11:01:46 EST 2001:<E> <HTTP> java.lang.NullPointerException
at weblogic.socket.TunnelContext.getServlet(Compiled Code)
at weblogic.socket.MuxableSocketHTTP.invokeServlet(Compiled Code)
at weblogic.socket.MuxableSocketHTTP.execute(Compiled Code)
at weblogic.t3.srvr.ExecuteThread.run(Compiled Code)
This seems to be a problem related to the use of performance pack and native
sockets during HTTP tunneling. In the latest crash, we got the exception
pasted above and the server crashed about 2 seconds after that. We have seen
things in the recent past where we get a HTTP tunneling related exception
and the server exits immediately afterwards.
Is anybody using the native performance packs in their production
environment on the above mentioned configuration? Has anybody seen this kind
of behaviour? Does anybody have any input on the use of native performance
pack (nativeIO) and its pitfalls in WLS 4.5.1 and sp 13 or otherwise?
Some research and investigation from our side has revealed the following
(hope it helps, but want to know more and for sure) -
In the weblogic.socket.NTSocketMuxer.java, there is an infinite loop that
tries to send HTTP tunnel keep alive messages to the clients. There it tries
to get the TunnelServlet to send the message. It gets a runtime exception
(the one in the trace above) but that is caught and the loop goes on. But on
the next pass through (in about 2 secs. or so), it makes a some native calls
(getNextSocket(), deleteIoStructure() and memcpy() which are all native
methods). Experience says that all native calls are prone to serious
problems.
So, does anybody have any info./solution/suggestion to shed some light on
this issue. This is an urgent issue as it is affection the production
environment.
Thanks in advance.
-SG

Answer 1

No, this is not currently available. FWIW, native io only helps when you
have a lot of sockets.
-- Rob
Larry Presswood wrote:
Is there any way to use the performance pack/native muxer in a client
weblogic
RMI program. We have a client callback into the WLAS server so we
were hoping that there was a way to use the performance pack for
the clients socket io???
I know about increasing thread pool size on a client and socket readers
but
was hoping we could use native.
This is weblogic 6.0 sp2
Also if this is not available in 6.0 is it available in 6.1 or going to
be available in > 6.1--
AVAILABLE NOW!: Building J2EE Applications & BEA WebLogic Server
by Michael Girdley, Rob Woollen, and Sandra Emerson
http://learnWebLogic.com
[att1.html]