|
From: Clever M. <clv...@gm...> - 2006-03-31 14:41:40
|
On 30-Mar-06, at 9:52 PM, Leif Mortenson wrote: > > I was not able to do a 3.2.0 release for AIX because the person > who normally > helps out couldn't do so this time. But the build should work. > This new version > contains a couple new properties which help with this kind of problem. > http://wrapper.tanukisoftware.org/doc/english/prop-monitor-thread- > count.html > http://wrapper.tanukisoftware.org/doc/english/prop-thread-count- > delay.html > I'll take a look. > The 3.1.2 release has had over 1000 AIX downloads and I have not > heard > any reports of this from other users. This makes me think that it > is not an AIX > wide problem, but rather a problem with a specific JVM implementation. > What is the value of the java.fullversion, java.runtime.version, > and java.vm.name > system properties that you are using? > You can get the information I need by copying the following from > your wrapper.log > with the wrapper.debug=true property set: > jvm 1 | Java Version : 1.4.2_08-b03 Java HotSpot(TM) Client VM > jvm 1 | Java VM Vendor : Sun Microsystems Inc. > This is the official Java 1.5 RC1 release from IBM. I suspect your other 3.1.2 users are launching Java 1.4 apps, and that something has changed a bit since. This is not out of the realm of possibilities, considering how new Java 1.5 is for AIX. The VM information is in my first email, no? The output I attached was created by running with debug=true. I'll repeat the information for your convenience: INFO | 2006/03/30 12:26:45 | Java Version : J2RE 1.5.0 IBM J9 2.3 AIX ppc-32 j9vmap3223ifx-20060310 (JIT enabled) INFO | 2006/03/30 12:26:45 | J9VM - 20060220_05389_bHdSMR INFO | 2006/03/30 12:26:45 | JIT - 20060220_2133_r8 INFO | 2006/03/30 12:26:45 | GC - 20060214_AA INFO | 2006/03/30 12:26:45 | Java VM Vendor : IBM Corporation The solution for us was to tweak the way that the shutdown thread is accounted for in WrapperManager. I can send the debugging output I enabled by recompiling WrapperManager with the trivial debugging (which was commented out) that shows it cycling through the threads looking for non-daemon alive threads. We only ever get "1" of these threads when the VM first comes up, so Wrapper kills the VM. Either the test is wrong, or the vlaue we are testing against (either 0 or 1) is wrong. It looks like Java 5 RC1 on AIX from IBM is more like the WebLogic VM in this regard. If your new properties control these values, then that should be sufficient for us. For this release we may have to run with my changes until I can update our thirdparty resources for next release. Thanks. -- John |