Jboss hang on HPUX

[david.wade]

JBoss 3.2.3 on HPUX 11 (dual CPU) on JDK 1.4.2_00

We have today seen two servers stop. Logging stops, everything stops and the CPU sits at 100% i.e. all of one processor. The application is very vanilla. There are no loops. Just plain JBoss.Net to SSB. We did have clustering enabled but soley for farmed deployment. Have removed farming just case the issue if jGroups releated.

Has anybody else seen this happen and does anybody have any ideas ? Have changed to capture stdout (was /dev/null) and will kill -3 if it happens again.

David

[david.wade]

Well, removing the clustering made no difference. Kill -3 produced no thread dump. The JVM is well and truely locks up and chews 100% of a CPU.

Have gone back to HPUX patches. Seems we don't have some critical patches for running Java reliably on HPUX. What I don't understand is why HP don't include these patches in their standard patch bundles. I mean, it appears their pthreads implementation is completely shot without these patches.

I see the HP now offically supports JBoss, but, wait for it, only when running on Linux... I wonder why ?

[david.wade]

Well, we installed the latest pthreads patch PHCO_30544, but the process still hangs...

[david.wade]

Hi all. I actually like talking to myself ;)

I guess noone else is daft enough to run JBoss on HP-UX. Well, we are finally all patched up on the HP-UX front, and, yes, still the JVM locks up at irregular intervals hours/days.

Is there anybody else out there who is daft enough to be attempting to run JBoss on the dying OS -HP-UX ?

[david.wade]

Its a bug in JDK 1.4.2_xx

Javasoft bug 4880633

The class XMLEncoder, which is meant to be threadsafe, uses the class java.beans.Statement which is also meant to be threadsafe, but isn't as it uses HashMap.

Must say, once we managed to get a coredump in the right format to HP, they found the cause of this issue within hours -and supplied a JAR which when put on the -Xbootclasspath fixes the problem by replacing the class java.beans.Statement with a version that uses Hashtable.

Well done HP Although, if SIGQUIT aka thread dump was not also broken in 1.4.2_00 (and upto _02) we would have found this out ourselves.

[aironyiu]

Hi Dave:

I am starting to test JBoss 4.0 on HP-UX 11i and would like to know more details about this problem (e.g. how to get the fixes). I talked to our local HP engineer and there seem to have no information.

Thks