Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 07 May 2010 23:21:02 +0200
From:      Peter Palmreuther <pitpalme@gmail.com>
To:        freebsd-java@freebsd.org
Subject:   Re: openjdk6 crashes
Message-ID:  <4BE4843E.6030901@gmail.com>
In-Reply-To: <201005070902.22700.achill@matrix.gatewaynet.com>
References:  <4BE11114.5030306@gmail.com>	<38DD588D-992A-4713-8D55-96B28E2D2AA2@moumantai.de>	<4BE3A331.20607@gmail.com> <201005070902.22700.achill@matrix.gatewaynet.com>

next in thread | previous in thread | raw e-mail | index | archive | help
Hello,

On 07.05.10 08:02, Achilleas Mantzios wrote:
> Hi, in your first message you said you got a "signal 6" SIGABRT, while with the fastdebug build
> you got a "signal 4" SIGILL, so you dont have consistent behaviour.

Yes, I've seen this too. My guess was the "normal" and the "fast-debug" build
differ and therefore behave slightly different.

> Are you sure your hardware is sane?

*hmmm* Depends on what level of certainty you'd call 'being sure'.

> Is your memory ok?

I thought and think so. I have not seen any behavior that makes me think my
memory is defect. The only thing crashing is OpenJDK and only when Sonatype
Nexus updates indexes.

> Did you run memtest86?

Not before your mail. But just to be sure I now did.
4 passes, not a single error. It took approximately 2 hours, so if CPU load,
and therefore system temperature, play a role, it should have been high
enough. Sonatype Nexus running crashes immediately when index update is
started, even if system is running only minutes.

> Are your disks ok?

Hope so. No signs of defect sectors. And as I reinstalled OpenJDK I'd guess
binaries and libraries are on different sectors, so a defect sector corrupting
them should not be very probable. At least the crash should look different,
doesn't it?

> Is your motherboard in a good condition? (no sounds, fans rotating fast, etc..)

It is. fans are normal speeding and silent. No "beeps", no sounds, no
recognizable anomalies.

> Is your CPU in a good condition? 
> Do you experience any high temperatures on your CPU?

I hope so. I haven't found any tool with reliable information about
temperature or fan speed. But there's no hot air blown out or something like
this. The machine I'm taking about runs day after day without any error. One
exception: I start tomcat manually after first crash and index refresh,
executed automatically if Tomcat runs, kicks in. So it's *ONLY* Sonatype Nexus
index update that kills *ONLY* OpenJDK.

> Note, that it is *100%* normal to have a problematic hardware, and the only
> instance that the hardware problem is manifested to happen when running a
> particular program or script (even consistently).

*erm* Maybe. For fixed conditions (which are seldom in my experience). Most
times I've seen systems crashing here and there with no fixed script or
program if hardware dies. Simply because it's not deterministic what
temperature the system will have and what program will be at which memory
address. The only fixed crashing point can be the hard disk, which I think I
can exclude due to the recompilation and therefore "relocation" on disk (i.e.
different inodes being used).

And to come to an end: if I see a program written in C++ (like OpenJDK/JVM)
constantly fails on an assertion (like in my case: 'Error: assert(sp() >=
inputs,"must have enough JVMS stack to execute")') I expect the assertion to
not match first and think about corrupt memory a few hundred positions later
on the "check list". That's at least what assertions are for: make the program
fail if something is recognizable wrong. If every "assert fail" would be a
memory defect unless 100% proven not to be that would render assertions
useless, wouldn't it?

But thanks for your help and input; Maybe somebody else has any idea about why
"sp()" is less than "input" when compiling (hotspot compiling?) this special
class in OpenJDK?!?
-- 
Best regards,

Peter Palmreuther



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4BE4843E.6030901>