Embedded Java and signal handlers

[Christian Heimes <li...@cheimes.de>]

Hello!

A couple of months ago we ran into a strange problem with signals in our
application stack. Whenever lucene was loaded, our application didn't
react to SIGINT (ctrl + c) properly. Instead of shutting down step by
step the server stopped immediately.

It took me a while to figure out that lucene.initVM() replaces Python's
SIGINT handler with a new signal handler. Python's SIGINT handler is
responsible for raising a KeyboardInterrupt exception when ctrl+c is
pressed. The JVM grabs a whole bunch of signals for its own purpose and
thus disables all signal handlers installed before initVM().

The fact is well documented at
http://www.ibm.com/developerworks/java/library/i-signalhandling/
together with an easy solution (-Xrs option). Sun's JVM behaves equally.

My Python package smc.lucene goes a slightly different way. It saves all
Python signal handlers before initVM() is called and restores them
afterwards. However that's not an appropriate solution for an embedded
Python interpreter (Python embedded in Apache via mod_python which
itself hosts an embedded JVM via JCC).

Andi, do you see a way to fix the problem inside initVM() anytime soon?

Christian

Re: Embedded Java and signal handlers

[Andi Vajda <va...@apache.org>]

On Mar 10, 2009, at 10:13, Christian Heimes <li...@cheimes.de> wrote:

> Hello!
>
> A couple of months ago we ran into a strange problem with signals in  
> our
> application stack. Whenever lucene was loaded, our application didn't
> react to SIGINT (ctrl + c) properly. Instead of shutting down step by
> step the server stopped immediately.
>
> It took me a while to figure out that lucene.initVM() replaces  
> Python's
> SIGINT handler with a new signal handler. Python's SIGINT handler is
> responsible for raising a KeyboardInterrupt exception when ctrl+c is
> pressed. The JVM grabs a whole bunch of signals for its own purpose  
> and
> thus disables all signal handlers installed before initVM().
>
> The fact is well documented at
> http://www.ibm.com/developerworks/java/library/i-signalhandling/
> together with an easy solution (-Xrs option). Sun's JVM behaves  
> equally.
>
> My Python package smc.lucene goes a slightly different way. It saves  
> all
> Python signal handlers before initVM() is called and restores them
> afterwards. However that's not an appropriate solution for an embedded
> Python interpreter (Python embedded in Apache via mod_python which
> itself hosts an embedded JVM via JCC).
>
> Andi, do you see a way to fix the problem inside initVM() anytime  
> soon?

It's not initVM's doing but the Java VM initialization code that it  
calls and initVM has some control over it via initialization  
parameters (equivalent to java's command line parameters).

You can call initVM(vmargs='-Xrs') but you seem to imply that this  
isn't good enough. I guess I'm not sure I completely understand your  
question.

Andi..

>
>
> Christian