Axis hanging when entropy expires

[Bob Arthur <re...@datacash.com>]

Hi,

A colleague was recently trying to set up Axis 1.2 RC2 on his 
development machine.  He got as far as seeing the root page, and having 
HappyAxis display OK.  Also, he was able to view the list of services, 
and download the wsdl from them

Unfortunately, attempting to call any end-point either from the browser 
or from a precompiled stub would simply hang for > 1 minute.

After much trawling through the source, making use of debug logging, we 
tracked the problem down to the SOAPPart constructor, where it called 
SessionUtils.generateSessionId().

It turns out that for some reason, his debian (woody) linux (2.4.23) 
machine was not refreshing its entropy, and had run out.  This caused 
SecureRandom.nextBytes to hang indefinitely while it tried to seed 
itself.  Rebooting the machine solved his problem.

I'm not convinced that this should be raised as a bug in Axis, as it was 
clearly caused by the OS.  However, I thought I'd let you know, partly 
so it's on record for others who may have the same problem, but also in 
case anyone thinks there's a case for somehow making the SecureRandom 
stuff more defensive.

For example, in SessionUtils.getRandom, when instantiating the 
SecureRandom object, we could immediately call nextBytes in a timed 
thread.  This would force the object to seed itself or throw an 
exception if it times out.

Many thanks,

Bob.

Re: Axis hanging when entropy expires

[Steve Loughran <st...@apache.org>]

Bob Arthur wrote:
> Hi,
> 
> A colleague was recently trying to set up Axis 1.2 RC2 on his 
> development machine.  He got as far as seeing the root page, and having 
> HappyAxis display OK.  Also, he was able to view the list of services, 
> and download the wsdl from them
> 
> Unfortunately, attempting to call any end-point either from the browser 
> or from a precompiled stub would simply hang for > 1 minute.
> 
> After much trawling through the source, making use of debug logging, we 
> tracked the problem down to the SOAPPart constructor, where it called 
> SessionUtils.generateSessionId().
> 
> It turns out that for some reason, his debian (woody) linux (2.4.23) 
> machine was not refreshing its entropy, and had run out.  This caused 
> SecureRandom.nextBytes to hang indefinitely while it tried to seed 
> itself.  Rebooting the machine solved his problem.
> 

"Debian unstable" is the platform nobody wants to see on a bugrep. 
Anything can have happened. I remember when Ant got a bugrep that <touch 
 > wouldnt work ...turns out that File.setlastModified() wasnt working.

I'm not in favour of adding checks to the code to deal with interim 
releases of any OS.

What you could do is add some more stuff to happyaxis.jsp...