[EMAIL PROTECTED] wrote:
> On Fri, 25 Aug 2000, Santiago Gala wrote:
>
> > I am experiencing a strange deadlock during cache updates. To me, it
> > happens only with tomcat, under FreeBSD 1.1.8 green threads, and not
> > under my other setups.
> >
> > It is related to DNS lookup. The symptoms are:
> >
> > After a while updating the cache, the VM hangs.
> >
> > In fact, this VM stops responding while there is a DNS lookup, but
> > sometimes it freezes. It looks as if the thread "missed" the query
> > response. This thread holds locks, including the IO Monitor lock, that
> > stop the whole virtual machine.
> >
> > I have not seen a similar behaviour with my other setups, identical
> > except:
> >
> > NT, jdk1.2.2, hotspot server
> > NT jdk 1.3, hotspot server
> > Linux, jdk1.1.7, with/without JIT, green threads
> > Linux jdk1.2.2, with JIT, native threads
> > Linux jdk1.3b9, hotspot client, native threads
> >
> > All ofthem work perfectly. The problem is the FreeBSD setup is the only
> > available in my public "virtual" server... :-(((((
>
> There is a java dns resolver that may work for you. It could
> be a bit of work to implement thou. The package is called
> dnsjava and a quick web search should find it. There is an
> example of it's use in JAMES ( mail server project on
> java.apache.org )
>
Thanks. I'll look at it. For the moment I have found a simpler solution:
substitute in the classes.zip a "hacked" version of jdk1.2 InetAddress.java I
found in the bugzilla of Sun, and adapt it to run in jdk1.1.8.
With my preliminary tests, it looks like it runs, although I'll need more
time to be sure.
It any of you is experiencing green threads VM deadlock with jdk1.1.X on ns
lookups, please tell me and I'll send you the class.
The bug is amazing. It was reported in 1997, and closed later on as "non
reproducible". Later, they acknowledged the bug (I think in 1999), and solved
it finally in jdk1.3. So, people that don't have jdk1.2/1.3 cannot even VOTE
for the bug, as it is formally solved. :-(
The reason why I had not seen it before (only sporadically with jdk1.1 under
Linux) is that probability of deadlock decreases with speed, and the
hisitech.com machine is fairly slow (5 times slower than my desktop running
jdk1.1, about 20 times slower running 1.3 Hotspot Client beta)
I'm testing a Jetspeed copy in http://hisitech.com:8080/servlet/jetspeed, in
case somebody want to look at it.
Regards
>
> Cheers,
>
> Pete
>
> *--------------------------------------------------*
> | Latrobe University, | Does the name 'Pavlov' |
> | Bundoora, Australia | ring a bell ? |
> *--------------------------------------------------*
>
> --
> --------------------------------------------------------------
> Please read the FAQ! <http://java.apache.org/faq/>
> To subscribe: [EMAIL PROTECTED]
> To unsubscribe: [EMAIL PROTECTED]
> Archives and Other: <http://java.apache.org/main/mail.html>
> Problems?: [EMAIL PROTECTED]
--
--------------------------------------------------------------
Please read the FAQ! <http://java.apache.org/faq/>
To subscribe: [EMAIL PROTECTED]
To unsubscribe: [EMAIL PROTECTED]
Archives and Other: <http://java.apache.org/main/mail.html>
Problems?: [EMAIL PROTECTED]
