[ic] Perl 5.8.0 + Interchange Clustering locking issues

[Dan Browning]

At 11:27 PM 9/9/2002 -0400, you wrote:
>Quoting Dan Browning (dbml@kavod.com):
> > At 10:01 PM 9/9/2002 -0400, you wrote:
> > >On Mon, Sep 09, 2002 at 06:45:02PM -0700, Barry Treahy, Jr. wrote:
> > > > Dan Browning wrote:
> > > >
> > > > >At 09:36 PM 9/8/2002 -0700, you wrote:
> > > > >
> > > > >>At 10:28 PM 7/27/2002 +0200, you wrote:
> > > > >>
> > > > >>>>Dan Browning wrote:
> > > > >>>>
> > > > >>>>>For fun, I was testing Perl 5.8.0 (default build options) on our
> > > > >>>>>Interchange cluster and found some file locking problems.  We'll
> > > > >>>>>be sticking with 5.6.1 for now, but I wanted to warn those who run
> > > > >>>>>clusters just in case they try it:
> > > > >>>>>x.x.x.x - - [27/July/2002:12:10:03 -0700] www11
> > > > >>>>>/cgi-bin/template/index.html Runtime error: Could not lock file:
> > > > >>>>>Interrupted system call
> > > > >>>>>It occurs even under very light loads, with MySQL sessions, and
> > > > >>>>>everything else shared over NFS3 (fcntl locks).  Perhaps it will
> > > > >>>>>be fixed in Perl 5.8.1...
> > >
> > >
> > >How do you know that is perl related and not NFS or NFS3?
> >
> > It could be a NFS configuration problem, but here is what I know:
> >
> >   * 5.6.1 works fine for our cluster (NFS)
> >
> >   * Barry doesn't use NFS, neither does he have a cluster, but is
> >     experiencing the same problem with 5.8.0.
> >
> > >At least
> > >in debian world, NFS defaults just changed to sync from async.  Perl
> > >changed from unthreaded to threaded.
> >
> > That was gutsy, but I guess the threading is supposed to be a lot 
> better in
> > 5.8.x.  Of course, I've known Debian to be everything except gutsy, so...
>
>Interchange does not like a threaded Perl, I can tell you that much. Of
>course that was with the old threads model, so if you have those defined
>I would undef them real quick. I doubt you do though, because IC won't
>even start with -DUSE_THREADS on Perl 5.6, depending on the phase of the
>moon.

Indeed I didn't use threads, in fact, I excluded threads from the 5.8 build 
I tested it with.

>But I suspect the problems have to do with changed signal code -- I hope
>for our sake that every OS doesn't start having the signal problems
>we have with BSD....

I know that 5.8.1 will have it all nipped in the bud.  How, you ask?  Why, 
due to my use of the cognitive behavioral theory:  I mentally picture it 
working, and Perl just behaves correctly.  [1]

/~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\
| Dan Browning, Kavod Technologies      <dan.browning@kavod.com> |
| (360) 882-7872 x7, 6700 NE 162nd Ave, Suite 210, Vancouver, WA |
\~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~/
Baruch's Observation:
         If all you have is a hammer, everything looks like a nail.


[1] Yes, I'll go back to the asylum now.