Mike Cantone wrote:
I'm optimistic I am on the crux of solving this problem with the help of some collective knowledge.

My ASP application can handle between 20-40 transactions - then it seems to turn into a runaway process. The memory consumption according to top is well under 1%, however the CPU for httpd goes into the high 90th percentile. My application calls swish-e indexing and I found a very interesting reference on the net which describes what seems to be the identical problem here: http://mathforum.org/epigone/modperl/fraplahswo . One of their recommendations is to put an AUTOLOAD in mod perls startup.pl to see what is happening and, as was the case in the thread, I can now see that lots of DESTROYs are not getting found.


I would recommend looking into BSD::Resource for per process limits setting. IF Apache::Resource has been ported to mod_perl 2.0, I would look at that too, which wraps around BSD::Resource nicely.

The majority of these DESTROYs however probably don't really exist, but a method that can't be found that looks like a real problem is Apache::Connection::fileno. Josh mentioned fixing this in some of his change notes for versions of modperl that didn't support this. I don't know who is looking for this method - I don't know whether it is ASP or not. I am using Apache 2.0.x which doesn't allow me to use modperl 1.0 (seems to want Apache 1.3), so I am using modperl "2.0" - the dev version. Where I can find the Apache::Connection.pm module, fileno truly is not defined, but what is not clear to me is what I can do about it. Moving to a 1.3 version of Apache is not an option for me - too many others rely on 2.0. Any ideas?

The code that references fileno should be safe, even if it does not exist:

# Apache::ASP::Response->IsClientConnected
        my $fileno = eval { $conn->fileno };

If you really want this code to not do what its doing,
then you could define this after loading Apache::ASP:

sub Apache::ASP::Response::IsClientConnected {
  shift->{IsClientConnected} = 1;
}

And your application should work fine still regardless of whether
this is a real fix for your problem.  I am guessing that the real
root of this problem is something else, and you will find your fix
best in BSD::Resource.  Also, if you are using worker mpm, I probably
wouldn't, and would be wary of using BSD::Resource in this case.
I would stick with the prefork mpm in Apache 2.

Regards,

Josh

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Reply via email to