I'm having some issues with running out of memory, and the $^M trick
didn't provide any insight (I still get "callback called exit").
So, I decided to try Apache::SizeLimit. Well, for BSD systems, it
seems that the computation of the shared size is quite incorrect. The
code is using the 2nd and 3rd return values from getrusage(), which
are, respectively, $maxrss and $ixrss. According to the system's
(FreeBSD 4.8) getrusage(2) man page, these are:
long ru_maxrss; /* max resident set size */
long ru_ixrss; /* integral shared text memory size */
Now, somehow I don't think that *any* perl code is in the text
area. So all we see as shared memory for these systems is the perl
interpreter and any system libraries that are shared. That is, pretty
much nothing compared to the data segment containing the perl code and
data.
However, it seems that the next two elements of the data structure
return the info we need to make an informed decision:
long ru_idrss; /* integral unshared data size */
long ru_isrss; /* integral unshared stack size */
The stack size is probably not significant, so it seems to me that the
following patch will actually do the right thing:
--- #SizeLimit.pm~ Tue Jun 10 15:17:55 2003
+++ SizeLimit.pm Tue Jun 10 15:17:55 2003
@@ -199,7 +199,7 @@
}
sub bsd_size_check {
- return (&BSD::Resource::getrusage())[2,3];
+ return (&BSD::Resource::getrusage())[2,4];
}
sub exit_if_too_big {
Any comments? Should I commit this prior to the upcoming mod_perl
release? Am I correct in my analysis?
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
