Mysterious coredumps on FreeBSD

From: Alex Rousskov (rousskov@ircache.net)
Date: Mon Jun 19 2000 - 17:51:54 MDT


Dear all,

        If you are using FreeBSD, please be advised that there appears
to be a memory management bug in FreeBSD 3.3 that gets triggered when
there are a lot of file descriptors in use. The OS bug often results in
older Polygraph versions dumping core. Polygraph 2.5.0 has a piece of
code that may warn you about the problem:

        Poll.cc:111: bug: invalid fd returned by poll(2): -1
        Poll.cc:111: bug: invalid fd returned by poll(2): -1074790400
        
However, that code cannot prevent the OS from [eventually] crashing the
test.

To fix the bug, I suggest that you apply the patch mentioned at
        http://www.FreeBSD.org/cgi/query-pr.cgi?pr=16568

In our limited tests, the PR 16568 patch alone fixes the kernel bug.

A better alternative would be to upgrade to the most recent 3.x release
of FreeBSD. If you are not a Unix guru and want to get the current
FreeBSD version with all Polygraph-specific patches, see
        http://www.measurement-factory.com/docs/FreeBSD/

N.B. I'd like to thank NetApp folks that have helped us a lot in
     debugging the problem and trying various patches.

Alex.



This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:13 MDT