On 2011/01/04 09:59 PM, Anatoly Ivanov wrote:
Hi,
I am running freeradius (2.1.8) with rlm_perl (5.10.1, USE_ITHREADS) on a Debian-Lenny system.
The problem is radius fails with segfault – periodically and intermittently. I have no way to reliably reproduce the problem – it happens only in production, and it is impossible to reliably predict when or backtrace why.
It seems that I am running into some kind of memory allocation error.
Coredump type #3 (see below) is the most "popular" one; coredumps with backtrace going into perl seem to be rather random (it fails in different parts of libperl.so) - again, see below.
I understand that freeradius has a newer version available - but I am hesitant to upgrade a production server without a very good reason. And I could not find such reason for an upgrade after reading the CHANGELOG for 2.1.10. But maybe I am wrong?
Any ideas?
A complete gut feel after reading this says you have a hardware problem - faulty ram. Has this happened from the beginning or suddenly now? You can try memtesting (http://www.memtest.org/) the server, or a trick that I've found works sometimes (if you can't take the server out of production) to show a ram problem, is to compile a kernel. I've seen compilation fail at different stages with faulty RAM. I realiuse the advice might sound ridiculous, but it has worked for me before. -- Johan Meiring Cape PC Services CC Tel: (021) 883-8271 Fax: (021) 886-7782