[M.C.S.]

You got some access data (please no abuse, my config isn't really secure for 3rd party shell access ;) ). Most important thing: The crashes only occur if there are people playing on the server. An empty server does never crash...

[david.vincent]

Originally Posted by BAILOPAN It's fishy that all the systems this happens on are using the NPTL (/lib/tls) libraries.

some backup software i use (bacula) says to get rid of the /lib/tls directory:

Get Rid of the /lib/tls Directory
The new pthreads library /lib/tls installed by default on recent Red Hat systems running kernel 2.4.x is defective. You must remove it or rename it then reboot your system before running Bacula otherwise after a week or so of running, Bacula will either block for long periods or deadlock entirely. The feedback that we have concerning 2.6 kernels is the same. However, on 2.6 systems, you may want to use the loader environment variable override rather than removing /lib/tls.

f you have a recent Red Hat Linux system running the 2.4.x kernel and you have the directory /lib/tls installed on your system (normally by default), bacula will NOT run. This is the new pthreads library and it is defective. You must remove this directory prior to running Bacula, or you can simply change the name to /lib/tls-broken) then you must reboot your machine (one of the few times Linux must be rebooted). If you are not able to remove/rename /lib/tls, an alternative is to set the environment variable "LD_ASSUME_KERNEL=2.4.19" prior to executing Bacula. For this option, you do not need to reboot, and all programs other than Bacula will continue to use /lib/tls.

The feedback that we have for 2.6 kernels is that the same problem exists. However, on 2.6 kernels, we would probably recommend using the environment variable override (LD_ASSUME_KERNEL=2.4.19) rather than removing /lib/tls.

that sound like the same problem?

i'll try the environment variable trick.

Originally Posted by BAILOPAN Better yet, anyone feel like giving me shell access to run GDB and track this down?

i'm off for a coulple of days in about 24 hours but if you get no other offers and have a static address i can give you ssh access.

[Sir Tom]

Originally Posted by BAILOPAN Compiling in all the necessary libraries is a bad idea... I will post some test binaries here later today though

It's a new day Anything new about the topic?

[M.C.S.]

Originally Posted by david.vincent Originally Posted by BAILOPAN It's fishy that all the systems this happens on are using the NPTL (/lib/tls) libraries. some backup software i use (bacula) says to get rid of the /lib/tls directory

Hmm, TLS has a lot of dependencies on Debian, MySQL for example. I think it won't be a good idea to rename / remove the tls directory

[Opacus]

This is getting me frustrated. Running 6 servers on Fedora Core 2, all crashing randomly and frequently with segmentation faults now, sometimes even just stopping, not reporting any error at all - just crashing. I really need a way around the JIT+Fedora problems - is there one? If anyone know of a 100% stable distro for AMXX+64bit support I'm willing to swap

[BAILOPAN]

There is no JIT on 64bit... hmmmm

[Opacus]

Even more akward then... Because they keep crashing like there's no tomorrow...

I have reverted to version TP4 - that's the most stable one for me. I've tried RC6 and 1.0 as well, but RC6 crash more than TP4, and 1.0 crash about once pr. hour.

[david.vincent]

Originally Posted by david.vincent i'll try the environment variable trick.

just before i was going away on a fedora fc3 server i set

export LD_ASSUME_KERNEL=2.4.19
export MALLOC_CHECK_=0

it crashed within the hour so changed all the servers back to adminmod. when i came back they'd all stayed up. last night i changed one server (the machine run's two cs servers) back to amxx with the above variables set. this morning when i checked it had been up 5 minutes. the adminmod server was up all night. the debug log says it's not using the /lib/tls/ libraries:

Using host libthread_db library "/lib/libthread_db.so.1".
Core was generated by `./hlds_i686 +port 27015 -pingboost 1 -steamuser xxx -steampass xxx''
Program terminated with signal 11, Segmentation fault.
#0 0x0060b08d in ?? ()
#0 0x0060b08d in ?? ()

two servers running amxx on suse 9.1 are running fine so it looks like it's something in fedora. any chance of those test binaries or still want a login?

thanx.

[M.C.S.]

As long as there is this big problem, I am officially running other plugins than AMX(X). Both have the same problem in the current version, so I have to choose other tools. It's breaking my heart

[DopeFish]

well, my two server are still running amx 0.9.8 and amxx 0.16. I've got a testserver running amxx 1.0, but it crashes randomly without any useful message (the standard Segfault message). If I manage to be able to reproduce a crash I could strace the bugger.