On Mon, 5 Dec 2005, Adam Grzesko wrote:
> We are running tens of hlds and srcds servers. Just after the latest
> update of srcds our srcds servers started crashing tremendously, even like
> 20-30 times a day. You can imagine whining and moaning of our young
> customers and our headache trying to solve all the problems and
> upgrading mani admin mod in the background :).
>
> Most of our systems are FreeBSD 5.4 stable and on several machines
> there's still good old FreeBSD 4.11 stable. All the instalations have:
> - SMP kernel
> - linux kernel emulation version 2.4.2
> (frankly, this is the only version available for FreeBSD).
>
> So far our installations had linux base 8 with glibc 2.3.2, which
> worked just fine....till there was this update.
>
> I have read somewhere at this list, that the cure might be upgrading
> glibc up to version 2.3.5. So I prepared custom linux base port which
> solved the problem of crashing srcds. There are just a few
> crashes daily per srcds process, so I assume this is just normal :).
There should not even be one crash per day if you restart the app every
morning like we do.. At 06.00 it restarts everyday as a workaround for the
memory leaks. We do this regardless of what game it is.
But afaik I rarely have any crashes on srcds. (FreeBSD 5.4-STABLE,
linux_base-rh-9 (glibc-2.3.2)) All servers are running without crashes
today.
> This custom linux base port is composed of:
> - linux base suse 9.3_2 (without it's glibc 2.3.2)
> - and fedora core 3.1 rpm's (glibc 2.3.5 i386, libstdc++ 3.4.4,
> libgcc 3.4.4)
why not use a plain FC4 linux_base? I rewrote rh9 to fc4. Didn't find any
real improvement and went back to rh9 on most servers. I also run suse9.3
and gentoo. All encounter crashes. gentoo runs with glibc 2.3.5.
> While srcds servers where now just working fine, with this new linux
> port, hlds (v27) servers started crashing like hell in the numbers
> going up to 40 times per day and before this change there were only
> few crashes daily.
Before the introduction of support for VAC2 (may/june) I hade rarely any
problems. And the problems have staggered after each patch.
> While doing some research on our installations of hlds servers I was
> blind to see that the defaults for these had the old metamod 1.17.1
> and the old AMX 0.9.9.
> We have switched to new metamod 1.18p26 and AMXX 1.60 and all
> instalations of hlds are running smooth and clean (so far :>) along
> with srcds servers.
I have not really tested amxx but I have been running amx w 1.17 and 1.18
and encounter crashes. I'll rewrite all my plugins for amxx later to test
this theory. But with v24 I could recreate all the crashes without metamod
enabled at all.
> **********
> The conclusion is : if you are running FreeBSD, upgrade your linux
> port to have glibc 2.3.5. At hlds installations check version of
> metamod and if necessary upgrade to the latest version.
> Don't upgrade to glibc 2.3.6 as we have experienced enourmously higher
> cpu utilization running 2.3.6.
> **********
>
> With this configuration in mind I smacked the latest upgrade of hlds -
> that is v28. My advice: don't try this at home,
> it welcomes you with a beautiful crash just right at startup :{.
My advice would rather to be that Valve go through all the changes they
have done to hlds this year, to see what code that causes these crashes on
several different distros. I thought it only was a FreeBSD issue until
lately when there have been reports of crashes running several different
linux distros. With this I dont mean a rollback, but go through each patch
to see if they can find out why it is creating these crashes. And for me
choke.
Or, for us that are using an real OS, release a FreeBSD native version of
the servers. But that will probably never happen.
/Bjorn
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
