Debriefing > Bug Reports & Fixes

[BUG] Server hangs periodically

(1/3) > >>

mookie:
So probably everyone has noticed this, that every once in a while (several times a day) a server will hang, or crash, or something. I've yet to see a MDMP file on either of our servers, probably because they restart automatically if they're unreachable for 15 minutes straight. It's getting to be a problem because last night, both hung at the same time, and maxed the CPUs, dropping our CSS pub from 500fps to 20fps. Is there any indication of what causes this or if it can be prevented?

It seems to happen on a variety of maps, and based on what I see in HLSW when they go down it is not tied to any obvious event (last log is not even always a kill). I see some gaps in one graph even from when players were not playing, meaning to me that probably it has hung sometimes either while empty or shortly after the first connect from empty.

We are running 1.0.1.0 on Windows/Xen using http://allnightersunite.com/gesource/GoldenEye_Source_v4_1_Server.zip

killermonkey:
If no mdmp is produced that almost surely points to a system level crash or a crash in valves hlds itself. A crash in the loaded binaries will produce a mdmp 95% of the time. Suggest you try reinstall of hlds.

mookie:
I don't think it's a crash per se, as a crash/exit would normally cause a quick restart, there would be a MDMP file, and the server maxes out the core its on after this happens (until the daemon manager forces it to restart). I'm going to install a server on my PC to see what happens but tentatively I don't think it's a fluke given that I see most of the other servers (through HLSW) go offline periodically, usually for a half hour or more at a time. I don't know if I've seen a Linux server go down or not.

killermonkey:
I am having a hard time believing its GES thats causing the hangs though.... considering nothing drastic has changed on the server from v4.1 to Beta 4 besides some extra python callbacks and cleanup of code. Unless your logs or a mdmp can point to a single point of failure or a common thread I can only consider it to be in Valve's underlying dedicated server code base (which is notoriously buggy.

I wish you could run a profile on the server to show where each frame is spending it's time... there might be a way to do that, but I am not sure what command it would be.

Try adding the -debug switch to your server profile, I believe this forces a dump every time the server is killed unexpectedly.

Wake[of]theBunT:
The guy who does our officials also is stating there is some random things occurring he can't explain.

I'm betting it's source dedicated server for mods being lame VERY rarely. Let's just hope its not like before in beta 3 and they started breaking things all the time!...

Navigation

[0] Message Index

[#] Next page