Build 161 server memory leak
devicenull
Join Date: 2003-04-30 Member: 15967Members, NS2 Playtester, Squad Five Blue
Join Date: 2003-04-30 Member: 15967Members, NS2 Playtester, Squad Five Blue
Comments
I would start uploading the logs.
<img src="http://img822.imageshack.us/img822/2575/graphm.png" border="0" class="linked-image" />
Notice the memory stays level, when it first had players tonight at 7pm, memory remained constant.
Looks like when the map changed or players left (?) the memory spiked up, then peaked a second time until it crashed at its limit of ~1.7GB or so.
Also, changed map? How do you do that?
And sorry I meant when the map resets (players usually leave after a game if they're gonna etc) not map change.
Also interesting that it chews 50% CPU (1 core) even when idle.
No idea what the spiky section was at 3pm, pretty strange I agree.
Let's see how it goes tomorrow I guess. Unfortunately the server crashes due to memory leaks again, not sure if your server is doing the same (rada sucks yeah?).
Looking at the graph the memory spikes a couple of times after 7.00pm. I'm guessing that's when it patched.
If it helps, I can provide process memory dumps but please send me a PM if you want these (developers).<!--QuoteEnd--></div><!--QuoteEEnd-->
Is there anything in the log? We saw a problem like this once in our testing, but after restarting the server is stayed at a consistent ~500 MB. We'll be looking into the cause of that for the next build.
A bunch of this: (8000+ lines of it)
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->Error: std::exception (bad allocation)
Error: std::exception (bad allocation)<!--c2--></div><!--ec2-->
Nothing else really strikes me as important from the log, it all looks pretty normal.
<!--c1--><div class='codetop'>CODE</div><div class='codemain'><!--ec1-->Error: std::exception (bad allocation)
Error: std::exception (bad allocation)<!--c2--></div><!--ec2-->
Nothing else really strikes me as important from the log, it all looks pretty normal.<!--QuoteEnd--></div><!--QuoteEEnd-->
Thanks, that's what I was expecting. I'm not really sure what's causing this yet (my best guess is something related to a player connecting), but it will be a top priority for 162.
No crashes here, but I've only had a few players and I doubt a full run at the map. Now that we might be needing the NS2Monitor/ProcessMemoryMonitor.exe again, could anyone with experience using it PM me? I'm a bit lost. And the <a href="http://forum.specialattack.net/viewtopic.php?f=94&t=7639" target="_blank">original instructions</a> aren't helping me much.
He sure does.
edit: I don't see memory leaks unless it happens at the start as described below.
Still havnt had a memory leak yet <b>EDIT: Just checked the log, it did reach 2gb this morning at 26 past midnight</b>, played some good games on Tram last night too, which seemed better, no black screens (possibly due to turrets being so expensive). One reproducible way I have noticed a memory leak is when clients are currently connected to the server, and when you restart it, they keep sending bad data and filling the memory untill they timeout, but it has normally crashed by then.
For example, if I restart the server while it has players currently connected, when the server comes back up, they must still be sending data to the server, because memory will grow at a rate relevant to how many clients were connected, however, if I wait untill all the users have timed out, restarting the server will start it at around 450MB and keep it steady there. If its possible for a previous client to keep sending rubbish data, is it possible for a different user to be the cause of it?
Recent Crash @ ~650MB with ~7 Players and ovrmind active:
"Faulting application server.exe, version 0.0.0.0, time stamp 0x4d2c794b, faulting module kernel32.dll, version 6.0.6001.18215, time stamp 0x4995344f, exception code 0x40010006, fault offset 0x0002f328, process id 0x988, application start time 0x01cbb3408d9339d0."
For example, if I restart the server while it has players currently connected, when the server comes back up, they must still be sending data to the server, because memory will grow at a rate relevant to how many clients were connected, however, if I wait untill all the users have timed out, restarting the server will start it at around 450MB and keep it steady there. If its possible for a previous client to keep sending rubbish data, is it possible for a different user to be the cause of it?<!--QuoteEnd--></div><!--QuoteEEnd-->
Thanks, that's very useful information.
That is why for now I've increased the cost of the Hydra so no-one goes around spamming them everywhere.
That is why for now I've increased the cost of the Hydra so no-one goes around spamming them everywhere.<!--QuoteEnd--></div><!--QuoteEEnd-->
Yes. A process can only use up to 2gb of memory. Once it tries to go above that, it crashes.
"Hydra Optimization (create 5+) "
its on the progress page. its in the works. nobody is friggin playing right now anyways unfortunately so might as just well wait for the patch next week.
maybe they could be nice and toss in Onos and allow pre ordering and ads on steam. revenue and features. but thats just my pipe dream :)
Okay i'll keep an eye on it. If you are referring to 2GB due to server.exe being a user process, you could use the /3gb switch to give you another GB. Although that's just going to delay the issue from happening and not fix the root cause.
<img src="http://dl.dropbox.com/u/2078961/ns2.png" border="0" class="linked-image" />
Note: The initial spike on the far left of the graph is from 161. The drop represents me restarting the server to apply 162.
It's some custom software that we wrote. Sorry, but it's not something you can download anywhere.
Thought it could have been from the web admin page. I have not yet had any leaks like are shown in that graph, had up to 12 players, currently about 6 and its holding steady.
Have you tried running everything stock?