[EvilEddy]

I've recently checked the server activity graphs for both of my servers, and both seem to be consuming great amounts of memory that never gets freed up again.

Has the server side got a memory leak or have I somehow done something to trigger this?



I anticipated it would crash eventually if the memory usage continued to increase, and tonight it did.

[RaWClan]

Interesting....
Ours seem to increase a little each day too but no where near as dramaticly as yours, Im guessing becasue yours is full 24/7 and ours for about 12-16hours a day. It does drop right back on a quick server restart tho.

I havent got a clue what does this. Im sure Justin will know and will shed some light on this.

Our mem useage on our most popular BC2 Server :
http://gp.killercrea...754bb356c7c.png

[EvilEddy]

RaWClan, on 06 September 2010 - 08:30 AM, said:

It does drop right back on a quick server restart tho.

Same here, as you can see from the R21 update and yesterdays death.

I'm hoping, in a roundabout way, that this is either something I have done or something KC have done... because then it can be fixed. Unfortunately my experience of KC is that they're far too professional and 'on the ball' for this to be their error, and I don't think it's something I could have caused. I'm looking in DICE's direction.

[KC-Justin]

Interesting. This used to happen in some of the very early BC2 builds. Did you just notice this since R21? Looks very much like a memory leak and probably from a specific map (as it's very gradual.)

- Can you post your map rotation for both the servers?
- Do you know which map the server crashed on?

[RaWClan]

I know its not quite as obvious on our server but it is still def there. When the server is populated it does increase mem useage over time until the server crashes or is reset as you can see below then it seems to drop back down (The server crashed Thursday night as you can see in the chart below.
You can also see that where our server has been full for the last 24 hours the mem is slowly creaping up (5th&6th Sept).
This server is running one map only (Valpariso) in Rush Mode 32 players.
Would be interested in knowing the casue for this and wether its something that we are doing to casue it. Some kind of logging or something to do with having an rcon tool running all the time?
Strange little issue

[EvilEddy]

Unfortunately I don't know what map it was on when it crashed. I'll see if any bandits were on at the time.

I hadn't checked prior to R21, but the graph shows the problem prior (I think that other 'drop' in memory usage was the R21 update).

The rotation is as follows;

Conquest - Laguna Alta, Arica Harbor, Laguna Presa, Panama Canal, Atacama Desert


Rush - Isla Inocentes, Valparaiso, Arica Harbor, Laguna Presa


Both definitely have the issue.

This post has been edited by VanKlomp: 06 September 2010 - 01:07 PM

[KC-Justin]

RaWClan, on 06 September 2010 - 11:54 AM, said:

I know its not quite as obvious on our server but it is still def there. When the server is populated it does increase mem useage over time until the server crashes or is reset as you can see below then it seems to drop back down (The server crashed Thursday night as you can see in the chart below.
You can also see that where our server has been full for the last 24 hours the mem is slowly creaping up (5th&6th Sept).
This server is running one map only (Valpariso) in Rush Mode 32 players.
Would be interested in knowing the casue for this and wether its something that we are doing to casue it. Some kind of logging or something to do with having an rcon tool running all the time?
Strange little issue

It's certainly a lot less obvious and still within the realms of ok. The typical mem usage for a BC2 starts at 150MB and runs to about 180-190 max. With VC's, it's hitting over 300MB and then falling over.

There is also a little known exploit that exists for maps Valparaiso and Port Valdez that can be used to crash servers. I don't know the cause at the mo but maybe your server has been the victim of that exploit.

[EvilEddy]

If it helps, we would be willing to run suspected maps for specific periods of time over the weekend, to identify if it is one or more maps that is causing the issue.

The only problem will be keeping he server full though, so if we could get a few players on from local KC clans to keep the server busy.

Would this help Justin?

[KC-Justin]

What we can do is keep your current rotation and enable runtime logging. This will have to be done the next time your server crashes, and when it does, please contact support asap so that an admin can enable it (via IRC would be pref.). Since both servers suffer this problem, I would suggest enabling logging on your 2nd server (least popular one.)

When it crashes with the runtime logging enabled, it will create a dump file which we can pass to DICE to look over.

[EvilEddy]

I expect it will empty out overnight, so if it does I'll get in touch with support tomorrow morning.

Thanks Justin.

[EvilEddy]

VanKlomp, on 07 September 2010 - 11:06 PM, said:

I expect it will empty out overnight, so if it does I'll get in touch with support tomorrow morning.

Would ya believe it? It didn't empty.

[EvilEddy]

Rush died. Fingers crossed!

This post has been edited by VanKlomp: 11 September 2010 - 05:12 PM

[Sgt.Trojan]

OUrs too Back up now though

[EvilEddy]

Sgt.Trojan, on 11 September 2010 - 05:46 PM, said:

OUrs too Back up now though

Yes, false alarm I think.

Now that the game panel graphs have moved on, I can see there was no drop in memory usage. The server didn't crash at all, so I think it was a famed plasma glitch.

I was hopeful it was time to get some logs off to DICE.

Bleh.

Onwards and upwards!

[EvilEddy]

There was, however, a drop in usage yesterday.



Justin, was that a crash?

[KC-Justin]

Hi,

Apparently not and your runtime log is now 0.5GB in size and growing (having issues opening with a text editor.)

What you might of had happen is where players get kicked but the server remains up and running and still responds to rcon queries. You can confirm this by checking the PB logs at the time of the suspected crash. Normally when this happens though, the server would require a restart for players to reconnect.

The above is an issue we've been trying to narrow down with DICE for ages. As there is no crash dump or even entries in the runtime log, and the only logs are of PB disconnecting players, it's very hard to track down. DICE are aware of it but have no idea why it's happening.

[EvilEddy]

Thanks Justin.

I can't believe it, I'm willing my server to die...

[EvilEddy]

I think it died.

Mission accomplished?

[EvilEddy]

VanKlomp, on 14 September 2010 - 10:31 PM, said:

I think it died.

Mission accomplished?

Just came off IRC, Godsbane is sending the dump file off.

[EvilEddy]

Will DICE feed back to you about this Justin?