PLS fix sever crashing at min 30+

[Atrik]

4 hours ago, Dr.Ahmadinejad said:

have a solution for Windows users: @Barcodes

1. Set up a virtual machine running Linux operating system (choose a light distribution for example Lubuntu); allocate 8GB of RAM, 4 CPU cores and 40GB disk space to it. 

2. Install 0A.D. inside the Linux virtual machine. You can find it at the first place in software centre or use the terminal  command: sudo apt-get install 0ad

3. Host a game using your Linux machine, but put yourself (hosting you) as unassigned spectator. 

4. To play, join your game from your real Windows machine as a guest player. 

5. Press start from the virtual machine when you are ready. 

 

You CAN play on windows! You just need to use Linux. EASY.

[Norse_Harold]

7 hours ago, Atrik said:

You CAN play on windows! You just need to use Linux. EASY.

Lol, yeah. For the about the same amount of effort and storage space, one might as well install a dual-boot of Linux on their computer so that the game runs at a normal framerate instead of trying to limp along with a virtual machine. Also, rejoining with Windows typically only gets a few minutes more of gameplay before the game runs out of memory again. And, I think that there are diminishing gains to the rejoining tactic because the only sources for free memory with a big match to simulate are garbage collection and memory leaks, of which there are not many if the process has only been running for a few minutes. This is a guess, though, I haven't checked in detail.

If someone wants help with setting up a dual-boot of Linux and Windows then talk to me on IRC or other means of communication.

Edited February 24 by Norse_Harold

[chrstgtr]

Mac host. Game crashed at min 42:11. For about the last time minutes of play, other players were dropping like every minute

2024-03-05_0011.zip

[Stan`]

@chrstgtr How much RAM does your Mac have ?

[Stan`]

https://trac.wildfiregames.com/ticket/6906

[real_tabasco_sauce]

@chrstgtr maybe see if one or two of the other players can post their replays here.

[chrstgtr]

 

19 hours ago, Stan` said:

@chrstgtr How much RAM does your Mac have ?

16GB

[Stan`]

Mmmh. If you replay the game on your machine does it still crash ?

[chrstgtr]

59 minutes ago, Stan` said:

Mmmh. If you replay the game on your machine does it still crash ?

Yes, I reach the normal "You have reached the end of the replay" message, which makes sense because the game never truly crashes--it goes OOS and makes it impossible to continue

[Stan`]

Wait you said the macos host crashed

[chrstgtr]

17 minutes ago, Stan` said:

Wait you said the macos host crashed

Yeah, it was unartful language by me. My fault. It is the OOS error @real_tabasco_sauce @Barcodes, and I mentioned earlier in the thread where all players go out of OOS around min 40. For the 10 minutes before that happened, various players went OOS or actually crashed.

Sorry.

[Stan`]

It's alright In this case this probably means the other went out of memory, so either it got corrupted in a way that was recoverable (hence the oos message) or it wasn't and therefore it crashed.

[real_tabasco_sauce]

Do you mean other players? Are there settings which minimize memory usage? maybe turning off model appearance randomization? Maybe I should recommend windows users try those settings to help prevent oos?

[Stan`]

Yeah once you reach 4GB of used RAM for the game, there is nowhere else to write data (Imagine having filled a notebook and the only place left blank is the table but your pen doesn't work on that surface. So at best you manage to find some blank spot in your notebook, or you write over something else and you go OOS.

I'm not sure what you can do to minimize RAM usage on windows. Since most games have no AIs they shouldn't be the culprit. Hopefully some bits of code aren't behaving and leaking memory and we can figure it out, if not this just means the simulation is complex and takes that much space. @wraitii or @Itms or @elexis might know more or how to get more info.

 

[chrstgtr]

1 hour ago, Stan` said:

Yeah once you reach 4GB of used RAM for the game, there is nowhere else to write data (Imagine having filled a notebook and the only place left blank is the table but your pen doesn't work on that surface. So at best you manage to find some blank spot in your notebook, or you write over something else and you go OOS.

I'm not sure what you can do to minimize RAM usage on windows. Since most games have no AIs they shouldn't be the culprit. Hopefully some bits of code aren't behaving and leaking memory and we can figure it out, if not this just means the simulation is complex and takes that much space. @wraitii or @Itms or @elexis might know more or how to get more info.

 

That makes sense. But why would it happen to all players at the exact same time? If it was one player causing it, wouldn't it just list that single player as OOS? 

[Feldfeld]

17 minutes ago, chrstgtr said:

That makes sense. But why would it happen to all players at the exact same time? If it was one player causing it, wouldn't it just list that single player as OOS? 

If the host has a simulation issue then that would show every other player OOS. Not saying that's necessarily the case here

[chrstgtr]

26 minutes ago, Feldfeld said:

If the host has a simulation issue then that would show every other player OOS. Not saying that's necessarily the case here

Got it. Sometimes game-ending OOS errors show that the host and only the host is OOS. But other times everyone, including the host, is OOS. 

[real_tabasco_sauce]

Windows 11 host (not me) with 4x32 ddr4 ram crashed today at ~30 minutes

metadata.json commands.txt

Everyone was in the OOS error and it all happened at the same time. Also, the game was not playable afterwards with everyone losing connection. The host said the game didn't crash.

Edited March 10 by real_tabasco_sauce

[hyperion]

On 08/03/2024 at 6:04 PM, real_tabasco_sauce said:

Do you mean other players? Are there settings which minimize memory usage? maybe turning off model appearance randomization? Maybe I should recommend windows users try those settings to help prevent oos?

Reports seems to have increased a lot since last community mod release, maybe try without. What should help a lot is reduced map size, less players or lower pop cap.

[real_tabasco_sauce]

It may be possible that the community mod might contribute, but I can't think of a mechanism. I think this is more likely a case of more reports arising from this discussion and trying to get to the bottom of it. For example, I only uploaded the replay and thought to ask about memory specs because I knew this discussion existed.

[chrstgtr]

1 hour ago, hyperion said:

Reports seems to have increased a lot since last community mod release, maybe try without. What should help a lot is reduced map size, less players or lower pop cap.

I don't think anything has changed. It has been a known problem for a long time. It is just being reported here now.

To the extent the mod did cause more occurrences, I think it is because the new mod makes some games last longer/more units to be produced/etc. (i.e., the mod is causing the game conditions under which the error occurs instead of the mod itself causing errors)

Edited March 11 by chrstgtr

[hyperion]

8 hours ago, chrstgtr said:

It is just being reported here now.

There were quite a few threads beside this one and even bug reports recently. Sure this doesn't mean there must be a correlation. Once the game is at max pop there shouldn't be much of additional memory be needed based on game length tho unless there is a leak, so I wouldn't dismiss the possibility.

Windows user are rare in my environment so I have not personally experienced this issue at all and hence I can't make any sensible statement of whether it having changed or not other than the frequency of reports.

[chrstgtr]

37 minutes ago, hyperion said:

There were quite a few threads beside this one and even bug reports recently. Sure this doesn't mean there must be a correlation. Once the game is at max pop there shouldn't be much of additional memory be needed based on game length tho unless there is a leak, so I wouldn't dismiss the possibility.

Windows user are rare in my environment so I have not personally experienced this issue at all and hence I can't make any sensible statement of whether it having changed or not other than the frequency of reports.

Sure. It’s possible. But I haven’t noticed any change. And two of the three people here are just echos of the just echos 

[Stan`]

19 hours ago, hyperion said:

Reports seems to have increased a lot since last community mod release, maybe try without. What should help a lot is reduced map size, less players or lower pop cap.

If you still have it around, it would be interesting to have the 64bit A26 build you made to see if that reduces the issues ? If it's that frequent, it would make it easy to rule it out ^^

Not sure if @vladislavbelov has a usable 64bit build.

[hyperion]

2 hours ago, Stan` said:

If you still have it around, it would be interesting to have the 64bit A26 build you made to see if that reduces the issues ? If it's that frequent, it would make it easy to rule it out ^^

Was based on svn before A26, closer to A25 I believe. The binary I deleted in the meantime but could be rebuilt, though might have to work around a certain python - sm issue.

From my limited knowledge of Windows memory model, it allows sort of overcommit if backed by hdd, maybe sort of dynamic swap is a better term. For me 0ad typically takes about 7 GB of virtual memory on Linux 64bit and some 1.5 GB residential memory. Therefore I think it's very reasonable to assume that a 32bit builds will have trouble with the limited address space but a 64 bit build would work reasonably well with 4GB physical ram even on Windows.

So I have no doubt that 32bit is an issue which people might run into occasionally, was just wondering if the last community mod might have moved the tipping point based on recent noise but according to @chrstgtr that doesn't seem to be the case. I think the last notable rise in requirements was with A24 and earlier version should be much less likely to run into this.