fixed Invalid Ballistics Objects being created and not cleaned up resulting in FPS impact.

[Tom P]

Chirp.

[Mike_Romeo]

Still after todays update ?

[rurounijones]

@BIGNEWY Ballistics objects are still accumulating in DCS 2.8.0.32937 Open Beta

This thread has been tagged "investigating" for over 2 months and 2 releases. If ED are having trouble replicating then please inform us so that we can provide more information.

Edited November 17, 2022 by rurounijones

[Flappie]

I'd like to help. Can you please provide a simple way to account for ballistic objects kept in memory? I mean by that a MIZ file with some scripting magic and a logger saying "15 ballistic objects in memory, 16 ballistic objects in memory, 32 ballistic...".

[abelian]

4 hours ago, Flappie said:

I'd like to help. Can you please provide a simple way to account for ballistic objects kept in memory? I mean by that a MIZ file with some scripting magic and a logger saying "15 ballistic objects in memory, 16 ballistic objects in memory, 32 ballistic...".

I wrote some software that might help you:

https://github.com/bobmoretti/dcs-tetrad

It will create a console and displays various statistics every 5 seconds. It also saves a compressed CSV file with frame rate and object/ballistics counts for every frame.

You can also enable a GUI that makes a running plot of this information over time (don't recommend running this on a server without a GPU).

Please ping me if you have any questions.

[rurounijones]

10 hours ago, Flappie said:

I'd like to help. <snip>

I would wait until we get some information from ED on what, if anything, needs to be provided before any more time is spent by the community.

They already have 3 Hoggit trackfiles and a liberation trackfile that is basically a miz by another name.

I am not sure providing more of the same will be useful without feedback from ED.

[Flappie]

8 minutes ago, rurounijones said:

I would wait until we get some information from ED on what, if anything, needs to be provided before any more time is spent by the community.

They already have 3 Hoggit trackfiles and a liberation trackfile that is basically a miz by another name.

I am not sure providing more of the same will be useful without feedback from ED.

Suit yourself...

Edited November 18, 2022 by Flappie

[Tom P]

Bump

[392198523]

The issue is somewhat acceptable on our server, which has around 400 Units and 30 players. The ballistics object increased to about 100 in each restart cycle which is 3 hours.

[Napillo]

On 11/17/2022 at 6:29 PM, abelian said:

I wrote some software that might help you:

Lua is garbage collected, if you're referencing all the variables at least once every 5 seconds, then you would be creating a memory leak.

A better way to test would be to call collectgarbage("count") which only gives you the statistics for memory used. 

Local variables cannot be garbage collected until they fall out of scope. Generally, you want local variables in a function, global variables elsewhere, that way when the function ends, you de-reference and mark for garbage collection the local variables.

Edited November 23, 2022 by Napillo

[rurounijones]

@Napillo@abelianJust in case you guys want to continue that conversation; could you do so in Private Messaging or another thread dedicated to the tool. I don't want to derail this thread from its original purpose, thanks.

Edited November 24, 2022 by rurounijones
spellng

[Tom P]

Soo I'm guessing this be labeled under investigating just means they know its there but nothing will happen for awhile.

[Mike_Romeo]

Pretty sure they will search for a solution even though it wont be high priority

[MetalStormGhost]

I wonder if this is related to ATGMs not being cleaned up after they have hit. I am experiencing this after updating to 2.8:

This used to happen with some WWII bombs also before 2.8, but now I'm also seeing it with multiple ATGMs (screenshot includes the AT-3, AT-5, AT-11 and the TOW).

Edited December 3, 2022 by MetalStormGhost

[Flappie]

Track please?

[rurounijones]

@MetalStormGhostMaybe but I doubt it since:  The coordinates don't match, the ballistics objects in thsi thread are not associated with shooting or weapon firing events and this has been going on a lot longer than 2.8. I would start a separate thread for investigation into what you are seeing.

[BIGNEWY]

Hi, 

the team have been unable to reproduce this issue. We will continue to watch for it in our sessions however. 

thank you

 

[rurounijones]

Just learnt of a hack to get some more information about the rogue ballistics.

They are allegedly "Category 3" weapons which translates to bombs, assuming that is not another default value like the map origin positioning and that it is using the Weapon.Category enum. If it is the Object.category then it is STATIC.

The APIs for the object seem a bit schizophrenic about whether the object exists or not.

local wep = { id_ = 33566465 } // The ID of one of the rogue ballistics objects.
return Weapon.getCategory(wep)

// getCategory = 3 (A bomb according to https://wiki.hoggitworld.com/view/DCS_Class_Weapon )

// isExist = false

// getDesc = Weapon doesn't exist

// getName = same as ID

// getLauncher = Weapon doesn't exist
// getTypeName = Empty string

// destroy = No error but Weapon is not destroyed

Edited December 6, 2022 by rurounijones

[MetalStormGhost]

On 12/3/2022 at 3:21 PM, Flappie said:

Track please?

I opened a new thread for the ATGMs in case it's an unrelated bug to this one. The track can be found here: https://forum.dcs.world/topic/314831-atgms-not-being-cleaned-up-on-impact/

[Tom P]

Any new or remaining issues since the new patch?

Edited December 22, 2022 by 162nd Pete

[Tom P]

Any changes?

[Flappie]

Not yet, but devs are looking into it.

[Harlikwin]

On 12/31/2022 at 2:01 AM, Flappie said:

Not yet, but devs are looking into it.

Point them to the MP server Engimas cold war... 

Have them check performance with the baseline control ~within 1h of reset (server resets every 4h)

And then after like 2-3 hours of play.

What they will see in VR is the following (shadows on)

CPU frame times right after reset are "normal" ~15ms CPU render time

After 2-3 hrs this will rise to 30-40ms when looking toward the "frontline" (this is terrible like 20fps in VR)

However if you turn off the shadow sharers in VR, it returns to normal ~15ms...

 

So something in 2.8 broke shadow sharers when looking at large numbers of "objects" that presumably aren't being cleaned up

 

 

[rurounijones]

On 12/31/2022 at 6:22 AM, 162nd Pete said:

Any changes?

Still occuring on the latest version. ED are unable to replicate even with trackfiles so unless they add some specific debugging code to the next version or someone figures out the cause some other way; I don't see this going away anytime soon.

Edited January 8, 2023 by rurounijones

[Flappie]

On 1/8/2023 at 2:24 AM, rurounijones said:

Still occuring on the latest version. ED are unable to replicate even with trackfiles so unless they add some specific debugging code to the next version or someone figures out the cause some other way; I don't see this going away anytime soon.

Devs were able to reproduce this issue, and it was internally reported. I'll ask for the Reported tag.

Edit: I'm being told by another tester that the issue described here may be different from the one that is already reported. Standby.

Edited January 9, 2023 by Flappie