Unity NetCode 0.0.2-preview.1

I understand, disagree and think this setup is bad. I feel "Server Build" should be replaced with "Headless Mode" in build settings, which is more generic for builds. If your logic is to have us users to add defines manually to specify a Client build, what is the problem with us specifying a Server build with defines as well? By this only UNITY_EDITOR takes precedence if anything and everything is modular, (subjective more)logical and extensible.

 

There seems to be a bug that the name of the gameobject is used when creating snapshot data, instead of what is defined in the Ghost Authoring Component.
This makes all instances of the prefab require to have the same name in the hierarchy.
This also happens when pressing the Generate Code button. It will mess up things if the object has spaces etc. in its name. Happens in 0.0.2 as well as 0.0.3.

 

Is there any way to get current clients on server?
Or should we count the connections when new client connects?

 

The problem is on Video RAM usage. I have 512mb of VRAM, while netcode sample gets 0.9Gb in editor. But why?

 

Hi @timjohansson. I see matchmaking package has been removed from repo. Is there any plan return and update the package again?

 

Just to make sure I understand that correctly:
Right now i have an NPC prefab which i generate authoring stuff on, and that authoring is added to my ghost collection. Then i create a gameobject instance of the prefab, naming it TestBoi and changing some values (render mesh etc). Right now this generates the error as shown above when pressing play. If i rename the instance back to NPC, it works, but the changed values dont get applied (mesh is taken from prefab, not from instance after conversion).

The intended behaviour would be the GO with name TestBoi is converted based on the NPCAuthoring as defined on the prefab (with a different name) and then getting its modified custom render mesh data etc. applied, right?

 

This has caused quite a lot confusion for some.
On that occasian I want to say, I'd much prefer the sample doesn't have stuff like this and sticks to conventional Unity rendering.
For this sample I expect a branch with pure ECS and one with hybrid GameObjects.

 

It's called "Server build" in the UI but it's actually a BuildOption named "EnableHeadlessMode" if you want to set it by script.
While the defines solution may or may not be something you like, all you have to do is to override ICustomBootstrap and create your own version. The only downside is that you lose the editor window that allows you to run multiple clients and other things.
You could even copy paste the whole ClientServerBootstrap and change the parts with the defines.

Talking about this, I feel like the NetCube sample is missing the DisconnectSystem (could use the same as the one in the asteroids sample).

I have 4 things you all should be careful about if you use latest version of every packages:

• SRPBatcher seems to fail (latest URP) and put everything at (0,0,0) with a scale of (1,1,1).
• Physics package version 0.2.5 has an issue preventing us from using it but a fix is on its way (check the DOTS physics forum about BlobAssetStore).
• Since the build manager don't have the correct build steps to create a server and client, you can create your own pipeline with a custom step that will handle adding the server option (check the attached file for a simple example but be aware that I removed livelinking for now). Don't forget to add UNITY_CLIENT define in your player settings.
• Last but not least: things are running in the update loop so don't forget to add a targetFramerate to your clients because, while it may work in the editor, it will run too fast in a build may break the networking (like input commands for example).

Also I have a question for you @timjohansson, I don't think there will be a way to selectively send the data to a client based on some condition (like they are too far, don't send all the world data to them, only what's around). We will have to do that ourselves from what I've heard. What would you think is better for the server:

• Having multiple worlds and duplicating all the systems ? Meaning that jobs may run on less entities.
• Find a way to put the netcode serialization / deserialization in specific worlds but things like physics in the same world (don't even know if it's possible with the netcode architecture, don't think so).
• Find a way to create subworlds that will not be sent to clients based on some conditions (haven't checked yet what I could use to prevent sending the data to a specific connection, may need to override some systems of the netcode package which I don't want to for obvious reasons).

Thanks for helping!

 

That helped. Thank you!

 

Yes, that it should be renamed to what it actually does is sort of what I was referring to. And I've already overriden and extended the bootstrap with my own definitions and worlds, and also created an extended multiplayer window for enabling my worlds as well. My post was merely feedback on my experience doing just that, which would reduce the amount of work someone would have to put in to extend the code.

 

My usecase is for a server simulating multiple small matches (few players, lots of matches) and since there is some Physics, putting them too far from the origin is not a good idea (IMO) but I will do what you said and look at a basic distance based importance solution.

Do you think it's possible (at least try) to make my own PVS without modifying the netcode package ? Or maybe netcode will have its own version in less than 3 months ? I'm just trying to estimate the time I should allocate to this feature.

 

Two suggestions/questions about the customization of spawn/despawn client entities:

1. The spawn Process on client side can easily by adjusted by creating a partial class of the auto generated GhostSpawnSystem. But all virtual methods (i.e. UpdateNewInterpolatedEntities, MarkPredictedGhosts) are called before the first snapshot data is applied. In case of a hybrid approach where i want to spawn a GameObject at the entity position, i need a virtual method to override after the snapshot data is applied. Otherwise the GameObject is spawned at (0,0,0). Is there a chance to also add a virtual method after the snapshot update?

2. Are there any Plans to customize the GhostDespawnSystem, too? This would be nice to fire a dying animation or to destroy the synced GameObject in a hybrid scenario.

 

I'd like to propose some changes to the netcode:
Right now, there are 3 types of spawn and they are handled like this: EntityManager.Instantiate(m_predictedPrefab, predictedEntities);
It's pretty much hardcoded to ECS only so I added 2 calls:

Code (CSharp):

1. protected virtual Entity SpawnGhost(EntityManager entityManager)
2. protected virtual Entity SpawnPredicted(EntityManager entityManager)

and replaced those instantiates with:

Code (CSharp):

1. for (int i = 0; i < m_CurrentPredictedSpawnList.Length; i++)
2.                     predictedEntities[i] = SpawnPredicted(EntityManager);

In the 2 new calls, the default model of the netcode is implemented. With overriding these 2 functions in a partial class from the GhostSpawnSystem this gives us devs a chance to write additonal pre/post processing code or, like in my case, write your own hybrid gameobjects spawn code.
It at least solves the hybrid gameobject problem some of us are having without much changes in the netcode.

I'd also like to know how spawning of different objects with the same snapshot archetype is handled in the 3rd person multiplayer demo. Tim suggested a component that is synced and has data for the representation to defer the loading so I guess you went that way. What I don't understand about it, why is that data part of the snapshot, the data that is only relevant on initialization.
A suggestion would be to have data we can mark, that is not in every snapshot but only when some call/flag/component is present. It would be always present on first inits but on subsequent sends this data would be empty till a server tick triggers any changes. Kind of like having full and partial snapshots.
I don't have too much insight in the inner netcode so sorry when this is already happening. The snapshots are still a little mysterious to me with memcpys and all.

And as @Flipps suggsted, having the latest snapshot available would be a good thing to have but I see the problem with it. The snapshot can't be applied to an entity that doesn't exist yet. I'm interested if there's a way we could still access that data from the spawn process another way..

 

This would help my current new issue: physics joints being entities created with conversion the workflow don't quite work with the ghosts, having this would make things a little better even if it won't completely solve the issue and isn't must have.

 

Congratulations on this release. Enjoy the fun of production hardening... It's the good kind of pain that only makes you stronger.

 

Thank you for taking the time to answer Tim!
I think your answer solves my questions. I'll move over to UpdateNewPredictedEntities, seems more reasonable.

Good to hear the compression kicks in! That's really nice this is already working.

 

I like the new RPC setup with the RpcCommandRequestSystem where you only have to add the RPCData and requestComponent, and it destroys itself. What would be nice however would be to add the EndSimulationCommandBufferSystem to RpcExecutor.Parameters as well. In many of my cases (and I think for other people as well) RPCs are one frame events and should be destroyed if nothing picked it up (in case of disconnects etc).

I guess one could add a second component and system to add the destroy to the end of frame instead, but might want to be the default behavior?

Also: Return the Entity created from the "ExecuteCreateRequestComponent" so we can add additional components^^ Else we're back at boiler plate code

 

Hi, I'd like to stop server in NetCube.
I called network.driver.dispose() in "OnUpdate" game.cs, however there are some errors that other entities or components still may access network.driver.
Is there any official way to stop server in netcode package or NetCube sample?

 

I am having some issues replicating the NetCube sample in URP/HDRP. Everything seems to work in the editor a cube will spawn and can be moved around. But once I build (and run the editor as the server), a cube will spawn on the built client (and a log message will output in editor that a client has connected) but the client cube will spawn at 0,0,0 and cannot be moved around. I am not sure if I have been doing something wrong or if something isn't functioning properly. I do know that if I take the same code and copy/paste it into a normal 3D project (that is not using any Render Pipeline Asset), build, and run (same as before), the cube will spawn above the plane and is able to be moved around. Any help is appreciated.

I am using 2019.3.0f1. But I have used 2019.3.0b11, 2019.3.0b12, 2020.1.0a14, and 2020.1.0a15 (URP/HRDP) with the same results.

EDIT: I am also using Unity NetCode 0.0.3-preview.2

 

You can't anymore for HDRP but on URP you can still disable SRPBatcher, that will fix your issue.

 

Thank you so much!! I can confirm this fixes the issue. Could you explain to me what this does and why this has an effect?

 

More info:
https://forum.unity.com/threads/are-unity-subscenes-and-lwrp-compatible.717422/#post-4794479
https://forum.unity.com/threads/dot...forms-are-reset-to-defaults-in-builds.757268/

 

I had two issues this morning following the code.
The main one is that when it generates the ghost collection components list it uses the project name.
The problem was my project has a space I it's name and so it tryed to created structs with a space in the name and... Well it didn't like that! Ha

edit: i found the PrefixName field which defaulted to the project name, and was the problem. removed the space and all good! - but likely worth doing a replace on invalid chars, or throwing an error on the generate.

 

i think you are looking for second paragraph here
https://docs.unity3d.com/Packages/com.unity.netcode@0.0/manual/network-connection.html

 

@TZ-
No, this is for client, not "server".
I'd like to stop server in Netcube.

 

Tried both your suggested solutions and both work like a charm! Thanks for the great feedback!

Have never used the SystemStateComponentData before but noticed that there is a "CleanupEntity" Tag added to the entity when the Entity is destroyed (and SystemStateComponents attached).



I thought it would be really nice to run systems on With<CleanupEntity> rather than to check for None<SnapshotData> but CleanupEntity seems not be public and can't be accessed.

 

@timjohansson Please, some ETA on the DOTS Sample Project ...

 

Just started on the project, still attempting to replicate the red cube demo in the video
If your project name have spaces, like "Untituled DOTS MMO", you get generated classes like "EnableUnity Untituled DOTS MMOGhostReceiveSystemComponent" which causes tons of compilations errors. For now I will remove white spaces from my project name

Edit: I completely missed the Name Prefix field on the GhostCollectionAuthoringComponent

 

https://github.com/Unity-Technologies/DOTSSample

 

@timjohansson It's released! So now you can point me at the code behind lag compensation?

 

I was taking a look at the DOTSSample and I'm not going to lie, It has become more complex than the FPSSample.
It is seems to be more modular than before. But everything just seems too much spread out.
Also as ECS is evolving, I'ts even taking me longer to understand how the code works and what it actually does.

Also the project should use URP instead of HDRP, as it is just a sample and to understand the FPS NetCode architecture also because I'm having long build times because of those HDRP Shaders, 13.50 min~

 

Ah, very true. One would need the EndSimulationCommandBuffer for the next frame, which I think isn't currently possible.

 

Why are the GhostSendSystem and BuildPhysicsWorld so exceedingly slow (These seem to be the biggest offenders on server/client side respectively, but it's hard to pinpoint exactly)? On a mostly empty project (DotsCube sample), one server and one client uses a 16ms per cycle
Setting for 4 clients and 32 sends it up to 132ms, even though it's exceedingly simple to have 40+ entities actually doing things on my other projects.

It is just some upfront cost, and it will hardly scale costlier as I make the game world more complex, or there is something wrong with my project? I understand that the project is simulating 36 client machines and a server, so I worry little about the 110ms cost of running the clients, but it's a huge issue if the server slows to a crawl just for having 36 people connected to it doing nothing at all

 

Unexpected exception Burst.Options.OptionException: Unexpected arguments: `--is-for-function-pointer --managed-function-pointer=0x000001CFD8544860`
at Burst.Compiler.IL.Jit.JitCompilerService.Compile (Burst.Compiler.IL.Jit.JitCompilerService+CompileJob job) [0x0012a] in <3179d4839c86430ca331f2949f40ede5>:0

While compiling job: Unity.NetCode.RpcSetNetworkId, Unity.NetCode, Version=0.0.0.0, Culture=neutral, PublicKeyToken=null::InvokeExecute(Unity.NetCode.RpcExecutor+Parameters&, Unity.NetCode, Version=0.0.0.0, Culture=neutral, PublicKeyToken=null)

Getting this error along with a few dozen not disposed errors after installing preview 0.0.4 and burst 1.20 preview 11.

Is there a valid burst and net combo that doesn't throw compile errors?

 

@timjohansson Any plan to always keep updating this Unity NetCode package to sync with latest Entities package? Currently latest Unity NetCode 0.0.4 still using older Entities 0.3.0-preview.4 package.

 

I'm running Burst 1.2.0, Collection 0.5.0, Entities 0.5.0, Jobs 0.2.3, NetCode 0.0.4 without errors in editor, or are you talking about building player?