Unity NetCode 0.0.2-preview.1

As my usecase for Unity is rather abnormal, I'd assume this doesn't have a DOTS solution to connecting to an existing TCP secure websocket correct? First glance of the documentation would indicate it has nothing related to that, but thought I might ask on the off chance there was something already built within the package

 

Yay! Thank you very much! Time to update!

 

I do not seem to be able to find the package in the package manager (even though I have preview-packages enabled).
Nor am I able to find any release / pre-release of Unity 2019.3 in Unity Hub, is this an issue anyone else is experiencing?

 

Tim, can you give me a summary of the bigger changes that happened with 0.0.2?
If you can't, no biggie, just interested.

 

There's no code for lag compensation, is there?

 

If I wouldn't have forgotten, then I will surely remind you . Thanks for that though.

By the way, any ETAs on when the sample project will be released?

 

I can't find the package either, I'm on version 2019.3.0f1 though.

 

How about com.unity.transport package? Is that package is available via Package manager too? Also, Is it safe to use com.unity.transport alone without netcode package?

 

"com.unity.entities": "0.2.0-preview.18",
"com.unity.netcode": "0.0.2-preview.1",
"com.unity.rendering.hybrid": "0.2.0-preview.18",
"com.unity.test-framework": "1.1.8",
"com.unity.transport": "0.2.1-preview.1",

 

I have the new netcode up and running and it was a much better process than the last time. I like most new changes. Less code-gen, no seperate prediction systems.

Sadly I still ran into an older problem I couldn't quite figure out. There seems to be a 1:1 relation between a type of ghost and its snapshot to a (converted) model.
If I want to have several moving platforms with a generic mover ghost I don't know how to make them have seperate meshes. There doesn't seem to be a pipeline for this and I'm wondering if it's better to wait because there is something up and coming that solves this or roll my own implementation.

 

GameObjects with GhostAuthoringComponent are converted twice in each of my client worlds, so I have two Entities per GameObject per World.
Is there a way to stop that from happening? Do I have to uncheck all the boxes for Interpolated Client / Predicted Client manually?
Why would a component need to be interpolated and predicted at the same time?

 

Any plan to add support for hybrid ECS (i.e. Convert and Inject Game Object conversion mode) for GhostAuthoringComponent? At least expose Convert and Inject Game Object conversion mode selection option just like how Convert To Entity does. Currently I have project implement in hybrid ECS way and I not able to use this netcode package as it needs to be in pure ECS. I think until there is most of Unity features available in dots, it's must have feature.

 

Tried to manipulate the child data of a GhostAuthoringComponent but the checkboxes are disabled.
I also disabled the root objects translation/rotation first in case there are any depdencies.
Is this a bug or i am doing something wrong?
Using the newest packages.

 

Oh, I see. The Problem is that one authoring component on my ghost prefab adds a singleton to the entity and thus I get an error that the singleton exists twice.

Maybe I am just using the authoring components wrongly. Probably should add that one on Instanciate then. Haven't used entity prefabs yet. Thanks for the clarification!

 

Me neither. Can someone explain please? Is this package present in 2019.3.0f1 or it is only in 2020 alpha?

 

I specifically downloaded 2019.3.0b12 because b11 was in https://github.com/Unity-Technologi...pleproject/ProjectSettings/ProjectVersion.txt

Have you checked "Show preview packages"?

 

I have 2019.3.0f1 which should be the newest version. Yes, I have checked "Show preview packages". I can see there for example:
"com.unity.entities": "0.2.0-preview.18",
"com.unity.rendering.hybrid": "0.2.0-preview.18",
"com.unity.test-framework": "1.1.8"

....but
"com.unity.transport": "0.2.1-preview.1",
"com.unity.netcode": "0.0.2-preview.1",

is nowhere to be found.

 

2019.3.0f1 >> I've been able to install these packages through the manifest.json but I couldn't find them in the Package Manager too.

 

Sorry, I've added them "manually" too because I copied the package manifest from the sample. Seems they don't show up when they are under testables.
https://github.com/Unity-Technologies/multiplayer/blob/master/sampleproject/Packages/manifest.json
Annoying, but easy enough to add them.

 

There's still no implementation of support for fragmentation

 

Just tried the samples and it all unusable on MacBook Pro 15 2012 (non retina) (10.14.6, Metal)

 

@timjohansson
Hi, I'd like to ask 1 question.
Is DOTS netcode for all people who creates networking game with unity?
or just for existed/future DOTS users?
We are currently working open world TPS game with unity but didn't use DOTS a lot.
I'd like to know whether we should use dots netcode or we should find another way.

 

Not sure if helpful to anyone. Trying to extend the package with some of my own worlds for other servers (like Login, Communication, Matchmaking). Internal servers only use RPCs. I generate simulation/initialization groups, and extract the RPC network part of the NetCode and add it in specified folders. Parse and remove unneeded code (which is very prone to what they might update it with). (But hopefully this will reduce my refactor time in future x)) This is the window script https://pastebin.com/TLSek3F2

Some things I've noticed for extending are that "Runtime/ClientAndServerWorld" could almost be it's own package which netCode is dependent on, could make using your own worlds and bootstrap a bit easier by simply replacing it with your own package. "UpdateInWorld.cs" should seperate the systemGroups into their own scripts.^^ I don't like the "UpdateInWorld" attribute as a compilment of "UpdateInGroup" as it can express same thing, but I can see how it is problematic without it as well. I've yet come to overwrite the bootstrap yet I'll see how that works out with combining it with my "UpdateInExtendedWorld".

Thanks for the awesome work! : )

Spoiler: Window

 

Thanks for the nice explanation.

I am struggling at creating a build for client only. What is the best way to do this (for now)?

My thoughts:

1. Using the new Build Pipeline with the new NetCodeConversionSettings. I think this is not implemented yet because i get a build where client + server worlds are running

2. Doing it like in the asteroids demo:
- import Settings to create a client.dll and server.dll
- delete client or server dll after the build run

EDIT:
Just figured out that you just have to declare UNITY_CLIENT or UNITY_SERVER in PlayerSettings Scripting Define Symbols

 

The quick start guide in the manual needs some attention; all the images are missing...

 

I'm at that point too, see my post here: https://forum.unity.com/threads/dots-netcode-copenhagen-updates.760112/page-3#post-5242976

I've tried something else: creating a custom bootstrap, 2 boot scenes (Client/Server scenes), creating a MB with a boolean that will create the corresponding worlds and then using the build pipeline to make a build (Client / Server profiles with corresponding scenes).
This should be working but it's not and there is something weird with Netcode gameobject conversion.

If I try to create the server / client world before the scene where I have my ghost collection it will not convert correctly my GO. I've looked more deeply into that issue and I've seen something that I find weird: in GameObjectConversionSettings when it will fork settings, the BlobAssetStore isn't kept and since it's a fork, you will get a null ref after that if you try to use it for the conversion which the GhostAuthoringSystem does.
Is this by design ? Like you must create the worlds in a scene where the GhostCollection already exists in order to prevent this issue ? (even then you would miss the awakes of the ConvertToEntity scripts I think).

Feels like I'm not understanding something about conversion here but not sure what.

I've also tried to change the netcode package directly to set a BlobAssetStore (after forking the settings and modifying the BlobAssetStore property to be able to set it inside another script) but the build just spam things like these warnings or it just crash (with a memory access error).

Internal: JobTempAlloc has allocations that are more than 4 frames old - this is not allowed and likely a leak

(Filename: C:\buildslave\unity\build\Runtime/Allocator/ThreadsafeLinearAllocator.cpp Line: 554)



Internal: deleting an allocation that is older than its permitted lifetime of 4 frames (age = 10)

Unity.Collections.LowLevel.Unsafe.UnsafeUtility:Free(Void*, Allocator)

 

[EDIT:] Problem solved by skipping the "Update component list" step and generating the code, then updating the components afterwards.

I tried to get a test project set up, using 2019.3.0f on Mac (Mojave, 10.14.6) but the editor crashed while setting up the Ghost Authoring Component after generating code. After that it continues to crash whenever I try to work with that component. I even tried deleting the test project and starting over, but as soon as I hit the "Update component list" button the editor crashes, every time.

The only clue I can find as to why is this assertion error in the logs:

* Assertion at metadata.c:1117, condition `idx < t->rows' not met

I've checked that all required packages are installed; I've tried both with and without installing updates to the couple of packages that have them (and yes, I fixed the invalid Collections package version to make the Jobs package happy).

Packages:

 

I'm having really really long stalls in the editor when trying to run the NetCode sample project, upwards of 1+ minute when simply hitting play - editor consumes one core fully and completely freeze, if i leave it for ~1 minute it will eventually start.

Completely clean project of the sample repo, on 2019.3.0f1.

Edit: The same happens if i install the netcode package into a clean project, and it goes away if i remove the netcode package.

 

I guess they really want us to use the experimental "Enter Play Mode Options"

 

With synchronous burst compilation - same behaviour. Also first run takes 30 sec + increase memory usage up to 4gb.

 

I have a similar problem, takes forever to start (30-60 seconds) and when started it's really really slow, getting like 10 fps on a threadripper 2950x

 

0.9 Gb of Video RAM usage? While I have 512mb. That's the reason of low fps on my machine.

 

I feel this is this counter intuitive, why does one have to take precedence and not just _ADD_ them together by default? Simpler to throw out a build with server which can be playtested, and let custom build options enabling/disabling the defines and thereby what worlds are used?

On a parallel note ,rewrote (extended) the bootstrap for multiple different server worlds; Barely any comments and weird code of using "foreach" when there is only max one element in array. =/

 

Would it not be possible to build headless thin clients for stress testing? Or is thin clients handled differently?