multiplayer

[twin juniper]

like store them in an array or something

[bitter oriole]

And at a glance, with about 1MB/s being rather on the good side of average upload speeds, you have to consider that streaming quality would be absolutely horrendous - like we're talking 240p here

[twin juniper]

[bitter oriole]

You can also write C++ to replicate PNG files, with the same caveat - one image would take about a second, depending on size

None of this is normally done

[twin juniper]

Gotchya, so we can't just replicate 2Dtexture file in Blueprint right?

[bitter oriole]

Anything visual is incredibly, drastically heavier than gameplay

[bitter oriole]

Nope

[twin juniper]

Thank you soo much Stranger

[sinful tree]

1. The item should be a replicated actor.
2. Any "Overlap" events should use a Has Authority (Authority) > at the beginning so they only trigger on the server. You don't want clients running any of the code themselves - they will have the location replicated to them automatically. This also means you don't need to have RPCs for anything to do with the pickup/chase system - the server will detect the overlap and you can have it move the objects.
3. Your "Score" variable should likely be getting set on the server, so again, no RPC there. You can make it a "RepNotify" variable and use the generated RepNotify function to call your event dispatcher.
4. You really should use a gate and tick like I sent in the images earlier to handle your "Chase" logic. #multiplayer message
5. You should not need to RPC anything to get this to work. The server can handled what it needs to without the clients telling it what to do, and the server replicates the score to the player which then can trigger the OnRep function which can then call your dispatcher to update your UI on the client.

[noble junco]

Q: For the last week, I have been trying to find example or outright libraries for how to implement multi-player character creator/customisation. Something akin to BD Online or Cyberpunk... or really anything half-descent really. Does anyone have a pointer to examples... or something already developed?

[graceful flame]

Make a ui of what you want to change then connect the ui values to the customization controls which make the characters unique and save it on a cloud database so it can be recalled later. It’s a lot of work and I’m doubtful that you’ll find any sort of tool that meets your game’s needs like that. Only thing close I can think of would be metahumans.

[weary laurel]

I have enabled port 7777 to listen in both Inbound Rules and OutboundRules over the firewall with TCP and UDP options, but my server still does not listen to port 7777, what could be the reason? Or can you suggest me a resource that explains this work thoroughly?

[true spade]

Hi, I'm trying to implement the online matchmaking, so I cannot find any steps since mostly projects/examples are outdated. Can anyone please write how to properly write matchmaking ?
For instance, that I can find servers join as a party with my friends, etc. 🙂

[bitter oriole]

That stuff hasn't changed much since Unreal 4.0, tbh

Create session, join session

[true spade]

Thanks for the answer, I'm trying to find a solution for the listen server and for the dedicated server, I wanna pick a best ping for the current player, but if I'm joining as a party of 4 friends (btw that example I cannot find anywhere) to join all 4 party members in the match, do you have an example or somehow explain that to me 😄

[bitter oriole]

At that point you probably want to write yourself a matchmaking server

And at that point you can simply implement your own online subsystem entirely

The engine's is simply a thin wrapper above Steam, EGS, GOG etc services

What they offer or not is outside the engine's control

[true spade]

Is it possible with Steam, join a group of players in the same server/session?

Without implementing own subsystem

[bitter oriole]

Steam gives you basic sessions tools - you can create one, players can search, list, choose sessions

What you build on top of that, whether you have a server or not, is quite outside Steam's responsibility

There's also non-Steam stuff that's more flexible: https://www.unrealengine.com/marketplace/en-US/product/matchmaking

[true spade]

Oh, so at first I can create only that you can join the lowest ping out there, but party joining will be almost impossible to make.

That's nice plugin

[past totem]

Advanced sessions plugin?

[true spade]

Is it possible create this within advance sessions plugin?

[past totem]

no clue what u want to do

google it and check

[true spade]

Join as a party into the match, for instance

[thin stratus]

Party Systems are done via Beacons

Steam only allows a player to be part of one Session at a time, so you can tell everyone to join whatever the Main Player (BeaconHost) wants, but that kills the Party. So you'd need to manually recreate the party when everyone leaves the actual Server again

Also this whole topic is #online-subsystems content

[true spade]

Thank you so much for this response. I'll take a look at that, this is what I'm looking for (I think). Thanks again

[marble thorn]

How do you pass variable values from a Client's menu widget to the server map's Gamemode in a dedicated server?

[dark edge]

Do the math on how much data that'd be

[dark edge]

You don't directly go from widget to gamemode

You'd communicate through the PlayerController

[marble thorn]

Hey, Much thanks for the reply. I am kind of stuck with this for the past two weeks. So this is my requirement.

0

I have a Game default map which has a widget. The players have 2 options, Create a session or join a session which is triggered with 2 buttons in the menu (Host and Join). In the menu, also the player selects the level he wants to play and number of players against which he wants to play. Once he selects map and enter number of players against which he wants to play, he clicks host button which will open level (in Ip Address) containing the server map which is a lobby. In the lobby gamemode, I have condition which will check if the lobby has n number of players that the host player has input in the menu before and then will load the level that the host player has selected in the menu before. The only problem which is frustrating me right now is the fact that I am not able to pass those menu variables to the server map (lobby) gamemode. I have tried options with OpenLevel and Casting values to Game Instance. but none is working with dedicated server. How exactly should I tackle this issue?

[marble thorn]

Can you pls elaborate as I am kind of a newbie wth this. Any help will go a long way mahn. I m too tired of this sh*t

[dark edge]

So are you doing dedicated server or listen server? You said dedicated but now you're talking about a host player which implies listen

[marble thorn]

No no... Host means a new player when he clicks on the host button, i will have a new game instance spinning up on AWS and the player will join the session. Then will wait in the lobby for other players to join via matchmaking. THis is how I am plannning the server and pls correct me if I am wrong

[little pumice]

It seems like FArchive doesn't care about property serialization order (I can flip line 30 with 31 one and get same result). How does it know during Deserialization what bytes corresponds to which property?

[slow hornet]

Can someone tell me more about how is the current state of online multiplayer in Unreal Engine 5?
I am getting back to development, and I was wondering about what is new out there and currently reliable.
I have found so many plugins on marketplace already and then there is this official plugin from Epic Games called (Online Subsystem EOS).
I am a bit lost on where to continue as a starting point, I have played around with an advanced steam dedicated plugin (dont really remember the name) and replication in general before in ue4, but that was quite a while, and it was actually working, but maybe there are better solutions atm no idea.

Also, before continuing much more on my project, I want to make a good multiplayer foundation.
Just need some bullet points on the current solutions pros/cons, like a overview of the current state and current plugins.
Maybe its asking too much I dont know.. but anything that you could talk about this would already be nice, just to try get things clear in my head.
Thank you!

[thin stratus]

The very first thing you should learn is that Replication and OnlineSubsystem stuff are two different things.

Replication happens when everyone is connect and playing your game.
OnlineSubsystems, like Steam, EOS, etc., happen around this. That stuff handles finding the games that you then join.

And similar things, like Friendlists etc.

In terms of replication not too much has changed. There is some push (no pun intended) for the PushModel, which is C++ only iirc, but default replication is still totally viable.

There is also some PredictionPlugin that is forever stuck

And some new things are yet to come that aren't usable yet

But generally you should be able to more or less code the same way you did 4 years ago

OnlineSubsystem depends on what you want to do

If you want to release on Steam, just use the SteamSubsystem.

If you need crossplatform, you should have a look at EOS + whatever platform you release on

But that's more #online-subsystems stuff

[thin stratus]

<< works both ways

If you flip the lines, they are flipped for saving and loading

So that does of course not have an effect then :P

The FArchive is an IsLoading or IsSaving if you wish to debug this or do special stuff for loading vs saving

[fluid summit]

Hi! i'm having a trouble with duplicated drops.
When my creeps die --> OnDestroyed.
I call an event on the GameMode to drop a single item.
This is being called only one (i checked with a break) but i still get 2 items dropped.

Does anyone know from where this second call may be coming?

All the drops are duplicated
Solved: The pin connected to the node "Drop item" was connected twice (one to the cast reference of GM and the other one to the Valid get of the GM)

[real ridge]

i have same problem

does anyone know solution?

[slow hornet]

Thank you for extensive description! I will look into SteamSubsystem then 😊
This is exactly what I was looking!

[small lark]

kind of a noob question but the online services EOS plugin vs online subsystem EOS plugin whats the difference...

[past totem]

hi how can I get the IP of the dedicated server

so I can save it to re-join the same server later

I'm joining through session bp nodes

how can I save it so I can rejoin the last server I was in?

[plucky prawn]

if you are running this from a vps, this is normally sent to you via email and accessible through a control panel. if its a local machine that you control, this can be found on windows a bunch of ways like through command prompt/power shell, or from your router's static ip list

[past totem]

I want to do it from bps

get the ip of the server I'm in / connecting to

or cpp if simple

[plucky prawn]

im not sure on that sorry

[past totem]

thx tho

[magic furnace]

why is this on rep being called

on all clients + server

but doesn't do anything

if I hook up a multicast

it does change the collision

so odd

[oak prawn]

I want to make a branch for my game so that the dialogs don't get mixed up. this branch will ensure that one dialog is working but the other dialog is not. But I've seen many times that the dialog is bugged while doing it online.

even when the dialogue was over it was showing that the dialogue was still going on

How can I remove this bug?

can it be fixed by ticking reliable?

[oak prawn]

if scene branch plays dialog correctly , it waits for dialog to end . But maybe there is a bug in the future, scene branch stays false

If I enable reliable, will branch work better?

[foggy idol]

wont server travel bring controllers and stuff

because i want to do it from a lobby

[past totem]

no

[cold moat]

How good are animations on a networked game in Unreal? For example I love making attack animations and calling code on the animation to trigger damage etc, can I set a notify in animation to run on the server? Or maybe make a notify that calls a replicated event?

[past totem]

it kinda works

generally it works

depends on how responsive u need it to be

[cold moat]

I mean, it would be okay to me if the attack trigger/notify and the animation were not in sync due to lag

but I would wish the trigger at least would get called

[past totem]

ye

[cold moat]

Then calling animations on the server and setting reliable maybe would do? 🤔

[past totem]

u don't need to set it reliable

[cold moat]

why not though?

[chrome bay]

You would already be connected to the lobby, traditionally the only way to keep the players in the same server is for that "lobby" to be a game/session you are connected to.

Alternatively you can form a "lobby" with beacons, then tell each player to join the game later - but beacons can be quite difficult to work with sometimes

[thin stratus]

Yeah e.g. Steam will need you to recreate the beacon lobby when everyone leaves

[wheat magnet]

how to test dedicated server without building client and dedicated server build each time in editor?

[fallow pendant]

Hello everyone. I've been struggling for a while.
What is the best practice for implementing team index and player index?
I don't know where to put it, Player State/Player Controller/Player Class
My game consist of 2 team (or more) and each team has 5 players
Please help me
Thanks guys

[bitter oriole]

Pretty sure player state already has a team index field

Not sure why you'd need a player index

But that should be there too

[fallow pendant]

Thank you Stranger,

[fallow pendant]

Thank you

But do you mean this?

[bitter oriole]

No, that's your unique platform net id

[thin stratus]

You can select "Play as Client" in the Editor

That spins up a DedicatedServer in the background basically.

[slow hornet]

Quick question guys, I am spawning an Actor on server side, however when I set to simulate physics on the component, I can see it moving in one client but not in the other client.
Actor is set to replicate, all chilld components are set to replicate too.
I don't need to multicast anything correct?

If I multicast the spawn of this actor it works, but I think that's not the correct way of doing it, because I would end up with different actor instances on each client, right?

this actor should live on the server, and all the clients should get its moving position (physics) from the server

[past totem]

Simulate physics is not replicated

Don't spawn an actor on multicast , that won't be synced

@slow hornet

[slow hornet]

exactly yes

so what is the proposal here?

[past totem]

To not use physics in multiplayer

U can replicate it with workarounds if u rly need but it'll be weird and laggy

[slow hornet]

Couldn't I spawn the actor on the server and then create copies for the clients (multicast) only to see those meshes falling?

Then any interaction that the client do on their multicasted actor with would be casted to the server and triggered there on the server actor

Wondering what is the proposal here to follow

[chrome bay]

no because if you spawn actors client side they cannot be resolved over the network

Replicated physics objects are always a bit jank, you just have to live with it. The engine does what it can to smooth things out

[past totem]

If you rly want, on server tick save mesh component translation to replicated var
And on client tick, use that var to set the transform of the mesh component

[chrome bay]

The correct way of doing this would be to spawn replicated actors server-side that are set to sim physics.

Don't do that, replicated movement already does this way more efficiently

And physics replication will automatically interpolate/smooth as best as possible

Trying to overrule it yourself will make it far worse

[past totem]

So u r saying to make a pawn child or something?

Ig that could work

[chrome bay]

No just make an actor that sims physics

and check "replicate movement" - that's it

[past totem]

But that doesn't work tbh

[chrome bay]

It works fine

But it needs to be replicated properly, if you spawn stuff client side independently it'll never work

[past totem]

Unless maybe it needs to be the first component or something to work i gyess

[chrome bay]

The root component is always the one that needs to simulate physics yeah

But that should always be the case anyway

[past totem]

I'll try too later cuz I do something similar thx

[summer tide]

When I set death to a character and play death animation, it effects all the characters created from that base BP/ Any idea how to prevent that? I thought casting should resove that.

[thin stratus]

That doesn't have anything to do with casting though

Casting is just checking if some reference is of that Type you are casting to.
And if it is, it allows you to access the properties of that Type/Class and its functions.

You'd need to show us how you actually set that death and anim stuff

[rich dune]

How would I handle a Multiplayer Stat and rank server the best way in UE5?

I want to centrally collect gamplay and rank data on a server. The question is the best way to do it. I could send the data directly from within the game to my MySQL database but this is probably not very safe. Would it be better to send and receive data via a socket connection to a server which it self connects to the database?

[summer tide]

https://gyazo.com/f08266f709909dfa35f692413c747583

[thin stratus]

And where is that code? In the Character?

[summer tide]

In my character BP. I have two of them placed in the map

[thin stratus]

Yeah that will play on both

Both fire BeginPlay of course

[summer tide]

How do I prevent that

[thin stratus]

That depends on what you actually want to achieve here

When should that DeathAnim be played?

[thin stratus]

If you don't have a Dedicated Server, then you should probably have some sort of webservice in between that manages your HTTP Requests.

Keep in mind that collecting data has to be opt in.

[summer tide]

My situation is complicated. But what should I do in that scenario of the screenshot?

[thin stratus]

I can't answer that. The scenario in your screenshot is way to vague and simple.

And I would still need to know what you would expect to happen in the Scenario of your screenshot

[rich dune]

How do you meant it with a dedicated server?

[summer tide]

I have a IsDeath variable in a component of that character. I set it to death and tried playing from blueprint and statemachine. Both has the same output

[thin stratus]

UE DedicatedServer

[thin stratus]

Yeah but the code in your screenshot is 100% clear why it's causing it

The code you actually should post is your gamepaly code

Cause that one has bugs

[summer tide]

How do I fix the one in the screenshot? Trying to understand

[rich dune]

Yeah sure the game is running on dedicated servers. but also gives the option to run on P2P and on both versions the game should be able to produce ranking data

[thin stratus]

Again, what do you expect to be the "fixed" state here?

[thin stratus]

Ranking data on P2P is kinda garbage

[summer tide]

Play the animation on the specific character

[thin stratus]

On which specific character

You have two in your level, which one should it play on and why?

[rich dune]

For the coop mode it's kind of usefull

[thin stratus]

Yeah but it can still be fully cheated

Only thing you can do is rate limiting and maybe limit how high the ranked values can go

[rich dune]

yeah thats the point . How do I atchieve the best result

[thin stratus]

Either way, have them talk to a Webservice

[summer tide]

Based on the IsDeath variable

[thin stratus]

Well, put a branch between BeginPlay and the Delay.
Connect the isDeath variable to the Branch

[summer tide]

I figured it out. Thanks for the hint. I used rep-notify on IsDeath

[fierce grove]

Hey there, does a TMap inside a replicated struct replicates?

[chrome bay]

no

In UE5 UHT will warn you that it needs to be marked "NotReplicated"

You can replicate it yourself using NetSerialize, but that ofc means the entire struct + TMap will be sent for any change

And if it's keyed with UObjects, it's unsafe

[chrome bay]

4 real

I wonder why that was ever allowed

[rotund cosmos]

Hi guys! I have a question in regards with timers and replication. I Have a UObject (an effect that can be applied to players.). It contains a name for the effect, etc. and A TimerHandle for the Assigned timer. The UObject is properly set up for replication (IsSupportedForNetworking() {return true}). I Have an array of these objects in the player class. It is set as replicated and I use ReplicateSubobjects for replicating the Effect objects through the array. When a new effect is added to the player, it creates a new effect object with specified properties on the server, sets the expiration timer handle with a given time and adds it to this array. I want to use the GetTimerRemaining() method on the clients, so I can set a progress bar accordingly. I dont replicate the timer handle, so it only works server side. How could I set a timer both on the client and the server after object creation? I tried it through a client RPC, but on the client, the Effect object is null / it hasn't been replicated yet. Any Ideas?

[loud frost]

Is there any limit on steamDevId 480 ??

Like number of players

[mortal sinew]

Can a listen server have connection with a dedicated server? while having clients connected

Lets say i want to have a global chat, per dedicated server, that would be accessible while you are playing in a session with a listen server

[chrome bay]

Timers are not replicated, they're local only

[rotund cosmos]

I know, thats why im asking, when should I create a timer on the client for visual only?

[chrome bay]

The easiest way to make it work, is to replicate a timestamp and a "length" - and the timestamp is made relative to GameState::ServerWorldTimeSeconds

[rotund cosmos]

hm, interesting

I will give it a try

[chrome bay]

It's not perfect and the timer may jitter now and again, but it's the only default synchronised game clock

[thin stratus]

"not perfect" :D absolute disaster that thing

[pallid mesa]

a better timer here: https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/ ✨

🤣

[oak prawn]

I have a very important boolean for my game. If it works incorrectly, the next dialogs of the game will not open. It works perfectly in singleplayer, but in multiplayer the boolean stays false after a while.

If I turn on reliable, can it be made to work more smoothly?

[slow hornet]

Maybe stupid question, but its not possible to spawn a decal to a StaticMeshComponent that lives on the server correct?
Even tho that StaticMeshComponent is ticked for replicate, since SpawnDecalAttached does not replicate?
Everytime I check if the MeshComponent is valid (before attaching the decal), it tells me false 😐

[oak prawn]

Because the boolean must be declared as true or false to the server in a very short time, otherwise bugs may occur.

[sinful tree]

That sounds like bad design. What happens if someone happens to have a ping spike or a bit of packet loss and they don't receive it within what you think the timespan should be?

[oak prawn]

I don't know, I think we had these problems yesterday. My friends tested my game.

exactly what i want. If a dialog is active, a new dialog should not start.

[chrome bay]

If the system is dependant on a delay it's going to break in MP at some point, that's inevitable.

The only proper solution really is to refactor the system so that you don't need the delays

[oak prawn]

I made a logic like this

[chrome bay]

And yes if an RPC has to be received, then it should of course be reliable. Reliable is meant for things you can't miss.

[oak prawn]

How can I set up the most seamless solution?

[chrome bay]

Like I said, remove the delays

A delay in multiplayer will break, it's only a matter of time

[oak prawn]

hmm

[chrome bay]

They're also a source of extremely hard to reproduce bugs

When you factor in real-world connections with latency, packet loss etc that you can't account for

[oak prawn]

so i need to use reliable right?

for smoother operation

[chrome bay]

If a client has to receive an RPC, then it should be reliable.

That doesn't mean it'll fix this

[elfin lintel]

If I send a UObject pointer over RPC (Server->Client) and store that pointer on the client, if a property on the UObject changes will the client see it? Or is the object effectively copied over the network?

[sinful tree]

Is that UObject replicated?

[chrome bay]

No to both. The property and object would both have to be replicated

client-side, that pointer would then be resolved to the clients local version of that object

[elfin lintel]

Ok thank you, I will have to think of another way to replicate that data. I am just trying to limit the amount of objects/data I am replicating and I was wondering if that would work. I may convert the object to a struct and then manually update/push that struct when changes are made to it (not often)

[chrome bay]

If you want to replicate arbitrary UObjects you can but there's a lot to it:
https://jambax.co.uk/replicating-uobjects/

If you can use a struct, that's by far the easiest method. Structs will only replicate the data that changes.

[oak prawn]

When I remove the delay, this time the dialog does not open in the other person.

[thin stratus]

Ah, casual "I used a delay to solve network latency." issue?

[oak prawn]

my problem is when a dialog is active the scene branch is true . scene branch drops false at the start of the dialog and becomes true again when the dialog ends

but if I don't put delay in between, the dialogue in the 2nd player doesn't appear because the scene branch is false

[dark edge]

State machine my man

Repnotify

[slow hornet]

Any clues on this?

[dark edge]

What's the actual problem?

What's not working?

[slow hornet]

I spawn an actor (server side) that has a component A, then I spawn a decal (client side) attached to that component A, I see the decal on my client, all good.
Then component A starts moving, and decal is not moving as expected (it was attached... i was expecting the decal to move too, matching component A position)

I am sending the component A object reference on the RPC to client to attach it, and it seems invalid I think

[dark edge]

is component A replicated and can you resolve it on both sides?

[slow hornet]

this is not correct, the decal gets attached properly, and component is valid. its only after starting moving that decal doesnt follow

[dark edge]

Do other things spawned and attached also not move or just decal?

[slow hornet]

the other things move correctly on all clients, on decal i stop seeing it

[dark edge]

Does a decal move with it in singleplayer or listen server?

[slow hornet]

not sure this answers your question, but i am testing using net mode: "Play as client"

i saw this working before in single player yes

before jumping into multiplayer

keep world transform on decal

[dark edge]

ya don't do that

keep world transform = don't move

or rather

Is it attaching with world or local? I know there's some UE5 stuff that you can do world attachment now

[slow hornet]

I am attaching to component, which is a StaticMeshComponent (root of an actor)

that component then gets simulated physics and decal doesnt follow

already tried Keep Relative and Keep world

[dark edge]

Try to just manually attach the thing after doing that. Maybe somethings broken in Spawn Decal Attached.

Is a decal a scene component?

[slow hornet]

ok will try

[slow hornet]

Dont really have enough knowledge to answer that yet sorry :/

[quasi tide]

scene component = a component with a transform
actor component = a component without a transform

[slow hornet]

I guess the answer is yes then, Decal has a Transform

[slow hornet]

tried, but also not following :/

will do some searches

[slow hornet]

I got confused sorry, the problem is actually that component A is not being sent properly (as variable) through the RPC to the client.
The component is ticket to replicate, but when calling the event on owning client and giving that component A as input, after the check if is valid, it says is null there.
And I believe it's because the owning client can't reach that that scene component object reference because its owned by the server?

[thin stratus]

When are you adding that Component @slow hornet ? DesignTime or Runtime?

[slow hornet]

runtime.
I spawn an actor in runtime that has that component

[thin stratus]

So the Component is added designtime?

[slow hornet]

the component is part of the actor already in design time yes

i just spawn that actor in runtime

[thin stratus]

Both Actor and Component are replicated?

[slow hornet]

i guess so, let me check agian

[thin stratus]

I mean, I don't actually know if it's needed to replicate the Component here, as the Client should be able to resolve it anyway.... maybe not

[slow hornet]

actor (replicates ticket)
component (component replciates ticket)

[thin stratus]

Hm, Components should count as subobject replication, and I'm relatively sure they are stably named like this

Not sure then atm

[slow hornet]

I made some progress here, but its still weird behaviour, maybe i need some switch on authority

player controller receives on server side the component where the decal should be attached

then i set with RepNotify that component

[thin stratus]

PlayerController?

[slow hornet]

yes

[thin stratus]

How does the PC play into all of this again?

Cause that only exists on the owning Client after all

And doesn't really follow any position

[slow hornet]

its just to have some place to draw the decals, and since PlayerController both runs in server side and client side

I was re-reading your compendium today and notice that PC lives on both places

[thin stratus]

I wouldn't use the PC for stuff like that tbh

I would also not replicate the Component but just an Actor if needed

Usually makes more sense to not even replicate the Actor and just have the Component being spawned on both ends by some other replication

[slow hornet]

where would u draw the decals then? on the pawn? because it needs to be client side i think

[thin stratus]

What kind of decal is that

[slow hornet]

this one

[thin stratus]

Yeah but, what is its purpose

[slow hornet]

ah

its like to draw an effect on my tree that will be falling down

[thin stratus]

Okay, and everyone has to see that?

[slow hornet]

yes i was trying that ^^

[thin stratus]

And your Tree is what? A replicated actor?

[slow hornet]

maybe overkill

yes

[thin stratus]

Because it sounds like you could get away with just having a "bIsFalling" RepNotify in which you add the Decal to the Tree

[slow hornet]

physics already replicate , the falling and stuff, using a StaticMeshActor

Ah

I think I got what u mean

inside the falling tree blueprint having that logic

instead of passing around the component object reference where i want to attach the decal

i was clearly overcomplicating.
let me try that, my current issue is that the decals were not following the physics movement.
but i believe that might work a way better!

[slow hornet]

If I got it right, this is not working for me :/
At least from my attempts, because the actor tree is not owned by the client, so I can't really spawn a decal (which is client only) there

[dark edge]

Show your setup in a repnotify

OnRep bIsFalling -> DoStuff

[cold moat]

Guys, I'm applying movement to a character server-only, on the client it seems very laggy, should I call the event both on client and server so the movement seems smooth on the client and if wrong, corrected by the server?

[dark edge]

Number one, how is the character moving? Is it a canonical unreal character with character movement component?

[cold moat]

Nope, I'm using set actor location as I'm using VInterp

it is a smooth movement from point a to b

the "Movement event" is set to "Run on Server", called by another event that is set to "Run only on Authority" (server afaik, this event also is set to Run On Server though)

although setting the movement to run on multicast still leaves it being very "choppy", laggy, it looks glitchy.

on server it is smooth like butter though, not very sure how to accomplish this smoothness on client as well

[dark edge]

That's the core problem of multiplayer.

Start by not trying to outsmart the engine. Just use replicate movement instead of multicasting the location. But you really want a movement component of some sort tho.

[cold moat]

I'm using the character movement component on my character

but I'm not sure if the movement from set actor location will be replicated just like the character movement's movement 🤔

[dark edge]

Ya prolly don't just set actor location. Use the movement component to do it.

What are you trying to do here? What's the gameplay mechanic.

[slow hornet]

its here

that print on return is false, its not spawning the decal

and my guess is because this actor runs on the server

[dark edge]

That should be an impulse btw

[dark edge]

And also you aren't checking if it's falling

That will trigger when it's NOT falling

Is Set Simulate Physics replicated or not? I don't recall if that state is replicated.

[slow hornet]

oh I thought that repnotify trigger when the var changes?

[dark edge]

Don't rely on it, it will fire on initial replication too I'm pretty certain

[slow hornet]

ok ok

[dark edge]

Or when going in and out of relevancy

You are trying to do valheim style trees, right?

[slow hornet]

physics is working, and tree falls, its just the decal thing not being applied

yes as prototype for the time being, it will help me further on on different scenarios

[thin stratus]

Can you try attaching it to something else than the physics simulating component?

There is not reason for this to not work

If the print prints on everyone

Then it should spawn the decal

There is no "This exists on the server" thing to prevent this

The OnRep would never work if the actor wouldn't replicate properly

[slow hornet]

ok thanks guys for the input, I am out now for a bit, neet to air my mind a bit on it.
I will try and recheck everything when I am back, and will just drop some lines on the feedkack I have.
Thank you both!

[vagrant grail]

what's the command to see how the player capsule gets corrected by the server compared to the client in terms of speed / position, and also how can I open the console to type that command ?

[dark edge]

Tilde opens console

Top left of keyboard

I don't recall the specific commands but it's something show corrections so you can Google that

[ember dagger]

Show net corrections something like that

Set it to 0 or 1

[dark edge]

I think it's
net showcorrections 1
Or something

[vagrant grail]

tried it but it doesn't show anything

tried both 1 and 0

[cold moat]

But is there a way to set location through the movement component? I don't think so :(

[slow hornet]

@dark edge @thin stratus
Thank you both for your patience today!
After some good fresh air, a bit of reworking on the blueprints and following your advices, I finally made it! ^^
Decals following the attached physics component 🙂
Here is the final solution I came up with, in case you wanna give some comments on it.

[dark edge]

Are you dedicated server only?

[slow hornet]

Well currently I am testing on net mode (play as client)

[dark edge]

Do you intend on listen server or dedicated?

[slow hornet]

dedicated I believe

[dark edge]

You shouldn't gate the decal if listen server.

[slow hornet]

yes I understand what u mean

it was just because, the dedicated server in this case wont have any decal to draw

so I just avoid that call, saving processing costs

but if listen server, then it has to run it, good point thanks!

[dark edge]

Dedicated server might just ignore it anyway, not sure

[slow hornet]

Ah I see

makes sense

[limber cloak]

If I want to have an account manager located on my server, will there be an issue if the game is placed on epic games store and steam.
I mean the account will be created by using the steam id or epic games id.

The issue I'm expecting is that the epic games account might match(be the same) as someone elses steam account id
What should I do? Theory please

[worn wagon]

Yo what's up guys, I'm wondering if any of you ever experimented with fake latency for the host of a listen server. My idea is give the host artificial latency matching the next lowest player to make the game more fair. Implementing this system-wide sounds impossible, but having it exist for specific systems such a hit registration seems feasible. What do you guys think?

[past totem]

this might be overcomplicating it but, I would try to see how the engine does it (network emulation setting)

might be something that can be exposed

but maybe not

[worn wagon]

Seems like a good place to start anyways, thanks

[dark edge]

That will make hosts mad more than it makes clients happy

Just favor the shooter and call it a day

[worn wagon]

It would be optional 🙂

[dark edge]

That's the nice thing about shooters. If you see your bullet hit the other guy, it should count 99% of the time.

[worn wagon]

I already favour the shooter but my friends and I like competitive stuff, I can easily see people saying "host advantage" every game

People would not want to host because they feel it's a crutch

[dark edge]

If your prediction is on point the host shouldn't have THAT big an advantage. But competitive and listen server don't really go together that great IMO

[worn wagon]

Yeah they don't 😄 It's not meant to be CS level, it's a party game, but the gunplay is semi competitive

When you have 1 shot headshots with some weapons I can see the advantage being noticeable

[dark edge]

The problem is you can't really do what you're wanting to do anyway. If the host is getting fake ping, then what is actually running the simulation?

the host IS the simulation

[worn wagon]

Ya I know, that's why I said doing it system wide isn't possible

[dark edge]

You'd need like a 2nd process/world or some really fancy code. Seems like a non-starter to me.

[worn wagon]

But you could certainly put some delay on hit registration for example

[dark edge]

If anyone lands a headshot it should kill who they hit, that's what prediction is for.

[hollow eagle]

that doesn't actually solve anything, proper prediction means the host doesn't have an advantage anyway

[dark edge]

You got instant shots or projectiles?

[worn wagon]

Instant shot

[dark edge]

Yeah then don't bother. Click head, kill guy (after server sanity checks it)

[worn wagon]

Hmm maybe I'm worrying over nothing

I haven't given it too much thought yet, I was curious your guys' opinions

[hollow eagle]

A proper prediction framework with rollback will give all players the same advantage as the host (very high pings excepted - you can't do anything about those).
In exchange you end up with peeker's advantage, which is something basically every competitive game just lives with.

[worn wagon]

Currently the client has full authority, so if they hit on their screen it is valid. Obviously this is highly abusable but it's a party game

[hollow eagle]

Right, that's the other route.

Which also means the host doesn't have an advantage.

[worn wagon]

Now that I'm thinking about it this is true 😄

[hollow eagle]

Doing some degree of sanity checking is probably a good idea, however.

[worn wagon]

What would you suggest? I currently only check for a few things, like ensuring hitresult arrays for spread weapons like shotguns have the correct number of items

[dark edge]

Yeah favor the shooter is basically the same but the server is checking that they didn't cheat it and COULD have hit the guy

in their view

[hollow eagle]

You could do something like checking with a trace against larger hitboxes than normal.

[dark edge]

Everyone is in the future and seeing others in the past compared to the server

[worn wagon]

I thought about a "close enough" distance check, which would stop teleport hacks

[hollow eagle]

Right

[worn wagon]

But regardless it's a party game yanno

Cheating isn't a big concern

[hollow eagle]

then I wouldn't worry about it

[dark edge]

Anyone here have any thoughts on how to extend this kind of system to slow projectiles?

Maybe 1s flight time projectiles

man i wish i had a multiplayer game idea that DID NOT require prediction. It's such a PITA

[worn wagon]

I'm glad the CMC handles that stuff because I know very little about prediction, outside of how source engine does it.

[worn wagon]

My dumb first idea would be to save all player position every x ticks (depending on how performant you want it to be) and interpolate between those values based on a timestamp given by the client

It would be all hitbox positions*

And redo the trace/hit calc based on that old position

[hollow eagle]

rollback solves everything (except peeker's advantage)
too bad it's a bunch of work

[worn wagon]

Is rollback what I said?

[hollow eagle]

no

rollback is recording the state of the world over time. When a client fires, it sends a timestamp to the server and the server rolls back the world to that time to process the result.

That way the server processes things as if the client fire command was received instantly

It's pretty much the industry standard way to do competitive FPS (CS, valorant, etc)

[dark edge]

Hell even New World doesn't predict projectiles lmao

not that it's an engineering marvel or anything

[hollow eagle]

Projectile prediction is pretty hard tbh. Look up the "I shot you first" GDC talk from bungie

they go over their solution for grenades

iirc they basically use the throwing animation to hide the grenade's position getting corrected by the server, but it's been a while since I've watched the talk

this one https://www.youtube.com/watch?v=h47zZrqjgLc

[dark edge]

I'm thinking something sort of like a fuzzy deproject.

Projectiles are local everwhere until a hit.
Server back-calculates the hit actor to where it was clientping/2 ms ago and does some sanity checks to ok it

[hollow eagle]

I think this is where they talk about networking grenades: https://youtu.be/h47zZrqjgLc?t=1568

relevant because grenades are just projectiles with a fancy animation

unfortunately it doesn't quite solve projectile weapons since weapons generally have a very short animation so there isn't much time available to hide projectile spawning

but it may give you some ideas

[dark edge]

You'd think someone somewhere would be working on a general prediction framework. I mean, it's all just boundary conditions and update rules.

[hollow eagle]

epic was at one point

[dark edge]

The whole CMC / GAS / Whatever else clusterfuck is a mess. It's all the same problem, why solve it over and over again.

[hollow eagle]

there's a WIP network prediction plugin that was under development at one point and then iirc the main engineer working on it left epic

GAS doesn't actually solve the same issue as CMC

CMC implements (client, not server) rollback, GAS solves instantaneous prediction.

actually pretty different problems

CMC stores movement in a buffer so that if the server corrects any movement it can replay newer movements on top of the correction

GAS just makes it so you can fire instantaneous actions and know when they've been acknowledged by the server

[dark edge]

I mean it's all sort of the same shape of the problem tho.
You got predicted actions, a chain of causalities, and corrections.

[hollow eagle]

It's a similar problem space but with very different solutions. You wouldn't want the complexity of CMC's movement buffer in GAS. You could do it and it'd be technically superior to the GAS solution, but the complexity of both the framework and code using it would skyrocket.

[dark edge]

Your saved move buffer and assembly of prediction keys are sort of the same thing. Yeah you'd def. want to parametrize whatever a generic solution would be to encapsulate both approaches.

[hollow eagle]

Nah, prediction keys aren't quite the same

GAS can't replay actions the way CMC can

prediction keys just make sure that an action being sent by the server wasn't already something the client did

CMC meanwhile will store a buffer of actions that haven't been confirmed by the server. If an action gets corrected but further actions have been taken by the client, then the further actions are instantly replayed on top of the corrected action leading to less rubber-banding.

CMC's solution is actually similar to what you need for full rollback, except inverted since it's clients performing the rollback instead of the server.

[dark edge]

So what happens in GAS if you do something analogous, like say, being OOM on the server, casting a heal that costs mana (prediction failed), and then casting a spell that costs half of your HP (successful on both sides)?

Does GAS basically just do a hard correction and sync the state at the end?

[hollow eagle]

I think it would cancel just the first ability that failed and restore that bit of state. The second wouldn't be affected as it wasn't affected by the first.

I'm not sure that the client would actually replay the second ability in that case - I think that the attribute set would just end up replicating and that'd be the correction.

I think that the piece GAS is missing compared to CMC (aside from CMC also handling interpolation from a bad state to a corrected one) is that it doesn't replay further prediction actions after a correction.

Which is the actual hard part about CMC.

[dark edge]

So what CMC does could be thought of as casting an ability every single frame. You'd def. want the ability to replay otherwise you'd never catch up to a good predicted state and be constantly clobbering your state.

[hollow eagle]

Exactly

It's why you can't use GAS for what CMC does

You can see the sheer amount of state CMC needs to save for predictions in FSavedMove_Character btw

there's quite a lot

and it has to know how to combine those states and replay them

The prediction plugin would have been awesome to solve trying to write some of this stuff for custom components. Even if it just pulled GAS' prediction framework out it'd be nice since it shouldn't really depend on GAS.

And if it supported CMC-style saved moves it would be even better.

[quasi tide]

Last I heard, they still want to work on the plugin, just no set time really.

[hollow eagle]

yeah 😦

[dark edge]

I just hope it's not just a 3rd prediction system

er 4th if you count physics extrapolation

"Hey I need a locally predicted ability that picks up physics objects and yeets them into Characters, damaging their health and knocking them back. Can you do it by tomorrow?"

[hollow eagle]

it actually had a WIP physics prediction component

and a set of new movement components

[iron crest]

Hello, I have a problem, I made a hotbar whereas you can drag and drop powers into and then you press a key that casts to the powers class and sees which class it is, then plays an event, my problem is that only seems to work in single player, when testing as client and I press the key, nothing happens and my print string states that the casts failed anyone have any clue why this isnt working?

[dark edge]

I have a physics based game project but that one will not be doing prediction. 15ms of lag doesn't really matter for an already slow-responding vehicular contraption and the thought of trying to predict user-built vehicle mechanics just makes me sick

I got it working in multiplayer with lag and it feels great. 2 clients at full speed colliding with each other behaves very well, considering how chaotic multiplayer physics can be.

Much easier to deal with 30ms of lag for actions than trying to predict Little Johnny's Clusterfuckmobile in a multiplayer setting.

[empty harness]

Im having issues with switching characters on client, server works fine, but when client unpossesses it doesnt repossess the newly spawned character and is just left basically stuck unable to move as it isnt controlling anything

[short arrow]

make sure you're not doing it on the same frame. I found that my issues were solved by spawning an actor... waiting 0.15 seconds before having the client possess it would work 100% of the time

for some reason for me It would never work if I tried to spawn an actor and possess it on the same frame. Idk if it's the same for others

[dark edge]

You're doing all of the possession etc on the server right?

[empty harness]

on begin overlap of an actor, then through a switch has authority and running off authority only

[dark edge]

show your code

[slow hornet]

Hello there, I am changing in runtime the mesh of static mesh component, and I am getting this error in console:

StaticMesh property overwritten for component StaticMeshComponent /Game/Levels/UEDPIE_1_Project_Level_1.Project_Level_1:PersistentLevel.Tree2_C_0.StaticMeshComponent0 without a call to NotifyIfStaticMeshChanged(). KnownStaticMesh (0000000000000000) != StaticMesh (00000B3651393F00 - StaticMesh /Game/Resources/Trees/p1.p1)

The game freezes for a bit but then it changes the mesh with success.
I tried looking into Notify / Changed node functions in blueprint for the mesh component but haven't found anything..
Any clues on this?

[slow hornet]

nevermind I will just create the actor with the desired mesh in construct script 🙂

[kindred gazelle]

You probably don't need to be doing all the events you have there. I'm no expert so take my advice with a grain of salt, but you don't need to do the check on a multicast. In fact you don't want to do a check on a multicast. A multicast will (if executed from server) fire on every client which I'm pretty sure you don't want to do, you just want this to fire on the client which pressed the button. So you'd probably want to only do the visual effect on a multicast without trying to replicate the check. This might not solve the issue as I don't fully understand why it would tell you the cast failed, but nobody else responded so I thought I'd help even if a bit late

[pallid mesa]

proper that's where native actions come into play... mini-activatable events based on tags in which u block and cancel a series of abilities based on the tag relationship table

[thin stratus]

GAS def doesn't do any buffering

You predict the Ability and if the Server says "Hold, you can't actually do this." it will just cancel the ability.
It will undo any side effects that are GAS related, but if you set a variable on your character to true, you gotta set it to false again when the ability ends/cancels.

I recently implemented a buffer, where I just listen to NotifyAbilityFailed or so in the ASC, and then add the Ability to an Array (custom Struct). Whenever NotifyAbilityEnded calls, I check if I have something to activate in that Array.
But that's more about buffering abilities for things like a MOBA.

In regards to Networking, CMC and GAS, what is really more complex are things that prohibit movement, like a Stun.

Because the SavedMoves don't know about the stun ;D

[quasi tide]

Got it, so don't support stuns 👍

[thin stratus]

Another thing I noticed is that if you have a delay in your GA and then perform a RootMotionSource, you gotta get corrected.

Which doesn't happen when you do it on Activate instantly.

That's because Epic flushes server moves on activate

To fix that you gotta expose the FlushServerMoves function and do it by hand in the frame you call the RootMotionSource

There is a lot of shit going on with CMC and GAS..

Sadly I never really have the time to write about those things. A good colleague of mine also figured out a good way on how to handle stacking shield buffs, which by default don't really work the way one would expect.
Took us quite a while to figure that out with GAS, but that's more #gameplay-ability-system

[pallid mesa]

this one is quite interesting, I wonder if this could be mitigated by adding a flushservermoves call in all the tasks that deal with rootmotion

[thin stratus]

It could

[past totem]

what is this 5.1

[quasi tide]

Exactly what it says. A new replication system.

Not much is actually known about it.

[woeful ferry]

I have a replicated UObject that I want to have some logic on the client after it has been destroyed on the server.

How would I approach that? Is a multicast for a "End Play" a good choice or how would I tackle it?

[pallid mesa]

OnDestroyed gets called everywhere @woeful ferry

also EndPlay gets called everywhere

[woeful ferry]

That's for Actors though

I'm talking about UObjects

[pallid mesa]

Oh right I missed that part haha

so if the multicast is reliable it will be called before the destruction

although there's no endplay on uobject

@woeful ferry

[woeful ferry]

Yea, that's why I was thinking on making a custom one 😄

With a reliable RPC

if that's a good idea 😛

[pallid mesa]

yes you could do that I don't see why not

although bear in mind that RPCs are for transient stuff

so if you want to make any stateful change then a RPC will not work here

[woeful ferry]

It's just to tell the client that a quest has been completed. Because if I remove the quest from the ActiveQuest array, the UObject will not replicate from that frame onward, since that array is a replicated Subobject.

I could also just remove the quest the next frame, so the client get the last bit of replication from that frame.

Although it seems a bit hacky, but it should work in theory 😄

[twin juniper]

Well, yes. There is the full code of it on github

in the 5.1 branch

[quasi tide]

Obviously....

Doesn't change my statement though.

[twin juniper]

You can know a lot about the concept and how it works by reading its code but okay

[quasi tide]

Obviously you can just read the code. I didn't say you couldn't. But that doesn't change the fact that not much is known at this point.

Yes that can change by simply reading the code

But as it stands right now - not much is known at this point

[thin stratus]

Hmmm, I have this feeling that there is something wrong overall in this setup.
Maybe it should just be a QuestArray, and the UObject has a State for Active or not? Then you don't have to clean it up and you can still send replicated data.

[woeful ferry]

I guess I can do that too! 😄

I was actually thinking about doing that

[vivid seal]

The way I've handled this is to have a bDestroyed boolean that is replicated. When I destroy the object on the server I set it to true, and disable any functionality, move it to a separate array, etc. then set a timer to actually destroy the object (remove all refs to it) in 1sec, and then the client handles cleanup stuff in the OnRep of that boolean

there's probably a better built-in way to do this though

i use that system for buffs/debuffs for example, when they are removed they go to a separate RecentlyRemovedBuffs array that isn't interacted with at all, stop exerting any effects on the player, and then 1sec later are destroyed after replicating a struct with info about how they were removed so the client can do any cleanup stuff.

I include the RecentlyRemovedBuffs in ReplicateSubobjects for the component that handles buffs, obviously

[dark edge]

You guys know any good approaches to handling state transitions with RepNotify?

That is.

Doing something at the edge of the transition? Like if you repnotify bIsDead, actually doing something when bIsDead changes?

That way you can use the edge as the event to play a sound and particles etc. But it won't happen if something becomes relevent and is dead

[vivid seal]

one trick i've heard for this is making a struct rather than just a flag for this, and adding a timestamp to it, and only playing effects if it happened recently

i think in c++ there's also a way to check if something just became relevant?

don't remember the code for it though, there's definitely a way to check what timestamp an actor became relevant at, and you can just check that timestamp is roughly equal to right now

looks like there's a "Get Game Time Since Creation" node in blueprints, you could test and see if that resets on leaving relevancy

[dark edge]

Yeah I'm curious about what the most elegent solution would be. I could imagine something like a door becoming relevent and then half a second later being opened. That should play the opening sound. But becoming relevent WHILE open should not.

Drive through a town and just hear hundreds of doors creaking

[vivid seal]

yeah your best bet is to tie "event" logic to "time since becoming relevant." like i said i KNOW there's a way to check that, i just cant remember what the name is/how to access it

[dark edge]

I think in C++ we have pre-post replication values, not sure what they'd be for becoming relevent tho

[vivid seal]

the pre-rep value passed in the OnRep is just whatever the client had, for initial replication I'd assume it's either a default value or just the same as the replicated value

probably wouldn't help for relevancy, more useful for checking transitions from one state to another (which isn't useful for booleans since you know its always the opposite of whatever it is now)

@dark edge searching through history for this channel it looks like that Get Time Since Creation function is what you're looking for, HOWEVER it won't work for objects that are placed in the map, since they don't despawn when out of relevancy range. It will work for anything spawned at runtime that despawns when leaving relevance

so you can use it for mobs that die, but probably not doors that are open

[dark edge]

So probably best to just split things between state and transients and call it a day.

isntead of implying transients from state

[vivid seal]

or just manually attach a timestamp to things if you really want it in one place

not a perfect solution by any means but it is pretty simple

[ember dagger]

Can anyone here help me out. Im having some issues with owner no see not working. I have a function called Update Held Item which takes in an actor class reference "new held item" which then calls a server rpc which then spawns the actor and then sets a reference to spawned actor as rep notify. In the rep notify function i add the item to my inventory and then attach it to my held item root component. The problem is in the held item actor. I have a static mesh with owner no see as true. and in the server rpc where i spawn the actor im connecting the owner to self. Everything spawns and attaches where it needs to fine, except the owner is still able to see the held item when it shouldnt. Any idea how to fix this?

[iron crest]

Hello, I have a problem, I made a hotbar whereas you can drag and drop powers into and then you press a key that casts to the powers class and sees which class it is, then plays an event, my problem is that only seems to work in single player, when testing as client and I press the key, nothing happens and my print string states that the casts failed anyone have any clue why this isnt working?

[plucky prawn]

your picture is very difficult to read but i would assume from reading the errors that Hud is invalid

[iron crest]

Ive chck though the hud is being refrence correctly, I can zoom in one sec

am I refrencing the hud incorrectly, PS: it works when only one client is present, but not when 2 are present

[scarlet olive]

Is this issue only occurring on the client?

[ember dagger]

no for both. I just went and set the visibility of the mesh locally instead, couldnt get it to work the otehr way

[sinful tree]

Widgets don't replicate and as such, you can't pass reference to a widget through RPCs like you're doing. You'd need to approach this differently.

[iron crest]

should I do this through the actual widget or through like boolean variables, idk how to go about this, how would you?

[sinful tree]

As a first step, perhaps make separate server RPCs for each ability and then call whichever one based on which ability class you've got slotted.

This keeps the logic of what to call based on the widget.

[iron crest]

Ok Ill tyr this out and see if I get ny errors

[iron crest]

this is the rror and this is the code

[sinful tree]

You can't reference widgets on other clients or on the server. Anything you want to read from the client's widgets has to be done before you call any server RPC. Determine which server RPC to call after you've pressed your input.

[iron crest]

so are you saying I have to RPC me getting the hud reference before anything? sorry havent worked much with stuff like this

[sinful tree]

Input > Determine which RPC to call (ie. read your hud widget and do your casts) > Call the RPC appropriate to the ability you want to activate.

[iron crest]

like ths?

dont know if this is relevant but any time I drag and drop a abilityt into my hotbar, it sets the hotbar ability to the abilitie that was dropped which is why im casting to the slot and getting the ability class, not sure if thats the right way to go abt it

[pallid mesa]

i use an enum for that

/dying/dead

and a replicated clock

[vagrant falcon]

what is a best place to store diablo like stats data (buffs, hp, attack speed, movement speed and etc.) in a top down multiplayer
thinking about proxy pawn or player state

my stats system

[dark edge]

smells like a component on pawn

why the proxy pawn?

[vagrant falcon]

@dark edge i found atm only this solution to make mouse click to move, is there any alternative?

[dark edge]

You definetly don't need a proxy pawn for that. Look at the click to move template

I would put your stats system into a component on pawn

reuse it for enemies etc

[vagrant falcon]

@dark edge ue4 top down template?

[dark edge]

IDK about your design but our game, there is literally zero difference between the player's character and an enemy. I can put an AIController on my character and possess and enemy and everything works and nothing is broken. The player's character isn't special. But YMMV

I don't remember the template name but there's one with click to move

I mean it's just click -> convert to world location -> pathfind to there

[vagrant falcon]

@dark edge ok, thanks i will take a look

[dark edge]

Also look into GAS before you try reinvent it, but be warned it's BIG

https://github.com/tranek/GASDocumentation

[vagrant falcon]

i was triyng GAS but the system is not flexible enough
so i'm making my own with lua injection

[dark edge]

What could GAS not do?

[latent heart]

Link error: CreateMMO() ?!

😄

[twin juniper]

it's one of the most flexible thing available out here x)

i mean it's enough flexible for a big game like Fortnite that has like 3 big gamemodes, survival, BR with a large map and creative where you create with you want

if that's not "flexible" for you then i don't know what it is 😛

[dark edge]

I can't think of a spell you can't make with GAS but idk

[modern cipher]

i wonder why pubg did not use gas xD

[bitter oriole]

Way too old for that

Not to mention it's a ShooterGame base

[twin juniper]

not sure if GAS was released when pubg was in development (late 2016 iirc)

[dark edge]

Now whether or not it's the best fit is another story. It's a big system. But it is surely capable of whatever you can think up.

[unkempt tiger]

Will GetWorld()->GetTimeSeconds()'s return value ever be affected by replication over the course of a match for a local client?

Or can the local client expect to use this safely to calculate how many real seconds passed (assuming time dilation is 1)?

[kindred widget]

Not 100%, but I think you're looking for AGameStateBase::GetServerWorldTimeSeconds

[unkempt tiger]

Oh no I'm trying to use GetTimeSeconds as a base to replace GetServerWorldTimeSeconds's return value

So that it's less prone to snapping with every replication

What do you mean by not 100%?

[kindred widget]

Meaning I wasn't sure.

[unkempt tiger]

Oh, right

[kindred widget]

I dunno. I normally interpolate anything related to networked values anyhow since you can't really be sure if you're going to get updates often anyhow.

[unkempt tiger]

I tried using a rolling average approach to smooth out a bias between the server's TS and the client's local TS, so I can override GetServerTimeSeconds to something like this:

So wanted to make sure I can rely on GetTimeSeconds, anyways thanks 👍

[nocturne glacier]

hey guys, following problem that drives me nuts:
i want to have a weapon c++ class, that calls functions when certain actors overlap with the weapon.
to do that i bind my callback function like this in BeginPlay only on the server and that works fine.

https://cdn.discordapp.com/attachments/820454594274394163/1027945180076195911/unknown.png

the callback function does currently nothing besides printing a message

https://cdn.discordapp.com/attachments/820454594274394163/1027945591994581073/unknown.png

now, if i test this in game with netmode listen server and 3 players, i see that when i overlap something with my weapon not only my weapon triggers the callback function but also the weapons of the 2 other players do that as well.

https://cdn.discordapp.com/attachments/820454594274394163/1027947389014777937/unknown.png

i really dont get how the hell this can happen. anyone any ideas? 😄

and the weapon class is super simple. nothing fancy there.

if i place the character blueprint in the level with no player controlling it, the weapon of that non player controlled pawn works fine and does not call the overlap function, when a player overlaps something with his weapon.

[serene kestrel]

One other thing to maybe look out for with your approach is that WorldTimeSeconds could also play out differently on Client/Server if you have a MaxUndilatedFrameTime set in your WorldSettings. Take a look at LevelTick.cpp->UWorld::Tick(ELevelTick, float DeltaSeconds) where GameDeltaSeconds is set to the fixed up delta seconds before incrementing TimeSeconds with that value

[unkempt tiger]

Oh, this is golden info, thanks for that heads up!

[pallid mesa]

@unkempt tiger why dont do this? https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

[serene kestrel]

Not sure of vio's use case, but the reliable RPC's in that implementation will run into problems when resending an old Client/Server timestamp so with some ill timed packet loss the sync could drift

[pallid mesa]

thats where the frequency comes into play

[unkempt tiger]

I see that he's still using GetWorld()->GetTimeSeconds() at some point

Also agreed @serene kestrel, I chose to use unreliable for that reason, put together with a cooldown that changes based on reply success rate

[pallid mesa]

Another approach that might work is using unreliable updates and have a higher frequency

[unkempt tiger]

Other than the reliable RPCs which had me raise an eyebrow I see no reason not to use this, any implementation that works is good enough depending on the dev needs I suppose

[pallid mesa]

but so far the trick relies on the ntp pong-back formula

[serene kestrel]

It seems like it's also only using the Delta from the absolute min round trip, if you do mess with time dilation, I think you could get weird values for round trip time if it's based off of world time.

[pallid mesa]

yes, it doesnt support time dilation!!!

so if u need time dilation

have this in mind

[unkempt tiger]

what does the ntp pong back formula end up looking like, what is that

[pallid mesa]

im on the phone rn, but look at ntp clock on Wikipedia

[serene kestrel]

Like the idea is good, but I've made something that actually syncs a local timer (not just a delta) in Unreal and after every sync, you throw out the old sync values and start a new sample

[pallid mesa]

sharing is caring haha 😄

[serene kestrel]

lol true, maybe I'll clean it up a bit and make the github public

[pallid mesa]

ping me and I'll link it in my article

specially if its non-destructive

if u know what I mean by that - because sometimes my wording is odd

[unkempt tiger]

How do you test your implementation? I set up a quick BP that visualizes the time sync

[unkempt tiger]

I don't, wdym by destructive?

[pallid mesa]

by non destructive I mean that it is an addon and doesn't require engine changes

so portable

or pluginified

[unkempt tiger]

Oh, got it, so any player controller subclass should be non destructive then

[pallid mesa]

mhm

so anything we could grab and use

[pallid mesa]

like that

i made a BP and compared the original one with the one i documented

run several experiments

[unkempt tiger]

what does the BP actually do?

[pallid mesa]

nothing just queries the server and client time

and compares both

at the bottom of the article you can find a table with a test sample

[unkempt tiger]

Yeah saw that, but how do you query the server time?

Or rather, when?

When something replicates?

[pallid mesa]

on tick

nono on tick i take a look at both clocks

or when someone late joins

[unkempt tiger]

With network emulation?

[pallid mesa]

mhm

yeah

also this clock you saw is battle-tested XD

buut

doesn't support time dilation

_<

i just hope that epic will improve some day the default clock they got

[unkempt tiger]

What is your usecase of the synced clock?

[pallid mesa]

everything

knowing when something happened

[unkempt tiger]

Okay so, for example, say you want to play a client animation

[pallid mesa]

mhm

[unkempt tiger]

so that it ends JUST when the client is supposed to receive some packet

[pallid mesa]

u can do frame skipping on the server based on this predictive action and the sent timestamp

[unkempt tiger]

e.g a respawn animation, and the packet is 'player respawned'

[pallid mesa]

mhhhh?

[unkempt tiger]

my brain stopped working welp

[pallid mesa]

hahaha dont worry

just bear in mind that its not perfect as it doesn't properly support time dilation

and you can have issues with these reliable RPCs

[unkempt tiger]

Im not really concerned with your implementation as much as I'm interested in what constitutes a correctly synced net clock

[pallid mesa]

Ah so the article provides a brief snippet on how it works

[chrome bay]

tbh there's not really any such thing

due to the laws of the physical universe that we occupy perfect sync is impossible

[pallid mesa]

hahaha

[chrome bay]

you can get good enough tho

[pallid mesa]

yee

[unkempt tiger]

What what test do you use to determine what's better?

[pallid mesa]

but we got one far better than what unreal does, which...well

[chrome bay]

see how much stuff breaks 😄

[pallid mesa]

if the time that has passed in the server is close to the time that has passed in the client

say... the server says 50

[chrome bay]

tbf we use the built-in engine one in HLL for quite a lot of stuff

[pallid mesa]

the client should say something close to 50

[chrome bay]

even hit reg

[serene kestrel]

Also for what it's worth, I only use Time dilation when testing things, not during actual gameplay so the time dilation support is really only important if you're going to use it with gameplay

[pallid mesa]

yes

thats what im saying that if u need time dilation

be careful haha

so vio

unreals clock says 49.20 when it should be 50

[unkempt tiger]

What I'm doing is: the server promises that in ServerTS = X, it'll send some packet, and then on the client, I do a progress-bar-like animation that is set to complete just when the client's version of ServerTS reaches X, and the test I perform is: did X arrive JUST when the animation ended, that is my test

[pallid mesa]

mine says 49.80 when it should be 50

just to make it simple

XD

[past basalt]

if i had the time I would look into this. nobody else in here?

[unkempt tiger]

So my question is, does that make for a good test, aka does it cover most/all usecases

[pallid mesa]

i guess but

a bit too cumbersome

ensuring that the clocks are close enough is already a good test

and as james said

most of the times...

unreal's default one is enough

(im using it aswell)

[unkempt tiger]

I'd be interested in sharing my quick and dirty test BP and see what other implementations are getting

[pallid mesa]

why dont u just print the clock values

in tick

😄

[unkempt tiger]

I wouldn't know what to do with them

[pallid mesa]

just print them to see if they are close enough, data analysis, u know

[unkempt tiger]

If I print them both on the client and the server, I would be getting two different prints

[pallid mesa]

correct yes

[unkempt tiger]

Printed to the screen at possibly two different times

Because internally UE needs to network the print call

[pallid mesa]

u can filter them out and compute the error tolerance

[unkempt tiger]

that adds bias to the test

[pallid mesa]

noooooo

[unkempt tiger]

does it not?

I think it does

[pallid mesa]

now ur making me doubt hahaha

im on the phone cant physically check anything

but if you perform the same test using the same conditions

using both clocks

the bias will be equivalent in both cases

dont you agree?

anyways i have to go 10 minutes

brb

tell me ur findings :))

and gl diggin vio!!

[unkempt tiger]

what findings D:

[pallid mesa]

@unkempt tiger I'm back hahaha

so your findings about both clocks

which one drifts less from the real server time

[unkempt tiger]

do you mean both implementations?

[nocturne glacier]

would really appreciate that

[pallid mesa]

yours and mine

mhm

or yours vs unreal's

[unkempt tiger]

ah I dont have time to switch implementations atm

[pallid mesa]

you overriden the unreal default one?

[unkempt tiger]

yeah

I guess I could undo the override

but I'm already certain mine produces better sync than unreal's default imp

[pallid mesa]

actually you can try to use the same netem settings I am using in my test

and print solely your clock

to see how it compares to mine and unreal's given that i documented that in the table

so no need to undo your implementation

just print/log some numbers

[unkempt tiger]

yeah I can try that

So literally just this?

How did you run it beyond net emulation settings? in standalone? under one process, launch server in separate process? etc

Was it in PIE or shipped etc? I'll just assume the UE version won't matter lol

[pallid mesa]

no

u dont need the Authority remote haha

PIE dc server 1 client

just print the value

and thats it

no need for the extra text

[unkempt tiger]

welp mine completely broke under the bad net emulation settings

[pallid mesa]

yikes, 3 seconds behind ~

then you know your clock won't work to send timestamps around

[unkempt tiger]

yeah lol

[pallid mesa]

so, try undoing your changes and printing out, it should improve

and then you can also grab my clock and fine tune its frequency and reliability

[unkempt tiger]

im more interested in learning what's failing with my approach

[pallid mesa]

that's very cool from your side

so... revise the concept of how an ntp clock works

and probably you can extrapolate what you might not be doing right

[unkempt tiger]

Oh, mine broke because the timing offset was so big that my client code discarded every server reply and never actually updated the timing offsets

[pallid mesa]

😄

@unkempt tiger how does it look now?

[unkempt tiger]

not fantastic, assuming the test is valid

It's actually just not fantastic without any assumptions, because my code doesn't pass the animation test as well

im surprised yours works so well

i'd imagine any kind of sync code would absolutely flop out when ping constantly varies between 400-800 ms and 5 packet loss

[pallid mesa]

if you base it off on replication yes

its just the ntp magic

yours its not bad, its likely comparable with the stock one

[unkempt tiger]

I suppose, the only real advantage I saw over the stock one is that mine doesn't snap as often whenever theres a sync

[pallid mesa]

nice

you have a researcher mindset

😄

[unkempt tiger]

[pallid mesa]

Laura don't come here to roast me come on

go back to your cave >:(

hahaha

[unkempt tiger]

@pallid mesa I'm looking at your article and two questions come to mind

I see you're sending the client timestamp over the network, wouldn't be less expensive to cache it locally on the client instead?

And second, I see you're locking into the ShortestRoundTripTime over time, how come? Is that not prone to failure as the match progresses?

For example the client can log into the server from a set of routers with optimal network conditions, but midway through the match, say due to sudden network congestion in the client's area, the client is rerouted to a different path, causing higher RTTs, and so now ServerWorldTimeDelta will never be set again

[pallid mesa]

those are three questions!!

let me read them

1. yes you can cache for when the client rpc kicks back in 😄

2. Let me check more in detail because I might very well forgot some extra details

@unkempt tiger 🍪 for you for finding bugs

@unkempt tiger im silly vio

the way it is is right

the trip with shortest round trip still provides most accurate delta to server's clock

network conditions may worsen, but time still flows at the same rate

[unkempt tiger]

where is that from?

[pallid mesa]

it's... quite evident

but that's me double checking with my friend

[unkempt tiger]

also, it wouldn't be crazy to assume that exists a packet that somehow, through magic, makes it home immediately, so then why not assume 0 shortest round trip time? and then we could just do ServerTimestamp - ClientTimestamp

[pallid mesa]

if you get a 0 rountriptime

oh boi

ur clocks will be perfectly synced

🤣

bear with me

this clocks only sync if there's a better rountrip time, thus a chance to improve

[twilit radish]

I would imagine the entire idea of shortest RTT is because of desyncs in PC clocks? PCs are never perfect synced up in time with everyone around the world.

[pallid mesa]

daft punk starts playing

[unkempt tiger]

idk, it's either that my brain is extremely farty atm, or something doesn't check out, because something feels odd both with the test and the minimized round trip assumption

(it may be my farty brain though)

[pallid mesa]

so

let me say this

time flows.... equally in both ends

meaning that... the clocks move equally

now... I send a correction from server to client to adjust the clock

only when the roundtriptime is better than my previous one

time flows.... equally in both ends

buuut, they got closer

if a new roundtriptime comes that is better than the previous one

they will be even closer

but the premise is that time flows.... equally in both ends

[twilit radish]

Now you are confusing me too

😆

[pallid mesa]

so the server does time + 1 every second

the client does the same

time + 1 every second

as you can see

the time flows in both ends the same way

so the rate is equal

the speed at which the clocks increase

[twilit radish]

But this assumption is kind of flawed, the clock on a PC is not perfectly in sync with a server clock. So while time is indeed moving at the same rate it may be further or more back depending per machine 🤔

[dark edge]

Yeah for a synced clock you just keep sending timestamp and ping and gradually dial the clocks in

They'll get to within a few ms of each other pretty quick

You need to do a statistical syncing, not base it all off of one packet

[unkempt tiger]

It feels like the kind of assumptions we are not allowed to make when creating a net synced clock

[pallid mesa]

correct time will drift, but not enough to be a problem, and over time is statically better than the stock one

[twilit radish]

But why assume time on a PC when you can assume the relative time?

[dark edge]

Hey my clock says 10 seconds and RTT to you is 31ms
Hey my clock says 11 seconds and RTT to you is 21ms
Hey my clock says 12 seconds and RTT to you is 50ms
Hey my clock says 13 seconds and RTT to you is 35ms
Hey my clock says 14 seconds and RTT to you is 32ms
Hey my clock says 15 seconds and RTT to you is 33ms
Hey my clock says 16 seconds and RTT to you is 33ms
Hey my clock says 17 seconds and RTT to you is 32ms

That's what the server should be saying to the client. That is enough

[twilit radish]

That would be more accurate than assuming that a PC and a server are perfectly lined up with each other.

[dark edge]

Client adjust their clock by statistically doing stuff to the last X sync messages from server

Filtering for ping spikes etc

Just like GPS

[pallid mesa]

the assumption here is that time increases in both ends equally

[dark edge]

It does

why wouldn't it

[pallid mesa]

thus the approach that I mentioned

but it doesn't always Adriel

we are in a quantum world

🤣

no out of goofs

you can assume it does, that's why the article I sent, just works™️

[twilit radish]

I should focus on work lol. I'll be back later, but I'm going to read your page about this again Vori 😛

[dark edge]

You will develop drift but that's why you are constantly updating the sync offset. Very gradually. Shouldn't ever adjust more than a few ms after initial sync

[pallid mesa]

that's what will naturally happen with the implementation

have you looked at it Adriel?

to be in context here: https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

[unkempt tiger]

I still think the test is misleading in that prints themselves have a net bias until they're read on the client's screen

[dark edge]

That's exactly what I've described but it syncs up every 10 seconds

[pallid mesa]

yup every 10 seconds and if it finds a better rtt it will update clockos

:3

[pallid mesa]

nonono vio

the test is right because even if there'd be replication in the midle

[dark edge]

The problem is you have uncertainty with RTT

[pallid mesa]

in the use case I'm doing it so with both clocks

[dark edge]

you need to do statistics on the RTT time to settle on a value

[pallid mesa]

well the shortest is the better, just set once and let every end simulate from then

no future updates needed

assuming shortest rtt is 0 (in a perfect world made out of flowers)

[dark edge]

You'll still need to periodically check and adjust if needed but that's the idea. You just don't KNOW if shortest is accurate.

consistent RTT is more important than short

[quasi tide]

Dang - two and a half hours later, we're still talking about syncing clock?

[pallid mesa]

i mean shortests gives me a shortests difference between client timestamp and servers... hence relying on shortests

HAHAHHA

sorry

assuming no dilations

[dark edge]

Yes but you do math on it.

Time = TimeFromServer - RTT/2

the uncertainty in RTT is where your error comes from

[zenith wyvern]

If an array replicates, does a 'set members in ...' on an element replicate? it seems like it doesnt, whereas adding does

[pallid mesa]

only when a better is found

[dark edge]

For later updates you need to check if you're still synced, if you sync to one magic short RTT packet and then constantly see that your calculations are consistently ahead or behind, you need to update your sync as it was obviously off by some amount of error

[pallid mesa]

yep this would by the theory of adjusting and removing remainders of error

but still, look at empirical data

it displays that the values or at least the tolerance between the real and simulated is pretty on point

[dark edge]

Right now are you doing the filtering and averaging or just doing a hard sync on the shortest RTT packet?

[pallid mesa]

I'm doing shift adjustments on the shortest RTT packet

as the article suggests

and then it stays there for the rest of the simulation if no other shortest is found

[dark edge]

What if the shortest was an outlier?

Which it most likely was

[pallid mesa]

you got a synced clock accidentally

the end goal is having the client values equivalent or close to the server values

and if the outlier within your net connection provided it, you were in luck

[unkempt tiger]

https://gfycat.com/DetailedWeakAltiplanochinchillamouse

This is my test

[pallid mesa]

thats a cool looking sphere

[dark edge]

I mean you do you but that sounds like a recipe for always being out of sync by ShortestRTT - AverageRTT ms

[unkempt tiger]

Its purpose is to simulate real world scenarios for clients using ServerWTS to estimate when things will get replicated from the server

With this in OnRep, I'd be glad if you could try it vori, the idea is that the growing sphere needs to make it exactly to the black sphere - and then restart, which happens on OnRep. The growth of the sphere is based on the ServerTS

[dark edge]

So why are you betting it all on the outlier and not the mean? If your RTTs look like
100
100
100
20
100
100
100
180
100
100

You'd assume 20ms for constant offset

not 100

and you'd constantly be 80ms off

[pallid mesa]

just because of the math hah

[dark edge]

You don't extrapolate your gas mileage by what you got when going down a long hill? Madness

[pallid mesa]

so in this case...

an rtt of 20 will have these implications

[dark edge]

I got 3 paychecks this month, if I extrapolate that out I'll get 36 this year instead of 24. I'm rich now.

[pallid mesa]

you'd adjust your clock in the server, and then the pong back won't likely be 20

thus the math wouldn't work

is that what you mean?

[dark edge]

You adjust on client but you need to use AVERAGE RTT, not shortest.

In perfect conditions they will be close

but across the internet they will not

[unkempt tiger]

notice that by doing that, you might see the ServerTS start to jitter with big enough ping variations (which wasn't the case as soon as a 'perfect' packet arrived previously)

[copper grove]

[2022.10.07-17.03.06:992][988]LogNetPlayerMovement: Warning: ServerMove: TimeStamp expired: 80.884956, CurrentTimeStamp: 82.064621, Character: BP_VRECharacter_Consumables_C_2147477988 i get alot of this while playing online and players are teleporting a lot

[dark edge]

IDK a lot about the CMC but that sounds like bad network conditions assuming you aren't doing anything fancy? THat's a 2s old move.

[pallid mesa]

Ima do some experiments

and look at data now

[dark edge]

What's your experimental setup? If it's on LAN this will all be moot

[pallid mesa]

netem pie

[dark edge]

or are you simulating conditions?

[pallid mesa]

BAD settings

everyone

the results I've sent are from bad settings

the test setup is printing out the serverclock in the server and in the client

keeping the shortest RTT so far has given the closest results... because time flows everywhere equally assuming no dilations

no I'm going to implement cumulative average discarding outliers

any recommendation so far? @whole grove @dark edge

how shall the heuristic look to discard outliers and such?

what's a reasonable threshold to consider an outlier?

[dark edge]

Just start with a rolling average and later you can toss outliers

[unkempt tiger]

you can also keep outliers but give them a smaller weight depending on how far they are from said rolling average, but that's probably unnecessarily complex

[pallid mesa]

@whole grove they are time values Laura

the server time value is the ground truth

an instance of time

[dark edge]

Just print out ServerTime - CalculatedClientTime every time you get a packet from server

[pallid mesa]

old clock is defaults unreal's networked clock in the gamestate

[dark edge]

you'll see the trends in your algorithm

a perfect algo would be close to 0

[pallid mesa]

yes quite considerably

[dark edge]

It's reduced for sure but whether or not its good enough for the application depends on the applicatoin

[pallid mesa]

you have it there in the table, I can calculate the delta error for you xD

[dark edge]

Can anyone intuit why
Average = 0.9 x Average + 0.1 x Sample
would NOT be a simple rolling average (size 10 in this case)?

This is geometrically decaying i think

[pallid mesa]

holy craaaap? <= 1ms error??????????

that's beyond the scope of what I need

i should add a new column to the table then with the error

it'll be better, thanks for the feedback

and about doing a moving average, I think if the code is simple I'll add it in the article as a post-mortem

UNLESS

Adriel you

or you Laura want to write an article about it 😄

I'm not telling you to write about doing a moving average xD I'm suggesting you to write about doing a net clock in unreal 😄

[unkempt tiger]

hm what is this 10-3-3 magic I hear

[dark edge]

You just copy this with the statistical filtering stuff
https://en.wikipedia.org/wiki/Network_Time_Protocol#Clock_synchronization_algorithm

[pallid mesa]

so my clock's error is about ~ 0,05 s

[dark edge]

50ms so about 3 frames

[pallid mesa]

And Unreal's ~ 0.4 s