multiplayer

[unkempt tiger]

Interesting, I bet this gives jitters a run for their money

[pallid mesa]

with bad network emulation btw

[dark edge]

What approach does Unreal use, just hard syncs every time or something?

[unkempt tiger]

i think so

[pallid mesa]

they use a replicated clock

just replication

[dark edge]

no accounting for rtt?

I suppose that makes sense because you often DON'T want a real sync, you want to know what your view of the server's time is

[pallid mesa]

I don't have the code in front of me sorry :c

yes that also makes sense, that's why I call mine non-destructive

it's explained in the article

[unkempt tiger]

I might just give this a go, ping jitters are refusing to let my rolling average stabilize

[pallid mesa]

i'll send u the technique employed in PMs

I've linked the article several times now ☹️

[pallid mesa]

nvm I can't

[dark edge]

We understand what you're doing, just saying it's not a great idea. You need an averaged RTT.

But if what you have is close enough just roll with it

[pallid mesa]

well I'd say that it's not the best solution!

but it can work

I think I'm going to take a look now at the averaged solution

and add it as a post mortem

keeping shortest rtt

[dark edge]

What about

OnRecieveServerTime
Error = Clamp(ServerTime - ClientTime, SomeRange)
Offset += Error
ClientTime = ServerTime + Offset

That's like a shitty PID right

I term only

[pallid mesa]

I'm a fan of SomeRange

aka magic numbers

but really a PID like shifting might work

but honestly I think that the next step is going to be directly the moving average

[unkempt tiger]

gonna give this a go

[pallid mesa]

oh no

quite literally min 3 max 7

😄

what about building this with unreal's rpc 😄

btw what do you mean by.... repeated all over the place Laura?

[unkempt tiger]

I think I did

[pallid mesa]

how much time between ping collections

between 2 consecutive measurements I mean

aight so i get that as fast as possible

right so... I'm going to do something to see if I bump those numbers

I'll send it back in here

thank you all, very valuable lesson learned!!

😄

[unkempt tiger]

Interesting, wonder how choosing the fastest 3 might have contributed to the performance of the solution

Less prone to interruptions maybe?

[pallid mesa]

the fastests 3 were averaged?

right its similar to what you described

i thought about doing a moving window with inbound packets capturing metrics every second

and computing average over that every second

my sample size would be the size of the window

right? XD

so not a moving avg but the average over the last 10 seconds

right but

this means u never stop updating the client clock

[unkempt tiger]

thats desirable

[pallid mesa]

ah no

u can stop

[unkempt tiger]

I just realized I started this network clock sync discussion 6 hours ago jfc

[pallid mesa]

u can stop if ur error is less than a threshold

okay all got it

[unkempt tiger]

ikr

[pallid mesa]

right but in a window its easier to get rid of outliers, right? XD

yes

the sliding window i like

but i need 100 measurements before i do the average

kekw

the global outliers would need to be based on magic numbers

:/

[unkempt tiger]

wait youre doing a 100-sized average window?

isn't that a little extreme

[pallid mesa]

kekw

so the 10 moving window idea can work just good

and its the easiest

what i dont like is the magic numbers

reeee

ill think about it as the universal constant

JUST WORKS™️

[unkempt tiger]

if you want it bad enough, every number can be your personal universal constant

[pallid mesa]

magic numbers: size of the window and threshold for outliers

[quasi tide]

Still on the clock stuff?

[dark edge]

Tune them to taste

Think "How much could I expect the actual ping to vary from average" and then clamp variance to that.

[pallid mesa]

mh

[dark edge]

I would clamp pretty tight honestly. It's ok if sync updates fairly slowly, it's far more important that it's smooth and constant

Or instead of a clamp do some power based thing to rescale

same as audio compression

[pallid mesa]

I'm trying now Laura's method, the approach I'm going for is... 10 recordings over a full second and once sliding window is full I average it excluding outliders (max and min from window)

and I repeat this every 10 seconds

what do you think?

then I'll go for the more intricate solution

but I want to try these ones out

[dark edge]

Server sends time once a second

Event ServerTimeRecieved
Remove oldest RTT, add new RTT
RTT_Average = Average 8 middle RTT of the 10
ClientTime = ServerTime + RTT_Average

Done

[pallid mesa]

ah so u suggest to correct every second

instead of... correct every 10 seconds

i suppose that doing 10 recordings over a second would generate a somewhat homogeneous set of samples

[dark edge]

You're better off spreading samples out rather than clustering them

[pallid mesa]

ye

i figured

I wanted to see if I could correct less often to.... avoid seeing too many jumps in the clocko

[dark edge]

If your RMS window is big enough your corrections should pretty much be super tiny

but you need a fairly high rate to be able to notice if network conditions change

neighbor starts torrenting and all of a sudden your average RTT goes from 20 to 120, you want to be able to notice it quickly

I'mma try it tonight with the PI approach, I think that might be very good because it'd capture the long term drift without being upset by short term changes in average RTT

[pallid mesa]

the PI approach is the pseudocode you just sent, right?

just gotta figure the rate at which it happens

[dark edge]

Nah like a PID but no D term

the faster the rate, the better, but the more network you use up. It's up to you

[pallid mesa]

ah with the clamping ~

oki, and... what do I do before the window is full

just use the new incoming rtt?

[dark edge]

Fill with 0 or first data point I'd say

[pallid mesa]

I wqas

about to write the same

hahaha

thanks Adriel, you are awesome.

[olive kraken]

Hey! general multiplayer question.
I've been researching about methods to pass client data to a server in the moment you join a session, i.e character selection in your local machine and retrieve that data once you join the multiplayer session.
Apparently game instance is the place to keep that data before/after joining session, and that is my specific problem. How to call the client game instance after I joined a session and avoid to call the server game instance instead?

[fossil spoke]

Use the Options string that can be passed when you load a level. If you are a Client joining a Server, this string gets sent to the Server.

You can then parse the options out of that string, its custom so you can define whatever data you like in the following format.

Option1=Value1?Option2=Value2?Option3=Value3....

[olive kraken]

Thanks for that!
I'm using however Join Session from advance sessions, and that automatically travels to the server level.

once this happens all gamestates, ps, instances, controllers, etc are overridden

[pallid mesa]

I reduced it to 0.02 🥳 -> 1 second per recording, 10 recordings to compute average

void ARBPlayerController::NewClientUpdateWorldTime_Implementation(float ClientTimestamp, float ServerTimestamp)
{
    const float RoundTripTime = GetWorld()->GetTimeSeconds() - ClientTimestamp;
    RTTSlidingWindow.Add(RoundTripTime);
    RTTSlidingWindow.RemoveAt(0);
    
    TArray<float> tmp = RTTSlidingWindow;
    tmp.Sort();
    float rttavg = 0;
    for (int i = 10; i < 90; ++i)
    {
        rttavg += tmp[i];
    }
    rttavg /= 80;
    ServerWorldTimeDelta = ServerTimestamp - ClientTimestamp - rttavg / 2.f;
}

^ I reduced it to 0.01 by simply increasing the rate and the window ^

[fossil spoke]

Never used that, sorry.

[serene kestrel]

One thing I meant to point out earlier, but when executing RPC's on the receiving end, the TimeSeconds value hasn't updated yet since that is done after the TickDispatch (check out UWorld::Tick()), so if you want to use the World time for the current frame (a.k.a the time you would get when ticking the frame where the RPC was received) you'll want to find some way to delay your timestamping to later in the tick or add in the global frame delta seconds (which won't get dilated so if you're using dilation beware)

[pallid mesa]

right

[mortal sinew]

Hello, i'd appreciate if someone is willing to help me understand how server world is meaning full

Is there any reason to run functions at the server other than to multicast?

[pallid mesa]

yes, you can run whatever you want at the server, and update replicated variables

[barren surge]

@pallid mesa btw I was able to get hard travels being seamless working. The majority of the work was dealing with loading. And yea the key difference between seamless travel and hard travels in UE is that's the only way UE can enforce stable names/ids for objects. Using our own custom stable naming system we are able to not have to hard reload a level.

[mortal sinew]

Does the client host(in a listen server) runs "twice" every frame? once as a client and once as a server?

[pallid mesa]

YES!!

that's fantastic news @barren surge don't let this die and write something about it! ✨

[barren surge]

Haha so many articles I'd like to write, hoping I find some time.

[pallid mesa]

this one is specially.... key there's nothing done like that ever before ZabirH, I'll keep encouraging you to do it!

it's such an achievement, and I'm glad it worked out

[barren surge]

Thanks!

[mortal sinew]

Is that possible to be connected to two servers at the same time?
Like a constant connection with a master server

[pallid mesa]

hi Khen take a look at Cedric's networking compendium, it'll clear things up in that end

[mortal sinew]

really appreciate it!!

[pallid mesa]

Basically take a look at the resources pinned in this channel

[mortal sinew]

Thank you for the directions

[pallid mesa]

any time, if you have any specific question after revising basics, we are here :)

[barren surge]

But I will say it's far from production ready. My test levels are pretty small. It looks like there is a bunch of loading work I still have to do and I've done a lot of this via engine changes.

[mortal sinew]

After all the time i've used discord i didn't knew that was a thing 😛

[pallid mesa]

hey it's a start, you think it's achiveable through overrides and such?

[barren surge]

Not ATM, I had to hack out a lot of the level loading flow.

[pallid mesa]

btw, I will probably release an article about data streaming checking against the reliable buffer, if that helps

[barren surge]

Basically I'm running all the connection establishment code but not triggering level loading cause I know everything will be named/ID stably. That's a big assumption for UE.

[pallid mesa]

WizardCell will love hearing this, he's on vacation now XD

[barren surge]

And then we use our custom protocol to tell the server of client state and server verifies or sends corrections.

[pallid mesa]

that's nice! I wonder if this new article will help you at all

its to avoid filling the reliable buffer when sending big chunks of data

[barren surge]

Well it's always great to read your stuff regardless of this specific work!

[pallid mesa]

oh no, just wondering if it would, to hurry up a bit 😄

in all honesty, I'm lazy ~

but if there's a drive, I jump into stuff

basically it all started changing ReplicateStreamingStatus, different use case but if you have a case in which you are required to send corrections

and specially tons of them....

it might be problematic

but you said you got a protocol, right?

[barren surge]

Yea, it collects all objects on a client and client assigned ID and asks the server for the real Names/IDs. Server uses type and few other properties to assume what actor the client is referring to.

The other thing we do is for client predicted spawned objects, we tell the server what the client spawned ID is and server will spawn using that ID (of course we use top bits to partition client specific ID so no two clients can generate same ID.)

[red musk]

I have an actor placed in the level and it is replicated. It can't RPC from server -> client, how come? I wouldn't expect it to need to be owned by a client to RPC that direction

[fathom aspen]

It has to be

[fathom aspen]

On my way back 😛

[fathom aspen]

So if I get it right, seamless travels don't keep actors that were forced to be stably named, stably named?

iirc Rename is called at some point when a (seamless) travel occurs, but yeah I haven't looked much at that part yet. I can see this be a problem for dynamic actors with stable names

[steady cape]

If it's Multicast then I might have had same issues, don't know wjat that's about

If it's Owning client, then the word "Owning" is not for no reason here

You need to call SetOwner on that actor with some actor that, directly or indirectly, owned by a player controller

[mortal sinew]

How do i set an actor as autonomous proxy on blueprints?

[thin stratus]

Multicast would work fine. client and server rpcs need an owning client. You need to perform the rpc in the character already if you are trying to interact via key press.

I usually call that the door problem when teaching some of the fundamentals

Opening a door in the level as a client needs a lot of the basics being applied. RPCs, OnReps, Ownership etc

[red musk]

In this case it’s server to client not client to server so I wouldn’t expect it to be an ownership or net authority issue.

Further testing makes me think the client simply didn’t exist yet to receive the RPC, since I was doing this on game start on a listen sever. Delaying the rpc as a test causes it to work, so bit of a false alarm. Thanks everyone!

[mortal elk]

Is it possible to set up online multiplayer using only blueprint? i have advanced sessions plugin but im getting join session problem

[pallid mesa]

@dark edge @whole grove https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/#based-on-a-moving-window-discarding-outliers
I've implemented avg rtt discarding outliers, and I've collected some new results. Thank you two for the help, got it integrated in the blog ^:

[pallid mesa]

add newest at the end

remove the oldest at the front

like a cyclical buffer

@whole grove

also btw same algorithm but instead the for loop doing from 1 to 9, doing from 0 to 3?

[olive kraken]

Hello! quick question.
In multiplayer, calling to the game instance will always return the local machine game instance instead of the server one? Is this the correct way to retrieve persistent data after travelling?

[twilit radish]

The game instance is indeed only available locally, so you could use it for that purpose yeah 🙂

[twilit radish]

Hey. Out of curiosity is this how the seamless travel between levels works that Unreal currently already does when the server switches maps or is it truly "seamless" as in the way that the client stays in a world and then loads all objects etc. given from the server? 🙂

Last one would definitely be very interesting 👀
Interesting in general though, don't get me wrong.

[barren surge]

The goal for me was to build something like Destiny or Division where matching make could happen and a different set of clients would connect to a new server without having a blocking load of a map. Out of the box, connection to new servers must be what UE calls "non-seamless travel". My guess was that UE requires non-seamless travel for new server connection to ensure object names/ids are stable. This is also why once connected to a server subsequent travels can be non-seamless. I had already come up with a way to spawn client side predicted objects that was consistently named (client actually tells the server of the event + spawned ids) that I was able to reuse for making new connections to server work with objects the client had already spawned but now connected to a new server.

[twilit radish]

So if I understand you correctly, same client world but different servers you can seamlessly connect to if needed?

[barren surge]

No the opposite - seamless travel works in UE because the server can rely on the fact the client loaded object will have the same name on server and clients. Maybe I'm miss-understanding your question?

[barren surge]

yes.

[twilit radish]

That's awesome, well done 😄

[finite goblet]

Does anyone know how the server tells the client to destroy a component that was destroyed on server ?

[barren surge]

Its very rudimentary atm. Very far from actually being like Destiny or Division.

[pallid mesa]

@whole grove avg error went up by picking the 3 fastests, with 4 fastests is kinda comparable with the middle-section method

[barren surge]

On the server the component ptr is nulled out, thus the "null reference" is serialized into a FBunch and sent to the clients. The clients then set the component ptr to null and GC actually cleans up the component some time later. Best of my recollection. If you want to confirm, put a OnRep on your component pointer and watch for when it fires and look at what was read from the FBunch to trigger your OnRep.

[finite goblet]

what about default objects created in constructors without the replication flag

addressable just by the path

[barren surge]

If no replication flag than, it will not replicate from server to client.

[finite goblet]

would defaultsuboject created in constructor of the actor won't destroy on the client if actor is replicated but not the component itself

[barren surge]

Yep, that is my understanding.

If a prop is not marked replicated, then no syncing between server client.

[finite goblet]

something about that tells me that it need more cleanup than just nulling it out , like the destroy callbacks , I think it'll be more fruitful to do a debug break on destroy for teh client

[barren surge]

Yep, that is def the ground truth! OnRep + UActorComponent::Unregister() are how components are removed from active play.

[barren surge]

I'm guessing there is logic that if the UPROP is a UActorComponent, before it is reset (right before OnRep is called). it will unregister it.

[pallid mesa]

20% culled, isn't so bad

culling the 40% of the samples is more aggresive :v

okay let's try to cull the 40% of the samples just for the shake of numbers

😄

I'll report back

[olive kraken]

Thank you!

Stream question.
Is there any risk of losing sync when using a stream under the same function. i.e the scale being set before a rotation and viceversa between server and local machines?

I noticed after around 100 world generations, there is an small chance 5% of having a different result in that instance generation between server/client, wondering why since there is no branching logic

[sinful tree]

I'd maybe set the output from the random functions into variables and then use them. Could maybe help just in case there's something strange going on with how its getting interpreted.

[dark edge]

How are you actually getting your data? How do you do the snapshot

[dark edge]

I don't know a lot about streams but don't they advance their state every time they're called?

so race condition?

[pallid mesa]

on tick querying the value every frame

[finite goblet]

This might be an engine bug but If I rejoin after component is destroyed on the server the component returns on the newly joining client and ticks like it always existed.
This is valid for components created in the constructor

maybe I need to set the constructor created pointer to replicate as well

Edit: Didn't change anything , so yeah the component will be recreated on join and tick again, Gonna try Dynamically created component
Edit2 : so it works if the component was created at runtime, gonna try on blueprint added component next, Same result

So for a component to be die clean on replication it has to be created at runtime and should be a replicated component, those are the two absolute requirement

Endgame is the GUID for the object going null on client causes it to be destoryed

[dark edge]

How can I add an onrep to an already existing replicated variable?

C++ variable

[finite goblet]

UPROPERTY(ReplicatedUsing=OnRep_something)

[quasi tide]

He's talkin' about one that is already replicated by the engine. Not a custom one.

At least, that's how I interpreted it 😅

[finite goblet]

ah

[dark edge]

Sheeit not too bad for first time trying this

Bad net settings

[quasi tide]

Yeah - so which one did you mean @dark edge? One that is being replicated by the engine already, or a custom one you created?

[dark edge]

Yes already replicated Ping

[quasi tide]

https://tenor.com/wW2p.gif

[dark edge]

you see my ghetto Onrep in that BP lol

[finite goblet]

I don't think you can do that without changing the source code for that variable

however

you can use something like PostNetReceive on a derived CPP class

which should get called when properties get replicated from server

so its kind of on rep , but for everything

[dark edge]

Yeah this was just a quick and dirty prototype to join the network clock shenanigans

[dark edge]

@pallid mesa

[finite goblet]

you could also just create a second variable in your derived class and replicate that with onrep and disable the first one with DOREPLIFETIME_ACTIVE_OVERRIDE

[pallid mesa]

yeee which one you using? 😄

the mini pid?

[olive kraken]

Yeah, it was a race condition, not in that very function, where the chain of position, rotation, scale will be always in this order, but a race from server to client replicating the integer that defines the seed, rep notify fix it.

[dark edge]

I posted the code. Just a simple smoothing on ping using exponential running average

[pallid mesa]

ah I cannot see the code, where did you post it? 👀

[dark edge]

Here

[pallid mesa]

Ahh very very simple!

I'm bookmarking it for reference

[dark edge]

I'll try a proper PID next, I feel lik that would lock in faster.

Also haveing a repnotify on ping would be great, I just did that nearly equal test on tick to detect when it replicated, quick and dirty

[pallid mesa]

mhm, yep, it'd be nice

also I had a chance to look at the gamestate implementation of their replicated clock

and it's odd

posting in here for you to see

[dark edge]

If it's overridable that'd be the best approach to improve the accuracy

then anything can just Get Server Game Time as if nothing changed

[pallid mesa]

i preffer non destructive, because it seems that servertimne is latent

but.... they are trying to do something...

look at this

[dark edge]

what does non destructive mean?

[pallid mesa]

void AGameStateBase::OnRep_ReplicatedWorldTimeSeconds()
{
    UWorld* World = GetWorld();
    if (World)
    {
        const float ServerWorldTimeDelta = ReplicatedWorldTimeSeconds - World->GetTimeSeconds();

        // Accumulate the computed server world delta
        SumServerWorldTimeSecondsDelta += ServerWorldTimeDelta;
        NumServerWorldTimeSecondsDeltas += 1.0;

        // Reset the accumulated values to ensure that we remain representative of the current delta
        if (NumServerWorldTimeSecondsDeltas > 250)
        {
            SumServerWorldTimeSecondsDelta /= NumServerWorldTimeSecondsDeltas;
            NumServerWorldTimeSecondsDeltas = 1;
        }

        double TargetWorldTimeSecondsDelta = SumServerWorldTimeSecondsDelta / NumServerWorldTimeSecondsDeltas;

        // Smoothly interpolate towards the new delta if we've already got one to avoid significant spikes
        if (ServerWorldTimeSecondsDelta == 0.0)
        {
            ServerWorldTimeSecondsDelta = TargetWorldTimeSecondsDelta;
        }
        else
        {
            ServerWorldTimeSecondsDelta += (TargetWorldTimeSecondsDelta - ServerWorldTimeSecondsDelta) * 0.5;
        }
    }
}

[pallid mesa]

keeping the vanilla functionality intact

I mean if you look at this... it looks like they are trying to.... compensate something

serverside polls at a fixed frequency of 0.1f

configurable in the CDO

[dark edge]

That kinda looks like a PID

with P term only

but with no latency compensation

is it should be accurate but always BEHIND by RTT/2

[pallid mesa]

let me check, actually!

😄

[dark edge]

So you probably can just do something like

ClientGameTime = ServerGameTime + RTT/2 and get something fairly accurate

although RTT changes all the time so you wanna filter/smooth it.

[pallid mesa]

sn't ServerWorldTimeDelta their RTT in their onrep?

[dark edge]

You basically have 2 values you're trying to calculate.

1. An accurate and representative RTT
2. Static time offset.

[dark edge]

I think that's the delta between recieved time and local time

[pallid mesa]

ah yes!

[dark edge]

They're doing static time offset calculations only. Nothing with RTT to try to get an accurate non-delayed synced clock.

[pallid mesa]

yes xD got myself confused there

I'm going to try the method I got for computing RTT

with the averages discarding outliers with this

[twilit radish]

Guys please send help. I just spend a total of an hour being confused how an AddActorLocalOffset was being triggered for the 2 other pawns out of 3. Turned out I'm an absolute idiot and didn't put anything in the level to compare the movement with, so it looked as if the other pawns were moving while in reality my controls were inverted.. So if I moved up it looked like the other 2 pawns were moving 🤣

Was adding checks everywhere, commenting out code, looking through engine code only to just make this dumb of a mistake LOL.

[pallid mesa]

@dark edge tried adding the RTT/2 to the return of the vanilla clock and.... didn't work

[dark edge]

It's not as accurate as what I made but it's pretty close on my end.

That's just GetServerGameTime + SmoothedPing/2

[pallid mesa]

excuse the dirtyness

[pallid mesa]

so it might be that your smooth ping is OP

or I don't really understand what is going on here XD

[dark edge]

You should start with trying to get a nice smooth ping value

Show how you're doing your 10 window average

[pallid mesa]

void ARBPlayerController::NewClientUpdateWorldTime_Implementation(float ClientTimestamp, float ServerTimestamp)
{
    const float RoundTripTime = GetWorld()->GetTimeSeconds() - ClientTimestamp;
    RTTSlidingWindow.Add(RoundTripTime);
    float AdjustedRTT = 0;
    if (RTTSlidingWindow.Num() == 10)
    {
        TArray<float> tmp = RTTSlidingWindow;
        tmp.Sort();
        for (int i = 2; i < 8; ++i)
        {
            AdjustedRTT += tmp[i];
        }
        AdjustedRTT /= 6;
        RTTSlidingWindow.RemoveAt(0);
    }
    else
    {
        AdjustedRTT = RoundTripTime;
    }
    
    ServerWorldTimeDelta = AdjustedRTT;
}

[dark edge]

Why aren't you getting Ping

You can't derive ping, you need to get it

unless you're calculating it by sending stuff from client to server but you already have a replicated Ping in Playerstate

You could do a sort of pingpong thing where you go like

Client->Server send Time
Server->Client return that time and server time
Client Do Math

that would bake in an implied ping from the timestamps

[pallid mesa]

mhm, that's what I'm doing here, the whole algorithm is here: https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/#based-on-a-moving-window-discarding-outliers

click and you can see it

scroll a bit until you reach the cpp part, based on => client -> server -> client

if (GetLocalRole() != ROLE_Authority)
    {
        RequestWorldTime_Internal();
        if (NetworkClockUpdateFrequency > 0.f)
        {
            FTimerHandle TimerHandle;
            GetWorld()->GetTimerManager().SetTimer(TimerHandle, this, &ThisClass::RequestWorldTime_Internal, NetworkClockUpdateFrequency, true);
        }
    }

now, brb

[muted perch]

Now multicasts do drain performance, but is there another way to show a particle effect on all clients without it?

using the spawn niagara system node

[pallid mesa]

use a multicast

make it unreliable

[muted perch]

is that the only way?

[pallid mesa]

no, but it's the best way as vfx are usually one-off events unless it's a stateful VFX

[twilit radish]

If something else triggers it you can do it through that, imagine you open a door you likely already have something telling the client to open that door. If there's nothing in place for being able to know that particles should play somewhere then you need a RPC I suppose.

[pallid mesa]

correct if you are already on a multicast execution context for whatever reason, you can reuse that

there are a couple of use cases that are convenient

event destroyed executes everywhere so you an play particles at actor destruction without need of extra logic

[dark edge]

You either multicast or you spawn the particle effect based on some already replicated state

Like you wouldn't multicast a bullet hit effect. You would spawn the hit effect on machines locally whenever the local bullet hits something

[pallid mesa]

lmao

i mean i think it's one of the most important topics that rarely gets spoken out

[fathom aspen]

Yeah your answer to Thom cleared it up. Hard travel ensures objects are stably named, while seamless doesn't. That's a big achievement till this becomes a thing:

[fathom aspen]

That is one way 🙂
More ways here: https://wizardcell.com/unreal/persistent-data/

[iron crest]

Hello, I have a problem, I made my hotbar work with multiplayer, its functional, but visually it isnt, Im having trouble with setting brush from texture, I will provide pictures of the blueprint, it only seems to change the brush in single player/standalone, not client, I wqas thinking about using interfacs for this but figured it wont work, am I casting incorrectly? Im having no errors, any help would be appreciated, Thank You!

[serene kestrel]

I assume that second image is your HUD widget construction, what's the value of Get Player Character when that is running? My first guess is that runs before you spawn your replicated character in multiplayer scenarios

[iron crest]

Oh that is in the widget aswell, its just on begin play I cast to my main character and make that a variable, I need that as a variable because in my main character I have a hud reference which I need to pull in order to get a reference to the slotsd

[serene kestrel]

Ah gotcha, didn't read your full question, what does your brush texture setting look like?

[iron crest]

one sec will provide image

First picture is how im setting the image second is the data table which I pull the image from, (not sure if the DT is relevant but oh well)

The individual slot is just a blank image and a button, and I need that image to be switched to the jutsu image, it works in single player not multi tho

here is the hud, as you can see im casting there because that where the slots are

just providing as much info as possible

[serene kestrel]

if you put a breakpoint on the SetBrush, is your texture a valid reference? How about the Slot and Image?

[iron crest]

let me test this one sec, I will put breakpoint

yep validated get printed as succesful for slot 1 and the slot works, before I had it so that you could change your ability I just had some crappy system that was in the slot that updates the ability after some delay, not sure if that is relevant, so in summary, the slot has no problems, and the brush didnt give any errors

Do you think perhaps I cant edit widgets in different widgets for multiplayer? cause if so couldnt I just use an interface to communicate between the two widgets, not to good at multiplayer but just thinking of some ideas

Also the actual ability works just not the image, not sure if I mentioned that

[serene kestrel]

You can definitely edit widgets in multiplayer, when you debug in editor, is there only one instance of your HUD widget? If your calls to set brush are succeeding, then my next guess would be maybe you're somehow creating multiple hud widgets and displaying one that you're not updating

[iron crest]

Hmm on both clients its showing 3 different mainhuds on both clients, here is a pic

[serene kestrel]

Your first picture above where you create the HUD widget, what blueprint is that in?

[iron crest]

I create the hud reference and overall hud in my MaincharBp/ThirdPersonCharacterBp

[serene kestrel]

Ah, yea so the problem is that when there's multiple characters, your creating a HUD for each of them, but you really only want to create a HUD for the local player's character. You'll either want to spawn the HUD from something like your Player Controller, or in your Character's Begin Play, only spawn the HUD if the player is controlled by the local player. I don't remember the exact blueprint nodes, but there might be a IsLocallyControlled or something similar

[iron crest]

Oh I know its just a branch and then the condition is (IsLocallyControlled) Should I add that and test it?

[serene kestrel]

Yea, I'd try that in your BeginPlay real quick just to see if that is the issue

[iron crest]

Woah! you fixed it, Thanks Man, I cant beleive something like that would have halted my project for a while, Thank You! 👍

[uncut atlas]

can a regular c++ array be replicated?
example:
UPROPERTY(Replicated)
AActor* Items[5];

[serene kestrel]

I'd recommend moving your HUD spawning out of your character in the future. In multiplayer, the Player Controller only exists on the Server and the Owning client, so that's generally a good place to initialize HUD things

[fossil spoke]

Only Unreal Types and common POD types support Replication.

[iron crest]

Agreed, id definently consider, Thank you

[fossil spoke]

You would need to use TArray to replicate an Array.

[uncut atlas]

riparoonie. thnx.

[serene kestrel]

If you change your setup to

UPROPERTY(Replicated)
TArray<AActor*> Items;

It would work, mostly the same, you can even initialize it to size 5

[uncut atlas]

yeah i know. its just the way the existing code was set up in this project i'm working on used regular array. i'm gonna have to use rpc unless i change things in numerous areas of the code 🫠

[shell forum]

So when you want to execute something on the client only character on an actor, is the best way to call an event on that player’s character or is there a better way? I’ve been finding myself creating too many arbitrary functions for different actor interactions in the character class.

[fathom aspen]

Either that or look for suitable client execution paths, for example BeginPlay client-side

That variable has already been propagated, i.e, part of the initial bunch.

It got set using the value from the CDO when the initial bunch was sent.

So the default value (probably false) is what got sent

Yes. Set that variable after you spawn that actor

Setting the variable has to happen at the same frame the actor is spawned

You can, if you do it in one of the functions that are part of the actor creation cycle

That or try doing it in BeginPlay server-side

No delays or calling other functions

Yeah I will likely do that

No worries. Checking this server search tool can be helpful too 🙂

[twin juniper]

Is there any way to implement rollback netcode to Unreal? Has anyone done it in any game?

Obviously it sounds like a daunting task - haven't found much info material on the matter (UE related)

I've been researching what Overwatch did and it's pretty interesting

[kindred widget]

Why would a replicated TArray of structs not call an OnRep when one of it's structs replicates a property?

Full breakdown:
I have a replicated component on a replicated actor, with a ReplicatedUsing TArray of structs. The struct is just a GameplayTag and a Float. In server code I change the float. This gets replicated to the client(I'm testing this by making the client UpdateUI on a timer instead of the onrep delegate), but does not actually run the OnRep associated with the array that the struct is in when client receives that float change. This same OnRep is used for two other properties, which it runs fine for, just not this.

[twin juniper]

that's interesting

[mortal elk]

hey guys im creating a online multiplayer game and im using blueprints ( i dont know c++) the thing is i need help with online server setup. which way is the easiest way to do this i dont want a advanced version i just want 2-3 player can connect and play the game i made together

[obtuse field]

Does ue has any build in system for setting up maximum amount of players on the server, or do I need to implement it myself?

[fathom aspen]

It does. Check AGameSession::ApproveLogin. The property is called AGameSession.MaxPlayers

[fathom aspen]

Check advanced sessions plugin and make it a lan session probably

[regal geyser]

Is there somewhere a proper documentation for using Steam Voice Chat? Or is it as simple asi putting few lines inside DefaultEngine/DefaultGame.ini and executing ToggleSpeaking 0/1 ?

[obtuse field]

Do you know any tutorials that explain AGameSession?

[fathom aspen]

Not that I know any. I would just dive through source

[mortal elk]

hey i can create LAN servers but i want to create a non-lan server (idk what is it called) like im in A city and my friend is in my B city and i want to test and play our game together

[fathom aspen]

Then you want to look for a host service, for example AWS, which will cost you money, or they may offer you free lite versions.

[silent valley]

Or use something like Hamachi to create a virtual lan

[mortal elk]

Thanks! maybe after some development I may need a server to release it. i can use AWS but i really dont know how to use it. when i buy the AWS is it easy to setup? how to set after getting a host server

also can these be done with blueprint? or do i need to know c++

[fathom aspen]

Not that I have experience, but you would need to check their manuals and SDK docs. I would assume cpp is needed, unless there are plugins out there that do it for you

[rotund cosmos]

Hi! Will the GameState persist if I use seamless travel? or is it only the GameMode?

[fathom aspen]

Neither one will persist

They do persist to the transition map not destination

[rotund cosmos]

I saw this in the docs

I thought it meant the gamemode will

[fathom aspen]

Sadly that's wrong 😦

Or not explained well

Check my baby: https://wizardcell.com/unreal/persistent-data/

[rotund cosmos]

I will, thanks

[rotund cosmos]

so how would you solve a situation like this: Players join the server, they select their squads. When the game starts, the server travels to the game map. How could I access all the player squads that have been created?

[short arrow]

So I'm thinking about making 2 sub actors for every base character in my game. One sub actor will have a high net update frequency which holds very important replicated properties, and the second sub actor will hold less important / less likely to change properties with a very low net update frequency and focus more on utilizing force net update calls. Should I do this with 2 components instead though? I feel like it's more limited if I do it on a component. I'm having a big big issue with a lot of replicated properties that don't always need to be checked for updates

[fathom aspen]

I would persist an actor that handles team management. I talk about that here (example provided): https://wizardcell.com/unreal/persistent-data/#5-getseamlesstravelactorlist

@rotund cosmos

[fathom aspen]

Thanks Ethan 🥳 ❤️

[jolly siren]

np ❤️

[fathom aspen]

I would just have the properties reside in their logical classes and have PushModel turned on

You can also go for a similar approach to what Fortnite did, if you feel you are replicating too much data, and just replicate the bare minimum, by having a struct that entails such properties and pushing those values through some other actor

https://epicgames.ent.box.com/s/m1egifkxv3he3u3xezb9hzbgroxyhx89

They call this "proxy"

[short arrow]

The latter sounds kind of like what I was thinking

Thanks for that. Also push model?

Is there docs on that too?

Never heard of ot

[fathom aspen]

Pretty sure there are none but what you can find in source

Hopefully I will get the time to write on that one soon

Check PushModel.h

[short arrow]

I'm a blueprint pleb. But my biggest issue is honestly just the server checking a lot of stuff that it didn't need to check

For example I have a replicated variable named isinair. And that variable hardly ever changes right?

But its still being checked a lot

Every server update cycle

[fathom aspen]

Yeah that is PushModel territory really.

Though you'll hardly notice any gains

You can go fancy by using net managers if you feel you are replicating too much actors

[short arrow]

Probably not with that one variable but with the base character... with 10 players in the world its taking 0.1 seconds to get through all those replicated properties

[fathom aspen]

https://vorixo.github.io/devtricks/network-managers/ - from this channel pins

[fathom aspen]

Yeah the more players the more it's noticed. It was made for Fortnite so you get the drill

[short arrow]

It said 153ms I believe which is a lot. So I'm assuming if I can bring that number down by a good 100ms my server performance could go up right? Correct me if I'm wrong but I figure

If the player has an update frequency of 100 and it takes 0.1 second for 10 characters

That's gotta be eating a big chunk of server performance

[obtuse field]

After I override AGameSession, where do I register new class, or what should I do with it?

[fathom aspen]

PushModel will help decrease that figure but I can't quite off say how much that would be. You will have to do your tests after utilizing PushModel and if that wasn't decreased by a big margin then you'll need to consider Net Managers but that's a cpp territory @short arrow

[short arrow]

Yeah thanks for 2 options. I'll definitely look into both ✊️. Also heard about net manager, I believe vorixo made a whole blog post about it

Unlucky c++

Thanks again though

[fathom aspen]

He did 🦆

No worries 🥳

[fathom aspen]

I went bankrupt, brb

[twilit radish]

But honestly I wish connecting servers and servers in general aren't such a pain. You either need to use something specific to the platform/store you're on. For example Steam's integration. Which often is free though but then you want crossplay and the madness begins, do you roll with some company that provides relay servers and all that stuff for often way more money than you would like to, do you go for EOS which requires an account which for small games that need every player they can get really is not ideal either. Do you role your own system? Which is quite complex beyond a few players.. Then we have the entire dedicated server madness. Why can't it just be easy lol.

[sinful tree]

"We ... do the other things, not because they are easy, but because they are hard"

[quasi tide]

Cross-platform play is going to require an account regardless of what route you take. So EOS helping in that regard is a non-issue.

Make creating an account super painless and 99% of people don't care.

[twilit radish]

Crossplay does not necessarily require a separate account though. If I want to play my co-op game with a friend who has it on Epic and I have it on Steam I could authenticate with the platform specific accounts.

But yes it's definitely more common to have.

[sinful tree]

Yes, just a few simple details...
Email address, login name, password, drivers license number, SSN, DoB, 3 security questions, and whether or not they want to use 2FA. Easy.

[twilit radish]

The thing is that linking an Epic account or a different account definitely scares off people. I have a friend who was playing a Steam demo. You needed an account for it, the moment he saw that just clicked away the game.

[sinful tree]

yeah I'm not keen on linking accounts all over the place myself.

[twilit radish]

And I'm not saying it's the majority of people, but I don't find it ideal either lol.

[sinful tree]

Is it not possible to use both and just have the engine detect which one is in use?

Err.. I guess it wouldn't work too well for sessions and stuff..

[twilit radish]

Sorry, detect what?

[sinful tree]

Which online subsystem you'd be using (Steam / EoS)

[twilit radish]

I know Satisfactory does it, but I'm not sure how it exactly works I must say.

[sinful tree]

They link I think

[twilit radish]

At the beginning of the game you can pick either one and then roll with that, but to change you need to contact their support I believe.

[sinful tree]

Been a year and a bit since I recall seeing that message, but that does sound right.

Can you actually look up sessions in Satisfactory now?

[twilit radish]

I don't remember.

I played it a little bit 😛

[sinful tree]

I know more recently they released a dedicated server build, but I was under the impression you needed to use an IP. I don't have it installed myself right now or I'd check XD

In terms of just using an ID to authenticate, I know with steam you can grab their steam session ticket and validate it on the server through a Steam API call, I would imagine EoS may have something similar.

[twilit radish]

Probably yeah.

[narrow nacelle]

i need a quick refresher on something

when calling a RPC which takes in a UObject pointer from the client, the UObject needs to exist in the server too right

also, what happens if i change a replicated TArray in a client?

[kindred widget]

It does need to exist on server as well, yes. It does not necessarily need to be replicated. For example a StaticMesh will resolve the same on a server and client and as an asset it's never replicated. If it's a runtime UObject, it would need to be replicated.

As for TArrays. Nothing will happen immediately. The client will just have differing data than the server. The server will not correct this unless whatever you specifically changed was changed on the server though. If you remove enties on the client, and then the server replicates some other change to the array, it'll just populate empty entries.

[narrow nacelle]

so changes in replicated variables propagate from server to clients

unless you do conditional stuff

[kindred widget]

Correct

[narrow nacelle]

i'm doing some kind of command pattern for RTS

been having issues with executing them on the server from clients 😄

because some of the logic does not need to be serverside and they're hot calls

for example when moving many units into a formation, holding right click will activate rotation mode for the final formation, which will probably run on tick or some fast timer

it should probably call a RPC with the final unit positions when you release the mouse button after rotating

[kindred widget]

synonymous to Total War?

[pallid mesa]

thankss :3

[narrow nacelle]

i can't remember how it worked in total war

but imagine when you right click to give a movement command, you can hold to give a specific rotation

to face the formation to a desired direction

when you release then the actual commands will be send to each unit with their respective position

[kindred widget]

Basically that. Total war also sized the formation's width based on how far away you moved the mouse from the original right click point.

[narrow nacelle]

i tried raw classes for the command object, which made everything much more complicated

[narrow nacelle]

oh, i was thinking about using the scroll wheel to adjust the spacing while holding

[narrow nacelle]

now i changed it to UObject to be able to benefit from the engine's network pointer magic

[kindred widget]

I'm not sure why you would need more than an RPC of structs. Struct has location, rotation, and intended unit width?

[obtuse field]

Does even AGameSession exist on Dedicated server?

[narrow nacelle]

the command class has virtual functions which can be overridden in derived classes

there are gonna be command objects for movement, attacking, patrolling etc

they are pretty encapsulated, they manage their own lifetime

so i need to create the command objects in the server

[kindred widget]

Would probably just keep that with the lifetime of the unit and replicate it to who needs to see it. When giving an order let client locally manipulate it. Send the order through the unit's owner(PlayerController) as a struct. Order type, array of intended movements as location,rotation,etc. Server can replicate the new state to everyone if everyone needs to see it or just keep it on server as where/what the units will move/do. Should be pretty light. Less object creation with constant order spam, etc.

[narrow nacelle]

the orders are created in the player controller, and then added to a command queue array in the unit's AI controller (which does not exist in clients)

the AI controller's command queue is kind of a FIFO buffer

it starts executing from the oldest command if there's something in the array

i guess for now i can do a base implementation which has most of the stuff running on the server

[kindred widget]

Just imagining visuals and stuff. Waypoints and whatnot. Just seems easier to attach an order object/actor to the unit and let them share lifetimes. The controller can even get it directly from it's unit if need be. Or the order thing can be a sub brain and send new orders to the controller once a current order has been met.

[narrow nacelle]

virtual void Execute();
virtual void Cancel();
virtual void Finish();

these reside inside the command object

the AI controller calls execute for the oldest command

that part works pretty well right now

i'm kinda lost on efficiently creating the commands

i think i'll move more logic in the player controller to server side

because when clients try to issue commands, the RPC gets a null command object

[thin stratus]

You could send a struct

But rpcs can't send child classes of structs

So... Hm

[timid tendon]

Hi guys. I've got a weird issue with my character animation glitching while in playing multiplayer. It does not happen if I open multiple session from UE editor as a listen server - only when testing over the network.

My only guess is that something is not replicating but for the life of me I don't know what.

[narrow nacelle]

my first implementation had raw classes wrapped with USTRUCT

which failed, i think i need to serialize the raw class anyway

[thin stratus]

Do your commands need custom data per command?

Like runtime

[narrow nacelle]

for example, move command has a location and a direction for the unit to face when it arrives at the destination

it also overrides Execute() from the base command class

[thin stratus]

Yeah that's tricky. Would need an rpc per data set which is shit

Does the server have all information needed by chance?

Aka can it generate the data

[narrow nacelle]

that's probably the part i'm having problems

there is an actor array in the player controller

for selected actors

[thin stratus]

I mean generally you only need to sync data generated by inputs

Mouse location for example

[shell forum]

How about SetOwner()? Or would that mess up server RPC calls?

[narrow nacelle]

when you right click, i'm doing a get hit results under cursor then determine what command we need to create

[thin stratus]

SetOwner only allows one owner at a time and needs to be called on the server

Despite that you can use that to allow rpcs

[narrow nacelle]

then for every selected actor, we create a command object and send it to the respective actor

[thin stratus]

Yeah so with GAS you usually just send the Mouse Location to the server and let the server do the targeting

You could also let the client do targeting but cheeeaaters

[narrow nacelle]

for some commands i need to create a command object for every unit because for example to move units into a formation they all need to receive a move command to their respective position

the formation positions are calculated from hit location, i was doing that in the client

i couldn't think of a reason to do that in the server

but i think i'll just move the whole thing to the server for now

call a rpc with the trace results, and determine command type in the server, then issue the commands

but then i need to move unit selection logic to the server too, as the server needs to know which units a player selected

i implemented the selection logic client side as it had no gameplay effects, you couldn't issue commands to units you don't own anyway ( even if you cheat to select them )

[narrow nacelle]

is it worth implementing gas for RTS?

now that i think of it, it handles network stuff too so lowkey makes sense

would it scale well with a few hundred units?

( probably scales better than my own implementations anyway, even if its marginal lmao )

[chrome mortar]

Hi, I'm new to MP, I'm currently trying to implement MP into my project that has GAS in it, I had to clean up a lot of stuff in order for it to work, but with this one particular problem I'm at my wits end... The character doesn't have anything fancy going on with Movement component etc... nice and simple movement, yet it lags both on Client and Listen server... any other Pawn/Character works perfectly fine, yet I'm strugglin with this one...

[narrow nacelle]

its gonna be the third rewrite of the command system if i'm gonna implement GAS

[chrome mortar]

Anyone any experience with this kind of Lag??

[shell forum]

You need to set the movement speed on both the client and server.

[chrome mortar]

I'm setting the base speed with attribute sets, thought that this way it would be set on both, how does one set it on both?

[shell forum]

https://nerivec.github.io/old-ue4-wiki/pages/authoritative-networked-character-movement.html

It unfortunately can’t be done in BP if you’re adding sprinting.

[chrome mortar]

I have that set up tho

Basically the whole MovementCOmponent

Although I took from the Tranek/GASDocumentation

Nearly identical

[serene kestrel]

just to verify it's corrections that are the problem, can you do p.NetShowCorrections how are you setting the base speed? If it's through a gameplay effect, it should be on both unless your attributes aren't replicating

[fathom aspen]

SetOwner(nullptr)? non-owned actors behave like server-owned.

[chrome mortar]

Whenever I try to use that command, it doesn't really do anything

Am I using it wrong perhaps? "p.NetShowCorrections 1"

[serene kestrel]

let me double check

yea that should be the command, if you're not seeing red/green capsules showing up, the the "lag" isn't due to movement corrections, it's something else

[chrome mortar]

Then I'm even more confused as to what could be causing it

I had a problem like 2 days ago where the attribute set refused to give me anything else other than 0, which resulted in an actual cluster ******* of movement lag on client, while the server was ok... But I just forgot to replicate MovementSpeed, which I fixed, but now I'm stuck with both client and server lagging just a tiny bit

[chrome mortar]

Any other idea you might find plausable?

Cause I straight up ran out of them, been testing 'till 8AM, got 3 hours of sleep and have been trying to fix it since

[serene kestrel]

Oh this even happens on your listen server's character? So not replication related then, probably something going wrong with your movement simulation. How did you integrate the attribute sets into your char movement comp?

[chrome mortar]

The component calls a getter for the movement speed, which just returns the value from the attribute set

Have just been testing if the function is an issue, it's not, so ignore the 600.f

The attribute set has delegates bound to it's attributes, so whenever I change the attribute, it changes the value of MaxWalkingSpeed

[serene kestrel]

In the video above with the laggy movement, is that just walking speed that was using the Owner->GetSpeed() before you hard coded it?

[chrome mortar]

yup

[serene kestrel]

And does GetSpeed() just return your MoveComp::MaxWalkSpeed value?

[chrome mortar]

And with the 600.f it's the same deal

Returns a value from the attribute set, which should be the exact same

[serene kestrel]

I'm confused what the SpeedChanged callback is doing then, if you're just reading attributes from the attribute set, what is that MaxWalkSpeed that you're setting?

[chrome mortar]

Well, if, let's say, a gameplayeffect changes the value of an Attribute "MovementSpeed", it will change the value of MaxWalkSpeed according to the attribute

wait

Doens't make sense, does it

Am I just doing this whole thing wrong?

I don't even know... Changing it did absolutely nothing xd

[serene kestrel]

In your GetMaxSpeed function, if you just change it to return the Speed Attribute's current value, do things stop lagging?

Also can you just paste what your Owner->GetSpeed() function does

[chrome mortar]

Idk, still continues

[split siren]

What is the overhead for encapsulating behaviour in a component? I have noticed that CharacterMovementComp is using Character for all of its replications and now I am not sure what is the most optimal way to handle encapsulating replication.

[serene kestrel]

When you send RPC's through a UObject (aka actor component) the payload uses an extra identifier to specify which object the Function needs to be called on. So if you're sending an RPC with very high frequency (every frame like movement component) then those extra identifier bytes can add up. I think that's really the only concern, so you could probably get away without doing it, but just paying more in bandwidth

[kindred widget]

It's not a lot. But also do be aware that CMC is older than Components having replication or networking. It was less of a design choice for that component.

[pallid mesa]

https://vorixo.github.io/devtricks/gas-replication-proxy/

[split siren]

Thanks for the explanation! I guess it's not something to worry about. @kindred widget I believe moving the replication from component was quite a recent change, this is why it was so concerning.

[kindred widget]

If it makes sense to keep it in a component, keep it in the component. Chances are there are likely many other ways to trim performance issues 'if' you actually run into them.

[blazing spruce]

Hi, I'm having an issue with my interaction, I have a base class that all interactable objects derive from, which is used to display some widgets indicating that an object can be interacted with, when a player overlaps with the interaction overlap box it adds the actor into a array of overlapping actors and then it would display the widgets

then when the player actually interacts with the object that they're overlapping, it checks if the actor is contained in that array, however for some reason this works fine for the server player but all clients are returning false on this if check as if they're not overlapping the box, which they are

any ideas why this wont work for clients?

Also my interaction set up

The print statement after being added into the overlapping actors array in the first image is telling me that the length of the array is 1, like it should be too

[fathom aspen]

You are populating the array client-side

[fathom aspen]

And checking if the element exists server-side

Which it doesn't

[blazing spruce]

Whats the best way to do it to push to the server? because all my interactable objects derive from this class so i need it to add the actor into the array from here i think

whats weird is that its been working perfectly for ages but for some reason stopped lol

[fathom aspen]

It works fine for server, but not for clients.

Keep it all server-side I guess

No IsLocallyControlled involved

Instead make sure you're on server (gate with authority for example)

[blazing spruce]

I need to keep the IsLocallyControlled check, without it if either the server or client is standing in the box overlap of an object it displays the interaction widget but everyone can see it (image below), with the IsLocallyControlled checks it makes it so only those within the overlaps see the widgets

[sinful tree]

Then you'll probably need to use a sequence. Top can use the Has Authority check, and the other that uses IsLocallyControlled to display your widget.

[fathom aspen]

Exactly, those are different execution paths

It's sad how this guy copied Cedric's compendium without even giving him credit: https://russianblogs.com/article/80371517154/

Unless I'm missing something

[sinful tree]

*1: Despite my own knowledge, I used the official Documentation, as stated at the
beginning of the Compendium.
*2: The main network framework, as well as the two (recreated by myself) images
of the distribution of common classes is documented by 'Nuno Afonso'
and can be seen here: www.nafonso.com Thanks for this amazing page!
*3: Nearly all examples, Blueprint Pictures and C++ Code, are created by myself.

Which is both in the compendium and on this website 😛

[fathom aspen]

Yeah that part made me laugh hard

That must be Cedric in disguise 🕵🏻

[sinful tree]

https://tenor.com/view/meme-our-now-gif-21036569

The guy is trying to make it our compendium.

[blazing spruce]

@fathom aspen awesome thats working perfectly now, thanks both!

[vocal current]

One way to do this could also be to allow the client to ask to interact with an object (populate the overlapping list on the client), but don't populate it on the server.

Instead on the server, just do some sanity checks to ensure that the player isn't cheating and can reasonably interact with that object (distance check, raycast to ensure that player isn't trying to interact with something on the other side of a wall, etc. Remember to add some leeway in the distance check because the player can be in a slightly different position on the client due to network latency)

You don't necessarily need to know if the player is overlapping the object on the server, you just need to make sure that the player isn't doing something dodgy with a hacked client (like trying to interact with something on the other side of the map or through a wall)

[graceful flame]

I'm following Vori's guide for a "non-destructive and better synced network clock", but I already have a player controller blueprint that I'm using. How can I use the timer with my existing Third Person Player Controller without having to make a new one? https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

Do I have to make a new PlayerController blueprint which inherits from the new CPP one and manually migrate all of the blueprint events, nodes and functions over?

Different questions each time though right?

Oh that sounds like it would save me lots of time, is there some sort of hidden away right click menu I need to click?

ahh i found it

Class settings, thanks!

omg I didnt even know reparenting was a thing

I thought it was more akin to IRL where you're just stuck with whatever parent LOL

[graceful flame]

So this gives me a float that counts up from 0, but how can I use this sever time to know when one second has passed and then decrease a widget timer? Currently I'm starting with an int like 500 or something then after a one second delay the pvp match timer is decremented by one. The replicated pvp match timer lives on the game state.

[pallid mesa]

hey lawlster! whatsup ~

you could use it as you'd use any other increasing float

I'd need more details

[vagrant grail]

Simple question :
Usually we tell people to not have input parameters on Server Events (RPCs) to avoid people cheating by sending the value they want for that input parameter, but why the node "Event AnyDamage" (That's the event receiving the damage from the "ApplyDamage" node) has an input parameter and is used on the server ?

Does that mean people could cheat ?

Please ping me if any answer 🙂

[graceful flame]

yeah thats what im not sure about how id use an increasing float to decrease a shared timer all clients can see

Like a pvp match timer for example

I wonder if its not the right tool for the task....

[sinful tree]

You can call Apply Damage only while running on the server. The idea being, you'd detected whatever hit on the server, and calculate whatever damage on the server, then apply the damage (which would be on the server)

[pallid mesa]

@graceful flame well you can do it by substracting the max-match time
Max(MaxMatchTime - RepliClock, 0)

[graceful flame]

Yes that makes sense

[vagrant grail]

Am I able to do the same by creating a server event that could be called only by the server ?

[fathom aspen]

Of course RPC can have input params. It's the developer's responsibility to make sure clients don't cheat. So after the RPC is sent, the server makes sure the client's provided values are intact.

ApplyDamage is BlueprintAuhtority function, i.e, can only be called from server, so it's not an RPC. Anyways, by same reasoning you don't call ApplyDamage on an actor unless you're sure that actor can be damaged (you're already on server by the time ApplyDamage is called so you can do your checks before that function is called)

[pallid mesa]

@graceful flame just be careful cause u cannot trust the clock right when you join!!

so u gotta wait until it syncs up

[graceful flame]

okay

[sinful tree]

Any event that you mark as "Run On Server" means you're allowing the client to ask to execute that event.
Therefore, any event that you don't explicitly call while running on a "Run On Server" event means that it will only ever execute wherever it was called, whether that be the server, or client.

[vagrant grail]

But how Epic succeed doing a node that could be called only by the server without the "Has Authority" node ? 🤔

[fathom aspen]

UHT does it.

i.e BlueprintAuthority

[sinful tree]

In C++

[vagrant grail]

but is it possible to do that through blueprint too ?

[sinful tree]

UFUNCTION(BlueprintCallable, BlueprintAuthorityOnly, Category="Game|Damage")
static float ApplyDamage(AActor* DamagedActor, float BaseDamage, AController* EventInstigator, AActor* DamageCauser, TSubclassOf<class UDamageType> DamageTypeClass);

Specifically the "BlueprintAuthorityOnly" tag is probably what does it.

[vagrant grail]

Oh nice

[fathom aspen]

Not without using that tag. So pure BP custom events can't have it

[vagrant grail]

I have another question, I have this code handling my health and shield system, and I need to modify it to handle 1 other thing. I have a boolean on my "Damage Zone" actor named "Shield Damage" which if set to true, the system takes the damage received and remove first the shield then the health (like in Fortnite, Apex,...etc) but now I want with my boolean "Shield Damage" if set to false, I need the system to not care about shield first and directly apply the damage to the health, what should I modify in my code ?

I tried to add a cast node to my "Damage Zone" to check if the Damage Causer is the Damage Zone and then retrieve that boolean variable but didn't work properly

[olive kraken]

Hey!
We are about to run the first alpha in our game CRETE. Was wondering if there is any plugin/API in the marketplace/Github to send data in-game into a database and bug reporter. Any tips about how you are monitoring gameplay, bugs and stats?
I'm aware of steam leader boards, thou wondering something more development oriented.

[sinful tree]

You'd probably be better off creating damage classes and start using them for applying damage. This way you don't have to care what specifically hit the player, rather the type of damage that is incoming dictates what happens. Then you'd just need to check the class of the damage type, or if you want to get more advanced, you create properties within your damage classes that you can then read and use in your damage calculations, like ignoring shields.

[vagrant grail]

That looks far complex for me as I don't have that knowledge I think. The issue is It's mendatory in my case to have that boolean on my Damage Zone instead of the player as each instance of Damage Zone has its own version of that variable being either true or false

[sinful tree]

If it's a damage zone, why would there need to be a true or false option on it?

It should just try to deal damage.

[vagrant grail]

Because maybe you want one zone to handle shield during damage and the second zone you don't want it to handle shield and directly hit the health

[sinful tree]

And therein lies why damage types would be better - you instead could define the damage type (by class) directly rather than using booleans.

[vagrant grail]

Wouldn't an Enum be better ? Because I don't see how to dod that with a class as I never saw something like that yet 🤔

[sinful tree]

You've been using classes the entire time. Every blueprint you use is a class. You define variables in a class, and you can read variables from classes. You can make child classes and set different values for specific classes.

The idea here is to get away from needing to know what caused the damage and instead focus on what type of damage was dealt.

[vagrant grail]

Yeah I know that a blueprint is class but I struggle to imagine how a damage type could be a class and what should I put in it (properties and methods). What makes more sense to me is an Enum containing different Damage Types

[sinful tree]

To create a DamageClass, you just need to create a blueprint based on DamageType class.

You put whatever properties in it that you want to help you define what "Damage" types there can be.

Then you create sub classes to help you fully define an actual damage type.

So if that right now is as simple as a "Ignore Shields" boolean, that's fine.

Leave it set as false by default. Create a child class and set it true.

When you go to apply damage, select the child class - you now are passing through the damage type that indicates it ignores shields.

On the "Any Damage" event, you can cast the Damage Type to your parent class - you can then read the variables from it.

If the cast fails, that indicates you haven't set up a damage class in the Apply Damage node, so you could log it for later revision, and just apply all damage as you would by default.

[vagrant grail]

my brain just melt with these instructions 🥲

I think it's above my knowledge to do that, so I think I would keep going with the way I wrote it but just I need help on where to put the part to bypass shield damage when "Shield Damage" is set to false

[sinful tree]

It's not above your level of knowledge if you know how to:
A) Create a blueprint.
B) How to create a child blueprint class (which is basically the same thing as A)
C) Setting some variables.
D) Casting and reading variables.

Then it's just a matter of feeding in variables where you need them, just like you would anything else.

[vagrant grail]

I did step A and B, what variables should i put inside now ?

somehow I already have some variables created by Epic

[past totem]

this node can only be called from the server ?

yea what wizardcell said - ApplyDamage is BlueprintAuhtority function, i.e, can only be called from server, so it's not an RPC

[vagrant grail]

Thanks and do you have a solution for my other problem please ? 🙂

[past totem]

@vagrant grail bro your problem is like 100 messages long

[vagrant grail]

@past totem Here's the problem you only need this message to get everything you need to understand my problem 😄 ☝️

[past totem]

why is this even in here? go #blueprint or something. this has nothing to do with replication or multiplaeyr ?

[past totem]

your question is phrased so unclearely. what is this shield damage boolean? so u mean that u want to make it so a special type of weapon will skip the shield damage and damage health directly no matter if there is shield or not ?

the correct way to do this would be using the DamageType, you can either create a new damage type like DamageType_SkipShield and then check if it's that class (or cast or whatever) and in that case skip shield. or have a function in a parent damage type that does checks and calculations for what u want which can be overriden in any child damgetype

or, you can just not use AnyDamage at all, and use ur own damage system and create ur own damage function with SkipShield bool, but probably not a good idea

[vagrant grail]

"shield damage" boolean is a boolean inside my damage zone, and if it's set to true, the damage zone will cause damage to the player buy bypass his shield if has some (as the player has some shield and some health and usually the shield gets consumed first then the health)

[past totem]

ye that's the solution then

[past totem]

vote for me issue : https://issues.unrealengine.com/issue/UE-149477

[vagrant grail]

voted !

[past totem]

thx

[sinful tree]

Whatever variables you want that differentiates your damage types. You can use booleans if you so desire like "SkipsShield". You can think of this parent class as the "base" for what all other damage types derive from and any child classes are just setting the values.

As I pointed out, you don't have to use variables here right now if you don't want to - you can just create classes and then switch your logic around by casting the damage type on the Any Damage event, but it gets a bit hard to manage. If you do use variables, then you only need to cast the damage type on the Any Damage event to your base damage type and grab the variables from it, so again, if you made a "SkipsShield" boolean, you can get it from the cast node.

[vagrant grail]

Do you have any tutorial showing up that, because if I don't see it I don't understand how to link stuff together. I always learn by seeing someone doing it first then apply it to my case then I understand how to do it 🙂

[vagrant grail]

Thank you it makes more sense to me like this 🙂

[lunar dirge]

Anyone know any free way to make a dedicated server for my steam game ? I have PTP set up already but my goal Is a dedicated server.

[undone kindle]

AWS, DigitalOcean, or similar free credits?

Or host it on a box in your house

[lunar dirge]

Nice thanks

[twin juniper]

Question for people working on a shooter game, i've been struggling with an issue for over a year soon with replication race condition, in some rare cases, player will spawn without equipping his weapon cause owner had not replicated yet, so i ended up adding a call to my equip function in OnRep_Owner annnnnnnnnnd then another issue came where simulated clients can't see the weapon in other player hands in rare case also, ending up with 10 call to my functions in a few OnRep and yeah i still have some race condition actually 🙂

So i saw UE4 was handling it like that, putting the weapon as pending while it has not replicate, my question is what is the best way to handle things like that, when switching to a not-yet replicated weapon ? Put a timer that runs until owner/instigator are valid and then switch to the weapon ? Do like UT4 and putting the weapon as pending and get 5 rpc to execute to switch to a weapon or yeah idk, feel free to suggest me anything 🤷‍♂️

I really don't like the way UT4 does it cause it runs a lot like A LOOOOOT of rpc to simply switch to a "valid/replicated" weapon

[jolly siren]

It's a common issue. I see simulated proxies without weapons in their hands a lot in fortnite too

[twin juniper]

I guess a bEquipped replicated property that attach/LinkAnimClassLayers in its OnRep would fix the issue permanently, actually it’s not replicated, my SwifchWeapon is like ShooterGame, every proxies execute the equip function

[twin juniper]

I fixed « attaching » by putting it to ClientGivenTo, so client is assured to always have a valid Character(instigator/owner) pointer

Now my issue remains with my LinkAnimClassLayers, so I’ll try to replicate my bEquipped and yeah I’ll see, it feels very dirty still, @jolly siren how u guys was handling that in Splitgate ?

[timid tendon]

I've got a weird issue I'm trying to get to the bottom of. My moving animation glitches heavily when the game is tested over the internet. This only happens on the client, the server runs smooth.

I'm not able to replicate it by upping the number of players in the editor, making ti extremely difficult to troubleshoot. I've tried posting to unreal forums and reddit but no luck.

I am pretty much at my wits end, could anyone point me in the right direction please?

[dark edge]

Show it with the capsule visible

Also show your animation BP

and confirm you aren't doing anything goofy like trying to replicate walk speed changes all the time.

Do you have an acceleration tilt mechanic?

[timid tendon]

I do have an acceleration tilt machanic, but not the tilt values are not replicated

[dark edge]

That's probably the culprit, show it

[timid tendon]

I'll take a video with the capsule tonight when I get home, as well as show the animation BP and the tilt mechanic. Unfortunately I'm at work now so it'll be a good couple of hours

[dark edge]

My guess right now is the acceleration tilt is reading some value that's not smooth on client. But just post everything and we'll take a look

[timid tendon]

How can I tell what the difference is server vs client? Just print a debug message and compare?

[dark edge]

The print will tell you what machine it's on

If you remember, how are you doing your acceleration tilt?

it SHOULD be derived from purely clientside info, either acceleration if provided by your movement solution, or computed from velocity and last frames velocity

[timid tendon]

These are the calculations involved - I feed the values to an additive blend space

[timid tendon]

I can't remember how I set my current velocity variable off the top of my head, but it's probably just pulled from the movement component

[dark edge]

Assuming that all works fine on server, check that you're getting clean values for CurrentVelocity

I wouldn't be doing all that vector breaking though, just work with vectors and zero out Z if/when you need to

Vector x 1,1,0 zeroes out Z

[timid tendon]

Alright, good advice. I'll check the velocity values tonight, thank you

[dim briar]

Why my replication Work in editor ( Local server ) But after build doesnt ? ( local )

Editor Correct Working
https://drive.google.com/uc?id=1wbOwOSYMMVn2Gh1wjKpEFYN6orQwj0Xj

Build Issue 😦
https://drive.google.com/uc?id=1bfgP2L4N1coHyA4pGoOxZkJH7OYQpamF

What is difference between local editor vs local builded ? Can someone explain me ? its frustrating...

[edgy escarp]

How do I prevent SetActorLocation and SetActorRotation from replicating when being executed by server? I've disabled bReplicatesMovement for the actor and all its owners.

[formal solar]

I'm trying to destroy an actor in a player controller BP in a server side function, is that illegal or something?

Meaning I run logic inside the player controller BP that destroys an actor I have a reference to

No error message but it just doesnt work

[short arrow]

why are clients still receiving updates from a dormant actor?

I have a barrel with a health variable that changes every x seconds, but it's replication is set to dormant all. So why are the clients still up to date on it's constantly changing health? 🤔

[chrome bay]

I suspect that the channel hasn't fully closed once it's changed again

[short arrow]

it starts off as initial and never changes

I suspected clients would never receive an update

but they receive it consistently

[chrome bay]

The client has to "ack" the most recent state before the connection goes dormant, but closing the actor channel usually takes a few seconds. Could be it changes too often for the channel to close

[short arrow]

it's set to update the health every 5 seconds

surely that's enough time 🤔

[chrome bay]

I'm not sure, it's quite variable - relevancy etc. is a good few seconds too

Also is this BP or C++?

I've no idea if setting a replicated property in BP means it automatically flushes dormancy

[short arrow]

BP it's just an actor that spawns in with initial. It's so weird I didn't have this problem in a different project

maybe I should give it a few seconds and then tell it to go dormant, I'll play with it more

[chrome bay]

try not changing the health and see if it goes dormant

[short arrow]

when I get the variable it says that it's dormant, not sure if it's reliable

[chrome bay]

But yeah IIRC, it won't truly go dormant until the client has acked the most recent state

[short arrow]

Update: gonna rip my hair out

😔

[bitter oriole]

I would also recommend against that

Wish we could actually recommend things

[quasi tide]

@whole grove has been bleeding into #multiplayer more frequently lately...what are you up to 🤔

[short arrow]

https://gyazo.com/693dc8d1a2a57bf78cc0f69166ba3e13 😔😔

also is it normal for their to be a server print for every client? This is actually the first time I'm seeing this

It literally says that it's dormant all and then proceeds to update for all clients wtf

[fathom aspen]

https://forums.unrealengine.com/t/net-dormancy-not-working-at-all/428661

Seems like a bug

Time to consider moving to cpp

[short arrow]

https://tenor.com/view/angry-angry-white-woman-angry-middle-aged-white-woman-upset-tantrum-gif-8128195

C++ is honestly my 2023 resolution. The game is so close to beta man. Im getting rekt by a stupid blueprint bug at the last minute

7 tickrate servers it is! Thanks for letting me know the problem wasn't me though, I can rest now

[twilit radish]

It's weird how they marked it as "won't fix" though.

But then again, Unreal doesn't seem to really like BPs for multiplayer in general I suppose.

[short arrow]

Ikr, I feel like epic is a bit disingenuous about being able to make full complete games with blueprint. Like I mean you can but soon as it gets slightly complicated youre screwed.

[twilit radish]

I believe they have said them selves that's not the intent of BPs. Any game should honestly use a mix of it, BP and C++ both have their ups and downs but when it comes to networking Unreal often just skips out on BPs which indeed makes anything that's slightly more complex than the basics impossible or not viable. Things like characters, optimizations, more advanced replication is all just not available in BPs.

Not hating on BPs btw, I think it's just fine to use but Epic is definitely not going to make it easy for you if you want a multiplayer game with just BPs 😅

[split siren]

Any idea how Fortnite is handling replication of firing of weapons? In one talk they about GAS they mentioned they have a custom way outside of GAS, but I don't think they mentioned details.
I am fairly certain they mentioned that damage is not predicted, but the rest is, and the prediction is what I am really curious about.

[quasi tide]

Don't try and predict damage.

[split siren]

No worries, that is one of the few details Epic mentioned. But the rest is a bit of a black box for me.

[latent heart]

It's the designer friendly version!

[fallow shadow]

as in

the actor is replicated

i guess its destruction should be broadcast to every client

aight

@long elkrq before i go do this

um

so this happens on standalone/listen server (server) too

so idk if this is the issue

[timid tendon]

I'm not sure exactly what you mean by clean value for velocity. Here are the screen prints of the client and server. They are different but not sure what to do to reconcile them? If that's what I need to do - I am very new to multiplayer. I have nevertheles addedd them as screenshots - this is my currentVelocity vector. I've also added the screenshot where it is set.

I disabled my leaning additive and the jerkiness stopped, so that is definitely the cause.

[obtuse field]

I setup a max player variable in my game session, but still more than 2 players can join

ALeapNGameSession::ALeapNGameSession()
{
    MaxPlayers = 2;
}

[quasi tide]

Ahh - the ol', give 'em what they give me treatment

[dark edge]

Draw debug line that velocity and make sure it's nice and smooth and pointing in the direction you're going

Then if that's smooth, do the same for your calculated acceleration and make sure that looks sane

[manic trellis]

can you please give some descriptive information.
This way the question is not very much answer able

[timid tendon]

OK great, I will try that. If it isn't smooth, what is the likely culprit? I can't check right now as the other devs that can help me test are offline so will most likely have to check it int he early morning hours

[dark edge]

You'll have to trace it down. Maybe you're getting intermittant 0 velocities, who knows

maybe you're getting delta time spikes

[timid tendon]

OK understood. It could be a wide variety of things. Once I've figured out what the problem is I will try and isolate the cause

[slim mist]

Is it possible to stream a render texture between two clients in a lobby (not session)?

[dark edge]

Maybe but that sounds pretty awful.

What are you trying to do?

[slim mist]

Imagine you're in a fighter jet cockpit that has MFD screens (the square digital display). You're buddy is playing by himself in his own server, but you both are in a lobby. You can look at your MFD to see what your buddy is up (a scene capture component attached to his vehicle in his separate server) on your MFD screen

except replace fighter jet with mechs

another idea is that despite being in separate levels, you might be able to set up artillery to provide support for your buddy, and having some kind of visual feed on your buddy's game world to see the effects of your artillery would be rewarding

[dark edge]

That is a HUGE undertaking.

You'll be much better off syncing whatever base state is represnted in that screen

You wouldn't be sending the video over the network, but the stuff used to render that video. Still a big undertaking if you're in separate servers.

Assuming you figure out how to send data to a client on a different server at all, you'd send the stuff needed to render that screen (models, transforms, etc), not the raw video itself.

[obtuse field]

Well, I setup MaxPlayers, so no more than 2 players should be able to join the server, but still more players can join for some reason

[fathom aspen]

I assume beacons are needed in such case

[obtuse field]

As far as I know, this should prevent more people from joining

[slim mist]

Looks like UMediaPlayer has functionality for replication. I wonder if I can use a render target as the source for it

[fathom aspen]

Debug PreLogin and figure out why it's letting more people in

@obtuse field

[manic trellis]

is you setting the session settings struct before creating a new session
there you can specify the max player to join the session or that specific session.

[cosmic yoke]

hello, i'm doing a multiplayer game and i'm having a problem where when a client crashes, its pawn still stays visible for the other clients for 30 s / 1 minute. is there a way to reduce the heartbeat frequency so the server game mode detects faster that the client is no longer there?

[dim briar]

Why my replication Work in editor ( Local server ) But after build doesnt ? ( local )

Editor Correct Working
https://drive.google.com/uc?id=1wbOwOSYMMVn2Gh1wjKpEFYN6orQwj0Xj

Build Issue 😦
https://drive.google.com/uc?id=1bfgP2L4N1coHyA4pGoOxZkJH7OYQpamF

What is difference between local editor vs local builded ? Can someone explain me ? its frustrating...

[oak oracle]

Hi devs i got a small issue , I got multicast rpc on the character , which runs when this character got eliminated , and i want to print on the screen of ONLY eliminated player a message , but this message got printed on all clients and server as well , Why ?
void ABlasterCharacter::Eliminated_Implementation() { if(this->GetLocalRole() == ENetRole::ROLE_AutonomousProxy) { GEngine->AddOnScreenDebugMessage(-1, 14.f, FColor::Blue, "You are eliminated"); } bEliminated = true; PlayEliminationMontage(); }

[kindred widget]

I tend to use Kismet's Print for network debugging. Shows which machine actually ran the print.

[oak oracle]

alright thanks a lot !

[unborn nimbus]

How can I get all relevant actors for a given connection?

[fathom aspen]

Iterate over actors server-side and check IsNetRelevantFor?

I guess you can iterate over actors client-side (without checking), though you will get static actors too

I would assume ReplicationGraph already does that

[obsidian oyster]

hi there. a question for anyone who knows. I've set up a tcp connection using FSockets. I've successfully connected, and can receive and send data over the connected socket. I'm getting some garbage unicode data at the end of my message. I've been using examples online for establishing the amount of data that should be recieved, but the unicode garbage still shows up. Here is my code on recv

        TArray<uint8> ReceivedData;
        uint32 Size;
        
        while (Socket->HasPendingData(Size))
        {
            ReceivedData.Init(FMath::Min(Size, 65507u), Size);
            int32 Read = 0;
            Socket->Recv(ReceivedData.GetData(), ReceivedData.Num(), Read);
            FString ReplyMessage = FString(UTF8_TO_TCHAR(ReceivedData.GetData()));
        }

i'm assuming its how i'm calling init on my ReceivedData object, but haven't worked with this stuff enough to know the solution.

[obsidian oyster]

i figured out the above. i was getting garbage because the data returned from RecievedData was not null terminated. i fixed this by adding ReceivedData.Add(0); before my conversion to FString.

[split siren]

If I have a replicated client-authoritative value on character (such as bIsAimingDownSights ), would you replicate it with COND_SimulatedOnly or COND_SkipOwner?

[fathom aspen]

I'm more used to see SkipOwner, but I guess it doesn't make a difference if you chose the former in case of Character

[fair coral]

Is it worth using raw sockets or should I use RPCs like they are sockets? Where I call and RPC and parse a struct containing all the data I need?

[twilit radish]

What are you trying to do?

[fair coral]

Want to do a rollback netcode implementation (done it in other langs before but using raw UDPSockets) so I am wandering whether to send structs across where I would do socket.send(bytes) I should just do CallRpc(bytes) ?

And use the CallRpc function as a top level base where everything is entered from

[twilit radish]

I would just use an RPC, it uses sockets under the hood but it’s more in an Unreal way. It’s a bit unnecessary to create another connection while you already have one IMO.

[fair coral]

👍 Cool, was just checking that I am not walking myself into any corners by doing it this way as I usually see RPCs used for 1 off updates rather than whole gamestate updates

[twilit radish]

I mean if you’re talking from client -> server RPCs are the only way. If we’re talking server -> client you can consider the replication system Unreal has 🙂

But it just depends on what it exactly is, but there’s not any need for a custom socket system for any server / client data as Unreal has systems for all that stuff 😄

[fair coral]

Just got two clients, though in unreal I guess it is just server client with client also having authority

[chrome bay]

server has auth not clients

[fair coral]

peer2peer setup, both have authority just need to keep the inputs in sync

[chrome bay]

if you're in Unreal one of those instances is the "server" though

I assume by authority you just mean the client is sending stuff to server and server is blindly accepting it

[fair coral]

Yes, the original question was about sockets vs rpcs for implementing a custom networking architecture

[tranquil yoke]

Is there any way, we can connect multiple game sessions with each other ?
I want some Pawns to be replicated to all the sessions exactly at the same time ?

[sweet ore]

hi everyone. i have problem with simple move to location in networking. when i click anywhere player movement become laggy then the server correct the location after player stop which cause stutter at the end of player stop move

anyone know how to solve this?

[plucky prawn]

Depending on how "connected" you want them to be, you can use normal non-ue sockets. For replication across servers I have no idea. Maybe you could use a proxy pawn and relay the input to the other servers

[dry pebble]

Anyone had issues calling RestartGame with steam listen server and having it not bring clients to the map?

[tranquil yoke]

@plucky prawnReplicate one Pawn movement with all the servers.
so all the sessions will see the pawn exactly in the same state.

[dry pebble]

@jolly siren I noticed you had the same issue 4 years ago with RestartGame and non seamless travel, did you end up fixing that or using seamless instead? As using seamless messes up the UI

[vapid junco]

Abnormal Flying Guns (Based on Lyra Starter Game)

any idea on what's happening here? Only occurs on the client side in a development package

Works perfectly on PIE, DebugGame, Standalone Windows for Debug and Dev, Shipping Build. Only occurs for the client when packaged as Development Build.

Logs:

[2022.10.11-07.16.01:299][120]LogNetPackageMap: Warning: Using None instead of replicated reference to B_WeaponSpawner_C /ShooterCore/Maps/BlasterMap_1/Generated/BlasterMap_1_MainGrid_L0_X-2_Y14_DL0.BlasterMap_1:PersistentLevel.B_WeaponSpawner_C_UAID_58112229E9901F3A01_2080049866 because the level it's in has not been made visible

[2022.10.11-07.16.31:828][786]LogNet: Warning: UActorChannel::ProcessBunch: SerializeNewActor failed to find/spawn actor. Actor: B_WeaponSpawner_C /ShooterCore/Maps/BlasterMap_1/Generated/BlasterMap_1_MainGrid_L0_X-2_Y14_DL0.BlasterMap_1:PersistentLevel.B_WeaponSpawner_C_UAID_58112229E9901F3A01_2066734865, Channel: 4
[2022.10.11-07.16.31:829][786]LogNet: Warning: UActorChannel::ProcessBunch: SerializeNewActor failed to find/spawn actor. Actor: B_WeaponSpawner_C /ShooterCore/Maps/BlasterMap_1/Generated/BlasterMap_1_MainGrid_L0_X-2_Y14_DL0.BlasterMap_1:PersistentLevel.B_WeaponSpawner_C_UAID_58112229E9901F3A01_2080049866, Channel: 5
[2022.10.11-07.16.31:829][786]LogNet: Warning: UActorChannel::ProcessBunch: SerializeNewActor failed to find/spawn actor. Actor: B_WeaponSpawner_C /ShooterCore/Maps/BlasterMap_1/Generated/BlasterMap_1_MainGrid_L0_X-4_Y14_DL0.BlasterMap_1:PersistentLevel.B_WeaponSpawner_C_UAID_58112229E9901F3A01_2097999868, Channel: 21
[2022.10.11-07.16.31:830][786]LogNet: Warning: UActorChannel::ProcessBunch: SerializeNewActor failed to find/spawn actor. Actor: B_WeaponSpawner_C /ShooterCore/Maps/BlasterMap_1/Generated/BlasterMap_1_MainGrid_L0_X-4_Y14_DL0.BlasterMap_1:PersistentLevel.B_WeaponSpawner_C_UAID_58112229E9901F3A01_2098003869, Channel: 49

[regal geyser]

VOIP isn’t working in shipped builds, in PIE/Standalone/Dev build it works just fine.

DefaultGame.ini
[/Script/Engine.GameSession]
bRequiresPushToTalk=true

DefaultEngine.ini
[Voice]
bEnabled=true
[OnlineSubsystem]
bHasVoiceEnabled=true

I add custom VOIPTalker, register it with PlayerState, and then I watch if VOIPTalker “Event begin Talking” triggers or not, but it doesn’t…

Also without VOIP, if I just use ToggleSpeaking 0/1 without anything else, it still doesn’t work only in shipped builds.

[obtuse field]

Why does "AtCapacity" return 0/false, even if there are 3 players and max players is set to 2? Do I miss something about this function?

ALeapNGameSession::ALeapNGameSession()
{
    MaxPlayers = 2;
    MaxPartySize = 2;
    MaxSpectators = 2;
    
}

FString ALeapNGameSession::ApproveLogin(const FString& Options)
{
    FString Message = Super::ApproveLogin(Options);
    
    UKismetSystemLibrary::PrintString(GetWorld(), "Calling Approve Login now!");
    UKismetSystemLibrary::PrintString(GetWorld(), FString::FromInt(AtCapacity(false)));
    
    return Message;
}

[hybrid crown]

Hello, i would like to setup a database for my games.
The game is started by clients, who can connect to small server for game, and can search into database for some information.
The client at some point should also be able to retrieve some data in the database (like level of the player).

Should i make a custom server who is in charge to handle communication between Client <-> Database and Game Server <-> Database ?
Is there solution who already exist working with UE ? or i need to do it from scratch ? if doing it from scratch, what is your advise for the protocol and quickly the architecture ? Is there a paper on the subject ?
Thanks !

Thanks, if you have any paper who can help on the subject, i take it !
Is it easy to integrate asp.net core to unreal engine ? Like there is already a native API for those thing ?

[bitter oriole]

Normally you'd simply send HTTP requests from Unreal or use the websocket library

You'd then have a web server acting as front for the database

Because you don't want untrusted clients connecting to your db

[hybrid crown]

Yeah, absolutely.

So something like a token generated for the session, i'm using steamwork for the authentification.

correct ?

[bitter oriole]

Not very knowledgeable of authentication but something like that

[hybrid crown]

I see, sound like exciting thing to learn.

We expect to have a "large" (At least for me) volume of player (300 000 on the release).
For the amount, do we need a complex setup or just one server for the DB will be enough ?

I know it's really dependant of the amount of data each client need to request/send, and it should not be to huge (data are requested only in the start of the game / end of each game).

I see, thanks for the answer @whole grove and @bitter oriole !

[scenic turtle]

hey all, how would I accomplish for a multiplayer game to get the closest player to me? Sort of like a look at each other so I would have like the heads turning towards each other.

[jolly siren]

@dry pebble I switched to seamless travel a long time ago

[twilit radish]

Just loop over all your pawns/characters and check the distance and compare that 🙂

[fathom aspen]

Depends on number of players ofc, but the brute force option would be an overkill to the server. I would go for something similar but have the client do it (that way you only get relevant pawns) and check distance against them (You will still need to consult server so they don't cheat)

Or maybe have the client do a sphere trace (with a radius set beforehand) and get closest player in that sphere (again will need to consult server)

[twilit radish]

If performance is your concern you don't have to do this in tick or anything, put a 0.5 second timer or something on it and absolutely no one will notice the difference. I also don't think a distance check every now and then on at most a few tens of players will matter. If you've more than that amount of players to worry about I don't think the question would have been asked here honestly 😛

Or just 'time slice' it and query a few players per tick or per X tens of a second and then next time do it for different players.

[fathom aspen]

The options are endless, though I like to deviate from the brute force option (CS vibes)

But yeah, the relevancy system kinda does something similar every tick

So why not utilize that 😛

[twilit radish]

A few distance checks really are not that expensive.

[fathom aspen]

Or use the RG with 0.x timer 🏅

[scenic turtle]

Thanks for the response Thorn and WizardCell, I am considering aroind 10 or so players. So I would have to do like a get all actor of class and check for each element their distance from the current player.. (self)?

[twilit radish]

The thing is Wizard, this is not a crucial aspect of the game. If it turns out to eat more performance than you want to, just change it. If this was your world generator for the next Minecraft sure it would be a totally different story. But it's just not worth trying to make everything the best and most optimised you can, time is money too 😛

And like I said, worst case you just put it on a 0.5 second timer or check only a few players every time you tick it 😄

[bitter oriole]

Taps "first rule of optimization" sign

[fathom aspen]

Time is 🎵 , I need to figure out how I work on tasks async

[twilit radish]

To answer your question btw. Basically yes. Just get whatever class you use, do some distance checks and store the lowest distance and then use that one 🙂

Also btw. How on earth do people deal with moving AI in games? Imagine the following scenario: I have my most generic twin stick spaceship shooter where you have a cool laser beam to hit moving enemies flying towards you. How do you deal with the scenarios where the person shooting hits the laserbeam on their end on an enemy but on the server it already moved past that point? In this case I'm not worried about the actual hit being valid or not, lets say we go with client authority here. But how do you match up that laser beam going past the enemy visually?

vs like this happening for other clients for example.

I can think of ways to hide the fact that it may look off, but is that really the only solution to this?

But then again, I suppose that's the entire thing with multiplayer.. Trying to hide the fact that worlds don't line up with each other xD

[fathom aspen]

I wonder how is this different from non-AI (PlayerControlled agents)?

[twilit radish]

It's not different. If I were to shoot at a different player it could have the same issue yeah.

Let me rephrase my question to "moving objects" then I guess 😛

[fathom aspen]

I see. I mean rewinding solves hit registration

But visually it doesn't add up anything I guess

Unless we rewind them visually

[twilit radish]

Like I said I'm not worried about hit registration, it's a small fun side project to just keep going with multiplayer stuff. It won't be released nor pvp. So client will say "I hit this thing" to make it feel good.

I guess I'll just go the route of trying to act like it hit if the clients says so. Except with players as I think that may be too much considering that players predict them selves.

Then it needs to go from Client1 -> Server -> Client2 while Client2 may have already moved way too much by then to apply tricks 😛

Considering there's no pvp/friendly fire it doesn't matter if it looks off I suppose as it won't damage the ship.

[fathom aspen]

I don't think it will be much noticed that the object is off (maybe it is in a space shooter type of game)

[twilit radish]

My concern with it is that with a bit of ping the laser goes beyond the enemy while still damaging the ship, that will look odd I feel like.

I think I'm just going to sneakily rotate the ship on the server when it's not the actual server to try and make it look like it actually hit the AI 😄

Although visually, not actually.

[fathom aspen]

Fair enough, would love to see the results 😄

[twilit radish]

👍

I can post a video for sure 🙂

[unborn nimbus]

If I want to "toggle" relevancy for an actor instance through RepGraph, how would I do that?

[chrome bay]

would need to add + remove it to an "always relevant" node

May also need to remove it from the spatial grid nodes because IIRC the distance relevancy trumps everything else anyway

[unborn nimbus]

so if it doesn't exist in a node, it's not relevant?

there should be nothing else calculating relevancy?

[chrome bay]

If it's not in any nodes it shouldn't replicate to anybody

[unborn nimbus]

makes sense, thanks!

[dry pebble]

Ahh I see! How did you handle all the HUD/UI widgets messing up after that? I don't fully understand why certain HUD elements show and others don't with seamless travel

[fathom aspen]

AFAIK all UMG widgets truly persist (with their data) a seamless travel

What do you mean by others/some don't?

[chrome bay]

Game HUD should really all be killed before travel

[fathom aspen]

Unless you have some chat widget (or something else I can't think of) you want to persist

[scenic turtle]

Yes Thanks Thom,
and do I have to do this logic from the player pawn BP, right? I am only asking because i did try to implement it earlier, but for example i wanted to get the shortest distance between 3 characters and the shortest distance always returned to self I think, I mean is there a way to ignore self and just check the distance from the other two characters.. sorry if I am dragging this out. I am following along based on this, https://www.youtube.com/watch?v=aKzjNTadrAc&t=4s but in the end I am not calling this from level blueprint, but from the PlayerPawnBP, which might be the source of my problem. Thanks!

[fathom aspen]

It is as easy as checking != Self

And no you're on right track, you don't need the level BP

It will cause you more headaches than present solutions

[scenic turtle]

Something like this WizardCell?
So you are saying that this approach is not ideal in this case? Or that running it from Level BP would be the inappropriate way. Thanks!

[fathom aspen]

Yes like you did in the picture

I'm saying keep things in the Pawn BP (i.e. no further steps needed)

No need to move things to level BP

[kindred widget]

Even that should be honestly be recreated. If I cared enough to want chat to persist loads, I'd store it somewhere in a file or outside of world in a game instance subsystem and reinit the widget with that data. I really personally dislike going against the general golden rule that UI is for displaying state and accepting input. Causing it to carry that state past loads turns it into the state holder and not the displayer.

[fathom aspen]

I somehow agree, I just tend for less effort and use what the engine presents. I was amazed to discover they persist out of the box, even if the persistent HUD gets destroyed before a new one gets created. I will still have to figure out how they persist though. I doubt that is a feature, but seems more like a glitch/hack hehe

It was talked about long ago on UDN and was published to the public without them figuring out how they persist: https://forums.unrealengine.com/t/cleaning-up-widgets-on-travel/320229/4 😄

I would guess they are deep linked into the GameViewportClient, though that doesn't explain how they are destroyed on hard travels

[kindred widget]

UWidgets aren't part of the actor system. They're not tied to world. They're just basic UObjects similar to GameInstance. The underlying slate they hold just gets redone when you add it back to another parent.

They're not exactly explicitly destroyed on hard travel so much that they are part of the viewport. You can store a Widget in GameInstance and put it back on screen after a hard travel for instance without recreating the UWidget. Seamless just doesn't remove the viewport.

The reason you don't have to clean them up from hard travel usually is because everything that holds a pointer to them is cleared. Without being part of the viewport hiearchy and no active actor holding a reference to them, they'll get collected.

[fathom aspen]

I've seen this a few times. When you say they are not tied to a world I guess you mean their outer isn't world like actors?

[fathom aspen]

I see, so hard travel does remove viewport.

[kindred widget]

Maybe. Actually let me look at that. Because it doesn't entirely clear everything. You wouldn't be able to do seamless movieplayer loading screens if it did.

[fathom aspen]

Last I checked, there are a few function versions to creating widgets. IIRC, by default their outer is GameInstance, and you can explicitly make it HUD or PlayerController.

[kindred widget]

There are a few. I don't think anything ends up as the outer besides World, GameInstance or other widgets.

The controller context is for coop. That's actually irrelevant to the outer, that's for setting the playercontext for which coop player the widget relates to.

[fathom aspen]

Makes sense. I will have to dedicate more time for this, though your insights have cleared things up for me, so thanks!

[feral falcon]

when working with steam online subsystem do you also want to be using the steam sockets plugin?

[pallid mesa]

you don't have to

[feral falcon]

okay I have been having serious issues with establishing a connection so I thought maybe that was my problem

[kindred widget]

@fathom aspenDid a basic test. Found FWorldDelegates::LevelRemovedFromWorld. This gets broadcast from UEngine::LoadMap.

When UUserWidget has AddToViewport called on it, it internally calls AddToScreen. This binds that previous delegate from LoadMap. That calls OnLevelRemovedFromWorld, which will make the UUserwidget call RemoveFromParent on itself.

I would wager a guess that Seamless does not call this since it does not actually tear down the persistent world.

That also explains why MoviePlayer isn't affected despite being able to use UUserWidgets. You never actually call AddToViewport on it that way, you actually have it construct it's underlying slate and pass that to the moveplayer.

AddToScreen is virtual though. It would be easy to bind to the PreSeamlessTravel handle and ditch them from there.

[fathom aspen]

Exactly. That was the function I traced last time OnLevelRemovedFromWorld and my guess was that it isn't called on seamless but iirc I was amazed that it is. Thanks for the breakdown though, now you made me really want to look into it again!

I'm not aware of the MoviePlayer part, what's that?

I guess its a UI related object

[kindred widget]

Movieplayer plays your game's default intro movies from these settings here. But it can also be used as an async loading screen display for hard level loading.

Read "Animated loading screens for hard world travel".

[fathom aspen]

Great, that's something cool to read on

[kindred widget]

You can also call it at runtime to block the gamethread for cinematics. Less useful in multiplayer though.

[fathom aspen]

So that's basically what games use to present their startup movies. Cool 😄

[shadow aurora]

Anyone have suggestions on the best way to sync doors in a networked multiplayer system? Most videos and stuff that I've seen don't really feel optimized for a good experience. They just sort of play a timeline whenever the event comes through on the client and server.

[dark edge]

Are doors open/closed or can they be in between?

either way, repnotify on state

[graceful flame]

Doors are one of the trickiest things to make. They might seem simple at first but can be an awful lot of work depending on how the door's complexity. For instance does the player click to open? Do they press a button? Should the character reach for a door knob? Can the character break the door down? Can the door be locked? Can the door be knocked on? Is there a door bell? Does the door just slide open based on proximity or with a key like in doom? https://www.youtube.com/watch?v=AYEWsLdLmcc

But if all you're trying to do is make sure an open door is open and a closed door is closed for all networked clients just use a boolean with repnotify like Adriel said.

I'm using sliding doors for my game.

[mystic sierra]

What's UE4 source code version size?

[pallid mesa]

2

In all seriousness, several hundreds of thousands of lines of code

[fathom aspen]

iirc it was millions last I checked

[pallid mesa]

that's what several hundreds of thousands LOCs mean

🤣

[fathom aspen]

I got outplayed

[cursive isle]

what do i have to do to port forward on ue4?

[glossy kettle]

NAT

[vagrant falcon]

i have weird issue, "simple move to location" works laggy when client moves on his instance 😭

It may not be so obvious on the video, but in the editor it is very noticeable

[rapid bronze]

You want it predicted

[vagrant falcon]

i though it would be automatically done via movement component on character

[rapid bronze]

It doesn't seem to be going through CMC

[vagrant falcon]

how can i do it with less pain

[rapid bronze]

Easiest way to test is running the same code locally on the client too

So like Run on Owning Client

After the Server one

[vagrant falcon]

running on owing client after server run doesnt help

[rapid bronze]

Then maybe try it fully locally so before the server

Forgot which was the way, lol

You just care for the pawn to have its own local sim and if it matches the server, it won't be corrected, giving you the smooth feeling

Unless the problem is somewhere else

[vagrant falcon]

i had to enable client Side Navigation in project settings, now it works

lol

but this makes a new security issue 😭

[rapid bronze]

Neato

Nah, not really

It you run it on the server, it'll get corrected

I think that only allows the sim to run on the client too for predictions like this

[vagrant falcon]

okay, thanks

[cursive isle]

what port does ue4 use

[vagrant falcon]

7777 ?

or you mean editor? for editor no idea

[fossil veldt]

Not sure when they changed it, I guess around 4.20 when the Fortnite stuff got merged in but I just noticed CMC doesn't do ServerMove() RPC anymore, they infact seemingly deprecated every single RPC relating to movement

so now i'm questioning, how does any of this movement code get called into on the server in the first place

I don't see where the client sends any of inputs to the server at all

Oh shit absolute madlads are manually sending the packed data in ServerSendMoveResponse 👀

ah nevermind, they moved it onto the character :p

But still crazy they manually pack the data that way

[vivid seal]

yeah all the CMC stuff is tied really tightly to the Character class, all RPCs are done through the Character class and then passed back to the CMC on the server

[dark edge]

Why would they do it like that?

[vivid seal]

there's a comment somewhere that implies there's some overhead to RPCing through a component directly

[dark edge]

Pedigree or is it actually a real design decision

Seems a bit weird to even have the CMC be a component then, except maybe for compatibility with other movement components

[vivid seal]

afaik its strictly an optimization

yeah i hate that about it, im sure there was a good reason they did that but i'd love the CMC to be a bit more "drag and drop" and usable with normal actors

[fathom aspen]

Seems like that was not low for a game like Fortnite

That is replication overhead, not to mention the RPC overhead

[fossil veldt]

yeah like the others said it's so you don't have to rep the component

which is costly

[hollow eagle]

given that CMC was already super tied to ACharacter it's not a huge deal. Sucks for more reusable components though.

[fathom aspen]

Sad day for the GMC (https://www.unrealengine.com/marketplace/en-US/product/general-movement-component). I would believe they have the component replicated with RPCs in there

In another context, Riot's dev blog posts are amazing: https://www.leagueoflegends.com/en-us/news/dev/dev-null-anti-cheat-kernel-driver/
From the article : In a partnered study lasting approximately 8 years and backed by $20 million in federal funding, leading scientists managed to chronologically place the invention of cheating somewhere between 3.5 billion BCE and November 20th, 1985. While its precise origin remains indeterminate, one reality has become accepted as established fact: Cheaters gonna cheat.

[open quail]

oh yeah they are

https://technology.riotgames.com/news/valorants-128-tick-servers