multiplayer

[fossil spoke]

This would be a time to override that

So you dont accidentally destroy that Pawn

Overriding that functionality is not unsafe

Its normal to do that

For certain circumstances

Like the one I mentioned above

Yes

[halcyon ore]

Seems so.

Actually, I've miss-lead you, and myself.
I don't actually see anywhere the DetachFromControllerPendingDestroy actually destroys the pawn?

That says "knowing it will be destroyed soon"
But, itself doesn't actually destroy.

Hence the confusion of your pointing out said function when debugging with a breakpoint.
Gotta find the exact line that calls destroy.

[fossil spoke]

You want APlayerController::PawnLeavingGame()

[halcyon ore]

Yeah.
I pointed that out.
So, its making me think your not actually using your C++ class, and still using base UE

[fossil spoke]

Which is called when the PC is destroyed

void ABLPlayerController::PawnLeavingGame()
{
    // Not calling super here is intended - we leave it up to each pawn to decide how it behaves when the controller is destroyed
    // Super::PawnLeavingGame();
}

This is ours

[halcyon ore]

Yeah.
Double check your controller parent.

Still has me sus, none the less. 😛

[nocturne quail]

not sure why in a standalone game the timer never stops...

FIRST TIME ENCOUNTER THIS ISSUE WITH TIMERS

void AWeapon::FireRifle_Internal() // called by server
{
    GetWorld()->GetTimerManager().ClearTimer(WeaponFiringTimer);
    GetWorld()->GetTimerManager().SetTimer(WeaponFiringTimer, this, &AWeapon::TryAutoFire_Internal, FiringInterval, true);
}

void AWeapon::TryAutoFire_Internal()
{
    if (bStoppedFiring)
        return;

    UE_LOG(LogTemp, Error, TEXT("=== TryAutoFire_Internal ==="));

    CurrentAmmoInClip -= 1;
    SpawnBullet();
    BulletCounter += 1;

    if (CurrentAmmoInClip == 1)
    {
        GetWorld()->GetTimerManager().ClearTimer(WeaponFiringTimer);
        BulletCounter = 0;
        bStoppedFiring = true;
    }
}

but if I play as client, it works fine

[nocturne quail]

solved, was related to bots, they were starting the timer but not exiting it 😄

[verbal dust]

When I spawn an actor it always spawns at an offset from where it is on server for the client

no matter if the rpc is called from client or sever

[dark edge]

Show code, and are you moving when doing this?

[verbal dust]

Ah you helped me fix it in the blueprint thread.

[crisp shard]

default values set in classes are "replicated" as in their intial set values are available to the clients, right?

[halcyon ore]

Not by network/ replication means, but yes.
The client knows of them.

[crisp shard]

right yea , that's what i mean, nice

[frosty harbor]

Anyone knows if there's a way to prevent CMC from affecting local player's camera? When doing crouching / uncrouching it's performing it instantly regardless of ping on the players camera, however, when the server actually performs the crouch / uncrouching the camera does another adjustment (even though I'm already back to where the camera was meant to be in anyways).

This is especially noticeable if you crouch / uncrouch instantly, because locally it happens right away but the "crouch" hasn't even arrived to the server yet even though you've already done crouch + uncrouch locally, so there's a big correction visually. If the player actually stays crouched for say 0.5s then there's no corrections at all but if it's an instant crouch / uncrouch then it triggers a big correction

Separating it to run crouch/uncrouch on the owning player only doens't even work either because this always sends the event to the server rpc regardless of happening locally or not

[sage lance]

check out how its done in lyra (my crouching works in multiplayer)

[frosty harbor]

its different, lyra is a 3rd person,im doing a first person game, my camera is a lot more sensitive to tiny changes. Mine also works in multiplayer its just that I want to finetune it not get the issue I mentioned above

[halcyon ore]

What am I missing for replicated sub objects? 😛
Owner Actor:
Replicates and Replicate Using Registered Sub Object List
Replicated Object:
(Image)

I see stuff talking about a ReplicateSubObjects function, but its comment says its un-used when using the Replicate Using Registered Sub Object List bool.

I spawn the actor, construct the object, AddReplicatedSubObject, and the server will print its on rep function call, so the object is created/ valid, and has the correct outer.
and, if I disable Replicate Using Registered Sub Object List, it throws the ensure about it being disabled, so what am I missing?

I also tried adding the replicated object to a replicated TArray, thinking maybe the var is needed to actually replicate, and the adding to the sub object list, just allows it.

[halcyon ore]

NVM, figured it out.
I messed up a bool check for my testing function.
So, it added to the sub object array, then removed it, instantly. 😛

[plain latch]

Has anyone here used AInfo for their inventory? I'm refractoring my inventory system and I've almost got it working, but I'm stuck on some client side stuff. I use an inventory actor that I spawn with an Inventorymanager component on the player controller, and to pass it around I use an interface. The system actually works for all net modes, I can add and remove fast array items, but the inventory pointer returns null when I go to fetch it for UI. Whats a good way to spawn AInfo classes?

If it helps, I spawn my info class the exact same way the Lyra Team Creation component does. (Waits for the experience, checks for authority, then spawns it inside a WITH_SEVRER_CODE block)

[thin stratus]

Spawning it on the Server if it's replicated should be all you have to do.
If it's invalid in the UI, then you might just have a race condition between it replicating and the UI asking for it.

[plain latch]

Yeah let me check. I was worried it had something to do with the player controller only being relevant on the server and owning client

[thin stratus]

That shouldn't matter, unless you pass the PC as the Owner and set the AInfo to share relevancy with it.

[plain latch]

Oh wait a minute I didn't actually pass anything in as the owner in the spawn params. That might have something to do with it

My idea for this is that the InventoryManager can be placed on any actor and it will just handle the creating it, then I can just pass it around like I would an ability system with an interface

So I might have to do some extra stuff for a player

[thin stratus]

Shouldn't matter. Not passing an Owner just affects Ownership an thus RPCs and Replication of properties.

The Actor should still replicate. You should just print/breakpoint the UI functions and your Actors BeginPlay to ensure they call in an expected order.

[plain latch]

I'll do some tests. I tried to nab it directly from the character for a print string where it also returned invalid, but the item granting still worked. I might've just done something wrong on the blueprint side there. Gonna look through my UI and that

Okay its looking more like a race condition. Might have to do something like an inventory ready gameplay message or something

The inventory manager itself is valid but the inventory actor isn't

I delayed the call a bunch and it still didn't show up. Silly question but is there some other REPNOTIFY macro I need to call or something in the OnRep function? Gonna mess around with DOREPLIFETIME too

[plain latch]

OnRep_Inventory is never getting called. I think I'm never setting the pointer on the client. Here's how I'm currently creating it

#if WITH_SERVER_CODE
void UTestInventoryManagerComponent::ServerCreateInventory()
{
      UWorld* World = GetWorld();
    check(World);

    FActorSpawnParameters SpawnInfo;
    SpawnInfo.SpawnCollisionHandlingOverride = ESpawnActorCollisionHandlingMethod::AlwaysSpawn;
    //SpawnInfo.ObjectFlags |= RF_Transient; // We never want inventories to save into a map.

    ATestInventory* NewInventory = World->SpawnActor<ATestInventory>(InventoryInfoClass, SpawnInfo);
    checkf(NewInventory != nullptr, TEXT("Failed to create inventory from class [%s]"), *GetNameSafe(*InventoryInfoClass));

    // This may be the problem! OnRep_Inventory is never getting called. I don't think this pointer is getting set on clients!
    SetInventory(NewInventory);
}
#endif //WITH_SERVER_CODE

[junior timber]

How can I implement microtransaction with steam and unreal engine 5.7
I am using advancedsession plugin for multiplayer. Is there any blueprint method to do it?
What should I look into?

[halcyon ore]

@haughty ingot Just to keep it in the correct section.

I looked it up.
How would this go with FastArray, thats is delta map based?

Does it like "work" with it?
Cuz, If I need to break it into bunches, but what if the values change while the bunches are being sent?
Data error/ miss-match?

[haughty ingot]

Well it's just to get the initial data over, you can still mark individual parts of the fast array dirty after and they will replicate correctly. You're just going around the normal property replication method

For the initial bunch

[halcyon ore]

Right, but given it happens in chunks (over time)
What if the data is changed, while being sent?
Won't the deltas be out of sync?

[haughty ingot]

It's kinda up to you to verify that, you could use the Dynamic replication condition to make sure that changes don't come through until after. There's a lot of edge-cases, but it's just a way to get a big bunch over

[halcyon ore]

I see, ok.
and, just to be sure (since it sounds like you've done it)
Google says to send these bunches through normal RPCs, or is there a more intended way, or reliable source of info?

[haughty ingot]

You could send a packet if you really wanted to, but an RPC is the easiest way.

[halcyon ore]

Well, thanks for that.
I didn't realize you could custom batch variable replication, and keep the actual var working like normal.
(I of course knew of batching, but I always thought it would make normal var replication have issues.

Still gotta do like 30% cursed stuff.
Cuz, of course var replication condition is everyone, or no one.
So, I need a middle man actor, for targeted player replication. 😛

[haughty ingot]

Hmmm that's not quite accurate, you can set any number of replication conditions. Including dynamic ones that you can change at runtime

[halcyon ore]

Correct, with COND_Custom.
Which has no context of person, and the UE page for it doesn't speak of anything for per controller/ person replication.

So, while.
Yes, dynamic.
It becomes a send var to everyone or send var to no one

One thing I did think of was Using the sub-objects netgroup replication type.
Doing some cursed ass like internal batch indexing.

But, I wasn't sure which method was more cursed, or not. 😛

[nova wasp]

https://dev.epicgames.com/documentation/en-us/unreal-engine/replicating-uobjects-in-unreal-engine#createanduseareplicationgroup

[halcyon ore]

Thats for sub-objects.
Not for just variables.

The sub-object method is more "cursed"
Compared to the more "correct" batching a Fast Array struct on its initial replication like Elliot suggests

[nova wasp]

for just individual variables you are going to need to make engine changes afaik

[halcyon ore]

I was just gonna attach an actor, and use the is relevant for function on it.

I can barely understand the network code, for UE, so ain't no way i'm gonna make engine changes for that.

[nova wasp]

also you can just replicate stuff down to individuals without any fancy anything by just using the player controller or actor components on it

[halcyon ore]

Won't the nature of Fast Array delta stuff, throw replication at every player either-way?

(when changes occur to the Fast Array I mean)

[nova wasp]

a fast array is an individual property

afaik there is not going to be a way to have nested replication conditions unless the things replicated by the fast array are object references that choose to not show up and even then I would very much say you probably want to have 2 fast arrays instead there lol

[halcyon ore]

No, its not a nested thing.

[nova wasp]

inside of what?

[halcyon ore]

Not sure I follow.
The CURRENT setup is a Fast Array in an inventory component.

My proposed working setup is a comp as a sort of "manager", but an attached actor has a Fast Array var on it, so I can use the is net relevant for function.

[nova wasp]

relevancy seems a bit overkill as it will have to constantly check them afaik, no?

but I don't know if net groups work well in old replication either

[halcyon ore]

But, net groups are replicated sub-object only.
Which leads to the same "issue" I will have.
Too much replication data in 1 network bunch, because UE doesn't batch sub-objects, like it does individual actors.

[nova wasp]

define too much replication data

is it concerning to tell users about data they should never know about here? that could make sense if the goal is to hide things

[halcyon ore]

This is for an inventory system, but with far more data storage, then I think most games do.
So, overall quality, individual stat quality (6 stats), quantity, some bools, string, and probably even more.
Those are just the basics.
Of course, not every item will use all of these, but it will very quickly hit the 65k limit.

[nova wasp]

have you ever bandwidth profiled this?

also why is a string even being sent...

[halcyon ore]

I profiled a couple.
Never enough to cause any issues.
But, I know the bandwith of a "basic" item, and a "complex" item.

A string is sent, for a custom name, if so desired by me, or the quality system. (Like I said, very data filled)

[nova wasp]

"if so desired by me?" is the string entirely arbitrary at runtime or is it something tied to a specific asset or object

[halcyon ore]

Arbitrary to whatever I want at runtime.

[nova wasp]

the important thing to think about here is INFORMATION I guess

for example if the string it uses can be derived from a simple ID you don't need to send a giant buffer of characters

In Iris you can kinda abuse this by using exported names I guess but not in regular replication

[halcyon ore]

No.
Its entirely arbitrary to whatever I may need it for.
I could rename wood to what the heck, for all I care.
Granted, I would never go that goofy.
But, it is just a string that accepts any value thrown at it.

[nova wasp]

please consider what I wrote there by "derived"

if you are the one who defines what the string is it's just in the game

so arguably you could make a dataasset or something per type of item, or even a custom asset etc

[halcyon ore]

I'd need to think on it, now that you give me that context.
I guess its just my want of being "direct"
I'd hate to be goofy as hell, and be like this says id 5, and go to another asset to find out wtf id 5 is.
When I can just type out cool wood, and know.

But, thats valid for the string usage.

[nova wasp]

but if it truly is just arbitrary it might be easier to just slam a string in there... I will say I have not really measured shorter strings bandwidth in detail mostly because I have not ever had to send one though so it might be fine

[nova wasp]

you can add a variable to a data asset that is a display name

[halcyon ore]

Are you suggesting I make individual data assets, for each "custom name".
I guess that would work, just seems kinda goofy.
But, it would solve the whole I don't wanna have to look up id 5

[nova wasp]

it depends on how many unique things there are

arguably if there are like 5 unique items it could just be an enum lol

[halcyon ore]

Oh, defenitily not 5. 😛

[nova wasp]

making a way to represent unchanging information about the item as a smaller handle (an asset object here) is something I find difficult to justify not doing in a serious project

arguably it could be the display name does change based on the instance info though, but even then passing down which type it is by asset would let you ask the class what name it should have based on current instance-specific data and not need to waste bandwidth sending basically a chat message for an inventory

[halcyon ore]

Perhaps you miss-understood.
The items do have a base name, such as wood, that is defined in the class, and not replicated over the network.
Unless, you do understand it as that.
Its just that there can also be a custom name that overrides the base name, that I want.

[nova wasp]

so the class is already sent?

is the custom name something that requires localization? how often does it change?

if it's rarely changed and just for a prototype thing it's probably fine, I am going to test out real quick how expensive small strings are to send

it might be that these aren't large enough to be concerning

[halcyon ore]

In theory it would be localized, so that its localized.
but, I haven't thought about localization that much.

Its one of those wants with my game.
While, yes I could define 15 different types of wood.
It seems more logical to make 1 wood, and custom override its name to however it was retrieved.

Rather then handle annoyances of item class conversion, or stacking different classes, and handle that.

[nova wasp]

even gameplay tags would be better here honestly...

you can't localize a raw string afaik

unless you have a local string table to hash over each time a new item shows up (oof)

to the FText

I have no idea what item class conversion has to do with this

[halcyon ore]

Yeah, I do current have a Gameplay tag thing setup.
For the quality system, and such.

Also, BTW
This is my current item network info setup:
A couple more are planned, but I haven't quite locked in what I need/ do want.
Still been working on the basics of systems, and getting them working.
Been going good, just reviewing some stuff, for better/ more correct ways to do it.

Class - item class
int64 - item ID
Fstring - custom name
CURRENTLY fstring array, but now planned to be GameplayTag array - dynamic keys
int32 - dynamic values

[nova wasp]

even smaller strings here nearly take up the entire packet bunch

if these are sparodic it won't matter I guess

[halcyon ore]

Yeah, thats about what I got.
Adding a basic 15 character string, brought my items from 58.2 bytes -> 73.2 bytes

[dark edge]

Been watching some stuff on Iris, so it doesn't rely on actors and actor channels?

[nova wasp]

no, it does not use actor channels

iris internals barely even know what a uobject is

[dark edge]

Does this make replicating internal state of subsystems a bit easier?

feels like it might

[nova wasp]

I am not sure... from the user gameplay code side it's 95% the same

its 1 cvar to turn on iris and try it I guess

[boreal bison]

what are the best optimizations to implement for spawning a lot of AI enemies (~100-200 at absolute max) for a multiplayer game in the style of something like Megabonk? the AI would only need to slowly move towards the players at first, maybe play a montage to perform a melee attack. I'm reading a lot about the CMC being fairly heavy, tick frequency, and stuff like that, and i know there are some related assets on FAB but I'm not sure which ones are worth getting, if any, and in general what optimizations allow a game like Marvel Rivals to run a multiplayer PvE mode that spawns hundreds of zombies at a time

[chrome bay]

They've only just added UObject support recently, AFAIK that's ue5-main stuff and above

Still early days in that regard

[nova wasp]

do you mean custom net factory stuff?

[chrome bay]

There's just a lot of recent, regular commits mentioning this stuff - so heed caution I guess

[torn hull]

Hey, can nested child components replicate in Blueprints? I’m having an issue where my sphere interaction collision is set to replicate, the actor itself is replicated, and the DefaultSceneRoot is replicated too. But when I disable the collision on the server, it doesn’t work on the clients. The collision is still active on their side. It only works if I do this using a rep notify, but I thought that if component replicates, I wouldn’t have to do that.

[kindred widget]

Collision settings aren't replicated I don't believe.

To answer your question directly, yes you can replicate that component. It'll work just fine.

But I don't believe that even if that component replicates correctly that collisions will. Usually this is handled at the Actor level based on a state that determines why that component does or doesn't need collision.

See? The code you're calling affects the BodyInstance.

void UPrimitiveComponent::SetCollisionEnabled(ECollisionEnabled::Type NewType)
{
    ECollisionEnabled::Type CurrentType = BodyInstance.GetCollisionEnabled();

    if (CurrentType != NewType)
    {
        BodyInstance.SetCollisionEnabled(NewType);

        EnsurePhysicsStateCreated();
        OnComponentCollisionSettingsChanged();

        if(IsRegistered() && BodyInstance.bSimulatePhysics && !IsWelded())
        {
            
            BodyInstance.ApplyWeldOnChildren();
        }

    }
}```

Body instance is not replicated.
```cpp
UPROPERTY(EditAnywhere, BlueprintReadOnly, Category=Collision, meta=(ShowOnlyInnerProperties, SkipUCSModifiedProperties))
FBodyInstance BodyInstance;```

[torn hull]

Ou I see, thanks for deeper answer. Ive tried also deactivating and destroying, but it didn’t work as well so I thought component wasnt replicated properly. With destroying component it was even stranger because it freezer in place on the client instead of deleting

[compact flame]

Im having a bit of an issue.. Im trying to make a kind of aiming system, but when I do the regular RPC setup it jitters on the client so Im trying to get it to do it on both client and server separately so that client looks smooth (It will use the servers rotation after so its just so it looks smooth on client)

Im prob doing it way wrong lol

Nvm figured it out

https://tenor.com/view/robert-downey-jr-damn-im-good-good-smile-look-gif-5074917

[bright summit]

what was the solution?

[compact flame]

Switch has authority didnt work so I makeshifted one

lol

[crisp shard]

i created a bullet pooling subsystem for guns and im trying to figure out the best method for spawning visuals (or really where/how to handle the hits). i currently do have damage being applied inside the subsystem, which would be fine if that's all i needed, but i'd like to handle hit VFX as well that are not related to dmg and more just visual of the bullet (like a bullet trail and an impact vfx regardless of what it hits). i can't do a multicast in the subsystem as far as i can tell,so im trying to figure out where/how would be the best method of handling the impacts outside of the subsystem

how about... a component on the game state????

[crisp shard]

@kindred widget my b, and no rush on this ping, figured since you suggested it, you may have an answer here.

i have made the cpp subsystem, and am not sure how to handle the hit's / visuals (hit and trail, so both). i have the whole thing pretty much working but just not sure how to handle the impact of the bullets from a visual POV, as you can't make a multicast from the subsystem, so curious of what would be the standard practice to handle the visuals / hit VFX for simulated clients?

[snow matrix]

Hello guys, I hope you are good, I am making a prototype of a cooperative game and I am encountering this problem. Does anyone know why this is happening?

Basically, when I rotate the character in the server window, the window vibrates, while when I rotate the character in the client window, the character rotates without problems. I think it's some kind of motion prediction and correction movement issue, but I don't understand it since the server itself is doing the rotation, so there shouldn't be these kinds of problems

Here is the code

I've been looking for solutions and even tried writing the code in the player controller, but the problem persists. So I was wondering if anyone has any idea what it could be. Thanks in advance for your help.

Btw in the OnRep function, I'm rotating the capsule component like this

[vapid fog]

Anyone using a lyra base, or otherwise, and upgrade to 5.6, have an issue where a host will ServerTravel to a new map, and clients will attempt to connect during travel, get refused, and then reload the default map?

I know what changelist is causing the issue, but I want to understand the root issue more clearly.

[nova wasp]

which changelist is causing the issue? (well, commit name would be what I need here given I'm not using p4 for reading main)

https://github.com/EpicGames/UnrealEngine/tree/ue5-main/Samples/Games/Lyra I am honestly impressed they still keep Lyra so up to date

I might have to check to see if there's anything in here I should update to

[nova wasp]

it kinda depends to me on how often they receive the info, how many there are, and how accurate it needs to be

also if you need relevancy for example to not send to a user when something happens reaaally far away

[vapid fog]

https://github.com/EpicGames/UnrealEngine/commit/9ffc263791a8e37c66145d0f652af5332821da9a

To be clear: I have not tried reverting the change myself to see if that's the issue for me, but the symptoms are the same.

And it happened when we updated from 5.5 -> 5.7. So similar engine update.

[crisp shard]

yea good questions, and good point about the relevancy, as i didn't think that far yet, but they are bullets from guns, so hard to really judge, but considering it's one of the more common weapon types, i'd say semi frequent? but could be very frequent, so potentially many shots at once, could be sniper where there is a cooldown, but could also be a machine gun where there are many bullets at a time.

currently i just made two* delegates in the subsystem and was thinking to bind to "on fired" and "on hit" (the two delegates) in the gamestate for the visuals (and potentially damage as well, but i probablay will handle that straight in the subsystem still, as the hit impact VFX are then different and i can do that stuff on the character that got hit) the bullet trace VFX would be fired every shot ideally (unreliably). could proabbly do this many diffetn wways but still new to cpp and super new to a custom subsystem so not sure of what might be best. the point about relevancy might be the most critical for just reducing any unnecessary events for irrelevant players. but doing on game state has me a bit wary as i wouldn't typically do something like this there, but it did seem like a suitable spot due to it always being available, and i wanted to avoid putting it on the charcter or anything like that. if the bullets were actors (which they aren't they are just data structs (new method i learned yesterday in here)), then i would have just done it in the bullet actors, but because they aren't im still thinkign through best places to handle the addtioinal logic

[vivid siren]

Does anyone know of a good place to see some ballpark "good" values we should be aiming for when it comes to network profiling stats?

When I see stuff like the average size in bytes being 200-300 for my pawns and 0-10 for generic items, i have no idea if that is good or bad. And when i see stuff like some actor having a Rep HZ of 60-70, I have no idea how to interpret that.

[vapid fog]

What seems to be happening is that the server initiates travel (this is ServerTravel, NOT SEAMLESS) which disconnects the clients and tells them "Hey, we are traveling to a new map!" and then BEFORE THE SERVER CAN FINISH TRAVELING TO A NEW MAP the clients attempt to connect, the server refuses them because its still on the old map, and so the client says "Oh well, heading back to main menu."

I have no idea why this worked before the changes in 5.6. But something clearly changed. Error handling? Maybe the depreciated socket address resolution stuff? It's unclear to me.

[nova wasp]

can you try to catch when the client first sends the "let me travel bro" request? or does the server not queue it up for later. This is a part of networking I am mostly clueless about so apologies if I gave the impression I have a strong idea of what is going on here... I was hoping it was just a clear difference in what Lyra is doing on their side but this is more fundamental to the connection it seems

[nova wasp]

honestly this is a great question I have wondered a lot myself... (The "what exactly is should I be aiming for bandwidth-per-second" part of it)
You could just look at what typical commercial games like what you want to make are sending in terms of up/down rate. Many games have network stats (even non-unreal games)

I would say in general the net update hz per-actor is something you can quickly tweak and observe how much it reduces quality just in the editor

generally speaking the idea is to send things that are important first at the best speed and reduce or remove things to send that are farther away. The player might not need to care as much about an actor updating 2000 miles away from where they are

[vivid siren]

Yeah i'm currently in the process of setting up a custom system to dynamically adjust things like the frequency of net updates, priority, and whatnot, but then i realized i don't even know what my target benchmarks should be lol

would be great if epic provided some example .nprof files for us to ponder

[nova wasp]

you don't need to do that

use repgraph (outdated and deprecated) or newer iris filters/prioritizers if you want something built in

be warned iris stuff is newer and more experimental but it is kinda purpose-built for doing stuff like that

@vapid fog do you change ServerTravelPause by any chance?

[/Script/Engine.Player]
; These numbers should match TotalNetBandwidth
ConfiguredInternetSpeed=200000000
ConfiguredLanSpeed=200000000

[/Script/OnlineSubsystemUtils.IpNetDriver]
MaxInternetClientRate=100000000 ; note that this is absurdly high lol, I don't think you want to do this in a real project but I think the defaults are too low
MaxClientRate=200000000
InitialConnectTimeout=10.0
ServerTravelPause=2.0

also note here I increase the default bandwidth rates a lot lol

[vivid siren]

i'd have to stick to repgraph, our project currently has a lot of experimental tech and we're trying to reduce the number of unknowns

[vivid siren]

what's the benefit of this change?

(i know you're talking to someone else about that, i'm just curious)

[nova wasp]

honestly this ini kinda applies to both of you lol

I don't know why the pause is reduce from 4 to 2, I am looking through random ini diffs to find out why

[vivid siren]

server travel has been my enemy recently, because i can get our project in a stable state (4 player co-op, physics based pawns), but things always seem to break down after a server travel

[vapid fog]

ha ha ha, you really wanted to make sure you didn't hit max bandwidth.

[sage lance]

ahh legendary for this i was wondering why it was so delayed

NOW THATS QUICK

xD

[vapid fog]

We don't have a ServerTravelPause.

[nova wasp]

It can change per net driver as well

[vapid fog]

Going to check the other ini files.

ah.

[sage lance]

yeah its not there by default but the default is like 4 seconds or something

[vapid fog]

BaseEngine.ini has it at 4.

[nova wasp]

I doubt simply delaying will help here I guess but just in case this is a place to investigate

[vapid fog]

Well, the delay would BE the problem. Since the client is trying to connect before the server has even initiated travel.

[sage lance]

removing that delay was clutch for me 😄

[vapid fog]

So, this will reduce the issue, but is unlikely to solve it entirely. The reason is because my client is throwing a socket error: SE_ECONNREFUSED.

Now, that's interesting because its only ever used in one place relevent for my setup. In FSocketEOS::SendTo.

    // Check for sending to an address we explicitly closed
    if (WasClosed(DestinationAddress))
    {
        UE_LOG(LogSocketSubsystemEOS, Warning, TEXT("Unable to send data to closed connection. DestinationAddress=[%s]"), *Destination.ToString(true));

        SocketSubsystem.SetLastSocketError(ESocketErrors::SE_ECONNREFUSED);
        return false;
    }

Which means that its trying to send to the socket that it already explicitly closed, and not using the new socket its opening (or perhaps the new socket is also getting closed).

[vapid fog]

Or the new net driver is trying to use the same socket or something?

This is all lower level than I'm used to. So still picking up how it works.

[quartz smelt]

Hey all! I'm having a timeline move an actor. The movement is replicated but the more clients are connected to the server, the faster the timeline ends? Any idea why this might be

[nova wasp]

does the frame time increase when more clients are connected?

[quartz smelt]

I just have the lenght of the timelet set to X seconds, I don't/didn't know i could change the frame time

[nova wasp]

this is basically me asking what framerate each case is running at

[quartz smelt]

With just a quick google search I now understand, haha thank you Im so silly

[quartz smelt]

@nova wasp Ay bro youre mad clutch, just adjusted it and it's all working. IDK how I didn't think about frame rates

[halcyon ore]

Surely, I'm missing something basic. 😛
Trying to setup a basic Fast Array test, that uses delta replication (I had previously setup a FastArray, but turns out, it didn't use deltas at all, or its related functions...

This is my .h file: (this is on a replicated actor component)
Super stupid basic.

USTRUCT(BlueprintType)
struct FEntry_Int_Test : public FFastArraySerializerItem
{
    GENERATED_BODY()
    
    UPROPERTY(BlueprintReadWrite)
    int32 Test_Int = 0;

    // Overriding PreReplicatedRemove and PostReplicatedAdd (optional)
    void PreReplicatedRemove(const struct FArray_Int_Test& InArraySerializer);
    void PostReplicatedAdd(const struct FArray_Int_Test& InArraySerializer);
    void PostReplicatedChange(const struct FArray_Int_Test& InArraySerializer);
};


USTRUCT(BlueprintType)
struct FArray_Int_Test : public FFastArraySerializer
{
    GENERATED_BODY()

    UPROPERTY(BlueprintReadWrite)
    TArray<FEntry_Int_Test> Items;

    UPROPERTY(BlueprintReadWrite)
    UObject* Owner_Object = nullptr;

    // Custom NetDeltaSerialize implementation
    bool NetDeltaSerialize(FNetDeltaSerializeInfo& DeltaParms)
    {
        return FFastArraySerializer::FastArrayDeltaSerialize<FEntry_Int_Test, FArray_Int_Test>(Items, DeltaParms, *this);
    }
};

template<>
struct TStructOpsTypeTraits<FArray_Int_Test> : public TStructOpsTypeTraitsBase2<FArray_Int_Test>
{
    enum { WithNetDeltaSerializer = true };
};

Then, a function that I run on a timer, to just have data change.

void UComponent_Fast_Test::Random_Entry()
{
    FEntry_Int_Test Entry = Actor_Array_Int.Items[FMath::RandRange(0, Actor_Array_Int.Items.Num() - 1)];
    Entry.Test_Int = FMath::Rand();
    Actor_Array_Int.MarkItemDirty(Entry);
}

A normal on rep gets fired off, when I change the value.
But, not its custom replicated functions for add, remove, change?

[nova wasp]

iris or normal replication (only changes the parent fast array iirc and maybe the function in delta, this is fine if not on iris)

[halcyon ore]

Normal

[nova wasp]

struct traits are wrong

you want to trait the fast array itself, not the item afaik

[halcyon ore]

It is.
Its targeting my FArray, not my FEntry
Sorry, that my names are all over the place.
I kinda expected this to work, and then I actually fill in data, and make it logical.

[nova wasp]

ah, got baited by the nearly identical names

[halcyon ore]

The only replication event, that works is the PostReplicatedAdd, when the actor becomes relevant.
But, when the data changes, the change function doesn't fire.

[nova wasp]

ah, I think I have a guess

you are copying in the item to mark dirty

void UComponent_Fast_Test::Random_Entry()
{
    FEntry_Int_Test& Entry = Actor_Array_Int.Items[FMath::RandRange(0, Actor_Array_Int.Items.Num() - 1)];
    Entry.Test_Int = FMath::Rand();
    Actor_Array_Int.MarkItemDirty(Entry);
}

do this instead

[halcyon ore]

I'll give it a try.
I thought I got it correct, cuz it yelled at me when I forgot to MarkDirty the item, I assumed it would do the same, if I did it wrong

Bruh.
Of course it was that simple. lmao

[nova wasp]

most ides will have a little warning when you copy something accidentally (or intentionally, not really a warning)

easy mistake to make though... I do that a fair bit

[halcyon ore]

I guess Rider don't care. 😛

[nova wasp]

Rider adds a little arrow that indicates a copy occured here

it's not really possible for Rider to know why you copied it and if you wanted to on purpose

[halcyon ore]

Oh, I thought you meant it would throw a full yellow warning.
Not this tiny little icon. 😛

[nova wasp]

arguably the MarkItemDirty impl could ensure upon the address being outside of any array allocator's range but idk if that fits well

yeah, that's there telling you what happened

[halcyon ore]

Well, thank you.
At least it was simple.
Now I can actually continue work.
Still getting used to reference, copy, especially when dealing with BP, where its like a fake reference or something.

I was under the impression copy/ reference stuff was between functions.
Not internal inside 1 function.

I guess from my BP experience that makes no sense, in hind sight, cuz BP specifically gets ref, or gets copy based on function use.

[nova wasp]

well, here the thing is the Item actually has internal state

FFastArraySerializerItem stores 3 separate integers for tracking state

you should definitely try to pay attention to things that have "out" variables etc

which is something passed in as a mutable reference

MarkItemDirty modifies the state of the passed in item

and you also would not really see the change in the array item if you only made a copy on the stack have a random number

unless passing in the item would also copy it and try to re-add it to the array (that would be weird)

// its assumed the array is not empty here, of course
    const int32 RandIndex = FMath::RandRange(0, Actor_Array_Int.Items.Num() - 1);
    Actor_Array_Int.Items[RandIndex].Test_Int = FMath::Rand();
    Actor_Array_Int.MarkItemDirty(Actor_Array_Int.Items[RandIndex]);

also could just keep it as array[i] instead but I also kinda like having local refs if this is harder to read

edited to make it a bit more correct but you get the idea

[halcyon ore]

Yeah, of course.
I just wanted a quick little test bed, to make sure it works.
So, just threw names, functions, etc together. 😛

[nova wasp]

also whenever you auto Thing = GetThing() auto NEVER tries to make a returned reference a reference

you must define it as auto& to get that

[wintry forge]

Can you nest a FFastArraySerializer inside another's FFastArraySerializerItem?
E.g. I have a gameplay tag stack that inherits FFastArraySerializer. And I have an instanced item that contains it, which is a FFastArraySerializeraItem used in my inventory List which is a FFastArraySerializer.
I see Lyra seems to use subobject replication for the same thing, but I was wondering if I could get away with just using structs

[meager spade]

not really no

oh i mean if its in a diff outer

then yes

[wintry forge]

So it'd at least still get serialized properly? Will it get delta serialized?
And since I have your excellency, would a UObject array for inventory items that have instanced properties and replicateSubObject (kinda like Lyra does) be better?

[meager spade]

this is all personal preferences, in our inventory we dont replicate the UObjects for inventory items (we build them all locally from the replicated fast array)

[halcyon ore]

Quick curiosity check, while unable to be at my PC.

Does Fast Array delta stuff work with by default with others structs inside the entry?
Or, can it not correctly delta structs?

[crisp shard]

Coming back to this, and to summarize the current state:

• gun fires “bullet” from subsystem pool of data driven bullets

This was for performance of having many bullets at once etc. happy with that part

The issue: I don’t know how to do a multicast event from this without having to create some sort of external actor to handle it because the subsystem has zero way to support a multicast so I have to delegate it out to something

I attempted the gamestate (a component on GS) and it works perfect except an issue someone else brought up which is that every time a player fires no matter where the event happened the game state is going to relevant therefore the events will always happen even when they “shouldn’t” as they would be out of range and have no need to see it.

My only thoughts are to do this on the character but I just see that as a cop out move as i know it’s always going to be valid for the time of the shot. I’d put it on the gun but while a bullet is in flight it’s possible they unequip the weapon making it invalid and won’t fire those events (which are purely visual). Damage I wouldn’t mind having on game state or game mode even still that event would be “fired” for anyone regardless of their relevancy for it.

Long winded but yea trying to figure out what alternative options I have in my case. Using a global visual manager actor would give me the same issue as the game stare as far as I can think but yea any thoughts would be great

[crisp shard]

i suppose the character is really the best place unless someone stops me

[crisp shard]

im putting it on the character and no one is going to stop me!!!!!!!!!!!!!!!!!!

unless someone wants to chime in

i lied that's dumb and going to give same issue

[crisp shard]

going back to the gamestate until someone saves the day with new info

[crisp shard]

im prob going to keep ranting on this until i find a solution, but again, how can i get these data struct bullets to have replicated visuals associated with them?

you can't multicast from the subsystem, so that's out

a delegate on gamestate or character, would fire for everyone that is bound to that meaning unneccesary amounts of events being sent, so thats out

what other option is there? how is anyone using a subsystem for object pooling and getting any sort of visual when the bullet or projectile is in the air?

this is seemingly is way more difficult than making the actual subsystem in cpp and i dont even know cpp like that , so i have to be missing something or there must be a better apporoach for this

[dark edge]

Why doesn't subsystem just spawn an actor to replicate through?

[crisp shard]

that would be what i would have considered, but with the struct data for the bullets, they aren't actors, so spawning an actor would be redundant to entirety of the point of the pooling non-actor bullets

you shared an example the other day of yours, so what are you doing to mediate the issue?

i've thought to use an interface as well for this, but idk again just seems like im doing too much to get the same result, but i do like the bullet pooling , and it does work, but no visuals

[halcyon ore]

1 actor to mediate the issue of 500 actors seems perfectly valid.

[crisp shard]

what do you mean 1 actor tho? im assuming he meant to spawn an actor for each bullet

[halcyon ore]

No
1 managing actor for replication.
Cuz subsystem can’t rep

Since he said “replicate through”

[crisp shard]

but i have the game state, the character i could use if that was the case. unless im missing the benefit of doing it in a seperately generated actor

[halcyon ore]

That works then.
But an actor helps for a closed 1 purpose system.

[crisp shard]

i'd also have to figure out relevancy with that actor, cause woudnt i need to set it to always relevant or move it around so that it would replicate the visuals to the proper players? like say it spawns this actor at 0,0,0, and player is 1000000 away, i would assume it wouldn't fire for them ?

[halcyon ore]

Hmmm
Unsure then.
Maybe he meant individual managing actors per player (only relevant to owner, and set owner to the 1 player)
Either methods work at the end of the day.
Just an idea of how people like to design stuff.

[crisp shard]

that would make more sense, but im currently leaning towards an interface that fires back to the attacking player after the shot and (if) a hit, and those two interface messages i can do things there

seems like the only option to eliminate all other issues but i haven't tested yet, so i'll see soon

actually, scratch the "shot fired" event, that's unnecessary as well, for the subysytem. i'd just need the "hit" event really, and tbh, it's ONLY for hit VFX, cause i could just do the damage directly in the subsystem event

[crisp shard]

@kindred widget im pinging you cause you suggested this. how would i get visuals to work with this bullet pooling, cause there are no easy solutions as far as i can tell with this subsystem method. i.e. getting the simulated proxies to see visuals of the actual bullet, where it actually is

prior i had a attached visual to the bullet (it was an actor prob was less performant) but this showed where the bullet actually was. doing this on the local client is simple because i can get the niagara data directly at once, but the server i can't, i have to send it after in a multicast, and if i were to just fire the subsystem bullet event via a multicast, then it would be 'late" becuase the server already fired and has to replicate info back to the simualted clients

[nova wasp]

the simple thing is to use instanced static meshes

and maybe niagara data channel for spawning the vfx

[crisp shard]

i actualy am using niagara data channel for the VFX (just learned about that too), but if i use ISMs or NDC, they both have the same question to me

which is, via the subsystem, i can't get the visuals to match the actual bullet without getting the math that is in the subsystem itself. so server event could fire, but the multicast would give simulated clients a false visual. i.e. they'd get damaged, THEN see a bullet visual fly at them

[nova wasp]

of course you still need to know about the initial fire on clients

not sure where I did not say they need to know about that

I suppose your question was more about the networking than the raw rendering then

[crisp shard]

im a bit twisted up as i've tried many things. it's really that the subsystem has no easy way of communicating directly to other clients, so if i were to tell the clients about the intital shot, i'd need the correct math of where that bullet is going

[nova wasp]

how complicated is the bullet simulation

[crisp shard]

prety sure it's like just fake phyics math

const FVector CurrentPos = Positions[i];
const FVector Gravity    = Data->GravityAcceleration;
const FVector NextPos    = CurrentPos + (Velocities[i] * DeltaTime) + (Gravity * (0.5f * DeltaTime * DeltaTime));
Velocities[i] += Gravity * DeltaTime;

[nova wasp]

so I would say sending the bullet position in realtime is possible but difficult unless you timestamp each new piece of bullet information (or in aggregate)

so one simple thing might be to send where the bullets start and their direction + what kind of bullet they are (to derive their velocity, or just the entire velocity raw if that is arbitrary)

[crisp shard]

i do have all of that in a DA in the gun itself, different for each gun, so that i would have, outside of the subsystem

[nova wasp]

subsystem/interfaces here are not really relevant as that's just a way to organize code... you could arguably make the subsystem spawn manager actors for each world cell or something but I actually think the idea of these being on the character (or a character's actor component) is a smart idea

any object can replicate if you want to suffer enough but we want to stick to actors and their components here to avoid complex ways of doing things that aren't nice

wrapping the way you ask the weapon to fire in an interface or whatever is fine, but interfaces are just a nice way to add common functionality to things and not really going to change how the actual data is sent around by unreal

[bright summit]

I need to test my game on separate game instances, using AdvancedSteamSessions. How can I connect to another game -nosteam option? open 127.0.0.1:17777 do not work. (I checked, this is the port for a server).

UE 5.6

[nova wasp]

@crisp shard
I think my general reasoning about how to make this work well depends on multiple hot takes:

The main thing here is balancing "knowing exactly what happened" about the bullets with the reality that sending ALL of the information might not be worth doing or necessary depending on a few things

I think of this as 4 distinct things I guess that all have different reasons for mattering:

• about the start position and state of each new bullet
• knowing about the "results" of the bullet hitting something, knowing about the visuals of where bullets that hit things that don't matter (perhaps bullets that miss damaging something are not as important as ones that hit players)
• bullets hitting players that or damaging them (does it matter to know about that in each individual case or just the results?)
• about the state of the bullet "in flight" (can something stop them? can they change direction or state?)

hot takes:

• if there are A LOT of bullets firing very quickly it might no longer be reasonable to send each one as an invidual "event" of something firing. in that case a "burst counter" is definitely kinda mandatory and should work fine even for slower things imo

• sending like an rpc per fire event is going to sound weird on clients as the timing of the audio (and maybe visuals) should just be done locally imo as you really just can't rely on rpcs to happen at a certain rate
  as you can tell there is a trend here of more bullets = being better to fake by running visuals locally for multiple reasons. Nobody will notice you firing 2 more bullets from a 40 round magazine I think

• If bullets fire VERY slowly or are "important" to know the true location of it might be better to just have atomic information about the state of one bullet in time. This would mean you want a sort of singular "I fired this" and some kind of way to track that object with some id, index or separate object if it's really darn important to have it adjusted in realtime

• if you onrep the positions of the bullet moving through space you are in for a difficult time to basically smooth the position of each frame. It would need a timestamp or frame index from the server to make it not basically vibes based when this arrives or you could just have it be "close enough" and only change the local fake one when it diverges a large amount

Locally simming the in-flight bullets and only having the result of them hitting things that matter be the result of replication on those objects seems simple to me and is what I think many games do

Also I think not having it on the weapon itself is not a bad idea... you do not want to be still firing a weapon you swapped away from. One advantage of this though is that sending it to that weapon means the weapon can derive the firing info based off of what kind of weapon it is I guess.
Arguably the client could just locally go "actually I don't care" about the incoming firing info OR the item swap could just wait until firing is finished but the more separate replicating objects and properties you have to more timing can be offset from one another and things will overlap in weird times.

[nova wasp]

sorry, dont want to crowd out Dexterowski's question about steam sessions

[bright summit]

don't worry, I will ask again in few next days if no body answers 🙂

[nova wasp]

Also I guess sending when the bullet fires and having it sim locally will of course depend on a few things about how you derive it locally. For example sending just the "the character fired now" will make it rely on the character's local view rotation on each client which will make the result wildly different but could be totally fine if generally speaking you can't rotate quickly

Sim proxy rotation is often smoothed out over time to not look weird (but arguably the smoothed result and the most recent rotation can be separate values)

however if you always use the most recent value and then have them fire the gun while turning you will see a sort of "wagon wheel" of rounds going in directions where the rotation stayed the same over N frames until the next onrep of the view rotation

Also if it helps being able to "buffer" the results of this information locally will let you get away with a LOT of smoothness that would otherwise be difficult. For example if you know the bullet went from

A ---------> B

from the server over the course of a 10th of a second or something you could just visualize interpolating between those two states with hermite interpolation

The disadvantage is that the client must know about and both these states which can only be known after both have happened, which adds inherent delay to what they see

extrapolating afterwards based on the most recent info and simming locally is possible too when you don't get new info, this is basically 90% of how cmc sim proxies work

https://developer.valvesoftware.com/wiki/Source_Multiplayer_Networking you likely do not need to go this far for bullets though as it's not like most of them change in-flight (the .1 second here is basically the "buffering time"... the space between states is arbitrary as long as it's a timestamp you can reason about locally timing-wise)

[crisp shard]

reading through everything and the "burst counter" doc as well (which seems to be the move for sure already).

to respond in order to your messages above as well:

• subsystem was meant to be solely for bullet pooling, i didn't know it was essentially a black box that i can't really interact with it's internal states easily (i can send events, etc, but not in a way that im used to), and i like the idea of handling things via a char component but yea what to choose depneds on your later questions
• definitely agree that i am not trying to replicate anything unnecessarily or make anything more complex, in fact was trying to do the opposite
• was mainly thinking of the interface only because i knew i could then at least have the relevnacy of the character that fired so that it would replicate like any other character replication var would and be relevant accordingly, but again only thought of that as it was the only method i could conjure to get out of the subsystem and back into a class that i know of

some things to mention about my chars and how much i care about x,y,z etc.

• characters are 2d sprites, so rotations are going to be wonky no matter what (in other words, i am not a stickler for something being "off" lokking when it comes to the characters)
• i want to have an accurate "hit" event regardless of it hitting a damagable actor (and when i hit anything damagable, thats actually handled in the actor / player that gets damage, so i dont really need to do anything hitfx related from the bullet itself, although, i did have both prior (standard hitfx + bloodvfx/damage vfx)
• not sure what would count for alot of bullets, but it's not a call of duty game, so i dont imagine THAT many bullets at once, most guns are single shot, like sniper (and it has a cooldown), but most dont have cooldowns
• doing the firing locally (per your one message) would be ideal to me, i just dont know how i'd trigger it with the current appraoch

running out of room so hold up

[nova wasp]

I guess a rough idea of the per second amount of bullets in the worst case would help

[crisp shard]

still typing up some other thoughts, but in my pool i have 2000 set, so i'd aruge there's never a case where there's that many bullets in air and active at once, but that's probably the most possible. 20 x 30 rounds per second, like 600? even if i assumed 50 players (which i dont even plan for) firing at 30 bullets a seoncd we're only at 1500

[nova wasp]

the pool here is not really important for networking, we don't need to care about the memory representation or how these tick when simmed locally for the networking part... it could be the actors asking the subsystem for a new pooled object whenever they need one or whatever

you can have an onrep of an integer spawn a whole new actor, it doesn't mean the object that has the onrep has to implement what the actor does

[crisp shard]

the "locally simming" is most likely what i want to do, just not sure how i'd do that with the curretn approch.

and yea i didn't want to do the callback/logic on the weapon as you could possibly dequip the weapon while bullet is mid flight, and that would case an issue postnetially, so i did think to just opt to use the character itself for handling any callback type events like a hit event for visuals (where i could safely multicast or handle damage)

accuracy, despite me wanting to have the bullets line up, is more about making sure the visual of the shot impact is directly happening when damage would happen, getting a proper tracer to follow the bullet is extra, but it was something i had working , but is was significnatly easier to setup, given it was an actor and i could attach a thing right to it

[nova wasp]

the collision is just... a scene query from where it was to where it is going (or is now)

if it helps there are many games that are co-op that simulate damage locally

for example Destiny and souls games

so you can arguably get away with damage for the local player being on their client (if you don't care about potential cheating here)

that way there is no desync ever with the visauls of the bullet hitting them and doing something

or being simple to dodge even at high latency

[crisp shard]

well i'd still want server auth damage, i only really am prediciting the shot / impact / vfx

[nova wasp]

how fast do these move

[crisp shard]

it does look a bit strange when you hit a palyer at mega lag, as you'll get a hit effect, THEN get the damage hit effect

[nova wasp]

so is this a pvp game or not

I missed that

[crisp shard]

uhhh like a bullet speed lmao v fast, i think i have sniper set at 120,000 or something

maybe less

[crisp shard]

i mean, yes, but not on the call of duty / fortnite level, but server does validate damage yes

[nova wasp]

predict locally and have a buffer of assumptions based on what you sent -> undo those assumptions when they aren't confirmed by server

or just don't show the hp bar moving until real authority damage happens

[crisp shard]

this is exactly what i have setup, i just changed the bullets to be fromt he subsystem

so all my visuals and damage i need to rethink on how to communicate from that subsystem out

that's really the big problem / confusion im having

which is where the interface or delegate idea came up

delegate was bad tho, as it's a subsystem, so everyone is going to get that event, which i dont want

[dark edge]

assuming you just need hits, just multicast through your net manager actor when hits happen

[nova wasp]

once again not sure I see the reason the subsystem or interface is important here, you can put this in a static array for all it matters as long as unreal replication is on a replicating property

[dark edge]

or just run them locally everywhere

[nova wasp]

multicasting for hits here also has multiple purposes... to tell the users that damage happened to them or others and visuals
those can be separate things with different reliability as needed (and you can just send multiple hit events in a buffer at once if needed)

[crisp shard]

maybe im just simply noob af and missing smoething basic, but the subsystem is what has caused my difficulties in the sense that there is no easy way for me to call a multicast event upon hit, without sending an event outside of the subsystem, i.e. to the character or the weapon (which we've discussed weapon wouldnt be ideal)

[dark edge]

subsystem spawns a proxy actor that it can network through

[nova wasp]

yeah, that would work

in that sense the subsystem is just a simple way to talk to a manager actor or a specific actor for an area... I don't think it is useful to replicate a subsystem here

[nova wasp]

I don't recall saying the character would not be ideal

[crisp shard]

no no, i meant the weapon(gun) in this case, character would be the ideal spot (up until this new proxy actor idea)

[dark edge]

starting with the subsystem is great because its lifetime is tied to the world, so it can spawn an actor to replicate through that shares that same lifetime, just all automagic

[crisp shard]

i did actually read about these today, would this be something that would have like the tempNetRevlanct thing

[nova wasp]

it being on character or a per-character weapon manager component thing is nice because it means relevancy Just Works™

[dark edge]

All depends on if you need relevency to be per player or not

[crisp shard]

that's exactly why i thought char would be ideal, is relevancy is handled based on char

[dark edge]

a sniper shooting across a valley might not be relevent but their bullet hitting next to your head sure would be

I just replicate it all, my maps are huge but ranges are long so there's not much saved in not replicating

[crisp shard]

ok real quick, from the subsystem, how would i communicate to the character? (without an interface or delegate)

[crisp shard]

yea that's something else i considered

[dark edge]

why do you gotta do that?

[crisp shard]

well this was still considering the Pre-proxy actor situation

[nova wasp]

yeah I'm super confused as to why we need a subsystem at all here

[crisp shard]

the subsystem is for pooling bullet's that all it is

it's an object pool

[dark edge]

should be just an array of data structs but sure yeah

[nova wasp]

then have the things that actually create the LOCALLY SIMMED bullets talk to the subsystem... it doesn't matter

[dark edge]

do you have a projectile problem that this is solving? how many bullets in flight are you expecting?

[crisp shard]

do you mean it shouldn't be a subsystem?

[crisp shard]

i want to have max bullets, so im preparing for max bullets

[crisp shard]

this is bit over my head on what you mean

[nova wasp]

given the speed of the bullets being more than 3x the speed of sound if what they said earlier is accurate the lifetime of a bullet in-flight will be very short unless we are shooting real damn far away

[crisp shard]

i feel like i should share the code exactly as it is, so we understand what's happening

[crisp shard]

mabye my number was wrong lmao

[nova wasp]

no that's honestly... pretty realistic real life bullet speed

[crisp shard]

ok pretty sure that's what i based it on

[nova wasp]

not trying to make it sound funny there, you can make them whatever speed you want

[crisp shard]

lmao no i got you, i thought it was funny tho, cuase thinking of it as 3x the speed of sound is hilarious

[dark edge]

Subsystem:
TArray<FProjectileStruct> Projectiles

FireProjectile(FProjectileInitialState InitialState) -> Projectiles.Add(InitialState)
Tick(float deltatime) -> for each Projectile in Projectiles-> update -> process hits

[nova wasp]

but them being very slow would mean we have a longer lifespan for each thingy, which would mean the total number of active things would increase drastically from people firing at things and the bullets taking a bit to finish hitting something

honest these are so damn fast that close-range you don't even need to simulate a separate thing

they would hit in the same frame at most server refresh rates

[dark edge]

The nice thing about this approach is that it devolves into hitscan if a proj is fast enough

same path

[nova wasp]

but a ticking manager is still a good idea as you can get fancy and do all traces in parallel :U

[dark edge]

just lives for one tick

[nova wasp]

yep, it's just resolved after the tick and that's it

generally speaking you will want to handle firing before the array is ticked though

that way results are immediate (in a frame sense)

[crisp shard]

the process hits, where does that happen ? or what does that do ?

[dark edge]

At the end of the tick you can multicast all the hits that happened

[nova wasp]

it would happen as the result of a scene query hit result hitting something

depending on what it hit you can do whatever

[dark edge]

I just handle hits locally, only actually applying damage for the server's view tho

[nova wasp]

it could be the person firing gets info, the thing hit, whatever

and yeah given this speed

I think players wont really be dodging bullets touhou style

[dark edge]

My project has up to like 5 second flight times so makes a bit more sense here

[crisp shard]

so does that mean that if i shoot a bullet, and don't hit you, but i hit the ground next to you, you don't hear or see anything?

[dark edge]

in my version of it I sim all bullets everywhere so I'd see it too

[nova wasp]

you will want the visual hits to happen relative to the players though! so you probably want a "damage vfx" send about the hit result on a character

for example you hit the head, you don't want to have it just pop out of existence anywhere... you want vfx coming out of their head locally based on it really saying I hit the head bone

[dark edge]

I don't have prediction so it's a lot simpler, shouldn't be much divergance at all

[nova wasp]

it will desync by nature if they are moving quickly as animations and their character pos are generally locally smoothed but it's better than not seeing the vfx on the real bone imo

it's possible when you play the game in turbo slow motion or watch a replay that things will look slightly off but imo it wont matter

[dark edge]

mine's all physics without prediction
https://youtu.be/ErruOO8Ailc?si=_hUzhlOCLbstuvGU&t=24
This'll give a better example of the scale

[crisp shard]

the prediction aspect is really not the issue for me but yea, one way thinking may simplify it. my actual issue would be with sim proxies i.e. the multicasting of the hit (regardless of damage or not)

[dark edge]

it kinda is because where ACharacters are isn't the same across all machines

[crisp shard]

yea things looking slightly off is fine, i think my issue was considering how to communicate out of the system i have, as i just want to be able to do the logic i want, but have the subsystem just handle the bullet pooling and "spawning" of bullets, as well as handlin the hit of course, which would need to fire a hit event of some sort to somewhere/someone, wethere i direcdtly do damage or i do an event where i take the hit result and then do damage

[dark edge]

everyone is in the future on their own screens and sees everyone else in the past

[crisp shard]

true, but that part i can accept, lag will be lag until im ready to do like lag compensation stuff, which sreally isn't that crazy sounding from me looking into it

[nova wasp]

hits on the ground don't matter much and can just be locally simmed

the server seeing something got hit damages that thing, and the result of the hit is sent as their hp changing and a cue or whatever...

it might be wise for locally simmed bullets that hit players to not make bloof vfx or even pass through them and only "true" authority hits be what makes blood vfx for sim proxy -> on sim proxy

you can have the local player predict damage vfx and verify it but this risks frustration if you reject a lot

[dark edge]

for an FPS I'd actually do clientside hit detection with server OK'ing it

[nova wasp]

and to be clear

[dark edge]

"I hit that guy at this location at this time"
"Aight bet, lemme check"
"Hey everyone, yeah he hit"

[nova wasp]

you DO NOT need to go crazy and have the server simuate the entire pair of bone animations... you can literally just compare if the two positions can be traced between against static objects and that both players were roughly in the area at the time

which is what I would do, you don't need to be valorant lol

[crisp shard]

yea this, i understand and am 100 percent coolio with, and this is what i do btw. like exactly

[crisp shard]

yea 100percent and i dont want to go crazy

[nova wasp]

having "fancy" verification of hits that go back in time and rewind is complcated and involves ALL game state being rollbackable and that is something that is a lot more complicated than 2 fvectors lol

[crisp shard]

i just want it to do what i had doing really, and if the new method means i cannot do the tracer like effect (bullet trail) as accurate, then that's fine, i mean i could even just sim the niagara effect with the same math as the bullet via the NDC, which seemss totes fine in my head

[nova wasp]

accuracy for what part?

[crisp shard]

lmao fair, yea haven't even tried but the math seemed semi understandable

[crisp shard]

accurate as it, where the bullet is, but it doesn't need to be exact exact, i just would like the trace to be like the actual trace , so that you can kinda see where the bullet came from

[dark edge]

So sorta on this topic, any thoughts on how to replicate my 1D physics system? It's architected like this for reasons:

BaseDeviceSubsystem
DeviceASubsystem
TArray<DeviceAModel>
DeviceZSubsystem
TArray<DeviceZModel>

There are many of these. Right now I'm thinking of making a replication proxy actor per device type A - Z, but curious if there's any thoughts on a simpler way to do it.
I don't need to replicate much state at all. For example, for a regular car drivetrain and controls, the only state I'd want to replicate would be the control state (throttle, brake, steer, aim direction), the current gear (since it's discrete), and that's it. The final physics state (position and velocity) is handled through physics replication. The entire sim runs on all machines, but as long as the inputs, outputs, and discrete intermediate states are replicated, it'll converge to being in sync.

[nova wasp]

for starters what is a device's state stored as?

[dark edge]

There's lots of intermediate state like the steering angle, current engine rpm, driveshaft rpm, etc, but that is all stable and derived from the inputs and outputs.

[nova wasp]

ideally the replicated state is a reflected type like a ustruct

so there's no need to have 2 separate bodge things that have to be changed every time you add a float

[dark edge]

https://github.com/AdrielKinnunen/SpectrelightPlugins/blob/main/Plugins/SLMechatronics/Source/SLMechatronics/Private/Devices/SLMDeviceGearbox.cpp#L44
This might help, it's the update function for one of these device subsystems.

each subsystem handles all of a device type in the world.

kinda like a half-assed ECS

[crisp shard]

i've never attempted any actual physics replication but there was a pretty recent talk from epic on network replicated physics that might be worth watching but it also might be to general

[dark edge]

If a vehicle is made up of all continuous state things, not discrete, then replicating the inputs and outputs would be enough.

it's the discrete state that gets tricky

[nova wasp]

what does do the to the simulation that is different than the client's version

is the result differnet or the entire state?

[dark edge]

By discrete state I mean things like whether or not a gun fired or a gearbox shifted, those are driven by continuous state (control signals), so there's situations where the gun could have fired or gear shifted on server but not on client, even if they are all running the same sim

It's like analog electronics sorta

[crisp shard]

it's actually based on character movement physic, based on mover, so actually prob not too helpful but i'll share link incase: https://youtu.be/_jRLlTDqoGI?si=PZz5f_YJPxZLXLxF

[nova wasp]

a character is a lot less complicated than an arbitrary chain of thingies here imo

but the idea is similar

[crisp shard]

yea and pretty sure it's based ocmpletely on mover if remember correctly

[dark edge]

The final movement isn't the problem, I got that synced up fine

it's just a bit funny when your car is stuck in 2nd on your screen but driving like it's in 3rd because it IS on the server

[nova wasp]

it depends to me on why the simulation of the gearbox would diverge

[dark edge]

look at how the gear is shifted

if signal > theshold

[nova wasp]

is the signal based on local input of the player

or it interacting with the world outside of it

[dark edge]

replicated input but yeah, not necessarily per player, it's all just a web of stuff

could be 2 players inputs added together, could be the result of a rangefinder, could be anything

Could have 10 players on one vehicle all "voting" on the steering angle, anything goes

[crisp shard]

is the gear shift a state itself? i'd assume that it's already replicated given the qeustion but that would be my first thought, is to have the gear shift itself justhave priority of overriding anything that wasn't correct locally

[dark edge]

There's not necessarily an actor here, the structure inside the subsystem is the root state

[nova wasp]

that would add inherent round trip delay to all results of the local state

[dark edge]

Nothing is predicted so everything has that delay

fine for giant contraptions

[nova wasp]

oh, it's all entirely auth based? okay

[dark edge]

yeah all auth

Replicated inputs -> Big web of stuff -> Replicated final movement

[nova wasp]

in that case it seems like the client needs to know only about

• the shape of the system and its parameters (which likely do not change often)
• the input from things outside the system
• the result in basic terms which you already have of course

[dark edge]

I want to opt in SOME of the big web of stuff to be synced

since some of it is sensitive. Say the shift signal only made it to 0.99 on my screen but 1.01 on the server

server would shift, my view would not. I need to sync that

[nova wasp]

what refresh rate is required to reach that intermediate state accurately on all sides

if this is not a fixed ticking simulation it is entirely vibes based

[dark edge]

Dunno, could be anything.

Someone coulda made a looping oscillator that adds to their inputs to make the vehicle nice and random, no clue

[nova wasp]

I can see why players need accurate info about the system to reason about wtf happened

[dark edge]

It's like analog Redstone

[nova wasp]

but I wonder if they need 1:1 accurate info or if they only need some info

[halcyon ore]

Quick curiosity check, while unable to be at my PC.

Does Fast Array delta stuff work with by default with others structs inside the entry?
Or, can it not correctly delta structs?

Like can I just stick my already made struct in it, or do I need to do some cursed conversion between Fast Array struct, and normal struct?

[crisp shard]

maybe overrride client numbers only after a certain threshold of difference ?

[nova wasp]

the delta is the individual element only sending itself as 1 replication element, which means it depends on what that element does. nested structs would be dragged in with it

[dark edge]

I'm trying to figure out how to just hard sync certain properties. I could probably make a net manager actor per subsystem that needs it.

[halcyon ore]

Ohhhh
It’s the array elements being delta’d
I always thought it was individual vars inside of it.

[nova wasp]

regular tarray replication always sends the entire array (which is fine for small things)

In iris this is not true anymore (thank god lol) but you still don't get the convenient per-element callbacks without a fast array

[nova wasp]

I think one crazy idea if you don't care about simulating the state locally you could send just what they need to know as a buffer of how it changed over time

that way for example if it peaked at 1.0001 you send that range and what it returned to after a split second

[dark edge]

Send how?

That's what I'm wondering

[nova wasp]

it depends to me on how inspecting these values works locally

[dark edge]

If this was actors I'd just have that integer be replicated and call it a day

[nova wasp]

no, that will not work here

[dark edge]

No i mean if everything was actors

[nova wasp]

regular property replication = you WILL miss intermediate positions of a state that changes over time

property replication is about the latest state being eventually the same value, it does not send in-between values at 100fps

[dark edge]

That's fine for the current gear

I'm not talking about replicating the gearboxes inputs

I'm talkign about replicating the integer Gear

USTRUCT(BlueprintType)
struct FSLMDeviceModelGearbox
{
    GENERATED_BODY()

    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    int32 NumForwardGears = 5;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    int32 NumReverseGears = 1;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    int32 CurrentGear = 0;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    float FirstGearRatio = 5.0;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    float RatioBetweenGears = 1.3;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    float GearSpreadExponent = 1.0;
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "SLMechatronics")
    float GearRatio = 0.0;
    UPROPERTY(VisibleInstanceOnly, BlueprintReadOnly, Category = "SLMechatronics")
    int32 Index_Rotation_Input = -1;
    UPROPERTY(VisibleInstanceOnly, BlueprintReadOnly, Category = "SLMechatronics")
    int32 Index_Rotation_Output = -1;
    UPROPERTY(VisibleInstanceOnly, BlueprintReadOnly, Category = "SLMechatronics")
    int32 Index_Signal_Shift = -1;

    float LastSignalValue = 0.0;
    bool bChangedGear;
    FSLMEvent OnChangedGear;
};

[nova wasp]

I guess what I am saying is you must send a tiny replay of how these changed if you can't simulate it locally and you need accuracy of what actually happened

[dark edge]

I could have 1,000 of these

[nova wasp]

if it's detereministic and you have the full state of the system and each input you could just tick it and see the result separately from the visual

[dark edge]

I've thought about trying to make it deterministic but I don't think so

since physics feeds back into it

[nova wasp]

it's honestly not as hard as it sounds if you don't need perfection

and you have full control of all inputs

[dark edge]

the net result of the entire sim depends on things as subtle as the load on a tire

[nova wasp]

yes, not saying that isn't true

but it if you simulate the system putting load on the tire locally and its the same input world state as the server it would be good enough
for example the result of the rangefinder is input into the simulation that can be a single signal into the graph, the REASON it went off can happen later and be told as history

this is kinda complicated to reason about but generally speaking only things that react to the world around them here are inputs into the system, whichever one is more expensive to send could change how this works

• a wheel hitting the ground as a hitresult might be relatively large information as a double vector that would lose precision when quantized (and probably still be okay)
• but the INTERACTION it sends up the graph could be relatively smaller float vector in local space so to speak

so each frame the graph has inputs sent into it from the world, you already made that work

this is honestly something that might start to get out of hand quickly but it would be quite simple to have a copy of the vehicle system that serves as the "scractchpad" of a locally simmed version with a buffer of true inputs + how long each frame took (ideally fixed but idk what the authority is here)

with actors this would be tedious as hell but you kind of skipped the hard part by making these a simple system with inputs and data

[nova wasp]

does a rangefinder tweak the value of something it sends as a floating point value or a binary true or false

[dark edge]

Everything is floats

Say you had:
Rangefinder -> trigger (input < 10,000) -> weapon

if RangeFinder outputs a 9000, trigger outputs a 1.0, weapon fires

Now of course weapon isn't a great example since the weapon itself can be replicated, doesn't make sense to have a weapon without an actor

[nova wasp]

okay, that means that for starters we are seeing 32 bits of information per input
but if the CONDITION of (input < 10,000) is unchanging or already known
we don't need to know anything but if the condition is true or false

obviously seeing the rangefinder in-context reading world data is useful to the user

but for the simulation result it might ultimately be something that is compressed down to the sort of true or false

[dark edge]

ok let's say it's something wihtout a world context.
It's a power network sim for a city builder, and you have a power-triggered switch. If too much power flows through it, it shuts off like a circuit breaker.
None of this is actors.
Assuming the inputs (the power plant state) and outputs (whos lights are on) are replicated in some way, you'd still want to know a bit of the internal state of the network since the switch is a discrete state.

[nova wasp]

but either way the number of inputs might be small enough this kind of compression is not useful

[nova wasp]

yes, but it depends on how much that discrete state must be known in detail vs results

[dark edge]

almost all discrete state needs to be known in detail since otherwise the results will just keep being wrong

instead of being temporarily wrong

[nova wasp]

then you must send a buffer of discrete states to the users or derive it locally based on a buffer of inputs and resim

for example they might never need to see the graph of power usage until reading a power meter

how to send a buffer of states is something I am not certain about but a fast array is ome simple way to get "good enough" results very quickly if the total size of the buffer or individual slements are not huge

[dark edge]

Say I had a:
FStruct:
float A
float B
float C (replicated)
float D
float Z

and I had a fastarray of these

in an actor

would I just automagically get C replicated but the rest able to do whatever?

[nova wasp]

it would only send float C while serializing, also you would have to mark the non-replicating structs as not replicating as a ustruct but you probably know that

for example

ustruct:
    UPROPERTY(NotReplicated)
    float B;
    UPROPERTY()
    float C;
    UPROPERTY(NotReplicated)
    float D;

however in some stupid situations these NotReplicated values will be zero'd like polymorphic struct serializers (because fuck you and GAS works that way with effect contexts) every time a replicated value arrives (only relevant when sending gameplay effect contexts or polymorphic struct serializers iirc)

but typical struct serialization will leave them alone afaik

[dark edge]

I might try moving the guts over to actors, just managed by a subsystem.

in the BP API I just need AddModel and RemoveModel and MakeConnection

[nova wasp]

actors here are only useful as a place rpcs and replicated properties go, the definitions for the structures and what happens on them can be done anywhere except for marking properties dirty easily and defining rpcs etc

[dark edge]

Yeah It'd just be for my sanity, I already have 2 classes and 2 structs per device type, and over 30 device types once I'm done

[nova wasp]

also you can get really insane with how you serialize the state as a buffer and could easily make a simple serialization scheme that sends deltas per state as an atomic thing so for example going from 1.0 -> 1.0 is cheap
will require actually making serilization though generally speaking

[dark edge]

Can a subsystem enumerate all subclasses of an actor and spawn exactly one of each?

[nova wasp]

in my case I send a buffer of inputs as a
InputStateDesc
{
Input
int32 NumAfterThatAreIdentical
}
TArray<InputStateDesc> Inputs
which requires 0 custom serializers and means sending copies just means incrementing the NumAfterThatAreIdentical

[dark edge]

The whole reason I reached for subsystems to begin with is for the automagic discovery, making a new device type is just a copy paste of a file and changing the guts out

[nova wasp]

sure? why does it need to?

I feel like each seprate system being an actor with owned properties seems okay here

[dark edge]

So MechatronicsSubsystem can spawn an actor per device type instead of me making a subsystem per device type.
Then I can get the sim logic and data storage and replication all in one spot.

[nova wasp]

and what I mean by SYSTEM is the GRAPH OF CONNECTED STATES

[dark edge]

oh i got that too

lol

[nova wasp]

how to make each distinct one do different things without a uproperty knowing about them as a type is going to mean some fancy stuff

[dark edge]

I haven't even mentioned domains

[nova wasp]

but honestly if you don't need perfection one fast way to make this work is replicating finstancedstructs

• this is relatively expensive when CHANGING them because it means you have to send the entire struct type each time + the data (oof)
• but it might be faster to set up and might be dwarfed rather quickly by the system state

but yeah honestly I wonder

[dark edge]

This whole thing would be so much easier if using objects scattered all over the goddamn place wasn't so slow

grrr cache

[nova wasp]

also for example the "elecritity meter graph" example does not need a unique actor property
you could have a generic UPROPERTY that serves a buffer of graph states and that could be... anything!
it could change to represent fuel on a fuel tank you look at just from an enum changing

remember that the thing you are replicating in a unique way are shapes of things, if the shape never changes neither does the way they replicate need to change

I would say just do transforms 100% by hand though, don't try to make that generic lol

because it is too important

but you already have that

[dark edge]

Yeah I hadn't thought of fuel or battery. I got it implemented but yes that is one continuous state that'd have to be replicated. At a lower frequency of course.

since that could diverge and stay diverged for a long time.

[nova wasp]

the state you need to see in third person and the state you need to see when staring at the UI of the car dashboard are distinct things

imo

[dark edge]

There's bits of state visible all over. Engine audio, shafts turning, audio everywhere.

[nova wasp]

which is not to disagree with what you just said, just trying to make what I am saying clear

the audio being here is kinda oof because that being desynced will be weird visually and that will be immediately obvious to viewers

[dark edge]

I'm pretty certain the client needs all state all the time, it's just a matter of how synced it needs to be or not. Engine RPM doesnt make sense to sync as it can't stay diverged for long and is very transient, it changes all the time.

[nova wasp]

the dashboard showing a graph of fuel chaning 200ms late is 100% fine but hearing a bonk 200ms later is awful

[dark edge]

That's why the approach is to run everything everywhere and just hard sync the things that can diverge

[dark edge]

Plus if the fuel is 200ms late you wouldn't even notice as it'd still be kept in general sync. What would really suck is running out of fuel locally but still having a few hundred yards worth on the server, you'd hear a stall locally but keep driving.

[nova wasp]

yes, which is something that should never diverge that far to matter visually

you would already have 100 new "results" of the full picture by then

it taking .1 second to start sputtering would not matter unless it was constantly flip flopping and never really happening

IMO the most simple way to have dual simulation with authority I have seen is to have predicted state and "true" state and only do server input based playback or authority on one of them at a time

for example Spelunky 2 has 2 game sims

• the one the user sees and does input and sim in in predictively
• the "true" one that is in the background and does only verified inputs and is 100% perfect, which serves as the authority to return to when things diverge

in your case add on top that you send the final results to the "true" one as well as input

this is nice because it means that you don't have to rollback everything but only the presented state to the "true" state when divergeance occurs

but this is a bit reliant on being able to sim the whole game quickly at the same hz as the server did

[dark edge]

Oh shit, this might be the play

[nova wasp]

okay nevermind then lol

idk if AInfo has a position that can work for prioritization

[dark edge]

no prioritization needed

whole sim runs everywhere, I can handle relevency in my own project via the actor's which hold components which register/deregister models.

replicated data for models which don't exist locally can just be ignored.

but for my own game there'll be no relevency, you can shoot all the way across the map so it won't save much

I'll cross the "model persists while component/actor comes and goes" later for the plugin. That'd be useful for stuff like a Satisfactory type thing

Haven't figured out a great way to do that yet, something with GUID will be the ticket I think

[nova wasp]

I feel like a static getter to get a manager actor is like the least of the issues here lol

[dark edge]

It's one of the reasons I was reaching for subsystems, if I can trim that out and get 2 birds stoned at once it'd be helpful

[nova wasp]

also just make the only place that spawns it actoriterator first and ensure if wrong where the real spawned one should have had a cached pointer... having 2 will entirely brick the game and should not be accepted

classic AI output I guess

[dark edge]

Yeah I'd never actually use its code, I didnt even know you could make static getters for actors, hadn't really thought about how GetGameState works under the hood.

[nova wasp]

if only place these can appear is from the subsystem lifecycle and nowhere else (which is true for AInfo as it is notplaceable etc)

[nova wasp]

a pointer can point to another pointer and that pointer can have actors in it

honestly I think subsystems should just have a template for a static getter at this point too because I have written that function 20 times now

I guess I can make a simple one

[dark edge]

you can just get subsystems

in bp

[nova wasp]

yeah but it's wordy to go by the type

I am not talking about BP here I guess

[dark edge]

yeah I care about the BP api a little bit because I'd like this to be just as usable from some BP chucking models and connections into the sim and feeding it data and getting results

[nova wasp]

BP subsystem code I make tends to just be passing in a world ptr to a bp function library and that just silently getting the subsystem each time (which is not great but is okay for most)

[nova wasp]

then why does it even need a subsystem if it just has a model input? Why can't the input just have the data it needs right there?

[dark edge]

I went with the static getters so it's trivially extendable. Copy paste file, change names and properties and logic, and bam, new device system

[nova wasp]

the only thing each model and connection should care about is probably the input and state unless it needs world context for qeuries like traces

but I might be misunderstanding what they can and can't do

of course you want to be able to debug print in the world etc easily

[dark edge]

Depends on the device, some are simple and fully in the sim, some need world stuff

a tire doesn't mean much if it doesn't get and set world state

[nova wasp]

but if they can randomly get and create other new devices outside of input is not possible to control the consistency at all unless its an entirely safe way to obtain an external device's state that knows about the current frame or something

or its a separate uworld (oh god)

[dark edge]

This sim is like a parallel physics engine in another dimension

inputs and physics are how it interfaces with the world

[nova wasp]

so the bp runs in parallel as well or just sends inputs?

[dark edge]

Like say you wanted to make a standard vehicle.
You could add an engine component, gearbox component, cockpit component, 4 tire components, giving them all their intial setup.
Or, you could just add the models at begin play and have no components at all. You'd of course have to feed the tire models their external state (loading, contact point, contact normal) and feed your throttle and steer and shift inputs on tick, but from there it's in C++ land doing its thing.

You could do a raycast vehicle with 1 static mesh component and as complex a drivetrain as you wanted.

in my project I have individual actors per vehicle part since it's a lego style building game, but it's all the same mechanics under the hood. The only thing that changes is how the models get put in the system and how they are fed their input data.

You could use the same system to make a power network sim for a city builder, even if there's no actors at all when the city is out of sight. The sim exists in its own world and just interfaces with actors at the edge

Just like a physics engine. Just with more constraint (device) and particle (domain) types.

[royal vault]

Hi if I send too many reliable client rpcs, everything works, but I can no longer posesss players and it takes a while before im allowed to move again. Any ideas?

[dark edge]

Why you sending so many?

What's in these RPCs?

[latent heart]

The engine starts dropping reliable rpcs if you send too many. Just don't.

[nova wasp]

too many reliable rpcs will force a client or server to send more and more data redundantly until it cannot any longer

[latent heart]

And the problem gets worse the less fps you have.

[dark edge]

Say I had a bunch of actors with arrays of properties, and I want to just spam those properties out to clients.
Is there a way to "batch" it so I just stream those or should each actor just do it itself?
I guess my question is more like, 1 actor with lots of data or many actors with a bit of data each?

[hollow breach]

i'm trying to create a procdural level loading system that loads a level instance at a 90 angle until it created a 4 quadrant level system

Server creates seed
Multicast loads level instances based on seed defined by server
All actors are already placed within the level.
All actors are replicated

End result:
Host actors all function properly
Remote actors don't move and don't replicate their information properly
This is ONLY the case with loaded level instanced, and works perfectly fine when all placed within a persistent level.

[dark edge]

I don't know any way to do this automagically, best I can think of is to replicate some state that drives the level instances and their transform data, then cruise the level serverside setting the actors to replicate.

What happens if you JUST spawn an instance on the server, does anything happen on client?

[hollow breach]

No, sadly

The only thing that I can think of is to create a system that will load the ground and static actors for server/remote, then load a level with the actors but ONLY on the server, then, have a system which stores the transforms of all of the actors.I have marked with a tag, delete those actors, then, spawn them again on server and have them replicate to the client

But that seems like such insane overkill

Like, there has to be a way for this to be simpler

[halcyon ore]

Did you turn off that net load on client, so the client doesn't attempt to make a client only copy in the spawned map?

[hollow breach]

Like on every actor?

[halcyon ore]

On the ones that need actual replication (NPC's, doors, etc)

No need to care about a wall or something.

[hollow breach]

I tried that with one of them, but they ended up never loading

[halcyon ore]

Damn.
I was hopeful. 😛

[nova wasp]

I honestly do not know because it feels like you have this idea of a giant global state that is going to get weird no matter what if segments appear one at a time

in my mind the ideal is you have an atomic packet that describes 1 isolated thing in entirety or at least a given area of the game

problem with having it be based on "manager actors" in cells is that while it seems simple at first I'm not sure how to handle data moving between them and arriving out of order (bake a frame number into it and just go from there?)

Some projects that REALLY need perfect full state send even just go the full "screw it, more data" and just have an http stream which is not nice but would basically work like downloading the state over time I guess

[royal vault]

sending voxel chunk data. it only needs to happen once:

(Tick)

    UActorChannel* Channel = NetConnection ? NetConnection->FindActorChannelRef(this) : nullptr;
    if (!Channel) return;

    int32 SlicesSentThisTick = 0;
    
    while (Server_BytesSent < Server_TotalBytes
        && Channel->NumOutRec < (static_cast<float>(RELIABLE_BUFFER) / 2.0f)
        && SlicesSentThisTick < MAX_SLICES_PER_TICK)
    {
        const int32 Remaining = Server_TotalBytes - Server_BytesSent;
        const int32 SliceSize = FMath::Min(Server_MaxSliceSize, Remaining);

        TArray<uint8> Slice;
        Slice.Append(Server_SerializedChunkDataSaves.GetData() + Server_BytesSent, SliceSize);

        // Send slice via reliable RPC
        Client_ReceiveChunkSlice(Server_BytesSent, Server_TotalBytes, Slice);

        Server_BytesSent += SliceSize;
        ++SlicesSentThisTick;
    }

    // Finished streaming
    if (Server_BytesSent >= Server_TotalBytes)
    {
        Server_SerializedChunkDataSaves.Empty();
        Client_ChunkStreamFinished();
    }

im already preventing sending too many rpc's at once

and the client RPC's get through correctly w/o lag

its just afterwards when i spawn my player- the player doesnt get posessed and the client cant move. but the game isnt frozen its just i cant use the player.

until a little while later (the photo shows just stuck inside my player eyes)

if i remove any client RPC's the posession and gameplay work completely fine

the client rpcs are sent only once, gameplay is just broken AFTER they have already been processed, until some strange amount of time later

[nova wasp]

well

is the reliable rpc to possess being sent during this massive mega rpc send?

also may as well just zip this bytebuffer if it must be sent

https://vorixo.github.io/devtricks/data-stream/

also check output log here I guess

which I assume you have already done? anything in there?

[royal vault]

the posesss rpc is sent only after the data is completely assembled from the client rpc's, but its weird that the possess takes >20s to execute

[royal vault]

o interesting thanks ill check it out

[royal vault]

ill take another look kk

[nova wasp]

I think you are trying to ask the actor channel for the number of reliables but

this is not per-actor channel

this is per connection

also this code is going to not work when Iris is the default

actor channels do not exist in Iris

also arguably you could avoid needing to pray it doesn't overflow by just not sending more data until an ack is returned from the other side

[royal vault]

yeah this makes sense- do you know any way to do that, i read the article you sent me i think it also uses the actor channels tho.

[royal vault]

will prolly end up trying this tho ty!

[nova wasp]

looks like your code is very similar, yeah

do you modify the default unreal network rates at all?

I guess it kinda depends on if Connection->GetOutgoingBunches() gets more stuff added to it or not from other channels

I'm not really sure but you can just breakpoint it and see fairly quickly

[nova wasp]

I am not sure and it's kind of surprising there isn't a simple way to go "will this reliable rpc break the connection or not?" I guess

there is technically a way to disable the network bandwidth limit in a scope

FScopedBandwidthLimitBypass in network prediction

why do they need this? because their input rpcs are huge I guess

it's really simple so you should probably just copy pasta it out

all it does is basically directly write the QueuedBits of the connection to be the same as it was before in the scope

[halcyon ore]

Still struggling a bit with this stupid FastArray shit...

So, I of course wanna handle the initial replication, manually.
But, how in the heck do I stop the initial replication, while still allowing replication after. 😛

My initial idea was to mess with the NetDeltaSerialize, but I can't figure out how to know if the Serialize is for the initial add, or from changes.

Or, am I going at this all wrong?
Cuz, I know someone suggested To use the rep condition, but that doesn't support multiple people (since rep condition targets everyone)

[nova wasp]

why is the initial replication bad here

[halcyon ore]

Cuz, it'll explode the rep limit.
hence the need to override, and batch it

[nova wasp]

read the (my earlier message) message right above yours

[halcyon ore]

I tried that.
It lagged shit to hell and back for a couple seconds, with a super simple test.
(movemenet and shit was out of sync for like 5 seconds)

[nova wasp]

what's actually in this fast array

[halcyon ore]

My items, this shit we spoke about last.

[nova wasp]

have you profiled it?

what is the actual direct error log that is happening

network tracing tl:dr

-NetTrace=4 -tracehost=localhost -trace=net,default,task launch with with this and press insights, look at shiny graph and see what is in packets

[halcyon ore]

I haven't debugged enough for the actual error log.
But, 1 item is 50 bytes at the current moment.
Which for a basic item like that, is 1300 items.

Let alone any other planned modifiers, that aren't in that.

[nova wasp]

okay I am not sure I can help if looking at the output window is not possible

it is really important to have this in your editor open at all times when doing anything technical imo

or some other live log thing

[halcyon ore]

Not sure I follow?
I already know the profile info.
Do you want me to re-re profile right this moment?

[nova wasp]

an actual insights profile? that is 3 seconds to make if you launched with the right args (use the ones I posted above)

sure? it might be separating it out into multiple bunches though

you can show me the old one if you want

to be clear I mean Unreal Insights and not stat net

stat net is still convenient sometimes though

if network insights is being strange it might be you need to start and stop it and then press play in some weird versions, it's kinda fickle

[halcyon ore]

What eactly do you wanna see from it right now?
A pic of one of my item reps, or?

[nova wasp]

sure? I guess

and I guess what the actual problem is with replication being too much, I don't know what the problem is
is disconnecting people? check the output log

[halcyon ore]

I don't have an actual error yet, I just know i'm gonna have issues, repping a wall of data filled items in 1 network bunch by base UE

[nova wasp]

I guess you could delay it being relevant or something?

I'm not sure I follow the idea of guessing here when we don't need to

if it's too much data then you can try to split it off into sub objects or something

but I'm not sure if that will really help reduce bandwidth and will actually add some overhead for some parts

I've made a 1k+ element fast array once and it just kinda reduced the send rate on the end clients

so you may as well just try adding random crap and pressing play

[halcyon ore]

Ok.
I'll get on that, for solid info and insights.

[nova wasp]

also keep in mind in newer versions you can have clients join later than play

etc

which can help figure out things from a new client showing up out of nowhere

[halcyon ore]

No, its no issue like that (right now at least)

[halcyon ore]

I realized I messed up my test.
I forgot to account for current, and max, my test only had current on it.

Ok, insights/ test results:
1 item - 788 bits
200 items - 129,031 bits
400 items - 257,231 bits
600 items - 385,396 bits
800 items - 513,631 bits (it started triggering bad ping, and ServerMove: TimeStamp expired: 20.660255, CurrentTimeStamp: 21.701906, Character: Player_Character_Female_BP_C_0

850 items - unknown bits (it triggered the Bunch size ensure)

Attempted to send bunch exceeding max allowed size. BunchSize=68228, MaximumSize=65536 Channel: [UActorChannel] Actor: Data_Tester_C /Game/ThirdPerson/Maps/UEDPIE_0_ThirdPersonMap.ThirdPersonMap:PersistentLevel.Data_Tester_C_5, Role: 3, RemoteRole: 1 [UChannel] ChIndex: 17, Closing: 0 [UNetConnection] RemoteAddr: 127.0.0.1:52873, Name: IpConnection_1, Driver: GameNetDriver IpNetDriver_0, IsServer: YES, PC: Controller_Player_BP_C_0, Owner: Controller_Player_BP_C_0, UniqueId: NULL:DESKTOP-M7EKB5K-5CFEED7248BFD81BBFABB69371097F8A

Data in test:
Class - item class
int64 - item ID
GamplayTag Array - 10 entries
int32 array - also 10 entries, that matches the gameplay tag length.

[nova wasp]

can we see a screenshot of 1 item in the sent packets

[halcyon ore]

The insights didn't split the values into there individual sizes (is it because its like 3 structs inside of itself?)
FastArray -> Fast Array entry -> custom struct.

[nova wasp]

oh, that's a partial bunch

does it show the true size right after?

instead you could just send 2/3 items or something that should not be forced to be a partial bunch

to just see 1 at a time

you could also show the writing half as well as that might not be all partials

[halcyon ore]

As in the server send you mean?

[nova wasp]

well okay

for starters what is the item id

a 64 bit integer that is unique to what? the kind of item? the instance?

what is the actual range of values this number can be in terms of min/max

[halcyon ore]

I wanted the item ID to be unique, no matter what happens.
and, one of my friends told me about discords snowflake ID system.
So, I basically made that.

Its 3 values stitched together.
Epoch time + hashed name + a counter that resets every couple seconds.

[nova wasp]

and just as I explained before you can see sending even small strings is brutal here and should be avoided if you plan on having many of these
2 duplicated (well not duplicated, you know what I mean) arrays of the same size is not wise as well as you are sending redundant information (the same size of n elements) but I am not sure having a nested array will ever really be nice here as this is not delta serialized

I have no clue if you can have fast arrays inside fast arrays or something but a gameplay tag container is fairly fancy in terms of serialization in comparison to a regular tarrat (just see its netserialize etc)

[nova wasp]

this is a good idea in theory but I just don't know if you can reasonably send 800 full sized 64 bit things... there is a reason network ids are small

[halcyon ore]

I know you can't reasonably send 800 full sized 64's.
Thats why I don't wanna do that. lol

Hence my want to batch it.

Which so far I've only been able to do with my tedious ass manual methods.

UE doesn't seem very supportive of it. 😛

[nova wasp]

one thing I'm not sure about is

why does the client need to see 800 items at once ever

if you don't want to use reasonably sized network ids or avoid strings when sending stuff this might be better off being sent only like when they open a chest to look inside something

[halcyon ore]

That is when it’s sent.
It’s not always known.

[nova wasp]

as for stuff like having a larger piece of data represented by a smaller piece of data that gets resolved on both sides iris has built-in stuff for this but it's a bit advanced and idk if I can safely advise using it

it still requires sending the true full thing once at least

I'm personally not convinced you need to send 64 bits of data to the client for them to have a locally unique thing... if this is a "forever" thing then maybe keep that a server conceit

for example

32 bit item id that increments for each new item would not wrap around until it used 4 billion entries

and at lower values is already compressed down to a lower range

perhaps clients can go "give me full info about THIS item" and the server sends it back as the full "true giga id" or from transcations involving picking stuff up

[halcyon ore]

Yeah.
If I can’t get batching, then some stuff will need to change.
Just the original idea of everything was built around batching, and custom manual shit.
I only recently heard that you could potentially do batching with Fast Array deltas, which is what I was planning in doing now.
And, thus an item ID on the client doesn’t matter as much, since the FastArray should keep stuff in sync.

[nova wasp]

it's already batching it, just kinda outside of your control afaik

[halcyon ore]

If it’s batching then why do I hit a limit?
Isn’t the whole point of batching, to not hit the limit, by using it over multiple network events?

[nova wasp]

because there is a limit

it already must batch because it physically cannot send this as 1 packet

notice how none of your packets are larger than the mtu etc

we could be evil and just... raise that limit I guess

[halcyon ore]

Definitely not doing that.

[nova wasp]

there is also a hardcoded size limit to array replication that exceeding will just break things I guess

its hard to think of a game that sends me the full internal info of 8000 things at a time that isnt like an rts game

do you have to see all of them at once visually? are they like in a giant graph of items in a box or something

[halcyon ore]

There in different boxes, as put in by the player.
You only know of them when opening said box.
There is cases where you do just “know of items”, but at max it’s like 8 per player.
Cuz, seeing dyed armor and shit.

[nova wasp]

uh

can we just send that info instead

8 items per player would be no issue here with a fast array

why do you need to know of all 800 items in a box immediately

[halcyon ore]

I think you miss-understand.
You do only know of items you open the box of.
But, shooting the client with even 300 items, while an entire game is going on around you, seems a bit much.

[nova wasp]

yeah, so they should receive information about what they need to know to represent what they need to see

as for how to filter this out a bit I'm not sure how old net groups work but repgroup should be able to filter out the chest inventory pages or something on an object basis... I would seriously consider Iris here because this is starting to have problems that can only really be done easily with iris filters imo

I'm sure there's something I'm missing though

[halcyon ore]

And here I thought this would be a simple ass “return false after this if check, and profit” lol

[nova wasp]

I kinda wish there was a nicer hierarchichal way to send information here that would kind of grow in a simple way... for example a sort of lod of what to send based on significance

[nova wasp]

well, if you had these items as objects that only appear when needed it would be pretty simple too

the server could kinda create these objects that are condition owner only or something I guess

more objects = bad news on old replication unless you are careful with not abusing polling (object polling is suffering for the server lol... always push model if possible)

[halcyon ore]

That was a thought.
But same overall issue. Client gets hit with 400 items in 1 hit.

I could TECHNICALLY make some like “paging” UI, rather then a scroll box, but that would feel so meh.

[dark parcel]

How does the client get hit with 400 items in 1 "hit"

[nova wasp]

yeah, of course but that would still need a way to send info atomically as some rpc per page request

a fast array of... a lot of 64 bit ids and strings

[halcyon ore]

Open a box with 400 items.
The client has to then be told of 400 items.

[nova wasp]

there's not really a nice way to defer sending a fast array over time I'm aware of unless they directly bonk the actor channel is there

[dark parcel]

Why do you have 400 unique items in a box? Isn't that tedious and too much data for the client to process, limitation aside.

[nova wasp]

it is hard to think of a game that streams in this much info in 1 packet

for 1 object without splitting it up somehow

even just not sending strings and not using massive ID would reduce this heavily to begin with too imo

[halcyon ore]

What I thought you were gonna tell me.
Is that in the fast array delta serialize function, you can check if it’s the initial replication, and just tell it to fuck off.
Then just put custom batching in there, and profit. 😛

[nova wasp]

I think one really quick and dirty way to have the "full info" come down would be to have it create or change a replicated object(s) for the local player that represents either

• the stuff they want to see on the current page in a chest
• the current item their mouse is hovering over or something

what is the actual information that must be known upon opening the chest

hovering items to see the name is not the same as looking at a wall of icons

[halcyon ore]

I mean.
I guess I could throw most of the info into the tooltip. 😛
it would’ve been nice and quicker to have it on the main item widget.
Cuz, then I would only need like quantity, class, and quality.

[nova wasp]

the tooltip can just be a replicated object or an rpc ack... is this string just for the tooltip?

quantity can be an insanely tiny number, the quality I don't know

of course you still need to know extra info about the item in some ways, so maybe the local player's inventory will have more info than the sort of "visual" replication

it's fine to go crazy there if they only have a few

[halcyon ore]

Feels like such annoying compromises for an overall easier dev experience for me.

[nova wasp]

okay, do the iris filtering setup then lol

[halcyon ore]

I might as well just do my custom/ manual shit.
Can’t trust iris at this rate. Lol

[nova wasp]

works for me :U

I am suggesting things because they seem simple to do and don't involve making heavy error-prone serialization code or engine changes but I don't know if waiting 100ms for knowing what at tooltip something is is somehow a non-starter

hell, you could literally locally cache the damn info for later and not re-request it if an ID didn't change

[halcyon ore]

Oh, I’m not denying your suggestions.
I’m more so annoyed that somehow nothing in UE has this (expect maybe iris, which I’ve never touched)

Surely someone would have run into needing to batch a lot of data.
Unless, maybe they did just rip apart the engine. (I’m trying not to do that, just trying all the suggested methods)

[nova wasp]

this engine is not made for mmos

or factory games or whatever

[halcyon ore]

Good thing it’s not an MMO

[nova wasp]

it's kind of pointless to say but generally speaking sim proxy info is lightweight visual info that is spatial so most of the engine is oriented around doing that

or things like "oh we, want to track this for owners but not non-owners" etc

or replays etc (not in iris though)

[halcyon ore]

This would probably be hellish simpler, if the stupid custom rep condition type. supported individual players, rather then affecting all players.
Cuz I could turn off rep, batch send items, turn on rep.

[nova wasp]

that's just praying it doesn't go above the buffer though in one frame

it's honestly surprising how awful it is to manage bandwidth on a per-thing basis even in iris though

weird oversite where the idea is to basically have priorities and then a 1d global buffer of send rate

there's not really a way I am aware of to have like a "budget" per-thing

I guess actor channel cheese could help though... honestly worth considering but it's a dead end if you plan to use future versions that might deprecate it

but that could be a loooong time in the future

[halcyon ore]

Apply damage ain’t removed yet. (Wasn’t that planned forever ago?)

Well.
As the last like 3 times. Thanks for the help/ suggestions.
I’ll need to mess with them tomorrow.

[nova wasp]

yeah I think it's fair to assume ye olden replication will be around for a while yet

I wish I had a fancy nicer answer to all this stuff but I have mainly dealt with networking stuff on a more individual realtime thingy bassis than huge buffers of potentially hierarchical things that need to be segmented... I hope someone reads my yapping and has some nicer ideas

would be worth looking around at other games and how they do inventory setups or things like gdc talks etc, even if it's not really simple to do in unreal there might be some inspiration

[frosty harbor]

this might be multiplayer related I'm not entirely sure but yea. I'm trying to make the animation stop the moment it detects a hit on a player by changing the playrate to 0, then reverting it after 0.1-0.2s. (I made it 10s to see what i was doing wrong in the screenshot). Essentially what action games call a "hit stop"

The issue is that I think this is not applying on the server at all and just on the owning client (local prediction) when in reality it should be applying on both. The reason why i think this way is that even though I put it to revert to the normal play rate only after 10s, it happens near instantly which makes me think that it's the server "correcting" the "stopped" animation by going past the local client's location & correcting the playrate.

Any idea what im doing wrong and how to improve this without making the GAS Ability replicate so I can use different rpcs inside it? (Which I think isn't what Epic games thinks is the right way to do things).

[halcyon ore]

Yeah
That’s where I got like the entire thing from.
The items variable, and Data types, and my current manual/ cursed method.
And it works fine, just takes a while to make.
But you guys talking about Fast Array, and replicated sub objects made me think this was some new tech that would’ve helped a ton. lol

[meager spade]

why would you set playrate to 0

[meager spade]

i personally would do loot tables that are kinda pre-picked items for chests, etc. then client just needs to know what loot table that chest items comes from, and does it all locally

[frosty harbor]

Hit stop effect when hitting a player. (Gives "weight" to the strike)

[meager spade]

yeah but you shouldnt need to do this

your animbp can handle this

[frosty harbor]

Hm

I assume you mean freeze pose when I flip a bool?

[thin stratus]

I mean, that's just the specifics of how you'd stop/pause/freeze the anim locally. Don't think that would resolve your issue, or?

[frosty harbor]

I haven't tested it yet but dont know if it will or not

[thin stratus]

Not sure why the server would do any corrections though. Montages aren't replicated by default.

[frosty harbor]

well i was assuming it was the server

maybe it isnt

but if it isnt then i have no clue why it doesnt work

[thin stratus]

Why assuming? Can't you print or breakpoint?

There might also be something about 0.0 playrate . Have you checked if there isn't just a Pause function instead?

[frosty harbor]

i mean idk what to print when the animaiton just continues and teleports around

it really looks like its catching up its why I think it was hte server correcting

but ill do some more tests

[frosty harbor]

Yep this was the issue

Pause function worked perfectly

playrate to 0 was causing it to skip ahead when resuming the playrate or something similar

[nova wasp]

this is what I suggested earlier with data assets or the like but apparently strings and dynamic data per element are required

[meager spade]

oh

like instanced structs?

[nova wasp]

that would help but it would mean extra data for all elements for the struct type

great for one-offs though

[gentle mauve]

Is it possible to change View Modes when playing as a Client in the editor? viewmode unlit does nothing

[halcyon ore]

If only this were these. 😛
these are gathered items all owned by the player. Just, thrown in boxes

[nova wasp]

I guess the point is if you can define the range of known types of things you can derive some of the information on the other side

but as we talked about earlier it's kinda arbitrary for some parts

[royal vault]

The solution was adding a check:

NetConnection->IsNetReady()

[north jacinth]

What would be the preferred pattern for handling interactions between pawns while using NPP?

[grand kestrel]

What do you define as an interaction, a collision?
-# I haven't used NPP and I wrote a GAS based solution so probably not gonna be helpful

[north jacinth]

attack resolution, essentially

[north jacinth]

GAS unfortunately would not work here, although it's great for what it does

[nova wasp]

show how you obtain the ping

use GetPingInMilliseconds

the player state

which version is this on?

in the future if you are on a really old version do not leave that info out

is bShouldUpdateReplicatedPing on?

well, it's kinda a running average if you read the source as well

is the ping average being updated in PingBucketV2 on clients?

or is that only on the server

no clue then... I am just browsing the 4.27 source on github

I recall there being an option to have ping include frametime in some versions but idk if that applies to 4

that doesn't seem very bad to me especially if it's unreliable

it would be a trivially small rpc

you could arguably include it with (well,inside) another commonly sent rpc to save even more

I guess my assumption is this is an rpc to the server from the client to just update some replicated value that's separate

[zealous wigeon]

persistent static data: postgres (sql)
persistent dynamic data: scylla (cql)
warm memory: redis (faster than postgres/scylla queries)
hot memory: the server's ram (latency sensitive)

@slate phoenix ^

[slate phoenix]

Thanks! I'm currently using Node.SJ for my MySQL API backend, and I'll take a look at what you wrote. Thank you 🙂

[north jacinth]

personally when I did that kind of thing I made it relatively frequent but unreliable. you can also adjust the frequency based on how stable the connection is

[thin stratus]

You can send unreliable RPCs on Tick just fine.

[haughty ingot]

No, you have to be worried about make believe performance issues you made up in your head.

[zealous wigeon]

but the little bird whispered to me in my dreams that every actor should always receive every replication

[thin stratus]

But, but, what about the children?!

[thin stratus]

https://tenor.com/view/anakin-star-wars-not-just-the-men-but-the-women-and-children-too-not-just-the-men-women-and-children-too-gif-17266839

[north jacinth]

Speaking of which, what’s the baseline bandwidth use for 1 player + 1 pawn + other game framework stuff? Like, not replicating movement, no extra RPCs or replicated properties other than the minimum to inform clients that it exists

[dark edge]

Amortizes to 0 I'd suppose.

[iron swan]

When I switch over to the server via RPC it seems a lot of my setup scripts for my vars and components haven't run on the server. I'm seeing invalid refs and triggering my invalid errors when I'm running within an RPC's impl (even when the var is replicated).
Is my understanding accurate?

This is within a PlayerController, for context

[thin stratus]

All I can say is that this can totally happen. It highly depends on when you call that RPC in relation to what variables you expect to be valid.

[royal vault]

Hi anyone know how to move a non-controlled character from a client?

context: controller moves character, which rides a mount. character forwards the movement to the mount- but movement-component doesnt know to use client-server movement.

[crisp shard]

uhhhhhh , i have two rep notifies (2 doors) in an actor and it seems that when a later joiner / non-relevant player gets in relevancy range, only 1 of these 2 variables get updated to the client. is this normal behavior?

[silent valley]

No

[crisp shard]

yea that's what i thought... not sure why this would be the case

[silent valley]

Likely causes might be, modifying value on client, different rep conditions, etc

[crisp shard]

there's currently no predictive action with the variable and they have same rep conditions, and server has correct value, and the client that initiated the opening sees the correct state, but it's other clients that will only get 1 of the two doors updated. not a clue rn what it could be

coudl the net update frequency have anythign to do with it? i have it set quite low in this case

[silent valley]

In my experience, when an actor becomes relevant, all the replicated variables are sent in the initial bunch. So idk what is happening for you. Best bet might be to add a c++ onrep and try and debug it as best you can.

[crisp shard]

geniunely, what would making the onrep in c++ do for me ?

[nova wasp]

Is one of them the default value?

[crisp shard]

yea, it's a bool (open or closed), and the door that isn't updated is just false aka closed in the late joiner's case

[nova wasp]

I guess my question here is why would it need to be sent if the value didn't change from the default on the client?

this is more of a rhetorical question but the main takeaway here is that you should get used to manually calling things that onreps call to make sure they get called when you actually want them to (especially in C++ on the authority as otherwise nothing would call it)

[silent valley]

So you can set a breakpoint and look up the callstack to figure out what caused this chain of events. Very useful method for figuring out unreal.

[crisp shard]

maybe i didn't understand the question, but what is happening is: 1 client opens both doors, 2nd client joins later, and walks over to the building, and only 1 of these 2 doors is opening , the other stays closed, but the client can walk right through it, beucase it's actually opened on ther server (as in, it has the correct value, but it isn't sending that value to the client).

i may have discovered the issue as i typed this

my doors open via a timeline

it's possible that the timeline is only diring for 1

[nova wasp]

this does not really help to decipher the absense of a call and they would have to get into serialization internals... this is a very indirect codepath that I doubt will be obvious from a callstack

[crisp shard]

so it doesn't actually move the door

that has to be it

[crisp shard]

in this case, would i call the events on like begin play ? or where would be the best place to call an event like that , not sure if there's an event that fires when it becomes relevant, other than begin play as far as im aware

[crisp shard]

my assumption here is, the timeline can't be called twice at 1 time.. but still confungles me as the client can still walk through the door, so it's "opened" but visually, it's not moved

[nova wasp]

relevancy is more of the actor literally being created from scratch in terms of what happens locally and I'm not even sure of a way to determine if the initial bunch was from relevancy distance or just it spawning as usual

[nova wasp]

use chaos visual debugger to see physics state of all worlds if you can't see the listen server's state (dedicated etc)

or maybe the outliner if is bieng cooperative in this case

If there's a nice "hey this was from relevancy and the actor has existed for a while actually on my end" I am not aware of it unless there's some timestamp I am unaware of

[crisp shard]

that would be nice, i've always considered begin play a bit like the actor's "on rep" as it does fire everytime a client gets into relevnacy again , but prob not the solution for my case. going to test to see if it is the timeline thing real quick

[nova wasp]

another thing can be incrementing things to indicate a change

if you find it's like missing fast-changing values between true/false

[crisp shard]

i havent fully proved it yet, but im 95p sure it's from the timeline, which is slightly annoying to figure out how to accomodate, cause like you mentioned, it would be nice to be able to determine if the variable change is coming from the client instigating it in realtime, vs a late joiner/relevnacy thing, cause if the latter, id' just set the transform and ignore the timeline completely

the timeline is really just so it looks semi decent instead of just teleporting to the opened state

i could make 2 timelines... lol

[nova wasp]

well, if the door opening was represented as the time it opened that would solve it I guess

it's arguably more data than you might want to represent something so simple but I would defend it as you aren't going to be opening 200 doors a frame

it can also be ground down to seconds etc and be VERY small

by this I mean you can replicate a timestamp for when the door last opened/changed etc

[crisp shard]

could i make a a seperate local bool that says if the doors are intialized on the client and if so , it can play the timeline, if not, just set them to whatever they need to be set at ? that might be similar approach to the timestamp

[nova wasp]

yes, that's not a bad idea either

this way the client can track what the door actually did on their side visually separately and they don't need to rely on only the replicated value I guess

very common

the timestamp I guess is more if the idea of it being old or recent is meaningful (or even getting really fancy with accurate timing... you probably just want the most recent state here though) no matter what but if a local bool does what you want that's plenty

[crisp shard]

nice that was the issue, and the local bool did fix it

good to know, that multiple things being updated at once using the same timeline will only visually be repreneted by the last changed variable that uses it

could also maybe make this a bit cleaner by simply never doing the timeline call in the on rep at all, teleport the door on server via the onrep (set it's actual state immediately), and have a multicast do the anim , so that way, it always fires and updates the correct position, and doesn't have to worry about if it's intilizaed or not for clients, but will playt he anim if you have the relevancy to see it. only issue w that is you might have a colliison issue with opening/closing but my door anim is quite fast either way, so prob not an issue

[nova wasp]

one potential issue is the animation playing when it pops-in but this might not matter in practice

[crisp shard]

yea main concern was just making sure door's state is appears as it actually is, but i've opt'd to use the bool, and have a small delay on begin play to set it to true, giving time for the doors to update. would have been a bit more of a lift to change it w the multicast given my current setup and i have a lot of diffeetn buildings w diffeten setups and diffetne numbers of doors etc.

[whole pollen]

Hey guys I want to learn Multiplayer, keeping in mind that I am not new to unreal engine. can you guys suggest some good learning material on youtube or any other place?

Would appreciate your help

[lament flax]

check pins

also https://notes.hzfishy.fr/Unreal-Engine/Networking/_Resources-(Networking)

[whole pollen]

Sure, Thanks Fishy

Thanks man

[crisp shard]

wanted to update, i've removed the bool and am now just using the timeline itself as a check to see if it is playing, and if it is (another door is being opened at the exact same time), then just set it's new transform instead of doing it via timeline. the bool had weird issues but this seems to work in all extreme scenarios

[slate phoenix]

Hello everyone,
I’m developing a game, but I’m stuck at this point.

Players will be able to create their own worlds or visit other players’ worlds. There will be an ownership system for worlds, and the number of worlds will be limited. For example, a size like 20x20x20, though the exact dimensions are not very important for now.

Players will progress and farm in their own worlds. However, what confuses me is this:
If there is only one dedicated server, do I need to run each world as a separate server? How should I handle this?

I’ve been thinking and researching this for about three days, but I couldn’t find a clear answer. Is there anyone who can explain this topic to me?

It will be a dedicated server, not a P2P game.
It will be something like an MMO.

[slate phoenix]

By “separate server,” I mean this:

For example, the main server:
192.168.1.1:8888 (I’m just using a random IP and port as an example)

When Player A creates a world → 192.168.1.1:8889

When Player B creates a world → 192.168.1.1:8890

So, do I need to run a separate server instance for each world on the same IP but different ports?

How to

ue need different world ?

Whenever a different world is created, the corresponding level map should be opened and the player should connect to the server.
I created a system like this, but it sends the player back to the world creation or world selection screen.

If I use Open Level, it works, but in this case the level is opened locally.
The player can see the main server and connects to the backend API; everything works fine.

However, when a second player enters the same world, they cannot see each other.
Even so, the actions performed are correctly saved through the backend API.

[slate phoenix]

If there's an easier way to manage it, don't hesitate to share.

[crisp shard]

I’d argue almost definitely you’d want to have a separate server per world unless there’s some sort of super server you’re making. Also having many worlds under 1 server assumingely wouldn’t run very well but again maybe there’s some super server you are aiming to have handle all of it

[slate phoenix]

For example, if my game is played by 10,000 people in the future, I’m afraid whether this port-based system will still make sense.
I’ve been thinking about this for 3 days, because once the game is released, there might be no turning back.

[slate phoenix]

I’m thinking about using a single server with a World Manager and Replication Filtering.
This might be more reasonable and also better in terms of cost, since I’m developing the game alone.

[thin stratus]

Single cloud server or single game server?

[thin stratus]

To generally answer this.

If your game needs DedicatedServers, then you need one (Unreal Engine Game) DedicatedServer per player world. And that DedicatedServer can usually handle a maximum of 100 players (with a lot of optimizations) before it will struggle with performance (more realistic is something around 60-70).

Running multiple DedicatedServers on one CloudServer is possible but probably limited to a handful before the CloudServer has to be too powerful. So if you plan for 10.000 players, and you'd expect most players to be in their own world while some are visited by a 2-4 other players, then you would need quite a lot of DedicatedServers and CloudServers.

If your game doesn't need any DedicatedServers and can work with ListenServers, then I would go with that instead.

DedicatedServers on the same CloudServer would share the IP, but have an incrementing port (starting at 7777). Individual CloudServers have their own IP.

Players, no matter if its their own World or not, would need to connect to that IP:Port via open <ip:port>.

For ListenServers, the hosting player would need to use OpenLevel <MapName>?listen, while the joining one would need to connect to their IP:Port. That usually requires opened ports on the player's end, which most people get around by using Steam or similar for the Session Management. That then also comes with the benefit of being able to join those Sessions and not having to share IP addresses.

[daring gorge]

im using a replicated variable to set the rotation of a turret to sync w where the player is looking at through the anim BP but when i spawn projectile and muzzle flash it spawns at the initial position of the guns barrel rather than where it has been rotated to which ofcourse looks weird, im also unable to use GetSocketRotation and get where the socket is facing, it just shoots forward as if the gun never rotated to begin with, i know anim BP isnt replicated but isnt the turret being rotated on all clients since its using a replicated variable to do that and doesnt that mean the socket should face the same direction in all instances?

[thin stratus]

If the Turret is rotated on everyone, then it should, in theory, also have the Socket positioned correctly.

Are you spawning the effect locally, or is it replicated?

Because the one thing I could imagine is that you are using a SkeletalMesh for the Turret and you don't have the Visibility Based Tick Option (or whatever it is called) on the Component set to "Always Tick and Refresh Bones".

[daring gorge]

it is rotated on all clients and server properly but when i use direct SocketRotation it just shoots forward as if it wasnt and when i use the replicated rotation it fires in the camera look at direction but the fx stays in the init socket position just rotated towards the direction

[thin stratus]

Which means if the Server isn't looking at it, it wouldn't update.

[daring gorge]

atm just spawning it on multicast but ill be changing that later to local spawning

[thin stratus]

And the multicast has the spawn transform as a param?

[daring gorge]

yes

[thin stratus]

Yeah, then try what I wrote above.

[daring gorge]

i also dont see this, im using 4.26

[thin stratus]

That should also have it

You can also test this by just making sure the server is looking at the turret.

[daring gorge]

is it this

[thin stratus]

If that still fails to spawn it correctly, then not sure.

[thin stratus]

Yeah

Maybe the socket isn't parented to the right bone?

[daring gorge]

i couldnt tell, ill try spawning w the server and looking at the turret once

still the same, this is the server looking at the turret

basically the turret has a replicatedplayer variable which is replicated and the player has a Rotation variable that's replicated, the same used for the TPS aim offset and such, also used when player is driving vehicle, the same rotation is being taken and promoted to another variable in turret and the anim bp uses that to rotate, so it perfectly rotates on all clients as it should, the socket is also parented to the right bone

wait no it isnt

no way

wow, im really sorry this is so dumb it wasnt parented right

[nova wasp]

kudos for posting a visual example to show context as it helps a lot

[thin stratus]

All good. Glad it was that easy :D

[daring gorge]

thanks guys

[nova wasp]

(I didnt do anything lol, I was just reading along)

[daring gorge]

also another question, how would you deal with high ping and input delays due to it? do you just do the local fx spawn and not care or are there buffers to handle this

i havent profiled my game but im worried about each shot being a seperate RPC and the listen server ping causing input delays

[nova wasp]

typical handling is to just simulate the projectiles locally and send results of what they do I guess back to the client
if you want to get a bit fancier the idea would be to locally predict a projectile and replace it with the "real" one from the server once it arrives seamlessly

if it is slow enough to justify that (a rocket or something etc in unreal tournament)

fairly open-ended problem that can kinda depend a lot on the perspective and how important it is to not miss things

for sim proxies some games use a replicated value that just incremenets if they are firing or not https://jambax.co.uk/better-burst-counters/

as a general rule of thumb it is important to understand that replication timing is entirely vibes base and will rarely look smooth when interpreted with no help to make it smoother

there are multiple built in-things for client prediction but they tend to not really be for this specific problem (GAS, cmc, network prediction/mover,newer chaos replication) as much as movement of physics bodies and characters

[nova wasp]

A weird tip about PIE network testing: you can actually have tickrates be different on different "machines" (well, PIE clients which are just ticked one after the other)

this is useful because you can see vastly different rpc/rep behaviour when you have a low vs higher tickrate on either side. Lower framerates will mean a longer time between consuming packets and producing them etc

So for example a client with a really bad framerate but decent internet might send the server a very sparse amount of inputs relative to one with a better framerate

Which is fine for many games but might cause jitter in others if they rely on a lot of info from that client etc

[royal vault]

for anyone else: look into the mover component !

[daring gorge]

this is amazing, thanks a lot!!!! i really appreciate it

[slate phoenix]

Let’s assume my game has 1000 concurrent players, and everyone can create their own world.
That means 1000 worlds.

1 world ≈ 400 MB RAM

Or if a world has 64 players ≈ 1.5 GB RAM

Because of this, the classic one world = one port / one Dedicated Server approach would consume too much RAM.

So I calculated a few scenarios:

Available RAM

Total RAM: 32 GB

Reserved for OS and other services: ~4 GB

Usable RAM: ~28 GB

Scenario 1 – Everyone plays alone

28 GB ÷ 30 MB (world + player)

≈ 900 players

Scenario 2 – Average 5 players per world

200 worlds × 50 MB = 10 GB

1000 players × 15 MB = 15 GB

Total ≈ 25 GB → ~1000 players ✅

Scenario 3 – Average 10 players per world

100 worlds × 100 MB = 10 GB

1000 players × 15 MB = 15 GB

Total ≈ 25 GB → ~1000 players ✅

The number of players in worlds will never be stable,
because players will constantly farm, trade, and visit other worlds.

Because of that, I’m considering a Single Server + World Manager architecture:

All worlds are handled inside a single server process

Actors are replicated only to relevant players

A Replication Filtering System is used to reduce unnecessary replication and RAM usage

I believe this approach will use significantly less RAM compared to running one Dedicated Server per world.

There may be 500 concurrent players, but the database can contain 2000 worlds.

Additionally, Listen Server is not an option.
The hosting player would be the server, and when that player disconnects, everything would be lost.
In other words, if the world owner leaves, all other players in that world would be disconnected as well.

Because of this, that approach is not viable.

[dark parcel]

but networking framework that comes with Unreal doesn't support multi world.

you were told this in the past.

A server can host a world and all client must connect to the same world.
Any world that client create will only exist on that specific machine and will not be known to server and other client.

The path of least resistance to have multi world is by using dedicated server to host each world.
A travel between world is simply joining a host with a specific map the player want to travel to.

[thin stratus]

You need to write that from scratch then.

As ColdSummer wrote, UE doesn't support multi word stuff and more than 60-100 players per UEServer isn't gonna run well. It's not even RAM related. The CPU is the bottleneck. And potentially even the bandwidth.

You do have a different definition of world I assume but it comes down to the same problem.

UEs Server setup is from a time where arena shooters were popular

Unreal tournament specifically

Meaning they are coded for short rounds with maybe 32 to 64 players max, and only the connected players see each other. 2 or more dedicated servers usually don't communicate with each other.

[humble idol]

From what I can gather, I'm counting about 3 layers of technological (networking) limitations that either aren't native or supported by UE. An ambitious AA studio will carefully consider projects that require even ONE additional layer of technological extension to the engine.

[thin stratus]

Yeah. This basically needs a custom setup, probably without even using native dedicated servers and handling replication outside of UEs native code.

[slate phoenix]

There will be no multiple level maps.
There will be only one level map, for example called M_World.

When a player creates a new world, the game will create that world using the M_World level as a template.
For example, Player A creates a world called “sohbet”, which is instantiated from M_World.

When Player B types and searches for the “sohbet” world, the database already contains that world, so the player can simply enter the world and visit it, because the world already exists.

For worlds that do not exist, the game will always use the M_World level to create a new one.

[slate phoenix]

I’ve already implemented some parts of the system, and as development progresses, I’ll share a video with you all to show the improvements. ❤️

[thin stratus]

Keep in mind that the term "World" has a meaning in UE

And almost everyone here will take that meaning over whatever you might be talking about.

[slate phoenix]

You're right, I needed to explain this.

[lament flax]

Seems like what you want is 1 dedicated server per player playing in his "world"

Which is insane

So yeah, lots of custom code and hardware setup to have it working

[latent heart]

Even if it's a "template", standard UE is still gonna have each of those levels need to be their own server (listen or dedicated) unless you write some sort of custom replication thing.

Maybe you implement phasing like mmos do these days.

[slate phoenix]

Yes, I tried it, but it didn’t work.
Either I’ll have to run separate servers by assigning a different port for each opened level, or I’ll need to buy a plugin like this.
https://www.fab.com/tr/listings/006026b4-9a0b-4f44-a918-a02f7241e997
The 6-hour process went to waste 😄

[dark parcel]

No, this plugin will not help with making your mmo.

read into the description, the secondary world doesn't support replication.

only the main world, so it's just business as usual in multiplayer context but client get to make their own world (which only exist on that specific client).

It's nice plugin to display pocket world like showing character or items to the local player.

[slate phoenix]

So how do I do it?

[latent heart]

I think you'd probably be best off doing what Cedric said. Or maybe just what I'm picking up from him: use UE just as a client. And maybe for connection. Have a totally separate system for the rest of the networking.

[dark parcel]

do you have the technical expertise and time? Multiplayer that comes with UE doesn't really fit for a server to support minecraft games for many players or multiworld.

The most realistic approach and one where you don't need to build sky scrapper is by hosting dedicated server per world.

using AWS fleet and serverless API calls, you can pretty much have the server on cloud. Hosting one per request and destroy it when there are no player left in the world.

[slate phoenix]

you said WORLD_A → Server 1 (Port 7777)
WORLD_B → Server 2 (Port 7778)
WORLD_C → Server 3 (Port 7779)
Do ?