Godot Rollback Netcode

This is an addon for implementing rollback and prediction netcode in the Godot game engine.

Beyond the basics (gathering input, saving/loading state, sending messages, detecting mismatches, etc) this library aims to provide support for many of the other aspects of implementing rollback in a real game, including timers, animation, random number generation, and sound - along with high-quality debugging tools to make solving problems easier.

Implementing rollback and prediction is HARD, and so every little bit of help is important. :-)

For C# support, please install the GodotRollbackNetcodeMono addon in addition to this addon. The Mono addon is separately maintained, and it adds C# wrappers and base classes to make easier to work with the GDScript parts of this addon.

Tutorials

I'm working on a series of video tutorials on YouTube - here's the playlist, and these are the parts that are published as of the last time this README was updated:

Installing

This addon is implemented as an editor plugin.

If you've never installed a plugin before, please see the official docs on how to install plugins.

However, the short version is:

Copy the addons/godot-rollback-netcode directory from this project into your Godot project at the exact same path. The easiest way to do this is in the AssetLib right in the Godot editor - search for "Godot Rollback Netcode".
Enable the plugin by clicking Project -> Project settings..., going to the "Plugins" tab, and clicking the "Enable" checkbox next to "Godot Rollback Netcode".

Games using this addon

If you release a game using this addon, please make an MR (Merge Request) to add it to the list!

Overview

This is a quick overview of the different pieces that the addon includes.

Singletons

res://addons/godot-rollback-netcode/SyncManager.gd: This is the core of the addon. It will be added to your project automatically when you enable the plugin. It must be named SyncManager for everything to work correctly.
res://addons/godot-rollback-netcode/SyncDebugger.gd: Adding this singleton will cause more debug messages to be printed to the console (and captured in the normal Godot logs) and make a debug overlay available. By default, the overlay can be shown by pressing F11, but you can assign any input event to the "sync_debug" action in the Input Map in your project's settings.
res://addons/godot-rollback-netcode/SyncReplay.gd: Adding this singleton will allow you to replay matches from log files, using the "Log Inspector" tool that is added to the Godot editor. See the "Setting up replay" sub-section below for more information.

Important properties, methods and signals on `SyncManager`

The SyncManager singleton is the core of this addon, and one of the primary ways that your game will interact with the addon. (The other primary way is via virtual methods that you'll implement in scripts on your nodes - see the section called "Virtual methods" below for more information.)

Properties:

current_tick: int: The current tick that we are executing. This will update during rollback to be the tick that is presently being re-executed.
input_tick: int: The tick we are currently gathering local input for. If there is an input delay configured in Project Settings, this will be ahead of current_tick by the number of frames of input delay. This doesn't change during rollback.
started: bool: will be true if synchronization has started; otherwise it'll be false. This property is read-only - you should call the start() or stop() methods to start or stop synchronizing.

Methods:

add_peer(peer_id: int) -> void: Adds a peer using its ID within Godot's High-Level Multiplayer API. Once a peer is added, the SyncManager will start pinging it right away. All peers should be added before calling SyncManager.start().
start() -> void: Starts synchronizing! This should only be called on the "host" (the peer with id 1), which will tell all the other clients to start as well. It's after calling this that the "Virtual methods" described below will start getting called.
stop() -> void: Stops synchronizing. If called on the "host" (the peer with id 1) it will tell all the clients to stop as well.
clear_peers() -> void: Clears the list of peers.
start_logging(log_file_path: String, match_info: Dictionary = {}) -> void: Starts logging detailed information about the current match to the given log file. The common convention is to put the log files under "user://detailed_logs/". The match_info is stored at the start of the log, and is used when loading a replay of the match. This method should be called before SyncManager.start() or in response to the "sync_started" signal.
stop_logging() -> void: Stops logging. This method should be called after SyncManager.stop() or in response to the "sync_stopped" signal.
spawn(name: String, parent: Node, scene: PackedScene, data: Dictionary = {}, rename: bool = true, signal_name: String = '') -> Node: Spawns a scene and makes a "spawn record" in state so that it can be de-spawned or re-spawned as the result of a rollback.

It returns the top-level node that was spawned, however, rather than doing most setup on the returned node, you should do it in response to the "scene_spawned" signal. This is because the scene could be re-spawned due to a rollback, and you probably want all the same setup to happen then as when it was originally spawned. (Note: there are rare cases when you do want to do setup only when spawned initially, and not when re-spawned.)
- name: The base name to use for the top-level node that is spawned.
- parent: The parent node the spawned scene will be added to.
- scene: The scene to spawn.
- data: Data that will be passed to _network_spawn_preprocess() and _network_spawn() on the top-level node. See the "Virtual methods" described below for more information.
- rename: If true, the actual name of the top-level node that is spawned will have an incrementing integer appended to it. If false, it'll try to use the name but this could lead to conflicts. Only set to false if you know for sure that no other sibling node will use that name.
- signal_name: If provided, this is the name that'll be passed to the "scene_spawned" signal; otherwise the name will be used.
despawn(node: Node) -> void: De-spawns a node that was previously spawned via SyncManager.spawn(). It calls _network_despawn() and removes the "spawn record" in state. By default, this will also remove the node from its parent and call node.queue_free(). However, if you have enabled "Reuse despawned nodes" in Project Settings, then the node will be saved and reused when the same scene needs to be spawned later. This makes it especially important to clean-up the nodes internal state in _network_despawn() so that the node is "like new" when reused.
play_sound(identifier: String, sound: AudioStream, info: Dictionary = {}) -> void: Plays a sound and records that we played this specific sound on the current tick, so that we won't play it again if we re-execute the same tick again due to a rollback.
- identifier: A unique identifier for the sound. Only one sound with this identifier will be played on the current tick. The common convention is to use the node path of the node playing the sound, with some sort of "tag" appended, for example:
```
SyncManager.play_sound(str(get_path()) + ':shoot', shoot_sound)
```
- sound: The sound resource to play.
- info: A set of optional parameters, including:
  - position: A Vector2 giving the position the sound should originate from. If omitted, positional audio won't be used.
  - volume_db: A float giving the volume in decibels.
  - pitch_scale: A float to scale the pitch.
  - bus: The name of the bus to play the sound to. If none is given, the default bus configured in Project Settings will be used.

Signals:

sync_started (): Emitted when synchronization has started, as a result of calling SyncManager.start() on the "host". Even though SyncManager.start() is only called on the "host", this signal will be emitted on all peers.
sync_stopped (): Emitted when synchronization has stopped for any reason - it could be due to an error (in which case "sync_error" will have been emitted before this signal) or SyncManager.stop() being called locally or on the "host".
sync_lost (): Emitted when this client has gone far enough out of sync with the other clients that we need to pause for a period of time and attempt to regain synchronization. Your game should show some indication (a message or loading icon) to the player so that they know why the match has suddenly come to a stop.
sync_regained (): Emitted if we've managed to regain sync after it had been lost. The message or icon your game showed to the player when "sync_lost" was emitted should be removed.
sync_error (msg: String): Emitted when a fatal synchronization error has occurred and the match cannot continue. This could be for a number of reasons, which will be identified in a human-readable message in msg.
scene_spawned (name: String, spawned_node: Node, scene: PackedScene, data: Dictionary): Emitted when a scene is spawned via SyncManager.spawn() or re-spawned due to a rollback. Connect to this signal when you want to do some setup on a scene that was spawned, and you need to ensure that that setup also happens if the scene is re-spawned during rollback (you want to do this most of the time).
scene_despawned (name: String, node: Node): Emitted when a scene is despawned via SyncManager.despawn() or due to a rollback. Connect to this signal when you want to do some cleanup on a scene that was despawned.
interpolation_frame (): If interpolation is enabled in Project Settings, the work of the SyncManager will be split between "tick frames" (where input is gathered, rollbacks are performed and ticks are executed) and a variable number of "interpolation frames" that happen between them. This signal is emitted at the end of each interpolation frame, so that you can perform some operations during a frame with more time budget to spare (a lot more needs to happen during tick frames).

Node types

This addon include a few rollback-aware node types:

NetworkTimer: A replacement for the Timer node. Unlike Timer, it doesn't wait for a number of seconds, but instead a number of ticks (the wait_ticks property).

If hash_state is set to false, then the timer's state won't be included in the hash used to detect state mismatches. This is useful if a timer may only run on a single client, rather than on all clients.
NetworkAnimationPlayer: Descends from the built-in AnimationPlayer but will only move forward each tick (rather than as time passes) and it supports rollback.

If auto_reset is set to true, it will automatically play the "RESET" animation every time state is loaded without an animation playing. This can help prevent issues where an animation started, but on rollback it's determined that it shouldn't have started, so the animation is left in an in-between state.
NetworkRandomNumberGenerator: For generating random numbers in a deterministic way that supports rollback. At the start of the match, the clients need to initialize the node with the same seed. Each time a random number is generated, it's internal state will change in a deterministic way, such that the sequence of numbers it generates on each client will be the same. When a rollback happens, it's internal state will rollback such that it'll generate the same sequence of numbers again.

One way to avoid having to share a seed for every NetworkRandomNumberGenerator in the game, is to share a single "mother seed" which is used for one NetworkRandomNumberGenerator that generates all the other seeds the game needs. I like to call this the "Johnny Appleseed approach" where "Johnny" is distributing seeds grown from the "mother seed". This will work so long as the nodes are always initialized in a deterministic order!

Virtual methods

For a node to participate in rollback, it must be in the "network_sync" group, which will cause SyncManager to call various virtual methods on the node:

_save_state() -> Dictionary: Returns the current node state. This same state will be passed to _load_state() when performing a rollback.

Warning: Don't put any References, Objects, Arrays or Dictionarys into state, unless you can duplicate them first. This is because changing the object later will change the data in the state buffer too. And NEVER put Nodes in state, instead use the node path, or some other way to locate the right node later, since the original node may no longer even exist. Any Objects put into state will need special support in your HashSerializer (see below) to prevent incorrectly detecting state mismatches.
_load_state(state: Dictionary) -> void: Called to roll the node back to a previous state, which originated from this node's _save_state() method.
_interpolate_state(old_state: Dictionary, new_state: Dictionary, weight: float) -> void: Updates the current state of the node using values interpolated from the old to the new state. This will only be called if "Interpolation" is enabled in project settings.
_get_local_input() -> Dictionary: Returns the local input that this node needs to operate. This will only be called for nodes whose "network master" (set via Node.set_network_master()) matches the peer id of the current client. Not all nodes need input, in fact, most do not. This is used most commonly on the node representing a player. This input will be passed into _network_process().
_predict_remote_input(previous_input: Dictionary, ticks_since_real_input: int) -> Dictionary: Returns predicted remote input based on the input from the previous tick, which may itself be predicted. This will only be called for nodes whose "network master" DOES NOT match the peer id of the current client. If this method isn't provided, the same input from the last tick will be used as-is. This input will be passed into _network_process() when using predicted input. If ticks_since_real_input is negative, we haven't received any remote inputs yet, which happens at the very start of the game.
_network_process(input: Dictionary) -> void: Processes this node for the current tick. The input will contain data from either _get_local_input() (if it's real user input) or _predict_remote_input() (if it's predicted). If this node doesn't implement those methods, it'll always be empty.

This method is meant to do what _process() or _physics_process() would have done in most other Godot games.
_network_preprocess(input: Dictionary) -> void: Works just like _network_process() but all the _network_preprocess() methods are run before all the _network_process() methods.
_network_postprocess(input: Dictionary) -> void: Works just like _network_process() but all the _network_postprocess() methods are run after all the _network_process() methods.

The following methods are only called on scenes that are spawned/de-spawned using SyncManager.spawn() and SyncManager.despawn():

_network_spawn_preprocess(data: Dictionary) -> Dictionary: Pre-processes the data passed to SyncManager.spawn() before it gets passed to _network_spawn(). The modified data returned by this method is what will get saved in state. This allows passing developer-friendly data to SyncManager.spawn() and this method can convert it into data that is safe to store in state.

For example, you may pass a Node object into SyncManager.spawn() and this method could convert it into a String representing the node path. This is important because storing a Node in state can cause weird problems, because it could continue to change (or even get deleted!) by the time that state needs to be loaded again.
_network_spawn(data: Dictionary) -> void: Called when a scene is spawned by SyncManager.spawn() or in rollback when this node needs to be re-spawned (ex. when we rollback to a tick before this node was de-spawned).
_network_despawn() -> void: Called when a node is de-spawned by SyncManager.despawn() or in rollback when this node needs to be de-spawned (ex. when we rollback to a tick before this node was spawned).

Project settings

The recommended way to configure SyncManager is via Project Settings (although, you can change its properties in code at runtime as well).

You can find its project settings under Network -> Rollback, after the plugin is enabled.

Max Buffer Size: The number of state and input frames to keep in the buffer. This defines the maximum number of ticks that the game can rollback.
Ticks To Calculate Advantage: The number of ticks that we collect advantage calculations from each peer before calculating an average, and possibly skipping some frames so that other clients can catch up.
Input Delay: The number of frames of input delay.
Ping Frequency: The number of seconds between each ping.
Interpolation: If enabled, we'll do 1 tick of interpolation. This allows the simulation frequency to be set lower than the rendering FPS, while still rendering smoothly.

Limits:

Max Input Frames Per Message: The maximum number of input frames that will be sent in a single message. If there are more input frames, then multiple messages will be sent.
Max Messages At Once: The maximum messages that will be sent to one peer at a time. If there are more messages, then the middle ones will be omitted (so the newest and oldest input frames will be sent).
Max Ticks To Regain Sync: If the clients get out of sync, and they don't manage to regain sync after this many ticks have been skipped, then SyncManager will emit "sync_error" and kill the match.
Min Lag To Regain Sync: If we've lost sync due to an input buffer underrun, then SyncManager won't start running again until the minimum lag with all clients is above this value. This is to prevent the game immediately losing sync right after it has regained it, meaning it will pause a slightly longer time, to avoid a series of smaller pauses.
Max State Mismatch Count: If more than this many state mismatches are detected in a row, it's considered a fatal state mismatch, and SyncManager will emit "sync_error" and kill the match.

Spawn Manager:

Reuse Despawned Nodes: If enabled, de-spawned nodes will be reused rather than instancing the scene every time. This is can provide a large performance boost in games that do a lot of spawning, but it requires the developer to carefully reset state during _network_despawn().

Sound Manager:

Default Sound Bus: The name of the default sound bus to use when playing sounds via SyncManager.play_sound(). If omitted, then "Master" will be used.

Classes:

These are all paths to classes that override the default implementations of the "Adaptor classes" described in the section of the same name below.

If any are omitted, then the default implementation will be used.

Debug:

WARNING: Do not keep these settings enabled in the release version of your game!

Rollback Ticks: If set, every tick, the game will rollback at least the number of ticks given here. This is a great setting to have enabled when implementing new game logic, as some problems won't be apparent until a series of rollbacks occur.
Random Rollback Ticks: If set, every tick, the game will rollback a random number of ticks up to the given value. This can help find bugs that only appear when two clients rollback a different number of ticks when executing the same logic. If "Rollback Ticks" is also set, it will effectively set a minimum value.
Message Bytes: If any message exceeds the given number of bytes, an error will be pushed to the editor and console, including the size of the offending message.
Skip Nth Message: If set, every nth message will not be sent. This is a quick way to simulate packet loss, but it isn't very accurate. System-level tools or a proxy that simulates packet loss will always make for more accurate testing.
Physics Process Msecs: If SyncManager._physics_process() takes more than this many milliseconds, then an error will be pushed to the editor and console, including how long it actually took.
Process Msecs: If SyncManager._process() takes more than this many milliseconds, then an error will be push to the editor and console, including how long it actually took.

Log Inspector:

Replay Match Scene Path: The path to the scene to load when launching the game as a replay client.
Replay Match Scene Method: The method on the above scene to call in order to setup the match to run as a replay. While this method can be named anything, it must accept the following arguments (in this order):
- my_peer_id: int: The peer id that the replay client will be displaying.
- peer_ids: Array: An array of the other peer ids in the match.
- match_info: Dictionary: The match_info Dictionary that was passed to SyncManager.start_logging() when creating the log that we are replaying from. This should be used to initialize the match to the same state as the match that was logged.
Replay Arguments: The arguments to pass to the game on the command-line when launching it as a replay client.
Replay Port: The TCP port used by the Godot editor to communicate with the replay client.

Other important settings not under "Rollback":

Physics -> Common -> Physics FPS: Since SyncManager uses _physics_process() to run tick frames, this controls the simulation frequency. If you're using interpolation, it's best if this value can evenly divide the rendering FPS, for example, if rendering at 60 fps, then 60, 30, 20, or 10 are all good values.

Adaptor classes

There are a few adaptor classes that can be used to modify the behavior of SyncManager.

`NetworkAdaptor`

All network communication from SyncManager passes through its NetworkAdaptor. The default implementation uses Godot RPCs.

You can replace the network adaptor with your own class, in order to customize network communications. This could even allow you to avoid using Godot's High-level Multiplayer API in favor of some other underlying communication layer, for example, Steam's peer-to-peer networking.

Parent class: res://addons/godot-rollback-netcode/NetworkAdaptor.gd

Default implementation: res://addons/godot-rollback-network/RPCNetworkAdaptor.gd

Additional implementations:

res://addons/godot-rollback-network/NakamaWebRTCNetworkAdaptor.gd: Integrates with the WebRTC and Nakama addon for Godot.
res://addons/godot-rollback-network/DummyNetworkAdaptor.gd: Allows using the addon without any real networking. It's used by the replay system in Log Inspector, and can also be used to implement an offline mode in your game - see the demo project for an example of how to do this.

`MessageSerializer`

The message serializer will convert input messages to bytes for sending to the other clients.

The default implementation is relatively wasteful (it will likely lead to messages exceeding the MTU) so you will ALMOST ALWAYS want to replace it with your own implementation that can pack the data as small as possible. This can only be done by knowing the structure and meaning of the data, which is why the game developer needs to do it.

Parent class and default implementation: res://addons/godot-rollback-netcode/MessageSerializer.gd

`HashSerializer`

The hash serializer will convert state or input into primitive types so that we can hash the Dictionary for use in comparisons. It is also used to convert back from these primitive types in the "Log inspector" tool.

The default implementation can't handle Objects in a smart way, so if you include any in your input or state, it could lead to SyncManager detecting a false state or input mismatch. Replace this with your own implementation to convert any of your objects into primitive types.

Parent class and default implementation: res://addons/godot-rollback-netcode/HashSerializer.gd

"Log Inspector" tool in Godot editor:

The "Log Inspector" can be opened by clicking Project -> Tools -> Log inspector... in the Godot editor.

It allows you to load the logs (generated by SyncManager.start_logging()) from all the clients in a match and examine the data in detail.

Note: Each client generates its own seperate log file for the match, so you'll need to collect them from the other players, and open ALL of them in the "Log Inspector" together.

At the top there is a view selector which lets you switch between a "Frame" and a "State/Input" view of the data.

Frame viewer:

The "Frame" viewer shows data about each frame executed, including tick frames, interpolation frames and skipped frames. The x-axis is milliseconds since the match began. It uses the system clock of the computer the client ran on, so data may not be line up correctly. You can correct for this by adding or subtracting milliseconds from each peer in the "Settings" dialog.

You can click anywhere on the graph, and it will show all the data logged on the most recent frame for each client in the lower part of the window. This data includes timing information, the number of rollbacks executed, messages received and much more.

If you click the "Previous frame" or "Next frame" buttons, the cursor will move forward or backward to the next frame on any peer. However, if "Seek only on replay peer" is checked, it'll only jump to the next or previous frame on the peer selected in the dropdown in the replay toolbar.

The arrows connect from the tick when input was generated, to the frame when the other peer receives that input. This gives a visual representation of the network traffic between those two peers. The pair of peers can be selected in the "Settings" dialog.

The orange line represents the number of rollbacks performed on a given "tick frame", allowing you to easily see how many mispredictions occurred and during which portions of the match.

State/Input viewer:

The "State/Input" viewer allows you to look at the final state and input on every tick in the match.

If there are any input or state mismatches (where one or more clients recorded state or input that was different than the other clients), the differences will be shown on the right side of the window.

Using the "Previous mismatch" or "Next mismatch" buttons, you can jump to the previous or next mismatch found in the data.

You are likely to spend a large amount of time chasing down issues in your game logic that lead to state mismatches. Input mismatches are much rarer, but they are a super critical problem if they do occur.

How to setup replay

The "Log Inspector" can allow you to replay the match from the logs, by either loading state (in the "State/Input viewer") or re-executing frames (in the "Frame viewer").

This can be very helpful in debugging issues, by allowing you to visually see the state on a particular peer, when there is a state mismatch, or re-execute frames as they happened during a match with breakpoints set in the Godot debugger.

This works by launching a special instance of the game (called a "replay client") which connects to the "Log Inspector" via a TCP port (this component is called the "replay server").

However, there is a little bit of setup required to allow your game to support this:

Add the res://addons/godot-rollback-netcode/SyncReplay.gd script as an autoload singleton in Project -> Project settings... and the "Autoload" tab.
Pass a match_info Dictionary to SyncManager.start_logging() with enough info to setup the match for replay.
Ensure that your game doesn't do any logging when launched as a replay client. You can do this by checking if SyncReplay.active is true.
Set "Replay Match Scene Path" in "Project settings" to the match scene to load when showing a replay. Frequently, my games have a "Match.tscn" that runs the match when played normally, so this is also what's used for showing a replay, but you could create a unique replay scene as well.
Set "Replay Match Scene Method" in "Project settings" to the method on the scene set in the previous step, which will be able to setup the match to show the replay using the match_info from step number 2. Its implementation will look something like:
```
func setup_match_for_replay(my_peer_id: int, peer_ids: Array, match_info: Dictionary) -> void:
  # Setup the match using 'match_info' and disable anything we don't
  # want or need during replay.
  pass
```
For more details on the individual arguments, see what the "Project Settings" section above says about this setting.
(Optional) If you have a custom HashSerializer, ensure that it can unserialize any of the custom data it serializes, since this is what will be used to load the state and input data from the logs.

After completing all of these steps, you should be able to click the "Launch" button on the replay toolbar in the "Log Inspector" and see your game launch as a replay client. Then, in the "State/Input viewer", if you view a particular tick, it should load the state from that tick in the replay client.

You can use the dropdown on the replay toolbar to configure which peer's data you would like to use in the replay.

Manually reading log files

Starting with v1.0.0-alpha8, the log files are stored in a binary format. However, a script is included in the addon to convert them to JSON, so that humans or external tools can read them, if necessary.

You can run it like this:

godot --no-window --script addons/godot-rollback-netcode/log2json.gd --input=INPUT.log --output=OUTPUT.json

... replacing INPUT.log with a full path to the log file, and OUTPUT.json with a path to the JSON file that will be created.

The most common "match flow"

While there's sure to be edge cases, this is this most common "match flow", or the process your game goes through to start, play and stop a match using this addon:

Get all players connected via Godot's High-Level Multiplayer API.
Call SyncManager.add_peer() for each peer in the match.
Initialize the match to its initial state on all clients. Sharing match configuration and letting the "host" know when each client is ready to start can be done using Godot's RPC mechanism. Make sure that all nodes representing players have their "network master" set (via set_network_master()) to the peer id of the client that is local for that player, so that the virtual _get_local_input() and _predict_remote_input() methods will be called.
Call SyncManager.start() on the "host".
Begin the match in all clients in response to the "sync_started" signal.
When the match is over, call SyncManager.stop() on the "host". (If a client needs to leave the match early, they should inform the other clients via an RPC or some other mechanism, and then call SyncManager.stop() locally.)
Clean-up after the match in all clients in response to the "sync_stopped" signal.
If these same players wish to play another match (which can be worked out over RPC), then return to step number 2.
If this client wishes to disconnect from these players entirely, call SyncManager.clear_peers() and disconnect from the High-Level Multiplayer API, possibly via get_tree().multiplayer.close_connection() (with ENet) or get_tree.multiplayer.close() (with WebRTC).

It's also a good idea to connect to the "sync_lost", "sync_regained" and "sync_error" signals so you can provide the player with useful error messages if something goes wrong.

If you are logging, you'll want to call SyncManager.start_logging() sometime before calling SyncManager.start() (but after all match setup is complete) or in response to the 'sync_started' signal, and call SyncManager.stop_logging() just after calling SyncManager.stop() or in response to the 'sync_stopped' signal. The logs are meant to contain data from just a single match, which is what the "Log inspector" tool will expect.

Logo credits

The logo is composed of these images:

https://pxhere.com/en/photo/1451861 (License: CC0)
https://godotengine.org/press (License: CC-BY-4.0 by Andrea Calabró)

License

Licensed under the MIT License.

Name		Name	Last commit message	Last commit date
Latest commit History 88 Commits
addons		addons
assets		assets
demo		demo
.gitattributes		.gitattributes
.gitignore		.gitignore
LICENSE.txt		LICENSE.txt
README.md		README.md
default_env.tres		default_env.tres
icon.png		icon.png
icon.png.import		icon.png.import
project.godot		project.godot

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Godot Rollback Netcode

Tutorials

Installing

Games using this addon