Having done lots of Minecraft modding a decade ago, it's wonderful to see that the community is still active enough for there to be inside jokes like these.
Given the size of the game, it's not an easy feat to build a Minecraft server in any language. Yet there are seven, in just Rust alone??
The protocol minecraft uses to communicate between server and client is relatively straightforward and 'dumb' (read: tolerant of missing or contradictory data), so it's quite easy to make a server that a client will connect to and work OK with. Making something that supports all the game mechanics, especially world generation (an area Mojang/Microsoft are a lot more protective of, besides) and bug-compatibility, is a lot harder.
If somebody could get a high performance mc server working that supported everything except world generation, that would be immensely useful to a lot of people. Worlds are often pregenerated and this can be done offline by an official java instance, then give to the alternative software which players actually connect to.
I suspect the hard part would be getting total parity with all the undocumented intricacies of mob spawning and AI, and block interactions. But if there are slight differences from Vanilla this isn't necessarily the end of the world for players. Popular server mods like Paper already tamper with some Minecraft "features" in an opinionated way and for the most part players don't notice.
Getting Redstone interactions to be bug compatible is no small task. Redstone has complex interactions with nearby blocks that are completely baffling to new players and still challenging to veterans.
Mob spawning and behavior shouldn't be that difficult, but if you want identical terrain generation you are going to be cursing life.
What would really make a third party server stand out is first class mod support.
Better performance is almost a given. Minecraft's engine has a lot of low hanging fruit that has yet to be picked despite it being theoretically a multi-billion dollar game. Just look at how shockingly CPU hungry hoppers are for example. Mob pathfinding also consumes an inordinate amount of resources and is still kinda lousy.
I get it has lots of computing to do for something like a server with large players, but even a server with a small amount of players that's technical focused can easily bring the game to a crawl.
Its funny how the best way to get great performance from Minecraft is getting a CPU with great single core performance, get lots of memory, and then use fabric mods to optimize the game/server.
The performance differences look pretty impressive from the benchmarks. I do notice that world generation and saving features are missing though, and these tend to be pretty expensive operations. Chunk gen especially can bring a weaker VPS to its knees. I'm sure the benchmarks were taken at an idle state, but I'd be curious to see how it compares once those features are included and being used.
I don't see it listed, but is there support for block breaking/placing yet? Presumably this would require light recalculation and a chunk update on the server.
Finally, do you plan to add advanced features like scoreboard, teams, or command block parsing in general? Mojang has at least open-sourced Brigadier for that.
Cool project. Hope to see it mature to the point of making servers easier to run on low-end hardware.
Yep block placing and breaking is already supported, but light currently don't so everything is dark :/, We are working on it.
Yep, We want to add all cool features like scoreboards, teams and we already have a API which is similar to use to Brigadier for our Commands.
Thanks, I would love to give players on low-end hardware the possibility to host servers. I think i may test Pumpkin on a raspberry PI or something one day
I remember Notch talking about the lighting calculations being one of the hardest parts to program, so I can understand that being a challenge.
Mojang made large improvements to the lighting engine in 1.20, bringing it in line with the performance mods Phosphor[1] and Starlight[2]. Despite being deprecated now, they might still offer some useful insight into how to approach such a system in a performant way. You'll need to be mindful of the licenses, but it's likely easier than reverse-engineering Minecraft (even with mappings).
Chunk gen makes sense to implement last or never. If you want a performant Minecraft server you need to pregen all the chunks anyway. You can still later regen chunks that have never been visited to get new chunkgen on updates since chunks store the inhabited time.
I think Minecraft server re-implementations are pretty neat and I like to see when a new one comes out. There are also specific purpose server impls like MCHPRS for doing fast redstone compilation for technical minecraft.
I think a high performance block-for-block compatible chunk generation program would be great for anarchy Minecraft servers or generally servers with an "infinite" minecraft world where pre-generating all chunks is not possible.
Oh yeah I agree. There's a lot of fun problems to solve with Minecraft servers. I didn't mean to imply that there are no reasons for good chunk gen. I'm primarily into technical survival so my personal priorities wouldn't be chunk gen.
Are there any benchmarks for it? How much faster is it than a vanilla server?
I know Minecraft servers tend to get extremely resource intensive as the player count creep and people run extremely beefy servers to handle the load and still offer poor TPS.
Minestom is not based on Kotlin, im guessing they went with a kotlin template instead. Minestom is a server implementation that doesn't use any Mojang code and by default doesn't do much for you but provides the utilities to add the things yourself.
I literally said Holy Shit out loud. This is an incredible improvement, and I'll refer to this in the future when I'm asked if we should make something new in Java.
Keeping in mind that this server appears to implement only a tiny subset of the features the ones it is benchmarked against do... No lighting, mob spawning, mob ai, redstone functionality, tree or plant growth, water/lava flow, etc.
More than a "yes, it should". While loading chunks you need to take redstone, nature and mobs into account. If you don’t - you need to find them and resolve at runtime with even bigger costs, relying solely on difference between programming languages in performance.
Yes but also consider the extensibility accessibility Java gave us. EVERYONE was building Minecraft mods back in the beta days. I might go as far as to say that extensibility is what made Minecraft so great.
What I always found surprising was how many minecraft mods there were despite mojang having absolutely no mod support for the game.
After learning that to make a minecraft mod the process was basically decompile minecraft fight the terrible names provided by the decompiler to make your changes then recompile it, I lost all interest.
No doubt about it. I don't think Minecraft would have gotten as far as fast in the public consciousness without content creators like Yogscast being able to produce so much novel content from modded Minecraft.
Currently Pumpkin has not much multi-threading but we want to go all in. There is already a good structure for multi-threading, We want to make everything multi-threaded what benefits from it
> Be a drop-in replacement for vanilla or other servers
It seems to me that unless it's a drop-in replacement its not a Minecraft server? Akin to how say an Uno deck isn't a drop-in replacement for a Hearts deck but still both card games but not both Uno decks.
Or is it just meaning that Pumpkin (besides the network) do things differently than vanilla and so you might not be able to open a vanilla created world using Pumpkin?
The common problem with Minecraft server implementations is that they are not bug-for-bug compatible, which will lead to certain techniques (especially redstone contraptions) breaking. The technical Minecraft community depends on many implementation details which not all servers support
I edited in scare quotes for "complete" to make that clearer, but I mean in terms of at least having matching blocks/mobs despite many differing details.
There are issues other than not ported bugs. Redstone in bedrock is know to be unreliable.
For example, in Java version if you take a circuit and activate it with a button/lever - it would always behave in the same way. In bedrock same setup could have random result. And "random" is something you don't want in a large sophisticated contraption.
I'd guess it's caused by some race-conditions in bedrock implementation, but alas it wasn't fixed in 7 years.
Minecraft has a lot of bugs or otherwise surprising behaviours that parts of the community have come to rely upon. This means that most non-vanilla minecraft servers aren't 100% drop in replacements. You have to make a decision what behaviours you want vs the performance and simplicity gains you will gain.
For example there there are tricks that allow you to delete bedrock blocks. Which then lets you either get onto the roof of the nether, or drop below the bottom of the world. Not all of these tricks will then work depending upon the specific minecraft server.
Another example is that in vanilla you can "bomb" people with experience orbs, the sheer number of orbs on the screen will grind their game to a halt since there are too many objects to track and render. Some minecraft servers work around this by grouping up experience orbs into a single bigger orb. That way you have fewer orbs on screen at once.
One bug abuse that blew my mind recently is the ability to have wireless redstone in vanilla [1]. I fell deep into that rabbit hole after a previous post on here about Bad Apple in Minecraft [2].
As someone who knew about this project from earlier (I had even joined their discord) (currently have just deleted my discord account for better state of mind)
Its really made me happy that hackernews really liked this project (140 upvotes is pretty good in my opinion)
From what I remember , there was one other server as well which also was written in rust but I am not exactly sure
Also , the last time I was at it , it was really really alpha software but it was getting developed at good rate , so I am not sure about its current state (I was there when the author had gone to take his exams IIRC)
Hi. Im so happy there are so many people liking the project, The Project is still pretty WIP but im really working hard on this, i finished my exmans last week and currently in holidays so commits are again back to normal :D
it already is, the entire protocol is reverse engineered, there are tools to automatically deobfuscate the code and there is already a full reimplementation of minecraft that also supports servers
If that's the case, how come nobody seems to be writing improved Minecraft clients?
Ever since I started playing it in the beta days I've been frustrated with how poorly Minecraft performs relative to what it's showing on the screen. (Not that that stopped me from pouring hundreds of hours into the damn thing.)
Well, they do? Sodium, for instance. It's a mod, not a full rewrite, rewriting the client from scratch would mean a lot of boring work like speaking with Mojang's server, but I understand Sodium basically rips out and replaces the entire graphics pipeline of the client.
Yeah, it was always weird how 32x32x48 extreme reactors lagged the game whenever you looked at them, but the moment you looked away everything was fine.
the client has to authenticate with a central server and present a ticket to the server it wants to connect to. otherwise clients could impersonate each other easily.
sure, iirc it used to even just be a setting? online-mode=false
most servers leave it enabled because preventing player impersonation is pretty important so people can't just easily grief each other. some piracy servers implemented their own auth on top.
I will give it a try, Thanks for this project. The performance of pure vanilla server jar is so bad. Thats one thing I wish mojang would improve. I know new chunk generation is multithread at least. I also wish they'd officially support some basic control for SMP servers. Something as basic as areas defined by two corner coordinates and basic permissions like place, break, interact. Just a basic config file is fine I can manage the requests/updates ot it using other tools - or a simple commandline utilities in game to ops. Something that lets me not have to always only use an excluse whitelist. I know Paper/waterfall and the others have plugsins for this but theres just something nice about staying pure vanilla. Ive been keep servers alive for a long time now.
Hey, Im happy you want to try it out just keep in mind its not done yet many features are missing, We want to have all the things you listed be configurable in a config file, so i think you will like it. If you have any problems may worth to take a look at https://snowiiii.github.io/Pumpkin/troubleshooting/common_is... or just ask for help on our discord, Have fun :D
Hilarious site but a little unjustified. Out of those, only Valence is a real "this already exists" argument. There's 7 total and the first 3 are dead and the latest 3 (incl pumpkin) are all 2024. And all of the 2024 ones explicitly differentiate themselves from Valence which is more of a framework that focuses on modularity. These latest 3 all look like healthy competitors that are all amicable towards each other
| | released | stars | last_update |
|----------|----------|-------|-------------|
| feather | 19aug | 2,614 | 2 years ago |
| mchprs | 20jun | 1,632 | last month |
| graphite | 22jun | 160 | 2 years ago |
| valence | 22sep | 2,731 | 3 days ago |
| hyperion | 24mar | 361 | today |
| pumpkin | 24jul | 965 | today |
| ferrumc | 24sep | 673 | today |
It's pretty funny how many of these are "Is it a Minecraft server? No. It's a framework with which you can build your own rewrite of Minecraft server."
This got me thinking - is Rust really the right tool for the job? And I thought, Elixir/Erlang feels like the perfect tool. Sure enough, someone did it! Great reference material for someone learning how to do things on the beam https://github.com/thecodeboss/minecraft
"Modern" seems to be used a loose adjective these days for "I rewrote $thing [in Rust]". Minecraft was created in 2011, and is Wikipedia says the last version of the 'classic' edition was released in 2017. So anything after 2017 is now defunct.
I don't mind people rewriting things in <insert-name-of-tech-I-like> but "modern" as a value seems pretty loose, and it's often at least arguable whether it's objectively better!
“Modern” more usually means some new JavaScript thing. In JS land, they consider anything that hasn’t had a commit on main branch in over 3 days to be a dead old project in need of being replaced with something new and “modern” that is up to date with the latest trends and breaking changes from the previous 24 hours of their world.
Usually the hyperbolic superlative for Rust projects is “blazing fast”. Of course, any kind of benchmarks or comparisons with other implementations are completely optional. It is simply enough to “cargo init” and start hammering out code. You don’t even need to consider the characteristics of the algorithms you choose to use! If it’s Rust, it’s “blazing fast”.
Your most starred repo is inferior to a shell one-liner lol. Talk about pot calling the kettle black. Just use the system dict, shuf, grep, and head.
It’s bad form to badmouth someone’s earnest work for sure. I wouldn’t do it normally since I think it’s nice that you actually did something. But if you’re going to sit in a glass house and throw stones you should expect some back.
Fortunately, my house is an underground burrow so I can throw stones with impunity. As ugly as it is to do.
Sorry, I may should not used the term Modern, Lets say the foundation is newer and more optimized than from the Original Minecraft server. Mojang developers have strict deadlines and do not care about performance (like basicly any big Studio today). This results in bad ugly code which only purpose it is to work nothing more. Minecraft was created 2009 btw
I'd argue they care about performance, but they also care about a whole slew of other things that also require prioritization to maintain the game and its cottage industry. Not a huge fan of the constant dogging on mojang everyone loves to engage in...
> vulnerabilities decay exponentially. They have a half-life. [...] A large-scale study of vulnerability lifetimes² published in 2022 in Usenix Security confirmed this phenomenon. Researchers found that the vast majority of vulnerabilities reside in new or recently modified code
A study limitation is that they looked only at security-relevant bugs (vulnerabilities). As someone who writes code, I would tend to think that this also goes for bugs without a direct security impact, but I don't have the data to back that notion up
For bugs, perhaps, but for vulnerabilities, new attacks and techniques are being found. Or just nobody is looking at most things most of the time and it's not really correlated with age that clearly. Imo it's good to have the data of what actually happens
Its written from the ground up and has a clean foundation (which is not the case in vanilla minecraft server code). We also want to use cool modern features like multi-threading or the rust language which is a modern language designed to fix mistakes from older languages
The limitation isn't the Java version, it's the way the MC code was architected.
Iirc part of original Minecraft's performance limitations were high object create/destroy rate leading to lots of garbage collection. With that in mind, picking a non-GCd language isn't completely crazy.
It was the other way around: say, the original code as written by Notch had functions like setPosition(x,y,z) and it was okay. When Jeb got in charge, he said "it's not object-oriented enough" and rewrote everything to setPosition(position). And boom, 1GB/sec allocations...
java 1.0 (even 0.9) did have. The 1st language with a memory model was java 5 (1.5), which was back ported to 1.4.2. 1.4(.2) already had non-blocking I/O as well.
Compatibility: refers to the ability for client of the latest mincraft server version to be able to connect to the server (pumpkin) unmodified with all mechanics working.
Drop-in replacement: refer to the ability of a server operator to simply exchange the current installation/executable of the server (be vanilla, paper, cuberite, etc...) for pumpkin while maintaining data, configuration, scripts and mods installed.
Tell you the truth, that doesn't help clarify to me much.
If it is compatible, then can't I 'drop it in'.
They sound like they are saying the same thing.
It is compatible so clients can connect to the server and be fully operational, and thus, I should be able to drop in this server, and use it as a server?
How can it be compatible if it doesn't maintain data, configuration, etc..
Edit:
or is this about Pumpkin files. Pumpkin will maintain it's own files, data, configuration. So it can't just use existing Minecraft data files.
So if it was a new world, Pumpkin would generate new pumpkin formatted files. But couldn't just 'drop it in' on an existing world and use the existing Minecraft data files.
Not sure of long term viability as far as effort, but if it is files, couldn't a converter from MineCraft to Pumpkin file structure, make the server 'drop in'?
"But from a client perspective, both implement HTTP/1.1."
This seems too low level example to apply for Minecraft.
Minecraft has a lot going on, back and forth. For something to be 'compatible' it would need to be so detailed an implementation of the server, that it could potentially also be a 'drop in'.
When it comes to game servers, especially for highly modified games like minecraft, there's the client facing side, including how the game behaves, and then there's the server administration side.
If I were to interpret the statement made here I would assume that not-a-drop-in while being vanilla compliant means things like:
* It won't support server side mods without you re-writing them in rust
* It won't support bukkit plugins without you re-writing them in rust
* It doesn't support the configuration file format that the java server uses
* Maybe it doesn't support the world file format (which would be server side only)
* Maybe it doesn't support the same administrative commands (which you would only see as an administrator)
etc.
So it can definitely be compatible from the client perspective while being woefully incompatible with any prior experience anyone running a minecraft server might have.
Drop-in means replacing an existing thing with a new one without changing anything.
This is not compatible with other servers plugins/data/configurations, so just replacing the binary and expecting to continue where you left off is not possible.
"Drop-in" is what enterprise software calls "bug-for-bug compatible" - e.g., replacing RedHat with CentOS (RIP) should work exactly the same, even if the CentOS team found bugs - they report them upstream and do NOT fix them themselves, because code may be relying on the bugs.
This is especially true with complicated vanilla Minecraft setups and red stone machines (Java Minecraft red stone has "bugs" that "shouldn't be there" but cannot be removed now since so much depends on it).
I would say that Minecraft servers by some qualities are really hard to implement (for example generate world as Java would - using Java's random number generator to generate exactly the same world in Rust) or even impossible. But other usages, like walking through existing world with 1000 of your colleagues might be worth of writing a very fast but not a "drop in replacement" server. Or a massive minigame maybe?
Hey, There were already efforts made to rewriting the same random generators from Java, And they work. Same seed as in vanilla, Same Result as in vanilla
It's only deterministic if you feed the seeds into the exact same pseudorandom number generator and fetch random values from it in the exact same order. If you take a single extra random value out of order, then everything ends up different.
Hey, With Compatibility i mean be compatible with existing Minecraft vanilla client's and also use vanilla logic.
With "Be a drop-in replacement for vanilla or other servers" i mean that if you just replace the existing server file with pumpkin, pumpkin will not load configs/plugins from vanilla/other servers
Hi, Valence is a framework (similar to Minestom in Java). You have to build everything you self. Pumpkin is not a framework :D. Also Valence is bit unactive (look commits)
That's why I said "clean room implementation". Any reason why Pumpkin does not build on top of Valence?
Would be great to see a more cohesive ecosystem for Minecraft servers in the rust community rather than reinventing the wheel
Impressive stuff, but I'll be more keen to see an high performance implementation that also supports the vast numbers of Minecraft mods. Forge servers are notoriously slow on bigger modpacks.
Right, that’s my take as well. But on the whole I don’t see Java itself as being fundamentally less performant, particularly with JDK 21+. Architecture choices in either language make or break performance.
Just to be clear this sever only works with the current vanilla version of Minecraft? I've been interested in playing again, but on the older Beta builds (1.7.3) prior to the full release versions. These versions aren't supported are they?
Curious what protocol you would use, or how you would do it differently? Are there small enhancements to the existing protocol you would do? Genuine open question for the sake of learning.
I would prefer the UDP Protocol over TCP like in Bedrock edition. Im pretty sure many PVP players would love this. Here is much non sense in the Minecraft protocol and things made to work not to be optimized (deadlines). At our discord we even already have a sticker :mojang_nonsense: which will be used quiet often. I also don't understand Mojang's tactic with packet changes, It sometimes looks like they care about Packets not being broken and being backwards compatible but them sometimes they change the whole Networking system (1.20)
Basically any significant Java software uses disproportionate amounts of RAM. My server can easily run a web server, database server, ftp server, memory cache, etc., times a few copies because various containers have their own web server or whatnot, all at the same time. But a single copy of Minecraft is too much, even when everything else is shut down.
That's why I struggle with the server: it's written in Java. Other game servers don't seem to have this problem. If the network protocol were all, that'd be great
(I didn't downvote any of your posts in case you now wonder if that was me)
> A Rust framework for building Minecraft: Java Edition servers.
Through there are currently 3 new rust impl. differentiating themself from Valence. So it's interesting to see how this will develop long term. Like will they be for different use case(1)? Will they merge? Will some die and developers migrate?
(1): There are actually quite a lot of different use cases, i.e. Java Edition is by now mostly about mods so providing different modding capabilities and various "common mods build-in" and similar can be a big target. But so can be wanting to run a mostly vanilla community MC server for quite a bunch of people etc.
