Strong Eventual Consistency – The Big Idea Behind CRDTs

2025-09-094:4616390lewiscampbell.tech

Missing the Forest for the Sequence Trees.

Show article

9/8/2025

CRDTs. Data structures that can be replicated across multiple nodes, edited independently, merged back together, and it all just works. But collaborative document editing and multiplayer TODO lists are just the tip of the iceberg - I believe the big application is distributed databases, and for that we need to talk about consistency.

CRDTs are a tool for Strong Eventual Consistency. Let's start with the definition of normal Eventual Consistency¹:

Eventual Delivery: An update delivered to one node will eventually reaches all nodes
Eventual Convergence: If two nodes have seen all the same updates, they will eventually have the same state.

Strong Eventual Consistency (SEC) replaces Eventual Convergence with Strong Convergence:

Strong Convergence: If two nodes have seen all the same updates, they will have the same state.

Spot the difference! In SEC two replicas have the same state as soon as the updates are processed, not eventually. If multiple nodes are updated independently, there will be conflicts - this is inevitable. What SEC guarantees is that these conflicts are solved automatically and deterministically.

This has massive implications. SEC means low latency, because nodes don't need to coordinate to handle reads and writes. It means incredible fault tolerance - every single node in the system bar one could simultaneously crash, and reads and writes could still happen normally. And it means nodes still function properly if they're offline or split from the network for arbitrary time periods.

Strong Eventual Consistency is Eventual Consistency that works. If you're doing local first, or low latency geo-replicated systems - accept no substitute. This is the context I think we should see CRDTs in - building blocks of Strongly Eventually Consistent systems. CRDTs as application state as one thing, CRDTs as entire databases are quite another.

I'm available for hire.

Read the original article

Comments

By judofyr 2025-09-098:515 reply

> This has massive implications. SEC means low latency, because nodes don't need to coordinate to handle reads and writes. It means incredible fault tolerance - every single node in the system bar one could simultaneously crash, and reads and writes could still happen normally. And it means nodes still function properly if they're offline or split from the network for arbitrary time periods.

Well, this all depends on the definition of «function properly». Convergence ensures that everyone observed the same state, not that it’s a useful state. For instance, The Imploding Hashmap is a very easy CRDT to implement. The rule is that when there’s concurrent changes to the same key, the final value becomes null. This gives Strong Eventual Consistency, but isn’t really a very useful data structure. All the data would just disappear!

So yes, CRDT is a massively useful property which we should strive for, but it’s not going to magically solve all the end-user problems.

By josephg 2025-09-0912:244 reply

Yeah; this has been a known thing for at least the 15 years I’ve been working in the collaborative editing space. Strong eventual consistency isn’t enough for a system to be any good. We also need systems to “preserve user intent” - whatever that means.

One simple answer to this problem that works almost all the time is to just have a “conflict” state. If two peers concurrently overwrite the same field with the same value, they can converge by marking the field as having two conflicting values. The next time a read event happens, that’s what the application gets. And the user can decide how the conflict should be resolved.

In live, realtime collaborative editing situations, I think the system just picking something is often fine. The users will see it and fix it if need be. It’s really just when merging long running branches that you can get in hot water. But again, I think a lot of the time, punting to the user is a fine fallback for most applications.

By joshcaughtfire 2025-09-0916:001 reply

good point. the reality is conflicts should often be handled in the business logic, not in the consensus logic, but not universally. For the former, having the conflict state be the consensus state is ideal, but you do risk polluting your upstream application with a bunch of unnecessary conflict handling for trivial state diffs.

With CRDT, you have local consistency and strong convergence, but no guarantee of semantic convergence (i.e. user intent). I would still hire OP, but I would definitely keep him in the backend and away from UX

By josephg 2025-09-0918:501 reply

My point is a good crdt should let you tune that on a per field / per instance basis. Sometimes you want automatic “good enough” merging. Sometimes you want user intervention. When you want each is not obvious at the moment. We haven’t really explored the UX state space yet.

In general the automatic merging works pretty well most of the time. Where things go wrong is - for example - when people think they can put JSON data into a text crdt and have the system behave well. Instead the automatic merging breaks the rules of JSON syntax and the system falls over.

By cubefox 2025-09-1117:48

We have LLMs now, couldn't they be used to merge conflicts in a more sensible way? It might get a little expensive I imagine.

By LAC-Tech 2025-09-0919:22

So the entire point of the (short) article I wrote was to get people to think outside of the the little box people put CRDTs in: javascript libraries and collaborative editing.

Yet here we are, circling back to collaborative editing...

At this point I think the term "CRDT" has too much baggage and I should probably stop using it, or at least not put it in blog post titles.

By evelant 2025-09-0914:111 reply

I've prototyped something attempting to solve this problem of preserving user intent and maintaining application semantics. See comment here https://news.ycombinator.com/item?id=45180325

By josephg 2025-09-102:311 reply

I've replied elsewhere, but on the face of it I can't see how this solves the problem of conflicts in any way. If you disagree, say more about how it solves this?

If two users concurrently edit the same word in a text document, how does your system help?

By evelant 2025-09-1011:361 reply

For a text document a normal CRDT is perfect. They're very good for that specific case. What I tried to solve is eventual consistency that _also_ preserves application semantics. For example a task tracker:

* first update sets task cancelled_at and cancellation_reason

* second update wants the task to be in progress, so sets started_at

CRDT's operate only at the column/field level. In this situation you'd have a task with cancelled_at, cancellation_reason, status in progress, and started_at. That makes no sense semantically, a task can't both be cancelled and in progress. CRDTs do nothing to solve this. My solution is aimed at exactly this kind of thing. Since it replicates _intentions_ instead of just data it would work like this:

action1: setCancelled(reason) action2: setInProgress

When reconciling total order of actions using logical clocks the app logic for setCancelled runs first then setInProgress runs second on every client once they see these actions. The app logic dictates what should happen, which depends on the application. You could have it discard action2. You could also have it remove the cancellation status and set in_progress. It depends on the needs of the application but the application invariants / semantics are preserved and user intentions are preserved maximally in a way that plain CRDTs cannot do.

By josephg 2025-09-1014:031 reply

Yes; I get all that from the readme. You pick an arbitrary order for operations to happen in. What I don't understand is how that helps when dealing with conflicts.

For example, lets say we have a state machine for a task. The task is currently in the IN_PROGRESS state - and from here it can transition to either CANCELLED or COMPLETE. Either of those states should be terminal. That is to say, once a task has been completed it can't be cancelled and vice versa.

The problem I see with your system is - lets say we have a task in the IN_PROGRESS state. One peer cancels a task and another tries to mark it complete. Lets say a peer sees the COMPLETE message first, so we have this:

    IN_PROGRESS -> COMPLETE

But then a peer sees the CANCEL message, and decides (unambiguously) that it must be applied before the completion event. Now we have this:

    IN_PROGRESS -> CANCELLED (-> COMPLETE ignored)

But this results in the state of the task visibly moving from the COMPLETE to CANCELLED state - which we said above the system should never do. If the task was complete, it can't be cancelled. There are other solutions to this problem, but it seems like the sort of thing your system cannot help with.

In general, CRDTs never had a problem arbitrarily picking a winner. One of the earliest documented CRDTs was a "Last-writer wins (LWW) register" which is a register (ie variable) which stores a value. When concurrent changes happen, the register chooses a winner somewhat arbitrarily. But the criticism is that this is sometimes not the application behaviour what we actually want.

You might be able to model a multi-value (MV) register using your system too. (Actually I'm not sure. Can you?) But I guess I don't understand why I would use it compared to just using an MV register directly. Specifically when it comes to conflicts.

By evelant 2025-09-1019:43

It does not pick an arbitrary order for operations. They happen in total (known at the time, eventually converging) order across all clients thanks to hybrid logical clocks. If events arrive that happened before events a client already has locally it will roll back to that point in time and replay all of the actions forward in total ordering.

As for the specific scenario, if a client sets a task as COMPLETE and another sets it as CANCELLED before seeing the COMPLETE from the other client here's what would happen.

Client1: { id: 1, action: completeTask, taskId: 123, clock: ...}

Client1: SYNC -> No newer events, accepted by server

Client2: { id: 2, action: cancelTask, taskId: 123, clock: ...}

Client2: SYNC -> Newer events detected.

Client2: Fetch latest events

Client2: action id: 1 is older than most recent local action, reconcile

Client2: rollback to action just before id: 1 per total logical clock ordering

Client2: Replay action { id: 1, action: completeTask, taskId: 123, clock: ...}

Client2: Replay action { id: 2, action: cancelTask, taskId: 123, clock: ...} <-- This is running exactly the same application logic as the first cancelTask. It can do whatever you want per app semantics. In this case we'll no-op since transition from completed -> cancelled is not valid.

Client2: SYNC -> no newer actions in remote, accepted

Client1: SYNC -> newer actions in remote, none local, fetch newer actions, apply action { id: 2, action: cancelTask, ...}

At this point client1, client2, and the central DB all have the same consistent state. The task is COMPLETE. Data is consistent and application semantics are preserved.

There's a little more to it than that to handle corner cases and prevent data growth, but that's the gist of it. More details in the repo.

The great thing is that state is reconciled by actually running your business logic functions -- that means that your app always ends up in a valid state. It ends up in the same state it would have ended up in if the app was entirely online and centralized with traditional API calls. Same outcome but works totally offline.

Does that clarify the idea?

You could argue that this would be confusing for Client2 since they set the task to cancelled but it ended up as complete. This isn't any different than a traditional backend api where two users take incompatible actions. The solution is the same, if necessary show an indicator in the UI that some action was not applied as expected because it was no longer valid.

edit: I think I should improve the readme with a written out example like this since it's a bit hard to explain the advantages of this system (or I'm just not thinking of a better way)

By ljlolel 2025-09-0912:302 reply

LLMs might be able to use context to auto resolve them often with correct user intent automatically

By johnecheck 2025-09-0913:17

LLMs could be good at this, but the default should be suggestions rather than automatic resolution. Users can turn on YOLO mode if their domain is non-critical or they trust the LLM to get it right.

By SkiFire13 2025-09-0917:101 reply

The issue is that to preserve the CRDT property the LLM has to resolve the conflicts in a deterministic and associative way. We can get the first property (although most popular LLMs do not uphold it) but we can hardy get the second one.

By johnecheck 2025-09-0919:58

I read the comment you're responding to as suggesting a way to resolve the conflicts layered atop the CRDT, not as a component of the CRDT itself. You're very right that LLMs are the wrong tool for CRDT implementation, but using them to generate conflict resolutions seems worth exploring.

By sethev 2025-09-0914:56

Joseph Hellerstein has a series of posts on CRDTs: https://jhellerstein.github.io/blog/crdt-intro/

He very much leans toward them being hard to use in a sensible way. He has some interesting points about using threshold functions over a CRDT to get deterministic reads (i.e. once you observe the value it doesn't randomly change out from under you). It feels a bit theoretical though, I wish there were examples of using this approach in a practical application.

By LAC-Tech 2025-09-0919:28

It's a bit like how a static type system provides useful guarantees, but you can still do:

``` fn add(x: num, y: num) = x * y ```

By torginus 2025-09-0914:152 reply

Why do we even need CRDTs? Why can't we have multi-user editors work like multiplayer video games?

The server has the authoritative state, users submit edits, which are then rejected or applied and the changes pushed to others. The users is always assumed to be online for multiplayer editing. No attempt is made to reconcile independent edits, or long periods of offline behavior.

To prevent data loss, when the user is offline and desyncs, he gets to keep his changes and manually merge them back.

I'm sure this isn't a Google genius worthy implementation and fails in the incredibly realistic scenario where thousands of people are editing the same spreadsheet at the same time, but its simple and fails in predictable ways.

By gritzko 2025-09-0914:291 reply

Once I was using Slack on a bad WiFi and it was an adventure. What I saw as "sent" others never saw.

By torginus 2025-09-0915:05

Yeah it's a common optimization technique I saw from both backend and frontend devs to hide errors and lie about the actual status.

By calvinmorrison 2025-09-0914:211 reply

sure, i mean that was how early group editing works, but generally you want to preserve state from both (if we both start typing in the same spot, we both add stuff). Also it prevents any offline editing or high...lag editing really. unlike gaming which needs to be realtime this is much softer.

but no you dont need it

By torginus 2025-09-0914:57

This needs to be as realtime as WhatsApp. If your internet connection gets bad often enough to have trouble supporting WhatsApp, then my heart goes out to you, but thankfully this is clearly not normal for the most of us most of the time.

And if this happens, your experience is going to be terrible anyway.

By the_duke 2025-09-098:504 reply

The big problem with CRDTs IMO is that they make it incredibly easy to break application semantics.

Just a basic example for a task tracker:

* first update sets task cancelled_at and cancellation_reason

* second update wants the task to be in progress, so sets started_at

If code just uses the timestamps to consider the task state, it would not assume the task is cancelled, unexpected since the later user update set it to in progress.

Easy fix, we just add a state field 'PENDING|INPROGRESS|CANCELLED|...'.

Okay, but now you have a task that is in progress, but also has a cancellation timestamp, which seems inconsistent.

The point is:

With CRDTs you have to consider how partial out of order merges affect the state, and make sure your logic is always written in a way so these are handled properly. That is *not easy*!

I'd love it if someone came up with a framework that allows defining application semantics on top of CRDTs, and have the framework ensure types remain consistent.

By tempodox 2025-09-099:192 reply

Do not separate the state field from its time stamp(s). Use a sum type (“tagged union”) where the time stamps are the payload for a selected state. Make invalid states unrepresentable.

By the_duke 2025-09-0910:532 reply

There are many ways to solve each individual problem.

The point is that you always have to think about merging behaviour for every piece of state.

By fauigerzigerk 2025-09-0912:121 reply

Yes, sort of like you have to think about your transaction boundaries in server-side code for every single task.

The difference is that coming up with a correct CRDT solution for application specific consistency requirements can be a research project. In many cases, no CRDT solution can exist.

By josephg 2025-09-0912:291 reply

Can you give some examples?

In my experience, 95% of applications are handled just fine by the sort of JSON types built in to Yjs or automerge. The problems I hear people complain about are things like performance, size on disk and library ergonomics. And the long tail of features - like ephemeral data support and binary assets.

But data mapping seems mostly fine?

I know of a couple of exceptions. Arbitrary nested tree reparenting can be a nightmare. And there aren’t many good rich text implementations out there.

What problems are you actually running into?

By fauigerzigerk 2025-09-0912:502 reply

There are two rather different issues.

One large class of problems I'm thinking of is simply outside the scope of CRDTs. The whole idea of _eventual_ consistency doesn't really work for things like payment systems or booking systems. A lot of OLTP applications have to be consistent at all times (hence the O). Money must not be double spent. Rooms or seats must not be double booked.

The other class of problems is more debatable. CRDTs can guarantee that collaborative text editing results in the same sequence of letters on all nodes. They cannot guarantee that this sequence makes sense. Authors can step on each other's toes.

Whether or not this is a problem depends on the specific workflow and I think it could be mitigated by choosing better units of storage/work (such as paragraphs rather than letters).

By gritzko 2025-09-0914:311 reply

The classical paper-ledger bookkeeping is pretty much eventually consistent. They did not have the Internet when they invented it.

Flight booking is often statistically consistent only. Overbooking, etc.

By fauigerzigerk 2025-09-0914:561 reply

>The classical paper-ledger bookkeeping is pretty much eventually consistent. They did not have the Internet when they invented it.

Absolutely. Bookkeeping is an offline activity (I'm only doing it once a year in my company, ha ha). You just have to make sure not to record the same transaction more than once, which could be non-trivial but shouldn't be impossible to do with CRDTs.

>Flight booking is often statistically consistent only. Overbooking, etc.

That may be acceptable in some cases but you still can't use CRDTs for it, because you need a way to limit the extent of overbooking. That requires a centralised count of bookings.

By josephg 2025-09-0919:44

Most complex crdts are built on top of the simple crdt of a grow only set. Ie, what we actually synchronise over the network is a big bag of commits / operations / something such that the network protocol makes sure everyone ends up with all of the operations known to any peer. Then the crdt takes that big set and produces some sort of sensible projection from it.

> You just have to make sure not to record the same transaction more than once

So this should be pretty easy. Have a grow only set of transactions. Give each one a globally unique ID at the point of creation. Order by date and do bookkeeping. One thing you can’t guarantee is that the balance is always positive. But otherwise - yeah.

By josephg 2025-09-0912:575 reply

> One large class of problems I'm thinking of is simply outside the scope of CRDTs. The whole idea of _eventual_ consistency doesn't really work for things like payment systems or booking systems.

Yes! I think of it as owned data and shared data. Owned data is data that is owned by one process or node. Eg my bank balance, the position of my mouse cursor, the temperature of my CPU. For this stuff, you don’t want a crdt. Use a database. Or a variable in memory or a file on disk. Broadcast updates if you want, but route all write requests through the data’s owner.

Then there’s shared data - like the source code for a project or an apple note. There, CRDTs might make sense - especially if you get branching and merging support along for the ride.

> Authors can step on each other's toes.

Yeah when merging long lived branches, the workflow most people want is what git provides - of humans manually resolving conflicts. There’s no reason a crdt couldn’t provide this. CRDTs have a superset of the information available to git. It’s weird nobody has coded a system like that up yet.

By sethev 2025-09-0913:51

Pijul is a version control system based on a CRDT: https://pijul.org/manual/theory.html#conflicts-and-crdts

It works like you describe, with humans manually resolving conflicts. The conflicts are represented in the data model, so the data model itself converges without conflicts...if that makes sense.

By withinboredom 2025-09-0914:351 reply

I think you have the right idea, but possibly the wrong perspective. You want your _source of truth_, which is the "owned data" to be strongly consistent. Your shared data is a "view of truth" which may be incomplete or in disagreement with the source of truth. For example, the color of the sky "right now" depends on where on the earth you are standing, but we can all agree that air is 'just barely blue' and it depends on the light shining into it and how much of there exists.

The _source of truth_ are these facts (like "the air is blue" or "the user inserted the letter A at position X" or "the CPU is 40 degrees"). The view of this source is what we see, and can be seen through a CRDT or any other lens.

By josephg 2025-09-0918:441 reply

The way I’m defining it, my shared state is the data we store in a crdt. And CRDTs have strong eventual consistency. That’s what makes them great. So we can have a data structure which shows all users an identical view of the world.

Normally we do that by storing something totally different under the hood. Eg, git actually stores a commit graph. But the system makes a determinism guarantee: we promise that all users who have the same version checked out will see exactly the same thing. At one level, we’re storing “a list of facts” (the commit graph). But at another level of abstraction, we’re just storing application data. It’s just also replicated between many peers. And editable locally without network access.

By withinboredom 2025-09-107:111 reply

> So we can have a data structure which shows all users an identical view of the world.

This is never true. You can prove that at some time now()-T where T > 0 you had the same view of the universe, but you cannot prove that you currently have the exact same view because even with the attempt of checking, T becomes greater than 0. Sometimes, this doesn't matter (T can be arbitrarily large and still effectively be zero -- like asking your friend if he is still married to that person. They can answer you days later, and it'll still be true), but sometimes even very small values of T cannot be assumed to be zero.

By josephg 2025-09-108:491 reply

Well yeah obviously you never know for sure that a remote peer doesn’t have some changes that they haven’t told you about yet. That’s also true with lots of platforms - like google docs and Notion and multiplayer video games. Seems fine though? I don’t understand why this matters for collaborative editing?

By withinboredom 2025-09-1010:411 reply

Have you ever worked on the same repo with >500 devs? 99% of the time, it doesn’t matter. People talk to people.

By josephg 2025-09-1113:08

Yes; but I have no idea how that connects to anything else we’ve been discussing here.

By johnecheck 2025-09-0913:21

Conflict-free is right in the name, layering conflicts on top of it would be blasphemy :p

By evelant 2025-09-0914:071 reply

See my comment below, I prototyped something like this. https://news.ycombinator.com/item?id=45180325

By josephg 2025-09-0919:391 reply

Interesting idea. As I understand it though, this wouldn’t give you the kind of conflict semantics I’m talking about out of the box. What I want is - if two users concurrently edit the same line of text, the system can “merge” those changes by storing the conflict. Subsequent readers of the document see a merge conflict and can resolve the conflict manually.

Your system looks like it just enforces a global order on the actions. This will give you SEC - but how do you preserve the information that these edits were concurrent - and thus conflict with one another?

By evelant 2025-09-1011:41

You're right, it's not the same as conflict/merge semantics, but you probably could implement those semantics on top of it. My idea was more about being able to merge offline states for arbitrary data without user intervention while also ensuring that application invariants / semantics are preserved. Preserving app semantics while as much as possible preserving user intentions.

By fauigerzigerk 2025-09-0913:02

>CRDTs have a superset of the information available to git. It’s weird nobody has coded a system like that up yet.

That's an interesting idea. I have to think about this.

By naasking 2025-09-0916:16

> The point is that you always have to think about merging behaviour for every piece of state.

CRDTs can't eliminate the requirement to think about what the consistent states are.

By shakna 2025-09-099:393 reply

If you want invalid states unrepresentable, and time as a primary key... How do you deal with time regularly becoming non-linear within the realm of computing?

By josephg 2025-09-0912:321 reply

The general answer is to accept that time isn’t linear. In a collaborative editing environment, every event happens after some set of other events based on what has been observed locally on that peer. This creates a directed acyclic graph of events (like git).

By shakna 2025-09-0920:301 reply

That requires a different primary key than the time then, no?

By josephg 2025-09-102:301 reply

It requires a different primary key than an autoincrementing integer. One popular choice is to use a tuple of (peer_guid, incrementing integer). Or a randomly generated GUID, or a hash of the associated data.

Then each event is associated with zero or more "parent events".

- An event has 0 parents if it is the first change

- An event has 1 parent if it simply came after that event in sequence

- And if an event merges 2 or more branches in history, it says it comes after all of those events

You can also think about it like a set. If I know about events {A, B, C} and generate event D, then D happens-after {A, B, C}. (Written {A,B,C} -> D). But if A->B, then I only need to explicitly record that {B,C} -> D because the relationship is transitive. A -> B -> D implies A -> D.

By shakna 2025-09-107:101 reply

And the moment you need to merge, unrepresentable states become possible.

There are techniques to make it less painful, but it will still be possible.

By tempodox 2025-09-108:021 reply

You mean, like attempting to merge contradictory states? You will need some resolution stategy then, but in general that would be application-specific, and sometimes it may not exist.

By shakna 2025-09-108:041 reply

Okay... But we're now back to invalid states being possible. Tagging with time isn't enough.

By josephg 2025-09-108:46

It isn’t enough for what? What are you trying to do?

There may be a way to solve whatever problem you have, but without specifics it’s impossible to tell.

By johnecheck 2025-09-0913:25

It might be nice if our universe conformed to our intuitions about time steadily marching forward at the same rate everywhere.

Einstein just had to come along and screw everything up.

Causality is the key.

By throwawaymaths 2025-09-0913:361 reply

logical clocks

By Phelinofist 2025-09-0915:48

Even with a hybrid logical clock they are not that useful in case of a network partition AND clock drift, no?

By evelant 2025-09-0910:582 reply

I prototyped exactly such a framework! It's designed to solve exactly the problem you mentioned. It’s a super interesting problem. https://github.com/evelant/synchrotron

The gist is:

* Replicating intentions (actions, immutable function call definitions that advance state) instead of just replicating state.

* Hybrid logical clocks for total ordering.

* Some client side db magic to make action functions deterministic.

This ensures application semantics are always preserved with no special conflict resolution considerations while still having strong eventual consistency. Check out the readme for more info. I haven’t gotten to take it much further beyond an experiment but the approach seems promising.

By the_duke 2025-09-0911:183 reply

Nice, will have a look!

I've had similar thoughts, but my concern was: if you have idempotent actions, then why not just encode them as actions in a log. Which just brings you to event sourcing, a quite well-known pattern.

If you go that route, then what do you need CRDTs for?

By ForHackernews 2025-09-0912:58

Doesn't event-sourcing imply that there's a single source-of-truth data store you can source them from? I'm not sure event sourcing says anything about resolving conflicts or consistency.

By evelant 2025-09-0912:22

The pattern I came up with is similar to event sourcing but with some CRDT and offline-first concepts mixed in. By using logical clocks and a client side postgres (pglite) it doesn't have to keep the entire event history for all time and the server side doesn't have to process actions/events at all beyond storing them. The clients do the resolution of state, not the server. Clients can operate offline as long as they like and the system still arrives at a consistent state. AFAIK this is different than most event sourcing patterns.

At least in my thinking/prototyping on the problem so far I think this solution offers some unique properties. It lets clients operate offline as long as they like. It delegates the heavy lifting of resolving state from actions/events to clients, requiring minimal server logic. It prevents unbounded growth of action logs by doing a sort of "rebase" for clients beyond a cutoff. It seems to me like it maximally preserves intentions without requiring specific conflict resolution logic. IMO worth exploring further.

By n0w 2025-09-0911:49

A CRDT is any data structure that meets the definition (associative, commutative, idempotent, etc...)

Event Sourcing is not strictly designed to achieve eventual consistency in the face of concurrent writes though. But that doesn't mean it can't be!

I've also been considering an intent based CRDT system for a while now (looking forward to checking out GPs link) and agree that it looks/sounds very much like Event Sourcing. It's worth while being clear on the definition/difference between the two though!

By SkiFire13 2025-09-0917:211 reply

I wonder how does this handle a modify-rename conflict? e.g. there's a file identified by its name `a` and one client renames it to `b` while another client tries to modify the contents of `a`. Once you replay it in this order does the intent of modifying the contents of what was once `a` remain?

I know you can use some unique persistent ids instead of names, but then you get into issues that two clients create two files with the same name: do you allow both or not? What if they initially create it equal? What if they do so but then they modify it to be different?

By evelant 2025-09-1112:12

It would be up to application logic. This prototype essentially offers the same behavior you would get with a traditional backend API except it works offline. The results would be the same as if clients made those calls to a backend api, that is, up to application logic. My idea was that it's essentially impossible to have generic "conflict resolution" that follows arbitrary business rules so I made the business rules _be_ the conflict resolution. For any given situation the answer to "how would it handle a then b then c" is "the same as any normal backend api, per regular business logic, except it works offline".

By littlecosmic 2025-09-0911:421 reply

Don’t you also have to consider this just as much without CRDT? Not saying it isn’t a real issue, but this example could easily be a problem with a more traditional style app - maybe users open the record on their web browser at same time and make different updates, or they update the different timestamp fields directly in a list of tasks.

By the_duke 2025-09-0912:19

Sure, but you can usually rely on database transactions to handle the hard part.

By filleokus 2025-09-0910:46

Yes!

Any many CRDT implantations have already solved this for the styled text domain (e.g bold and cursive can be additive but color not etc).

But something user definable would be really useful

By gritzko 2025-09-0910:411 reply

The big idea behind CRDTs is that a data structures can have replicas synchronizing on a best-effort basis. That is much closer to the physical reality: server here, client there, phones all over the place.

The basic CRDT ideas are actually pretty easy to implement: add some metadata here, keep some history there. The difficulty, for the past 20 years or so, is making the overheads low, and the APIs understandable.

Many projects revolve around some JSON-ish data format that is also a CRDT:

- Automerge https://automerge.org (the most tested one, but feels like legacy at times, the design is ~10yrs old, there are more interesting new ways)

- JsonJoy https://jsonjoy.com/

- RDX (mine) https://replicated.wiki/ https://github.com/gritzko/go-rdx/

- Y.js https://yjs.dev/

Others are trying to retrofit CRDTs into SQLite or Postgres. IMO, those end up using last-write-wins in most cases. Relational logic steers you that way.

By petralithic 2025-09-1014:451 reply

Could you explain more about Automerge being legacy? They recently released a new major version and revamped their algorithm I believe. What is better about your version?

By gritzko 2025-09-1021:082 reply

Automerge is based on design decisions from 2014-2017. I remember that epoch and what everybody thought back then. The views have evolved since, but Automerge is a large project, many things have being built on top. It is unrealistic to change any of the basic assumptions at this point.

I may go into the technical details, assuming my hourly rate is respected.

By diarrhea 2025-09-1021:56

Go on. I respect your hourly rate.

By petralithic 2025-09-113:581 reply

How is your version better?

By gritzko 2025-09-116:05

That's a long list.

- RDX has a spec, so it can have compatible implementations. The result of a merge is specified to a bit. Automerge works the way Orion coded it (this time).

- There are equivalent text and binary formats, JDR and RDX.

- RDX palette of types is richer. Automerge is narrower than JSON.

- RDX can work in any commodity LSM db, natively, in the core (and it does).

- and so on...