Hacker News new | past | comments | ask | show | jobs | submit login
Creating a collaborative app using CRDTs (medium.com)
60 points by phedkvist 17 days ago | hide | past | web | favorite | 14 comments

I found this hard to read.

You state “I will explain later” a number of times. It would be better to hold off introducing the topic until it is more contextual to the reader. It breaks the flow of your writing.

The text could be broken up into more paragraphs.

You could use some more headings to break up sections:

1. What is a CRDT 2. What problem do they solve and some simple examples. 3. State based solutions 4. Operator based solutions 5. Join-semi lattice 6. Putting it all together.

In your second join-semilattice diagram it’s not obvious that the two blue nodes are being merged into the green.

The pseduo code you’ve included for the g-counter, 2p-set and or-set are quite unintuative imo.

I’m probably not your target audience but I read a lot of academic writing now and I think this could be better.

Content is great though, didn’t know what a join-semi lattice was before hand so thanks a lot!

Honestly, CRDTs are quite complicated to understand. No amount of blog posts, research papers or diagrams will truly click (for most).

And like somebody else said in comments, most articles on CRDTs assumes the audience are math people. I'm working on a series of blog posts that slowly uncover how real-time collaboration algorithms and data-structures work – with all the necessary math concepts explained in simple terms first. Will publish them out as soon as they take a good shape.

I agree it is quite difficult to explain these concepts in simple terms. I spent a lot of time reading academic papers, as well as reading blog posts in order to grasp these concepts.

Good luck with your articles, where can I read them once you have published them?

True but I was commenting on the writing, not the concepts.

Thanks for the feedback.

Great that you learned something. The CRDT papers don't explain the join-semilattice that much, it's only mentioned briefly, even though it's important in order to understand the theory behind CRDTs.

I don't really understand CRDTs, but I'd like to. The problem I keep encountering is that resources either assume I'm better at math than I am, or they come across as trying to over simplify/poorly analogize a hard thing (reminiscent of monad tutuorials ~2011). Can anyone point me to some resources that slowly,but thoroughly explain the concepts to me?

This is one of the best videos I've seen on CRDTs: https://www.youtube.com/watch?v=pMMDVphop40

CRDTs are just a way to send to send data between parties so that all parties end up with the same thing after receiving all messages, even if they receive them out of order. The way the messages are encoded is the math part, ensuring that you can keep applying these functions to your existing state to always get to the final state. That's where the associative/commutative requirements come from.

Note that there are operation-based (sending how something should be updated) and stated-based (sending what it should be updated to) CRDTs and they have different semantics and use-cases.

Wow, that is a fantastic video, thanks!

Here's my attempt at explaining what CRDTs are in 30 seconds:

CRDT are a type (e.g. a String, a Vector, an Integer) that has limitations with how it can be mutated.

For example, you can create an Integer CRDT that can only be `increased` and `decreased`, or a String CRDT that can only have characters `inserted` and `deleted` at given positions.

These limited mutating capabilities are chosen so that there never can be any conflict.

This a promising field but the article doesn't mention (at least that I could find) the challenges in implementing CRDT techniques for data structures that humans tend to use such as documents diagrams and spreadsheets, while maintaining acceptable size and performance.

I had a lot of hope for crdt as a more general technique for client server communication, but it seems that usage of crdt for anything other that collaborative documents is pretty rare

I'm working on this. It is challenging. Even for documents there are challenges.

I think CRDTs can be applied to more than just documents, and they can for sure be applied to client-server communication, and peer-to-peer for that matter.

I used CRDT to create a collaborative graph tool using the 2P2P-Graph. There is a pretty good Wikipedia page about CRDTs, which I recommend you read. It can definitely be applied to many use cases were the datatypes such as counters, sets, maps, sequences, etc are used.

CRDTs allow multiple concurrent mutations to the same data. If it's just between a single client and server then there's no need for this. There are some databases that are now using it for replication though.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact