Discussions related to Shared Editing, Distributed Apps, and Yjs. Questions about Yjs should go to discuss.yjs.dev. Bug reports to the issue tracker.
thanks @dmonad : will try that again, ran into some weird errors and just ran way. Basically I'm looking to have something running in the browser in a way that I can just write the DOM the old fashioned way and wire the javascript in. My challenge with webpack to date is that I can't hack around in the browser, or I don't know how to do it yet.
Hey @dmonad , I'm wondering about the pratcical limitations of Yjs in a single doc-- if the server is behind a load balancer, containerized and each has an instance of redis... it should scale pretty far, no? (I'm using a combination of Websockets and WebRTC, only single ws server atm). I think the limitations of React may become problematic (at some point)? Each user has a live cursor throttled at 30 updates / sec.
With gfx applications, is there a way to prioritize data that is on the screen or near the user? (I think "no") What kinds of things might you recommend to look at before a scale test, and what kind of things during?
With gfx applications, is there a way to prioritize data that is on the screen or near the user? (I think "no") What kinds of things might you recommend to look at before a scale test, and what kind of things during?
I suppose I'd have the database server acting as a persistent server that doesn't die? The rest would be ephemeral containers? or am I thinking about that the wrong way...
Any others with advice, also appreciated.
@evilfant if
ydoc.store.pendingDs === null && ydoc.store.pendingStructs === null that you can be certain that you are missing some updates. Otherwise, you simply have to initialize a fresh sync.
After losing connection, you must create a fresh sync to the other peer. You will only exchange the missing updates (it exchanges a few bytes in case there are no updates at all). You can see examples on docs.yjs.dev ("document updates")
@meatflavourdev That should scale pretty well. With redis you can basically scale indefinitely. But note that all peers will still receive all document updates / awareness updates. So something you could look into is to separate your (I assume - large document) into smaller fragments that users can subscribe to when they need to. This is what subdocuments are for. But you can also just create several Yjs documents to reflect your application-specific data.
the canvas is infinite so i think it prudent to control document size and transfer rate (by extension)
am I the only one facing this issue after updating y-webrtc?
TypeError: Cannot read property 'encrypt' of undefined
Module.encrypt
49 | return /** @type {PromiseLike<Uint8Array>} */ (promise.resolve(data))
50 | }
51 | const iv = crypto.getRandomValues(new Uint8Array(12))
> 52 | return crypto.subtle.encrypt(
| ^ 53 | {
54 | name: 'AES-GCM',
55 | iv
Hi @nc, I created the Awareness CRDT because the Yjs document does not deal very well with cursor positions. When you move your cursor across the document, you could easily create thousands of changes per second that cannot be efficiently shared using the Yjs CRDT. If possible, I recommend to add at least some kind of timeout before writing to Yjs. If possible, you should store the cursor in a different data structure when other users are not concerned about it. E.g. localStorage might be perfect if you just want to be able to restore cursors across browser-sessions.
Other users probably don't care about the "last" cursor positions from other users.
Or more plausibly, a reference to functional programming with an inverted lambda
Hey @cbenz_gitlab,
that's really funny and is actually what I thought when I started the project. I was a huge fan of Haskell, but that's pretty useless if you want to create collaborative applications. But why did it have to be JavaScript?? WHYYY JS!!
Now I really like JavaScript. It's a really fun language.
The actual reason is that I named the underlying CRDT algorithm YATA - Yet Another Transformation Approach. So Yjs is the js implementation of the Y(ATA) CRDT. Also, the Y kinda looks like two branches merging, which is what Yjs does. So when I name a variable ytext or ymodel, I immediately know that the data is observable and automatically merges with other peers :)
am I the only one facing this issue after updating y-webrtc?
TypeError: Cannot read property 'encrypt' of undefined Module.encrypt 49 | return /** @type {PromiseLike<Uint8Array>} */ (promise.resolve(data)) 50 | } 51 | const iv = crypto.getRandomValues(new Uint8Array(12)) > 52 | return crypto.subtle.encrypt( | ^ 53 | { 54 | name: 'AES-GCM', 55 | iv
[encrypt] this library only work for https
@MatrizLoin1210_twitter
I create counter , but feels not so right .
The best approach at the moment is probably to maintain a Y.Array and add all addititons & substractions to the Y.Array. You must not delete entries, as concurrent deletes would still decrement by only one.
Counter CRDTs are super popular. However, I believe that in many cases, you want the replace semantics. If user1 incremented a counter to 4 - because they think that 4 is the correct number. And user2 also increments the counter to 4 (because that other user is making the same calculation), it might be confusing when the conter is incremented to 8 (for no apparent reason)
Although, in that use-case, you definitely want to store who pressed the "like button". Otherwise, people could like the same post many times.
So you also end-up with an array.
I'm just saying that Y.Array and Y.Map should be preferred in a lot of cases. There are definitely valid use-cases.
Hi All, I'm working with y-websocket right now, and i'm wondering if anyone has experience adding messages to the server. I want clients to be aware when someone adds another "room", so that an added room gets added to other connected clients. Initially I thought i could just use the same WebsocketProvider to wrap the other providers to sync the client state using custom message types. I've gotten started but i keep running into issues with the persistence layer and i feel stuck. Any ideas?
Hi All, I'm working with y-websocket right now, and i'm wondering if anyone has experience adding messages to the server. I want clients to be aware when someone adds another "room", so that an added room gets added to other connected clients. Initially I thought i could just use the same WebsocketProvider to wrap the other providers to sync the client state using custom message types. I've gotten started but i keep running into issues with the persistence layer and i feel stuck. Any ideas?
The issue is if i just send a regular message like Websocket.send("stateUpdate") the server crashes with an Index out of bounds error.
You can either implement your own protocol based on the binary protocol that Yjs uses (which is fairly complicated) or use subdocument which seem perfect for your usecase https://docs.yjs.dev/api/subdocuments ;)
@dmonad thank you for the quick reply. I saw that in the documentation - but i was thinking it might be overkill. Something simple like the awareness CRDT makes more sense to me. I'm really just trying to maintain a list collaborative of docs, and if someone added a doc, then that gets propagated to everyone.... but now that i type that out it seems like exactly what subdocument documentation is describing haha
Suppose I have a
Y.Map and have set up a Y.UndoManager to track updates to it. I now make a series of, for instance, 20 adds, deletes and updates to the Y.Map. Then I decide that I want to undo these most of these changes, back to, say, the 3rd change. I could do this by calling undoManager.undo 17 times, but is there any way to revert straight to the 3rd change (e.g. by saving a pointer to the undoManager state before each change)? The use case is that I want to give the user a menu of the last n changes, so that the user can revert straight back to a previous state, without having to unwind each action one at a time.