so anyone trying to fix this well known long-term bug?
so anyone trying to fix this well known long-term bug?
Jbasically my point is if you have to pass it in to the DO - I might as well just include the data outside of the DO since that's where it's coming from anyways.
LYou can track this bug in https://github.com/cloudflare/workerd/issues/2240
JSince it is never populated inside the DO (and I suppose this won't be fixed short-term), I think it would be better and cause less confusion to have different type definitions for each place, where the one in the DO instance wouldn't have the
.name property
NTrying to move a KV DO to SQLite and I know I'll have to delete it and create a new one. But how do you do that? what does the migration look like? Do I have to completely remove the binding from wrangler, build then readd it?
Ndo i simply do this?
LJust got issues with SQLite today in my DO where I stored user tables. "Error: Exceeded allowed rows written in Durable Objects free tier". Which is really strange since I'm only storing a few rows (8 users!) and 100 rows in a table called "transcripts".
JSo they just call
setIdentifier every time they "get" an Actor instance (which is a Durable Object) and refer to this.identifier rather than try to parse the name out of the id. It's very little code and works the same in dev and production.LPersonally, I suggest against this for what's worth. Doing explicitly as it suits your use-case, either once at the beginning/creation, or in every request.
LI'm basically using all RPC calls I thinkI hear people say this and I always wonder how users enter the system? It's got to be HTTP or WebSockets, no?
LWhat exactly are you recommending against and why?
LDoing a 2nd call to the DO all the time for every request to set the name identifier.
Jlike I was thinking
firstDO.withIdentifier('user:123').index()Jso I use a normal Hono router on a worker for the routing.
I use the DO's
everything else is done through RPC
I use the DO's
fetch()everything else is done through RPC
JJso really I use the durable object as a class with a store in it, kinda like you'd use something like Pinia for on the frontend.
Ja store has it's state, actions, getters
Jidk why I went with this instead of doing it all through
fetch () and routing it that wayJ@Larry and the
BindDurableObjectMiddleware is what sets the DO to c.var.notifications
1You can't turn an old KV namespace into an SQLite one
CSome DOs we use purely as a data store with some collocated code. so they are never accessed via http or websockets
CEssentially d1 esk functionality but with benefit of collocated code and easier to have db per customer/user
LI'm not a hono user, but if you set a header in one of your middleware's with the name, then in your fetch inside your DO you grab that header and store it, you are golden. It even works if you are doing a WebSocket upgrade request because that goes through the same fetch handler inside your DO. That's how I always do it for DOs that are only one hop away from the user. For deeper layers, I would have to resort to some of the things people are mentioning here, but I decided a while ago that I would just use the string form of the id instead of the name for those layers so I never need the name for them.
LYeah, as @1984 Ford Laser said, there is no automatic way to do it. I did it lazily when I had to do it the first system I migrated but I never did it that way after that because I had a panic at one point that I had lost data due to a bug in my code. Down time and cold turkey cut over is probably best.
JA bit related to this: Types for RPC functions are incredibly fragile. With common things like branded types they just turn into ‘never’ and make you question your own sanity
Nim aware but i figured i could just clear it and make a new one but nope! I had to delete it, remove the binding, deploy, then readd it.
GMy work around was to setup a container that my worker calls. The container reads all the data from the durable object into its own sqlite database and then saves that to R2
GIt's less elegant then wrangler but it doesn't block operations so thats a plus, and its also a proper database file and not a .sql file so its a lot more compact and easy to work with
PBackups are not an option neither a feature. It is an integrity and security core aspect born with computers. Cloudflare is great but Jesus the mindset behind few thing really needs to be revisited. This is why I was happy that open source altertives like Rivet would come to shake these nonsenses that make such a great feature a no-go for many businesses and projects.
JThere are a couple of things that look like really weird decisions to me as well. The ones that cause me the most frustration are around creating too much "magic" like Smart Placement without the possibility of setting a hint, and Queues that don't let you set concurrency manually. If it "just worked" it would be incredible but it was not or still isn't the case for well over a year. Hyperdrive did make things better, though.
LI've never used it but have you looked at: https://developers.cloudflare.com/realtime/sfu/introduction/
Cloudflare Docs
Cloudflare Realtime can be used to add realtime audio, video and data into your applications. Cloudflare Realtime uses WebRTC, which is the lowest latency way to communicate across a broad range of platforms like browsers, mobile, and native apps.
CUnfortunately it’s kinda a typescript shortcoming, in the cloudflare type definitions RPC maps all functions of an object to make them async. Typescript doesn’t preserve generics when you do this, therefore you lose things like branded types.
CWorkaround we have used is to recast to correct type. Which if you have some kinda wrapper around your DOs is workable since you can isolate the hack to one place.
MDoes anyone know how to get approved for the Dynamic Worker Loader Beta? We could really use the invite actually. Thank you.
1MYes
Mwhen using durable object
versus what you'd normally expect:
script_name option for local dev, the generated binding lacks a type parameter, usually provided to the namespace generic, ie.versus what you'd normally expect:
JI am using DO and CF Containers to execute on demand user scripts on our platform. Is there any way to get access to "Dynamic worker loader beta". Have filled the form on announcement date.
LJust to be clear. D1 and SQLite DOs support Point in Time recovery for any point in time up to 30 days in the past.
What I a mentioned above was for things like backups elsewhere, to different DBs, or offloading backups to another product like R2 or even different providers.
What I a mentioned above was for things like backups elsewhere, to different DBs, or offloading backups to another product like R2 or even different providers.
LAs you will see over the next few weeks/months, a lot of these things will be improved. As a platform Cloudflare evolved a lot over the years, and yes, a lot of the mindset and priorities changed due to how customers are using the platform and how certain "magic" as you say didn't come to fruition. (Personal opinion, but I personally dislike anything "magic", if you could see it from my suggestions over the past year, so I am with you on that one 
)
So, as I always say, giving constructive feedback in what you would like to see change, or even new things, and justifying with legit use-cases is the best way to steer the platform towards delivering those things.
)So, as I always say, giving constructive feedback in what you would like to see change, or even new things, and justifying with legit use-cases is the best way to steer the platform towards delivering those things.
LThese are my "rules" for working with DOs. I'm looking for feedback/pushback/thoughts. They were written for my AI Coding Agent but I believe they are good rules regardless.
POh absolutely, PIT was already part of my answer, but thanks for clarifying! Still, PIT is only a first-level safety net and indeed doesn’t solve the real issue : offloaded backups that ensure data integrity and security and easy way to import/export. Without these "features", backups and recovery in case of emergency become far too uncertain, left entirely to the user’s discretion and prone to human error.
SQLite definitely makes things more portable than KV, though. I just hope dumping the SQL file somewhere won’t take months (or years!) to arrive.
SQLite definitely makes things more portable than KV, though. I just hope dumping the SQL file somewhere won’t take months (or years!) to arrive.
JIf I have a transaction going on with SQLite (Output gate closed) and inside the transaction I make a network request (fetch, Queues
sendBatch, etc), what should happen? Should it just deadlock/timeout?LIf I have a transaction going on with SQLiteWhat do you mean by this? How do you have it going? Is it an implicit transaction, or do you use the
transaction()/transactionSync() functions?JThat would be with
transaction()/transactionSync()JFor example we could have an outbox table:
- Transaction starts;
- Change/delete a row (Output Gate closed);
- Send the message to an external system (Request blocked by the Output Gate. Deadlock?);
- Receive a response;
- Transaction commit/rollback depending on the response;
JRight now I think we would have to lease the message and send it outside of the transaction, and then the consumer would have to
ack it later.LTLDR: If the callback given to those functions throws, the transaction is rolled back. Anythinh done outside in external systems is up to you to stay consistent.
JSo there is no way currently to make outgoing connections inside a transaction?
JWith SQLite