Of all the syntax options they could've gone with, they settled on what I would say is arguably the worst. If you want a one-liner, decorators are widely used across different languages and Typescript supports them as well.
So at it's core this is "just" a toolkit to add automatic retries to functions inside another function?
I don't know if the audience Vercel is targetig knows about idempotency as well as they should before plastering all their functions with "use workflow".
I guess in the end it's another abstraction layer for queues or state machines and another way to lock you into Vercel.
Somewhat related since this about "workflows" and not cloud function, but are there any practical benefits to cloud functions other than the fact that it's cheaper for the providers as they don't have to run an entire bespoke runtime for every customer/application?
This is actually pretty cool. We have a similar custom library at Xbox that's used extensively across all of our services.
I do wish that there was some kind of self-hostable World implementation at launch. If other PAAS providers jump onto this, I could see this sticking around.
Lost me at "use workflow" directive. This and Next16 expanding the set of directives just makes me question if I'm the mad man for thinking they are absolutely terrible.
I'd rather be explicit about what's going on at each step. That way idempotent functions can be handled differently, retry limits can be applied, and no separate preprocessor is required.
export async function welcome(userId: string) {
const user = await retry(() => getUser(userId));
const { subject, body } = await retry(() => generateEmail({
name: user.name, plan: user.plan
}));
const { status } = await retry(() => sendEmail({
to: user.email,
subject,
body,
}), 2);
return { status, subject, body };
}
looking at the docs and examples, I see Workflows and Steps and Retries, but I don't see any Durable yet. none of the examples really make it clear how or where anything gets stored
This seems pretty similar to Cloudflare Workflows (https://developers.cloudflare.com/workflows/), but with code-rewriting to use syntax magic (functions annotated with "use workflow" and "use step") instead of Cloudflare's `step.do` and `step.sleep` function calls. (I think I lightly prefer Cloudflare's model for not relying on a code-rewriting step, partly because I think it's easier for programmers to see what's going on in the system.) Workflow's Hooks are similar to Cloudflare's `step.waitForEvent` + `instance.sendEvent`. It's kind of exciting to see this programming model get more popular. I wonder if the ecosystem can evolve into a standardized model.
So this seems similar to https://temporal.io/, am I reading this right? I used that briefly a few years ago and it was pretty nice at the time. It did make some features much easier to model like their welcome email example. Would love to hear from someone with extensive temporal experience, iirc the only drawback was on the infra side of things.
It's not really clear how you "update" a workflow/step method?
What happens if you saw a bug, or you want to update and change a workflow? Is there a way to discard / upgrade the existing in-memory workflows that are being executed (and correspond to the previous version) so they are now "updated"?
I'm excited about this because durable workflows are really important for making AI applications production ready :) Disclaimer: I'm working on DBOS, a durable workflow library built on Postgres, which looks complementary to this.
I asked their main developer Dillon about the data/durability layer and also the compilation step. I wonder if adding a "DBOS World" will be feasible. That way, you get Postgres-backed durable workflows, queues, messaging, streams, etc all in one package, while the "use workflow" interface remains the same.
Here is the response from Dillon, and I hope it's useful for the discussion here:
> "The primary datastore is dynamodb and is designed to scale to support tens of thousands of v0 size tenants running hundreds of thousands of concurrent workflows and steps."
> "That being said, you don't need to use Vercel as a backend to use the workflow SDK - we have created a interface for anyone to implements called 'World' that you can use any tech stack for https://github.com/vercel/workflow/blob/main/packages/world/..."
> "you will require a compiler step as that's what picks up 'use workflow' and 'use step` and applies source transformations. The node.js run time limitations only apply to the outer wrapper function w/ `use workflow`"
I can almost with 100% certainty see this being one of those things that ultimately, after years of just blantantly ignoring something as simple as basic syntax rules, being redefined to something that is actually valid JavaScript / TypeScript.
> For instance, Math.random and Date constructors are fixed in workflow runs, so you are safe to use them and the framework ensures that the values don't change across replays.
How do you create an environment where everything is deterministic? Do they invoke every supported non deterministic function when creating the environment and rewrite those functions to return the values from the environment's creation time? Is there something more complex happening?
33 comments
[ 3.0 ms ] story [ 46.1 ms ] threadI don’t like having custom bundler logic for my code.
I guess in the end it's another abstraction layer for queues or state machines and another way to lock you into Vercel.
Starting to scatter magic strings throughout a code base feels like a real step back.
At least in any other framework library I can just command click and see why things are not working, place breakpoints and even modify code.
I do wish that there was some kind of self-hostable World implementation at launch. If other PAAS providers jump onto this, I could see this sticking around.
just build state machines folks
looking at the docs and examples, I see Workflows and Steps and Retries, but I don't see any Durable yet. none of the examples really make it clear how or where anything gets stored
"use blackBoxWrapperForEverything";
or is this some extra compilation step to rewrite the code?
What happens if you saw a bug, or you want to update and change a workflow? Is there a way to discard / upgrade the existing in-memory workflows that are being executed (and correspond to the previous version) so they are now "updated"?
- where does the state and telemetry get stored?
- if something is sleeping for 7 days, and you release a new version in that time, what is invoked?
- how do you configure retries? Looks like it retries forever
And I echo the hatred of the magic strings. Terrible dx
I asked their main developer Dillon about the data/durability layer and also the compilation step. I wonder if adding a "DBOS World" will be feasible. That way, you get Postgres-backed durable workflows, queues, messaging, streams, etc all in one package, while the "use workflow" interface remains the same.
Here is the response from Dillon, and I hope it's useful for the discussion here:
> "The primary datastore is dynamodb and is designed to scale to support tens of thousands of v0 size tenants running hundreds of thousands of concurrent workflows and steps."
> "That being said, you don't need to use Vercel as a backend to use the workflow SDK - we have created a interface for anyone to implements called 'World' that you can use any tech stack for https://github.com/vercel/workflow/blob/main/packages/world/..."
> "you will require a compiler step as that's what picks up 'use workflow' and 'use step` and applies source transformations. The node.js run time limitations only apply to the outer wrapper function w/ `use workflow`"
How do you create an environment where everything is deterministic? Do they invoke every supported non deterministic function when creating the environment and rewrite those functions to return the values from the environment's creation time? Is there something more complex happening?