No. That also poses a challenge: how long would we allow? What if termination needs to happen faster

No. That also poses a challenge: how long would we allow? What if termination needs to happen faster or is due to machine failure (which would restart your workflow from the last good step elsewhere), or exceeding CPU or memory?

“Finalizers” are challenging as they set an expectation of 100% but not always the case.

Matt SilverlockOP•11/23/24, 12:24 PM

We might revisit this but not straightforward today.

MMatt Silverlock No. That also poses a challenge: how long would we allow? What if termination ne...

森

森@VerseLink (RW)•11/23/24, 12:36 PM

I see, thank you for your reply.

Alternatively, allow a workflow to "stop" (so we have a separate call workflow.get(id).stop()workflow.get(id).stop()) instead of terminate as a means to softly shutdown the workflow. So the user could gracefully close the workflow (for example the workflow could be a long running polling workflow that doesn't have an end state).

User could decide when the stop process should timeout, and call terminate()terminate() to kill the workflow and abruptly kill the workflow in case it's stuck in a inf-loop and we want to kill it for sure.

森

森@VerseLink (RW)•11/23/24, 12:38 PM

If I'm not mistaken, terminateterminate is the only way to stop a long running / no end-state workflow, so it would be nice to gracefully or signal the workflow to transition to a potential end-state or cleanup

an•11/23/24, 2:23 PM

Hi, currently the cpu time limit on each step is 30s, does that mean the step can live up to 30s?

My step can run up to 10 minutes due to a sleep function in my dependency, but since it's not using cpu I'm not sure I can use step for this

Migrating library to workflow logic is not doable, though.

Aan Hi, currently the cpu time limit on each step is 30s, does that mean the step ca...

Seekerdasbatatas•11/23/24, 4:00 PM

the step can live indefinetly (sleep, wait for I/O, etc..) - those 30 seconds regard actual CPU time: think of it as 30 seconds worth of cpu cycles

Seekerdasbatatas•11/23/24, 4:01 PM

if you are sleeping or waiting for I/O, you are not consuming cpu time

an•11/23/24, 4:39 PM

got it, thank you

laurynas•11/23/24, 9:58 PM

Any thoughts or plans about adding a websocket/SSE or similar reporting endpoint that we could subscribe to when running a Workflow? To my mind I currently need to either poll the .status().status() method (which should be fine for now) or set some outside state from inside the step callback (seems like a code smell).

森

森@VerseLink (RW)•11/24/24, 5:15 AM

btw, Maximum steps per Workflow 256Maximum steps per Workflow 256, does this mean only for Workers Free? what about the limitation for Workers Paid

森

森@VerseLink (RW)•11/24/24, 5:16 AM

because I just realized this would prohibit the idea of using workflow for a long running no end-state polling process

森

森@VerseLink (RW)•11/24/24, 5:16 AM

which workflow could be used for

Llaurynas Any thoughts or plans about adding a websocket/SSE or similar reporting endpoint...

森

森@VerseLink (RW)•11/24/24, 5:24 AM

I really liked the concept of workflow; since workflow should be implemented on top of durable object (correct me if i'm wrong).
It feels like websocket implementation should be possible.
Looking through workflows documentation, it did gave me inspiration to refactor some of my internal codes so that it's a drop in replacement of workflow for Durable Objects.
(and once workflow is matured with more signals or queues then switch back)
The only down side is that the billing would be DO (so wall time and all) but it should give you all the granular steps you could ask for
(start, pause, restart, graceful stop, termination)
Let me know if you are interested!

森

森@VerseLink (RW)•11/24/24, 5:44 AM

Currently I think workflow lacks a few functionality, cf team let me know if there's a better place to post suggestions:

Signal: Signal for the workflow itself to pause / resume / restart / stop; currently there's no way in a workflow to observe any signal natively, my prototype implementation involves using KV to write and read signals between each steps; still a hassle, and terminated workflow has no way to clear KV flags, so a cron job or expiring KV flags are needed.
Long running task: It looks like the workflow has maximum steps limitation, along with the tendency to store all the results in the status, which is good for finite states; but for long, non-end state (unless interrupted) workflow, it might hit the limitation of max steps or max storage eventually since there's no way of clearing or discarding old results.
Events: There's no way for a workflow to natively send events to other processes or back to the client, like SSE or Websocket. The workaround would be to implement would be using Queue or call upon a DO to monitor the results. (which kind of defeats the purpose of workflow).

## Proposal

Signals:

First proposal: Allow the WorkflowEvent<T>WorkflowEvent<T> to change; the workflow would accept the original WorkflowEvent as some of sort constructor (so it could be accessed via this.initEventthis.initEvent etc; then the arguments in the
```
run
```
```
run
```
indicates a modified arguments that the workflow can compare to; a change of value can indicate a signal
Second proposal: Alternatively, allow WorkflowEvent<T>WorkflowEvent<T> to have signalsignal property, this property has special value that could be checked and changed unlike payloadpayload indicating the original value; the signalsignal property can have built-in signals like shouldStopshouldStop to indicate a built-in option for the workflow to be informed that it should gracefully transition into a stop or cancelled state with proper clean up.

### Long running task

First proposal: Allow a previous state results to be discarded or overwritten if the same steps repeat itself (indicating a loop), this should be an opt-in behaviour and WorkflowStepConfigWorkflowStepConfig can accept a option to overrideResultoverrideResult; this may be harder to implement then the second porposal.
Second proposal: Allow discard of staled state results, only preserve the newest 256 results and any older results are discarded. This would transfer the responsibility to preserve old data to new results to the user end; it's easier to be implemented, and requires minimal breaking change, where first proposal can be implementated later.

### Events
Allow a workflow to accept websocket, just like DO.
I don't think there's a good way to persist Response for SSE over step.sleepstep.sleep

DDaniel Jakobsen Hey is there typescript support for the param in the workflow? Right now `Workf...

can•11/25/24, 8:16 AM

isn't params something you pass to the workflow? i believe we have to define it and pass it to the WorkflowEntrypoint generics
https://developers.cloudflare.com/workflows/get-started/guide/

ddsgadget•11/25/24, 1:12 PM

We have a workflow that is stuck in Queued status. We can't do anything to get it out of that status. Pausing it or Terminating the instance in the dashboard does nothing (the UI just hangs). Trying to pause terminate it via the API also throws errors saying that it can moved to a different state because it is in the queed state. How do you delete or terminate a queued state? Seems like you can't get out of it.

DDaniel Jakobsen Hey is there typescript support for the param in the workflow? Right now `Workf...

Matt SilverlockOP•11/25/24, 1:23 PM

unknown

unknown

is correct - you pass your own type: https://developers.cloudflare.com/workflows/build/workers-api/#workflowinstancecreateoptions

Cloudflare Docs

Workers API | Cloudflare Workflows docs

This guide details the Workflows API within Cloudflare Workers, including methods, types, and usage examples.

Dddsgadget We have a workflow that is stuck in Queued status. We can't do anything to get i...

Matt SilverlockOP•11/25/24, 1:23 PM

Possibly a bug (it’s a beta!). Can you share your account ID & instance ID here?

Ccan isn't params something you pass to the workflow? i believe we have to define it ...

Matt SilverlockOP•11/25/24, 1:26 PM

You can also see more here: https://developers.cloudflare.com/workflows/build/events-and-parameters/#typescript-and-type-parameters

What you pass to create({params: YourType})create({params: YourType}) is the payloadpayload property of WorkflowEvent.

You don’t need to cast it with as - you’re passing it, and

as

as

doesn’t tell Workflows anything.

In your Workflow you type the WorkflowEventWorkflowEvent by passing your type as the type parameter as per the docs I linked.

Cloudflare Docs

Events and parameters | Cloudflare Workflows docs

When a Workflow is triggered, it can receive an optional event. This event can include data that your Workflow can act on, including request details, user data fetched from your database (such as D1 or KV) or from a webhook, or messages from a Queue consumer.

MMatt Silverlock Possibly a bug (it’s a beta!). Can you share your account ID & instance ID here?

ddsgadget•11/25/24, 1:58 PM

instance is: 079dd27c-cb60-45c9-a0f1-8bad4b169c97

Dddsgadget instance is: 079dd27c-cb60-45c9-a0f1-8bad4b169c97

Matt SilverlockOP•11/25/24, 2:13 PM

Need both.

ddsgadget•11/25/24, 2:24 PM

you can probably find my account id, i don't want to add it here

Dddsgadget you can probably find my account id, i don't want to add it here

Hello, I’m Allie!•11/25/24, 3:04 PM

Account IDs are safe to share, as without credentials, they cannot be used for anything

DDaniel Jakobsen I think a better type api would be to use the generic type parameters, something...

Matt SilverlockOP•11/25/24, 3:20 PM

Thanks for the PR, although I don't get the benefit here - if you're passing your type as params, the type information doesn't carry through to the Workflow itself (it can't by definition: types don't work like that in JS).

Typing the paramsparams property in the createcreate call doesn't propagate through to WorkflowEventWorkflowEvent because there is a serialization step in the middle.

What are you actually trying to achieve here?

Matt SilverlockOP•11/25/24, 3:22 PM

as

as

isn't a validation either: it casts to that type. If you need to validate the params, use something like

zod

zod

DDaniel Jakobsen Code organization, I have this structure ``` src/index.ts src/route/some_route.t...

Matt SilverlockOP•11/25/24, 3:28 PM

I'm suggesting doing neither: typing the paramsparams property doesn't propagate type information anywhere, because the createcreate call doesn't return anything about your params.

DDaniel Jakobsen Workflows is really powerful! reduces a bunch of headache 🫠 Is it expected to s...

Matt SilverlockOP•11/25/24, 5:17 PM

Yes, we don't take things away from the free tier. It'll just have the limits we advertise: great for prototyping/hobby projects, but if you want to ship something production scale, you'll likely need to upgrade to paid.

HHello, I’m Allie!Account IDs are safe to share, as without credentials, they cannot be used for a...

ddsgadget•11/25/24, 5:29 PM

Yeah, I don't know. i don't feel comfortable sharing Account IDs here. Anyway, the bug is that if a workflow is in a queued status it is impossible to get it out of that status. can't terminate it, pause it or do anything.

ddsgadget•11/25/24, 5:31 PM

BTW, Workflows are simply amazing. This is the greatest thing, in my opinion, to happen to Workers since I started using them a few years ago. I didn't realize it was beta, and I already moved alot of functionality to Workflows and I am just hoping not much changes when it is out of beta.

Dddsgadget Yeah, I don't know. i don't feel comfortable sharing Account IDs here. Anyway, t...

Matt SilverlockOP•11/25/24, 5:44 PM

Account IDs are safe to share. You can't do anything with them. d458dbe698b8eef41837f941d73bc5b3d458dbe698b8eef41837f941d73bc5b3 is mine.

Dddsgadget Yeah, I don't know. i don't feel comfortable sharing Account IDs here. Anyway, t...

Matt SilverlockOP•11/25/24, 5:45 PM

Without an account ID we can't easily track it down: IDs are scoped to an account.

森

森@VerseLink (RW)•11/25/24, 8:45 PM

Uncaught SyntaxError: The requested module 'cloudflare:workers' does not provide an export named 'WorkflowStep'

uhm, did I break something...?

MMatt Silverlock I'm suggesting doing neither: typing the `params` property doesn't propagate typ...

Focux•11/25/24, 9:44 PM

I'd love to use generics when defining the workflow binding so I get the autocomplete when calling createcreate. This would also match other Cloudflare bindings like DurableObjects, which also uses generics.

alex (he/him)•11/26/24, 5:49 AM

Concurrent Workflow instances (executions)Concurrent Workflow instances (executions) is that per Workflow or per Worker Script? So if I define multiple workflows in the same worker script, do all of them have the concurrency. Would Workflow B be blocked by a lot of Workflow A instances?

alex (he/him)•11/26/24, 5:57 AM

And does anyone has thoughts about these two options. Is there any difference between them, and if so what is the difference?

Original message was deleted

森

森@VerseLink (RW)•11/26/24, 6:12 AM

i was on wrangler 3.90.0 (i did try to update it, and delete the .wrangler file)

森

森@VerseLink (RW)•11/26/24, 6:12 AM

i copied the type definition as a hack for now

Aalex (he/him)`Concurrent Workflow instances (executions)` is that per Workflow or per Worker ...

Matt SilverlockOP•11/26/24, 11:44 AM

It’s account wide (all of your Workflows). We’ll be bringing per-Workflow controls, and the limit on the paid plan will increase as we go.

MMatt Silverlock It’s account wide (all of your Workflows). We’ll be bringing per-Workflow contro...

apolodoro•11/26/24, 6:58 PM

Hey Matt, Is sleeping until event (webhook) planned for this year?

Seekerdasbatatas•11/27/24, 6:33 PM

We've bumped up step limits to 512 per workflow: https://developers.cloudflare.com/workflows/reference/limits/

happy thanksgiving to y'all

Cloudflare Docs

Limits | Cloudflare Workflows docs

Limits that apply to authoring, deploying, and running Workflows are detailed below.

kchro3•11/28/24, 1:57 AM

I've seen several comments about this error response:

✘ [ERROR] An RPC stub was not disposed properly. You must call dispose() on all stubs in order to let the other side know that you are no longer using them. You cannot rely on the garbage collector for this because it may take arbitrarily long before actually collecting unreachable objects. As a shortcut, calling dispose() on the result of an RPC call disposes all stubs within it.

✘ [ERROR] An RPC stub was not disposed properly. You must call dispose() on all stubs in order to let the other side know that you are no longer using them. You cannot rely on the garbage collector for this because it may take arbitrarily long before actually collecting unreachable objects. As a shortcut, calling dispose() on the result of an RPC call disposes all stubs within it.

but i don't think anyone's responded to them. Any idea what this is about?

Kkchro3 I've seen several comments about this error response: ``` ✘ [ERROR] An RPC stub...

1984 Ford Laser•11/28/24, 3:04 AM

I think if you need a stub after it has returned its value, you should dup()dup() it then dispose()dispose() of the original?

No. That also poses a challenge: how long would we allow? What if termination needs to happen faster

Signals:

Similar Threads

Similar Threads

Signals:

Similar Threads