at the moment yes, but that's an implementation detail of Workflows so the "minimum" sleep might (an
at the moment yes, but that's an implementation detail of Workflows so the "minimum" sleep might (and probably will) change over time
AIs there a rough guide? Just trying to consider whether hacking around a “wait for event” is worthwhile as I wouldn’t want to have say a 5 second sleep that is effectively polling for something happening and it eat into my concurrency
Sthe minimum sleep required atm to pass it to waiting is in the minutes range (and the transition happens after that time has passed) - so I don't know if it's worthwhile hacking around that. I've personally seen people pausing and resuming an instance to "emulate" events, but on a lighter note, we're working on native events :soontm: 
AThe use case is let’s say it takes 10-30 seconds for something to happen, I don’t want to sleep for say 45 (to be safe) or “poll” at 10 seconds or it would eat concurrency. I could set it to 5 mins, let’s say, but then there’s an unneccessary gap. I have a little solution using a DO (pause workflow -> call DO that sets alarm that effectively does the polling -> DO resumes workflow when event happens) and it works nicely until native events, but wondered if it was worthwhile (but if it’s minutes, I think so for now)
SConsidering that you probably don't have stuff running parallel to the wait for event call, that seems to work nicely
AYeah, what would have been super cool was if I could define the callback (well, the logic to determine if to continue) in the Workflow and pass it to the DO, but alas the “connection” doesn’t naturally survive hibernation 
Ahttps://github.com/apeacock1991/workflows-wait-for-action/ this was surprisingly easy (I wrote it at the weekend, hadn't had the time to check if it was useful until just then) for anyone that wants to pause/resume workflow using a DO and "protect" your concurrency limit
GitHub
Contribute to apeacock1991/workflows-wait-for-action development by creating an account on GitHub.
WHi folks - having some trouble with throwing errors inside a workflow.
First, it seems throwing a NonRetryableError, is retried as many times as specified
Second, I can't seem to find out where the throw error messages are being returned in the workflow status. The docs seem to say
here is the code I'm using to reproduce:
This logs "trying step 2" three times
First, it seems throwing a NonRetryableError, is retried as many times as specified
Second, I can't seem to find out where the throw error messages are being returned in the workflow status. The docs seem to say
const wf = await MY_WORKFLOW.get(id) and then await wf.status() should return errors, but they are always empty.here is the code I'm using to reproduce:
This logs "trying step 2" three times
JYeah, that would def. work-out, but is not that elegant / simple again 
Sticking with using upstash's queueing service for now, where I have 1 scheduled task that calls my api/tasks/generate-reports endpoint every day, which then queries my database and batch-enqueues one task for each report I want to generate & send out
Sticking with using upstash's queueing service for now, where I have 1 scheduled task that calls my api/tasks/generate-reports endpoint every day, which then queries my database and batch-enqueues one task for each report I want to generate & send out
NHi all! I'm running into the following error when using Workflows:
I see that
I'm a little confused because this error is only thrown after some calls to
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.I see that
step.do expects to return Rpc.Serializable<unknown>, so I did a little bit of reading about the RPC lifecycle here and found that in the context of Worker Service Bindings, you should either use the experimental using syntax, or explicitly call object[Symbol.dispose](). I'm a little confused because this error is only thrown after some calls to
step.do, and not others. If someone could provide some insight as to what's going on, that would be great. Happy to provide more details alsoCloudflare Docs
Memory management, resource management, and the lifecycle of RPC stubs.
Tcan anyone help me understand the cost difference of a workflow, while the workflow is running does this add all cpu time during the total run time or just during events that trigger CPU time effectively making the cost that same as using wokers native or roughly the same
I'm curious for scale if I implemented in two ways
Option 1. I send post req with a worker webhook url and then when the job is completed webhook is hit generating new worker instance to handle post job completion duties
Option 2. I create a workflow that for a particular "generation" or request tracks/handles everything, initial http post req as well as waiting for job completion
thank you in advance if anyone can help!
I'm curious for scale if I implemented in two ways
Option 1. I send post req with a worker webhook url and then when the job is completed webhook is hit generating new worker instance to handle post job completion duties
Option 2. I create a workflow that for a particular "generation" or request tracks/handles everything, initial http post req as well as waiting for job completion
thank you in advance if anyone can help!
Ahttps://blog.cloudflare.com/building-workflows-durable-execution-on-workers/#so-how-much-does-it-cost it’s exactly the same as Workers pricing
The Cloudflare Blog
Cloudflare Workflows is now in open beta! Workflows allows you to build reliable, repeatable, long-lived multi-step applications that can automatically retry, persist state, and scale out. Read on to learn how Workflows works, how we built it on top of Durable Objects, and how you can deploy your first Workflows application.
AIf I’m understanding correctly, that’s how Workflows are designed, you can just start a Workflow per file (as an example) and they will run in parallel (up to the max concurrency limit, which is current 100 Workflows at any one time - will likely be raised during beta)
FWhen using
sleep locally with a number value, it treats it like ms instead of secondsFI'm not sure if that's a local development issue or if the documentation is outdated, but it says in the doc that when passing a number value to
sleep it's seconds not ms
Fgood to know, thank you
FI see that there's been an open PR since november to fix the docs, it'd be good to get it merged so people don't get confused by the docs
https://github.com/cloudflare/cloudflare-docs/pull/18369
https://github.com/cloudflare/cloudflare-docs/pull/18369
DHey!
I'm trying to use Workflows within an existing Workers project. The project is using vite to build a remix app.
But I'm getting this error if I try to add a workflows configuration into wrangler.jsonc:
I'm trying to use Workflows within an existing Workers project. The project is using vite to build a remix app.
But I'm getting this error if I try to add a workflows configuration into wrangler.jsonc:
Acan someone confirms AsyncLocalStorage will not behave as expected in a step fn unless the callback is defined inline ?
TCan you just
return?
MOr throw a
NonRetryableError Cloudflare Docs
This guide details the Workflows API within Cloudflare Workers, including methods, types, and usage examples.
JQuestion: has anyone encountered problems doing image transforms in workflows? I'm encountering a strange issue where my image transformations aren't working at all.
I have a pretty basic request, fetching an image from an external domain. I can confirm that the
I'm calling this from within a Workflow, in case that matters. I do have Image Transformation enabled for the zone in which the worker is running. I'm not running this on a
I have a pretty basic request, fetching an image from an external domain. I can confirm that the
Cf-Resized header is nullContent-Type header is un-changed (image/jpeg).I'm calling this from within a Workflow, in case that matters. I do have Image Transformation enabled for the zone in which the worker is running. I'm not running this on a
worker.dev domain, this is in the expected zone (but on a sub-domain). I have confirmed that Image Transformations are working with another worker within the same zone, so the only thing difference is the use of a Workflow. Is there something that I'm missing here? Thanks!
UIt's not possible for a workflow to run on a custom zone it will always run as if in the "workers.dev" zone. So I don't think it's possible to use image transforms through workflows unless they made it possible to use transforms in workers regardless of zone since I last checked.,
JOh interesting - that's very helpful, thank you! This makes me wonder if I could make another worker in the right zone and use service bindings to talk with it?
JThis does make me wonder - is it not possible to have more than 50 subrequests from a workflow, if it's in the workers.dev zone? https://developers.cloudflare.com/workers/platform/limits/
USubrequests goes by your account subscription. As long as you pay $5/mo for workers subscription all the workers limits will go by the workers paid plan.
Image transforms are just tied to zones still and not account so for something like a worker running a workflow it wont work.
Image transforms are just tied to zones still and not account so for something like a worker running a workflow it wont work.
JGotcha - ok, thank you! I'll do some experimentation later today to see if I can get it to work with another worker w/ service bindings - and if not, probably another worker doing a normal
fetch()
THey Everyone! 
Building a Cloudflare-Powered File Sharing App – Seeking Feedback on Architecture
I’m working on a file-sharing application where users can upload files and get a shareable link for others to download or upload additional files. My goal is to make it efficient, scalable, and cost-effective using Cloudflare’s ecosystem.
High-Level Architecture:
Cloudflare R2 stores uploaded files.
Event notifications send messages to a Cloudflare Queue when a file is uploaded.
A Cloudflare Worker consumes these messages and triggers a Cloudflare Workflow to handle file transfers.
The workflow:
Initiates a multipart upload to a cheaper storage bucket.
Reads the R2 object as a stream and uploads it in chunks.
Processes parts in bulk, each bulk handled in a separate step.
Sends a complete command if successful or an abort command if any step fails.
Questions I’d Love Feedback On:
How many workflows can run in parallel?
I trigger the workflow from a worker bound to a queue—how long can a workflow run?
What’s the best approach for handling uploads from the client?
whats counts for sub-requests in a workflow ?
Option 1: Upload to a worker endpoint (via Cloudflare Access) and trigger a workflow to handle multipart upload. Is this possible?
Option 2: Upload directly to R2 using se a PUT request for file under 50 MIB, for larger files use multipart upload:
(a) Temporary credentials (Access Key/Secret Key) scoped to the user's prefix (username).
(b) Pre-signed URLs (short expiration) for uploads/downloads.
(c) Worker with access for authenticated user with zero trust.
Would love to hear your thoughts! Looking forward to the discussion.
Building a Cloudflare-Powered File Sharing App – Seeking Feedback on Architecture
I’m working on a file-sharing application where users can upload files and get a shareable link for others to download or upload additional files. My goal is to make it efficient, scalable, and cost-effective using Cloudflare’s ecosystem.
High-Level Architecture:
Cloudflare R2 stores uploaded files.
Event notifications send messages to a Cloudflare Queue when a file is uploaded.
A Cloudflare Worker consumes these messages and triggers a Cloudflare Workflow to handle file transfers.
The workflow:
Initiates a multipart upload to a cheaper storage bucket.
Reads the R2 object as a stream and uploads it in chunks.
Processes parts in bulk, each bulk handled in a separate step.
Sends a complete command if successful or an abort command if any step fails.
Questions I’d Love Feedback On:
How many workflows can run in parallel? I trigger the workflow from a worker bound to a queue—how long can a workflow run? What’s the best approach for handling uploads from the client? whats counts for sub-requests in a workflow ?Option 1: Upload to a worker endpoint (via Cloudflare Access) and trigger a workflow to handle multipart upload. Is this possible?
Option 2: Upload directly to R2 using se a PUT request for file under 50 MIB, for larger files use multipart upload:
(a) Temporary credentials (Access Key/Secret Key) scoped to the user's prefix (username).
(b) Pre-signed URLs (short expiration) for uploads/downloads.
(c) Worker with access for authenticated user with zero trust.
Would love to hear your thoughts! Looking forward to the discussion.
Tis a step in a workflow can be think of a single worker? does every step have it's own limitation of sub-request and open connection, cpu time, or the quota limitation is for the entire workflow ? its very unclear from the documentation.
SSub-request and open connection limits are per workflow instance. We automatically retry when you reach CPU time and memory limits, but it's not really in a per-step basis (because you can have steps in parallel)
T@Luís Duarte 
and what is an instance of a workflow ?
and what is an instance of a workflow ?
IA single run of the workflow.
II.e. you run the workflow, that's 1 instance, you run it again, that's 2 instances.
IThe CPU time and memory may be shared between runs (the same with any Worker) but the subrequests and concurrent requests are per run.
TSub-request and open connection limits are per workflow instance?
workflow limits are the same of a worker? cause Im looking for sub-requests and open connections here
https://developers.cloudflare.com/workflows/reference/limits/
and i can't find anything explicit
workflow limits are the same of a worker? cause Im looking for sub-requests and open connections here
https://developers.cloudflare.com/workflows/reference/limits/
and i can't find anything explicit
Cloudflare Docs
Limits that apply to authoring, deploying, and running Workflows are detailed below.
Ssubrequest and open connection limits are inherited from the worker limits
T@Seekerdasbatatas if a step retries its counts for another round of this step subrequests from the workflow quota ? :NotLikeThis:
Tim trying to find how to split the file into multipart without hiting the limit
Tstep 1: init multipart upload command
step 2: creating an array of steps, each step does. promise.all([promises of uploading a part]) -> maximum 2 concurent uploads
step 3. complete request
if failed abort reqeust
step 2: creating an array of steps, each step does. promise.all([promises of uploading a part]) -> maximum 2 concurent uploads
step 3. complete request
if failed abort reqeust
Sif you mean R2's multipart upload, you could calculate the part size by dividing the total file size by 990 sub-requests (if you have the file length, i also don't do a 1000 to give you some leeway for other stuff ): https://developers.cloudflare.com/r2/objects/multipart-objects/
Cloudflare Docs
R2 supports S3 API's Multipart Upload with some limitations.
Sthe maximum is 5GiB so using this method you could upload 5TiB (ish)
FHey folks, been using workflows is a lot more lately, mostly as encapsulation of agentic flows.
Wondering if there are either plans or existing strategies to visualize AI Gateway calls/entries related to workflow runs?
(same w/ sub-request or triggering sub-workflows?)
So far, thinking of just adding some metadata to the LLM calls... and then manual filter, but was thinking as you folks have control over the gateway + knowldge of the workflow/worker Instances there was something better.
(Would be awesome to be able to shove metadata/custom data to workflow metrics)
Wondering if there are either plans or existing strategies to visualize AI Gateway calls/entries related to workflow runs?
(same w/ sub-request or triggering sub-workflows?)So far, thinking of just adding some metadata to the LLM calls... and then manual filter, but was thinking as you folks have control over the gateway + knowldge of the workflow/worker Instances there was something better.
(Would be awesome to be able to shove metadata/custom data to workflow metrics)
MWhat kind of metadata?
THey everyone what is the road map of the workflows? is it something i can start using on production ?
TIt's not public. It's beta so inherently no, it's not business critical ready. Yes some people probably use it for prod systems, depends on risk appetite
TIm using one workflow to for multipart upload, and with current limitation the largest file i can support is 15 GiB and not 5TiB file.
taking into consideration 15 minutes time limit, and if every upload part in worst case will retry.
can you take a look at my calculator https://rb.gy/tdpydr, maybe im wrong ? i have seen a lot of question regards limits, it would be nice to create a helper so developers can use this to manage their quota limits.
taking into consideration 15 minutes time limit, and if every upload part in worst case will retry.
can you take a look at my calculator https://rb.gy/tdpydr, maybe im wrong ? i have seen a lot of question regards limits, it would be nice to create a helper so developers can use this to manage their quota limits.
Need to shorten a URL? Use RB.GY for free, memorable shortened links that are easy to track. Ideal for social media, emails, and ads.
