It's more that I would like to have retries on the fetch part only
It's more that I would like to have retries on the fetch part only
MLike, if the KV or validation fails, I don't want to redo the fetching
I know I can implement this myself within the step, but I'm looking to learn Workflows and that would feel unintuitive
I know I can implement this myself within the step, but I'm looking to learn Workflows and that would feel unintuitive
IHow expensive is the fetch?
MIt's to a resource I don't own so I'd rather not do any more than I need to. It's also several fetches per run
Cis there a recommended number of workflows per worker. the indefinite queued state is becoming an issue so i'm exploring have each workflow with it's own worker (not an ideal situation and not sure if it might improve things).
Pis the concurrent execution limit likely to be raised significantly (like over 100k) once the beta is complete? (Trying to figure out if I'm better off using DOs or workflows)
YHey so i have a rag workflow pipeline for a hackathon the issue is the steps are failing with
"Too many subrequests." for generating embeddings and storing it into neon db(its a requirement thats why i cant use service binding to D1). are there any subrequest limits in workflows ??
-free tier account
"Too many subrequests." for generating embeddings and storing it into neon db(its a requirement thats why i cant use service binding to D1). are there any subrequest limits in workflows ??
-free tier account
YOk found it 50 subrequests only
MSame as Workers: 50 on free, 1000 on Paid.
MCloudflare Docs
Limits that apply to authoring, deploying, and running Workflows are detailed below.
YThats so bad
YI think I'll have to invoke other workflows from the current one to bypass it
YI hope ?
MYou still have a daily limit on total invocations (it's a free plan). 50 subrequests per Workflow instance is still a fair few.
YIt's 100 invocations per day right. I need to process 50~200 documents at max so maybe send them in a batch of 25-40 to other workflows
MIt’s 100,000 per day (see the docs I linked)
YNice 
YCan I call the worker's endpoint from the workflow in the same application? How do I use service binding in the same worker.
Does using a service binding like this, I can avoid reaching 50 subrequests limit ?
Does using a service binding like this, I can avoid reaching 50 subrequests limit ?
PKeen to understand what kind of limit is envisioned longer term for "Concurrent Workflow instances" if anyone from CF can speak to that?
Oi don't mean to be rude in any way, but is there any ETA on the ghost executions? i really find workflows to be the best thing for my usecase, but i can't use them if a certain amount of executions just get killed and not even retried
Oor if there is anything on my side that i can do to resolve the issue i'd be happy to do so
MFor Paid: ballpark 1k-5k per Workflow.
DWe are aware of such phenomenon and we are actively working on a fix. Since last week, the chance of getting a ghost execution should have reduced significantly. Further, we are currently testing mechanisms for the system to evict ghost instances. More updates will arrive soon.
OThank you so much!
PThanks good to know - Out of interest, when I look at the DO API, I don't immediately see why it would not be unlimited eventually? Curious what the constraints are
MWe’re still assessing limits. Workflows isn’t just a single DO - we’re doing all of the orchestration, metrics, etc.
Tcan i write workflows in rust
MNot yet. Something we want to land closer to the GA. Focusing on the foundations first.
Tgotcha, good luck ^^
DAre there any updates on getting the Workflow return value through the binding, or Is that on the roadmap?
Adding the Cloudflare API SDK just to get the workflow return value seems to be a bit too much.
I currently have to do a workaround by adding an additional unnecessary step to the end of my workflows, just so I can access the value from the
Adding the Cloudflare API SDK just to get the workflow return value seems to be a bit too much.
I currently have to do a workaround by adding an additional unnecessary step to the end of my workflows, just so I can access the value from the
InstanceStatusoutput
MWhat do you consider the "return" value here? Steps return values. The
runThanks for the reply Matt. The workflow run method definition has an return value that you can type (it isn't defined as Promise<void>) and the Workflow output (what the
run
DIn the Cloudflare API "Get Logs And Status From Instance" endpoint, the workflow output is also returned separately from the "steps" results. But confusingly, the
outputInstanceStatus
MNot quite - something we're iterating on, specifically because returning a single (size limited) string can be confusing here.
Dhmm okay, from my testing the type of the API docs are wrong. It does indeed return an object value
SBut confusingly, the output property of the Binding InstanceStatus contains the step results, not the workflow output.
You mean in local dev, right? what's the behaviour on a real workflow?
Dhmm it seems like you are right! I just tested it again with my deployed workflow and I do indeed get the workflow output there.
Ssorry about that 
it should be an unknown there - i'll look into it but I think it's an openapi 3.0 limitation
it should be an unknown there - i'll look into it but I think it's an openapi 3.0 limitationDThat is good to know! I hope miniflare will be updated soon to support the same behavior
Syhep i'm looking into it right now
Shttps://github.com/cloudflare/workers-sdk/pull/7575 there you go
GitHub
Previously, in local dev, the output field would return the list of successful steps outputs in the workflow. This is not expected behavior compared to production workflows (where the output is the...
Dthat was quick looking forward to it! Good idea to keep the step outputs available for debugging locally
looking forward to it! Good idea to keep the step outputs available for debugging locally
AAny updates or issues I can subscribe/follow to, to track progress on this?
Aseems like there might be some sort of outage? when i create new workflows, they have no steps, and when I try to pause or terminate the instances in the dashboard, those API calls return 500 errors like:
Athey also all have a start time of "pending"
Aoh wait - i think i'm running up against the 100 concurrency limit 
Ai would have assumed that workflows that aren't actually running (and are just sleeping) would not count towards the 100 concurrency limit...?
Ahow long is "eventually"? I'm seeing many instances, whose first step is a simple
await step.sleep('wait 24 hours', '24 hours')this is apparently deadlocking my whole account, since these instances which should be
WaitingRunning