Doc: how to upgrade workflows

[maelp]

It's not really clear from the doc 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"?

[pranaygp]

Great callout on this. We are currently building something for versioning so that you can upgrade workflows in-flight from one deployment to another

[adpadillar]

On this topic, i don't think the docs are clear how workflows work across deployments. My current understanding is that if a run is in flight and a new version of the app is deployed (and using Vercel World) the workflow will execute with the original code (probably leveraging vercel skew protection)

This still leaves me with a few questions:

1. Is this assumption right at all? I can't seem to find it in docs
2. Is this "keep running against previous versions" implemented at all in the core or is this all in the vercel-world implementation?
3. If implementing a new world (let's assume aws-world), would this need to be implemented using a deployment ID within aws-world?
4. If implementing skew protection in new world implementations (aws-world) were unfeasible, and deployment ID was static (like in local-world), would anything break if new versions of the app were deployed with inflight workflows and no changes to any interfaces? (step in and out shape)

I understand these things might not be covered in docs yet because the library just shipped, so this is just me trying to add to the conversation for when this is eventually considered

[codybrouwers]

I've hit this limitation too. Is there a timeline for a solution?

A few questions:

1. Why use deployment ID instead of deployment URL (which stays stable across deploys)?
2. Could we get a helper function that restarts all active workflows on new deployment and cancels them on the old one? Would this be a good approach?

I also think this limitation should be documented more prominently - it's currently not clear that workflows stay pinned to the old deployment even after deploying new code.

[pranaygp]

Thanks for these great points! i just wrote a short response here with our thinking #99 (comment)

I'll try to write up a longer RFC for this very soon - we're hoping to address this limitation ASAP