Failing Vercel Preview Protection with context.call

[tombeckenham]

DO NOT merge

I have modified email-analyzer-o1 to include the recommended setup for Vercel preview protection. This shows that the first 2 workflow steps work, but the context.call steps fail on vercel preview branches

[tombeckenham]

See screenshot of the failing workflow on preview branches

[CahidArda]

Hi @tombeckenham,

Thanks for bringing this to our attention.

The simplest fix to get around this issue is to pass:

• Upstash-Callback-Forward-X (for context.call)
• Upstash--Failure-Callback-Forward-X (for failure function)

const { POST: handler } = serve<string>(async (context) => {  
  await context.api.openai.call("call", {
    baseURL: "https://testing-ma-feat.requestcatcher.com/",
    body: {
      messages: [],
      model: "gpt-3.5-turbo"
    },
    operation: "chat.completions.create",
    token: "sk-xxxx",
  })
}, {
  qstashClient: new Client({
    token: process.env.QSTASH_TOKEN!,
    baseUrl: process.env.QSTASH_URL!,
    headers: {
      'X-Vercel-Protection-Bypass': 'my-custom-value',
      'upstash-callback-forward-X-Vercel-Protection-Bypass': 'my-custom-value',
      'upstash-failure-callback-forward-X-Vercel-Protection-Bypass': 'my-custom-value'
    }
  })
})

You shouldn't encounter this issue after updating like this.

I will look into whether we can simplify this. If I can't find a way, I will update the docs

[tombeckenham]

I can confirm the workaround works - provided I included the headers on the triggerWorkflow call as well as the qstashClient. Any chance you could make vercel support simpler? I now have to add 4x headers to the trigger call and the serve function. I'm about to look at using webhooks for another workflow. Is there an equivilent header to add for that?

[tombeckenham]

context.waitForWebhook seems to fail as well. You can see from the below screenshot that the response is missing the vercel header.

[CahidArda]

I see. I now understand the issue.

We are planning to change the trigger logic so that the headers passed in trigger with upstash-forward are sent to all requests hitting the endpoint as part of a workflow run. This should fix the issue once and for all.

We are planning to release a fix to resolve this next week.

Let us know if you have any questions!