Activation

By default a production deploy goes live as soon as it's ready. Deferred activation splits that in two, so you can check the new version before any visitor sees it.

Stage, check, go liveCopy Anchor Link

import { Effect, Schedule } from "effect"
import { Deploykit, Platform, Provider } from "@deploykit/core"

export const release = (appId: string, attemptId: string) =>
  Effect.gen(function* () {
    const deploykit = yield* Deploykit.Deploykit

    // 1. Stage.
    const staged = yield* deploykit.deployDirectory(appId, "./my-app", {
      target: "production",
      activation: "deferred",
      operationId: attemptId
    })

    // 2. Wait and check.
    const ready = yield* deploykit.getDeployment(appId, staged.id).pipe(
      Effect.repeat({
        until: d => Provider.isTerminal(d.status),
        schedule: Schedule.spaced("1 second")
      })
    )
    yield* Platform.waitUntilServing(ready.url!)
    yield* runSmokeTests(ready.url!)

    // 3. Go live, then 4. confirm.
    yield* deploykit.activateDeployment(appId, staged.id)
    return yield* deploykit.getActivation(appId, staged.id).pipe(
      Effect.repeat({
        until: a => a.state === "active",
        schedule: Schedule.spaced("2 seconds").pipe(Schedule.upTo({ duration: "1 minute" }))
      })
    )
  })

RulesCopy Anchor Link

• Only a ready production deployment of the same project can be activated. Anything else fails with UnsupportedError. A preview can't be promoted; deploy again with target: "production".
• activateDeployment returning means Vercel accepted the request. Only getActivation returning active confirms traffic moved.
• Activation is never retried automatically. If it fails with an unknown outcome, check getActivation before trying again.
• deploykit doesn't stop two callers activating at once. If that can happen, decide in your own code which one wins, for example with a lock per app.

Rolling backCopy Anchor Link

Activate the previous deployment. It's still a ready production deployment of the project, so the same call applies:

await deploykit.activateDeployment(appId, previousDeploymentId)

Recovering a lost responseCopy Anchor Link

If the connection drops right after Vercel receives a deploy, you get a ProviderError with recovery: "reconcile". The deployment may or may not exist. Don't deploy again. Look it up by the operationId you passed:

const recover = (appId: string, attemptId: string) =>
  Effect.gen(function* () {
    const deploykit = yield* Deploykit.Deploykit
    const result = yield* deploykit.reconcileDeployment(appId, attemptId)
    return result._tag === "Recovered" ? result.deployment : undefined
  })

Recovered only comes back when exactly one deployment matches in a complete search. Anything less certain is Unknown, and the next step is yours: wait and reconcile again, or alert someone.

deploykit tags each deployment with your operationId in its metadata, and searches recent deployments for it. Save the operationId before calling deploy, and the deployment id as soon as you get it. The operation id helps you find a deployment; it doesn't guarantee only one is created.