Move delay for Heroku Preboot to before marking deployment successful (#21736)

* Move delay for Heroku Preboot from post-deploy to pre-success

* Shorten arbitrary delay before first Fastly purge

.github/workflows/prod-build-deploy.yml

@@ -129,9 +129,10 @@ jobs:
           HYDRO_ENDPOINT: ${{ secrets.HYDRO_ENDPOINT }}
           HYDRO_SECRET: ${{ secrets.HYDRO_SECRET }}
           SOURCE_BLOB_URL: ${{ steps.build-source.outputs.download_url }}
+          DELAY_FOR_PREBOOT: 'true'
         with:
           script: |
-            const { GITHUB_TOKEN, HEROKU_API_TOKEN, SOURCE_BLOB_URL } = process.env
+            const { GITHUB_TOKEN, HEROKU_API_TOKEN, SOURCE_BLOB_URL, DELAY_FOR_PREBOOT } = process.env
 
             // Exit if GitHub Actions PAT is not found
             if (!GITHUB_TOKEN) {
@@ -158,6 +159,7 @@ jobs:
             try {
               await deployToProduction({
                 octokit,
+                includeDelayForPreboot: DELAY_FOR_PREBOOT !== 'false',
                 // These parameters will ONLY be set by Actions
                 sourceBlobUrl: SOURCE_BLOB_URL,
                 runId: context.runId
@@ -205,7 +207,6 @@ jobs:
           FASTLY_TOKEN: ${{ secrets.FASTLY_TOKEN }}
           FASTLY_SERVICE_ID: ${{ secrets.FASTLY_SERVICE_ID }}
           FASTLY_SURROGATE_KEY: 'all-the-things'
-          DELAY_FOR_PREBOOT: 'true'
         with:
           script: |
             // Workaround to allow us to load ESM files with `require(...)`
@@ -215,9 +216,7 @@ jobs:
             const { default: purgeEdgeCache } = require('./script/deployment/purge-edge-cache')
 
             try {
-              await purgeEdgeCache({
-                includeDelayForPreboot: process.env.DELAY_FOR_PREBOOT !== 'false'
-              })
+              await purgeEdgeCache()
             } catch (error) {
               console.error(`Failed to purge the edge cache: ${error.message}`)
               console.error(error)

script/deployment/deploy-to-production.js

@@ -6,6 +6,7 @@ import { setOutput } from '@actions/core'
 
 const SLEEP_INTERVAL = 5000
 const HEROKU_LOG_LINES_TO_SHOW = 25
+const DELAY_FOR_PREBOOT_SWAP = 135000 // 2:15
 
 // Allow for a few 404 (Not Found) or 429 (Too Many Requests) responses from the
 // semi-unreliable Heroku API when we're polling for status updates
@@ -13,6 +14,7 @@ const ALLOWED_MISSING_RESPONSE_COUNT = 5
 
 export default async function deployToProduction({
   octokit,
+  includeDelayForPreboot = true,
   // These parameters will only be set by Actions
   sourceBlobUrl = null,
   runId = null,
@@ -332,11 +334,20 @@ export default async function deployToProduction({
       )} seconds.`
     )
 
-    //
-    // TODO:
-    // Should we consider adding an explicit 2-minute pause here to allow for
-    // Heroku Preboot to actually swap in the new dynos?
-    //
+    // IMPORTANT:
+    // If Heroku Preboot is enabled, then there is an additional delay of at
+    // least 2 minutes before the new dynos are swapped into active serving.
+    // If we move off Heroku in the future, this should be revisited and
+    // updated/removed as relevant to align with the new hosting platform.
+    if (includeDelayForPreboot) {
+      console.log(`Waiting for Heroku Preboot to swap dynos (${DELAY_FOR_PREBOOT_SWAP} ms)...`)
+      await sleep(DELAY_FOR_PREBOOT_SWAP)
+
+      // TODO:
+      // Is there a faster alternative than this arbitrary delay? For example,
+      // is there some Heroku API we can query to see when this release is
+      // considered to be the live one, or when the old dynos are shut down?
+    }
 
     // Report success!
     const successMessage = `Deployment succeeded after ${Math.round(

script/deployment/purge-edge-cache.js

@@ -2,7 +2,6 @@ import sleep from 'await-sleep'
 import got from 'got'
 
 const ONE_SECOND = 1000
-const ONE_MINUTE = 60 * ONE_SECOND
 
 async function purgeFastlyBySurrogateKey({ apiToken, serviceId, surrogateKey }) {
   const key = surrogateKey
@@ -17,17 +16,10 @@ async function purgeFastlyBySurrogateKey({ apiToken, serviceId, surrogateKey })
   return got.post(requestPath, { headers, json: true })
 }
 
-// This delay (includeDelayForPreboot) can potentially be removed in the
-// future if the deployment workflow is updated to include a delay to offset
-// Heroku Preboot before this script runs.
-export default async function purgeEdgeCache({ includeDelayForPreboot = true } = {}) {
-  // If Heroku Preboot is enabled, then there is an additional delay of at
-  // least 2 minutes before the new dynos are swapped into active serving.
-  const delayForPrebootSwap = 2 * ONE_MINUTE + 30 * ONE_SECOND
-
+export default async function purgeEdgeCache() {
   // Give the app some extra time to wake up before the thundering herd of
   // Fastly requests.
-  const delayBeforeFirstPurge = ONE_MINUTE
+  const delayBeforeFirstPurge = 30 * ONE_SECOND
 
   // Evidence has shown that it's necessary to purge twice to ensure all
   // customers see fresh content.
@@ -47,11 +39,6 @@ export default async function purgeEdgeCache({ includeDelayForPreboot = true } =
     surrogateKey: FASTLY_SURROGATE_KEY,
   }
 
-  if (includeDelayForPreboot) {
-    console.log('Waiting for Heroku Preboot to swap dynos...')
-    await sleep(delayForPrebootSwap)
-  }
-
   console.log('Waiting extra time to prevent a Thundering Herd problem...')
   await sleep(delayBeforeFirstPurge)