👷 do not rely on hardcoded list of DCs

.gitlab/deploy-auto.yml

@@ -17,7 +17,7 @@ stages:
     - VERSION=$(node -p -e "require('./lerna.json').version")
     - yarn
     - yarn build:bundle
-    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH --check-telemetry-errors
+    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $DATACENTER --check-telemetry-errors
 
 step-1_deploy-prod-minor-dcs:
   when: manual
@@ -26,38 +26,47 @@ step-1_deploy-prod-minor-dcs:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: minor-dcs
+    DATACENTER: minor-dcs
 
-step-2_deploy-prod-eu1:
+step-2_deploy-prod-private-regions:
   needs:
     - step-1_deploy-prod-minor-dcs
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: eu1
+    DATACENTER: private-regions
 
-step-3_deploy-prod-us1:
+step-3_deploy-prod-eu1:
   needs:
-    - step-2_deploy-prod-eu1
+    - step-2_deploy-prod-private-regions
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: us1
+    DATACENTER: eu1
 
-step-4_deploy-prod-gov:
+step-4_deploy-prod-us1:
   needs:
-    - step-3_deploy-prod-us1
+    - step-3_deploy-prod-eu1
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: root
+    DATACENTER: us1
 
-step-5_publish-npm:
+step-5_deploy-prod-gov:
   needs:
-    - step-4_deploy-prod-gov
+    - step-4_deploy-prod-us1
+  extends:
+    - .base-configuration
+    - .deploy-prod
+  variables:
+    DATACENTER: gov
+
+step-6_publish-npm:
+  needs:
+    - step-5_deploy-prod-gov
   stage: deploy
   extends:
     - .base-configuration
@@ -66,9 +75,9 @@ step-5_publish-npm:
     - yarn
     - node ./scripts/deploy/publish-npm.ts
 
-step-6_publish-developer-extension:
+step-7_publish-developer-extension:
   needs:
-    - step-5_publish-npm
+    - step-6_publish-npm
   stage: deploy
   extends:
     - .base-configuration
@@ -77,9 +86,9 @@ step-6_publish-developer-extension:
     - yarn
     - node ./scripts/deploy/publish-developer-extension.ts
 
-step-7_create-github-release:
+step-8_create-github-release:
   needs:
-    - step-6_publish-developer-extension
+    - step-7_publish-developer-extension
   stage: deploy
   extends:
     - .base-configuration

.gitlab/deploy-manual.yml

@@ -19,37 +19,44 @@ stages:
     - VERSION=$(node -p -e "require('./lerna.json').version")
     - yarn
     - yarn build:bundle
-    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH --no-check-telemetry-errors
+    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $DATACENTER --no-check-telemetry-errors
 
 step-1_deploy-prod-minor-dcs:
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: minor-dcs
+    DATACENTER: minor-dcs
 
-step-2_deploy-prod-eu1:
+step-2_deploy-prod-private-regions:
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: eu1
+    DATACENTER: private-regions
 
-step-3_deploy-prod-us1:
+step-3_deploy-prod-eu1:
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: us1
+    DATACENTER: eu1
 
-step-4_deploy-prod-gov:
+step-4_deploy-prod-us1:
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: root
+    DATACENTER: us1
 
-step-5_publish-npm:
+step-5_deploy-prod-gov:
+  extends:
+    - .base-configuration
+    - .deploy-prod
+  variables:
+    DATACENTER: gov
+
+step-6_publish-npm:
   stage: deploy
   extends:
     - .base-configuration
@@ -59,7 +66,7 @@ step-5_publish-npm:
     - yarn
     - node ./scripts/deploy/publish-npm.ts
 
-step-6_publish-developer-extension:
+step-7_publish-developer-extension:
   stage: deploy
   extends:
     - .base-configuration
@@ -69,7 +76,7 @@ step-6_publish-developer-extension:
     - yarn
     - node ./scripts/deploy/publish-developer-extension.ts
 
-step-7_create-github-release:
+step-8_create-github-release:
   stage: deploy
   extends:
     - .base-configuration
@@ -80,7 +87,7 @@ step-7_create-github-release:
     - node scripts/release/create-github-release.ts
 
 # This step is used to deploy the SDK to a new datacenter.
-# the `UPLOAD_PATH` variable needs to be provided as an argument when starting the manual job
+# the `DATACENTER` variable needs to be provided as an argument when starting the manual job
 optional_step-deploy-to-new-datacenter:
   extends:
     - .base-configuration

scripts/deploy/deploy-prod-dc.spec.ts

@@ -3,20 +3,23 @@ import path from 'node:path'
 import { beforeEach, before, describe, it, mock, type Mock } from 'node:test'
 import { browserSdkVersion } from '../lib/browserSdkVersion.ts'
 import type { CommandDetail } from './lib/testHelpers.ts'
-import { mockModule, mockCommandImplementation } from './lib/testHelpers.ts'
+import { mockModule, mockCommandImplementation, mockFetchHandlingError } from './lib/testHelpers.ts'
 
 const currentBrowserSdkVersionMajor = browserSdkVersion.split('.')[0]
 
 describe('deploy-prod-dc', () => {
   const commandMock = mock.fn()
   const checkTelemetryErrorsMock: Mock<(datacenters: string[], version: string) => Promise<void>> = mock.fn()
+  const fetchHandlingErrorMock = mock.fn()
 
   let commands: CommandDetail[]
   let checkTelemetryErrorsCalls: Array<{ version: string; datacenters: string[] }>
 
   before(async () => {
+    mockFetchHandlingError(fetchHandlingErrorMock)
     await mockModule(path.resolve(import.meta.dirname, '../lib/command.ts'), { command: commandMock })
     await mockModule(path.resolve(import.meta.dirname, '../lib/executionUtils.ts'), {
+      fetchHandlingError: fetchHandlingErrorMock,
       timeout: () => Promise.resolve(),
     })
     await mockModule(path.resolve(import.meta.dirname, './lib/checkTelemetryErrors.ts'), {
@@ -62,31 +65,39 @@ describe('deploy-prod-dc', () => {
     ])
   })
 
-  it('should only check telemetry errors before deploying if the upload path is root', async () => {
-    await runScript('./deploy-prod-dc.ts', 'v6', 'root', '--check-telemetry-errors')
+  it('should deploy all minor datacenters', async () => {
+    await runScript('./deploy-prod-dc.ts', 'v6', 'minor-dcs', '--no-check-telemetry-errors')
 
-    // Should only call checkTelemetryErrors once (no gating for root)
-    assert.strictEqual(checkTelemetryErrorsCalls.length, 1)
-    assert.deepEqual(checkTelemetryErrorsCalls[0], {
-      version: `${currentBrowserSdkVersionMajor}.*`,
-      datacenters: ['root'],
-    })
+    // Should not call checkTelemetryErrors when --no-check-telemetry-errors is used
+    assert.strictEqual(checkTelemetryErrorsCalls.length, 0)
 
     assert.deepEqual(commands, [
-      { command: 'node ./scripts/deploy/deploy.ts prod v6 root' },
-      { command: 'node ./scripts/deploy/upload-source-maps.ts v6 root' },
+      { command: 'node ./scripts/deploy/deploy.ts prod v6 ap1,ap2,us3,us5' },
+      { command: 'node ./scripts/deploy/upload-source-maps.ts v6 ap1,ap2,us3,us5' },
     ])
   })
 
-  it('should deploy all minor datacenters', async () => {
-    await runScript('./deploy-prod-dc.ts', 'v6', 'minor-dcs', '--no-check-telemetry-errors')
+  it('should deploy all private regions', async () => {
+    await runScript('./deploy-prod-dc.ts', 'v6', 'private-regions', '--no-check-telemetry-errors')
 
     // Should not call checkTelemetryErrors when --no-check-telemetry-errors is used
     assert.strictEqual(checkTelemetryErrorsCalls.length, 0)
 
     assert.deepEqual(commands, [
-      { command: 'node ./scripts/deploy/deploy.ts prod v6 us3,us5,ap1,ap2,prtest00' },
-      { command: 'node ./scripts/deploy/upload-source-maps.ts v6 us3,us5,ap1,ap2,prtest00' },
+      { command: 'node ./scripts/deploy/deploy.ts prod v6 prtest00,prtest01' },
+      { command: 'node ./scripts/deploy/upload-source-maps.ts v6 prtest00,prtest01' },
+    ])
+  })
+
+  it('should deploy gov datacenters to the root upload path and skip all telemetry error checks', async () => {
+    await runScript('./deploy-prod-dc.ts', 'v6', 'gov', '--check-telemetry-errors')
+
+    // gov datacenters should not be checked for telemetry errors
+    assert.strictEqual(checkTelemetryErrorsCalls.length, 0)
+
+    assert.deepEqual(commands, [
+      { command: 'node ./scripts/deploy/deploy.ts prod v6 root' },
+      { command: 'node ./scripts/deploy/upload-source-maps.ts v6 root' },
     ])
   })
 })

scripts/deploy/deploy-prod-dc.ts

@@ -1,7 +1,7 @@
 import { parseArgs } from 'node:util'
 import { printLog, runMain, timeout } from '../lib/executionUtils.ts'
 import { command } from '../lib/command.ts'
-import { siteByDatacenter } from '../lib/datacenter.ts'
+import { getAllMinorDcs, getAllPrivateDcs } from '../lib/datacenter.ts'
 import { browserSdkVersion } from '../lib/browserSdkVersion.ts'
 import { checkTelemetryErrors } from './lib/checkTelemetryErrors.ts'
 
@@ -14,22 +14,13 @@ const ONE_MINUTE_IN_SECOND = 60
 const GATE_DURATION = 30 * ONE_MINUTE_IN_SECOND
 const GATE_INTERVAL = ONE_MINUTE_IN_SECOND
 
-// Major DCs are the ones that are deployed last.
-// They have their own step jobs in `deploy-manual.yml` and `deploy-auto.yml`.
-const MAJOR_DCS = ['root', 'us1', 'eu1']
-
-// Minor DCs are all the DCs from `siteByDatacenter` that are not in `MAJOR_DCS`.
-function getAllMinorDcs(): string[] {
-  return Object.keys(siteByDatacenter).filter((dc) => !MAJOR_DCS.includes(dc))
-}
-
 if (!process.env.NODE_TEST_CONTEXT) {
   runMain(() => main(...process.argv.slice(2)))
 }
 
 export async function main(...args: string[]): Promise<void> {
   const {
-    values: { 'check-telemetry-errors': shouldCheckTelemetryErrors },
+    values: { 'check-telemetry-errors': checkTelemetryErrorsFlag },
     positionals,
   } = parseArgs({
     args,
@@ -44,32 +35,60 @@ export async function main(...args: string[]): Promise<void> {
   })
 
   const version = positionals[0]
-  const uploadPath = positionals[1] === 'minor-dcs' ? getAllMinorDcs().join(',') : positionals[1]
+  const datacenters = await getDatacenters(positionals[1])
 
-  if (!uploadPath) {
-    throw new Error('UPLOAD_PATH argument is required')
+  if (!datacenters) {
+    throw new Error('DATACENTER argument is required')
   }
 
+  // Skip all telemetry error checks for gov datacenter deployments
+  const shouldCheckTelemetryErrors = checkTelemetryErrorsFlag && !datacenters.every((dc) => dc === 'gov')
+
   if (shouldCheckTelemetryErrors) {
     // Make sure system is in a good state before deploying
     const currentBrowserSdkVersionMajor = browserSdkVersion.split('.')[0]
-    await checkTelemetryErrors(uploadPath.split(','), `${currentBrowserSdkVersionMajor}.*`)
+    await checkTelemetryErrors(datacenters, `${currentBrowserSdkVersionMajor}.*`)
   }
 
-  command`node ./scripts/deploy/deploy.ts prod ${version} ${uploadPath}`.withLogs().run()
-  command`node ./scripts/deploy/upload-source-maps.ts ${version} ${uploadPath}`.withLogs().run()
+  const uploadPathTypes = toDatacenterUploadPathType(datacenters).join(',')
+
+  command`node ./scripts/deploy/deploy.ts prod ${version} ${uploadPathTypes}`.withLogs().run()
+  command`node ./scripts/deploy/upload-source-maps.ts ${version} ${uploadPathTypes}`.withLogs().run()
 
-  if (shouldCheckTelemetryErrors && uploadPath !== 'root') {
-    await gateTelemetryErrors(uploadPath)
+  if (shouldCheckTelemetryErrors) {
+    await gateTelemetryErrors(datacenters)
   }
 }
 
-async function gateTelemetryErrors(uploadPath: string): Promise<void> {
-  printLog(`Check telemetry errors for ${uploadPath} during ${GATE_DURATION / ONE_MINUTE_IN_SECOND} minutes`)
+async function gateTelemetryErrors(datacenters: string[]): Promise<void> {
+  printLog(`Check telemetry errors for ${datacenters.join(',')} during ${GATE_DURATION / ONE_MINUTE_IN_SECOND} minutes`)
   for (let i = 0; i < GATE_DURATION; i += GATE_INTERVAL) {
-    await checkTelemetryErrors(uploadPath.split(','), browserSdkVersion)
+    await checkTelemetryErrors(datacenters, browserSdkVersion)
     process.stdout.write('.') // progress indicator
     await timeout(GATE_INTERVAL * 1000)
   }
+
   printLog() // new line
 }
+
+async function getDatacenters(datacenterGroup: string): Promise<string[]> {
+  if (datacenterGroup === 'minor-dcs') {
+    return await getAllMinorDcs()
+  }
+
+  if (datacenterGroup === 'private-regions') {
+    return await getAllPrivateDcs()
+  }
+
+  return datacenterGroup.split(',')
+}
+
+function toDatacenterUploadPathType(datacenters: string[]): string[] {
+  return datacenters.map((datacenter) => {
+    if (datacenter === 'gov') {
+      return 'root'
+    }
+
+    return datacenter
+  })
+}

scripts/deploy/lib/checkTelemetryErrors.spec.ts

@@ -67,6 +67,21 @@ describe('check-telemetry-errors', () => {
       fetchHandlingError: fetchHandlingErrorMock,
     })
 
+    await mockModule(path.resolve(import.meta.dirname, '../../lib/datacenter.ts'), {
+      getSite: (datacenter: string) => {
+        const siteByDatacenter: Record<string, string> = {
+          us1: 'datadoghq.com',
+          eu1: 'datadoghq.eu',
+          us3: 'us3.datadoghq.com',
+          us5: 'us5.datadoghq.com',
+          ap1: 'ap1.datadoghq.com',
+          ap2: 'ap2.datadoghq.com',
+          prtest00: 'prtest00.datad0g.com',
+        }
+        return Promise.resolve(siteByDatacenter[datacenter])
+      },
+    })
+
     checkTelemetryErrors = (await import('./checkTelemetryErrors.ts')).checkTelemetryErrors
   })
 

scripts/deploy/lib/checkTelemetryErrors.ts

@@ -3,7 +3,7 @@
  */
 import { printLog, fetchHandlingError } from '../../lib/executionUtils.ts'
 import { getTelemetryOrgApiKey, getTelemetryOrgApplicationKey } from '../../lib/secrets.ts'
-import { siteByDatacenter } from '../../lib/datacenter.ts'
+import { getSite } from '../../lib/datacenter.ts'
 
 const TIME_WINDOW_IN_MINUTES = 5
 
@@ -41,7 +41,7 @@ export async function checkTelemetryErrors(datacenters: string[], version: strin
   const queries = getQueries(version)
 
   for (const datacenter of datacenters) {
-    const site = siteByDatacenter[datacenter]
+    const site = await getSite(datacenter)
 
     if (!site) {
       printLog(`No site is configured for datacenter ${datacenter}. skipping...`)