[test optimization] Add Dynamic Instrumentation to jest retries (#4876)

Author: juan-fernandez

LICENSE-3rdparty.csv

@@ -31,6 +31,7 @@ require,retry,MIT,Copyright 2011 Tim Koschützki Felix Geisendörfer
 require,rfdc,MIT,Copyright 2019 David Mark Clements
 require,semver,ISC,Copyright Isaac Z. Schlueter and Contributors
 require,shell-quote,mit,Copyright (c) 2013 James Halliday
+require,source-map,BSD-3-Clause,Copyright (c) 2009-2011, Mozilla Foundation and contributors
 dev,@apollo/server,MIT,Copyright (c) 2016-2020 Apollo Graph, Inc. (Formerly Meteor Development Group, Inc.)
 dev,@types/node,MIT,Copyright Authors
 dev,autocannon,MIT,Copyright 2016 Matteo Collina

integration-tests/ci-visibility/dynamic-instrumentation/dependency.js

@@ -0,0 +1,7 @@
+module.exports = function (a, b) {
+  const localVariable = 2
+  if (a > 10) {
+    throw new Error('a is too big')
+  }
+  return a + b + localVariable - localVariable
+}

integration-tests/ci-visibility/dynamic-instrumentation/test-hit-breakpoint.js

@@ -0,0 +1,15 @@
+/* eslint-disable */
+const sum = require('./dependency')
+
+// TODO: instead of retrying through jest, this should be retried with auto test retries
+jest.retryTimes(1)
+
+describe('dynamic-instrumentation', () => {
+  it('retries with DI', () => {
+    expect(sum(11, 3)).toEqual(14)
+  })
+
+  it('is not retried', () => {
+    expect(sum(1, 2)).toEqual(3)
+  })
+})

integration-tests/ci-visibility/dynamic-instrumentation/test-not-hit-breakpoint.js

@@ -0,0 +1,17 @@
+/* eslint-disable */
+const sum = require('./dependency')
+
+// TODO: instead of retrying through jest, this should be retried with auto test retries
+jest.retryTimes(1)
+
+let count = 0
+describe('dynamic-instrumentation', () => {
+  it('retries with DI', () => {
+    const willFail = count++ === 0
+    if (willFail) {
+      expect(sum(11, 3)).toEqual(14) // only throws the first time
+    } else {
+      expect(sum(1, 2)).toEqual(3)
+    }
+  })
+})

integration-tests/jest/jest.spec.js

@@ -33,7 +33,11 @@ const {
   TEST_SOURCE_START,
   TEST_CODE_OWNERS,
   TEST_SESSION_NAME,
-  TEST_LEVEL_EVENT_TYPES
+  TEST_LEVEL_EVENT_TYPES,
+  DI_ERROR_DEBUG_INFO_CAPTURED,
+  DI_DEBUG_ERROR_FILE,
+  DI_DEBUG_ERROR_SNAPSHOT_ID,
+  DI_DEBUG_ERROR_LINE
 } = require('../../packages/dd-trace/src/plugins/util/test')
 const { DD_HOST_CPU_COUNT } = require('../../packages/dd-trace/src/plugins/util/env')
 const { ERROR_MESSAGE } = require('../../packages/dd-trace/src/constants')
@@ -2399,4 +2403,205 @@ describe('jest CommonJS', () => {
       })
     })
   })
+
+  context('dynamic instrumentation', () => {
+    it('does not activate dynamic instrumentation if DD_TEST_DYNAMIC_INSTRUMENTATION_ENABLED is not set', (done) => {
+      receiver.setSettings({
+        itr_enabled: false,
+        code_coverage: false,
+        tests_skipping: false,
+        flaky_test_retries_enabled: true,
+        early_flake_detection: {
+          enabled: false
+        }
+        // di_enabled: true // TODO
+      })
+      const eventsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/citestcycle'), (payloads) => {
+          const events = payloads.flatMap(({ payload }) => payload.events)
+
+          const tests = events.filter(event => event.type === 'test').map(event => event.content)
+          const retriedTests = tests.filter(test => test.meta[TEST_IS_RETRY] === 'true')
+
+          assert.equal(retriedTests.length, 1)
+          const [retriedTest] = retriedTests
+
+          assert.notProperty(retriedTest.meta, DI_ERROR_DEBUG_INFO_CAPTURED)
+          assert.notProperty(retriedTest.meta, DI_DEBUG_ERROR_FILE)
+          assert.notProperty(retriedTest.metrics, DI_DEBUG_ERROR_LINE)
+          assert.notProperty(retriedTest.meta, DI_DEBUG_ERROR_SNAPSHOT_ID)
+        })
+      const logsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/logs'), (payloads) => {
+          if (payloads.length > 0) {
+            throw new Error('Unexpected logs')
+          }
+        }, 5000)
+
+      childProcess = exec(runTestsWithCoverageCommand,
+        {
+          cwd,
+          env: {
+            ...getCiVisAgentlessConfig(receiver.port),
+            TESTS_TO_RUN: 'dynamic-instrumentation/test-hit-breakpoint'
+          },
+          stdio: 'inherit'
+        }
+      )
+
+      childProcess.on('exit', (code) => {
+        Promise.all([eventsPromise, logsPromise]).then(() => {
+          assert.equal(code, 0)
+          done()
+        }).catch(done)
+      })
+    })
+
+    it('runs retries with dynamic instrumentation', (done) => {
+      receiver.setSettings({
+        itr_enabled: false,
+        code_coverage: false,
+        tests_skipping: false,
+        flaky_test_retries_enabled: true,
+        early_flake_detection: {
+          enabled: false
+        }
+        // di_enabled: true // TODO
+      })
+      let snapshotIdByTest, snapshotIdByLog
+      let spanIdByTest, spanIdByLog, traceIdByTest, traceIdByLog
+      const eventsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/citestcycle'), (payloads) => {
+          const events = payloads.flatMap(({ payload }) => payload.events)
+
+          const tests = events.filter(event => event.type === 'test').map(event => event.content)
+          const retriedTests = tests.filter(test => test.meta[TEST_IS_RETRY] === 'true')
+
+          assert.equal(retriedTests.length, 1)
+          const [retriedTest] = retriedTests
+
+          assert.propertyVal(retriedTest.meta, DI_ERROR_DEBUG_INFO_CAPTURED, 'true')
+          assert.propertyVal(
+            retriedTest.meta,
+            DI_DEBUG_ERROR_FILE,
+            'ci-visibility/dynamic-instrumentation/dependency.js'
+          )
+          assert.equal(retriedTest.metrics[DI_DEBUG_ERROR_LINE], 4)
+          assert.exists(retriedTest.meta[DI_DEBUG_ERROR_SNAPSHOT_ID])
+
+          snapshotIdByTest = retriedTest.meta[DI_DEBUG_ERROR_SNAPSHOT_ID]
+          spanIdByTest = retriedTest.span_id.toString()
+          traceIdByTest = retriedTest.trace_id.toString()
+
+          const notRetriedTest = tests.find(test => test.meta[TEST_NAME].includes('is not retried'))
+
+          assert.notProperty(notRetriedTest.meta, DI_ERROR_DEBUG_INFO_CAPTURED)
+        })
+
+      const logsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/logs'), (payloads) => {
+          const [{ logMessage: [diLog] }] = payloads
+          assert.deepInclude(diLog, {
+            ddsource: 'dd_debugger',
+            level: 'error'
+          })
+          assert.equal(diLog.debugger.snapshot.language, 'javascript')
+          assert.deepInclude(diLog.debugger.snapshot.captures.lines['4'].locals, {
+            a: {
+              type: 'number',
+              value: '11'
+            },
+            b: {
+              type: 'number',
+              value: '3'
+            },
+            localVariable: {
+              type: 'number',
+              value: '2'
+            }
+          })
+          spanIdByLog = diLog.dd.span_id
+          traceIdByLog = diLog.dd.trace_id
+          snapshotIdByLog = diLog.debugger.snapshot.id
+        })
+
+      childProcess = exec(runTestsWithCoverageCommand,
+        {
+          cwd,
+          env: {
+            ...getCiVisAgentlessConfig(receiver.port),
+            TESTS_TO_RUN: 'dynamic-instrumentation/test-hit-breakpoint',
+            DD_TEST_DYNAMIC_INSTRUMENTATION_ENABLED: 'true'
+          },
+          stdio: 'inherit'
+        }
+      )
+
+      childProcess.on('exit', () => {
+        Promise.all([eventsPromise, logsPromise]).then(() => {
+          assert.equal(snapshotIdByTest, snapshotIdByLog)
+          assert.equal(spanIdByTest, spanIdByLog)
+          assert.equal(traceIdByTest, traceIdByLog)
+          done()
+        }).catch(done)
+      })
+    })
+
+    it('does not crash if the retry does not hit the breakpoint', (done) => {
+      receiver.setSettings({
+        itr_enabled: false,
+        code_coverage: false,
+        tests_skipping: false,
+        flaky_test_retries_enabled: true,
+        early_flake_detection: {
+          enabled: false
+        }
+        // di_enabled: true // TODO
+      })
+      const eventsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/citestcycle'), (payloads) => {
+          const events = payloads.flatMap(({ payload }) => payload.events)
+
+          const tests = events.filter(event => event.type === 'test').map(event => event.content)
+          const retriedTests = tests.filter(test => test.meta[TEST_IS_RETRY] === 'true')
+
+          assert.equal(retriedTests.length, 1)
+          const [retriedTest] = retriedTests
+
+          assert.propertyVal(retriedTest.meta, DI_ERROR_DEBUG_INFO_CAPTURED, 'true')
+          assert.propertyVal(
+            retriedTest.meta,
+            DI_DEBUG_ERROR_FILE,
+            'ci-visibility/dynamic-instrumentation/dependency.js'
+          )
+          assert.equal(retriedTest.metrics[DI_DEBUG_ERROR_LINE], 4)
+          assert.exists(retriedTest.meta[DI_DEBUG_ERROR_SNAPSHOT_ID])
+        })
+      const logsPromise = receiver
+        .gatherPayloadsMaxTimeout(({ url }) => url.endsWith('/api/v2/logs'), (payloads) => {
+          if (payloads.length > 0) {
+            throw new Error('Unexpected logs')
+          }
+        }, 5000)
+
+      childProcess = exec(runTestsWithCoverageCommand,
+        {
+          cwd,
+          env: {
+            ...getCiVisAgentlessConfig(receiver.port),
+            TESTS_TO_RUN: 'dynamic-instrumentation/test-not-hit-breakpoint',
+            DD_TEST_DYNAMIC_INSTRUMENTATION_ENABLED: 'true'
+          },
+          stdio: 'inherit'
+        }
+      )
+
+      childProcess.on('exit', (code) => {
+        Promise.all([eventsPromise, logsPromise]).then(() => {
+          assert.equal(code, 0)
+          done()
+        }).catch(done)
+      })
+    })
+  })
 })

package.json

@@ -113,6 +113,7 @@
     "rfdc": "^1.3.1",
     "semver": "^7.5.4",
     "shell-quote": "^1.8.1",
+    "source-map": "^0.7.4",
     "tlhunter-sorted-set": "^0.1.0"
   },
   "devDependencies": {

packages/datadog-instrumentations/src/jest.js

@@ -237,7 +237,6 @@ function getWrappedEnvironment (BaseEnvironment, jestVersion) {
             name: removeEfdStringFromTestName(testName),
             suite: this.testSuite,
             testSourceFile: this.testSourceFile,
-            runner: 'jest-circus',
             displayName: this.displayName,
             testParameters,
             frameworkVersion: jestVersion,
@@ -274,13 +273,18 @@ function getWrappedEnvironment (BaseEnvironment, jestVersion) {
         }
       }
       if (event.name === 'test_done') {
+        const probe = {}
         const asyncResource = asyncResources.get(event.test)
         asyncResource.runInAsyncScope(() => {
           let status = 'pass'
           if (event.test.errors && event.test.errors.length) {
             status = 'fail'
-            const formattedError = formatJestError(event.test.errors[0])
-            testErrCh.publish(formattedError)
+            const numRetries = this.global[RETRY_TIMES]
+            const numTestExecutions = event.test?.invocations
+            const willBeRetried = numRetries > 0 && numTestExecutions - 1 < numRetries
+
+            const error = formatJestError(event.test.errors[0])
+            testErrCh.publish({ error, willBeRetried, probe, numTestExecutions })
           }
           testRunFinishCh.publish({
             status,
@@ -302,6 +306,9 @@ function getWrappedEnvironment (BaseEnvironment, jestVersion) {
             }
           }
         })
+        if (probe.setProbePromise) {
+          await probe.setProbePromise
+        }
       }
       if (event.name === 'test_skip' || event.name === 'test_todo') {
         const asyncResource = new AsyncResource('bound-anonymous-fn')
@@ -310,7 +317,6 @@ function getWrappedEnvironment (BaseEnvironment, jestVersion) {
             name: getJestTestName(event.test),
             suite: this.testSuite,
             testSourceFile: this.testSourceFile,
-            runner: 'jest-circus',
             displayName: this.displayName,
             frameworkVersion: jestVersion,
             testStartLine: getTestLineStart(event.test.asyncError, this.testSuite)