dd-trace 6.17.0 → 6.19.0

package/packages/datadog-instrumentations/src/vitest-worker.js

@@ -4,6 +4,7 @@ const fs = require('node:fs')
 const path = require('node:path')
 const { performance } = require('node:perf_hooks')
 const { fileURLToPath } = require('node:url')
+const { deserialize } = require('node:v8')
 const { isMainThread, parentPort } = require('node:worker_threads')
 
 const { channel } = require('dc-polyfill')
@@ -47,6 +48,7 @@ const {
 
 const EFD_SUITE_ADMISSION_TIMEOUT_MS = 5000
 const logSubmissionFlushCh = channel('ci:log-submission:flush')
+const wrappedTestRunners = new WeakSet()
 const taskToCtx = new WeakMap()
 const taskToTestProperties = new WeakMap()
 const taskToStatuses = new WeakMap()
@@ -412,6 +414,14 @@ function finishEfdSuiteAdmissionRequest (requestId, allowed) {
  * @param {unknown} message
  */
 function handleEfdSuiteAdmissionResponse (message) {
+  // Older Vitest 4 forks serialize parent messages explicitly; newer forks and threads send plain values.
+  if (Buffer.isBuffer(message) || (message?.type === 'Buffer' && Array.isArray(message.data))) {
+    try {
+      message = deserialize(Buffer.from(message))
+    } catch {
+      return
+    }
+  }
   if (!Array.isArray(message) || message[0] !== VITEST_WORKER_EFD_SUITE_ADMISSION_RESPONSE_CODE) return
 
   const { allowed, requestId } = message[1] || {}
@@ -520,6 +530,11 @@ function wrapSuiteHookFn (hookType, fn, fallbackTask) {
 }
 
 function wrapVitestTestRunner (VitestTestRunner) {
+  // ESM hooks can expose the same runner through multiple module identities.
+  // Wrapping its lifecycle methods twice would report each retry twice.
+  if (wrappedTestRunners.has(VitestTestRunner)) return
+  wrappedTestRunners.add(VitestTestRunner)
+
   // `onBeforeRunTask` is run before any repetition or attempt is run
   // `onBeforeRunTask` is an async function
   shimmer.wrap(VitestTestRunner.prototype, 'onBeforeRunTask', onBeforeRunTask => async function (task) {
@@ -623,6 +638,13 @@ function wrapVitestTestRunner (VitestTestRunner) {
   shimmer.wrap(VitestTestRunner.prototype, 'onAfterRunTask', onAfterRunTask => function (task) {
     const { isTestManagementTestsEnabled } = getProvidedContext()
 
+    // Vitest 5 aggregates failed repetitions into the final state. Preserve the last attempt's actual result.
+    if ((attemptToFixTasks.has(task) || efdRetryTasks.has(task)) &&
+        task.result.state === 'fail' && taskToCtx.get(task)?.status === 'pass' &&
+        (task.result.errors?.length ?? 0) === (taskToReportedErrorCount.get(task) ?? 0)) {
+      task.result.state = 'pass'
+    }
+
     if (isTestManagementTestsEnabled) {
       const isAttemptingToFix = attemptToFixTasks.has(task)
       const isQuarantined = quarantinedTasks.has(task)
@@ -745,7 +767,12 @@ function wrapVitestTestRunner (VitestTestRunner) {
       }
     }
 
-    const lastExecutionStatus = task.result.state
+    let lastExecutionStatus = task.result.state
+    if (lastExecutionStatus === 'run' && numRepetition > 0) {
+      // Vitest 5 resets the state before each repetition but retains errors from earlier attempts.
+      const reportedErrorCount = taskToReportedErrorCount.get(task) ?? 0
+      lastExecutionStatus = (task.result.errors?.length ?? 0) > reportedErrorCount ? 'fail' : 'pass'
+    }
     const isAtf = attemptToFixTasks.has(task)
     const isEfd = efdRetryTasks.has(task)
     const shouldTrackStatuses = isEfd || isAtf
@@ -1101,14 +1128,18 @@ function getStartTestsWrapper (frameworkVersion) {
             isDisabled: disabledTasks.has(task),
             isQuarantined: quarantinedTasks.has(task),
           })
-        } else if (state === 'pass' && !isSwitchedStatus) {
+        } else if ((state === 'pass' && !isSwitchedStatus) || switchedStatus === 'pass') {
+          // Attempt-to-fix can pass its last attempt while an earlier failure still fails the suite.
+          if (switchedStatus === 'pass' && errors?.length) {
+            testSuiteError = testError
+          }
           if (testCtx) {
             const isSkippedByTestManagement =
               !attemptToFixTasks.has(task) && (disabledTasks.has(task) || quarantinedTasks.has(task))
             const promises = {}
             testPassCh.publish({
               task,
-              finalStatus: isSkippedByTestManagement ? 'skip' : 'pass',
+              finalStatus: isSkippedByTestManagement ? 'skip' : state,
               earlyFlakeAbortReason: efdSlowAbortedTasks.has(task) ? 'slow' : undefined,
               promises,
               ...testCtx.currentStore,

package/packages/datadog-instrumentations/src/webdriverio.js

@@ -63,6 +63,7 @@ const baseReporterWaitForSyncCh = tracingChannel('orchestrion:@wdio/runner:BaseR
 const runnerRunCh = tracingChannel('orchestrion:@wdio/runner:Runner_run')
 const executeAsyncCh = tracingChannel('orchestrion:@wdio/utils:executeAsync')
 const launcherStartInstanceCh = tracingChannel('orchestrion:@wdio/cli:Launcher_startInstance')
+const configParserShardCh = tracingChannel('orchestrion:@wdio/config:ConfigParser_shard')
 const localRunnerRunCh = tracingChannel('orchestrion:@wdio/local-runner:LocalRunner_run')
 const localRunnerShutdownCh = tracingChannel('orchestrion:@wdio/local-runner:LocalRunner_shutdown')
 const testFrameworkFnWrapperCh = tracingChannel('orchestrion:@wdio/utils:testFrameworkFnWrapper')
@@ -87,6 +88,7 @@ if (loadCh.hasSubscribers) {
 }
 
 const coordinatorStates = new WeakMap()
+const emptyShardConfigurations = new WeakSet()
 const localRunnerVersions = new WeakMap()
 const rumBrowsers = new Set()
 const rumCorrelationBrowsers = new Set()
@@ -936,6 +938,7 @@ function waitForRumTestStart (context) {
  * @property {string} framework
  * @property {string|undefined} rootDir
  * @property {typeof process.env|undefined} runnerEnv
+ * @property {{total: number}|undefined} shard
  */
 
 /**
@@ -1806,17 +1809,18 @@ function getSessionStatus (state) {
 }
 
 /**
- * Returns whether every started worker reported that it discovered no tests.
+ * Explains a skipped session using the workers' discovery results.
  *
  * @param {CoordinatorState} state
  */
-function isExpectedEmptySession (state) {
-  if (state.workers.size === 0) return false
-
+function getTestSessionEmptyReason (state) {
+  if (state.workers.size === 0 && emptyShardConfigurations.has(getRunnerConfiguration(state.localRunner))) {
+    return 'zero_test_shard'
+  }
   for (const workerRecord of state.workers) {
-    if (workerRecord.hasTests !== false) return false
+    if (workerRecord.hasTests !== false) return 'all_tests_skipped'
   }
-  return true
+  return 'zero_tests'
 }
 
 /**
@@ -1832,9 +1836,9 @@ function finishCoordinator (state, error, onDone) {
     return
   }
   if (!state.sessionStarted) {
-    if (!error && getSessionStatus(state) !== 'fail') {
-      onDone()
-      return
+    if (!error && state.workers.size === 0 &&
+      !emptyShardConfigurations.has(getRunnerConfiguration(state.localRunner))) {
+      error = new Error('No test files were found.')
     }
     initializeCoordinator(state, () => finishCoordinator(state, error, onDone))
     return
@@ -1861,7 +1865,7 @@ function finishCoordinator (state, error, onDone) {
   const status = error ? 'fail' : getSessionStatus(state)
   testSessionFinishCh.publish({
     status,
-    isExpectedEmptySession: status === 'skip' && isExpectedEmptySession(state),
+    testSessionEmptyReason: status === 'skip' ? getTestSessionEmptyReason(state) : undefined,
     error,
     isEarlyFlakeDetectionEnabled: state.configuration.isEarlyFlakeDetectionEnabled,
     isEarlyFlakeDetectionFaulty: state.configuration.isEarlyFlakeDetectionFaulty,
@@ -1976,6 +1980,17 @@ jasmineAdapterInitCh.subscribe({
 
 // dc-polyfill supports partial tracing-channel subscribers, unlike the Node.js type definition.
 // @ts-expect-error
+configParserShardCh.subscribe({
+  end (context) {
+    const config = context.self._config
+    if (testFinishCh.hasSubscribers && config?.shard?.total > 1 &&
+      context.arguments[0].length > 0 && context.result?.length === 0) {
+      emptyShardConfigurations.add(config)
+    }
+  },
+})
+
+// @ts-expect-error See the partial tracing-channel subscriber above.
 launcherStartInstanceCh.subscribe({
   start (context) {
     const localRunner = context.self?.runner
@@ -2050,10 +2065,11 @@ localRunnerRunCh.subscribe({
 // @ts-expect-error See the partial tracing-channel subscriber above.
 localRunnerShutdownCh.subscribe({
   asyncEnd (context) {
-    const state = coordinatorStates.get(context.self)
-    if (!state) {
+    const runnerConfiguration = getRunnerConfiguration(context.self)
+    if (!testFinishCh.hasSubscribers || !SUPPORTED_FRAMEWORK_ADAPTERS.has(runnerConfiguration?.framework)) {
       return
     }
+    const state = getCoordinatorState(context.self)
 
     // Orchestrion uses the callback for the matching settlement path to delay LocalRunner.shutdown.
     const waitForCoordinator = onDone => {

package/packages/datadog-instrumentations/src/ws.js

@@ -94,9 +94,22 @@ function wrapSend (send) {
       return send.apply(this, arguments)
     }
 
+    const socket = this._sender?._socket
+    // No socket is attached before the handshake completes or after it is torn down.
+    if (!socket) {
+      return send.apply(this, arguments)
+    }
+
     const [data, options, cb] = arguments
 
-    const ctx = { data, socket: this._sender?._socket }
+    // `ws` stringifies numbers, then spreads `options` over its own inferred default, so an
+    // explicitly passed `binary` wins even when it is nullish and the frame goes out as text.
+    const payload = typeof data === 'number' ? data.toString() : data
+    const binary = options != null && Object.hasOwn(options, 'binary')
+      ? Boolean(options.binary)
+      : typeof payload !== 'string'
+    const byteLength = sentDataLength(payload)
+    const ctx = { data, binary, socket, byteLength }
 
     return typeof cb === 'function'
       ? producerCh.traceCallback(send, undefined, ctx, this, data, options, cb)
@@ -241,6 +254,33 @@ addHook({
   return Receiver
 })
 
+/**
+ * Byte length of an outgoing payload, following the coercion `ws` applies in `Sender.send`:
+ * strings and blobs are measured directly and everything else goes through `toBuffer()`, which
+ * yields one byte per element for plain arrays rather than concatenating them. Blob-likes carry
+ * neither a `byteLength` nor an array shape, so they are matched last to keep the common
+ * buffer path first.
+ *
+ * @param {WebSocketMessageData | number | undefined} data
+ */
+function sentDataLength (data) {
+  if (typeof data === 'string') {
+    return Buffer.byteLength(data)
+  }
+  // An array is one byte per element even when it carries an unrelated `byteLength`.
+  if (Array.isArray(data)) {
+    return data.length
+  }
+  // Covers views as well as `ArrayBuffer` and `SharedArrayBuffer`.
+  if (typeof (/** @type {{ byteLength?: unknown }} */ (data)?.byteLength) === 'number') {
+    return /** @type {{ byteLength: number }} */ (data).byteLength
+  }
+  // Blobs and files, detected the way `ws` does: structurally, so shimmed and cross-realm
+  // implementations it accepts are measured instead of silently reported as empty.
+  const { size } = /** @type {{ size?: unknown }} */ (data) ?? {}
+  return typeof size === 'number' ? size : 0
+}
+
 /**
  * @param {WebSocketMessageData} data
  */

package/packages/datadog-plugin-aws-sdk/src/services/bedrockruntime/utils.js

@@ -23,6 +23,15 @@ const PROVIDER = {
   MISTRAL: 'MISTRAL',
 }
 
+/**
+ * @typedef {{
+ *   inputTokens?: number,
+ *   outputTokens?: number,
+ *   cacheReadTokens?: number,
+ *   cacheWriteTokens?: number,
+ * }} StreamedUsage
+ */
+
 /**
  * Coerce the chunks into a single response body.
  *
@@ -44,10 +53,7 @@ function extractTextAndResponseReasonFromStream (chunks, modelProvider, modelNam
   }
 
   let message = ''
-  let inputTokens = 0
-  let outputTokens = 0
-  let cacheReadTokens = 0
-  let cacheWriteTokens = 0
+  let usage
 
   for (const { chunk: { bytes } } of chunks) {
     const body = JSON.parse(Buffer.from(bytes).toString('utf8'))
@@ -56,8 +62,6 @@ function extractTextAndResponseReasonFromStream (chunks, modelProvider, modelNam
       case PROVIDER.AMAZON: {
         if (body?.outputText) {
           message += body?.outputText
-          inputTokens = body?.inputTextTokenCount
-          outputTokens = body?.totalOutputTextTokenCount
         } else if (body?.contentBlockDelta?.delta?.text) {
           message += body.contentBlockDelta.delta.text
         }
@@ -79,9 +83,6 @@ function extractTextAndResponseReasonFromStream (chunks, modelProvider, modelNam
           message += body.delta.text
         }
 
-        if (body.message?.usage?.input_tokens) inputTokens = body.message.usage.input_tokens
-        if (body.message?.usage?.output_tokens) outputTokens = body.message.usage.output_tokens
-
         break
       }
       case PROVIDER.COHERE: {
@@ -101,24 +102,61 @@ function extractTextAndResponseReasonFromStream (chunks, modelProvider, modelNam
       }
     }
 
-    // by default, it seems newer versions of the AWS SDK include the input/output token counts in the response body
-    const invocationMetrics = body['amazon-bedrock-invocationMetrics']
-    if (invocationMetrics) {
-      inputTokens = invocationMetrics.inputTokenCount
-      outputTokens = invocationMetrics.outputTokenCount
-      cacheReadTokens = invocationMetrics.cacheReadInputTokenCount
-      cacheWriteTokens = invocationMetrics.cacheWriteInputTokenCount
+    usage = mergeStreamedUsage(usage, body, modelProviderUpper)
+  }
+
+  return new Generation({ message, role: 'assistant', ...usage })
+}
+
+/**
+ * Merge the token counts one streamed `invokeModel` frame reports into the running totals. Each
+ * provider spells them differently, and most report none until the trailing invocation metrics,
+ * so a frame carrying no counts leaves the totals as they were.
+ *
+ * @param {StreamedUsage} [usage]
+ * @param {object} body parsed chunk body
+ * @param {string} modelProviderUpper
+ * @returns {StreamedUsage | undefined}
+ */
+function mergeStreamedUsage (usage, body, modelProviderUpper) {
+  // by default, it seems newer versions of the AWS SDK include the input/output token counts in
+  // the response body; any provider can send them, and they supersede what the frames reported
+  const invocationMetrics = body['amazon-bedrock-invocationMetrics']
+  if (invocationMetrics) {
+    return {
+      inputTokens: invocationMetrics.inputTokenCount,
+      outputTokens: invocationMetrics.outputTokenCount,
+      cacheReadTokens: invocationMetrics.cacheReadInputTokenCount,
+      cacheWriteTokens: invocationMetrics.cacheWriteInputTokenCount,
     }
   }
 
-  return new Generation({
-    message,
-    role: 'assistant',
-    inputTokens,
-    outputTokens,
-    cacheReadTokens,
-    cacheWriteTokens,
-  })
+  switch (modelProviderUpper) {
+    case PROVIDER.AMAZON: {
+      const { inputTextTokenCount, totalOutputTextTokenCount } = body
+      if (inputTextTokenCount === undefined && totalOutputTextTokenCount === undefined) return usage
+
+      return {
+        ...usage,
+        inputTokens: inputTextTokenCount ?? usage?.inputTokens,
+        outputTokens: totalOutputTextTokenCount ?? usage?.outputTokens,
+      }
+    }
+    case PROVIDER.ANTHROPIC: {
+      // `message_start` reports the initial counts under `message.usage`; the closing
+      // `message_delta` reports the final output count at the top level
+      const chunkUsage = body.message?.usage ?? body.usage
+      if (!chunkUsage) return usage
+
+      return {
+        ...usage,
+        inputTokens: chunkUsage.input_tokens ?? usage?.inputTokens,
+        outputTokens: chunkUsage.output_tokens ?? usage?.outputTokens,
+      }
+    }
+  }
+
+  return usage
 }
 
 class Generation {
@@ -568,13 +606,27 @@ function resolveToolResultItem (item) {
   return `[Unsupported content type(s): ${getContentBlockType(item)}]`
 }
 
+/**
+ * Normalize a Converse usage object onto the LLMObs metric names. Returns undefined when the
+ * response reported no counts at all, so a caller cannot mistake an empty record for a measurement.
+ *
+ * @param {object} [usage]
+ * @returns {{
+ *   inputTokens?: number, outputTokens?: number, cacheReadTokens?: number, cacheWriteTokens?: number
+ * } | undefined}
+ */
 function buildUsage (usage = {}) {
-  return {
-    inputTokens: usage.inputTokens,
-    outputTokens: usage.outputTokens,
-    cacheReadTokens: usage.cacheReadInputTokens ?? usage.cacheReadInputTokenCount,
-    cacheWriteTokens: usage.cacheWriteInputTokens ?? usage.cacheWriteInputTokenCount,
+  const inputTokens = usage.inputTokens
+  const outputTokens = usage.outputTokens
+  const cacheReadTokens = usage.cacheReadInputTokens ?? usage.cacheReadInputTokenCount
+  const cacheWriteTokens = usage.cacheWriteInputTokens ?? usage.cacheWriteInputTokenCount
+
+  if (inputTokens === undefined && outputTokens === undefined &&
+      cacheReadTokens === undefined && cacheWriteTokens === undefined) {
+    return
   }
+
+  return { inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens }
 }
 
 /**
@@ -714,6 +766,8 @@ function extractTextAndResponseReasonConverseFromStream (chunks) {
 
 module.exports = {
   Generation,
+  buildUsage,
+  mergeStreamedUsage,
   RequestParams,
   extractTextAndResponseReasonFromStream,
   parseModelId,

package/packages/datadog-plugin-cucumber/src/index.js

@@ -73,7 +73,7 @@ class CucumberPlugin extends CiPlugin {
       isEarlyFlakeDetectionFaulty,
       isTestManagementTestsEnabled,
       isParallel,
-      isExpectedEmptySession,
+      testSessionEmptyReason,
       error,
       onDone,
     }) => {
@@ -118,12 +118,11 @@ class CucumberPlugin extends CiPlugin {
 
       this.testSessionSpan.setTag(TEST_STATUS, status)
       this.testModuleSpan.setTag(TEST_STATUS, status)
-      if (isExpectedEmptySession) {
+      if (testSessionEmptyReason) {
         setExpectedEmptyTestSessionTags(
           this.testSessionSpan,
           this.testModuleSpan,
-          'No scenarios were executed',
-          'zero_tests'
+          testSessionEmptyReason
         )
       }
       if (error) {
@@ -450,7 +449,8 @@ class CucumberPlugin extends CiPlugin {
     })
 
     this.addSub('ci:cucumber:is-modified-test', ({
-      scenarios,
+      gherkinNodes,
+      gherkinScopeRanges,
       testFileAbsolutePath,
       modifiedFiles,
       stepIds,
@@ -458,11 +458,25 @@ class CucumberPlugin extends CiPlugin {
       setIsModified,
     }) => {
       const testScenarioPath = getTestSuitePath(testFileAbsolutePath, this.repositoryRoot || process.cwd())
-      for (const scenario of scenarios) {
+      for (const [startLine, endLine] of gherkinScopeRanges) {
+        if (isModifiedTest(testScenarioPath, startLine, endLine, modifiedFiles, 'cucumber')) {
+          setIsModified(true)
+          return
+        }
+      }
+      for (const gherkinNode of gherkinNodes) {
+        const lastStep = gherkinNode.steps.at(-1)
+        let endLine = lastStep?.location.line ?? gherkinNode.location.line
+        if (lastStep?.dataTable?.rows.length) {
+          endLine = lastStep.dataTable.rows.at(-1).location.line
+        } else if (lastStep?.docString) {
+          const { content, location } = lastStep.docString
+          endLine = location.line + (content ? content.split('\n').length + 1 : 1)
+        }
         const isModified = isModifiedTest(
           testScenarioPath,
-          scenario.location.line,
-          scenario.steps.at(-1).location.line,
+          gherkinNode.location.line,
+          endLine,
           modifiedFiles,
           'cucumber'
         )

package/packages/datadog-plugin-cypress/src/cypress-plugin.js

@@ -94,6 +94,7 @@ const {
 const { ORIGIN_KEY, COMPONENT } = require('../../dd-trace/src/constants')
 const { RESOURCE_NAME } = require('../../../ext/tags')
 const getConfig = require('../../dd-trace/src/config')
+const { getValueFromEnvSources } = require('../../dd-trace/src/config/helper')
 const {
   SCREENSHOT_UPLOAD_RESULT_ERROR,
   SCREENSHOT_UPLOAD_RESULT_UPLOADED,
@@ -242,7 +243,7 @@ function getSessionStatus (summary, testSuiteStatuses) {
   if (summary.status === 'failed' || summary.failures > 0 || summary.totalFailed > 0) {
     return 'fail'
   }
-  if (summary.totalSkipped !== undefined && summary.totalSkipped === summary.totalTests) {
+  if ((summary.totalSkipped ?? 0) + (summary.totalPending ?? 0) === summary.totalTests) {
     return 'skip'
   }
   return 'pass'
@@ -381,7 +382,7 @@ function getSuiteStatus (suiteStats) {
     return 'fail'
   }
   if (suiteStats.tests !== undefined &&
-    (suiteStats.tests === suiteStats.pending || suiteStats.tests === suiteStats.skipped)) {
+    suiteStats.tests === (suiteStats.pending ?? 0) + (suiteStats.skipped ?? 0)) {
     return 'skip'
   }
   return 'pass'
@@ -451,6 +452,8 @@ function getFinalStatus ({
 }
 
 class CypressPlugin {
+  #pendingTestSuiteSpans = new Map()
+
   _isInit = false
   testEnvironmentMetadata = getTestEnvironmentMetadata(TEST_FRAMEWORK_NAME)
 
@@ -548,6 +551,7 @@ class CypressPlugin {
    *
    */
   resetRunState () {
+    this.#pendingTestSuiteSpans.clear()
     this._isInit = false
     this.finishedTestsByFile = {}
     this.hasTestsReported = false
@@ -854,6 +858,12 @@ class CypressPlugin {
 
     const testOptimizationConfig = getConfig().testOptimization
     this.rumFlushWaitMillis = testOptimizationConfig.DD_CIVISIBILITY_RUM_FLUSH_WAIT_MILLIS
+    if (getValueFromEnvSources('DD_TEST_FAILURE_SCREENSHOTS_ENABLED', true) === true) {
+      cypressConfig.screenshotOnRunFailure = true
+    }
+    if (getValueFromEnvSources('DD_TEST_FAILURE_VIDEOS_ENABLED', true) === true) {
+      cypressConfig.video = true
+    }
     this.warnIfMisconfiguredTestFailureScreenshots(cypressConfig, tracer, testOptimizationConfig)
     this.warnIfMisconfiguredTestFailureVideos(cypressConfig, tracer, testOptimizationConfig)
 
@@ -1048,7 +1058,7 @@ class CypressPlugin {
       }
     }
 
-    return this.tracer.startSpan(`${TEST_FRAMEWORK_NAME}.test_suite`, {
+    const testSuiteSpan = this.tracer.startSpan(`${TEST_FRAMEWORK_NAME}.test_suite`, {
       childOf: this.testModuleSpan,
       tags: {
         [COMPONENT]: TEST_FRAMEWORK_NAME,
@@ -1057,9 +1067,13 @@ class CypressPlugin {
       },
       integrationName: TEST_FRAMEWORK_NAME,
     })
+    this.#pendingTestSuiteSpans.set(testSuite, testSuiteSpan)
+    return testSuiteSpan
   }
 
-  getTestSpan ({ testName, testSuite, isUnskippable, isForcedToRun, testSourceFile, isDisabled, isQuarantined }) {
+  getTestSpan ({
+    testName, testSuite, isUnskippable, isForcedToRun, testSourceFile, isDisabled, isQuarantined, startTime,
+  }) {
     const testSuiteTags = {
       [TEST_MODULE]: TEST_FRAMEWORK_NAME,
     }
@@ -1116,6 +1130,7 @@ class CypressPlugin {
 
     return this.tracer.startSpan(`${TEST_FRAMEWORK_NAME}.test`, {
       childOf,
+      startTime,
       tags: {
         [COMPONENT]: TEST_FRAMEWORK_NAME,
         [ORIGIN_KEY]: CI_APP_ORIGIN,
@@ -1348,6 +1363,20 @@ class CypressPlugin {
   }
 
   afterRun (suiteStats, error) {
+    const pendingFinalizations = []
+    // Late manual initialization cannot intercept a later replacement of after:spec.
+    // Run results let us finalize the suites whose browser hooks already created spans.
+    if (this._isInit && suiteStats?.runs) {
+      for (const result of suiteStats.runs) {
+        const testSuiteSpan = this.#pendingTestSuiteSpans.get(result.spec?.relative)
+        if (!testSuiteSpan) continue
+        this.testSuiteSpan = testSuiteSpan
+        const endedAt = Date.parse(result.stats?.endedAt)
+        const finalization = this.afterSpec(result.spec, result, undefined,
+          Number.isFinite(endedAt) ? endedAt : undefined)
+        if (finalization) pendingFinalizations.push(finalization)
+      }
+    }
     const hasPendingVideoSpans = this.pendingVideoUploads.length > 0
     const videoUploadsPromise = this.uploadPendingTestSuiteVideos()
     if (!this._isInit) {
@@ -1358,7 +1387,11 @@ class CypressPlugin {
       }
       return
     }
-    const finalizationPromise = this.#finalizeRun(suiteStats, error, hasPendingVideoSpans)
+    // Recovered screenshot-only tests must finish before the run closes any remaining trace spans.
+    const finalizeRun = () => this.#finalizeRun(suiteStats, error, hasPendingVideoSpans)
+    const finalizationPromise = pendingFinalizations.length > 0
+      ? Promise.all(pendingFinalizations).then(finalizeRun)
+      : finalizeRun()
     if (videoUploadsPromise) {
       return Promise.all([videoUploadsPromise, finalizationPromise])
         .then(() => this.#flushExporter(false))
@@ -1384,12 +1417,11 @@ class CypressPlugin {
 
       this.testModuleSpan.setTag(TEST_STATUS, testStatus)
       this.testSessionSpan.setTag(TEST_STATUS, testStatus)
-      if (testStatus !== 'fail' && hasNoTests) {
+      if (testStatus !== 'fail' && (hasNoTests || testStatus === 'skip')) {
         setExpectedEmptyTestSessionTags(
           this.testSessionSpan,
           this.testModuleSpan,
-          'No tests were executed',
-          'zero_tests'
+          hasNoTests ? 'zero_tests' : 'all_tests_skipped'
         )
       }
       if (error) {
@@ -1504,10 +1536,10 @@ class CypressPlugin {
     }
   }
 
-  afterSpec (spec, results, error) {
+  afterSpec (spec, results, error, finishTime) {
     const { tests, stats, screenshots, video, error: resultError } = results || {}
     const cypressTests = tests || []
-    if (cypressTests.length > 0) this.hasTestsReported = true
+    if (cypressTests.length > 0 || stats?.tests > 0) this.hasTestsReported = true
     const specScreenshots = screenshots || []
     const finishedTests = this.finishedTestsByFile[spec.relative] || []
     const screenshotUploadPromises = []
@@ -1540,7 +1572,10 @@ class CypressPlugin {
         ? getTestSuitePath(spec.absolute, this.repositoryRoot)
         : spec.relative
 
-      const skippedTestSpan = this.getTestSpan({ testName: cypressTestName, testSuite: spec.relative, testSourceFile })
+      // Recovered tests that never ran have no execution time; keep them at the recorded suite end.
+      const skippedTestSpan = this.getTestSpan({
+        testName: cypressTestName, testSuite: spec.relative, testSourceFile, startTime: finishTime,
+      })
       skippedTestSpan.setTag(TEST_FINAL_STATUS, 'skip')
 
       skippedTestSpan.setTag(TEST_STATUS, 'skip')
@@ -1578,7 +1613,7 @@ class CypressPlugin {
         })
       }
 
-      testSpanFinishes.push({ testSpan: skippedTestSpan, finishTime: this._now() })
+      testSpanFinishes.push({ testSpan: skippedTestSpan, finishTime: finishTime ?? this._now() })
     }
 
     // Make sure that reported test statuses are the same as Cypress reports.
@@ -1749,7 +1784,7 @@ class CypressPlugin {
       }
     }
 
-    const testSuiteFinishTime = this._now()
+    const testSuiteFinishTime = finishTime ?? this._now()
     const suiteError = error || resultError || latestError
     const suiteStatus = suiteError ||
       cypressTests.some(test => CYPRESS_STATUS_TO_TEST_STATUS[test.state] === 'fail')
@@ -1771,6 +1806,7 @@ class CypressPlugin {
       testSuiteSpan.setTag(TEST_STATUS, suiteStatus)
       if (suiteError) testSuiteSpan.setTag('error', suiteError)
       this.testSuiteSpan = null
+      this.#pendingTestSuiteSpans.delete(spec.relative)
     }
 
     let testSpansPromise
@@ -2034,7 +2070,8 @@ class CypressPlugin {
           rumTestExecutionIdCookieName: RUM_TEST_EXECUTION_ID_COOKIE_NAME,
         }
 
-        this.testSuiteSpan ||= this.getTestSuiteSpan({ testSuite, testSuiteAbsolutePath })
+        this.testSuiteSpan = this.#pendingTestSuiteSpans.get(testSuite) ||
+          this.getTestSuiteSpan({ testSuite, testSuiteAbsolutePath })
         return suitePayload
       },
       'dd:beforeEach': (test) => {