dd-trace 4.18.0 → 4.22.0

package/packages/dd-trace/src/profiling/profilers/events.js

@@ -0,0 +1,264 @@
+const { performance, constants, PerformanceObserver } = require('node:perf_hooks')
+const { END_TIMESTAMP_LABEL } = require('./shared')
+const semver = require('semver')
+const { Function, Label, Line, Location, Profile, Sample, StringTable, ValueType } = require('pprof-format')
+const pprof = require('@datadog/pprof/')
+
+// Format of perf_hooks events changed with Node 16, we need to be mindful of it.
+const node16 = semver.gte(process.version, '16.0.0')
+
+// perf_hooks uses millis, with fractional part representing nanos. We emit nanos into the pprof file.
+const MS_TO_NS = 1000000
+
+// While this is an "events profiler", meaning it emits a pprof file based on events observed as
+// perf_hooks events, the emitted pprof file uses the type "timeline".
+const pprofValueType = 'timeline'
+const pprofValueUnit = 'nanoseconds'
+
+function labelFromStr (stringTable, key, valStr) {
+  return new Label({ key, str: stringTable.dedup(valStr) })
+}
+
+function labelFromStrStr (stringTable, keyStr, valStr) {
+  return labelFromStr(stringTable, stringTable.dedup(keyStr), valStr)
+}
+
+class GCDecorator {
+  constructor (stringTable) {
+    this.stringTable = stringTable
+    this.reasonLabelKey = stringTable.dedup('gc reason')
+    this.kindLabels = []
+    this.reasonLabels = []
+    this.flagObj = {}
+
+    const kindLabelKey = stringTable.dedup('gc type')
+
+    // Create labels for all GC performance flags and kinds of GC
+    for (const [key, value] of Object.entries(constants)) {
+      if (key.startsWith('NODE_PERFORMANCE_GC_FLAGS_')) {
+        this.flagObj[key.substring(26).toLowerCase()] = value
+      } else if (key.startsWith('NODE_PERFORMANCE_GC_')) {
+        // It's a constant for a kind of GC
+        const kind = key.substring(20).toLowerCase()
+        this.kindLabels[value] = labelFromStr(stringTable, kindLabelKey, kind)
+      }
+    }
+  }
+
+  decorateSample (sampleInput, item) {
+    const { kind, flags } = node16 ? item.detail : item
+    sampleInput.label.push(this.kindLabels[kind])
+    const reasonLabel = this.getReasonLabel(flags)
+    if (reasonLabel) {
+      sampleInput.label.push(reasonLabel)
+    }
+  }
+
+  getReasonLabel (flags) {
+    if (flags === 0) {
+      return null
+    }
+    let reasonLabel = this.reasonLabels[flags]
+    if (!reasonLabel) {
+      const reasons = []
+      for (const [key, value] of Object.entries(this.flagObj)) {
+        if (value & flags) {
+          reasons.push(key)
+        }
+      }
+      const reasonStr = reasons.join(',')
+      reasonLabel = labelFromStr(this.stringTable, this.reasonLabelKey, reasonStr)
+      this.reasonLabels[flags] = reasonLabel
+    }
+    return reasonLabel
+  }
+}
+
+class DNSDecorator {
+  constructor (stringTable) {
+    this.stringTable = stringTable
+    this.operationNameLabelKey = stringTable.dedup('operation')
+    this.hostLabelKey = stringTable.dedup('host')
+    this.addressLabelKey = stringTable.dedup('address')
+    this.portLabelKey = stringTable.dedup('port')
+  }
+
+  decorateSample (sampleInput, item) {
+    const labels = sampleInput.label
+    const stringTable = this.stringTable
+    function addLabel (labelNameKey, labelValue) {
+      labels.push(labelFromStr(stringTable, labelNameKey, labelValue))
+    }
+    const op = item.name
+    addLabel(this.operationNameLabelKey, item.name)
+    const detail = item.detail
+    switch (op) {
+      case 'lookup':
+        addLabel(this.hostLabelKey, detail.hostname)
+        break
+      case 'lookupService':
+        addLabel(this.addressLabelKey, detail.host)
+        labels.push(new Label({ key: this.portLabelKey, num: detail.port }))
+        break
+      case 'getHostByAddr':
+        addLabel(this.addressLabelKey, detail.host)
+        break
+      default:
+        if (op.startsWith('query')) {
+          addLabel(this.hostLabelKey, detail.host)
+        }
+    }
+  }
+}
+
+class NetDecorator {
+  constructor (stringTable) {
+    this.stringTable = stringTable
+    this.operationNameLabelKey = stringTable.dedup('operation')
+    this.hostLabelKey = stringTable.dedup('host')
+    this.portLabelKey = stringTable.dedup('port')
+  }
+
+  decorateSample (sampleInput, item) {
+    const labels = sampleInput.label
+    const stringTable = this.stringTable
+    function addLabel (labelNameKey, labelValue) {
+      labels.push(labelFromStr(stringTable, labelNameKey, labelValue))
+    }
+    const op = item.name
+    addLabel(this.operationNameLabelKey, op)
+    if (op === 'connect') {
+      const detail = item.detail
+      addLabel(this.hostLabelKey, detail.host)
+      labels.push(new Label({ key: this.portLabelKey, num: detail.port }))
+    }
+  }
+}
+
+// Keys correspond to PerformanceEntry.entryType, values are constructor
+// functions for type-specific decorators.
+const decoratorTypes = {
+  gc: GCDecorator
+}
+// Needs at least node 16 for DNS and Net
+if (node16) {
+  decoratorTypes.dns = DNSDecorator
+  decoratorTypes.net = NetDecorator
+}
+
+/**
+ * This class generates pprof files with timeline events sourced from Node.js
+ * performance measurement APIs.
+ */
+class EventsProfiler {
+  constructor (options = {}) {
+    this.type = 'events'
+    this._flushIntervalNanos = (options.flushInterval || 60000) * 1e6 // 60 sec
+    this._observer = undefined
+    this.entries = []
+  }
+
+  start () {
+    function add (items) {
+      this.entries.push(...items.getEntries())
+    }
+    if (!this._observer) {
+      this._observer = new PerformanceObserver(add.bind(this))
+    }
+    this._observer.observe({ entryTypes: Object.keys(decoratorTypes) })
+  }
+
+  stop () {
+    if (this._observer) {
+      this._observer.disconnect()
+    }
+  }
+
+  profile (startDate, endDate) {
+    if (this.entries.length === 0) {
+      // No events in the period; don't produce a profile
+      return null
+    }
+
+    const stringTable = new StringTable()
+    const locations = []
+    const functions = []
+
+    // A synthetic single-frame location to serve as the location for timeline
+    // samples. We need these as the profiling backend (mimicking official pprof
+    // tool's behavior) ignores these.
+    const locationId = (() => {
+      const fn = new Function({ id: functions.length + 1, name: stringTable.dedup('') })
+      functions.push(fn)
+      const line = new Line({ functionId: fn.id })
+      const location = new Location({ id: locations.length + 1, line: [line] })
+      locations.push(location)
+      return [location.id]
+    })()
+
+    const decorators = {}
+    for (const [eventType, DecoratorCtor] of Object.entries(decoratorTypes)) {
+      const decorator = new DecoratorCtor(stringTable)
+      decorator.eventTypeLabel = labelFromStrStr(stringTable, 'event', eventType)
+      decorators[eventType] = decorator
+    }
+    const timestampLabelKey = stringTable.dedup(END_TIMESTAMP_LABEL)
+
+    const dateOffset = BigInt(Math.round(performance.timeOrigin * MS_TO_NS))
+    const lateEntries = []
+    const perfEndDate = endDate.getTime() - performance.timeOrigin
+    const samples = this.entries.map((item) => {
+      const decorator = decorators[item.entryType]
+      if (!decorator) {
+        // Shouldn't happen but it's better to not rely on observer only getting
+        // requested event types.
+        return null
+      }
+      const { startTime, duration } = item
+      if (startTime >= perfEndDate) {
+        // An event past the current recording end date; save it for the next
+        // profile. Not supposed to happen as long as there's no async activity
+        // between capture of the endDate value in profiler.js _collect() and
+        // here, but better be safe than sorry.
+        lateEntries.push(item)
+        return null
+      }
+      const endTime = startTime + duration
+      const sampleInput = {
+        value: [Math.round(duration * MS_TO_NS)],
+        locationId,
+        label: [
+          decorator.eventTypeLabel,
+          new Label({ key: timestampLabelKey, num: dateOffset + BigInt(Math.round(endTime * MS_TO_NS)) })
+        ]
+      }
+      decorator.decorateSample(sampleInput, item)
+      return new Sample(sampleInput)
+    }).filter(v => v)
+
+    this.entries = lateEntries
+
+    const timeValueType = new ValueType({
+      type: stringTable.dedup(pprofValueType),
+      unit: stringTable.dedup(pprofValueUnit)
+    })
+
+    return new Profile({
+      sampleType: [timeValueType],
+      timeNanos: endDate.getTime() * MS_TO_NS,
+      periodType: timeValueType,
+      period: 1,
+      durationNanos: (endDate.getTime() - startDate.getTime()) * MS_TO_NS,
+      sample: samples,
+      location: locations,
+      function: functions,
+      stringTable: stringTable
+    })
+  }
+
+  encode (profile) {
+    return pprof.encode(profile)
+  }
+}
+
+module.exports = EventsProfiler

package/packages/dd-trace/src/profiling/profilers/shared.js

@@ -0,0 +1,39 @@
+'use strict'
+
+const { isMainThread, threadId } = require('node:worker_threads')
+
+const END_TIMESTAMP_LABEL = 'end_timestamp_ns'
+const THREAD_NAME_LABEL = 'thread name'
+const OS_THREAD_ID_LABEL = 'os thread id'
+const THREAD_ID_LABEL = 'thread id'
+const threadNamePrefix = isMainThread ? 'Main' : `Worker #${threadId}`
+const eventLoopThreadName = `${threadNamePrefix} Event Loop`
+
+function getThreadLabels () {
+  const pprof = require('@datadog/pprof')
+  const nativeThreadId = pprof.getNativeThreadId()
+  return {
+    [THREAD_NAME_LABEL]: eventLoopThreadName,
+    [THREAD_ID_LABEL]: `${threadId}`,
+    [OS_THREAD_ID_LABEL]: `${nativeThreadId}`
+  }
+}
+
+function cacheThreadLabels () {
+  let labels
+  return () => {
+    if (!labels) {
+      labels = getThreadLabels()
+    }
+    return labels
+  }
+}
+
+module.exports = {
+  END_TIMESTAMP_LABEL,
+  THREAD_NAME_LABEL,
+  THREAD_ID_LABEL,
+  threadNamePrefix,
+  eventLoopThreadName,
+  getThreadLabels: cacheThreadLabels()
+}

package/packages/dd-trace/src/profiling/profilers/space.js

@@ -1,6 +1,7 @@
 'use strict'
 
 const { oomExportStrategies } = require('../constants')
+const { getThreadLabels } = require('./shared')
 
 function strategiesToCallbackMode (strategies, callbackMode) {
   return strategies.includes(oomExportStrategies.ASYNC_CALLBACK) ? callbackMode.Async : 0
@@ -33,7 +34,7 @@ class NativeSpaceProfiler {
   }
 
   profile () {
-    return this._pprof.heap.profile(undefined, this._mapper)
+    return this._pprof.heap.profile(undefined, this._mapper, getThreadLabels)
   }
 
   encode (profile) {

package/packages/dd-trace/src/profiling/profilers/wall.js

@@ -2,21 +2,19 @@
 
 const { storage } = require('../../../../datadog-core')
 
-const dc = require('../../../../diagnostics_channel')
+const dc = require('dc-polyfill')
 const { HTTP_METHOD, HTTP_ROUTE, RESOURCE_NAME, SPAN_TYPE } = require('../../../../../ext/tags')
 const { WEB } = require('../../../../../ext/types')
 const runtimeMetrics = require('../../runtime_metrics')
 const telemetryMetrics = require('../../telemetry/metrics')
+const { END_TIMESTAMP_LABEL, getThreadLabels } = require('./shared')
 
 const beforeCh = dc.channel('dd-trace:storage:before')
 const enterCh = dc.channel('dd-trace:storage:enter')
+const spanFinishCh = dc.channel('dd-trace:span:finish')
 const profilerTelemetryMetrics = telemetryMetrics.manager.namespace('profilers')
 
-const threadName = (function () {
-  const { isMainThread, threadId } = require('node:worker_threads')
-  const name = isMainThread ? 'Main' : `Worker #${threadId}`
-  return `${name} Event Loop`
-})()
+const MemoizedWebTags = Symbol('NativeWallProfiler.MemoizedWebTags')
 
 let kSampleCount
 
@@ -29,30 +27,6 @@ function getStartedSpans (context) {
   return context._trace.started
 }
 
-function generateLabels ({ context: { spanId, rootSpanId, webTags, endpoint }, timestamp }) {
-  const labels = { 'thread name': threadName }
-  if (spanId) {
-    labels['span id'] = spanId
-  }
-  if (rootSpanId) {
-    labels['local root span id'] = rootSpanId
-  }
-  if (webTags && Object.keys(webTags).length !== 0) {
-    labels['trace endpoint'] = endpointNameFromTags(webTags)
-  } else if (endpoint) {
-    // fallback to endpoint computed when sample was taken
-    labels['trace endpoint'] = endpoint
-  }
-  // Incoming timestamps are in microseconds, we emit nanos.
-  labels['end_timestamp_ns'] = timestamp * 1000n
-
-  return labels
-}
-
-function getSpanContextTags (span) {
-  return span.context()._tags
-}
-
 function isWebServerSpan (tags) {
   return tags[SPAN_TYPE] === WEB
 }
@@ -64,6 +38,38 @@ function endpointNameFromTags (tags) {
   ].filter(v => v).join(' ')
 }
 
+function getWebTags (startedSpans, i, span) {
+  // Are web tags for this span already memoized?
+  const memoizedWebTags = span[MemoizedWebTags]
+  if (memoizedWebTags !== undefined) {
+    return memoizedWebTags
+  }
+  // No, we'll have to memoize a new value
+  function memoize (tags) {
+    span[MemoizedWebTags] = tags
+    return tags
+  }
+  // Is this span itself a web span?
+  const context = span.context()
+  const tags = context._tags
+  if (isWebServerSpan(tags)) {
+    return memoize(tags)
+  }
+  // It isn't. Get parent's web tags (memoize them too recursively.)
+  // There might be several webspans, for example with next.js, http plugin creates the first span
+  // and then next.js plugin creates a child span, and this child span has the correct endpoint
+  // information. That's why we always use the tags of the closest ancestor web span.
+  const parentId = context._parentId
+  while (--i >= 0) {
+    const ispan = startedSpans[i]
+    if (ispan.context()._spanId === parentId) {
+      return memoize(getWebTags(startedSpans, i, ispan))
+    }
+  }
+  // Local root span with no web span
+  return memoize(null)
+}
+
 class NativeWallProfiler {
   constructor (options = {}) {
     this.type = 'wall'
@@ -71,13 +77,27 @@ class NativeWallProfiler {
     this._flushIntervalMillis = options.flushInterval || 60 * 1e3 // 60 seconds
     this._codeHotspotsEnabled = !!options.codeHotspotsEnabled
     this._endpointCollectionEnabled = !!options.endpointCollectionEnabled
-    this._withContexts = this._codeHotspotsEnabled || this._endpointCollectionEnabled
+    this._timelineEnabled = !!options.timelineEnabled
+    // We need to capture span data into the sample context for either code hotspots
+    // or endpoint collection.
+    this._captureSpanData = this._codeHotspotsEnabled || this._endpointCollectionEnabled
+    // We need to run the pprof wall profiler with sample contexts if we're either
+    // capturing span data or timeline is enabled (so we need sample timestamps, and for now
+    // timestamps require the sample contexts feature in the pprof wall profiler.)
+    this._withContexts = this._captureSpanData || this._timelineEnabled
     this._v8ProfilerBugWorkaroundEnabled = !!options.v8ProfilerBugWorkaroundEnabled
     this._mapper = undefined
     this._pprof = undefined
 
-    // Bind to this so the same value can be used to unsubscribe later
-    this._enter = this._enter.bind(this)
+    // Bind these to this so they can be used as callbacks
+    if (this._withContexts) {
+      if (this._captureSpanData) {
+        this._enter = this._enter.bind(this)
+        this._spanFinished = this._spanFinished.bind(this)
+      }
+    }
+    this._generateLabels = this._generateLabels.bind(this)
+
     this._logger = options.logger
     this._started = false
   }
@@ -115,15 +135,20 @@ class NativeWallProfiler {
     })
 
     if (this._withContexts) {
-      this._profilerState = this._pprof.time.getState()
       this._currentContext = {}
       this._pprof.time.setContext(this._currentContext)
-      this._lastSpan = undefined
-      this._lastStartedSpans = undefined
-      this._lastSampleCount = 0
 
-      beforeCh.subscribe(this._enter)
-      enterCh.subscribe(this._enter)
+      if (this._captureSpanData) {
+        this._profilerState = this._pprof.time.getState()
+        this._lastSpan = undefined
+        this._lastStartedSpans = undefined
+        this._lastWebTags = undefined
+        this._lastSampleCount = 0
+
+        beforeCh.subscribe(this._enter)
+        enterCh.subscribe(this._enter)
+        spanFinishCh.subscribe(this._spanFinished)
+      }
     }
 
     this._started = true
@@ -144,11 +169,17 @@ class NativeWallProfiler {
 
     const span = getActiveSpan()
     if (span) {
+      const context = span.context()
       this._lastSpan = span
-      this._lastStartedSpans = getStartedSpans(span.context())
+      const startedSpans = getStartedSpans(context)
+      this._lastStartedSpans = startedSpans
+      if (this._endpointCollectionEnabled) {
+        this._lastWebTags = getWebTags(startedSpans, startedSpans.length, span)
+      }
     } else {
       this._lastStartedSpans = undefined
       this._lastSpan = undefined
+      this._lastWebTags = undefined
     }
   }
 
@@ -163,21 +194,17 @@ class NativeWallProfiler {
         context.rootSpanId = rootSpan.context().toSpanId()
       }
     }
-    if (this._endpointCollectionEnabled) {
-      const startedSpans = this._lastStartedSpans
-      // Find the first webspan starting from the end:
-      // There might be several webspans, for example with next.js, http plugin creates a first span
-      // and then next.js plugin creates a child span, and this child span haves the correct endpoint information.
-      for (let i = startedSpans.length - 1; i >= 0; i--) {
-        const tags = getSpanContextTags(startedSpans[i])
-        if (isWebServerSpan(tags)) {
-          context.webTags = tags
-          // endpoint may not be determined yet, but keep it as fallback
-          // if tags are not available anymore during serialization
-          context.endpoint = endpointNameFromTags(tags)
-          break
-        }
-      }
+    if (this._lastWebTags) {
+      context.webTags = this._lastWebTags
+      // endpoint may not be determined yet, but keep it as fallback
+      // if tags are not available anymore during serialization
+      context.endpoint = endpointNameFromTags(this._lastWebTags)
+    }
+  }
+
+  _spanFinished (span) {
+    if (span[MemoizedWebTags]) {
+      span[MemoizedWebTags] = undefined
     }
   }
 
@@ -193,12 +220,12 @@ class NativeWallProfiler {
 
   _stop (restart) {
     if (!this._started) return
-    if (this._withContexts) {
+    if (this._captureSpanData) {
       // update last sample context if needed
       this._enter()
       this._lastSampleCount = 0
     }
-    const profile = this._pprof.time.stop(restart, this._withContexts ? generateLabels : undefined)
+    const profile = this._pprof.time.stop(restart, this._generateLabels)
     if (restart) {
       const v8BugDetected = this._pprof.time.v8ProfilerStuckEventLoopDetected()
       if (v8BugDetected !== 0) {
@@ -208,6 +235,38 @@ class NativeWallProfiler {
     return profile
   }
 
+  _generateLabels (context) {
+    if (context == null) {
+      // generateLabels is also called for samples without context.
+      // In that case just return thread labels.
+      return getThreadLabels()
+    }
+
+    const labels = { ...getThreadLabels() }
+
+    const { context: { spanId, rootSpanId, webTags, endpoint }, timestamp } = context
+
+    if (this._timelineEnabled) {
+      // Incoming timestamps are in microseconds, we emit nanos.
+      labels[END_TIMESTAMP_LABEL] = timestamp * 1000n
+    }
+
+    if (spanId) {
+      labels['span id'] = spanId
+    }
+    if (rootSpanId) {
+      labels['local root span id'] = rootSpanId
+    }
+    if (webTags && Object.keys(webTags).length !== 0) {
+      labels['trace endpoint'] = endpointNameFromTags(webTags)
+    } else if (endpoint) {
+      // fallback to endpoint computed when sample was taken
+      labels['trace endpoint'] = endpoint
+    }
+
+    return labels
+  }
+
   profile () {
     return this._stop(true)
   }
@@ -220,10 +279,14 @@ class NativeWallProfiler {
     if (!this._started) return
 
     const profile = this._stop(false)
-    if (this._withContexts) {
+    if (this._captureSpanData) {
       beforeCh.unsubscribe(this._enter)
       enterCh.unsubscribe(this._enter)
+      spanFinishCh.unsubscribe(this._spanFinished)
       this._profilerState = undefined
+      this._lastSpan = undefined
+      this._lastStartedSpans = undefined
+      this._lastWebTags = undefined
     }
 
     this._started = false

package/packages/dd-trace/src/proxy.js

@@ -10,6 +10,7 @@ const PluginManager = require('./plugin_manager')
 const remoteConfig = require('./appsec/remote_config')
 const AppsecSdk = require('./appsec/sdk')
 const dogstatsd = require('./dogstatsd')
+const spanleak = require('./spanleak')
 
 class Tracer extends NoopProxy {
   constructor () {
@@ -35,6 +36,19 @@ class Tracer extends NoopProxy {
         setInterval(() => {
           this.dogstatsd.flush()
         }, 10 * 1000).unref()
+
+        process.once('beforeExit', () => {
+          this.dogstatsd.flush()
+        })
+      }
+
+      if (config.spanLeakDebug > 0) {
+        if (config.spanLeakDebug === spanleak.MODES.LOG) {
+          spanleak.enableLogging()
+        } else if (config.spanLeakDebug === spanleak.MODES.GC_AND_LOG) {
+          spanleak.enableGarbageCollection()
+        }
+        spanleak.startScrubber()
       }
 
       if (config.remoteConfig.enabled && !config.isCiVisibility) {
@@ -62,11 +76,14 @@ class Tracer extends NoopProxy {
         // do not stop tracer initialization if the profiler fails to be imported
         try {
           const profiler = require('./profiler')
-          profiler.start(config)
+          this._profilerStarted = profiler.start(config)
         } catch (e) {
           log.error(e)
         }
       }
+      if (!this._profilerStarted) {
+        this._profilerStarted = Promise.resolve(false)
+      }
 
       if (config.runtimeMetrics) {
         runtimeMetrics.start(config)
@@ -104,6 +121,13 @@ class Tracer extends NoopProxy {
     return this
   }
 
+  profilerStarted () {
+    if (!this._profilerStarted) {
+      throw new Error('profilerStarted() must be called after init()')
+    }
+    return this._profilerStarted
+  }
+
   use () {
     this._pluginManager.configurePlugin(...arguments)
     return this

package/packages/dd-trace/src/ritm.js

@@ -3,7 +3,7 @@
 const path = require('path')
 const Module = require('module')
 const parse = require('module-details-from-path')
-const dc = require('../../diagnostics_channel')
+const dc = require('dc-polyfill')
 
 const origRequire = Module.prototype.require
 

package/packages/dd-trace/src/service-naming/schemas/v0/storage.js

@@ -37,6 +37,11 @@ const redisConfig = {
 
 const storage = {
   client: {
+    aerospike: {
+      opName: () => 'aerospike.command',
+      serviceName: ({ tracerService, pluginConfig }) =>
+        pluginConfig.service || `${tracerService}-aerospike`
+    },
     'cassandra-driver': {
       opName: () => 'cassandra.query',
       serviceName: ({ tracerService, pluginConfig, system }) =>

package/packages/dd-trace/src/service-naming/schemas/v1/storage.js

@@ -22,6 +22,10 @@ function withFunction ({ tracerService, pluginConfig, params }) {
 
 const storage = {
   client: {
+    aerospike: {
+      opName: () => 'aerospike.command',
+      serviceName: configWithFallback
+    },
     'cassandra-driver': {
       opName: () => 'cassandra.query',
       serviceName: configWithFallback

package/packages/dd-trace/src/span_processor.js

@@ -138,6 +138,10 @@ class SpanProcessor {
       }
     }
 
+    for (const span of trace.finished) {
+      span.context()._tags = {}
+    }
+
     trace.started = active
     trace.finished = []
   }