newrelic 13.6.2 → 13.6.3

package/NEWS.md

@@ -1,14 +1,31 @@
-### v13.6.2 (2025-10-29)
-
-#### Bug fixes
-
-* Updated express instrumentation to properly instrument an array of middleware defined on a route ([#3473](https://github.com/newrelic/node-newrelic/pull/3473)) ([e383c4e](https://github.com/newrelic/node-newrelic/commit/e383c4efc7171f02aef41d6c657455338b4b7258))
-
-#### Documentation
-
-* documented how to use the subscriber based instrumentation ([#3432](https://github.com/newrelic/node-newrelic/pull/3432)) ([3b3ab72](https://github.com/newrelic/node-newrelic/commit/3b3ab72eee3b4c1a12592d92eab59da2a48b1f46))
-* Updated compatibility report ([#3471](https://github.com/newrelic/node-newrelic/pull/3471)) ([964543b](https://github.com/newrelic/node-newrelic/commit/964543bae1a7c1fd5af32f02ba40d4bf68737d11))
-
+### v13.6.3 (2025-11-11)
+
+#### Bug fixes
+
+* Updated `@google/genai` to properly handle assigning token counts to LlmCompletionSummary message ([#3495](https://github.com/newrelic/node-newrelic/pull/3495)) ([f805d38](https://github.com/newrelic/node-newrelic/commit/f805d382ed0a898f784b053fb9855e9da97dea98))
+* Updated AWS Bedrock instrumentation to properly assign token counts ([#3491](https://github.com/newrelic/node-newrelic/pull/3491)) ([b1687f6](https://github.com/newrelic/node-newrelic/commit/b1687f6dcb8782623a92dacee740a537ad791fdc))
+* Updated openai instrumentation to properly handle streaming when `stream_options.include_usage` is set ([#3494](https://github.com/newrelic/node-newrelic/pull/3494)) ([37c43f5](https://github.com/newrelic/node-newrelic/commit/37c43f5b448f851b749b95b032c4fb52785f2ab0))
+
+#### Code refactoring
+
+* Update `mysql` and `mysql2` instrumentation to subscribe to events emitted ([#3423](https://github.com/newrelic/node-newrelic/pull/3423)) ([2e38914](https://github.com/newrelic/node-newrelic/commit/2e38914ffd8510dc610bd629a00b61e604dca2db))
+
+#### Documentation
+
+* Cleaned up README links and outdated verbiage ([#3484](https://github.com/newrelic/node-newrelic/pull/3484)) ([51e6d7a](https://github.com/newrelic/node-newrelic/commit/51e6d7ace2fb62b907113ec01f63a8d8e4be4e34))
+* Updated compatibility report ([#3477](https://github.com/newrelic/node-newrelic/pull/3477)) ([859f27f](https://github.com/newrelic/node-newrelic/commit/859f27febe4d4afd46bf7d080a2db51ae9acc0cc))
+
+### v13.6.2 (2025-10-29)
+
+#### Bug fixes
+
+* Updated express instrumentation to properly instrument an array of middleware defined on a route ([#3473](https://github.com/newrelic/node-newrelic/pull/3473)) ([e383c4e](https://github.com/newrelic/node-newrelic/commit/e383c4efc7171f02aef41d6c657455338b4b7258))
+
+#### Documentation
+
+* documented how to use the subscriber based instrumentation ([#3432](https://github.com/newrelic/node-newrelic/pull/3432)) ([3b3ab72](https://github.com/newrelic/node-newrelic/commit/3b3ab72eee3b4c1a12592d92eab59da2a48b1f46))
+* Updated compatibility report ([#3471](https://github.com/newrelic/node-newrelic/pull/3471)) ([964543b](https://github.com/newrelic/node-newrelic/commit/964543bae1a7c1fd5af32f02ba40d4bf68737d11))
+
 ### v13.6.1 (2025-10-28)
 
 #### Bug fixes
@@ -8022,3 +8039,4 @@ Special thanks to Ryan Copley (@RyanCopley) for the contribution.
 
 
 
+

package/README.md

@@ -5,7 +5,7 @@
 
 This package instruments your application for performance monitoring with [New Relic](https://newrelic.com).
 
-In order to take full advantage of this package, make sure you have a [New Relic account](https://newrelic.com) before starting. Available features, such as slow transaction traces, will vary [based on account level](https://newrelic.com/application-monitoring/features).
+In order to take full advantage of this package, make sure you have a [New Relic account](https://newrelic.com) before starting. 
 
 As with any instrumentation tool, please test before using in production.
 
@@ -15,7 +15,7 @@ To use New Relic's Node.js agent entails these three steps, which are described
 
 - Install [the `newrelic` package](https://www.npmjs.com/package/newrelic)
 - Create a base configuration file
-- Require the agent in your program
+- Load the agent in your program
 
 1. To install the agent for performance monitoring, use your favorite npm-based package manager and install the `newrelic` package into your application:
 
@@ -95,7 +95,7 @@ node -r newrelic your-program.js
 
 If your application is written with `import` and `export` statements in javascript, you are using [ES Modules](https://nodejs.org/api/esm.html#modules-ecmascript-modules) and must bootstrap the agent in a different way.
 
-The New Relic Node.js agent includes ***_experimental_*** support for ES Modules. The agent is reliant on an experimental feature in Node.js in order to appropriately register instrumentation. Until the Node.js API for [ES Module Loaders](https://nodejs.org/api/esm.html#loaders) is stable, breaking changes may occur when updating Node.js. Lastly, the ESM loader does not follow the same [supported Node.js versions](https://docs.newrelic.com/docs/apm/agents/nodejs-agent/getting-started/compatibility-requirements-nodejs-agent#system) as the agent. The minimum supported version of Node.js is `v16.12.0`.
+The New Relic Node.js agent includes support for ES Modules. 
 
 ### Setup
 
@@ -237,10 +237,9 @@ If the issue has been confirmed as a bug or is a feature request, please file a
 **Support Channels**
 
 * [New Relic Documentation](https://docs.newrelic.com/docs/agents/nodejs-agent/getting-started/introduction-new-relic-nodejs): Comprehensive guidance for using our platform
-* [New Relic Community](https://forum.newrelic.com/): The best place to engage in troubleshooting questions
-* [New Relic Developer](https://developer.newrelic.com/): Resources for building a custom observability applications
+* [New Relic Community](https://support.newrelic.com/): The best place to engage in troubleshooting questions
+* [New Relic Technical Support Offerings](https://docs.newrelic.com/docs/licenses/license-information/general-usage-licenses/support-plan): Support offerings.
 * [New Relic University](https://learn.newrelic.com/): A range of online training for New Relic users of every level
-* [New Relic Technical Support](https://support.newrelic.com/) 24/7/365 ticketed support. Read more about our [Technical Support Offerings](https://docs.newrelic.com/docs/licenses/license-information/general-usage-licenses/support-plan).
 
 
 ## Privacy
@@ -250,10 +249,6 @@ We define “Personal Data” as any information relating to an identified or id
 
 Please review [New Relic’s General Data Privacy Notice](https://newrelic.com/termsandconditions/privacy) for more information.
 
-## Roadmap
-
-See our [roadmap](./ROADMAP_Node.md), to learn more about our product vision, understand our plans, and provide us valuable feedback.
-
 ## Contribute
 
 We encourage your contributions to improve the Node.js agent! Keep in mind when you submit your pull request, you'll need to sign the CLA via the click-through using CLA-Assistant. You only have to sign the CLA one time per project.
@@ -268,7 +263,7 @@ If you believe you have found a security vulnerability in this project or any of
 
 If you would like to contribute to this project, review [these guidelines](./CONTRIBUTING.md).
 
-To [all contributors](https://github.com/newrelic/node-newrelic/graphs/contributors), we thank you!  Without your contribution, this project would not be what it is today.  We also host a community project page dedicated to [New Relic Node Agent](https://opensource.newrelic.com/projects/newrelic/node-newrelic).
+To [all contributors](https://github.com/newrelic/node-newrelic/graphs/contributors), we thank you!  Without your contribution, this project would not be what it is today.
 
 ## License
 

package/lib/config/default.js

@@ -1166,6 +1166,32 @@ defaultConfig.definition = () => {
           formatter: int,
           default: 10
         },
+        // The root sampling options are defined in an unusual way compared to the rest of our configuration but this is done to
+        // allow compatibility with how OpenTelemetry sets their samplers and be backwards compatible as well. Root can either be
+        // a string or an object.
+        //
+        // Example setting root sampler via config to a string value -
+        // root: 'always_on'
+        //
+        // Example setting root sampler via config to trace id ratio based -
+        // root: {
+        //   trace_id_ratio_based: {
+        //     ratio: 0.5
+        //   }
+        // }
+        //
+        root: {
+          formatter: allowList.bind(null, ['trace_id_ratio_based', 'adaptive', 'always_on', 'always_off', 'default']),
+          default: 'default',
+          // if setting root to trace_id_ratio_based via env vars, set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_ROOT: 'trace_id_ratio_based'
+          // and then set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_ROOT_TRACE_ID_RATIO_BASED_RATIO to the desired ratio
+          trace_id_ratio_based: {
+            ratio: {
+              formatter: float,
+              env: 'NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_ROOT_TRACE_ID_RATIO_BASED_RATIO'
+            }
+          }
+        },
 
         /**
          * When set to `always_on`, the sampled flag in the `traceparent` header
@@ -1178,8 +1204,16 @@ defaultConfig.definition = () => {
          * setting.
          */
         remote_parent_sampled: {
-          formatter: allowList.bind(null, ['always_on', 'always_off', 'default']),
-          default: 'default'
+          formatter: allowList.bind(null, ['trace_id_ratio_based', 'adaptive', 'always_on', 'always_off', 'default']),
+          default: 'default',
+          // if setting remote_parent_sampled to trace_id_ratio_based via env vars, set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_SAMPLED: 'trace_id_ratio_based'
+          // and then set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_SAMPLED_TRACE_ID_RATIO_BASED_RATIO to the desired ratio
+          trace_id_ratio_based: {
+            ratio: {
+              formatter: float,
+              env: 'NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_SAMPLED_TRACE_ID_RATIO_BASED_RATIO'
+            }
+          }
         },
 
         /**
@@ -1193,8 +1227,16 @@ defaultConfig.definition = () => {
          * is set to 0.
          */
         remote_parent_not_sampled: {
-          formatter: allowList.bind(null, ['always_on', 'always_off', 'default']),
-          default: 'default'
+          formatter: allowList.bind(null, ['trace_id_ratio_based', 'adaptive', 'always_on', 'always_off', 'default']),
+          default: 'default',
+          // if setting remote_parent_not_sampled to trace_id_ratio_based via env vars, set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_NOT_SAMPLED: 'trace_id_ratio_based'
+          // and then set NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_NOT_SAMPLED_TRACE_ID_RATIO_BASED_RATIO to the desired ratio
+          trace_id_ratio_based: {
+            ratio: {
+              formatter: float,
+              env: 'NEW_RELIC_DISTRIBUTED_TRACING_SAMPLER_REMOTE_PARENT_NOT_SAMPLED_TRACE_ID_RATIO_BASED_RATIO'
+            }
+          }
         }
       }
     },

package/lib/config/index.js

@@ -1060,10 +1060,13 @@ Config.prototype._fromPassed = function _fromPassed(external, internal, arbitrar
 
   for (const key of Object.keys(external)) {
     // if it's not in the defaults, it doesn't exist
-    if (!arbitrary && !(key in internal)) {
+    if (!arbitrary && Object.hasOwn(internal, key) === false) {
       continue
     }
 
+    // Handle trace_id_ratio_based for distributed tracing sampler configuration
+    this._buildTraceIdRatioSamplers(key, internal, external[key])
+
     if (key === 'ssl' && !isTruthular(external.ssl)) {
       logger.warn(SSL_WARNING)
       continue
@@ -1087,6 +1090,54 @@ Config.prototype._fromPassed = function _fromPassed(external, internal, arbitrar
   }
 }
 
+/**
+ * Builds the trace_id_ratio_based for distributed tracing sampler configuration structure
+ *
+ * @param {string} key - The configuration key being processed
+ * @param {object} internal - The internal configuration object being modified
+ * @param {object} samplerConfig - Sampler configuration object from defaultsConfig
+ */
+Config.prototype._buildTraceIdRatioSamplers = function _buildTraceIdRatioSamplers(key, internal, samplerConfig) {
+  if (key !== 'sampler') {
+    return
+  }
+
+  const samplers = Object.keys(samplerConfig)
+
+  // user can set multiple samplers
+  for (const sampler of samplers) {
+    if (this._isTraceIdRatioBasedConfig(sampler, samplerConfig)) {
+      const ratioValue = samplerConfig[sampler].trace_id_ratio_based.ratio
+      if (ratioValue) {
+        internal[key][sampler] = {
+          trace_id_ratio_based: {
+            ratio: ratioValue
+          }
+        }
+        logger.trace('Setting trace id ratio based sampler on %s', samplers)
+      } else {
+        logger.error('Not setting trace id ratio based sampler on %s.', samplers)
+      }
+    }
+  }
+}
+
+/**
+ * Checks if the external config contains a trace_id_ratio_based for distributed tracing
+ * sampler configuration
+ *
+ * @param {string} selectedSampler - The selected sampler key
+ * @param {object} samplerConfig - The sampler configuration object
+ * @returns {boolean} true if it's a trace_id_ratio_based config
+ */
+Config.prototype._isTraceIdRatioBasedConfig = function _isTraceIdRatioBasedConfig(selectedSampler, samplerConfig) {
+  const samplers = ['root', 'remote_parent_sampled', 'remote_parent_not_sampled']
+
+  return samplers.includes(selectedSampler) &&
+   typeof samplerConfig[selectedSampler] === 'object' &&
+    'trace_id_ratio_based' in samplerConfig[selectedSampler]
+}
+
 /**
  * Some values should be picked up only if they're not otherwise set, like
  * the Windows / Azure application name. Don't set it if there's already
@@ -1163,6 +1214,53 @@ function setFromEnv({ config, key, envVar, formatter, paths }) {
   }
 }
 
+/**
+ * Assigns the value of the distributed tracing sampler env var as an trace_id_ratio_based
+ * object to the sampler in the config.
+ *
+ * @param {object} params object passed to fn
+ * @param {string} params.key key of the sampler
+ * Example: 'root' or 'remote_parent_sampled'
+ * @param {object} params.value value of the sampler
+ * Example: { default: 'default', formatter: someFunc, trace_id_ratio_based: {ratio} }
+ * @param {object} params.config agent config
+ * @param {Array} params.paths list of leaf nodes leading to the sampling configuration value
+ * Example: ['distributed_tracing', 'sampler']
+ */
+function setTraceIDRatioSampler({ key, value, config, paths }) {
+  // trace id ratio based sampler is nested in leaf nodes under distributed_tracing > samplers > key
+  // so don't continue if path is empty
+  if (paths.length === 0) {
+    return
+  }
+
+  const samplers = ['root', 'remote_parent_sampled', 'remote_parent_not_sampled']
+  if (samplers.includes(key) && paths[paths.length - 1] === 'sampler') {
+    // config.distributed_tracing.sampler.[TYPE OF SAMPLER] needs to be set to trace_id_ratio_based
+    // in order to continue
+    if (config.distributed_tracing.sampler[key] !== 'trace_id_ratio_based') {
+      return
+    }
+
+    // set the trace id based ratio
+    const envVar = value.trace_id_ratio_based.ratio.env
+    const formatter = value.trace_id_ratio_based.ratio.formatter
+    const setting = process.env[envVar]
+    if (setting) {
+      const formattedSetting = formatter(setting)
+      config.distributed_tracing.sampler[key] = {
+        trace_id_ratio_based: {
+          ratio: formattedSetting
+        }
+      }
+      logger.trace('Setting %s environment variable', envVar)
+    } else {
+      logger.error('Not setting %s environment variable. Setting sampler root to default.', envVar)
+      config.distributed_tracing.sampler.root = 'default'
+    }
+  }
+}
+
 /**
  * Recursively visit the nodes of the config definition and look for environment variable names, overriding any configuration values that are found.
  *
@@ -1201,6 +1299,17 @@ Config.prototype._fromEnvironment = function _fromEnvironment(
     if (Object.prototype.hasOwnProperty.call(value, 'default') === true) {
       const envVar = deriveEnvVar(key, paths)
       setFromEnv({ config, key, paths, envVar, formatter: value.formatter })
+
+      // Handle custom configuration of the sampler if it's set to trace_id_ratio_based
+      // since sampler was set to a string but now has to be converted to an object.
+      // This is called here because we need set the sampler to a value first in config before setting
+      // trace id ratio sampler (if the user set that)
+      setTraceIDRatioSampler({
+        key,
+        value,
+        config,
+        paths
+      })
       continue
     }
 

package/lib/instrumentation/@google/genai.js

@@ -121,23 +121,26 @@ function recordChatCompletionMessages({
 
 function instrumentStream ({ shim, request, response, segment, transaction }) {
   let err
-  let content
-  let modelVersion
-  let finishReason
   let entireMessage = ''
+  let isDone = false
+  let cachedResult = {}
   shim.wrap(response, 'next', function wrapNext(shim, originalNext) {
     return async function wrappedNext(...nextArgs) {
       let result = {}
       try {
         result = await originalNext.apply(response, nextArgs)
+        // When the stream is done we get {value: undefined, done: true}
+        // we need to cache the composed value and add the entire message
+        // back in later
+        if (result.done === true) {
+          isDone = true
+        } else {
+          cachedResult = result.value
+        }
+
         if (result?.value?.text) {
-          modelVersion = result.value.modelVersion
-          content = result.value.candidates[0].content
           entireMessage += result.value.text // readonly variable that equates to result.value.candidates[0].content.parts[0].text
         }
-        if (result?.value?.candidates?.[0]?.finishReason) {
-          finishReason = result.value.candidates[0].finishReason
-        }
       } catch (streamErr) {
         err = streamErr
         throw err
@@ -146,30 +149,18 @@ function instrumentStream ({ shim, request, response, segment, transaction }) {
         // time it took to handle the stream
         segment.touch()
 
-        // result will be {value: undefined, done: true}
-        // when the stream is done, so we need to create
-        // a mock GenerateContentResponse object with
-        // the entire message
-        //
         // also need to enter this block if there was an
         // error, so we can record it
-        if (result?.done || err) {
-          if (content) {
-            content.parts[0].text = entireMessage
-            result.value = {
-              candidates: [
-                { content, finishReason }
-              ],
-              modelVersion
-            }
+        if (isDone || err) {
+          if (cachedResult?.candidates?.[0]?.content?.parts) {
+            cachedResult.candidates[0].content.parts[0].text = entireMessage
           }
-
           recordChatCompletionMessages({
             agent: shim.agent,
             segment,
             transaction,
             request,
-            response: result?.value,
+            response: cachedResult,
             err
           })
         }

package/lib/instrumentations.js

@@ -30,7 +30,6 @@ module.exports = function instrumentations() {
     langchain: { module: './instrumentation/langchain' },
     memcached: { type: InstrumentationDescriptor.TYPE_DATASTORE },
     mongodb: { type: InstrumentationDescriptor.TYPE_DATASTORE },
-    mysql: { module: './instrumentation/mysql' },
     next: { module: './instrumentation/nextjs' },
     q: { type: null },
     redis: { type: InstrumentationDescriptor.TYPE_DATASTORE },

package/lib/llm-events/aws-bedrock/bedrock-response.js

@@ -73,6 +73,30 @@ class BedrockResponse {
     return this.#parsedBody
   }
 
+  get inputTokenCount() {
+    if (this.#isConverse) {
+      return parseInt(this.#originalResponse?.output?.usage?.inputTokens || 0, 10)
+    }
+
+    return parseInt(this?.headers?.['x-amzn-bedrock-input-token-count'] || 0, 10)
+  }
+
+  get outputTokenCount() {
+    if (this.#isConverse) {
+      return parseInt(this.#originalResponse?.output?.usage?.outputTokens || 0, 10)
+    }
+
+    return parseInt(this?.headers?.['x-amzn-bedrock-output-token-count'] || 0, 10)
+  }
+
+  get totalTokenCount() {
+    if (this.#isConverse) {
+      return parseInt(this.#originalResponse?.output?.usage?.totalTokens || 0, 10)
+    }
+
+    return this.inputTokenCount + this.outputTokenCount
+  }
+
   /**
    * The prompt responses returned by the model.
    *
@@ -164,7 +188,7 @@ class BedrockResponse {
     } else if (cmd.isClaude3() === true) {
       if (body?.type === 'message_stop') {
         // Streamed response
-        this.#completions = body.completions
+        this.#completions.push(body.completions)
       } else {
         this.#completions = [stringifyClaudeChunkedMessage(body?.content)]
       }

package/lib/llm-events/aws-bedrock/chat-completion-message.js

@@ -6,9 +6,6 @@
 'use strict'
 
 const LlmEvent = require('./event')
-const { setTokenFromCallback } = require('../utils')
-const { tokenUsageAttributesExist, tokenUsageHeadersExist } = require('./utils')
-
 /**
  * @typedef {object} LlmChatCompletionParams
  * @augments LlmEventParams
@@ -71,9 +68,8 @@ class LlmChatCompletionMessage extends LlmEvent {
       const completionContent = this.bedrockResponse?.completions?.join(' ')
 
       if (promptContent && completionContent) {
-        setTokenFromCallback(
+        this.setTokenFromCallback(
           {
-            context: this,
             tokenCB,
             reqModel: this.bedrockCommand.modelId,
             resModel: this.bedrockCommand.modelId,
@@ -85,11 +81,7 @@ class LlmChatCompletionMessage extends LlmEvent {
       return
     }
 
-    // If no token count callback is available, we need to check the response object
-    // or response headers for usage information and set token_count to 0 if all usage attributes are present.
-    if (tokenUsageAttributesExist(this.bedrockResponse) || tokenUsageHeadersExist(this.bedrockResponse)) {
-      this.token_count = 0
-    }
+    this.setTokenInCompletionMessage({ promptTokens: this.bedrockResponse.inputTokenCount, completionTokens: this.bedrockResponse.outputTokenCount })
   }
 }
 

package/lib/llm-events/aws-bedrock/chat-completion-summary.js

@@ -7,8 +7,6 @@
 
 const LlmEvent = require('./event')
 
-const { setUsageTokens } = require('./utils')
-const { setTokenUsageFromCallback } = require('../utils')
 /**
  * @typedef {object} LlmChatCompletionSummaryParams
  * @augments LlmEventParams
@@ -50,21 +48,19 @@ class LlmChatCompletionSummary extends LlmEvent {
       const promptContent = this.bedrockCommand?.prompt?.map((msg) => msg.content).join(' ')
       const completionContent = this.bedrockResponse?.completions?.join(' ')
 
-      if (promptContent && completionContent) {
-        setTokenUsageFromCallback(
-          {
-            context: this,
-            tokenCB,
-            reqModel: this.bedrockCommand.modelId,
-            resModel: this.bedrockCommand.modelId,
-            promptContent,
-            completionContent
-          }
-        )
-      }
+      this.setTokenUsageFromCallback(
+        {
+          tokenCB,
+          reqModel: this.bedrockCommand.modelId,
+          resModel: this.bedrockCommand.modelId,
+          promptContent,
+          completionContent
+        }
+      )
       return
     }
-    setUsageTokens(this.bedrockResponse, this)
+
+    this.setTokensInResponse({ promptTokens: this.bedrockResponse.inputTokenCount, completionTokens: this.bedrockResponse.outputTokenCount, totalTokens: this.bedrockResponse.totalTokenCount })
   }
 }
 

package/lib/llm-events/aws-bedrock/converse-stream-handler.js

@@ -65,6 +65,11 @@ class ConverseStreamHandler {
         message: {
 
         }
+      },
+      usage: {
+        inputTokens: null,
+        outputTokens: null,
+        totalTokens: null
       }
     }
   }
@@ -119,9 +124,7 @@ async function * handleConverse() {
     for await (const event of this.stream) {
       yield event
       this.updateHeaders(event)
-      if (event.messageStart?.role) {
-        this.role = 'assistant'
-      } else if (event.contentBlockStart?.start) {
+      if (event.contentBlockStart?.start) {
         // Handles a Content block start event. Tool use only.
         const blockStartData = event.contentBlockStart.start
         if (blockStartData.toolUse) {
@@ -133,6 +136,7 @@ async function * handleConverse() {
           activeChunk = null
         }
       } else if (event.contentBlockDelta?.delta) {
+        // There are also deltas for tool use (stringified inputs) but we don't currently record them so we just ignore for now
         if (event.contentBlockDelta.delta.text) {
           // It seems like the first streamed chunk does not always start with a contentBlockStart message
           // If the stream starts with a delta, assume the current chunk is text
@@ -141,9 +145,12 @@ async function * handleConverse() {
           }
           activeChunk.text += event.contentBlockDelta.delta.text
         }
-        // There are also deltas for tool use (stringified inputs) but we don't currently record them so we just ignore for now
+      // used in `handleResponse` to determine why the model stopped
       } else if (event.messageStop) {
         this.stopReason = event.messageStop?.stopReason
+      // sends token usage info at the end of the stream
+      } else if (event.metadata?.usage) {
+        this.response.output.usage = event.metadata.usage
       }
     }
   } finally {

package/lib/llm-events/aws-bedrock/embedding.js

@@ -6,7 +6,6 @@
 'use strict'
 
 const LlmEvent = require('./event')
-const { validCallbackTokenCount, calculateCallbackTokens } = require('../utils')
 
 /**
  * @typedef {object} LlmEmbeddingParams
@@ -44,28 +43,12 @@ class LlmEmbedding extends LlmEvent {
         return
       }
 
-      const totalTokenCount = calculateCallbackTokens(tokenCB, this['request.model'], content)
-      const hasValidCallbackCounts = validCallbackTokenCount(totalTokenCount)
-
-      if (hasValidCallbackCounts) {
-        this['response.usage.total_tokens'] = Number(totalTokenCount)
-      }
+      const totalTokens = this.calculateCallbackTokens(tokenCB, this['request.model'], content)
+      this.setTokensOnEmbeddingMessage(totalTokens)
       return
     }
 
-    const totalTokens = this.getTotalTokens()
-    if (totalTokens) {
-      this['response.usage.total_tokens'] = Number(totalTokens)
-    }
-  }
-
-  getTotalTokens() {
-    // We record the input token count as total tokens
-    const totalToken =
-      this.bedrockResponse?.usage?.input_tokens || this.bedrockResponse?.usage?.inputTokens ||
-      this.bedrockResponse?.headers['x-amzn-bedrock-input-token-count']
-
-    return totalToken
+    this.setTokensOnEmbeddingMessage(this.bedrockResponse.inputTokenCount)
   }
 }
 

package/lib/llm-events/aws-bedrock/event.js

@@ -6,7 +6,7 @@
 'use strict'
 
 const { randomUUID } = require('crypto')
-const { DESTINATIONS } = require('../../config/attribute-filter')
+const BaseEvent = require('../event')
 
 /**
  * @typedef {object} LlmEventParams
@@ -29,7 +29,7 @@ const defaultParams = {
 /**
  * Baseline object representing a LLM event.
  */
-class LlmEvent {
+class LlmEvent extends BaseEvent {
   /**
    * All parameters that were passed in to the constructor after they have
    * been merged with the constructor's defaults.
@@ -43,6 +43,7 @@ class LlmEvent {
    * @param {LlmEventParams} params Construction parameters.
    */
   constructor(params = defaultParams) {
+    super()
     params = Object.assign({}, defaultParams, params)
     this.constructionParams = params
 
@@ -64,24 +65,6 @@ class LlmEvent {
     this['request.max_tokens'] = null
   }
 
-  /**
-   * Pull user set `llm.*` attributes from the current transaction and
-   * add them to the event.
-   *
-   * @param {object} agent The New Relic agent that provides access to the
-   * transaction.
-   */
-  set metadata(agent) { // eslint-disable-line accessor-pairs
-    const tx = agent.tracer.getTransaction()
-    const attrs = tx?.trace?.custom.get(DESTINATIONS.TRANS_SCOPE) || {}
-    for (const [k, v] of Object.entries(attrs)) {
-      if (k.startsWith('llm.') === false) {
-        continue
-      }
-      this[k] = v
-    }
-  }
-
   /**
    * Removes the complex objects from the event
    * This will be called right before the event is enqueued to the custom event aggregator