feat: Added New Relic Control health check (#2841)

Author: jsumners-nr

index.js

@@ -5,6 +5,8 @@
 
 'use strict'
 
+const HealthReporter = require('./lib/health-reporter')
+
 // Record opening times before loading any other files.
 const preAgentTime = process.uptime()
 const agentStart = Date.now()
@@ -154,6 +156,7 @@ function createAgent(config) {
       'New Relic requires that you name this application!\n' +
       'Set app_name in your newrelic.js or newrelic.cjs file or set environment variable\n' +
       'NEW_RELIC_APP_NAME. Not starting!'
+    agent.healthReporter.setStatus(HealthReporter.STATUS_MISSING_APP_NAME)
     throw new Error(message)
   }
 
@@ -167,6 +170,7 @@ function createAgent(config) {
 
   agent.start(function afterStart(error) {
     if (error) {
+      agent.healthReporter.setStatus(HealthReporter.STATUS_INTERNAL_UNEXPECTED_ERROR)
       const errorMessage = 'New Relic for Node.js halted startup due to an error:'
       logger.error(error, errorMessage)
 

lib/agent.js

@@ -40,6 +40,7 @@ const {
 const synthetics = require('./synthetics')
 const Harvester = require('./harvester')
 const { createFeatureUsageMetrics } = require('./util/application-logging')
+const HealthReporter = require('./health-reporter')
 
 // Map of valid states to whether or not data collection is valid
 const STATES = {
@@ -162,6 +163,8 @@ function Agent(config) {
     throw new Error('Agent must be created with a configuration!')
   }
 
+  this.healthReporter = new HealthReporter({ agentConfig: config })
+
   // The agent base attributes which last throughout its lifetime.
   this._state = 'stopped'
   this.config = config
@@ -326,6 +329,7 @@ Agent.prototype.start = function start(callback) {
   if (this.config.agent_enabled !== true) {
     logger.warn('The New Relic Node.js agent is disabled by its configuration. ' + 'Not starting!')
 
+    this.healthReporter.setStatus(HealthReporter.STATUS_AGENT_DISABLED)
     this.setState('stopped')
     return process.nextTick(callback)
   }
@@ -342,17 +346,21 @@ Agent.prototype.start = function start(callback) {
         'Has a license key been specified in the agent configuration ' +
         'file or via the NEW_RELIC_LICENSE_KEY environment variable?'
     )
+    this.healthReporter.setStatus(HealthReporter.STATUS_LICENSE_KEY_MISSING)
 
     this.setState('errored')
     sampler.stop()
     return process.nextTick(function onNextTick() {
-      callback(new Error('Not starting without license key!'))
+      agent.healthReporter.stop(() => {
+        callback(new Error('Not starting without license key!'))
+      })
     })
   }
   logger.info('Starting New Relic for Node.js connection process.')
 
   this.collector.connect(function onStartConnect(error, response) {
     if (error || response.shouldShutdownRun()) {
+      agent.healthReporter.setStatus(HealthReporter.STATUS_CONNECT_ERROR)
       agent.setState('errored')
       sampler.stop()
       callback(error || new Error('Failed to connect to collector'), response && response.payload)
@@ -476,23 +484,26 @@ Agent.prototype.stop = function stop(callback) {
 
   sampler.stop()
 
-  if (this.collector.isConnected()) {
-    this.collector.shutdown(function onShutdown(error) {
-      if (error) {
-        agent.setState('errored')
-        logger.warn(error, 'Got error shutting down connection to New Relic:')
-      } else {
-        agent.setState('stopped')
-        logger.info('Stopped New Relic for Node.js.')
-      }
-
-      callback(error)
-    })
-  } else {
-    logger.trace('Collector was not connected, invoking callback.')
+  this.healthReporter.setStatus(HealthReporter.STATUS_AGENT_SHUTDOWN)
+  this.healthReporter.stop(() => {
+    if (agent.collector.isConnected()) {
+      agent.collector.shutdown(function onShutdown(error) {
+        if (error) {
+          agent.setState('errored')
+          logger.warn(error, 'Got error shutting down connection to New Relic:')
+        } else {
+          agent.setState('stopped')
+          logger.info('Stopped New Relic for Node.js.')
+        }
+
+        callback(error)
+      })
+    } else {
+      logger.trace('Collector was not connected, invoking callback.')
 
-    process.nextTick(callback)
-  }
+      process.nextTick(callback)
+    }
+  })
 }
 
 /**

lib/collector/api.js

@@ -9,6 +9,7 @@ const CollectorResponse = require('./response')
 const facts = require('./facts')
 const logger = require('../logger').child({ component: 'collector_api' })
 const RemoteMethod = require('./remote-method')
+const HealthReporter = require('../health-reporter')
 
 const NAMES = require('../metrics/names')
 
@@ -221,6 +222,7 @@ CollectorAPI.prototype._retry = function _retry(ctx, error, response) {
     ctx.errors.push(error)
   } else if (response && SUCCESS.has(response.status)) {
     dumpErrors(ctx.errors, 'connect')
+    this._agent.healthReporter.setStatus(HealthReporter.STATUS_HEALTHY)
     ctx.callback(null, CollectorResponse.success(response.payload))
     return
   }
@@ -231,6 +233,7 @@ CollectorAPI.prototype._retry = function _retry(ctx, error, response) {
   // Retry everything except for an explicit Disconnect response code.
   if (response.status === 410 || response.agentRun === AGENT_RUN_BEHAVIOR.SHUTDOWN) {
     logger.error('The New Relic collector rejected this agent.')
+    this._agent.healthReporter.setStatus(HealthReporter.STATUS_FORCED_DISCONNECT)
     return ctx.callback(null, CollectorResponse.fatal(response.payload))
   } else if (response.status === 401) {
     logger.warn(
@@ -240,6 +243,7 @@ CollectorAPI.prototype._retry = function _retry(ctx, error, response) {
         ' (status code %s)',
       response.status
     )
+    this._agent.healthReporter.setStatus(HealthReporter.STATUS_INVALID_LICENSE_KEY)
   } else if (this._isProxyMisconfigured(error)) {
     logger.warn(
       error,
@@ -248,6 +252,17 @@ CollectorAPI.prototype._retry = function _retry(ctx, error, response) {
         'SSL(https). If your proxy is configured to accept connections over http, try ' +
         'setting `proxy` to a fully qualified URL(e.g http://proxy-host:8080).'
     )
+    this._agent.healthReporter.setStatus(HealthReporter.STATUS_HTTP_PROXY_MISCONFIGURED)
+  } else {
+    // Sometimes we get a `CollectorResponse` instance instead of an
+    // `http.ServerResponse`. In such cases, we do not have access to the
+    // status code.
+    let msg = 'Unexpected error communicating with New Relic backend.'
+    if (response.status) {
+      msg = `Received error status code from New Relic backend: ${response.status}.`
+    }
+    logger.warn(error, msg)
+    this._agent.healthReporter.setStatus(HealthReporter.STATUS_BACKEND_ERROR)
   }
 
   const backoff = BACKOFFS[Math.min(ctx.attempts, ctx.max) - 1]

lib/config/default.js

@@ -147,6 +147,45 @@ defaultConfig.definition = () => ({
     formatter: boolean,
     default: true
   },
+
+  /**
+   * Collects configuration related to New Relic Agent Control, i.e. centralized
+   * agent management in container based environments.
+   */
+  agent_control: {
+    /**
+     * Indicates that the agent is being managed by Agent Control. Must be set
+     * to true health monitoring.
+     */
+    enabled: {
+      formatter: boolean,
+      default: false
+    },
+
+    /**
+     * Settings specific to the health monitoring aspect of Agent Control.
+     */
+    health: {
+      /**
+       * A string file path to a directory that the agent is expected to write
+       * health status files to. Must be set for health monitoring to be
+       * enabled.
+       */
+      delivery_location: {
+        default: 'file:///newrelic/apm/health'
+      },
+
+      /**
+       * The time, in seconds, that the agent should wait between writing
+       * updates to its health status. The default interval is 5 seconds.
+       */
+      frequency: {
+        formatter: int,
+        default: 5
+      }
+    }
+  },
+
   /**
    * The default Apdex tolerating / threshold value for applications, in
    * seconds. The default for Node is apdexT to 100 milliseconds, which is