feat: Added support for getting container ids from ECS metadata API

lib/utilization/docker-info.js

@@ -5,61 +5,162 @@
 
 'use strict'
 
-const logger = require('../logger').child({ component: 'docker-info' })
+const fs = require('node:fs')
+const http = require('node:http')
+const log = require('../logger').child({ component: 'docker-info' })
 const common = require('./common')
 const NAMES = require('../metrics/names')
 const os = require('os')
 let vendorInfo = null
+
 const CGROUPS_V1_PATH = '/proc/self/cgroup'
 const CGROUPS_V2_PATH = '/proc/self/mountinfo'
+const BOOT_ID_PROC_FILE = '/proc/sys/kernel/random/boot_id'
 
 module.exports.getVendorInfo = fetchDockerVendorInfo
 module.exports.clearVendorCache = function clearDockerVendorCache() {
   vendorInfo = null
 }
 
-module.exports.getBootId = function getBootId(agent, callback) {
+module.exports.getBootId = function getBootId(agent, callback, logger = log) {
   if (!/linux/i.test(os.platform())) {
     logger.debug('Platform is not a flavor of linux, omitting boot info')
     return setImmediate(callback, null, null)
   }
 
-  common.readProc('/proc/sys/kernel/random/boot_id', function readProcBootId(err, data) {
-    if (!data) {
-      bootIdError()
-      return callback(null, null)
+  fs.access(BOOT_ID_PROC_FILE, fs.constants.F_OK, (err) => {
+    if (err == null) {
+      // The boot id proc file exists, so use it to get the container id.
+      return common.readProc(BOOT_ID_PROC_FILE, (_, data, cbAgent = agent) => {
+        readProcBootId({ data, agent: cbAgent, callback })
+      })
     }
 
-    data = data.trim()
-    const asciiData = Buffer.from(data, 'ascii').toString()
+    logger.debug('Container boot id is not available in cgroups info')
 
-    if (data !== asciiData) {
-      bootIdError()
+    if (hasAwsContainerApi() === false) {
+      // We don't seem to have a recognized location for getting the container
+      // identifier.
+      logger.debug('Container is not in a recognized ECS container, omitting boot info')
+      recordBootIdError(agent)
       return callback(null, null)
     }
 
-    if (data.length !== 36) {
-      bootIdError()
-      if (data.length > 128) {
-        data = data.substring(0, 128)
+    getEcsContainerId({ agent, callback, logger })
+  })
+}
+
+/**
+ * Queries the AWS ECS metadata API to get the boot id.
+ *
+ * @param {object} params Function parameters.
+ * @param {object} params.agent Newrelic agent instance.
+ * @param {Function} params.callback Typical error first callback. The second
+ * parameter is the boot id as a string.
+ * @param {object} [params.logger] Internal logger instance.
+ */
+function getEcsContainerId({ agent, callback, logger }) {
+  const ecsApiUrl =
+    process.env.ECS_CONTAINER_METADATA_URI_V4 || process.env.ECS_CONTAINER_METADATA_URI
+  const req = http.request(ecsApiUrl, (res) => {
+    let body = Buffer.alloc(0)
+    res.on('data', (chunk) => {
+      body = Buffer.concat([body, chunk])
+    })
+    res.on('end', () => {
+      try {
+        const json = body.toString('utf8')
+        const data = JSON.parse(json)
+        if (data.DockerId == null) {
+          logger.debug('Failed to find DockerId in response, omitting boot info')
+          recordBootIdError(agent)
+          return callback(null, null)
+        }
+        callback(null, data.DockerId)
+      } catch (error) {
+        logger.debug('Failed to process ECS API response, omitting boot info: ' + error.message)
+        recordBootIdError(agent)
+        callback(null, null)
       }
-    }
+    })
+  })
 
-    return callback(null, data)
+  req.on('error', () => {
+    logger.debug('Failed to query ECS endpoint, omitting boot info')
+    recordBootIdError(agent)
+    callback(null, null)
   })
 
-  function bootIdError() {
-    agent.metrics.getOrCreateMetric(NAMES.UTILIZATION.BOOT_ID_ERROR).incrementCallCount()
+  req.end()
+}
+
+/**
+ * Inspects the running environment to determine if the AWS ECS metadata API
+ * is available.
+ *
+ * @see https://docs.aws.amazon.com/AmazonECS/latest/developerguide/ec2-metadata.html
+ *
+ * @returns {boolean}
+ */
+function hasAwsContainerApi() {
+  if (process.env.ECS_CONTAINER_METADATA_URI_V4 != null) {
+    return true
   }
+  return process.env.ECS_CONTAINER_METADATA_URI != null
+}
+
+/**
+ * Increments a supportability metric to indicate that there was an error
+ * while trying to read the boot id from the system.
+ *
+ * @param {object} agent Newrelic agent instance.
+ */
+function recordBootIdError(agent) {
+  agent.metrics.getOrCreateMetric(NAMES.UTILIZATION.BOOT_ID_ERROR).incrementCallCount()
+}
+
+/**
+ * Utility function to parse a Docker boot id from a cgroup proc file.
+ *
+ * @param {Buffer} data The information from the proc file.
+ * @param {object} agent Newrelic agent instance.
+ * @param {Function} callback Typical error first callback. Second parameter
+ * is the boot id as a string.
+ *
+ * @returns {*}
+ */
+function readProcBootId({ data, agent, callback }) {
+  if (!data) {
+    recordBootIdError(agent)
+    return callback(null, null)
+  }
+
+  data = data.trim()
+  const asciiData = Buffer.from(data, 'ascii').toString()
+
+  if (data !== asciiData) {
+    recordBootIdError(agent)
+    return callback(null, null)
+  }
+
+  if (data.length !== 36) {
+    recordBootIdError(agent)
+    if (data.length > 128) {
+      data = data.substring(0, 128)
+    }
+  }
+
+  return callback(null, data)
 }
 
 /**
  * Attempt to extract container id from either cgroups v1 or v2 file
  *
  * @param {object} agent NR instance
  * @param {Function} callback function to call when done
+ * @param {object} [logger] internal logger instance
  */
-function fetchDockerVendorInfo(agent, callback) {
+function fetchDockerVendorInfo(agent, callback, logger = log) {
   if (!agent.config.utilization || !agent.config.utilization.detect_docker) {
     return callback(null, null)
   }
@@ -93,8 +194,9 @@ function fetchDockerVendorInfo(agent, callback) {
  *
  * @param {string} data file contents
  * @param {Function} callback function to call when done
+ * @param {object} [logger] internal logger instance
  */
-function parseCGroupsV2(data, callback) {
+function parseCGroupsV2(data, callback, logger = log) {
   const containerLine = new RegExp('/docker/containers/([0-9a-f]{64})/')
   const line = containerLine.exec(data)
   if (line) {
@@ -110,8 +212,9 @@ function parseCGroupsV2(data, callback) {
  * e.g. - `4:cpu:/docker/f37a7e4d17017e7bf774656b19ca4360c6cdc4951c86700a464101d0d9ce97ee`
  *
  * @param {Function} callback function to call when done
+ * @param {object} [logger] internal logger instance
  */
-function findCGroupsV1(callback) {
+function findCGroupsV1(callback, logger = log) {
   common.readProc(CGROUPS_V1_PATH, function getCGroup(err, data) {
     if (!data) {
       logger.debug(`${CGROUPS_V1_PATH} not found, exiting parsing containerId.`)

test/unit/facts.test.js

@@ -6,6 +6,8 @@
 'use strict'
 
 const tap = require('tap')
+const fs = require('fs')
+const fsAccess = fs.access
 const os = require('os')
 const hostname = os.hostname
 const networkInterfaces = os.networkInterfaces
@@ -518,6 +520,7 @@ tap.test('boot_id', (t) => {
     startingOsPlatform = os.platform
 
     os.platform = () => 'linux'
+    fs.access = (file, mode, cb) => cb(null)
   })
 
   t.afterEach(() => {
@@ -530,6 +533,7 @@ tap.test('boot_id', (t) => {
     sysInfo._getDockerContainerId = startingDockerInfo
     common.readProc = startingCommonReadProc
     os.platform = startingOsPlatform
+    fs.access = fsAccess
 
     startingGetMemory = null
     startingGetProcessor = null
@@ -562,7 +566,9 @@ tap.test('boot_id', (t) => {
             break
 
           case 'input_boot_id':
-            mockReadProc = (file, cb) => cb(null, testValue)
+            mockReadProc = (file, cb) => {
+              cb(null, testValue, agent)
+            }
             break
 
           // Ignore these keys.

test/unit/utilization/aws-ecs-api-response.json

@@ -0,0 +1,50 @@
+{
+  "DockerId": "1e1698469422439ea356071e581e8545-2769485393",
+  "Name": "fargateapp",
+  "DockerName": "fargateapp",
+  "Image": "123456789012.dkr.ecr.us-west-2.amazonaws.com/fargatetest:latest",
+  "ImageID": "sha256:1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcd",
+  "Labels": {
+    "com.amazonaws.ecs.cluster": "arn:aws:ecs:us-west-2:123456789012:cluster/testcluster",
+    "com.amazonaws.ecs.container-name": "fargateapp",
+    "com.amazonaws.ecs.task-arn": "arn:aws:ecs:us-west-2:123456789012:task/testcluster/1e1698469422439ea356071e581e8545",
+    "com.amazonaws.ecs.task-definition-family": "fargatetestapp",
+    "com.amazonaws.ecs.task-definition-version": "7"
+  },
+  "DesiredStatus": "RUNNING",
+  "KnownStatus": "RUNNING",
+  "Limits": {
+    "CPU": 2
+  },
+  "CreatedAt": "2024-04-25T17:38:31.073208914Z",
+  "StartedAt": "2024-04-25T17:38:31.073208914Z",
+  "Type": "NORMAL",
+  "LogDriver": "awslogs",
+  "LogOptions": {
+    "awslogs-create-group": "true",
+    "awslogs-group": "/ecs/fargatetestapp",
+    "awslogs-region": "us-west-2",
+    "awslogs-stream": "ecs/fargateapp/1e1698469422439ea356071e581e8545"
+  },
+  "ContainerARN": "arn:aws:ecs:us-west-2:123456789012:container/testcluster/1e1698469422439ea356071e581e8545/050256a5-a7f3-461c-a16f-aca4eae37b01",
+  "Networks": [
+    {
+      "NetworkMode": "awsvpc",
+      "IPv4Addresses": [
+        "10.10.10.10"
+      ],
+      "AttachmentIndex": 0,
+      "MACAddress": "06:d7:3f:49:1d:a7",
+      "IPv4SubnetCIDRBlock": "10.10.10.0/20",
+      "DomainNameServers": [
+        "10.10.10.2"
+      ],
+      "DomainNameSearchList": [
+        "us-west-2.compute.internal"
+      ],
+      "PrivateDNSName": "ip-10-10-10-10.us-west-2.compute.internal",
+      "SubnetGatewayIpv4Address": "10.10.10.1/20"
+    }
+  ],
+  "Snapshotter": "overlayfs"
+}