Required dependencies

npm install @elastic/ecs-pino-format @opentelemetry/api @opentelemetry/core \
  @opentelemetry/sdk-metrics after-all-results agentkeepalive \
  async-value-promise basic-auth breadth-filter cookie core-util-is \
  end-of-stream error-callsites error-stack-parser escape-string-regexp \
  fast-safe-stringify fast-stream-to-buffer http-headers \
  import-in-the-middle json-bigint lru-cache measured-reporting \
  module-details-from-path monitor-event-loop-delay object-filter-sequence \
  pino require-in-the-middle readable-stream semver stream-chopper \
  source-map

No native addons, no pod install, no Android linking required. Node.js >=14.5 is required; Node >=18 is recommended for AsyncLocalStorage run-context support and native fetch instrumentation.

Project setup

1. Copy the source/ directory into your project, e.g. src/apm-core/.
2. In tsconfig.json add a path alias if using TypeScript:

   {
     "compilerOptions": {
       "paths": {
         "apm-core/*": ["./src/apm-core/*"]
       }
     }
   }
   

3. Set environment variables before your process starts:

   ELASTIC_APM_SERVER_URL=https://your-apm-server:8200
   ELASTIC_APM_SECRET_TOKEN=your_token
   ELASTIC_APM_SERVICE_NAME=my-service
   ELASTIC_APM_ENVIRONMENT=production
   

4. The agent must be required/imported before any other application module so that the instrumentation shimmer can wrap third-party modules at load time.
5. If you use ESM, the agent must be loaded via --import or --loader (see upstream docs), because import-in-the-middle hooks require the module loader to be active before user modules resolve.

Public API

HttpApmClient

import { HttpApmClient } from './apm-core/apm-client/http-apm-client/index.js';

new HttpApmClient(options: {
  agentName: string;
  agentVersion: string;
  serviceName: string;
  userAgent: string;
  serverUrl: string;
  secretToken?: string;
  apiKey?: string;
  // ...additional transport options
}): WritableStream & EventEmitter

The primary intake transport. Construct one instance per agent lifecycle; it manages a persistent keepalive connection to APM Server, handles NDJSON serialization, gzip compression, and stream chopping. Use it when you need direct control over the APM intake stream rather than going through the high-level Agent facade.

CloudMetadata

import { CloudMetadata } from './apm-core/cloud-metadata/index.js';

class CloudMetadata {
  constructor(
    cloudProvider: 'auto' | 'aws' | 'gcp' | 'azure' | 'none',
    logger: Logger,
    serviceName: string
  );
  getCloudMetadata(callback: (err: Error | null, metadata: object | null) => void): void;
}

Probes cloud-provider IMDS endpoints with aggressive timeouts (100 ms socket connect, 1 s HTTP). Call getCloudMetadata once at agent startup to enrich every event with provider, region, instance ID, etc. Returns null when not running in a recognized cloud or when cloudProvider is 'none'.

Metrics (start / stop / getOrCreateCounter)

import Metrics from './apm-core/metrics/index.js';

class Metrics {
  constructor(agent: Agent);
  start(refTimers: boolean): void;
  stop(): void;
  getOrCreateCounter(name: string, dimensions?: object): Counter | undefined;
  getOrCreateGauge(name: string, collect: () => number, dimensions?: object): void;
  getOrCreateHistogram(name: string, dimensions?: object): Histogram | undefined;
}

Wraps measured-reporting in an APM-aware registry. Call start(false) (unref timers) in serverless or CLI contexts so the process can exit naturally. Always call stop() on agent shutdown to flush final metric snapshots.

Working examples

Scenario - Minimal agent bootstrap with custom config

Boot the agent from source, passing config programmatically instead of via env vars. This is useful in monorepo setups where multiple services share one config loader.

import path from 'path';

// MUST be the very first import in the entry point
const Agent = require('./src/apm-core/agent');

const agent = new Agent();
agent.start({
  serviceName: 'order-service',
  serviceVersion: '2.1.0',
  serverUrl: process.env.APM_SERVER_URL ?? 'http://localhost:8200',
  secretToken: process.env.APM_SECRET_TOKEN,
  environment: process.env.NODE_ENV ?? 'development',
  logLevel: 'info',
  captureBody: 'errors',
  transactionSampleRate: 0.25,
});

// Agent is now active; subsequent requires are auto-instrumented
const express = require('express');
const app = express();

app.get('/health', (_req, res) => res.json({ status: 'ok' }));
app.listen(3000, () => console.log('listening'));

Scenario - Fetch cloud metadata at startup

Enrich a custom metadata payload with cloud provider details before registering the service.

import { CloudMetadata } from './src/apm-core/cloud-metadata/index';
import pino from 'pino';

const logger = pino({ level: 'warn' });
const cloud = new CloudMetadata('auto', logger, 'inventory-service');

cloud.getCloudMetadata((err, metadata) => {
  if (err) {
    console.warn('Cloud metadata unavailable:', err.message);
  } else {
    console.log('Running on:', metadata);
    // { provider: 'aws', region: 'us-east-1', instance: { id: 'i-0abc...' } }
  }
  // Continue application bootstrap
  startApp();
});

function startApp() { /* ... */ }

Scenario - Record a custom counter metric

Use the Metrics class to track a business event counter that will be flushed on the configured metricsInterval.

// Assumes `agent` is a started Agent instance from agent.js
import Metrics from './src/apm-core/metrics/index';

const metrics = new Metrics(agent);
metrics.start(false); // unref timers

const ordersCounter = metrics.getOrCreateCounter('orders.placed', {
  region: 'eu-west-1',
});

function handleOrder(order: Order) {
  processOrder(order);
  ordersCounter?.inc();
}

process.on('SIGTERM', () => {
  metrics.stop();
  process.exit(0);
});

File walkthrough

• agent.js - Root Agent class; owns the lifecycle (start, destroy) and wires together instrumentation, metrics, cloud metadata, and the APM client.
• apm-client/ - HttpApmClient manages the TCP intake stream to APM Server; NoopApmClient is a silent drop replacement used when the agent is disabled.
• cloud-metadata/ - One module per provider (aws.js, azure.js, gcp.js) plus CallbackCoordination for racing concurrent probes with a global timeout.
• config/ - config.js reads all sources (env, options object, central config); schema.js declares every supported option; normalizers.js coerces types and validates values.
• filters/ - sanitize-field-names.js applies wildcard patterns to redact sensitive header/body field names before they leave the process.
• instrumentation/ - index.js sets up RITM/IITM hooks and owns the module patcher registry; transaction.js and span.js are the core trace-event models; run-context/ provides async context propagation via AsyncLocalStorage or AsyncHooks.
• metrics/ - index.js exports Metrics; registry.js wraps measured-reporting; queue.js creates Bull/BullMQ queue depth gauges.
• middleware/ - Express/Connect middleware that starts transactions from incoming HTTP requests.
• opentelemetry-bridge/ - Intercepts OTel API calls and routes them into APM spans, allowing OTel-instrumented libraries to appear in APM traces without a collector.
• opentelemetry-metrics/ - Registers an OTel SDK MetricReader that forwards measurements to the APM metrics pipeline.
• tracecontext/ - Parses and serializes W3C traceparent and tracestate headers for distributed tracing propagation.
• constants.js - Single source of truth for string literals shared across subsystems (e.g. CONTEXT_MANAGER_ASYNCHOOKS).
• errors.js - Captures Error objects, extracts stack frames, and normalizes them into APM error intake format.
• lambda.js - Detects the Lambda handler path from _HANDLER env var and wraps it for cold-start and invocation tracking.
• logging.js - Builds a pino logger with ECS formatting; exposes NoopLogger for test use.
• parsers.js - Extracts URL, method, headers, user-agent, and body context from Node.js IncomingMessage/ServerResponse pairs.
• stacktraces.js - Collects V8 stack frames, applies source-map translation, and culls internal frames.
• symbols.js - Exports a shared Symbol registry so instrumentation code and user code reference identical private property keys.
• wildcard-matcher.js - Implements the *-glob matching used by transactionIgnoreUrls, sanitizeFieldNames, and similar list-type config options.

Common pitfalls and fixes

• Agent started after other imports: Instrumentation will miss already-loaded modules. Fix: make the agent require/import the absolute first statement in your entry file, before Express, Knex, Redis, etc.
• ESM projects omitting --import: require-in-the-middle cannot hook ES module loads without the loader. Fix: add --import elastic-apm-node/start.js (or equivalent source path) to your node invocation.
• metricsInterval: 0 with Metrics.start(): The start method asserts metricsInterval > 0. Fix: guard with if (agent._conf.metricsInterval > 0) metrics.start(false).
• Cloud metadata probes adding startup latency: Each provider probe has a 100 ms socket + 100 ms DNS + 1 s HTTP timeout. Fix: set cloudProvider: 'none' or hardcode the provider if the deployment environment is known.
• HttpApmClient missing required options: The client throws if agentName, agentVersion, serviceName, or userAgent are absent. Fix: always supply all four requiredOpts even in custom/minimal configurations.
• TypeScript path aliases not resolved at runtime: tsconfig paths are compile-time only. Fix: add tsconfig-paths/register or use tsc-alias in the build step so runtime require resolves apm-core/* correctly.

Vibe-coding prompt template

I have the Elastic APM Node.js agent core library source copied into `src/apm-core/`
in my project. The integration guide is in `USAGE.md` at the project root.
The upstream package is `user@example.com`.

Please help me integrate this APM source into my project step by step:

1. Read `USAGE.md` and `src/apm-core/agent.js` to understand the Agent API.
2. Install all required dependencies listed in USAGE.md § "Required dependencies".
3. Add agent bootstrap code at the top of my entry file (`src/index.ts`) using
   the real `Agent` class from `src/apm-core/agent.js`. Use env vars for
   `serverUrl`, `secretToken`, and `serviceName`.
4. Wire cloud metadata fetching using `CloudMetadata` from
   `src/apm-core/cloud-metadata/index.js` so startup logs show the provider.
5. Add a custom counter metric using `Metrics` from `src/apm-core/metrics/index.js`
   to track [describe your business event].
6. Ensure the agent is stopped cleanly on `SIGTERM`.
7. Show me the final diff. Do not invent any API methods; use only what is
   documented in USAGE.md and visible in the source files under `src/apm-core/`.

License and credit

The source is licensed under the BSD 2-Clause License (Elasticsearch B.V. and contributors). See source/LICENSE if present, or the license header in every source file. Upstream package: elastic-apm-node — source repository: github.com/elastic/apm-agent-nodejs.

This product’s unique contribution is clean room implementation of valuable code block into AVCP compatible standard.

完整安装指南与集成提示词将在购买后解锁。

New Relic Node.js Performance Monitoring Agent

Automation, Utilities & Developer Tools

免费