@google-cloud/trace-agent

Stackdriver Trace Agent for Node.js

Beta. This is a Beta release of the Stackdriver Trace agent for Node.js. These libraries might be changed in backward-incompatible ways and are not subject to any SLA or deprecation policy.

This module provides automatic tracing for Node.js applications with Stackdriver Trace. Stackdriver Trace is a feature of Google Cloud Platform that collects latency data (traces) from your applications and displays it in near real-time in the Google Cloud Console.

Usage

The Trace Agent supports Node 4+.

Note: Using the Trace Agent requires a Google Cloud Project with the Stackdriver Trace API enabled and associated credentials. These values are auto-detected if the application is running on Google Cloud Platform. If your application is not running on GCP, you will need to specify the project ID and credentials either through the configuration object, or with environment variables. See Setting Up Stackdriver Trace for Node.js for more details.

Simply require and start the Trace Agent as the first module in your application:

require('@google-cloud/trace-agent').start();
// ...

Optionally, you can pass a configuration object to the start() function as follows:

require('@google-cloud/trace-agent').start({
  samplingRate: 500, // sample one trace every half-second.
  ignoreUrls: [ /^\/ignore-me#/ ] // ignore the "/ignore-me" endpoint.
});
// ...

The object returned by start() may be used to create custom trace spans:

const traceApi = require('@google-cloud/trace-agent').start();
traceApi.runInRootSpan({ name: 'my-root-span' }, (rootSpan) => {
  // ...
  rootSpan.endSpan();
});

Note: If your source code contains untranspiled async/await (introduced in Node 7.6), please see this section on enabling experimental tracing for async functions.

What gets traced

The trace agent can do automatic tracing of the following web frameworks:

• express (version 4)
• gRPC server (version ^1.1)
• hapi (versions 8 - 16)
• koa (version 1)
• restify (versions 3 - 6)

The agent will also automatically trace RPCs from the following modules:

• Outbound HTTP requests through http, https, and http2 core modules
• grpc client (version ^1.1)
• mongodb-core (version 1)
• mongoose (version 4)
• mysql (version ^2.9)
• mysql2 (version 1)
• pg (versions 6 - 7)
• redis (versions 0.12 - 2)

You can use the Custom Tracing API to trace other modules in your application.

To request automatic tracing support for a module not on this list, please file an issue. Alternatively, you can write a plugin yourself.

Tracing Additional Modules

To load an additional plugin, specify it in the agent's configuration:

  require('@google-cloud/trace-agent').start({
    plugins: {
      // You may use a package name or absolute path to the file.
      'my-module': '@google-cloud/trace-agent-plugin-my-module',
      'another-module': path.join(__dirname, 'path/to/my-custom-plugins/plugin-another-module.js')
    }
  });

This list of plugins will be merged with the list of built-in plugins, which will be loaded by the plugin loader. Each plugin is only loaded when the module that it patches is loaded; in other words, there is no computational overhead for listing plugins for unused modules.

Custom Tracing API

The custom tracing API can be used to create custom trace spans. A span is a particular unit of work within a trace, such as an RPC request. Spans may be nested; the outermost span is called a root span, even if there are no nested child spans. Root spans typically correspond to incoming requests, while child spans typically correspond to outgoing requests, or other work that is triggered in response to incoming requests.

For any of the web frameworks for which we provide built-in plugins, a root span is automatically started whenever an incoming request is received (in other words, all middleware already runs within a root span). If you wish to record a span outside of any of these frameworks, any traced code must run within a root span that you create yourself.

Accessing the API

Calling the start function returns an instance of TraceApi, which provides an interface for tracing:

const traceApi = require('@google-cloud/trace-agent').start();

It can also be retrieved by subsequent calls to get elsewhere:

// after start() is called
const traceApi = require('@google-cloud/trace-agent').get();

A TraceApi object is guaranteed to be returned by both of these calls, even if the agent is disabled.

A fully detailed overview of the TraceApi object is available here.

How does automatic tracing work?

The Trace Agent automatically patches well-known modules to insert calls to functions that start, label, and end spans to measure latency of RPCs (such as mysql, redis, etc.) and incoming requests (such as express, hapi, etc.). As each RPC is typically performed on behalf of an incoming request, we must make sure that this association is accurately reflected in span data. To provide a uniform, generalized way of keeping track of which RPC belongs to which incoming request, we rely on the continuation-local-storage module to keep track of the "trace context" across asynchronous boundaries.

continuation-local-storage, which relies on async-listener to preserve continuations over asynchronous boundaries, works great in most cases. However, it does have some limitations that can prevent us from being able to properly propagate trace context:

• It is possible that a module does its own queuing of callback functions – effectively merging asynchronous execution contexts. For example, one may write an http request buffering library that queues requests and then performs them in a batch in one shot. In such a case, when all the callbacks fire, they will execute in the context which flushed the queue instead of the context which added the callbacks to the queue. This problem is called the pooling problem or the user-space queuing problem, and is a fundamental limitation of JavaScript. If your application uses such code, you will notice that RPCs from many requests are showing up under a single trace, or that certain portions of your outbound RPCs do not get traced. In such cases we try to work around the problem through monkey patching, or by working with the library authors to fix the code to properly propagate context. However, finding problematic code is not always trivial.
• Presently, it is not possible for async-listener to keep track of transitions across await-ed lines in ES7 async functions that are available with Node 7.6+. If your application uses untranspiled async functions, we will not be properly track RPCs.

Tracing with async/await

Starting in module version 2.2, the Trace Agent ships with an experimental implementation (using the Node 8 async_hooks API) that supports async/await. To enable this implementation, run your application with the environmental variable GCLOUD_TRACE_NEW_CONTEXT set:

# Requires Node 8+
$ GCLOUD_TRACE_NEW_CONTEXT=1 npm start

We are actively looking for feedback on this new implementation. Please file an issue if you encounter unexpected or unwanted behavior.

Contributing changes

• See CONTRIBUTING.md

Licensing

• See LICENSE