New Relic for Node.js

This package instruments your application for performance monitoring with New Relic.

Make sure you have a New Relic account before starting. To see all the features, such as slow transaction traces, you will need a New Relic Pro subscription (or equivalent).

As with any instrumentation tool, please test before using in production.

Getting started
Security
Configuration
RUM / browser timings
Transactions and request naming
Licensing
Contributions
Known issues

Getting started

Install node. The agent runs on v0.8 and higher, but some features (e.g. error tracing) depend on features introduced in v0.10. Development work on this module is done with the latest non-development release of Node.
Verify your version of node came with a new enough version of npm using npm -v. We require version 1.4.28 or newer, and recommend using the latest release. Read more about upgrading npm here.
Install this module via npm install newrelic for the application you want to monitor.
Copy newrelic.js from node_modules/newrelic into the root directory of your application.
Edit newrelic.js and replace license_key's value with the license key for your account.
Add require('newrelic'); as the first line of the app's main module.

If you wish to keep the configuration for the module separate from your application, the module will look for newrelic.js in the directory referenced by the environment variable NEW_RELIC_HOME if it's set.

When you start your app, New Relic should start up with it and start reporting data that will appear within the New Relic UI after a few minutes. Because the agent minimizes the amount of bandwidth it consumes, it only reports data once a minute, so if you require the module in tests that take less than a minute to run, it won't have time to report data to New Relic. The module will write its log to a file named newrelic_agent.log in the application directory. If New Relic doesn't send data or crashes your app, the log can help New Relic determine what went wrong, so be sure to send it along with any bug reports or support requests.

Upgrading npm

If you're running on a version of npm before 1.4.28, or are interested in moving up to latest follow these steps:

Run npm -v to make sure you have npm installed and working.
If you are on linux/smartos/osx/*nix run ls -l $(which npm) and check to see if the file is owned by "root" or "admin". If so, prefix the next command with sudo.
Run npm install -g npm@latest to upgrade npm itself.

Warning: An existing installation of npm can break if it is used to upgrade itself without root privileges while the npm executable is owned by root.

Security

We take security (and the protection of your and your users' privacy) very seriously. See SECURITY.md for details, but the brief version is that if you feel you've found a security issue, contact us at security@newrelic.com.

Configuring the module

The module can be tailored to your app's requirements, both from the server and via the newrelic.js configuration file you created. For complete details on what can be configured, refer to lib/config.default.js, which documents the available variables and their default values.

In addition, for those of you running in PaaS environments like Heroku or Microsoft Azure, all of the configuration variables in newrelic.js have counterparts that can be set via environment variables. You can mix and match variables in the configuration file and environment variables freely; environment variables take precedence.

Here's the list of the most important variables and their values:

NEW_RELIC_LICENSE_KEY: Your New Relic license key. This is a required setting with no default value.
NEW_RELIC_APP_NAME: The name of this application, for reporting to New Relic's servers. This value can be also be a comma-delimited list of names. This is a required setting with no default value. (NOTE: as a convenience to Azure users, the module will use APP_POOL_ID as the application name if it's set, so you can use the name you chose for your Azure Web Server without setting it twice.)
NEW_RELIC_NO_CONFIG_FILE: Inhibit loading of the configuration file altogether. Use with care. This presumes that all important configuration will be available via environment variables, and some log messages assume that a config file exists.
NEW_RELIC_HOME: path to the directory in which you've placed newrelic.js.
NEW_RELIC_USE_SSL: Use SSL for communication with New Relic's servers. Enabled by default.
NEW_RELIC_LOG: Complete path to the New Relic agent log, including the filename. The agent will shut down the process if it can't create this file, and it creates the log file with the same umask of the process. Setting this to stdout will write all logging to stdout, and stderr will write all logging to stderr.
NEW_RELIC_LOG_LEVEL: Logging priority for the New Relic agent. Can be one of error, warn, info, debug, or trace. debug and trace are pretty chatty; unless you're helping New Relic figure out irregularities with the module, you're probably best off using info or higher.

For completeness, here's the rest of the list:

NEW_RELIC_ENABLED: Whether or not the agent should run. Good for temporarily disabling the agent while debugging other issues with your code. It doesn't prevent the module from bootstrapping its instrumentation or setting up all its pieces, it just prevents it from starting up or connecting to New Relic's servers. Defaults to true.
NEW_RELIC_ERROR_COLLECTOR_ENABLED: Whether or not to trace errors within your application. Values are true or false. Defaults to true.
NEW_RELIC_ERROR_COLLECTOR_IGNORE_ERROR_CODES: Comma-delimited list of HTTP status codes to ignore. Maybe you don't care if payment is required? Ignoring a status code means that the transaction is not renamed to match the code, and the request is not treated as an error by the error collector. Defaults to ignoring 404.
NEW_RELIC_IGNORE_SERVER_CONFIGURATION: Whether to ignore server-side configuration for this application. Defaults to false.
NEW_RELIC_TRACER_ENABLED: Whether to collect and submit slow transaction traces to New Relic. Values are true or false. Defaults to true.
NEW_RELIC_TRACER_THRESHOLD: Threshold of web transaction response time (in seconds) at which a transaction trace will count as slow and be sent to New Relic. Can also be set to apdex_f, at which point it will set the trace threshold to 4 times the current ApdexT. Defaults to apdex_f.
NEW_RELIC_APDEX: Set the initial Apdex tolerating / threshold value in seconds. This is more often than not set from the server. Defaults to 0.100.
NEW_RELIC_CAPTURE_PARAMS: Whether to capture request parameters on slow transaction or error traces. Defaults to false.
NEW_RELIC_IGNORED_PARAMS: Some parameters may contain sensitive values you don't want being sent out of your application. This setting is a comma-delimited list of names of parameters to ignore. Defaults to empty.
NEW_RELIC_NAMING_RULES: A list of comma-delimited JSON object literals: NEW_RELIC_NAMING_RULES='{"pattern":"^t","name":"u"},{"pattern":"^u","name":"t"}' See the section on request and transaction naming for details. Defaults to empty.
NEW_RELIC_IGNORING_RULES: A list of comma-delimited patterns: NEW_RELIC_IGNORING_RULES='^/socket\.io/.*/xhr-polling,ignore_me' Note that currently there is no way to escape commas in patterns. Defaults to empty.
NEW_RELIC_TRACER_TOP_N: Increase this number to increase the diversity of slow transaction traces sent to New Relic. Defaults to 1. See the description in lib/config.default.js, as this feature is exceedingly hard to summarize.
NEW_RELIC_HOST: Hostname for the New Relic collector. You shouldn't need to change this.
NEW_RELIC_PORT: Port number on which the New Relic collector will be listening. You shouldn't need to change this either.
NEW_RELIC_PROXY_URL: A fully-qualified URL to an http/https proxy. The proxy URL may include basic authentication. The use of NEW_RELIC_PROXY_URL overrides other proxy settings.
NEW_RELIC_PROXY_HOST: Proxy hostname
NEW_RELIC_PROXY_PORT: Proxy port.
NEW_RELIC_PROXY_USER: Proxy user name (basic auth only).
NEW_RELIC_PROXY_PASS: Proxy password.
NEW_RELIC_DEBUG_METRICS: Whether to collect internal supportability metrics for the agent. Don't mess with this unless New Relic asks you to.
NEW_RELIC_DEBUG_TRACER: Whether to dump traces of the transaction tracer's internal operation. It's unlikely to be informative unless you're a New Relic Node.js engineer and it has a significant performance cost, so use with care.
NEW_RELIC_BROWSER_MONITOR_ENABLE: Whether to generate browser timing (RUM) headers or not.
NEW_RELIC_LABELS: Sets the label names and values to associate with the application. The list is a semi-colon delimited list of colon-separated name and value pairs

Browser timings (RUM / Real User Monitoring)

New Relic's instrumentation can extend beyond your application into the client's browser. The newrelic module can generate <script> headers which, when inserted into your HTML templates, will capture client-side page load times.

Headers must be manually injected, but no extra configuration is necessary to enable browser timings.

Basics

Insert the result of newrelic.getBrowserTimingHeader() into your html page.
The browser timing headers should be placed in the beginning of your <head> tag.
- As an exception to the above, for maximum IE compatability, the results of getBrowserTimingHeader() should be placed after any X-UA-COMPATIBLE HTTP-EQUIV meta tags.
Do not cache the header, call it once for every request.

Example

Below is an example using express and jade; Express is a popular web application framework, and jade is a popular template module. Although the specifics are different for other frameworks, the general approach described below should work in most cases.

The simplest way to insert browser timing headers is to pass the newrelic module into your template, and call newrelic.getBrowserTimingHeader() from within the template.

app.js:

var newrelic = require('newrelic');
var app = require('express')();

// In Express, this lets you call newrelic from within a template.
app.locals.newrelic = newrelic;

app.get('/user/:id', function (req, res) {
  res.render('user');
});
app.listen(process.env.PORT);

layout.jade:

doctype html
html
  head
    != newrelic.getBrowserTimingHeader()
    title= title
    link(rel='stylesheet', href='/stylesheets/style.css')
  body
    block content

By defaults calls to newrelic.getBrowserTimingHeader() should return valid headers. You can disable header generation without removing your template code. In your newrelic.js file, add the following to disable header generation.

exports.config = {
  // ... other config
  browser_monitoring : {
    enable : false
  }
};

You can also set the environment variable NEW_RELIC_BROWSER_MONITOR_ENABLE=false.

It is safe to leave the header generation code in place even when you're not using it. If browser timings are disabled, or there is an error such that working headers cannot be generated, the newrelic module will generate an innocuous HTML comment. If the newrelic module is disabled entirely no content will be generated.

Transactions and request naming

In order to get the most value out of New Relic for Node.js, you may have to do a little work to help us figure out how your application is structured. New Relic works on the assumption that it can group requests to your application into transactions, which are defined by giving one or more request paths a name. These names are used to visualize where your app is spending its time (in transaction breakdowns), to identify slow requests, and to group scoped metrics. For example, this can identify slow database queries by showing which requests are spending a long time waiting on the database.

If you're using Express, Restify or Hapi with their default routers and are satisfied with your application being represented by those frameworks' route matchers, you may not need to do anything. However, if you want more specific names than are provided by your framework, you may want to use one or more of the tools described further on.

The simplest way to tell that you need to read further in this document is if you feel too many of your requests are being lumped together under the catch-all name /*. All requests that aren't otherwise named by the module will end up grouped under /*.

Background

If you've been working with Node for a while, you're probably accustomed to thinking of your application's requests in terms of raw URLs. One of the great things about Node is that it makes it so easy and simple to work with HTTP, and that extends to things like parsing URLs and creating your own strategies for naming and routing requests for services like RESTful APIs. This presents a challenge for New Relic, because we need to keep the number of names we're tracking small enough that we can keep the New Relic user experience snappy, and also so we don't overwhelm you with so much data that it's difficult for you to see the problem spots in your applications. URLs are not a good fit for how New Relic sees performance.

Another of Node's great strengths is that it provides a lot of tools that build on top of the http module to simplify writing web services. Unfortunately, that variety greatly complicates things for us, with our limited resources, and so we offer a few different tools to help you give us the information we need to provide you useful metrics about your application:

we can read the route names from the Express, Restify, and Hapi routers, if you're using them (and as said above, for many of you, this may be all you need)
we offer an API for naming the current request, either with simple names or, if you prefer, grouped into controllers with actions
and we support rules stored in your module's configuration that can mark requests to be renamed or ignored based on regular expressions matched against the request's raw URLs (also available as API calls)

Requests are mapped to transaction names using a deterministic process:

On an incoming request, the module creates a New Relic transaction that includes the HTTP request method and other metadata. As a fallback (for use in the error tracer), each transaction is named after its raw request URL.
When a framework using a supported router is in use, the agent copies the framework name, route path and request method onto the transaction as its transaction name.
Within handlers, you can make calls to the module's API to explicitly set the name of the route to whatever you want, or mark the route to be ignored (or explicitly not ignored, see the below discussion of rules). As noted below, the last API call wins when there is more than one.
When the transaction is finished (i.e. the response has been completely streamed out to the client), any naming or ignoring rules you have specified in your configuration are applied. NOTE: Older versions of the module only applied these rules if nothing else had named the transaction, which made using rules with router-based automatic naming impossible. You must be using the latest version of New Relic to combine naming and ignoring rules with naming set by the router.
Finally, New Relic may have its own rules that it applies to the transaction name, either because the request is for something New Relic rolls up by default (i.e. static assets like images or CSS files), or because New Relic is applying rules to remedy metric grouping issues.
If nothing has named the transaction after steps 1-5, the transaction is named /*, which puts that request in the undifferentiated, default bucket for requests. This will typically only happen in situations where your application uses either no framework, or a framework with a router that isn't currently supported by New Relic.

Let's go through the naming tools one at a time.

Router introspection

Express is the most popular web framework in use within the Node community, and a number of important services are also using Restify. Both frameworks map routes to handlers, and both use a similar pattern to do so: they match one or more HTTP methods (e.g. GET or the ever-popular OPTIONS – let's hear it for CORS) along with a potentially parameterized path (e.g. /user/:id) or a regular expression (e.g. /^/user/([-0-9a-f]+)$/). New Relic will capture both those pieces of information in the request name. If you have support for slow transaction traces and have enabled capture_params, the transaction trace will also have the request's parameters and their values attached to it. Likewise, request parameters and their values will be attached to any errors recorded by the agent.

The only important thing to know about New Relic's support for Express, Restify, or Hapi is that if you're dissatisfied with the names it comes up with, you can use the API calls described below to come up with more descriptive names. Also, if you use a different web framework or router and would like to see support for it added, please let us know.

Request naming with the module API

The API is what's handed back from require('newrelic'), so

var newrelic = require('newrelic');

is all you need. Please note that you still need to ensure that loading the New Relic module is the first thing your application does, as it needs to bootstrap itself before the rest of your application loads, but you can safely require the module from multiple modules in your application – it will only initialize itself once.

newrelic.setTransactionName(name)

Name the current request. You can call this function anywhere within the context of an HTTP request handler, at any time after handling of the request has started, but before the request has finished. A good rule of thumb is that if the request and response objects are in scope, you can set the name.

Explicitly calling newrelic.setTransactionName() will override any names set by Express, Restify or Hapi routes. Calls to newrelic.setTransactionName() and newrelic.setControllerName() will overwrite each other. The last call made before the request ends wins.

VERY IMPORTANT NOTE: Do not include highly variable information like GUIDs, numerical IDs, or timestamps in the request names you create. If your request is slow enough to generate a transaction trace, that trace will contain the original URL. If you enable parameter capture, the parameters will also be attached to the trace. The request names are used to group requests for New Relic's many charts and tables, and those visualizations' value drops as the number of different request names increases. If you have 50 or so different transaction names, you're probably pushing it. If you have more than a couple hundred, you need to rethink your naming strategy.

newrelic.setControllerName(name, [action])

Name the current request using a controller-style pattern, optionally including the current controller action. If the action is omitted, New Relic will include the HTTP method (e.g. GET, POST) as the action. The rules for when you can call newrelic.setControllerName() are the same as they are for newrelic.setTransactionName().

Explicitly calling newrelic.setControllerName() will override any names set by Express, Restify, or Hapi routes. Calls to newrelic.setTransactionName() and newrelic.setControllerName() will overwrite each other. The last one to run before the request ends wins.

See the above note on newrelic.setTransactionName(), which also applies to this function.

Rules for naming and ignoring requests

If you don't feel like putting calls to the New Relic module directly into your application code, you can use pattern-based rules to name requests. There are two sets of rules: one for renaming requests, and one to mark requests to be ignored by New Relic.

If you're using socket.io, you will have a use case for ignoring rules right out of the box. You'll probably want to add a rule like the following:

// newrelic.js
exports.config = {
  // other configuration
  rules : {
    ignore : [
      '^/socket.io/.*/xhr-polling'
    ]
  }
};

This will keep socket.io long-polling from dominating your response-time metrics and blowing out the apdex metrics for your application.

rules.name

A list of rules of the format {pattern : "pattern", name : "name"} for matching incoming request URLs to pattern and naming the matching New Relic transactions name. When automatically naming rules from the router configuration, the patterns are still applied to the URL path, and not the name returned by the router. The pattern can be set as either a string or a JavaScript regular expression literal. Both pattern and name are required. Additional attributes are ignored.

Can also be set via the environment variable NEW_RELIC_NAMING_RULES, with multiple rules passed in as a list of comma-delimited JSON object literals: NEW_RELIC_NAMING_RULES='{"pattern":"^t","name":"u"},{"pattern":"^u","name":"t"}'

rules.ignore

A list of patterns for matching incoming request URLs to be ignored. When using ignoring rules with instrumented routers, the matches are still made against the URL paths, not the name returned by the router. Patterns may be strings or regular expressions.

Can also be set via the environment variable NEW_RELIC_IGNORING_RULES, with multiple rules passed in as a list of comma-delimited patterns: NEW_RELIC_IGNORING_RULES='^/socket\.io/\*/xhr-polling,ignore_me' Note that currently there is no way to escape commas in patterns.

API for adding naming and ignoring rules

newrelic.addNamingRule(pattern, name)

Programmatic version of rules.name above. Naming rules can not be removed until the Node process is restarted. They can also be added via the module's configuration. Both parameters are mandatory.

newrelic.addIgnoringRule(pattern)

Programmatic version of rules.ignore above. Ignoring rules can not be removed until the Node process is restarted. They can also be added via the module's configuration. The pattern is mandatory.

Other API calls

newrelic.addCustomParameter(name, value)

Set a custom parameter value to be attached to a transaction trace and/or error in the New Relic UI. This must be called within the context of a transaction, so it has a place to set the custom parameters.

newrelic.setIgnoreTransaction(ignored)

Tell the module explicitly whether or not a given request should be ignored. Allows you to explicitly filter out long-polling routes or requests you know are going to be time-consuming in an uninteresting way, and also allows you to gather metrics for requests that would otherwise be ignored. Note that to prevent a transaction from being ignored with this function, you must pass false as the parameter – in this case null or undefined will be ignored.

newrelic.noticeError(error, customParameters)

If your app is doing its own error handling with domains or try/catch clauses, but you want all of the information about how many errors are coming out of the app to be centrally managed, use this call. Unlike most of the calls here, this function can be used outside of route handlers, but will have additional context if called from within transaction scope. If custom parameters are passed in on an object literal, they will be passed back to New Relic for display.

newrelic.shutdown([options], callback)

Use this method to gracefully shut down the agent. When called with options.collectPendingData set to true, the agent will send any pending data to the New Relic servers before shutting down. This is useful when you want to shut down the Node process and make sure that all transactions and/or errors are captured by New Relic.

Example of collecting pending data before shutting down the process:

newrelic.shutdown({ collectPendingData: true }, function(error) {
  process.exit()
})

Custom Instrumentation

Custom transaction should be used for instrumenting socket.io or other varieties of socket servers, and background jobs. These are things that the agent can't automatically instrument because without your knowledge of your application, the agent can't tell when they should begin and end.

newrelic.createWebTransaction(url, handle)

url is the name of the web transaction. It should be pretty static, not including anything like user ids or any other data that is very specific to the request. handle is the function you'd like to wrap in the web transaction. Both custom and auto instrumentation will be captured as part of the transaction.

If called within an active web transaction, it will act as a nested tracer. If called within an active background transaction, it will create a new, independent transaction and any calls within the handle will be bound to the new web transaction.

Custom transactions must be ended manually by calling endTransaction(). Timing for custom transaction starts from when the returned wrapped function is called until endTransaction() is called.

newrelic.createBackgroundTransaction(name, [group], handle)

name is the name of the job. It should be pretty static, and not include job ids or anything very specific to that run of the job. group is optional, and allows you to group types of jobs together. This should follow similar rules as the name. handle is a function that encompases your background job. Both custom and auto instrumentation will be captured as part of the transaction.

If called within an active background transaction, it will act as a nested tracer. If called within an active web transaction, it will create a new transaction and any calls within the handle will be bound to the new, independent background transaction.

newrelic.endTransaction()

This takes no arguments and must be called to end any custom transaction. It will detect what kind of transaction was active and end it.

newrelic.createTracer(name, handle)

name is the name of the tracer. It will show up as a segment in your transaction traces and create its own metric. handle is the function to be bound to the tracer.

Timing is from when createTracer is called until the handle done executing. This should be called inside of a transaction to get data. If it is called outside of a transaction it will just pass through.

newrelic.recordMetric(name, value)

name is the metric name to record, it must be a string that begins with Custom/ typically followed by segments for category and label. (eg.Custom/my_category/my_label). value is either a numerical value to associate with the metric sample, or an object representing multiple samples for the metric. If value is an object, it must include keys for count, total, min, max, and sumOfSquares.

newrelic.incrementMetric(name[, amount])

name is the metric name to record, it must be a string that beings with Custom/ typically followed by segments for category and label. (eg.Custom/my_category/my_label). amount is optional, but must be an integer if provided. amount is the number of times to increment the metrics count, it defaults to 1.

The fine print

This is the Node-specific version of New Relic's transaction naming API documentation. The naming API exists to help us deal with the very real problem that trying to handle too many metrics will make New Relic slow for everybody, not just the account with too many metrics. If, in conversation with New Relic Support, you see discussion of "metric explosion", this is what they're talking about.

While we have a variety of strategies for dealing with these issues, the most severe is simply to blacklist offending applications. The main reason for you to be careful in using our request-naming tools is to prevent that from happening to your applications. We will do everything in our power to ensure that you have a good experience with New Relic even if your application is causing us trouble, but sometimes this will require manual intervention on the part of our team, and this can take a little while.

Contributions

We owe a debt to all of the beta testers and users who have provided us with feedback, and in some cases significant pieces of code. (If you wish to contribute, please see CONTRIBUTING.md in this directory.) In particular, we're indebted to these people:

Hernan Silberman, for his work on the memcached instrumentation.
Jeff Howell <jhowell@kabam.com>, for coming up with a much simpler way to instrument node-mongodb-native, as well as pointing out a problem with the Connect instrumentation.

Recent changes

Information about changes to the module are in NEWS.md.

Known issues:

New Relic for Node is only supported on Node.js 0.8 and newer. Some features may behave differently between 0.8, 0.10 and 0.11 / 0.12. The agent is optimized for newer versions of Node.
There are irregularities around transaction trace capture and display. If you notice missing or incorrect information from transaction traces, let us know.
There are ~~over 20,000~~ ~~30,000~~ ~~40,000~~ A LOT of modules on npm. We can only instrument a tiny number of them. Even for the modules we support, there are a very large number of ways to use them. If you see data you don't expect on New Relic and have the time to produce a reduced version of the code that is producing the strange data, it will be used to improve the module and you will have the Node team's gratitude.
The CPU and memory overhead incurred by New Relic for Node is relatively minor (~1-10%, depending on how much of the instrumentation your apps end up using). GC activity is significantly increased while the agent is active, due to the large number of ephemeral objects created by metrics gathering.
When using Node's included clustering support, each worker process will open its own connection to New Relic's servers, and will incur its own overhead costs.

New Relic features available for other platforms not yet in Node.js

custom metrics
explain plans
garbage collector instrumentation
thread profiling
X-ray transactions (depends on thread profiling)
capacity planning

LICENSE

New Relic for Node is free-to-use, proprietary software. Please see the full license (found in LICENSE in this distribution) for details on its license and the licenses of its dependencies.

peaksnail/node-newrelic