Agent API

edit

The Elastic APM Node.js agent is a singleton. You get the agent instance by requiring either elastic-apm-node or elastic-apm-node/start. The agent is also returned by the .start() method, which allows you to require and start the agent on the same line:

const apm = require('elastic-apm-node').start(...)

If you need to access the Agent in any part of your codebase, you can simply require elastic-apm-node to access the already started singleton. You therefore don’t need to manage or pass around the started Agent yourself.

apm.start([options])

edit

Starts the Elastic APM agent for Node.js and returns itself.

For the APM agent to automatically instrument Node.js modules, it must be started before those modules are loaded. See Starting the agent for details and possible surprises with compilers/transpilers/bundlers.

See the Configuration documentation for available options.

apm.isStarted()

edit

Added in: v1.5.0

Use isStarted() to check if the agent has already started. Returns true if the agent has started, otherwise returns false.

apm.getServiceName()

edit

Added in: v3.11.0

Get the configured serviceName. If a service name was not explicitly configured, this value may have been automatically determined. The service name is not determined until agent.start(), so will be undefined until then. A misconfigured agent can have a null service name.

apm.setFramework(options)

edit

Added in: v2.8.0

Set or change the frameworkName or frameworkVersion after the agent has started. These config options can also be provided as part of the regular agent configuration.

apm.addFilter(fn)

edit

Added in: v0.1.0

Use addFilter() to supply a filter function.

Each filter function will be called just before data is being sent to the APM Server. This will allow you to manipulate the data being sent, for instance to remove sensitive information like passwords etc. (Note: Filters added via addFilter are not applied to the "metadata" object sent to the APM Server — use addMetadataFilter instead.)

Each filter function will be called in the order they were added, and will receive a payload object as the only argument, containing the data about to be sent to the APM Server.

The format of the payload depends on the event type being sent. For details about the different formats, see the events intake API docs.

The filter function is synchronous and should return the manipulated payload object. If a filter function doesn’t return any value or returns a falsy value, the remaining filter functions will not be called and the payload will not be sent to the APM Server.

Example usage:

apm.addFilter(function redactSecretHeader(payload) {
  if (payload.context &&
      payload.context.request &&
      payload.context.request.headers &&
      payload.context.request.headers['x-secret']) {
    // redact sensitive data
    payload.context.request.headers['x-secret'] = '[REDACTED]'
  }

  // remember to return the modified payload
  return payload
})

A set of built-in filters are added by default. See filterHttpHeaders for details.

Though you can also use filter functions to add new contextual information to the user and custom properties, it’s recommended that you use apm.setUserContext() and apm.setCustomContext() for that purpose.

apm.addErrorFilter(fn)

edit

Added in: v2.0.0

Similar to apm.addFilter(), but the fn will only be called with error payloads.

apm.addTransactionFilter(fn)

edit

Added in: v2.0.0

Similar to apm.addFilter(), but the fn will only be called with transaction payloads.

apm.addSpanFilter(fn)

edit

Added in: v2.0.0

Similar to apm.addFilter(), but the fn will only be called with span payloads.

apm.addMetadataFilter(fn)

edit

Added in: v3.14.0

Use addMetadataFilter(fn) to supply a filter function for the metadata object sent to the APM Server. This will allow you to manipulate the data being sent, for instance to remove possibly sensitive information.

Each filter function will be called in the order they were added, and will receive a metadata object as the only argument. The filter function is synchronous and must return the manipulated object. Example usage:

apm.addMetadataFilter(function dropArgv(metadata) {
  if (metadata.process && metadata.process.argv) {
    delete metadata.process.argv
  }
  return metadata
})

Warning: It is the responsibility of the author to ensure the returned object conforms to the metadata schema otherwise all APM data injest will fail. A metadata filter that breaks the metadata will result in error logging from the agent, something like:

ERROR (elastic-apm-node): APM Server transport error (400): Unexpected APM Server response
APM Server accepted 0 events in the last request
Error: validation error: 'metadata' required
  Document: {"metadata":null}

apm.setUserContext(context)

edit

Added in: v0.1.0

Call this to enrich collected performance data and errors with information about the user/client. This function can be called at any point during the request/response life cycle (i.e. while a transaction is active).

The given context will be added to the active transaction. If no active transaction can be found, false is returned. Otherwise true.

It’s possible to call this function multiple times within the scope of the same active transaction. For each call, the properties of the context argument are shallow merged with the context previously given.

If an error is captured, the context from the active transaction is used as context for the captured error, and any custom context given as the 2nd argument to apm.captureError takes precedence and is shallow merged on top.

The provided user context is stored under context.user in Elasticsearch on both errors and transactions.

apm.setCustomContext(context)

edit

Added in: v0.1.0

  • context <Object> Can contain any property that can be JSON encoded.

Call this to enrich collected errors and transactions with any information that you think will help you debug performance issues or errors. This function can be called at any point while a transaction is active (e.g. during the request/response life cycle of an incoming HTTP request).

The provided custom context is stored under context.custom in APM Server pre-7.0, or transaction.custom and error.custom in APM Server 7.0+.

The given context will be added to the active transaction. If no active transaction can be found, false is returned. Otherwise true.

It’s possible to call this function multiple times within the scope of the same active transaction. For each call, the properties of the context argument are shallow merged with the context previously given.

If an error is captured, the context from the active transaction is used as context for the captured error, and any custom context given as the 2nd argument to apm.captureError takes precedence and is shallow merged on top.

Before using custom context, ensure you understand the different types of metadata that are available.

apm.setLabel(name, value[, stringify = true])

edit

Added in: v0.1.0
Renamed from apm.setTag() to apm.setLabel(): v2.10.0
Added stringify argument in: v3.11.0

  • name <string> Any periods (.), asterisks (*), or double quotation marks (") will be replaced by underscores (_), as those characters have special meaning in Elasticsearch
  • value <string> | <number> | <boolean>
  • stringify <boolean> Defaults to true. When true, if a non-string value is given, it is converted to a string before being sent to the APM Server.

Set a label on the current transaction. You can set multiple labels on the same transaction. If an error happens during the current transaction, it will also get tagged with the same label.

Labels are key/value pairs that are indexed by Elasticsearch and therefore searchable (as opposed to data set via apm.setCustomContext()). Before using custom labels, ensure you understand the different types of metadata that are available.

Avoid defining too many user-specified labels. Defining too many unique fields in an index is a condition that can lead to a mapping explosion.

apm.addLabels({ [name]: value }[, stringify = true])

edit

Added in: v1.5.0
Renamed from apm.addTags() to apm.addLabels(): v2.10.0
Added stringify argument in: v3.11.0

  • labels <Object> Contains key/value pairs:

    • name <string> Any periods (.), asterisks (*), or double quotation marks (") will be replaced by underscores (_), as those characters have special meaning in Elasticsearch
    • value <string> | <number> | <boolean>
  • stringify <boolean> Defaults to true. When true, if a non-string value is given, it is converted to a string before being sent to the APM Server.

Add several labels on the current transaction. You can add labels multiple times. If an error happens during the current transaction, it will also get tagged with the same labels.

Labels are key/value pairs that are indexed by Elasticsearch and therefore searchable (as opposed to data set via apm.setCustomContext()). Before using custom labels, ensure you understand the different types of metadata that are available.

Avoid defining too many user-specified labels. Defining too many unique fields in an index is a condition that can lead to a mapping explosion.

apm.setGlobalLabel(name, value)

edit

Added in: v3.47.0

Extends the globalLabels configuration. It allows setting labels that are applied to all transactions. A potential use case is to specify a label with the state of your application: 'initializing' | 'available' | 'unhealthy'.

Labels are key/value pairs that are indexed by Elasticsearch and therefore searchable (as opposed to data set via apm.setCustomContext()). Before using custom labels, ensure you understand the different types of metadata that are available.

Avoid defining too many user-specified labels. Defining too many unique fields in an index is a condition that can lead to a mapping explosion.

apm.captureError(error[, options][, callback])

edit

Added in: v0.1.0

  • error - Can be either an <Error> object, a message string, or a special parameterized message object
  • options <Object> The following options are supported:

    • timestamp <number> The time when the error happened. Must be a Unix Time Stamp representing the number of milliseconds since January 1, 1970, 00:00:00 UTC. Sub-millisecond precision can be achieved using decimals. If not provided, the current time will be used
    • message - If the error argument is an <Error> object, it’s possible to use this option to supply an additional message string that will be stored along with the error message under log.message
    • user - See metadata section for details about this option
    • custom - See metadata section for details about this option
    • request <http.IncomingMessage> You can associate an error with information about the incoming request to gain additional context such as the request url, headers, and cookies. However, in most cases, the agent will detect if an error was in response to an http request and automatically add the request details for you. See http requests section for more details.
    • response <http.ServerResponse> You can associate an error with information about the http response to get additional details such as status code and headers. However, in most cases, the agent will detect if an error occured during an http request and automatically add response details for you. See http responses section for more details.
    • handled <boolean> Adds additional context to the exception to show whether the error is handled or uncaught. Unhandled errors are immediately flushed to APM server, in case the application is about the crash. Default: true.
    • labels <Object> Add additional context with labels, these labels will be added to the error along with the labels from the current transaction. See the apm.addLabels() method for details about the format.
    • captureAttributes <boolean> Whether to include properties on the given <Error> object in the data sent to the APM Server (as error.exception.attributes). Default: true.
    • skipOutcome <boolean> Whether to skip setting the outcome value for the current span to failure. See Span outcome for more information. Default: false.
    • parent Transaction | Span | null - A Transaction or Span instance to make the parent of this error. If not given (or undefined), then the current span or transaction will be used. If null is given, then no span or transaction will be used. (Added in v3.33.0.)
  • callback - Will be called after the error has been sent to the APM Server. It will receive an Error instance if the agent failed to send the error, and the id of the captured error.

Send an error to the APM Server:

apm.captureError(new Error('boom!'))

Message strings

edit

Instead of an Error object, you can log a plain text message:

apm.captureError('Something happened!')

This will also be sent as an error to the APM Server, but will not be associated with an exception.

Parameterized message object

edit

Instead of an Error object or a string, you can supply a special parameterized message object:

apm.captureError({
  message: 'Could not find user %s with id %d in the database',
  params: ['Peter', 42]
})

This makes it possible to better group error messages that contain variable data like ID’s or names.

Metadata

edit

To ease debugging it’s possible to send some extra data with each error you send to the APM Server. The APM Server intake API supports a lot of different metadata fields, most of which are automatically managed by the Elastic APM Node.js Agent. But if you wish you can supply some extra details using user or custom. For more details on the properties accepted by the events intake API see the events intake API docs.

To supply any of these extra fields, use the optional options argument when calling apm.captureError().

Here are some examples:

// Sending some extra details about the user
apm.captureError(error, {
  user: {
    id: 'unique_id',
    username: 'foo',
    email: '[email protected]'
  }
})

// Sending some arbitrary details using the `custom` field
apm.captureError(error, {
  custom: {
    some_important_metric: 'foobar'
  }
})

To supply per-request metadata to all errors captured in one central location, use apm.setUserContext() and apm.setCustomContext().

HTTP requests

edit

Besides the options described in the metadata section, you can use the options argument to associate the error with an HTTP request:

apm.captureError(err, {
  request: req // an instance of http.IncomingMessage
})

This will log the URL that was requested, the HTTP headers, cookies and other useful details to help you debug the error.

In most cases, this isn’t needed, as the agent is pretty smart at figuring out if your Node.js app is an HTTP server and if an error occurred during an incoming request. In which case it will automate this processes for you.

HTTP responses

edit

Besides the options described in the metadata section, you can use the options argument to associate the error with an HTTP response:

apm.captureError(err, {
  response: res // an instance of http.ServerResponse
})

This will log the response status code, headers and other useful details to help you debug the error.

In most cases, this isn’t needed, as the agent is pretty smart at figuring out if your Node.js app is an HTTP server and if an error occurred during an incoming request. In which case it will automate this processes for you.

apm.middleware.connect()

edit

Added in: v0.1.0

Returns a middleware function used to collect and send errors to the APM Server.

const apm = require('elastic-apm-node').start()
const connect = require('connect')

const app = connect()

// your regular middleware:
app.use(...)
app.use(...)

// your main HTTP router
app.use(function (req, res, next) {
  throw new Error('Broke!')
})

// add Elastic APM in the bottom of the middleware stack
app.use(apm.middleware.connect())

app.listen(3000)

apm.middleware.connect must be added to the middleware stack before any other error handling middleware functions or there’s a chance that the error will never get to the agent.

apm.startTransaction([name][, type][, subtype][, action][, options])

edit

Added in: v0.1.0
Transaction subtype and action deprecated in: v3.25.0

  • name <string> The name of the transaction. You can always set this later via transaction.name or apm.setTransactionName(). Default: unnamed
  • type <string> The type of the transaction. You can always set this later via transaction.type.
  • subtype <string> The subtype of the transaction. You can alternatively set this via transaction.subtype. The transaction subtype field is deprecated: it is not used and will be removed in the next major version.
  • action <string> The action of the transaction. You can alternatively set this via transaction.action. The transaction action field is deprecated: it is not used and will be removed in the next major version.
  • options <Object> The following options are supported:

    • startTime <number> The time when the transaction started. Must be a Unix Time Stamp representing the number of milliseconds since January 1, 1970, 00:00:00 UTC. Sub-millisecond precision can be achieved using decimals. If not provided, the current time will be used
    • childOf <string> A W3C trace-context "traceparent" string, typically received from a remote service call.
    • tracestate <string> A W3C trace-context "tracestate" string.
    • links <Array> Span links. A transaction can refer to zero or more other transactions or spans (separate from its parent). Span links will be shown in the Kibana APM app trace view. The links argument is an array of objects with a single "context" field that is a Transaction, Span, or W3C trace-context traceparent string. For example: apm.startTransaction('aName', { links: [{ context: anotherSpan }] }).

Start a new transaction.

Use this function to create a custom transaction. Note that the agent will do this for you automatically whenever your application receives an incoming HTTP request. You only need to use this function to create custom transactions.

There’s a special type called request which is used by the agent for the transactions automatically created when an incoming HTTP request is detected.

See the Transaction API docs for details on how to use custom transactions.

apm.endTransaction([result][, endTime])

edit

Added in: v0.1.0

  • result <string> Describes the result of the transaction. This is typically the HTTP status code, or e.g. "success" or "failure" for a background task
  • endTime <number> The time when the transaction ended. Must be a Unix Time Stamp representing the number of milliseconds since January 1, 1970, 00:00:00 UTC. Sub-millisecond precision can be achieved using decimals. If not provided, the current time will be used

Ends the active transaction. If no transaction is currently active, nothing happens.

Note that the agent will do this for you automatically for all regular HTTP transactions. You only need to use this function to end custom transactions created by apm.startTransaction() or if you wish the end a regular transaction prematurely.

Alternatively you can call end() directly on an active transaction object.

apm.currentTransaction

edit

Added in: v1.9.0

Get the currently active transaction, if used within the context of a transaction.

If there’s no active transaction available, null will be returned.

apm.currentSpan

edit

Added in: v2.0.0

Get the currently active span, if used within the context of a span.

If there’s no active span available, null will be returned.

apm.currentTraceparent

edit

Added in: v2.9.0

Get the serialized traceparent string of the current transaction or span.

If there’s no active transaction or span available, null will be returned.

apm.setTransactionName(name)

edit

Added in: v0.1.0

  • name <string> Set or overwrite the name of the current transaction.

If you use a supported router/framework the agent will automatically set the transaction name for you.

If you do not use Express, hapi, koa-router, Restify, or Fastify or if the agent for some reason cannot detect the name of the HTTP route, the transaction name will default to METHOD unknown route (e.g. POST unknown route).

Read more about naming routes manually in the Get started with a custom Node.js stack article.

apm.startSpan([name][, type][, subtype][, action][, options])

edit

Added in: v1.1.0

  • name <string> The name of the span. You can alternatively set this via span.name. Default: unnamed
  • type <string> The type of the span. You can alternatively set this via span.type.
  • subtype <string> The subtype of the span. You can alternatively set this via span.subtype.
  • action <string> The action of the span. You can alternatively set this via span.action.
  • options <Object> The following options are supported:

    • startTime <number> The time when the span started. Must be a Unix Time Stamp representing the number of milliseconds since January 1, 1970, 00:00:00 UTC. Sub-millisecond precision can be achieved using decimals. If not provided, the current time will be used
    • exitSpan <boolean> Make an "exit span". Exit spans represent outgoing communication. They are used to create a node in the Service Map and a downstream service in the Dependencies Table. The provided subtype will be used as the downstream service name.
    • links <Array> Span links. A span can refer to zero or more other transactions or spans (separate from its parent). Span links will be shown in the Kibana APM app trace view. The links argument is an array of objects with a single "context" field that is a Transaction, Span, or W3C trace-context traceparent string. For example: apm.startSpan('aName', { links: [{ context: anotherSpan }] }).

Start and return a new custom span associated with the current active transaction. This is the same as getting the current transaction with apm.currentTransaction and, if a transaction was found, calling transaction.startSpan(name, type, options) on it.

When a span is started it will measure the time until span.end() is called.

See Span API docs for details on how to use custom spans.

If there’s no active transaction available, null will be returned.

apm.handleUncaughtExceptions([callback])

edit

Added in: v0.1.0

By default, the agent will terminate the Node.js process when an uncaught exception is detected. Use this function if you need to run any custom code before the process is terminated.

apm.handleUncaughtExceptions(function (err) {
  // Do your own stuff... and then exit:
  process.exit(1)
})

The callback is called after the event has been sent to the APM Server with the following arguments:

  • err <Error> the captured exception

This function will also enable the uncaught exception handler if it was disabled using the captureExceptions configuration option.

If you don’t specify a callback, the node process is terminated automatically when an uncaught exception has been captured and sent to the APM Server.

It is recommended that you don’t leave the process running after receiving an uncaught exception, so if you are using the optional callback, remember to terminate the node process.

apm.flush([callback])

edit

Added in: v0.12.0

// with node-style callback
apm.flush(function (err) {
  // Flush complete
})

// with promises
apm.flush().then(function () {
  // Flush complete
}).catch(function (err) {
  // Flush returned an error
})

// inside of an async function
try {
  await apm.flush()
  // Flush complete
} catch (err) {
  // Flush returned an error
}

Manually end the active outgoing HTTP request to the APM Server. The HTTP request is otherwise ended automatically at regular intervals, controlled by the apiRequestTime and apiRequestSize config options.

If an optional callback is provided as the first argument to this method, it will call callback(flushErr) when complete. If no callback is provided, then a Promise will be returned, which will either resolve with void or reject with flushErr.

The callback is called (or the Promise resolves if no callback argument is provided) after the active HTTP request has ended. The callback is called even if no HTTP request is currently active.

apm.lambda([type, ]handler)

edit

Added in: v1.4.0

exports.hello = apm.lambda(function (event, context, callback) {
  callback(null, `Hello, ${payload.name}!`)
})

Manually instrument an AWS Lambda function to form a transaction around each execution. Optionally, a type may also be provided to group lambdas together. By default, "lambda" will be used as the type name.

Read more lambda support in the Lambda article.

apm.addPatch(modules, handler)

edit

Added in: v2.7.0

  • modules <string> | <string[]> Name of module(s) to apply the patch to, when required.
  • handler <Function> | <string> Must be a patch function or a path to a module exporting a patch function

    • exports <Object> The original export object of the module
    • agent - The agent instance to use in the patch function
    • options <Object> The following options are supported:

Register a module patch to apply on intercepted require calls.

A module can have any number of patches and will be applied in the order they are added.

apm.addPatch('timers', (exports, agent, { version, enabled }) => {
  const setTimeout = exports.setTimeout
  exports.setTimeout = (fn, ms) => {
    const span = agent.startSpan('set-timeout')
    return setTimeout(() => {
      span.end()
      fn()
    }, ms)
  }

  return exports
})

// or ...

apm.addPatch(['hapi', '@hapi/hapi'], (exports, agent, { version, enabled }) => {
  const setTimeout = exports.setTimeout
  exports.setTimeout = (fn, ms) => {
    const span = agent.startSpan('set-timeout')
    return setTimeout(() => {
      span.end()
      fn()
    }, ms)
  }

  return exports
})

// or ...

apm.addPatch('timers', './timer-patch')

apm.removePatch(modules, handler)

edit

Added in: v2.7.0

Removes a module patch. This will generally only be needed when replacing an existing patch. To disable instrumentation while keeping context propagation support, see disableInstrumentations.

apm.removePatch('timers', './timers-patch')

// or ...

apm.removePatch(['timers'], './timers-patch')

// or ...

apm.removePatch('timers', timerPatchFunction)

apm.clearPatches(modules)

edit

Added in: v2.7.0

Clear all patches for the given module. This will generally only be needed when replacing an existing patch. To disable instrumentation while keeping context propagation support, see disableInstrumentations.

apm.clearPatches('timers')

// or ...

apm.clearPatches(['timers'])

apm.currentTraceIds

edit

Added in: v2.17.0

apm.currentTraceIds produces an object containing trace.id and either transaction.id or span.id when a current transaction or span is available. When no transaction or span is available it will return an empty object. This enables log correlation to APM traces with structured loggers.

{
  "trace.id": "abc123",
  "transaction.id": "abc123"
}
// or ...
{
  "trace.id": "abc123",
  "span.id": "abc123"
}

apm.registerMetric(name[, labels], callback)

edit

This functionality is in technical preview and may be changed or removed in a future release. Elastic will work to fix any issues, but features in technical preview are not subject to the support SLA of official GA features.

  • name <string> Name of the metrics.
  • labels <Object> Contains key/value pairs. Optional labels. Omittable.
  • callback <Function> Must be a function that returns the current metric value.

Register a metric callback.

Take care not to use the names of built-in metrics.

apm.registerMetric( 'ws.connections' , () => {
  return wss.clients.size;
})

// or, to additionally label the metric with "module: 'ws'":

apm.registerMetric( 'ws.connections' , {module : 'ws'}, () => {
  return wss.clients.size;
})

apm.setTransactionOutcome(outcome)

edit

Added in: v3.12.0

Will set the outcome property on the current transaction.

See the Transaction Outcome docs for more information.

apm.setSpanOutcome(outcome)

edit

Added in: v3.12.0

Will set the outcome property on the current span.

See the Span Outcome docs for more information.