seek-datadog-custom-metrics

🐶 Datadog Custom Metrics

Helpers for sending Datadog custom metrics via hot-shots.

yarn add seek-datadog-custom-metrics

Tagging convention

All custom metrics are prefixed by AppConfig.name. Two global tags are also added to every custom metric:

• AppConfig.environment becomes env:${value}
• AppConfig.version becomes version:${value}

These tags are consistent with tags sent by Gantry via Datadog's AWS integration.

API reference

createStatsDClient

createStatsDClient creates a hot-shots client configured with our tagging convention. This is intended for containerized services, particularly those deployed with Gantry.

import { StatsD } from 'hot-shots';
import { createStatsDClient } from 'seek-datadog-custom-metrics';

// Expects `name`, `version`, `environment` and `metricsServer` properties
import config from '../config';

// This example assumes Bunyan/pino
import { rootLogger } from '../logger';
const errorHandler = (err: Error) => {
  rootLogger.error('StatsD error', err);
};

// Returns a standard hot-shots StatsD instance
const metricsClient = createStatsDClient(StatsD, config, errorHandler);

createLambdaExtensionClient

createLambdaExtensionClient creates a Lambda extension client. This is intended for AWS Lambda functions and is a replacement for createCloudWatchClient.

This client will only submit metrics as a distribution which enables globally accurate aggregations for percentiles (p50, p75, p90, etc).

import { createLambdaExtensionClient } from 'seek-datadog-custom-metrics';

// Expects `name` and `metrics` properties
import config from '../config';

// Returns a standard hot-shots StatsD instance
const { metricsClient, withLambdaExtension } =
  createLambdaExtensionClient(config);

export const handler = withLambdaExtension((event, _ctx) => {
  try {
    logger.info('request');

    await lambdaFunction(event);
  } catch (err) {
    logger.error({ err }, 'request');

    metricsClient.increment('invocation_error');

    throw new Error('invoke error');
  }
});

createNoOpClient

createNoOpClient returns a no-op client. This is intended for use where a MetricsClient interface is expected but you do not wish to provide one, e.g in tests.

import { createNoOpClient } from 'seek-datadog-custom-metrics';

// Returns a `MetricsClient` subset of the full StatsD interface
const metricsClient = createNoOpClient();

createTimedSpan

createTimedSpan wraps an asynchronous closure and records custom Datadog metrics about its performance. This is intended as a lightweight alternative to APM where nested spans aren't required.

import { createTimedSpan } from 'seek-datadog-custom-metrics';

// Takes a StatsD instance or `MetricsClient`
const timedSpan = createTimedSpan(metricsClient);

const loadPrivateKey = async (): Promise<PrivateKey> =>
  await timedSpan(
    // Prefix for the custom metrics
    'secrets.load_private_key',
    // Closure to be timed
    () => client.getSecretValue({ SecretId }).promise(),
  );

httpTracingConfig

The dd-trace package can instrument your application and trace its outbound HTTP requests. However, its emitted trace.http.request metric only captures the HTTP method against the resource.name tag, which is not useful if your application makes HTTP requests to multiple resources and you want to inspect latency by resource.

This configuration object adds a hook to replace the resource.name with a HTTP method and semi-normalised URL. For example, if your application makes the following HTTP request:

PUT https://www.example.com/path/to/123?idempotencyKey=c1083fb6-519c-42bf-8619-08dfd6229954

The trace.http.request metric will see the following tag change:

- resource_name:put
+ resource_name:put_https://www.example.com/path/to/number?idempotencyKey=uuid

Apply the configuration object where you bootstrap your application with the Datadog tracer:

import { httpTracingConfig } from 'seek-datadog-custom-metrics';

// DataDog/dd-trace-js#1118
datadogTracer?.use('http', httpTracingConfig);

This configuration may be superseded in future if the underlying dd-trace implementation is corrected.

DataDog/dd-trace-js#1118