Prometheus API Monitoring

• Goal
• Features
• Usage
  • Options
  • Access the metrics
• Custom Metrics
  • Note
• Additional Metric Labels
• Request.js HTTP request duration collector
  • Usage
  • Initialize
  • Options
  • request
  • request-promise-native
  • axios
• Usage in koa
• Test
• Prometheus Examples Queries
  • Apdex
  • 95th Response Time by specific route and status code
  • Median Response Time Overall
  • Median Request Size Overall
  • Median Response Size Overall
  • Average Memory Usage - All services
  • Average Eventloop Latency - All services

Goal

API and process monitoring with Prometheus for Node.js micro-service

Note: Prometheus (prom-client) is a peer dependency since 1.x version

Features

• Collect API metrics for each call
  • Response time in seconds
  • Request size in bytes
  • Response size in bytes
  • Add prefix to metrics names - custom or project name
  • Exclude specific routes from being collect
  • Number of open connections to the server
• Process Metrics as recommended by Prometheus itself
• Endpoint to retrieve the metrics - used for Prometheus scraping
  • Prometheus format
  • JSON format (${path}.json)
• Support custom metrics
• Http function to collect request.js HTTP request duration

Usage

const apiMetrics = require('prometheus-api-metrics');
app.use(apiMetrics())

Options

Access the metrics

To get the metrics in Prometheus format use:

curl http[s]://<host>:[port]/metrics

To get the metrics in JSON format use:

curl http[s]://<host>:[port]/metrics.json

Note:

1. If you pass to the middleware the metricsPath option the path will be the one that you chose.

2. If you are using express framework and no route was found for the request (e.g: 404 status code), the request will not be collected. that's because we'll risk memory leak since the route is not a pattern but a hardcoded string.

Custom Metrics

You can expand the API metrics with more metrics that you would like to expose. All you have to do is:

Require prometheus client

const Prometheus = require('prom-client');

Create new metric from the kind that you like

const checkoutsTotal = new Prometheus.Counter({
  name: 'checkouts_total',
  help: 'Total number of checkouts',
  labelNames: ['payment_method']
});

Update it:

checkoutsTotal.inc({
  payment_method: paymentMethod
})

The custom metrics will be exposed under the same endpoint as the API metrics.

For more info about the Node.js Prometheus client you can read here

Note

This will work only if you use the default Prometheus registry - do not use new Prometheus.Registry()

Additional Metric Labels

You can define additional metric labels by using additionalLabels and extractAdditionalLabelValuesFn options.

For instance:

const apiMetrics = require('prometheus-api-metrics');
app.use(apiMetrics({
  additionalLabels: ['customer', 'cluster'],
  extractAdditionalLabelValuesFn: (req, res) => {
      const { headers } = req.headers;
      return {
        customer: headers['x-custom-header-customer'],
        cluster: headers['x-custom-header-cluster']
      }
  }
}))

Request.js HTTP request duration collector

This feature enables you to easily process the result of Request.js timings feature.

Usage

Initialize

You can choose to initialized this functionality as a Class or not

Class:

const HttpMetricsCollector = require('prometheus-api-metrics').HttpMetricsCollector;
const collector = new HttpMetricsCollector();
collector.init();

Singleton:

const HttpMetricsCollector = require('prometheus-api-metrics').HttpMetricsCollector;
HttpMetricsCollector.init();

Options

• durationBuckets - the histogram buckets for request duration.
• countClientErrors - Boolean that indicates whether to collect client errors as Counter, this counter will have target and error code labels.
• useUniqueHistogramName - Add to metrics names the project name as a prefix (from package.json)
• prefix - A custom metrics names prefix, the package will add underscore between your prefix to the metric name.

For Example:

request

request({ url: 'http://www.google.com', time: true }, (err, response) => {
    Collector.collect(err || response);
});

request-promise-native

return requestPromise({ method: 'POST', url: 'http://www.mocky.io/v2/5bd9984b2f00006d0006d1fd', route: 'v2/:id', time: true, resolveWithFullResponse: true }).then((response) => {
    Collector.collect(response);
}).catch((error) => {
    Collector.collect(error);
});

Notes:

1. In order to use this feature you must use { time: true } as part of your request configuration and then pass to the collector the response or error you got.
2. In order to use the timing feature in request-promise/request-promise-native you must also use resolveWithFullResponse: true
3. Override - you can override the route and target attribute instead of taking them from the request object. In order to do that you should set a metrics object on your request with those attribute:

   request({ method: 'POST', url: 'http://www.mocky.io/v2/5bd9984b2f00006d0006d1fd', metrics: { target: 'www.google.com', route: 'v2/:id' }, time: true }, (err, response) => {...};
   });

axios

const axios = require('axios');
const axiosTime = require('axios-time');

axiosTime(axios);

try {
    const response = await axios({ baseURL: 'http://www.google.com', method: 'get', url: '/' });
    Collector.collect(response);
} catch (error) {
    Collector.collect(error);
}

Notes:

• In order to collect metrics from axios client the axios-time package is required.

Usage in koa

This package supports koa server that uses koa-router and koa-bodyparser

const { koaMiddleware } = require('prometheus-api-metrics')

app.use(koaMiddleware())

Test

npm test

Prometheus Examples Queries

Apdex

(sum(rate(http_request_duration_seconds_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>">, route="<ROUTE_NAME>", le="0.05"}[10m])) by (<SERVICE_LABEL_FIELD>) + sum(rate(http_request_duration_seconds_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>", route="<ROUTE_NAME>", le="0.1"}[10m])) by (<SERVICE_LABEL_FIELD>)) / 2 / sum(rate(http_request_duration_seconds_count{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>", route="<ROUTE_NAME>"}[10m])) by (<SERVICE_LABEL_FIELD>)

95th Response Time by specific route and status code

histogram_quantile(0.95, sum(rate(http_request_duration_seconds_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>", route="<ROUTE_NAME>", code="200"}[10m])) by (le))

Median Response Time Overall

histogram_quantile(0.50, sum(rate(http_request_duration_seconds_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>"}[10m])) by (le))

Median Request Size Overall

histogram_quantile(0.50, sum(rate(http_request_size_bytes_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>"}[10m])) by (le))

Median Response Size Overall

histogram_quantile(0.50, sum(rate(http_response_size_bytes_bucket{<SERVICE_LABEL_FIELD>="<SERVICE_LABEL>"}[10m])) by (le))

Average Memory Usage - All services

avg(nodejs_external_memory_bytes / 1024 / 1024) by (<SERVICE_LABEL_FIELD)

Average Eventloop Latency - All services

avg(nodejs_eventloop_lag_seconds) by (<SERVICE_LABEL_FIELD)