This is a middleware for express servers, that expose metrics for prometheus.
The metrics exposed allows to calculate common RED (Request, Error rate, Duration of requests), and USE (Utilisation, Error rate, and Saturation), metrics
npm i --save @zachery2006/express-prometheus-middleware
Name | Description | Default |
---|---|---|
metricsPath | Url route that will expose the metrics for scraping. | /metrics |
metricsApp | Express app that will expose metrics endpoint, if an app is provided, use it, instead of instantiating a new one | null |
collectDefaultMetrics | Whether or not to collect prom-client default metrics. These metrics are usefull for collecting saturation metrics, for example. |
true |
collectGCMetrics | Whether or not to collect garbage collection metrics via module prometheus-gc-stats . Dependency prometheus-gc-stats is marked as optional, hence if this option is set to true but npm/yarn could not install the dependency, no garbage collection metric will be collected. |
false |
requestDurationBuckets | Buckets for the request duration metrics (in seconds) histogram | Uses prom-client utility: Prometheus.exponentialBuckets(0.05, 1.75, 8) |
requestLengthBuckets | Buckets for the request length metrics (in bytes) histogram | no buckets (The request length metrics are not collected): [] |
responseLengthBuckets | Buckets for the response length metrics (in bytes) histogram | no buckets (The response length metrics are not collected) [] |
normalizePath | Optional, list of regexes to be used as argument to url-value-parser, this will cause extra route params, to be replaced later argument. | no normalizePath: [] |
authenticate | Optional authentication callback, the function should receive as argument, the req object and return truthy for sucessfull authentication, or falsy, otherwise. This option supports Promise results. |
null |
suffix | Optional suffix for the metrics name | no suffix added |
customLabels | Optional Array containing extra labels, used together with transformLabels |
no extra labels: [] |
transformLabels | Optional function(labels, req, res) adds to the labels object dynamic values for each label in customLabels |
null |
normalizeStatus | Optional parameter to disable normalization of the status code. Example of normalized and non-normalized status code respectively: 4xx and 422. | true |
const express = require('express');
const promMid = require('express-prometheus-middleware');
const app = express();
const PORT = 9091;
app.use(promMid({
metricsPath: '/metrics',
collectDefaultMetrics: true,
requestDurationBuckets: [0.1, 0.5, 1, 1.5],
requestLengthBuckets: [512, 1024, 5120, 10240, 51200, 102400],
responseLengthBuckets: [512, 1024, 5120, 10240, 51200, 102400],
normalizePath: [
['^/test/.*', '/httpMetricTest'],
['^/status/.*', '/status'],
['^/hello*', '/hello'],
['^/birds/.*', '/#birds'],
// From deep to surface
['^/customers/.*/companies/.*/users/.*', '/customers/#customer/companies/#company/users/#user'],
['^/customers/.*/companies/.*', '/customers/#customer/companies/#company'],
['^/customers/.*/add/.*', '/customers/#customer/add/#add'],
],
/**
* Uncomenting the `authenticate` callback will make the `metricsPath` route
* require authentication. This authentication callback can make a simple
* basic auth test, or even query a remote server to validate access.
* To access /metrics you could do:
* curl -X GET user:password@localhost:9091/metrics
*/
// authenticate: req => req.headers.authorization === 'Basic dXNlcjpwYXNzd29yZA==',
/**
* The prefix option will cause all metrics to have the given prefix.
* E.g.: `app_prefix_http_requests_total`
*/
// prefix: 'app_prefix_',
/**
* Can add custom labels with customLabels and transformLabels options
*/
// customLabels: ['contentType'],
// transformLabels(labels, req) {
// // eslint-disable-next-line no-param-reassign
// labels.contentType = req.headers['content-type'];
// },
}));
// curl -X GET localhost:9091/hello?name=Chuck%20Norris
app.get('/hello', (req, res) => {
console.log('GET /hello');
const { name = 'Anon' } = req.query;
res.json({ message: `Hello, ${name}!` });
});
app.listen(PORT, () => {
console.log(`Example api is listening on http://localhost:${PORT}`);
});
- Default metrics from prom-client
http_requests_total
: Counter for total requests received, has labelsroute
,method
,status
http_request_duration_seconds
: - Duration of HTTP requests in seconds, has labelsroute
,method
,status
The labels route
and status
are normalized:
route
: will normalize id like route paramsstatus
: will normalize to status code family groups, like2XX
or4XX
.
In the examples below, Suppose you tagged your application as "myapp", in the prometheus scrapping config.
sum(up{app="myapp"})
Rate of http status code 5XX responses
sum(rate(http_requests_total{status="5XX", app="myapp"}[5m]))
histogram_quantile(0.95, sum(rate(http_request_duration_seconds_bucket{app="myapp"}[5m])) by (le))
histogram_quantile(0.95, sum(rate(http_request_length_bytes_bucket{app="myapp"}[5m])) by (le))
histogram_quantile(0.95, sum(rate(http_response_length_bytes_bucket{app="myapp"}[5m])) by (le))
sum(rate(http_request_duration_seconds_sum{app="myapp"}[5m])) by (instance) / sum(rate(http_request_duration_seconds_count{app="myapp"}[5m])) by (instance)
sum(rate(http_requests_total{app="myapp"}[5m])) by (instance)
In this example we are removing some health/status-check routes, replace them with your needs.
sum(rate(http_requests_total{app="myapp", route!~"/|/healthz"}[5m])) by (instance, route)
rate(process_cpu_system_seconds_total{app="myapp"}[5m])
rate(process_cpu_user_seconds_total{app="myapp"}[5m])
nodejs_heap_size_total_bytes{app="myapp"}
nodejs_heap_size_used_bytes{app="myapp"}