node-svc-base

Bringing some consistent form and best-practices to node apps, especially when it comes to metrics.

Basic usage

Create a svc.js:

const { Svc } = require('@balena/node-svc-base');

const { metricsConfig } = require('./metrics');
const { healthCheck, handleSIGTERM } = require('./lifecycle');

const svc = new Svc('mock-svc',
    {
        healthCheck,
        handleSIGTERM,
        metricsConfig
    }
);

module.exports = svc;

This requires arguments of types:

    healthCheck: () => Promise<boolean>,
    handleSIGTERM: (terminate : () => void) => void,
    metricsConfig: MetricsConfig

(for the definition of the type MetricsConfig see ./src/lib/metrics.ts)

This allows use in other files, like:

const svc = require('./svc');
const { app, metrics, server } = svc;

app.get('/hello', (req, res) => {
    metrics.inc('hello_total', 1);
    ses.send('hello');
});
server.timeout = 90000;
svc.setInitialized();

or, in typescript,

import * as svc from './svc';
...

NOTE: It's important to call setInitialized() when your app is ready to serve, or it won't be marked healthy by k8s.

See ./automation/usr/src/app/ for the full example app.

Params

The full params available when constructing a service can be seen in src/index.ts:

interface Params {
	healthCheck: () => Promise<boolean>;
	metricsConfig: MetricsConfig;
	handleSIGTERM: (() => void) => void;
	monitoringDir?: string;
	app?: express.Application;
}

MetricsConfig

The metricsConfig parameter is important since it allows one to specify many different but related aspects of metrics, as an object with top-level keys: metrics, recording_rules, alerts, dashboards. See ./src/lib/metrics.ts for the type signatures of the objects under these keys.

Graceful Shutdown

When the process receives SIGTERM, the handleSIGTERM function will be called, and svc.isTerminating() will begin to return true. When the work has been done to gracefully handle SIGTERM, the user code can call svc.terminate() or - which is the same function - can call terminate : () => void passed to handleSIGTERM, which will call process.exit(0).

Ports

• 8080 express app (components add their routes, middlewares here)
• 9090 prometheus metrics (from @balena/node-metrics-gatherer)
• 9393 service-discovery for dashboards and alerts

SIGTERM

The sigterm handler function will be run, and on completion (it should return a promise that resolves), a file /var/node-svc-base/graceful-shutdown-${name} will be written. A process waiting for graceful shutdown to complete can watch for this file's creation (for example, a kubernetes preStop hook can send systemctl kill -s SIGTERM ${service} and then wait on this file.

Service discovery of dashboards / alerts

The following will automatically served, to be discovered by our monitoring system:

• Any .json files representing grafana dashboards in /etc/monitoring/dashboards/

• Any prometheus alerts defined in /etc/monitoring/alerting-rules.yml see Prometheus docs - "Alerting Rules" for syntax.