7. Running in production

Status: Runnable. Everything on this page works today.

What Avi.js does for you in production, and the settings that control it: body limits, safe errors, client disconnects, graceful shutdown, timeouts and streaming.

These defaults are part of Avi.js from the start, not add-ons:

Concern	Default behavior
Request bodies	Limited to 1 MiB (bodyLimit). Oversized bodies get 413, never read past it.
Error responses	Generic 5xx bodies in every environment. Details go to your logger.
Client disconnects	ctx.signal aborts so handlers can stop work early; streams stop being read.
Shutdown	app.close() drains in-flight requests, forces after shutdownTimeout, then disposes providers in reverse order.
Slow clients	headersTimeout and requestTimeout close connections that send too slowly.
Malformed input	Bad URL encodings get 400; Node rejects broken HTTP and oversized headers (431).
Streaming	Streamed responses respect backpressure from slow clients.

Streaming responses

Return a stream instead of a value. Avi.js sends it chunked, reads from it only as fast as the client receives, and stops reading if the client disconnects:

import { createReadStream } from 'node:fs';
import { reply } from '@avijs/core';

route({
  method: 'GET',
  path: '/export.csv',
  handler: () =>
    reply(createReadStream('./export.csv'), {
      headers: { 'content-type': 'text/csv; charset=utf-8' },
    }),
});

Supported return values:

Return	Sent as
a Node Readable	chunked, application/octet-stream unless you set a type
a web ReadableStream	the same
a web Response	its status, headers and body; ctx.setHeader headers are added

HEAD requests get the headers without reading the stream. If a stream fails after the response has started, the status can no longer change: Avi.js logs the error and closes the connection, so the client sees a truncated response instead of a silently wrong one.

A production entry point

import { createAvi } from '@avijs/core';

const app = createAvi({
  bodyLimit: 256 * 1024,
  shutdownTimeout: 15_000,
  headersTimeout: 10_000,
  requestTimeout: 30_000,
  logger: { error: (message, error) => console.error(message, error) },
  onResponse: ({ method, route, status, duration }) => {
    metrics.observe({ method, route, status }, duration);
  },
});

// app.register(...)

await app.listen(Number(process.env['PORT'] ?? 3000), '0.0.0.0');

const shutdown = () => {
  void app.close().then(() => process.exit(0));
};
process.once('SIGINT', shutdown);
process.once('SIGTERM', shutdown);

• Set shutdownTimeout below your platform's kill grace period (Kubernetes defaults to 30 seconds) so in-flight requests finish before the process is killed.
• Behind a load balancer, set keepAliveTimeout above the balancer's idle timeout so it never reuses a connection Node has just closed.
• requestTimeout, headersTimeout, keepAliveTimeout and connectionsCheckingInterval are passed straight to node:http.

Performance

The repository's benchmarks/ folder measures routing, startup, HTTP throughput and validation overhead, with the method and machine for every number. They are local measurements, not comparisons with other frameworks; Avi.js makes no speed claims until those exist.

Next

Chapter 8: Testing your app.