Node.js runtime upgraded to Node 22

The atime, mtime, ctime, and birthtime properties on Stats objects are now lazy-loaded and no longer appear when you enumerate the object, for example with spread syntax or JSON.stringify(). To keep the old behavior, copy them explicitly:

{ ...stats, mtime: stats.mtime, atime: stats.atime, ctime: stats.ctime, birthtime: stats.birthtime }

Node 22 defines a global Iterator. Your own code is unlikely to be affected, but polyfills that patch Iterator.prototype unconditionally can throw TypeError: Cannot redefine property. Check your dependencies for iterator-helper polyfills such as es-iterator-helpers or core-js iterator shims.

Node 22 defines a global WebSocket. Guarded assignments like if (!global.WebSocket) { global.WebSocket = require('ws') } now silently do nothing, so the service might use Node’s built-in implementation instead of the one it expected. Check your dependencies for this pattern.

Node 22 defines a global navigator. Code that uses typeof navigator !== 'undefined' to detect a browser now misidentifies Node as a browser. Use a check such as typeof window !== 'undefined' instead.

URL and URLSearchParams instances can no longer be cloned or transferred. Passing one through a structured clone, for example MessagePort.postMessage() in worker_threads, throws a DataCloneError. Convert the object to a string with toString() before sending it, and reconstruct it on the receiving end.

The bundled llhttp HTTP parser is stricter and more spec-compliant. Malformed input that Node previously tolerated, such as invalid characters in header or URL fields or embedded raw \n in headers, is now rejected with a parse error. Confirm that clients sending traffic to your service’s HTTP server still parse cleanly.

When you use both response.setHeader() and response.writeHead(), duplicate headers are now preserved instead of collapsed. Avoid mixing the two calls, or deduplicate headers before passing them.

The default highWaterMark increased from 16 KiB to 64 KiB, so streams buffer more data by default. Check services with tight memory budgets, or tests that assert exact chunk sizes or counts on data events.