Streams trong Node.js là gì? Có những loại stream nào?

Stream xử lý data theo chunks — memory efficiency cực cao: đọc file 1GB chỉ tốn ~64KB RAM (default highWaterMark) thay vì 1GB.

• Bốn loại: Readable (fs.createReadStream, HTTP req, process.stdin), Writable (fs.createWriteStream, HTTP res, process.stdout), Duplex (TCP socket — đọc và ghi độc lập), Transform (zlib.createGzip, crypto.createCipher — transform data flow through). pipeline() API (Node.js 10+) thay thế .pipe(): await pipeline(readStream, transformStream, writeStream) — tự cleanup và propagate errors (.pipe() không handle errors tốt).
• Stream consumers với for await...of: for await (const chunk of readableStream) { process(chunk); } — ergonomic nhất.

Ví dụ thực: serve 1GB file với 0 timeout và thấp RAM: const src = fs.createReadStream(filePath); const gzip = zlib.createGzip(); await pipeline(src, gzip, res) — data chảy qua, không buffer toàn bộ.

• Backpressure: khi Writable chậm hơn Readable, pipeline tự pause Readable tránh OOM — đây là lý do không dùng .pipe() thủ công trong production.

Streams process data in chunks — extreme memory efficiency: reading a 1GB file uses only ~64KB of RAM (default highWaterMark) instead of 1GB.

• Four types: Readable (fs.createReadStream, HTTP req, process.stdin), Writable (fs.createWriteStream, HTTP res, process.stdout), Duplex (TCP socket — read and write independently), Transform (zlib.createGzip, crypto.createCipher — transforms data flowing through). pipeline() API (Node.js 10+) replaces .pipe(): await pipeline(readStream, transformStream, writeStream) — auto-cleanup and error propagation (.pipe() doesn't handle errors well).
• Stream consumers with for await...of: for await (const chunk of readableStream) { process(chunk); } — the most ergonomic approach.
• Practical example: serve a 1GB file with zero timeout and minimal RAM: const src = fs.createReadStream(filePath); const gzip = zlib.createGzip(); await pipeline(src, gzip, res) — data flows through without buffering the entire file.
• Backpressure: when the Writable is slower than the Readable, pipeline automatically pauses the Readable to prevent OOM — this is why you should not use .pipe() manually in production.