Backpressure trong Node.js streams là gì? Tại sao quan trọng?

• Backpressure: Readable produce data nhanh hơn Writable consume — nếu không xử lý, data buffer trong memory vô hạn cho đến khi OOM crash. highWaterMark: ngưỡng buffer size (default 16KB cho object streams, 16384 bytes cho byte streams) — khi buffer vượt ngưỡng, stream signals slow down.
• Cơ chế manual: writable.write(chunk) trả về false khi internal buffer đầy → stop reading; lắng nghe writable.on('drain', resume) để tiếp tục.

Ví dụ file upload: client upload 1GB/s, DB write chỉ 100MB/s — không có backpressure thì 900MB/s tích lũy trong RAM. pipe() tự xử lý nhưng không handle errors tốt. pipeline() API (stream/promises): await pipeline(req, transform, fs.createWriteStream(dest)) — tự xử lý backpressure VÀ cleanup tất cả streams khi error/complete, không leak.

• Transform streams: implement _transform(chunk, enc, callback), gọi callback() khi sẵn sàng nhận chunk tiếp theo — tự nhiên tạo backpressure.
• Lưu ý: .on('data', handler) puts stream in flowing mode bỏ qua backpressure — dùng async iteration for await (const chunk of readable) thay thế.

• Backpressure: a Readable produces data faster than a Writable can consume it — without handling this, data buffers in memory indefinitely until an OOM crash. highWaterMark: the buffer size threshold (default 16KB for object streams, 16384 bytes for byte streams) — when the buffer exceeds this, the stream signals to slow down.
• Manual mechanism: writable.write(chunk) returns false when the internal buffer is full → stop reading; listen for writable.on('drain', resume) to continue.
• File upload example: client uploads at 1GB/s, DB writes only at 100MB/s — without backpressure, 900MB/s accumulates in RAM. pipe() handles this automatically but doesn't handle errors well. pipeline() API (stream/promises): await pipeline(req, transform, fs.createWriteStream(dest)) — handles backpressure AND cleans up all streams on error/completion, no leaks.
• Transform streams: implement _transform(chunk, enc, callback), call callback() when ready for the next chunk — naturally creates backpressure.

Pitfall: .on('data', handler) puts the stream in flowing mode, bypassing backpressure — use async iteration for await (const chunk of readable) instead.