Thread pool là gì? Tại sao cần thread pool và cách sizing đúng cách?

Thread pool là tập hợp các worker threads được tạo sẵn và tái sử dụng, thay vì tạo/hủy thread mới cho mỗi task — vì tạo thread tốn ~1MB stack, time để OS allocate, và context switch overhead.

• Hoạt động: task queue nhận công việc, worker thread nhàn rỗi lấy task từ queue, thực thi, rồi chờ task tiếp theo.
• ThreadPoolExecutor (Java), worker_threads pool (Node.js), Goroutine worker pattern (Go).
• Sizing là bài toán quan trọng: với CPU-bound tasks, pool size = số CPU cores (thêm thread không giúp, chỉ tốn context switch); với I/O-bound tasks, pool size lớn hơn số CPU (vì thread chờ I/O không chiếm CPU) — formula tham khảo: threads = cores * (1 + wait_time/service_time).
• Pool quá nhỏ: task queue dài, high latency.
• Pool quá lớn: memory tốn kém, nhiều context switch, performance giảm.
• Java Tomcat default pool size 200; Node.js libuv I/O thread pool default 4 (UV_THREADPOOL_SIZE).

A thread pool is a set of pre-created worker threads that are reused, rather than spawning and destroying a new thread for every task — because thread creation costs ~1MB of stack space, OS allocation time, and context-switch overhead.

How it works: a task queue receives incoming work, an idle worker thread picks up a task, executes it, then waits for the next one. Examples: ThreadPoolExecutor (Java), worker_threads pool (Node.js), goroutine worker pattern (Go). Sizing is critical: for CPU-bound tasks, pool size = number of CPU cores (adding more threads only increases context switching with no gain); for I/O-bound tasks, the pool can be larger than the core count (since threads waiting on I/O do not consume CPU) — a reference formula is threads = cores * (1 + wait_time / service_time). Too small a pool: long task queues and high latency. Too large a pool: wasted memory, excessive context switching, and degraded performance. Java Tomcat defaults to a pool size of 200; Node.js libuv's I/O thread pool defaults to 4 (configurable via UV_THREADPOOL_SIZE).