Go: What is the GMP scheduler?

GMP — Goroutine (G), Machine (M), Processor (P). Go uses this three-level scheduler to efficiently multiplex goroutines onto OS threads.

G (Goroutine): Represents a single unit of work. Contains its own stack (starts at 2KB, grows dynamically), program counter, and state (runnable, running, waiting, syscall).

M (Machine): An OS thread created by the runtime. M threads actually execute goroutine code. The Go runtime limits the number of M threads (default max 10,000).

P (Processor): A logical processor that holds a local run queue of goroutines. The number of P processors is set by GOMAXPROCS (defaults to number of CPU cores).

How scheduling works:

1. Each P has a local queue of runnable goroutines (capacity 256).
2. When a P's queue is empty, it tries to steal goroutines from other P queues.
3. If stealing fails, it checks the global run queue.
4. If the global queue is empty, it checks the network poller.
5. M threads bind to P processors to execute goroutines.

1// Check and set GOMAXPROCS
2fmt.Println("GOMAXPROCS:", runtime.GOMAXPROCS(0))
3
4// Set to 4 processors for 4-core machine
5runtime.GOMAXPROCS(4)
6
7// Goroutine scheduling
8runtime.Gosched() // Yield to scheduler

Work stealing: When a P has nothing to do, it randomly selects another P and takes half its goroutines. This prevents idle processors.

When to use GMP knowledge:

• Tuning GOMAXPROCS for CPU-bound workloads.
• Debugging performance issues with goroutine scheduling.
• Understanding why runtime.LockOSThread() behaves differently.

Common mistakes:

• Setting GOMAXPROCS too high, causing excessive context switching.
• Not understanding that runtime.GOMAXPROCS(0) returns the CURRENT value without changing it.
• Assuming goroutines always run in parallel — they only run in parallel when GOMAXPROCS > 1 and multiple Ms are available.
• Using runtime.LockOSThread() without understanding it pins the goroutine to one OS thread.