Go: What is the GMP model?

GMP — Goroutine (G), Machine (M), Processor (P). Go's scheduler architecture that enables efficient concurrency by multiplexing goroutines onto OS threads.

Component details:

• G (Goroutine): Execution unit with stack, PC, and scheduling state. States: _Gidle, _Grunnable, _Grunning, _Gsyscall, _Gwaiting.
• M (Machine): OS thread. Created by runtime, limited to 10,000. Binds to P to execute goroutines.
• P (Processor): Logical processor with local run queue (256 capacity). Count = GOMAXPROCS (default: runtime.NumCPU()).

Scheduling algorithm:

1// Check runtime configuration
2fmt.Println("GOMAXPROCS:", runtime.GOMAXPROCS(0))
3fmt.Println("NumCPU:", runtime.NumCPU())
4fmt.Println("NumGoroutine:", runtime.NumGoroutine())
5
6// Scheduler hints
7runtime.Gosched()    // Yield to scheduler
8runtime.Goexit()    // Exit current goroutine

Work stealing process:

1. P's local queue empty → try global queue.
2. Global queue empty → try to steal from random P.
3. Steal half of victim's queue.
4. If nothing → check network poller.
5. If still nothing → M may park/sleep.

Handoff (syscall handling):

1. M enters blocking syscall.
2. M's P is detached.
3. P is given to another idle M (or new M created).
4. Goroutines in P's queue continue running.

When GMP knowledge helps:

• Tuning for CPU-bound vs I/O-bound workloads.
• Debugging scheduling latency.
• Container environment configuration.
• Optimizing goroutine placement.

Common mistakes:

• Not adjusting GOMAXPROCS in containers.
• Confusing GOMAXPROCS(0) (read) with GOMAXPROCS(n) (set).
• Assuming goroutines always run in parallel.
• Using LockOSThread() without need.
• Not understanding work stealing randomness.