Understanding the Go Runtime: The Scheduler

5 days ago

The Go scheduler uses the GMP model: G (goroutine), M (OS thread), and P (processor).
Goroutines (G) are lightweight, starting with a 2KB stack, and can be recycled for efficiency.
M represents an OS thread, executing goroutines, with a special g0 goroutine for runtime tasks.
P is a scheduling context, holding local run queues and memory caches, enabling efficient work distribution.
The scheduler's global state, schedt, manages shared resources like the global run queue and idle lists.
Goroutines transition through states: _Grunnable, _Grunning, _Gwaiting, and _Gdead, managing their own state changes.
Blocking operations like channel receives or system calls involve goroutines parking themselves.
System calls may lead to P detachment if prolonged, allowing other goroutines to continue running.
Stack growth and preemption (cooperative and asynchronous) ensure goroutines don't monopolize resources.
The scheduling loop prioritizes runtime work, checks local and global queues, and employs work stealing to balance load.
Spinning threads briefly stay active to handle new work, optimizing responsiveness without wasting CPU.
Goroutine context switches are fast (~50-100ns), leveraging minimal state saving for efficiency.

Hasty Briefsbeta