Calling fork in a go program

I recently read the popular post about Redis architecture. It had a chapter about forking - the process of how they create a disk backup of the in-memory database contents. The chapter addresses some same system programming topics as the linux-insides book I've been glancing over recently.

As the article points out, the POSIX fork system call creates a duplicate of the calling process but does not copy the memory pages. If the parent or any child process access a memory page, it points to the read-only shared memory until the process changes any values of that page. The calling process then gets a writable copy of the accessed memory pages (copy-on-write).

DALL·E 2 did not know how to draw a go gopher but close enough!

Only single-threaded applications support fork, but I still decided to try it out on a go app. Go runtime does not support fork due to go programs being multithreaded and other reasons I don't even try to understand. There is a ForkExec call in the standard library that first forks but then runs exec, replacing everything, including the runtime.

As a curiosity, a go program can be forked but works only in trivial cases. I'm unsure what happens under the hood, TBH but seems to work as expected. Here is an example gist.

	package main
	import (
	"fmt"
	"syscall"
	)
	func main() {
	a := 10
	// NOTE: this does not actually work in any non-trivial applications.
	id, _, _ := syscall.Syscall(syscall.SYS_FORK, 0, 0, 0)
	if id == 0 {
	fmt.Println("In child:", a, &a)
	a++
	fmt.Println("In child:", a, &a)
	} else {
	fmt.Println("In parent:", a, &a)
	a = a + 10
	fmt.Println("In parent:", a, &a)
	}
	// In parent: 10 0xc0000180b0
	// In parent: 20 0xc0000180b0
	// In child: 10 0xc0000180b0
	// In child: 11 0xc0000180b0
	}

view raw fork.go hosted with ❤ by GitHub