About fatal error when triggering gc

[lynin]

hello, I found a number of errors, which were seems concentrated in the gc of go runtime.

The following is a summary of the error messages:

1 =>
runtime: gp: gp=0xc00041e820, goid=347, gp->atomicstatus=1
runtime:  g:  g=0x1392220, goid=0,  g->atomicstatus=0
fatal error: bad g->status in ready

2 =>
fatal error: casgstatus: waiting for Gwaiting but is Grunnable

When I solved the above problems, the next fatal error came:

fatal error: unexpected signal during runtime execution
[signal SIGSEGV: segmentation violation code=0x1 addr=0x330 pc=0x1038847]

The error stack is too long, so complete information is not provided here🤔

BTW, The tested go version is 1.19.2 and 1.19.3

Here is my test code:

package itogami_test

import (
	"github.com/alphadose/itogami"
	"runtime/debug"
	"sync"
	"testing"
)

func doing() {
	times := 1000
	size := 100
	task := func(i int) {
		_ = i + 1
	}

	pool := itogami.NewPoolWithFunc(uint64(size), task)

	var wg sync.WaitGroup
	wg.Add(size)

	for i := 0; i < size; i++ {
		go func() {
			for n := 0; n < times; n++ {
				pool.Invoke(n)
			}
			wg.Done()
		}()
	}
	wg.Wait()
}

func TestTriggerGC(t *testing.T) {
	debug.SetMemoryLimit((1 << 20) * 100) // 100m

	for i := 0; i < 100; i++ {
		doing()
	}
}

Hope to get your help, thank you!

[alphadose]

@lynin this library is experimental as it tries to schedule and manage goroutines manually as can be seen in this line https://github.com/alphadose/itogami/blob/master/lib_runtime_linkage.go#L160

Scheduling goroutines manually grants better performance than the golang runtime scheduler but it also doesn't play well with golang runtime because of some niche edge cases here and there leading to exceptions and crashes

The first error you posted occurs when a goroutine is scheduled to be ready at the same time a GC occurs leading to an exception in goroutine state
The second error occurs when a lot of goroutines are in waiting state and we do mcall(gosched_m) (runtime.GoSched()) https://github.com/alphadose/itogami/blob/master/pool.go#L50
As for the 3rd issue I am lacking a bit of context without the full error log hence I cannot comment on that

Ultimately this library is just a POC for a better golang scheduler and I have opened an issue for the same in golang core golang/go#53398