runtime/debug: FreeOSMemory() not working on Android 9

[YouROK]

What version of Go are you using (go version)?

GO 1.12 - 1.14

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

On android 9, on other all works

What did you do?

func main() {
	bigArr := make([]byte, 250*1024*1024)
	for i := 0; i < len(bigArr); i++ {
		bigArr[i] = byte(i)
	}
	bigArr = nil
	runtime.GC()
	debug.FreeOSMemory()
	for true {
		//Some thing works in program
		time.Sleep(time.Second)
	}
}

What did you expect to see?

I expected to see the memory in the system, but it remains in the program. If you allocate memory several times, then the program is killed by the system, but should clear it

In version 1.11.X all works in android 9, memory returning to os, but not execute on android 10. On android 5,6,7,8 all works of new golang version

Compile cmd: GOARM=7 GOOS=linux GOARCH=arm go build -o test main

[odeke-em]

Thank you for the report @YouROK and welcome to the Go project!

/cc @mknyszek @aclements @eliasnaur @cherrymui

[mknyszek]

Considering the combination of Android version and Go version here (Android 10+, Go 1.12+), I suspect this is another case of MADV_FREE not updating the kernel's per-process RSS numbers. Android 10's minimum Linux kernel version is 4.9 whereas Android 9's minimum is 4.4. MADV_FREE was introduced in Linux 4.5.

Sorry for the very late reply on this. Could you try running your code with GODEBUG=madvdontneed=1? I'm not sure how difficult that is to do on Android.

[elgatito]

@mknyszek It is great we can use GODEBUG flags from environment variables. But it would be alto great to be able to hardcode that into the code, or define, depending on the build parameters, which is needed, if you cannot control how the binary is ran by user.

[networkimprov]

Filed #40870

[aclements]

It's really disappointing if Android kills applications for memory use without first trying to clean up their freed pages... If that is the case, then I would consider MADV_FREE quite broken on Android and we shouldn't use it.

@YouROK, your example code is likely working as expected, but you mention "If you allocate memory several times, then the program is killed by the system". That sounds like the real issue here. Could you post code that reproduces that?

[aclements]

Ping @YouROK . Could you try the GODEBUG=madvdontneed=1 environment variable? And can you post the code that demonstrates "allocating memory several times"?

[gopherbot]

Change https://golang.org/cl/267100 mentions this issue: runtime: default to MADV_DONTNEED on Linux

[gopherbot]

Timed out in state WaitingForInfo. Closing.

(I am just a bot, though. Please speak up if this is a mistake or you have the requested information.)