nothing will be simple with reaper + os/exec in the same process

cmd/dkl-system-init/reaper.go

@@ -1,36 +1,41 @@
 package main
 
 import (
+	"fmt"
 	"os"
 	"os/signal"
+	"sync"
 	"syscall"
 
 	"golang.org/x/sys/unix"
 	"novit.nc/direktil/pkg/log"
 )
 
+var reapLock = sync.RWMutex{}
+
 func handleChildren() {
+	sigchld := make(chan os.Signal, 2048)
+	signal.Notify(sigchld, syscall.SIGCHLD)
+
 	// set us as a sub-reaper
 	if err := unix.Prctl(unix.PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0); err != nil {
 		initLog.Taintf(log.Error, "reaper: failed to set myself a child sub-reaper: %v", err)
 	}
 
-	sigchld := make(chan os.Signal, 2048)
-	signal.Notify(sigchld, syscall.SIGCHLD)
-
 	for range sigchld {
 		reapChildren()
 	}
 }
 
 func reapChildren() {
+	reapLock.Lock()
+	defer reapLock.Unlock()
 	for {
 		pid, err := syscall.Wait4(-1, nil, syscall.WNOHANG, nil)
-		if err != nil {
-			if err == unix.ECHILD {
-				break
-			}
+		if err != nil && err != syscall.ECHILD {
 			initLog.Taintf(log.Warning, "reaper: wait4 failed: %v", err)
+			fmt.Printf("reaper: wait4 failed: %v\n", err)
+			break
 		}
 		if pid <= 0 {
 			break

cmd/dkl-system-init/reaper_test.go

@@ -0,0 +1,44 @@
+package main
+
+import (
+	"os"
+	"os/exec"
+	"sync"
+	"testing"
+)
+
+func TestReap(t *testing.T) {
+	truePath, err := exec.LookPath("true")
+	if err != nil {
+		t.Log("true binary not found, ignoring this test.")
+		return
+	}
+
+	go handleChildren()
+
+	count := 1000
+
+	wg := &sync.WaitGroup{}
+	wg.Add(count)
+	for i := 0; i < count; i++ {
+		i := i
+		go func() {
+			cmd := exec.Command(truePath)
+			cmd.Stdout = os.Stdout
+			cmd.Stderr = os.Stderr
+			cmd.Stdin = os.Stdin
+			if err := cmd.Run(); err != nil {
+				t.Errorf("[%d] %v", i, err)
+			}
+			wg.Done()
+		}()
+	}
+
+	wg.Wait()
+
+	cmd := exec.Command("sh", "-c", "ps aux |grep Z")
+	cmd.Stdout = os.Stdout
+	cmd.Run()
+
+	t.Fail()
+}