Fix 2 bugs in pid1

1) Was calling Wait() on the child process, but also calling Wait4,
which would race, causing an occasional error or panic.

2) In testing (1), I observed occasional hangs.  Tracing it down to a
SIGWINCH, which masked a SIGCHLD, causing it to hang.  Both seem fixed.

Added a manual test script.

cmd/git-sync/main.go

@@ -319,7 +319,7 @@ func main() {
 	}
 
 	// From here on, output goes through logging.
-	log.V(0).Info("starting up", "args", os.Args)
+	log.V(0).Info("starting up", "pid", os.Getpid(), "args", os.Args)
 
 	// Startup webhooks goroutine
 	var webhook *Webhook

pkg/pid1/pid1.go

@@ -25,27 +25,30 @@ func ReRun() error {
 	if err := cmd.Start(); err != nil {
 		return err
 	}
-	go runInit(cmd.Process.Pid)
-	return cmd.Wait()
+	runInit(cmd.Process.Pid)
+	return nil
 }
 
-// runInit runs a bare-bones init process.  This will never return.  In case of
-// truly unknown errors it will panic.
-func runInit(pid int) {
+// runInit runs a bare-bones init process.  This will return when firstborn
+// exits.  In case of truly unknown errors it will panic.
+func runInit(firstborn int) {
 	sigs := make(chan os.Signal, 8)
 	signal.Notify(sigs)
 	for sig := range sigs {
-		if sig == syscall.SIGCHLD {
-			sigchld()
-		} else {
+		if sig != syscall.SIGCHLD {
 			// Pass it on to the real process.
-			syscall.Kill(pid, sig.(syscall.Signal))
+			syscall.Kill(firstborn, sig.(syscall.Signal))
+		}
+		// Always try to reap a child - empirically, sometimes this gets missed.
+		if sigchld(firstborn) {
+			return
 		}
 	}
 }
 
-// sigchld handles a SIGCHLD.
-func sigchld() {
+// sigchld handles a SIGCHLD.  This will return true when firstborn exits.  In
+// case of truly unknown errors it will panic.
+func sigchld(firstborn int) bool {
 	// Loop to handle multiple child processes.
 	for {
 		var status syscall.WaitStatus
@@ -53,10 +56,15 @@ func sigchld() {
 		if err != nil {
 			panic(fmt.Sprintf("failed to wait4(): %v\n", err))
 		}
+
+		if pid == firstborn {
+			return true
+		}
 		if pid <= 0 {
 			// No more children to reap.
 			break
 		}
 		// Must have found one, see if there are more.
 	}
+	return false
 }

pkg/pid1/test/fast-exit/.gitignore

@@ -0,0 +1 @@
+fast-exit

pkg/pid1/test/fast-exit/Dockerfile

@@ -0,0 +1,4 @@
+FROM debian
+RUN apt-get update && apt-get install -y bash procps psmisc psutils
+COPY fast-exit /fast-exit
+ENTRYPOINT ["/fast-exit"]

pkg/pid1/test/fast-exit/main.go

@@ -0,0 +1,28 @@
+// A do-nothing app to test pid1.ReRun().
+package main
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+
+	"k8s.io/git-sync/pkg/pid1"
+)
+
+func main() {
+	// In case we come up as pid 1, act as init.
+	if os.Getpid() == 1 {
+		fmt.Printf("detected pid 1, running as init\n")
+		err := pid1.ReRun()
+		if err == nil {
+			os.Exit(0)
+		}
+		if exerr, ok := err.(*exec.ExitError); ok {
+			os.Exit(exerr.ExitCode())
+		}
+		fmt.Printf("unhandled pid1 error: %v\n", err)
+		os.Exit(127)
+	}
+	fmt.Printf("main app\n")
+	os.Exit(42)
+}

pkg/pid1/test/fast-exit/test.sh

@@ -0,0 +1,10 @@
+#!/bin/sh
+
+go build
+docker build -t example.com/fast-exit .
+
+# In the past we have observed hangs and missed signals.  This *should* run
+# forever.
+while true; do
+    docker run -ti --rm example.com/fast-exit
+done