Merge pull request #1445 from mlaventure/fix-exec-orphans

Fix exec orphans

.travis.yml

@@ -5,7 +5,6 @@ language: go
 
 go:
   - 1.8.x
-  - tip
 
 go_import_path: github.com/containerd/containerd
 
@@ -17,7 +16,7 @@ addons:
       - curl
 
 env:
-  - SECCOMP_VERSION=2.3.1 RUNC_COMMIT=9a01140955fba11a1c2927a9b273e9c837e3e30a
+  - SECCOMP_VERSION=2.3.1 RUNC_COMMIT=d40db12e72a40109dfcf28539f5ee0930d2f0277
 
 
 install:

Dockerfile

@@ -59,7 +59,7 @@ RUN set -x \
 	&& rm -rf "$SECCOMP_PATH"
 
 # Install runc
-ENV RUNC_COMMIT 992a5be178a62e026f4069f443c6164912adbf09
+ENV RUNC_COMMIT d40db12e72a40109dfcf28539f5ee0930d2f0277
 RUN set -x \
 	&& export GOPATH="$(mktemp -d)" \
     && git clone git://github.com/opencontainers/runc.git "$GOPATH/src/github.com/opencontainers/runc" \

containerd-shim/main.go

@@ -7,10 +7,11 @@ import (
 	"os/signal"
 	"path/filepath"
 	"runtime"
+	"sync"
 	"syscall"
 
-	"github.com/containerd/containerd/osutils"
 	"github.com/containerd/console"
+	"github.com/containerd/containerd/osutils"
 )
 
 func writeMessage(f *os.File, level string, err error) {
@@ -111,7 +112,10 @@ func start(log *os.File) error {
 	if runtime.GOOS == "solaris" {
 		return nil
 	}
-	var exitShim bool
+	var (
+		exitShim  bool
+		cleanOnce sync.Once
+	)
 	for {
 		select {
 		case s := <-signals:
@@ -128,15 +132,19 @@ func start(log *os.File) error {
 			}
 			// runtime has exited so the shim can also exit
 			if exitShim {
-				// kill all processes in the container incase it was not running in
-				// its own PID namespace
-				p.killAll()
-				// wait for all the processes and IO to finish
-				p.Wait()
-				// delete the container from the runtime
-				p.delete()
-				// the close of the exit fifo will happen when the shim exits
-				return nil
+				cleanOnce.Do(func() {
+					go func() {
+						// kill all processes in the container incase it was not running in
+						// its own PID namespace
+						p.killAll()
+						// wait for all the processes and IO to finish
+						p.Wait()
+						// delete the container from the runtime
+						p.delete()
+						// the close of the exit fifo will happen when the shim exits
+						os.Exit(0)
+					}()
+				})
 			}
 		case msg := <-msgC:
 			switch msg.Type {

integration-test/exec_test.go

@@ -458,3 +458,56 @@ func (cs *ContainerdSuite) TestExecKillShimNoPanic(t *check.C) {
 	t.Assert(err, checker.Equals, nil)
 	t.Assert(len(containers), checker.Equals, 0)
 }
+
+func (cs *ContainerdSuite) TestBusyboxTopExecShDefunct(t *check.C) {
+	bundleName := "busybox-top"
+	if err := CreateBusyboxBundle(bundleName, []string{"top"}); err != nil {
+		t.Fatal(err)
+	}
+
+	var (
+		err   error
+		initp *ContainerProcess
+	)
+
+	containerID := "top"
+	initp, err = cs.StartContainer(containerID, bundleName)
+	t.Assert(err, checker.Equals, nil)
+
+	execID := "sh"
+	p, err := cs.AddProcessToContainer(initp, execID, "/", []string{"PATH=/bin"}, []string{"sh"}, 0, 0)
+	t.Assert(err, checker.Equals, nil)
+
+	for _, evt := range []types.Event{
+		{
+			Type:   "start-container",
+			Id:     containerID,
+			Status: 0,
+			Pid:    "",
+		},
+		{
+			Type:   "start-process",
+			Id:     containerID,
+			Status: 0,
+			Pid:    execID,
+		},
+	} {
+		ch := initp.GetEventsChannel()
+		e := <-ch
+		evt.Timestamp = e.Timestamp
+		t.Assert(*e, checker.Equals, evt)
+	}
+
+	p.io.stdinf.Write([]byte("sleep 1000&\n"))
+	p.io.stdinf.Write([]byte("sleep 1&\n"))
+	p.io.stdinf.Write([]byte("exit\n"))
+	p.CloseStdin()
+
+	<-time.After(2 * time.Second)
+
+	b, err := exec.Command("ps", "afx").CombinedOutput()
+	t.Assert(err, checker.Equals, nil)
+	t.Assert(string(b), checker.Not(checker.Contains), "<defunct>")
+
+	cs.KillContainer(containerID)
+}