Merge pull request #426 from gitido/pressure_level

libcontainer: Add support for memcg pressure notifications

Author: Mrunal Patel

libcontainer/container_linux.go

@@ -104,6 +104,12 @@ type Container interface {
 	// errors:
 	// Systemerror - System error.
 	NotifyOOM() (<-chan struct{}, error)
+
+	// NotifyMemoryPressure returns a read-only channel signaling when the container reaches a given pressure level
+	//
+	// errors:
+	// Systemerror - System error.
+	NotifyMemoryPressure(level PressureLevel) (<-chan struct{}, error)
 }
 
 // ID returns the container's unique ID
@@ -357,6 +363,10 @@ func (c *linuxContainer) NotifyOOM() (<-chan struct{}, error) {
 	return notifyOnOOM(c.cgroupManager.GetPaths())
 }
 
+func (c *linuxContainer) NotifyMemoryPressure(level PressureLevel) (<-chan struct{}, error) {
+	return notifyMemoryPressure(c.cgroupManager.GetPaths(), level)
+}
+
 // XXX debug support, remove when debugging done.
 func addArgsFromEnv(evar string, args *[]string) {
 	if e := os.Getenv(evar); e != "" {

libcontainer/notify_linux.go

@@ -12,39 +12,40 @@ import (
 
 const oomCgroupName = "memory"
 
-// notifyOnOOM returns channel on which you can expect event about OOM,
-// if process died without OOM this channel will be closed.
-// s is current *libcontainer.State for container.
-func notifyOnOOM(paths map[string]string) (<-chan struct{}, error) {
-	dir := paths[oomCgroupName]
-	if dir == "" {
-		return nil, fmt.Errorf("There is no path for %q in state", oomCgroupName)
-	}
-	oomControl, err := os.Open(filepath.Join(dir, "memory.oom_control"))
+type PressureLevel uint
+
+const (
+	LowPressure PressureLevel = iota
+	MediumPressure
+	CriticalPressure
+)
+
+func registerMemoryEvent(cgDir string, evName string, arg string) (<-chan struct{}, error) {
+	evFile, err := os.Open(filepath.Join(cgDir, evName))
 	if err != nil {
 		return nil, err
 	}
 	fd, _, syserr := syscall.RawSyscall(syscall.SYS_EVENTFD2, 0, syscall.FD_CLOEXEC, 0)
 	if syserr != 0 {
-		oomControl.Close()
+		evFile.Close()
 		return nil, syserr
 	}
 
 	eventfd := os.NewFile(fd, "eventfd")
 
-	eventControlPath := filepath.Join(dir, "cgroup.event_control")
-	data := fmt.Sprintf("%d %d", eventfd.Fd(), oomControl.Fd())
+	eventControlPath := filepath.Join(cgDir, "cgroup.event_control")
+	data := fmt.Sprintf("%d %d %s", eventfd.Fd(), evFile.Fd(), arg)
 	if err := ioutil.WriteFile(eventControlPath, []byte(data), 0700); err != nil {
 		eventfd.Close()
-		oomControl.Close()
+		evFile.Close()
 		return nil, err
 	}
 	ch := make(chan struct{})
 	go func() {
 		defer func() {
 			close(ch)
 			eventfd.Close()
-			oomControl.Close()
+			evFile.Close()
 		}()
 		buf := make([]byte, 8)
 		for {
@@ -61,3 +62,28 @@ func notifyOnOOM(paths map[string]string) (<-chan struct{}, error) {
 	}()
 	return ch, nil
 }
+
+// notifyOnOOM returns channel on which you can expect event about OOM,
+// if process died without OOM this channel will be closed.
+func notifyOnOOM(paths map[string]string) (<-chan struct{}, error) {
+	dir := paths[oomCgroupName]
+	if dir == "" {
+		return nil, fmt.Errorf("path %q missing", oomCgroupName)
+	}
+
+	return registerMemoryEvent(dir, "memory.oom_control", "")
+}
+
+func notifyMemoryPressure(paths map[string]string, level PressureLevel) (<-chan struct{}, error) {
+	dir := paths[oomCgroupName]
+	if dir == "" {
+		return nil, fmt.Errorf("path %q missing", oomCgroupName)
+	}
+
+	if level > CriticalPressure {
+		return nil, fmt.Errorf("invalid pressure level %d", level)
+	}
+
+	levelStr := []string{"low", "medium", "critical"}[level]
+	return registerMemoryEvent(dir, "memory.pressure_level", levelStr)
+}

libcontainer/notify_linux_test.go

@@ -13,24 +13,25 @@ import (
 	"time"
 )
 
-func TestNotifyOnOOM(t *testing.T) {
-	memoryPath, err := ioutil.TempDir("", "testnotifyoom-")
+type notifyFunc func(paths map[string]string) (<-chan struct{}, error)
+
+func testMemoryNotification(t *testing.T, evName string, notify notifyFunc, targ string) {
+	memoryPath, err := ioutil.TempDir("", "testmemnotification-"+evName)
 	if err != nil {
 		t.Fatal(err)
 	}
-	oomPath := filepath.Join(memoryPath, "memory.oom_control")
+	evFile := filepath.Join(memoryPath, evName)
 	eventPath := filepath.Join(memoryPath, "cgroup.event_control")
-	if err := ioutil.WriteFile(oomPath, []byte{}, 0700); err != nil {
+	if err := ioutil.WriteFile(evFile, []byte{}, 0700); err != nil {
 		t.Fatal(err)
 	}
 	if err := ioutil.WriteFile(eventPath, []byte{}, 0700); err != nil {
 		t.Fatal(err)
 	}
-	var eventFd, oomControlFd int
 	paths := map[string]string{
 		"memory": memoryPath,
 	}
-	ooms, err := notifyOnOOM(paths)
+	ch, err := notify(paths)
 	if err != nil {
 		t.Fatal("expected no error, got:", err)
 	}
@@ -40,7 +41,14 @@ func TestNotifyOnOOM(t *testing.T) {
 		t.Fatal("couldn't read event control file:", err)
 	}
 
-	if _, err := fmt.Sscanf(string(data), "%d %d", &eventFd, &oomControlFd); err != nil {
+	var eventFd, evFd int
+	var arg string
+	if targ != "" {
+		_, err = fmt.Sscanf(string(data), "%d %d %s", &eventFd, &evFd, &arg)
+	} else {
+		_, err = fmt.Sscanf(string(data), "%d %d", &eventFd, &evFd)
+	}
+	if err != nil || arg != targ {
 		t.Fatalf("invalid control data %q: %s", data, err)
 	}
 
@@ -63,9 +71,9 @@ func TestNotifyOnOOM(t *testing.T) {
 	}
 
 	select {
-	case <-ooms:
+	case <-ch:
 	case <-time.After(100 * time.Millisecond):
-		t.Fatal("no notification on oom channel after 100ms")
+		t.Fatal("no notification on channel after 100ms")
 	}
 
 	// simulate what happens when a cgroup is destroyed by cleaning up and then
@@ -79,18 +87,42 @@ func TestNotifyOnOOM(t *testing.T) {
 
 	// give things a moment to shut down
 	select {
-	case _, ok := <-ooms:
+	case _, ok := <-ch:
 		if ok {
-			t.Fatal("expected no oom to be triggered")
+			t.Fatal("expected no notification to be triggered")
 		}
 	case <-time.After(100 * time.Millisecond):
 	}
 
-	if _, _, err := syscall.Syscall(syscall.SYS_FCNTL, uintptr(oomControlFd), syscall.F_GETFD, 0); err != syscall.EBADF {
-		t.Error("expected oom control to be closed")
+	if _, _, err := syscall.Syscall(syscall.SYS_FCNTL, uintptr(evFd), syscall.F_GETFD, 0); err != syscall.EBADF {
+		t.Error("expected event control to be closed")
 	}
 
 	if _, _, err := syscall.Syscall(syscall.SYS_FCNTL, uintptr(eventFd), syscall.F_GETFD, 0); err != syscall.EBADF {
 		t.Error("expected event fd to be closed")
 	}
 }
+
+func TestNotifyOnOOM(t *testing.T) {
+	f := func(paths map[string]string) (<-chan struct{}, error) {
+		return notifyOnOOM(paths)
+	}
+
+	testMemoryNotification(t, "memory.oom_control", f, "")
+}
+
+func TestNotifyMemoryPressure(t *testing.T) {
+	tests := map[PressureLevel]string{
+		LowPressure:      "low",
+		MediumPressure:   "medium",
+		CriticalPressure: "critical",
+	}
+
+	for level, arg := range tests {
+		f := func(paths map[string]string) (<-chan struct{}, error) {
+			return notifyMemoryPressure(paths, level)
+		}
+
+		testMemoryNotification(t, "memory.pressure_level", f, arg)
+	}
+}