Merge pull request #1577 from shaloulcy/notify_systemd

feature: add systemd notify functionality

apis/server/server.go

@@ -34,7 +34,7 @@ type Server struct {
 }
 
 // Start setup route table and listen to specified address which currently only supports unix socket and tcp address.
-func (s *Server) Start() (err error) {
+func (s *Server) Start(readyCh chan bool) (err error) {
 	router := initRoute(s)
 	errCh := make(chan error)
 
@@ -50,6 +50,7 @@ func (s *Server) Start() (err error) {
 	if s.Config.TLS.Key != "" && s.Config.TLS.Cert != "" {
 		tlsConfig, err = httputils.GenTLSConfig(s.Config.TLS.Key, s.Config.TLS.Cert, s.Config.TLS.CA)
 		if err != nil {
+			readyCh <- false
 			return err
 		}
 		if s.Config.TLS.VerifyRemote {
@@ -61,6 +62,7 @@ func (s *Server) Start() (err error) {
 	for _, one := range s.Config.Listen {
 		l, err := getListener(one, tlsConfig)
 		if err != nil {
+			readyCh <- false
 			return err
 		}
 		logrus.Infof("start to listen to: %s", one)
@@ -71,6 +73,9 @@ func (s *Server) Start() (err error) {
 		}(l)
 	}
 
+	// the http server has set up, send Ready
+	readyCh <- true
+
 	// not error, will block and run forever.
 	return <-errCh
 }

cri/criservice.go

@@ -14,37 +14,42 @@ import (
 )
 
 // RunCriService start cri service if pouchd is specified with --enable-cri.
-func RunCriService(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr, stopCh chan error) {
+func RunCriService(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr, stopCh chan error, readyCh chan bool) {
 	var err error
 
 	defer func() {
 		stopCh <- err
 		close(stopCh)
 	}()
 	if !daemonconfig.IsCriEnabled {
+		// the CriService has been disabled, so send Ready
+		readyCh <- true
 		return
 	}
 	switch daemonconfig.CriConfig.CriVersion {
 	case "v1alpha1":
-		err = runv1alpha1(daemonconfig, containerMgr, imageMgr)
+		err = runv1alpha1(daemonconfig, containerMgr, imageMgr, readyCh)
 	case "v1alpha2":
-		err = runv1alpha2(daemonconfig, containerMgr, imageMgr)
+		err = runv1alpha2(daemonconfig, containerMgr, imageMgr, readyCh)
 	default:
+		readyCh <- false
 		err = fmt.Errorf("invalid CRI version,failed to start CRI service")
 	}
 	return
 }
 
 // Start CRI service with CRI version: v1alpha1
-func runv1alpha1(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr) error {
+func runv1alpha1(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr, readyCh chan bool) error {
 	logrus.Infof("Start CRI service with CRI version: v1alpha1")
 	criMgr, err := criv1alpha1.NewCriManager(daemonconfig, containerMgr, imageMgr)
 	if err != nil {
+		readyCh <- false
 		return fmt.Errorf("failed to get CriManager with error: %v", err)
 	}
 
 	service, err := servicev1alpha1.NewService(daemonconfig, criMgr)
 	if err != nil {
+		readyCh <- false
 		return fmt.Errorf("failed to start CRI service with error: %v", err)
 	}
 
@@ -59,6 +64,9 @@ func runv1alpha1(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, ima
 		logrus.Infof("CRI Stream server stopped")
 	}()
 
+	// the criservice has set up, send Ready
+	readyCh <- true
+
 	// Check for error
 	for i := 0; i < cap(errChan); i++ {
 		if err := <-errChan; err != nil {
@@ -71,15 +79,17 @@ func runv1alpha1(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, ima
 }
 
 // Start CRI service with CRI version: v1alpha2
-func runv1alpha2(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr) error {
+func runv1alpha2(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, imageMgr mgr.ImageMgr, readyCh chan bool) error {
 	logrus.Infof("Start CRI service with CRI version: v1alpha2")
 	criMgr, err := criv1alpha2.NewCriManager(daemonconfig, containerMgr, imageMgr)
 	if err != nil {
+		readyCh <- false
 		return fmt.Errorf("failed to get CriManager with error: %v", err)
 	}
 
 	service, err := servicev1alpha2.NewService(daemonconfig, criMgr)
 	if err != nil {
+		readyCh <- false
 		return fmt.Errorf("failed to start CRI service with error: %v", err)
 	}
 
@@ -94,6 +104,9 @@ func runv1alpha2(daemonconfig *config.Config, containerMgr mgr.ContainerMgr, ima
 		logrus.Infof("CRI Stream server stopped")
 	}()
 
+	// the criservice has set up, send Ready
+	readyCh <- true
+
 	// Check for error
 	for i := 0; i < cap(errChan); i++ {
 		if err := <-errChan; err != nil {

daemon/daemon.go

@@ -18,6 +18,8 @@ import (
 	"github.com/alibaba/pouch/pkg/meta"
 	"github.com/alibaba/pouch/pkg/system"
 
+	systemddaemon "github.com/coreos/go-systemd/daemon"
+	systemdutil "github.com/coreos/go-systemd/util"
 	"github.com/gorilla/mux"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
@@ -203,16 +205,30 @@ func (d *Daemon) Run() error {
 	// set image proxy
 	ctrd.SetImageProxy(d.config.ImageProxy)
 
+	httpReadyCh := make(chan bool)
+	criReadyCh := make(chan bool)
+
 	httpServerCloseCh := make(chan struct{})
 	go func() {
-		if err := d.server.Start(); err != nil {
+		if err := d.server.Start(httpReadyCh); err != nil {
 			logrus.Errorf("failed to start http server: %v", err)
 		}
 		close(httpServerCloseCh)
 	}()
 
 	criStopCh := make(chan error)
-	go criservice.RunCriService(d.config, d.containerMgr, d.imageMgr, criStopCh)
+	go criservice.RunCriService(d.config, d.containerMgr, d.imageMgr, criStopCh, criReadyCh)
+
+	httpReady := <-httpReadyCh
+	criReady := <-criReadyCh
+
+	if httpReady && criReady {
+		notifySystemd()
+	}
+
+	// close the ready channel
+	close(httpReadyCh)
+	close(criReadyCh)
 
 	err = <-criStopCh
 	if err != nil {
@@ -322,3 +338,18 @@ func (d *Daemon) addSystemLabels() error {
 
 	return nil
 }
+
+func notifySystemd() {
+	if !systemdutil.IsRunningSystemd() {
+		return
+	}
+
+	sent, err := systemddaemon.SdNotify(false, "READY=1")
+	if err != nil {
+		logrus.Errorf("failed to notify systemd for readiness: %v", err)
+	}
+
+	if !sent {
+		logrus.Errorf("forgot to set Type=notify in systemd service file?")
+	}
+}

hack/package/deb/systemd/pouch.service

@@ -4,6 +4,7 @@ Description=pouch
 [Service]
 ExecStart=/usr/bin/pouchd
 ExecReload=/bin/kill -HUP $MAINPID
+Type=notify
 
 # Having non-zero Limit*s causes performance problems due to accounting overhead
 # in the kernel. We recommend using cgroups to do container-local accounting.

hack/package/rpm/service/pouch.service

@@ -4,6 +4,7 @@ Description=pouch
 [Service]
 ExecStart=/usr/local/bin/pouchd
 ExecReload=/bin/kill -HUP $MAINPID
+Type=notify
 
 # Having non-zero Limit*s causes performance problems due to accounting overhead
 # in the kernel. We recommend using cgroups to do container-local accounting.