fix daemon restore after kill -HUP

daemon/pod/container.go

@@ -902,6 +902,39 @@ func (c *Container) addToSandbox() error {
 	return nil
 }
 
+func (c *Container) associateToSandbox() error {
+	c.Log(DEBUG, "try to associate container %s to sandbox", c.Id())
+	alive, err := c.p.sandbox.AssociateContainer(c.Id())
+	if err != nil {
+		return err
+	}
+	// FIXME missing container status history here.
+	if alive {
+		c.status.State = S_CONTAINER_RUNNING
+		c.status.StartedAt = time.Now()
+	} else {
+		c.status.State = S_CONTAINER_CREATED
+		c.status.CreatedAt = time.Now()
+	}
+
+	c.streams.NewInputPipes()
+	tty := &hypervisor.TtyIO{
+		Stdin:  c.streams.Stdin(),
+		Stdout: c.streams.Stdout(),
+		Stderr: c.streams.Stderr(),
+	}
+	err = c.p.sandbox.Attach(tty, c.Id())
+	if err != nil {
+		return err
+	}
+
+	go c.waitFinish(-1)
+
+	c.startLogging()
+
+	return nil
+}
+
 func (c *Container) initLogger() {
 	if c.logger.Driver != nil {
 		return

daemon/pod/decommission.go

@@ -313,6 +313,10 @@ func (p *XPod) RemoveContainer(id string) error {
 	if err = p.saveLayout(); err != nil {
 		return err
 	}
+	// save sandbox change
+	if err = p.saveSandbox(); err != nil {
+		return err
+	}
 
 	return nil
 }
@@ -566,6 +570,11 @@ func (p *XPod) cleanup() {
 		p.Log(ERROR, "pod stopping failed, failed to decommit the resources: %v", err)
 	}
 
+	err = p.removeSandboxFromDB()
+	if err != nil {
+		p.Log(ERROR, "pod stopping failed, failed to remove sandbox persist data: %v", err)
+	}
+
 	p.Log(DEBUG, "tag pod as stopped")
 	p.statusLock.Lock()
 	if p.status != S_POD_NONE {

daemon/pod/migration.go

@@ -85,7 +85,7 @@ func MigrateLagecyPersistentData(db *daemondb.DaemonDB, podFactory func() *PodFa
 				Id:          vmID,
 				PersistInfo: vmData,
 			}
-			err = saveMessage(db, fmt.Sprintf(SB_KEY_FMT, podID), &sandboxInfo, nil, "sandbox info")
+			err = saveMessage(db, fmt.Sprintf(SB_KEY_FMT, podSpec.Id), &sandboxInfo, nil, "sandbox info")
 			if err != nil {
 				return err
 			}

daemon/pod/persist.go

@@ -2,6 +2,7 @@ package pod
 
 import (
 	"fmt"
+	"strings"
 
 	"github.com/golang/protobuf/proto"
 	"github.com/hyperhq/hypercontainer-utils/hlog"
@@ -84,40 +85,58 @@ func LoadXPod(factory *PodFactory, layout *types.PersistPodLayout) (*XPod, error
 	if err != nil {
 		return nil, err
 	}
+	// prevent using incomplete pod
+	defer func() {
+		if err != nil {
+			p.releaseNames(spec.Containers)
+		}
+	}()
 
 	for _, ix := range layout.Interfaces {
-		if err := p.loadInterface(ix); err != nil {
+		if err = p.loadInterface(ix); err != nil {
 			return nil, err
 		}
 	}
 
 	for _, vid := range layout.Volumes {
-		if err := p.loadVolume(vid); err != nil {
+		if err = p.loadVolume(vid); err != nil {
 			return nil, err
 		}
 	}
 
 	for _, cid := range layout.Containers {
-		if err := p.loadContainer(cid); err != nil {
+		if err = p.loadContainer(cid); err != nil {
 			return nil, err
 		}
 	}
 
 	err = p.loadSandbox()
 	if err != nil {
-		//remove vm from daemonDB
-		return nil, err
+		if !strings.Contains(err.Error(), "leveldb: not found") {
+			p.removeSandboxFromDB()
+			return nil, err
+		}
+		p.status = S_POD_STOPPED
+	}
+
+	// if sandbox is running, set all volume INSERTED
+	if p.status == S_POD_RUNNING {
+		for _, v := range p.volumes {
+			v.status = S_VOLUME_INSERTED
+		}
 	}
 
 	err = p.loadPodMeta()
 	if err != nil {
 		return nil, err
 	}
 
-	//resume logging
+	//associate containers
 	if p.status == S_POD_RUNNING {
 		for _, c := range p.containers {
-			c.startLogging()
+			if err = c.associateToSandbox(); err != nil {
+				return nil, err
+			}
 		}
 	}
 
@@ -377,11 +396,28 @@ func (inf *Interface) removeFromDB() error {
 }
 
 func (p *XPod) saveSandbox() error {
-	var sb types.SandboxPersistInfo
-	if p.sandbox != nil {
-		// TODO: dump sandbox info from runv VM
+	var (
+		sb  types.SandboxPersistInfo
+		err error
+	)
+	stop_status := map[PodState]bool{
+		S_POD_NONE:     true,
+		S_POD_STOPPED:  true,
+		S_POD_STOPPING: true,
+		S_POD_ERROR:    true,
+	}
+	p.statusLock.RLock()
+	defer p.statusLock.RUnlock()
+	if !stop_status[p.status] {
+		sb.Id = p.sandbox.Id
+		sb.PersistInfo, err = p.sandbox.Dump()
+		if err != nil {
+			hlog.HLog(ERROR, p, 2, "failed to dump sandbox %s: %v", sb.Id, err)
+			return err
+		}
+		return saveMessage(p.factory.db, fmt.Sprintf(SB_KEY_FMT, p.Id()), &sb, p, "sandbox info")
 	}
-	return saveMessage(p.factory.db, fmt.Sprintf(SB_KEY_FMT, p.Id()), &sb, p, "sandbox info")
+	return nil
 }
 
 func (p *XPod) loadSandbox() error {