Merge pull request #6744 from Junnplus/restart-policy

Add restart policy for enhanced restart manager

runtime/restart/monitor/change.go

@@ -20,10 +20,12 @@ import (
 	"context"
 	"fmt"
 	"net/url"
+	"strconv"
 	"syscall"
 
 	"github.com/containerd/containerd"
 	"github.com/containerd/containerd/cio"
+	"github.com/containerd/containerd/runtime/restart"
 	"github.com/sirupsen/logrus"
 )
 
@@ -38,6 +40,7 @@ func (s *stopChange) apply(ctx context.Context, client *containerd.Client) error
 type startChange struct {
 	container containerd.Container
 	logURI    string
+	count     int
 
 	// Deprecated(in release 1.5): but recognized now, prefer to use logURI
 	logPath string
@@ -61,6 +64,15 @@ func (s *startChange) apply(ctx context.Context, client *containerd.Client) erro
 			s.logPath, s.logURI)
 	}
 
+	if s.count > 0 {
+		labels := map[string]string{
+			restart.CountLabel: strconv.Itoa(s.count),
+		}
+		opt := containerd.WithAdditionalContainerLabels(labels)
+		if err := s.container.Update(ctx, containerd.UpdateContainerOpts(opt)); err != nil {
+			return err
+		}
+	}
 	killTask(ctx, s.container)
 	task, err := s.container.NewTask(ctx, log)
 	if err != nil {

runtime/restart/monitor/monitor.go

@@ -19,6 +19,7 @@ package monitor
 import (
 	"context"
 	"fmt"
+	"strconv"
 	"sync"
 	"time"
 
@@ -72,6 +73,7 @@ func init() {
 			},
 		},
 		InitFn: func(ic *plugin.InitContext) (interface{}, error) {
+			ic.Meta.Capabilities = []string{"no", "always", "on-failure", "unless-stopped"}
 			opts, err := getServicesOpts(ic)
 			if err != nil {
 				return nil, err
@@ -217,15 +219,29 @@ func (m *monitor) monitor(ctx context.Context) ([]change, error) {
 			return nil, err
 		}
 		desiredStatus := containerd.ProcessStatus(labels[restart.StatusLabel])
-		if m.isSameStatus(ctx, desiredStatus, c) {
+		task, err := c.Task(ctx, nil)
+		if err != nil && desiredStatus == containerd.Stopped {
 			continue
 		}
+		status, err := task.Status(ctx)
+		if err != nil && desiredStatus == containerd.Stopped {
+			continue
+		}
+		if desiredStatus == status.Status {
+			continue
+		}
+
 		switch desiredStatus {
 		case containerd.Running:
+			if !restart.Reconcile(status, labels) {
+				continue
+			}
+			restartCount, _ := strconv.Atoi(labels[restart.CountLabel])
 			changes = append(changes, &startChange{
 				container: c,
 				logPath:   labels[restart.LogPathLabel],
 				logURI:    labels[restart.LogURILabel],
+				count:     restartCount + 1,
 			})
 		case containerd.Stopped:
 			changes = append(changes, &stopChange{
@@ -235,15 +251,3 @@ func (m *monitor) monitor(ctx context.Context) ([]change, error) {
 	}
 	return changes, nil
 }
-
-func (m *monitor) isSameStatus(ctx context.Context, desired containerd.ProcessStatus, container containerd.Container) bool {
-	task, err := container.Task(ctx, nil)
-	if err != nil {
-		return desired == containerd.Stopped
-	}
-	state, err := task.Status(ctx)
-	if err != nil {
-		return desired == containerd.Stopped
-	}
-	return desired == state.Status
-}