save snapshot of max revision periodically

cmd/backup/main.go

@@ -13,7 +13,7 @@ var (
 )
 
 func init() {
-	flag.StringVar(&cfg.MasterHost, "master", "", "API Server addr, e.g. ' - NOT RECOMMENDED FOR PRODUCTION - http://127.0.0.1:8080'. Omit parameter to run in on-cluster mode and utilize the service account token.")
+	flag.StringVar(&masterHost, "master", "", "API Server addr, e.g. ' - NOT RECOMMENDED FOR PRODUCTION - http://127.0.0.1:8080'. Omit parameter to run in on-cluster mode and utilize the service account token.")
 	flag.StringVar(&clusterName, "etcd-cluster", "", "")
 	// TODO: parse policy
 	flag.Parse()

hack/build_backup.sh

@@ -1,5 +1,9 @@
 #!/usr/bin/env bash
 
+set -o errexit
+set -o nounset
+set -o pipefail
+
 pushd "./cmd/backup"
 go build .
 docker build --tag gcr.io/coreos-k8s-scale-testing/kubeetcdbackup:latest .

pkg/backup/backup.go

@@ -1,6 +1,11 @@
 package backup
 
 import (
+	"fmt"
+	"io"
+	"io/ioutil"
+	"log"
+	"os"
 	"time"
 
 	"golang.org/x/net/context"
@@ -28,6 +33,7 @@ func New(kclient *unversioned.Client, clusterName string, policy Policy) *Backup
 }
 
 func (b *Backup) Run() {
+	lastSnapRev := int64(0)
 	for {
 		// TODO: add interval to backup policy
 		time.Sleep(20 * time.Second)
@@ -41,23 +47,96 @@ func (b *Backup) Run() {
 		if err != nil {
 			panic(err)
 		}
-		for i := range pods.Items {
-			m := etcdutil.Member{Name: pods.Items[i].Name}
-			cfg := clientv3.Config{
-				Endpoints:   []string{m.ClientAddr()},
-				DialTimeout: 5 * time.Second,
-			}
-			etcdcli, err := clientv3.New(cfg)
-			if err != nil {
-				logrus.Errorf("clientv3.New failed: %v", err)
-				continue
-			}
-			resp, err := etcdcli.Get(context.TODO(), "/", clientv3.WithSerializable())
-			if err != nil {
-				logrus.Errorf("etcdcli.Get failed: %v", err)
-				continue
-			}
-			logrus.Infof("member: %s, revision: %d", m.Name, resp.Header.Revision)
+		if len(pods.Items) == 0 {
+			logrus.Warning("no running pods found.")
+			continue
 		}
+		member, rev, err := getMemberWithMaxRev(pods)
+		if err != nil {
+			logrus.Error(err)
+		}
+		if member == nil {
+			logrus.Warning("no reachable member")
+			continue
+		}
+		if rev == lastSnapRev {
+			logrus.Info("skipped creating new backup: no change since last time")
+			continue
+		}
+
+		log.Printf("saving backup for cluster (%s)", b.clusterName)
+		if err := writeSnap(member, rev); err != nil {
+			logrus.Errorf("write snapshot failed: %v", err)
+			continue
+		}
+		lastSnapRev = rev
 	}
 }
+
+func writeSnap(m *etcdutil.Member, rev int64) error {
+	cfg := clientv3.Config{
+		Endpoints:   []string{m.ClientAddr()},
+		DialTimeout: 5 * time.Second,
+	}
+	etcdcli, err := clientv3.New(cfg)
+	if err != nil {
+		return fmt.Errorf("failed to create etcd client (%v)", err)
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+
+	rc, err := etcdcli.Maintenance.Snapshot(ctx)
+	if err != nil {
+		return fmt.Errorf("failed to receive snapshot (%v)", err)
+	}
+	defer rc.Close()
+
+	// TODO: custom backup dir
+	tmpfile, err := ioutil.TempFile("", "snapshot")
+	n, err := io.Copy(tmpfile, rc)
+	if err != nil {
+		tmpfile.Close()
+		os.Remove(tmpfile.Name())
+		return fmt.Errorf("failed to save snapshot: %v", err)
+	}
+	cancel()
+	tmpfile.Close()
+	nextSnapshotName := makeFilename(rev)
+	err = os.Rename(tmpfile.Name(), nextSnapshotName)
+	if err != nil {
+		os.Remove(tmpfile.Name())
+		return fmt.Errorf("rename snapshot from %s to %s failed: %v", tmpfile.Name(), nextSnapshotName, err)
+	}
+	log.Printf("saved snapshot %s (size: %d) successfully", nextSnapshotName, n)
+	return nil
+}
+
+func makeFilename(rev int64) string {
+	return fmt.Sprintf("%016x.backup", rev)
+}
+
+func getMemberWithMaxRev(pods *api.PodList) (*etcdutil.Member, int64, error) {
+	var member *etcdutil.Member
+	maxRev := int64(0)
+	for i := range pods.Items {
+		m := &etcdutil.Member{Name: pods.Items[i].Name}
+		cfg := clientv3.Config{
+			Endpoints:   []string{m.ClientAddr()},
+			DialTimeout: 5 * time.Second,
+		}
+		etcdcli, err := clientv3.New(cfg)
+		if err != nil {
+			return nil, 0, fmt.Errorf("failed to create etcd client (%v)", err)
+		}
+		resp, err := etcdcli.Get(context.TODO(), "/", clientv3.WithSerializable())
+		if err != nil {
+			return nil, 0, fmt.Errorf("etcdcli.Get failed: %v", err)
+		}
+		logrus.Infof("member: %s, revision: %d", m.Name, resp.Header.Revision)
+		if resp.Header.Revision > maxRev {
+			maxRev = resp.Header.Revision
+			member = m
+		}
+	}
+	return member, maxRev, nil
+}

pkg/cluster/cluster.go

@@ -2,9 +2,6 @@ package cluster
 
 import (
 	"fmt"
-	"io"
-	"io/ioutil"
-	"os"
 	"strconv"
 	"strings"
 	"time"
@@ -205,49 +202,6 @@ func (c *Cluster) removePodAndService(name string) error {
 	return nil
 }
 
-func (c *Cluster) backup() error {
-	clientAddr := "todo"
-	nextSnapshotName := "todo"
-
-	cfg := clientv3.Config{
-		Endpoints: []string{clientAddr},
-	}
-	etcdcli, err := clientv3.New(cfg)
-	if err != nil {
-		panic(err)
-	}
-
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
-
-	log.Println("saving snapshot from cluster", c.name)
-
-	rc, err := etcdcli.Maintenance.Snapshot(ctx)
-	cancel()
-	if err != nil {
-		return err
-	}
-
-	tmpfile, err := ioutil.TempFile(c.backupDir, "snapshot")
-	n, err := io.Copy(tmpfile, rc)
-	if err != nil {
-		tmpfile.Close()
-		os.Remove(tmpfile.Name())
-		log.Printf("saving snapshot from cluster %s error: %v", c.name, err)
-		return err
-	}
-
-	err = os.Rename(tmpfile.Name(), nextSnapshotName)
-	if err != nil {
-		os.Remove(tmpfile.Name())
-		log.Printf("renaming snapshot from cluster %s error: %v", c.name, err)
-		return err
-	}
-
-	log.Printf("saved snapshot %v (size: %d) from cluster %s", n, nextSnapshotName, c.name)
-
-	return nil
-}
-
 func (c *Cluster) monitorPods() {
 	opts := k8sapi.ListOptions{
 		LabelSelector: labels.SelectorFromSet(map[string]string{