etcdserver: add cluster id check for hashKVHandler

Signed-off-by: caojiamingalan <alan.c.19971111@gmail.com>
Signed-off-by: Marek Siarkowicz <siarkowicz@google.com>

tests/e2e/cluster_test.go

@@ -181,6 +181,17 @@ type etcdProcessClusterConfig struct {
 // newEtcdProcessCluster launches a new cluster from etcd processes, returning
 // a new etcdProcessCluster once all nodes are ready to accept client requests.
 func newEtcdProcessCluster(t testing.TB, cfg *etcdProcessClusterConfig) (*etcdProcessCluster, error) {
+	epc, err := initEtcdProcessCluster(t, cfg)
+	if err != nil {
+		return nil, err
+	}
+
+	return startEtcdProcessCluster(epc, cfg)
+}
+
+// initEtcdProcessCluster initializes a new cluster based on the given config.
+// It doesn't start the cluster.
+func initEtcdProcessCluster(t testing.TB, cfg *etcdProcessClusterConfig) (*etcdProcessCluster, error) {
 	skipInShortMode(t)
 
 	etcdCfgs := cfg.etcdServerProcessConfigs(t)
@@ -199,7 +210,11 @@ func newEtcdProcessCluster(t testing.TB, cfg *etcdProcessClusterConfig) (*etcdPr
 		}
 		epc.procs[i] = proc
 	}
+	return epc, nil
+}
 
+// startEtcdProcessCluster launches a new cluster from etcd processes.
+func startEtcdProcessCluster(epc *etcdProcessCluster, cfg *etcdProcessClusterConfig) (*etcdProcessCluster, error) {
 	if cfg.rollingStart {
 		if err := epc.RollingStart(); err != nil {
 			return nil, fmt.Errorf("Cannot rolling-start: %v", err)

tests/e2e/corrupt_test.go

@@ -96,6 +96,91 @@ func corruptTest(cx ctlCtx) {
 	waitReadyExpectProc(proc, []string{fmt.Sprintf("etcdmain: %016x found data inconsistency with peers", id0)})
 }
 
+func TestInPlaceRecovery(t *testing.T) {
+	basePort := 20000
+	BeforeTest(t)
+
+	// Initialize the cluster.
+	epcOld, err := newEtcdProcessCluster(t,
+		&etcdProcessClusterConfig{
+			clusterSize:      3,
+			initialToken:     "old",
+			keepDataDir:      false,
+			CorruptCheckTime: time.Second,
+			basePort:         basePort,
+		},
+	)
+	if err != nil {
+		t.Fatalf("could not start etcd process cluster (%v)", err)
+	}
+	t.Cleanup(func() {
+		if errC := epcOld.Close(); errC != nil {
+			t.Fatalf("error closing etcd processes (%v)", errC)
+		}
+	})
+	t.Log("old cluster started.")
+
+	//Put some data into the old cluster, so that after recovering from a blank db, the hash diverges.
+	t.Log("putting 10 keys...")
+
+	oldCc := NewEtcdctl(epcOld.EndpointsV3(), clientNonTLS, false, false)
+	for i := 0; i < 10; i++ {
+		err := oldCc.Put(testutil.PickKey(int64(i)), fmt.Sprint(i))
+		assert.NoError(t, err, "error on put")
+	}
+
+	// Create a new cluster config, but with the same port numbers. In this way the new servers can stay in
+	// contact with the old ones.
+	epcNewConfig := &etcdProcessClusterConfig{
+		clusterSize:         3,
+		initialToken:        "new",
+		keepDataDir:         false,
+		CorruptCheckTime:    time.Second,
+		basePort:            basePort,
+		initialCorruptCheck: true,
+	}
+	epcNew, err := initEtcdProcessCluster(t, epcNewConfig)
+	if err != nil {
+		t.Fatalf("could not init etcd process cluster (%v)", err)
+	}
+	t.Cleanup(func() {
+		if errC := epcNew.Close(); errC != nil {
+			t.Fatalf("error closing etcd processes (%v)", errC)
+		}
+	})
+
+	newCc := NewEtcdctl(epcNew.EndpointsV3(), clientNonTLS, false, false)
+	assert.NoError(t, err)
+
+	// Rolling recovery of the servers.
+	t.Log("rolling updating servers in place...")
+	for i, newProc := range epcNew.procs {
+		oldProc := epcOld.procs[i]
+		err = oldProc.Close()
+		if err != nil {
+			t.Fatalf("could not stop etcd process (%v)", err)
+		}
+		t.Logf("old cluster server %d: %s stopped.", i, oldProc.Config().name)
+		err = newProc.Start()
+		if err != nil {
+			t.Fatalf("could not start etcd process (%v)", err)
+		}
+		t.Logf("new cluster server %d: %s started in-place with blank db.", i, newProc.Config().name)
+		t.Log("sleeping 5 sec to let nodes do periodical check...")
+		time.Sleep(5 * time.Second)
+	}
+	t.Log("new cluster started.")
+
+	alarmResponse, err := newCc.AlarmList()
+	assert.NoError(t, err, "error on alarm list")
+	for _, alarm := range alarmResponse.Alarms {
+		if alarm.Alarm == etcdserverpb.AlarmType_CORRUPT {
+			t.Fatalf("there is no corruption after in-place recovery, but corruption reported.")
+		}
+	}
+	t.Log("no corruption detected.")
+}
+
 func TestPeriodicCheckDetectsCorruption(t *testing.T) {
 	checkTime := time.Second
 	BeforeTest(t)

tests/integration/hashkv_test.go

@@ -52,13 +52,14 @@ func TestCompactionHash(t *testing.T) {
 		},
 	}
 
-	testutil.TestCompactionHash(context.Background(), t, hashTestCase{cc, clus.Members[0].GRPCURL(), client}, 1000)
+	testutil.TestCompactionHash(context.Background(), t, hashTestCase{cc, clus.Members[0].GRPCURL(), client, clus.Members[0].s}, 1000)
 }
 
 type hashTestCase struct {
 	*clientv3.Client
-	url  string
-	http *http.Client
+	url    string
+	http   *http.Client
+	server *etcdserver.EtcdServer
 }
 
 func (tc hashTestCase) Put(ctx context.Context, key, value string) error {
@@ -72,7 +73,7 @@ func (tc hashTestCase) Delete(ctx context.Context, key string) error {
 }
 
 func (tc hashTestCase) HashByRev(ctx context.Context, rev int64) (testutil.KeyValueHash, error) {
-	resp, err := etcdserver.HashByRev(ctx, tc.http, "http://unix", rev)
+	resp, err := etcdserver.HashByRev(ctx, tc.server.Cluster().ID(), tc.http, "http://unix", rev)
 	return testutil.KeyValueHash{Hash: resp.Hash, CompactRevision: resp.CompactRevision, Revision: resp.Header.Revision}, err
 }