Golang Cluster.NumNodes方法代码示例

// CheckGossip fetches the gossip infoStore from each node and invokes the given
// function. The test passes if the function returns 0 for every node,
// retrying for up to the given duration.
func CheckGossip(
	ctx context.Context, t testing.TB, c cluster.Cluster, d time.Duration, f CheckGossipFunc,
) {
	err := util.RetryForDuration(d, func() error {
		select {
		case <-stopper.ShouldStop():
			t.Fatalf("interrupted")
			return nil
		case <-time.After(1 * time.Second):
		}

		var infoStatus gossip.InfoStatus
		for i := 0; i < c.NumNodes(); i++ {
			if err := httputil.GetJSON(cluster.HTTPClient, c.URL(ctx, i)+"/_status/gossip/local", &infoStatus); err != nil {
				return errors.Wrapf(err, "failed to get gossip status from node %d", i)
			}
			if err := f(infoStatus.Infos); err != nil {
				return errors.Errorf("node %d: %s", i, err)
			}
		}

		return nil
	})
	if err != nil {
		t.Fatal(errors.Errorf("condition failed to evaluate within %s: %s", d, err))
	}
}

func testGossipPeeringsInner(
	ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig,
) {
	num := c.NumNodes()

	deadline := timeutil.Now().Add(cfg.Duration)

	waitTime := longWaitTime
	if cfg.Duration < waitTime {
		waitTime = shortWaitTime
	}

	for timeutil.Now().Before(deadline) {
		CheckGossip(ctx, t, c, waitTime, HasPeers(num))

		// Restart the first node.
		log.Infof(ctx, "restarting node 0")
		if err := c.Restart(ctx, 0); err != nil {
			t.Fatal(err)
		}
		CheckGossip(ctx, t, c, waitTime, HasPeers(num))

		// Restart another node (if there is one).
		var pickedNode int
		if num > 1 {
			pickedNode = rand.Intn(num-1) + 1
		}
		log.Infof(ctx, "restarting node %d", pickedNode)
		if err := c.Restart(ctx, pickedNode); err != nil {
			t.Fatal(err)
		}
		CheckGossip(ctx, t, c, waitTime, HasPeers(num))
	}
}

func testBuildInfoInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	checkGossip(t, c, 20*time.Second, hasPeers(c.NumNodes()))

	var details serverpb.DetailsResponse
	util.SucceedsSoon(t, func() error {
		select {
		case <-stopper:
			t.Fatalf("interrupted")
		default:
		}
		return httputil.GetJSON(cluster.HTTPClient, c.URL(0)+"/_status/details/local", &details)
	})

	bi := details.BuildInfo
	testData := map[string]string{
		"go_version":   bi.GoVersion,
		"tag":          bi.Tag,
		"time":         bi.Time,
		"dependencies": bi.Dependencies,
	}
	for key, val := range testData {
		if val == "" {
			t.Errorf("build info not set for \"%s\"", key)
		}
	}
}

// checkGossip fetches the gossip infoStore from each node and invokes the given
// function. The test passes if the function returns 0 for every node,
// retrying for up to the given duration.
func checkGossip(t *testing.T, c cluster.Cluster, d time.Duration, f checkGossipFunc) {
	err := util.RetryForDuration(d, func() error {
		select {
		case <-stopper:
			t.Fatalf("interrupted")
			return nil
		case <-time.After(1 * time.Second):
		}

		var infoStatus gossip.InfoStatus
		for i := 0; i < c.NumNodes(); i++ {
			if err := httputil.GetJSON(cluster.HTTPClient, c.URL(i)+"/_status/gossip/local", &infoStatus); err != nil {
				return err
			}
			if err := f(infoStatus.Infos); err != nil {
				return errors.Errorf("node %d: %s", i, err)
			}
		}

		return nil
	})
	if err != nil {
		t.Fatal(errors.Errorf("condition failed to evaluate within %s: %s", d, err))
	}
}

func mustGetHosts(t *testing.T, c cluster.Cluster) ([]iptables.IP, map[iptables.IP]int) {
	var addrs []iptables.IP
	addrsToNode := make(map[iptables.IP]int)
	for i := 0; i < c.NumNodes(); i++ {
		addr := iptables.IP(c.InternalIP(i).String())
		addrsToNode[addr] = i
		addrs = append(addrs, addr)
	}
	return addrs, addrsToNode
}

func restoreNetwork(t *testing.T, c cluster.Cluster) []error {
	var errs []error
	for i := 0; i < c.NumNodes(); i++ {
		for _, cmd := range iptables.Reset() {
			if err := c.ExecRoot(i, cmd); err != nil {
				errs = append(errs, err)
			}
		}
	}
	return errs
}

func testClusterRecoveryInner(
	ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig,
) {
	num := c.NumNodes()

	// One client for each node.
	initBank(t, c.PGUrl(ctx, 0))

	start := timeutil.Now()
	state := testState{
		t:        t,
		errChan:  make(chan error, num),
		teardown: make(chan struct{}),
		deadline: start.Add(cfg.Duration),
		clients:  make([]testClient, num),
	}

	for i := 0; i < num; i++ {
		state.clients[i].Lock()
		state.initClient(ctx, t, c, i)
		state.clients[i].Unlock()
		go transferMoneyLoop(ctx, i, &state, *numAccounts, *maxTransfer)
	}

	defer func() {
		<-state.teardown
	}()

	// Chaos monkey.
	rnd, seed := randutil.NewPseudoRand()
	log.Warningf(ctx, "monkey starts (seed %d)", seed)
	pickNodes := func() []int {
		return rnd.Perm(num)[:rnd.Intn(num)+1]
	}
	go chaosMonkey(ctx, &state, c, true, pickNodes, 0)

	waitClientsStop(ctx, num, &state, stall)

	// Verify accounts.
	verifyAccounts(t, &state.clients[0])

	elapsed := timeutil.Since(start)
	var count uint64
	counts := state.counts()
	for _, c := range counts {
		count += c
	}
	log.Infof(ctx, "%d %.1f/sec", count, float64(count)/elapsed.Seconds())
}

func testNodeRestartInner(
	ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig,
) {
	num := c.NumNodes()
	if minNum := 3; num < minNum {
		t.Skipf("need at least %d nodes, got %d", minNum, num)
	}

	// One client for each node.
	initBank(t, c.PGUrl(ctx, 0))

	start := timeutil.Now()
	state := testState{
		t:        t,
		errChan:  make(chan error, 1),
		teardown: make(chan struct{}),
		deadline: start.Add(cfg.Duration),
		clients:  make([]testClient, 1),
	}

	clientIdx := num - 1
	client := &state.clients[0]
	client.Lock()
	client.db = makePGClient(t, c.PGUrl(ctx, clientIdx))
	client.Unlock()
	go transferMoneyLoop(ctx, 0, &state, *numAccounts, *maxTransfer)

	defer func() {
		<-state.teardown
	}()

	// Chaos monkey.
	rnd, seed := randutil.NewPseudoRand()
	log.Warningf(ctx, "monkey starts (seed %d)", seed)
	pickNodes := func() []int {
		return []int{rnd.Intn(clientIdx)}
	}
	go chaosMonkey(ctx, &state, c, false, pickNodes, clientIdx)

	waitClientsStop(ctx, 1, &state, stall)

	// Verify accounts.
	verifyAccounts(t, client)

	elapsed := timeutil.Since(start)
	count := atomic.LoadUint64(&client.count)
	log.Infof(ctx, "%d %.1f/sec", count, float64(count)/elapsed.Seconds())
}

func testAdminLossOfQuorumInner(
	ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig,
) {
	if c.NumNodes() < 2 {
		t.Logf("skipping test %s because given cluster has too few nodes", cfg.Name)
		return
	}

	// Get the ids for each node.
	nodeIDs := make([]roachpb.NodeID, c.NumNodes())
	for i := 0; i < c.NumNodes(); i++ {
		var details serverpb.DetailsResponse
		if err := httputil.GetJSON(cluster.HTTPClient, c.URL(ctx, i)+"/_status/details/local", &details); err != nil {
			t.Fatalf("failed to get local details from node %d: %s", i, err)
		}
		nodeIDs[i] = details.NodeID
	}

	// Leave only the first node alive.
	for i := 1; i < c.NumNodes(); i++ {
		if err := c.Kill(ctx, i); err != nil {
			t.Fatal(err)
		}
	}

	// Retrieve node statuses.
	var nodes serverpb.NodesResponse
	if err := httputil.GetJSON(cluster.HTTPClient, c.URL(ctx, 0)+"/_status/nodes", &nodes); err != nil {
		t.Fatal(err)
	}

	for _, nodeID := range nodeIDs {
		var nodeStatus status.NodeStatus
		if err := httputil.GetJSON(cluster.HTTPClient, c.URL(ctx, 0)+"/_status/nodes/"+strconv.Itoa(int(nodeID)), &nodeStatus); err != nil {
			t.Fatal(err)
		}
	}

	// Retrieve time-series data.
	nowNanos := timeutil.Now().UnixNano()
	queryRequest := tspb.TimeSeriesQueryRequest{
		StartNanos: nowNanos - 10*time.Second.Nanoseconds(),
		EndNanos:   nowNanos,
		Queries: []tspb.Query{
			{Name: "doesn't_matter", Sources: []string{}},
		},
	}
	var queryResponse tspb.TimeSeriesQueryResponse
	if err := httputil.PostJSON(cluster.HTTPClient, c.URL(ctx, 0)+"/ts/query",
		&queryRequest, &queryResponse); err != nil {
		t.Fatal(err)
	}

	// TODO(cdo): When we're able to issue SQL queries without a quorum, test all
	// admin endpoints that issue SQL queries here.
}

func testRepairInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	testStopper := stop.NewStopper()
	dc := newDynamicClient(c, testStopper)
	testStopper.AddCloser(dc)
	defer testStopper.Stop()

	// Add some loads.
	for i := 0; i < c.NumNodes()*2; i++ {
		ID := i
		testStopper.RunWorker(func() {
			insertLoad(t, dc, ID)
		})
	}

	// TODO(bram): #5345 add repair mechanism.

	select {
	case <-stopper:
	case <-time.After(cfg.Duration):
	}
}

// BidirectionalPartitionNemesis is a nemesis which randomly severs the network
// symmetrically between two random groups of nodes. Partitioned and connected
// mode take alternating turns, with random durations of up to 15s.
func BidirectionalPartitionNemesis(t *testing.T, stop <-chan struct{}, c cluster.Cluster) {
	randSec := func() time.Duration { return time.Duration(rand.Int63n(15 * int64(time.Second))) }
	log.Infof(context.Background(), "cleaning up any previous rules")
	_ = restoreNetwork(t, c) // clean up any potential leftovers
	log.Infof(context.Background(), "starting partition nemesis")
	for {
		ch := make(chan struct{})
		go func() {
			select {
			case <-time.After(randSec()):
			case <-stop:
			}
			close(ch)
		}()
		cutNetwork(t, c, ch, randomBidirectionalPartition(c.NumNodes())...)
		select {
		case <-stop:
			return
		case <-time.After(randSec()):
		}
	}
}

func testPutInner(ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	db, err := c.NewClient(ctx, 0)
	if err != nil {
		t.Fatal(err)
	}

	errs := make(chan error, c.NumNodes())
	start := timeutil.Now()
	deadline := start.Add(cfg.Duration)
	var count int64
	for i := 0; i < c.NumNodes(); i++ {
		go func() {
			r, _ := randutil.NewPseudoRand()
			value := randutil.RandBytes(r, 8192)

			for timeutil.Now().Before(deadline) {
				k := atomic.AddInt64(&count, 1)
				v := value[:r.Intn(len(value))]
				if err := db.Put(ctx, fmt.Sprintf("%08d", k), v); err != nil {
					errs <- err
					return
				}
			}
			errs <- nil
		}()
	}

	for i := 0; i < c.NumNodes(); {
		baseCount := atomic.LoadInt64(&count)
		select {
		case <-stopper.ShouldStop():
			t.Fatalf("interrupted")
		case err := <-errs:
			if err != nil {
				t.Fatal(err)
			}
			i++
		case <-time.After(1 * time.Second):
			// Periodically print out progress so that we know the test is still
			// running.
			loadedCount := atomic.LoadInt64(&count)
			log.Infof(ctx, "%d (%d/s)", loadedCount, loadedCount-baseCount)
			c.Assert(ctx, t)
			if err := cluster.Consistent(ctx, c, 0); err != nil {
				t.Fatal(err)
			}
		}
	}

	elapsed := timeutil.Since(start)
	log.Infof(ctx, "%d %.1f/sec", count, float64(count)/elapsed.Seconds())
}

func checkRangeReplication(t *testing.T, c cluster.Cluster, d time.Duration) {
	if c.NumNodes() < 1 {
		// Looks silly, but we actually start zero-node clusters in the
		// reference tests.
		t.Log("replication test is a no-op for empty cluster")
		return
	}

	wantedReplicas := 3
	if c.NumNodes() < 3 {
		wantedReplicas = c.NumNodes()
	}

	log.Infof(context.Background(), "waiting for first range to have %d replicas", wantedReplicas)

	util.SucceedsSoon(t, func() error {
		// Reconnect on every iteration; gRPC will eagerly tank the connection
		// on transport errors. Always talk to node 0 because it's guaranteed
		// to exist.
		client, dbStopper := c.NewClient(t, 0)
		defer dbStopper.Stop()

		select {
		case <-stopper:
			t.Fatalf("interrupted")
			return nil
		case <-time.After(1 * time.Second):
		}

		foundReplicas, err := countRangeReplicas(client)
		if err != nil {
			return err
		}

		if log.V(1) {
			log.Infof(context.Background(), "found %d replicas", foundReplicas)
		}
		if foundReplicas >= wantedReplicas {
			return nil
		}
		return fmt.Errorf("expected %d replicas, only found %d", wantedReplicas, foundReplicas)
	})

	log.Infof(context.Background(), "found %d replicas", wantedReplicas)
}

func testFreezeClusterInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	minAffected := int64(server.ExpectedInitialRangeCount())

	const long = time.Minute
	const short = 10 * time.Second

	mustPost := func(freeze bool) serverpb.ClusterFreezeResponse {
		reply, err := postFreeze(c, freeze, long)
		if err != nil {
			t.Fatal(errors.Errorf("%v", err))
		}
		return reply
	}

	if reply := mustPost(false); reply.RangesAffected != 0 {
		t.Fatalf("expected initial unfreeze to affect no ranges, got %d", reply.RangesAffected)
	}

	if reply := mustPost(true); reply.RangesAffected < minAffected {
		t.Fatalf("expected >=%d frozen ranges, got %d", minAffected, reply.RangesAffected)
	}

	if reply := mustPost(true); reply.RangesAffected != 0 {
		t.Fatalf("expected second freeze to affect no ranges, got %d", reply.RangesAffected)
	}

	if reply := mustPost(false); reply.RangesAffected < minAffected {
		t.Fatalf("expected >=%d thawed ranges, got %d", minAffected, reply.RangesAffected)
	}

	num := c.NumNodes()
	if num < 3 {
		t.Skip("skipping remainder of test; needs at least 3 nodes")
	}

	// Kill the last node.
	if err := c.Kill(num - 1); err != nil {
		t.Fatal(err)
	}

	// Attempt to freeze should get stuck (since it does not get confirmation
	// of the last node receiving the freeze command).
	// Note that this is the freeze trigger stalling on the Replica, not the
	// Store-polling mechanism.
	acceptErrs := strings.Join([]string{
		"timed out waiting for Range",
		"Timeout exceeded while",
		"connection is closing",
		"deadline",
		// error returned via JSON when the server-side gRPC stream times out (due to
		// lack of new input). Unmarshaling that JSON fails with a message referencing
		// unknown fields, unfortunately in map order.
		"unknown field .*",
	}, "|")
	if reply, err := postFreeze(c, true, short); !testutils.IsError(err, acceptErrs) {
		t.Fatalf("expected timeout, got %v: %v", err, reply)
	}

	// Shut down the remaining nodes and restart them.
	for i := 0; i < num-1; i++ {
		if err := c.Kill(i); err != nil {
			t.Fatal(err)
		}
	}
	for i := 0; i < num; i++ {
		if err := c.Restart(i); err != nil {
			t.Fatal(err)
		}
	}

	// The cluster should now be fully operational (at least after waiting
	// a little bit) since each node tries to unfreeze everything when it
	// starts.
	if err := util.RetryForDuration(time.Minute, func() error {
		if _, err := postFreeze(c, false, short); err != nil {
			if testutils.IsError(err, "404 Not Found") {
				// It can take a bit until the endpoint is available.
				return err
			}
			t.Fatal(err)
		}

		// TODO(tschottdorf): moving the client creation outside of the retry
		// loop will break the test with the following message:
		//
		//   client/rpc_sender.go:61: roachpb.Batch RPC failed as client
		//   connection was closed
		//
		// Perhaps the cluster updates the address too late after restarting
		// the node.
		db, dbStopper := c.NewClient(t, 0)
		defer dbStopper.Stop()

		if _, err := db.Scan(context.TODO(), keys.LocalMax, roachpb.KeyMax, 0); err != nil {
			t.Fatal(err)
		}
		return nil
	}); err != nil {
		t.Fatal(err)
	}
//.........这里部分代码省略.........

func testEventLogInner(
	ctx context.Context, t *testing.T, c cluster.Cluster, cfg cluster.TestConfig,
) {
	num := c.NumNodes()
	if num <= 0 {
		t.Fatalf("%d nodes in cluster", num)
	}

	var confirmedClusterID uuid.UUID
	type nodeEventInfo struct {
		Descriptor roachpb.NodeDescriptor
		ClusterID  uuid.UUID
	}

	// Verify that a node_join message was logged for each node in the cluster.
	// We expect there to eventually be one such message for each node in the
	// cluster, and each message must be correctly formatted.
	util.SucceedsSoon(t, func() error {
		db := makePGClient(t, c.PGUrl(ctx, 0))
		defer db.Close()

		// Query all node join events. There should be one for each node in the
		// cluster.
		rows, err := db.Query(
			"SELECT targetID, info FROM system.eventlog WHERE eventType = $1",
			string(csql.EventLogNodeJoin))
		if err != nil {
			return err
		}
		seenIds := make(map[int64]struct{})
		var clusterID uuid.UUID
		for rows.Next() {
			var targetID int64
			var infoStr gosql.NullString
			if err := rows.Scan(&targetID, &infoStr); err != nil {
				t.Fatal(err)
			}

			// Verify the stored node descriptor.
			if !infoStr.Valid {
				t.Fatalf("info not recorded for node join, target node %d", targetID)
			}
			var info nodeEventInfo
			if err := json.Unmarshal([]byte(infoStr.String), &info); err != nil {
				t.Fatal(err)
			}
			if a, e := int64(info.Descriptor.NodeID), targetID; a != e {
				t.Fatalf("Node join with targetID %d had descriptor for wrong node %d", e, a)
			}

			// Verify cluster ID is recorded, and is the same for all nodes.
			if (info.ClusterID == uuid.UUID{}) {
				t.Fatalf("Node join recorded nil cluster id, info: %v", info)
			}
			if (clusterID == uuid.UUID{}) {
				clusterID = info.ClusterID
			} else if clusterID != info.ClusterID {
				t.Fatalf(
					"Node join recorded different cluster ID than earlier node. Expected %s, got %s. Info: %v",
					clusterID, info.ClusterID, info)
			}

			// Verify that all NodeIDs are different.
			if _, ok := seenIds[targetID]; ok {
				t.Fatalf("Node ID %d seen in two different node join messages", targetID)
			}
			seenIds[targetID] = struct{}{}
		}
		if err := rows.Err(); err != nil {
			return err
		}

		if a, e := len(seenIds), c.NumNodes(); a != e {
			return errors.Errorf("expected %d node join messages, found %d: %v", e, a, seenIds)
		}

		confirmedClusterID = clusterID
		return nil
	})

	// Stop and Start Node 0, and verify the node restart message.
	if err := c.Kill(ctx, 0); err != nil {
		t.Fatal(err)
	}
	if err := c.Restart(ctx, 0); err != nil {
		t.Fatal(err)
	}

	util.SucceedsSoon(t, func() error {
		db := makePGClient(t, c.PGUrl(ctx, 0))
		defer db.Close()

		// Query all node restart events. There should only be one.
		rows, err := db.Query(
			"SELECT targetID, info FROM system.eventlog WHERE eventType = $1",
			string(csql.EventLogNodeRestart))
		if err != nil {
			return err
		}

//.........这里部分代码省略.........