Golang cluster.Cluster类代码示例

func cutNetwork(t *testing.T, c cluster.Cluster, closer <-chan struct{}, partitions ...[]int) {
	defer func() {
		if errs := restoreNetwork(t, c); len(errs) > 0 {
			t.Fatalf("errors restoring the network: %+v", errs)
		}
	}()
	addrs, addrsToNode := mustGetHosts(t, c)
	ipPartitions := make([][]iptables.IP, 0, len(partitions))
	for _, partition := range partitions {
		ipPartition := make([]iptables.IP, 0, len(partition))
		for _, nodeIndex := range partition {
			ipPartition = append(ipPartition, addrs[nodeIndex])
		}
		ipPartitions = append(ipPartitions, ipPartition)
	}
	log.Warningf(context.TODO(), "partitioning: %v (%v)", partitions, ipPartitions)
	for host, cmds := range iptables.Rules(iptables.Bidirectional(ipPartitions...)) {
		for _, cmd := range cmds {
			if err := c.ExecRoot(addrsToNode[host], cmd); err != nil {
				t.Fatal(err)
			}
		}
	}
	<-closer
	log.Warningf(context.TODO(), "resolved all partitions")
}

func postFreeze(c cluster.Cluster, freeze bool, timeout time.Duration) (serverpb.ClusterFreezeResponse, error) {
	httpClient := cluster.HTTPClient
	httpClient.Timeout = timeout

	var resp serverpb.ClusterFreezeResponse
	log.Infof("requesting: freeze=%t, timeout=%s", freeze, timeout)
	cb := func(v proto.Message) {
		oldNum := resp.RangesAffected
		resp = *v.(*serverpb.ClusterFreezeResponse)
		if oldNum > resp.RangesAffected {
			resp.RangesAffected = oldNum
		}
		if (resp != serverpb.ClusterFreezeResponse{}) {
			log.Infof("%+v", &resp)
		}
	}
	err := util.StreamJSON(
		httpClient,
		c.URL(0)+"/_admin/v1/cluster/freeze",
		&serverpb.ClusterFreezeRequest{Freeze: freeze},
		&serverpb.ClusterFreezeResponse{},
		cb,
	)
	return resp, err
}

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

	util.SucceedsSoon(t, func() error {
		select {
		case <-stopper:
			t.Fatalf("interrupted")
			return nil
		default:
		}
		var r struct {
			BuildInfo map[string]string
		}
		if err := getJSON(c.URL(0), "/_status/details/local", &r); err != nil {
			return err
		}
		for _, key := range []string{"goVersion", "tag", "time", "dependencies"} {
			if val, ok := r.BuildInfo[key]; !ok {
				t.Errorf("build info missing for \"%s\"", key)
			} else if val == "" {
				t.Errorf("build info not set for \"%s\"", key)
			}
		}
		return nil
	})
}

func testStatusServerInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	// Get the ids for each node.
	idMap := make(map[int]string)
	for i := 0; i < c.NumNodes(); i++ {
		var detail details
		if err := getJSON(c.URL(i), "/_status/details/local", &detail); err != nil {
			t.Fatal(err)
		}
		idMap[i] = detail.NodeID.String()
	}

	// Check local response for the every node.
	for i := 0; i < c.NumNodes(); i++ {
		checkNode(t, c, i, idMap[i], "local", idMap[i])
		get(t, c.URL(i), "/_status/nodes")
		get(t, c.URL(i), "/_status/stores")
	}

	// Proxy from the first node to the last node.
	firstNode := 0
	lastNode := c.NumNodes() - 1
	firstID := idMap[firstNode]
	lastID := idMap[lastNode]
	checkNode(t, c, firstNode, firstID, lastID, lastID)

	// And from the last node to the first node.
	checkNode(t, c, lastNode, lastID, firstID, firstID)

	// And from the last node to the last node.
	checkNode(t, c, lastNode, lastID, lastID, lastID)
}

// 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) {
	util.SucceedsWithin(t, d, func() error {
		select {
		case <-stopper:
			t.Fatalf("interrupted")
			return nil
		case <-time.After(1 * time.Second):
		}

		for i := 0; i < c.NumNodes(); i++ {
			var m map[string]interface{}
			if err := getJSON(c.URL(i), "/_status/gossip/local", &m); err != nil {
				return err
			}
			infos, ok := m["infos"].(map[string]interface{})
			if !ok {
				return errors.New("no infos yet")
			}
			if err := f(infos); err != nil {
				return util.Errorf("node %d: %s", i, err)
			}
		}

		return nil
	})
}

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

	var details server.DetailsResponse
	util.SucceedsSoon(t, func() error {
		select {
		case <-stopper:
			t.Fatalf("interrupted")
		default:
		}
		return util.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)
		}
	}
}

func testStatusServerInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	// Get the ids for each node.
	idMap := make(map[int]roachpb.NodeID)
	for i := 0; i < c.NumNodes(); i++ {
		var details server.DetailsResponse
		if err := util.GetJSON(cluster.HTTPClient, c.URL(i)+"/_status/details/local", &details); err != nil {
			t.Fatal(err)
		}
		idMap[i] = details.NodeID
	}

	// Check local response for the every node.
	for i := 0; i < c.NumNodes(); i++ {
		id := idMap[i]
		checkNode(t, c, i, id, id, id)
		get(t, c.URL(i), "/_status/nodes")
	}

	// Proxy from the first node to the last node.
	firstNode := 0
	lastNode := c.NumNodes() - 1
	firstID := idMap[firstNode]
	lastID := idMap[lastNode]
	checkNode(t, c, firstNode, firstID, lastID, lastID)

	// And from the last node to the first node.
	checkNode(t, c, lastNode, lastID, firstID, firstID)

	// And from the last node to the last node.
	checkNode(t, c, lastNode, lastID, lastID, lastID)
}

// 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 := util.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 postFreeze(c cluster.Cluster, freeze bool) (server.ClusterFreezeResponse, error) {
	httpClient := cluster.HTTPClient()
	httpClient.Timeout = 10 * time.Second

	var resp server.ClusterFreezeResponse
	err := postJSON(httpClient, c.URL(0), "/_admin/v1/cluster/freeze",
		&server.ClusterFreezeRequest{Freeze: freeze}, &resp)
	return resp, err
}

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 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 postFreeze(c cluster.Cluster, freeze bool, timeout time.Duration) (server.ClusterFreezeResponse, error) {
	httpClient := cluster.HTTPClient
	httpClient.Timeout = timeout

	var resp server.ClusterFreezeResponse
	err := util.PostJSON(
		httpClient,
		c.URL(0)+"/_admin/v1/cluster/freeze",
		&server.ClusterFreezeRequest{Freeze: freeze},
		&resp,
	)
	return resp, err
}

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

	// One client for each node.
	initBank(t, c.PGUrl(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(t, c, i)
		state.clients[i].Unlock()
		go transferMoneyLoop(i, &state, *numAccounts, *maxTransfer)
	}

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

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

	waitClientsStop(num, &state, cfg.Stall)

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

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

func testPutInner(t *testing.T, c cluster.Cluster, cfg cluster.TestConfig) {
	db, dbStopper := c.NewClient(t, 0)
	defer dbStopper.Stop()

	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(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:
			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(context.Background(), "%d (%d/s)", loadedCount, loadedCount-baseCount)
			c.Assert(t)
			cluster.Consistent(t, c)
		}
	}

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

// 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))) }
	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()):
		}
	}
}