Golang testutils.TeardownCluster函数代码示例

func TestRaftQuorumRecovery(t *testing.T) {
	t.Parallel()

	// Bring up a 5 nodes cluster
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Lose a majority
	for i := uint64(1); i <= 3; i++ {
		nodes[i].Server.Stop()
		nodes[i].Shutdown()
	}

	raftutils.AdvanceTicks(clockSource, 5)

	// Restore the majority by restarting node 3
	nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)

	delete(nodes, 1)
	delete(nodes, 2)
	raftutils.WaitForCluster(t, clockSource, nodes)

	// Propose a value
	value, err := raftutils.ProposeValue(t, raftutils.Leader(nodes))
	assert.NoError(t, err)

	for _, node := range nodes {
		raftutils.CheckValue(t, clockSource, node, value)
	}
}

func testRaftRestartCluster(t *testing.T, stagger bool) {
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Propose a value
	values := make([]*api.Node, 2)
	var err error
	values[0], err = raftutils.ProposeValue(t, nodes[1], "id1")
	assert.NoError(t, err, "failed to propose value")

	// Stop all nodes
	for _, node := range nodes {
		node.Server.Stop()
		node.Shutdown()
	}

	raftutils.AdvanceTicks(clockSource, 5)

	// Restart all nodes
	i := 0
	for k, node := range nodes {
		if stagger && i != 0 {
			raftutils.AdvanceTicks(clockSource, 1)
		}
		nodes[k] = raftutils.RestartNode(t, clockSource, node, false)
		i++
	}
	raftutils.WaitForCluster(t, clockSource, nodes)

	// Propose another value
	values[1], err = raftutils.ProposeValue(t, raftutils.Leader(nodes), "id2")
	assert.NoError(t, err, "failed to propose value")

	for _, node := range nodes {
		assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
			var err error
			node.MemoryStore().View(func(tx store.ReadTx) {
				var allNodes []*api.Node
				allNodes, err = store.FindNodes(tx, store.All)
				if err != nil {
					return
				}
				if len(allNodes) != 2 {
					err = fmt.Errorf("expected 2 nodes, got %d", len(allNodes))
					return
				}

				for i, nodeID := range []string{"id1", "id2"} {
					n := store.GetNode(tx, nodeID)
					if !reflect.DeepEqual(n, values[i]) {
						err = fmt.Errorf("node %s did not match expected value", nodeID)
						return
					}
				}
			})
			return err
		}))
	}
}

func TestRaftBootstrap(t *testing.T) {
	t.Parallel()

	nodes, _ := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	assert.Equal(t, 3, len(nodes[1].GetMemberlist()))
	assert.Equal(t, 3, len(nodes[2].GetMemberlist()))
	assert.Equal(t, 3, len(nodes[3].GetMemberlist()))
}

// This test rotates the encryption key and restarts the node - the intent is try to trigger
// race conditions if there is more than one node and hence consensus may take longer.
func TestRaftEncryptionKeyRotationStress(t *testing.T) {
	t.Parallel()

	// Bring up a 3 nodes cluster
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)
	leader := nodes[1]

	// constantly propose values
	done, stop, restart, clusterReady := make(chan struct{}), make(chan struct{}), make(chan struct{}), make(chan struct{})
	go func() {
		counter := len(nodes)
		for {
			select {
			case <-stop:
				close(done)
				return
			case <-restart:
				// the node restarts may trigger a leadership change, so wait until the cluster has 3
				// nodes again and a leader is selected before proposing more values
				<-clusterReady
				leader = raftutils.Leader(nodes)
			default:
				counter += 1
				raftutils.ProposeValue(t, leader, DefaultProposalTime, fmt.Sprintf("id%d", counter))
			}
		}
	}()

	for i := 0; i < 30; i++ {
		// rotate the encryption key
		nodes[3].KeyRotator.QueuePendingKey([]byte(fmt.Sprintf("newKey%d", i)))
		nodes[3].KeyRotator.RotationNotify() <- struct{}{}

		require.NoError(t, raftutils.PollFunc(clockSource, func() error {
			if nodes[3].KeyRotator.GetKeys().PendingDEK == nil {
				return nil
			}
			return fmt.Errorf("not done rotating yet")
		}))

		// restart the node and wait for everything to settle and a leader to be elected
		nodes[3].Server.Stop()
		nodes[3].ShutdownRaft()
		restart <- struct{}{}
		nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)
		raftutils.AdvanceTicks(clockSource, 1)

		raftutils.WaitForCluster(t, clockSource, nodes)
		clusterReady <- struct{}{}
	}

	close(stop)
	<-done
}

func TestCanRemoveMember(t *testing.T) {
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Stop node 2 and node 3 (2 nodes out of 3)
	nodes[2].Server.Stop()
	nodes[2].Shutdown()
	nodes[3].Server.Stop()
	nodes[3].Shutdown()

	// Node 2 and Node 3 should be listed as Unreachable
	assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
		members := nodes[1].GetMemberlist()
		if len(members) != 3 {
			return fmt.Errorf("expected 3 nodes, got %d", len(members))
		}
		if members[nodes[2].Config.ID].Status.Reachability == api.RaftMemberStatus_REACHABLE {
			return fmt.Errorf("expected node 2 to be unreachable")
		}
		if members[nodes[3].Config.ID].Status.Reachability == api.RaftMemberStatus_REACHABLE {
			return fmt.Errorf("expected node 3 to be unreachable")
		}
		return nil
	}))

	// Removing node 3 should fail
	ctx, _ := context.WithTimeout(context.Background(), 10*time.Second)
	err := nodes[1].RemoveMember(ctx, 3)
	assert.Error(t, err)
	assert.Equal(t, err, raft.ErrCannotRemoveMember)
	members := nodes[1].GetMemberlist()
	assert.Equal(t, len(members), 3)

	// Restart node 2 and node 3
	nodes[2] = raftutils.RestartNode(t, clockSource, nodes[2], false)
	nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)
	raftutils.WaitForCluster(t, clockSource, nodes)

	// Removing node 3 should succeed
	ctx, _ = context.WithTimeout(context.Background(), 10*time.Second)
	err = nodes[1].RemoveMember(ctx, nodes[3].Config.ID)
	assert.NoError(t, err)
	members = nodes[1].GetMemberlist()
	assert.Nil(t, members[nodes[3].Config.ID])
	assert.Equal(t, len(members), 2)

	// Removing node 2 should fail
	ctx, _ = context.WithTimeout(context.Background(), 10*time.Second)
	err = nodes[1].RemoveMember(ctx, nodes[2].Config.ID)
	assert.Error(t, err)
	assert.Equal(t, err, raft.ErrCannotRemoveMember)
	assert.Equal(t, len(members), 2)
}

func TestRaftLeaderDown(t *testing.T) {
	t.Parallel()

	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Stop node 1
	nodes[1].Stop()

	newCluster := map[uint64]*raftutils.TestNode{
		2: nodes[2],
		3: nodes[3],
	}
	// Wait for the re-election to occur
	raftutils.WaitForCluster(t, clockSource, newCluster)

	// Leader should not be 1
	assert.NotEqual(t, nodes[2].Leader(), nodes[1].Config.ID)

	// Ensure that node 2 and node 3 have the same leader
	assert.Equal(t, nodes[3].Leader(), nodes[2].Leader())

	// Find the leader node and a follower node
	var (
		leaderNode   *raftutils.TestNode
		followerNode *raftutils.TestNode
	)
	for i, n := range newCluster {
		if n.Config.ID == n.Leader() {
			leaderNode = n
			if i == 2 {
				followerNode = newCluster[3]
			} else {
				followerNode = newCluster[2]
			}
		}
	}

	require.NotNil(t, leaderNode)
	require.NotNil(t, followerNode)

	// Propose a value
	value, err := raftutils.ProposeValue(t, leaderNode)
	assert.NoError(t, err, "failed to propose value")

	// The value should be replicated on all remaining nodes
	raftutils.CheckValue(t, clockSource, leaderNode, value)
	assert.Equal(t, len(leaderNode.GetMemberlist()), 3)

	raftutils.CheckValue(t, clockSource, followerNode, value)
	assert.Equal(t, len(followerNode.GetMemberlist()), 3)
}

func TestRaftLeader(t *testing.T) {
	t.Parallel()

	nodes, _ := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	assert.True(t, nodes[1].IsLeader(), "error: node 1 is not the Leader")

	// nodes should all have the same leader
	assert.Equal(t, nodes[1].Leader(), nodes[1].Config.ID)
	assert.Equal(t, nodes[2].Leader(), nodes[1].Config.ID)
	assert.Equal(t, nodes[3].Leader(), nodes[1].Config.ID)
}

func TestRaftLogReplication(t *testing.T) {
	t.Parallel()

	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Propose a value
	value, err := raftutils.ProposeValue(t, nodes[1])
	assert.NoError(t, err, "failed to propose value")

	// All nodes should have the value in the physical store
	raftutils.CheckValue(t, clockSource, nodes[1], value)
	raftutils.CheckValue(t, clockSource, nodes[2], value)
	raftutils.CheckValue(t, clockSource, nodes[3], value)
}

func TestRaftUnreachableNode(t *testing.T) {
	t.Parallel()

	nodes := make(map[uint64]*raftutils.TestNode)
	var clockSource *fakeclock.FakeClock
	nodes[1], clockSource = raftutils.NewInitNode(t, tc, nil)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	// Add a new node
	nodes[2] = raftutils.NewNode(t, clockSource, tc, raft.NodeOptions{JoinAddr: nodes[1].Address})

	err := nodes[2].JoinAndStart(ctx)
	require.NoError(t, err, "can't join cluster")

	go nodes[2].Run(ctx)

	// Stop the Raft server of second node on purpose after joining
	nodes[2].Server.Stop()
	nodes[2].Listener.Close()

	raftutils.AdvanceTicks(clockSource, 5)
	time.Sleep(100 * time.Millisecond)

	wrappedListener := raftutils.RecycleWrappedListener(nodes[2].Listener)
	securityConfig := nodes[2].SecurityConfig
	serverOpts := []grpc.ServerOption{grpc.Creds(securityConfig.ServerTLSCreds)}
	s := grpc.NewServer(serverOpts...)

	nodes[2].Server = s
	raft.Register(s, nodes[2].Node)

	go func() {
		// After stopping, we should receive an error from Serve
		assert.Error(t, s.Serve(wrappedListener))
	}()

	raftutils.WaitForCluster(t, clockSource, nodes)
	defer raftutils.TeardownCluster(t, nodes)

	// Propose a value
	value, err := raftutils.ProposeValue(t, nodes[1], DefaultProposalTime)
	assert.NoError(t, err, "failed to propose value")

	// All nodes should have the value in the physical store
	raftutils.CheckValue(t, clockSource, nodes[1], value)
	raftutils.CheckValue(t, clockSource, nodes[2], value)
}

func TestRaftLogReplicationWithoutLeader(t *testing.T) {
	t.Parallel()
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Stop the leader
	nodes[1].Stop()

	// Propose a value
	_, err := raftutils.ProposeValue(t, nodes[2])
	assert.Error(t, err)

	// No value should be replicated in the store in the absence of the leader
	raftutils.CheckNoValue(t, clockSource, nodes[2])
	raftutils.CheckNoValue(t, clockSource, nodes[3])
}

func TestRaftFollowerLeave(t *testing.T) {
	t.Parallel()

	// Bring up a 5 nodes cluster
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Node 5 leaves the cluster
	// Use gRPC instead of calling handler directly because of
	// authorization check.
	cc, err := dial(nodes[1], nodes[1].Address)
	assert.NoError(t, err)
	raftClient := api.NewRaftMembershipClient(cc)
	defer cc.Close()
	ctx, _ := context.WithTimeout(context.Background(), 10*time.Second)
	resp, err := raftClient.Leave(ctx, &api.LeaveRequest{Node: &api.RaftMember{RaftID: nodes[5].Config.ID}})
	assert.NoError(t, err, "error sending message to leave the raft")
	assert.NotNil(t, resp, "leave response message is nil")

	raftutils.ShutdownNode(nodes[5])
	delete(nodes, 5)

	raftutils.WaitForPeerNumber(t, clockSource, nodes, 4)

	// Propose a value
	value, err := raftutils.ProposeValue(t, nodes[1], DefaultProposalTime)
	assert.NoError(t, err, "failed to propose value")

	// Value should be replicated on every node
	raftutils.CheckValue(t, clockSource, nodes[1], value)
	assert.Len(t, nodes[1].GetMemberlist(), 4)

	raftutils.CheckValue(t, clockSource, nodes[2], value)
	assert.Len(t, nodes[2].GetMemberlist(), 4)

	raftutils.CheckValue(t, clockSource, nodes[3], value)
	assert.Len(t, nodes[3].GetMemberlist(), 4)

	raftutils.CheckValue(t, clockSource, nodes[4], value)
	assert.Len(t, nodes[4].GetMemberlist(), 4)
}

func TestRaftJoinTwice(t *testing.T) {
	t.Parallel()

	nodes, _ := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Node 3 tries to join again
	// Use gRPC instead of calling handler directly because of
	// authorization check.
	cc, err := dial(nodes[3], nodes[1].Address)
	assert.NoError(t, err)
	raftClient := api.NewRaftMembershipClient(cc)
	defer cc.Close()
	ctx, _ := context.WithTimeout(context.Background(), 10*time.Second)
	_, err = raftClient.Join(ctx, &api.JoinRequest{})
	assert.Error(t, err, "expected error on duplicate Join")
	assert.Equal(t, grpc.Code(err), codes.AlreadyExists)
	assert.Equal(t, grpc.ErrorDesc(err), "a raft member with this node ID already exists")
}

func TestRaftWipedState(t *testing.T) {
	t.Parallel()

	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Stop node 3
	nodes[3].Server.Stop()
	nodes[3].ShutdownRaft()

	// Remove its state
	os.RemoveAll(nodes[3].StateDir)

	raftutils.AdvanceTicks(clockSource, 5)

	// Restart node 3
	nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)

	// Make sure this doesn't panic.
	raftutils.PollFuncWithTimeout(clockSource, func() error { return errors.New("keep the poll going") }, time.Second)
}

func TestRaftNewNodeGetsData(t *testing.T) {
	t.Parallel()

	// Bring up a 3 node cluster
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Propose a value
	value, err := raftutils.ProposeValue(t, nodes[1])
	assert.NoError(t, err, "failed to propose value")

	// Add a new node
	raftutils.AddRaftNode(t, clockSource, nodes, tc)

	time.Sleep(500 * time.Millisecond)

	// Value should be replicated on every node
	for _, node := range nodes {
		raftutils.CheckValue(t, clockSource, node, value)
		assert.Equal(t, len(node.GetMemberlist()), 4)
	}
}

func TestRaftQuorumFailure(t *testing.T) {
	t.Parallel()

	// Bring up a 5 nodes cluster
	nodes, clockSource := raftutils.NewRaftCluster(t, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	raftutils.AddRaftNode(t, clockSource, nodes, tc)
	defer raftutils.TeardownCluster(t, nodes)

	// Lose a majority
	for i := uint64(3); i <= 5; i++ {
		nodes[i].Server.Stop()
		nodes[i].Stop()
	}

	// Propose a value
	_, err := raftutils.ProposeValue(t, nodes[1])
	assert.Error(t, err)

	// The value should not be replicated, we have no majority
	raftutils.CheckNoValue(t, clockSource, nodes[2])
	raftutils.CheckNoValue(t, clockSource, nodes[1])
}