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].ShutdownRaft()
}
raftutils.AdvanceTicks(clockSource, 5)
// Restore the majority by restarting node 3
nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)
raftutils.ShutdownNode(nodes[1])
delete(nodes, 1)
raftutils.ShutdownNode(nodes[2])
delete(nodes, 2)
raftutils.WaitForCluster(t, clockSource, nodes)
// Propose a value
value, err := raftutils.ProposeValue(t, raftutils.Leader(nodes), DefaultProposalTime)
assert.NoError(t, err)
for _, node := range nodes {
raftutils.CheckValue(t, clockSource, node, value)
}
}
func TestRaftJoinWithIncorrectAddress(t *testing.T) {
t.Parallel()
nodes := make(map[uint64]*raftutils.TestNode)
var clockSource *fakeclock.FakeClock
nodes[1], clockSource = raftutils.NewInitNode(t, tc, nil)
defer raftutils.ShutdownNode(nodes[1])
// Try joining a new node with an incorrect address
n := raftutils.NewNode(t, clockSource, tc, raft.NodeOptions{JoinAddr: nodes[1].Address, Addr: "1.2.3.4:1234"})
defer raftutils.CleanupNonRunningNode(n)
err := n.JoinAndStart(context.Background())
assert.NotNil(t, err)
assert.Contains(t, grpc.ErrorDesc(err), "could not connect to prospective new cluster member using its advertised address")
// Check if first node still has only itself registered in the memberlist
assert.Len(t, nodes[1].GetMemberlist(), 1)
}
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 TestRaftJoinWithIncorrectAddress(t *testing.T) {
t.Parallel()
nodes := make(map[uint64]*raftutils.TestNode)
var clockSource *fakeclock.FakeClock
nodes[1], clockSource = raftutils.NewInitNode(t, tc, nil)
defer raftutils.ShutdownNode(nodes[1])
// Try joining a new node with an incorrect address
n := raftutils.NewNode(t, clockSource, tc, raft.NewNodeOptions{JoinAddr: nodes[1].Address, Addr: "1.2.3.4:1234"})
defer raftutils.CleanupNonRunningNode(n)
err := n.JoinAndStart()
assert.NotNil(t, err)
assert.Equal(t, grpc.ErrorDesc(err), raft.ErrHealthCheckFailure.Error())
// Check if first node still has only itself registered in the memberlist
assert.Equal(t, len(nodes[1].GetMemberlist()), 1)
}
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 all nodes should fail
for i := 1; i <= 3; i++ {
ctx, _ := context.WithTimeout(context.Background(), 10*time.Second)
err := nodes[1].RemoveMember(ctx, uint64(i))
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)
// Node 2 and Node 3 should be listed as Reachable
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_UNREACHABLE {
return fmt.Errorf("expected node 2 to be reachable")
}
if members[nodes[3].Config.ID].Status.Reachability == api.RaftMemberStatus_UNREACHABLE {
return fmt.Errorf("expected node 3 to be reachable")
}
return nil
}))
// Stop Node 3 (1 node out of 3)
nodes[3].Server.Stop()
nodes[3].Shutdown()
// 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[3].Config.ID].Status.Reachability == api.RaftMemberStatus_REACHABLE {
return fmt.Errorf("expected node 3 to be unreachable")
}
return nil
}))
// Removing node 2 should fail (this would break the quorum)
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)
members := nodes[1].GetMemberlist()
assert.Equal(t, len(members), 3)
// Removing node 3 works fine because it is already unreachable
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)
// Add back node 3
raftutils.ShutdownNode(nodes[3])
delete(nodes, 3)
raftutils.AddRaftNode(t, clockSource, nodes, tc)
// Node 2 and Node 3 should be listed as Reachable
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 reachable")
}
if members[nodes[3].Config.ID].Status.Reachability != api.RaftMemberStatus_REACHABLE {
return fmt.Errorf("expected node 3 to be reachable")
}
return nil
}))
// 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 succeed
ctx, _ = context.WithTimeout(context.Background(), 10*time.Second)
err = nodes[1].RemoveMember(ctx, nodes[2].Config.ID)
assert.NoError(t, err)
members = nodes[1].GetMemberlist()
assert.Nil(t, members[nodes[2].Config.ID])
assert.Equal(t, len(members), 1)
}
func TestRaftSnapshotForceNewCluster(t *testing.T) {
t.Parallel()
// Bring up a 3 node cluster
nodes, clockSource := raftutils.NewRaftCluster(t, tc, &api.RaftConfig{SnapshotInterval: 10, LogEntriesForSlowFollowers: 0})
defer raftutils.TeardownCluster(t, nodes)
nodeIDs := []string{"id1", "id2", "id3", "id4", "id5"}
// Propose 3 values.
for _, nodeID := range nodeIDs[:3] {
_, err := raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeID)
assert.NoError(t, err, "failed to propose value")
}
// Remove one of the original nodes
// 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[2].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[2])
delete(nodes, 2)
// Nodes shouldn't have snapshot files yet
for _, node := range nodes {
dirents, err := ioutil.ReadDir(filepath.Join(node.StateDir, "snap-v3-encrypted"))
assert.NoError(t, err)
assert.Len(t, dirents, 0)
}
// Trigger a snapshot, with a 4th proposal
_, err = raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeIDs[3])
assert.NoError(t, err, "failed to propose value")
// Nodes should now have a snapshot file
for nodeIdx, node := range nodes {
assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
dirents, err := ioutil.ReadDir(filepath.Join(node.StateDir, "snap-v3-encrypted"))
if err != nil {
return err
}
if len(dirents) != 1 {
return fmt.Errorf("expected 1 snapshot, found %d on node %d", len(dirents), nodeIdx+1)
}
return nil
}))
}
// Join another node
nodes[4] = raftutils.NewJoinNode(t, clockSource, nodes[1].Address, tc)
raftutils.WaitForCluster(t, clockSource, nodes)
// Only restart the first node with force-new-cluster option
nodes[1].Server.Stop()
nodes[1].ShutdownRaft()
nodes[1] = raftutils.RestartNode(t, clockSource, nodes[1], true)
delete(nodes, 3)
delete(nodes, 4)
raftutils.WaitForCluster(t, clockSource, nodes)
// The memberlist should contain exactly one node (self)
memberlist := nodes[1].GetMemberlist()
require.Len(t, memberlist, 1)
// Propose a 5th value
_, err = raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeIDs[4])
require.NoError(t, err)
}
