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
}))
}
}
// 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 TestStress(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)
// number of nodes that are running
nup := len(nodes)
// record of nodes that are down
idleNodes := map[int]struct{}{}
// record of ids that proposed successfully or time-out
pIDs := []string{}
leader := -1
for iters := 0; iters < 1000; iters++ {
// keep proposing new values and killing leader
for i := 1; i <= 5; i++ {
if nodes[uint64(i)] != nil {
id := strconv.Itoa(iters)
_, err := raftutils.ProposeValue(t, nodes[uint64(i)], id)
if err == nil {
pIDs = append(pIDs, id)
// if propose successfully, at least there are 3 running nodes
assert.True(t, nup >= 3)
// only leader can propose value
assert.True(t, leader == i || leader == -1)
// update leader
leader = i
break
} else if strings.Contains(err.Error(), "context deadline exceeded") {
// though it's timing out, we still record this value
// for it may be proposed successfully and stored in Raft some time later
pIDs = append(pIDs, id)
}
}
}
if rand.Intn(100) < 10 {
// increase clock to make potential election finish quickly
clockSource.Increment(200 * time.Millisecond)
time.Sleep(10 * time.Millisecond)
} else {
ms := rand.Intn(10)
clockSource.Increment(time.Duration(ms) * time.Millisecond)
}
if leader != -1 {
// if propose successfully, try to kill a node in random
s := rand.Intn(5) + 1
if _, ok := idleNodes[s]; !ok {
id := uint64(s)
nodes[id].Server.Stop()
nodes[id].Shutdown()
idleNodes[s] = struct{}{}
nup -= 1
if s == leader {
// leader is killed
leader = -1
}
}
}
if nup < 3 {
// if quorum is lost, try to bring back a node
s := rand.Intn(5) + 1
if _, ok := idleNodes[s]; ok {
id := uint64(s)
nodes[id] = raftutils.RestartNode(t, clockSource, nodes[id], false)
delete(idleNodes, s)
nup++
}
}
}
// bring back all nodes and propose the final value
for i := range idleNodes {
id := uint64(i)
nodes[id] = raftutils.RestartNode(t, clockSource, nodes[id], false)
}
raftutils.WaitForCluster(t, clockSource, nodes)
id := strconv.Itoa(1000)
val, err := raftutils.ProposeValue(t, raftutils.Leader(nodes), id)
assert.NoError(t, err, "failed to propose value")
pIDs = append(pIDs, id)
// increase clock to make cluster stable
time.Sleep(500 * time.Millisecond)
clockSource.Increment(500 * time.Millisecond)
ids, values := raftutils.GetAllValuesOnNode(t, clockSource, nodes[1])
// since cluster is stable, final value must be in the raft store
find := false
for _, value := range values {
if reflect.DeepEqual(value, val) {
find = true
break
}
}
assert.True(t, find)
// all nodes must have the same value
raftutils.CheckValuesOnNodes(t, clockSource, nodes, ids, values)
// ids should be a subset of pIDs
for _, id := range ids {
find = false
for _, pid := range pIDs {
if id == pid {
find = true
break
}
}
assert.True(t, find)
}
}
func TestRaftForceNewCluster(t *testing.T) {
t.Parallel()
nodes, clockSource := raftutils.NewRaftCluster(t, tc)
// 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")
// The memberlist should contain 3 members on each node
for i := 1; i <= 3; i++ {
assert.Equal(t, len(nodes[uint64(i)].GetMemberlist()), 3)
}
// Stop all nodes
for _, node := range nodes {
node.Server.Stop()
node.Shutdown()
}
raftutils.AdvanceTicks(clockSource, 5)
toClean := map[uint64]*raftutils.TestNode{
2: nodes[2],
3: nodes[3],
}
raftutils.TeardownCluster(t, toClean)
delete(nodes, 2)
delete(nodes, 3)
// Only restart the first node with force-new-cluster option
nodes[1] = raftutils.RestartNode(t, clockSource, nodes[1], true)
raftutils.WaitForCluster(t, clockSource, nodes)
// The memberlist should contain only one node (self)
assert.Equal(t, len(nodes[1].GetMemberlist()), 1)
// Add 2 more members
nodes[2] = raftutils.NewJoinNode(t, clockSource, nodes[1].Address, tc)
raftutils.WaitForCluster(t, clockSource, nodes)
nodes[3] = raftutils.NewJoinNode(t, clockSource, nodes[1].Address, tc)
raftutils.WaitForCluster(t, clockSource, nodes)
newCluster := map[uint64]*raftutils.TestNode{
1: nodes[1],
2: nodes[2],
3: nodes[3],
}
defer raftutils.TeardownCluster(t, newCluster)
// The memberlist should contain 3 members on each node
for i := 1; i <= 3; i++ {
assert.Equal(t, len(nodes[uint64(i)].GetMemberlist()), 3)
}
// 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 TestGCWAL(t *testing.T) {
t.Parallel()
// Additional log entries from cluster setup, leader election
extraLogEntries := 5
// Number of large entries to propose
proposals := 8
// Bring up a 3 node cluster
nodes, clockSource := raftutils.NewRaftCluster(t, tc, &api.RaftConfig{SnapshotInterval: uint64(proposals + extraLogEntries), LogEntriesForSlowFollowers: 0})
for i := 0; i != proposals; i++ {
_, err := proposeLargeValue(t, nodes[1], DefaultProposalTime, fmt.Sprintf("id%d", i))
assert.NoError(t, err, "failed to propose value")
}
time.Sleep(250 * time.Millisecond)
// Snapshot should have been triggered just as the WAL rotated, so
// both WAL files should be preserved
assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
dirents, err := ioutil.ReadDir(filepath.Join(nodes[1].StateDir, "snap"))
if err != nil {
return err
}
if len(dirents) != 1 {
return fmt.Errorf("expected 1 snapshot, found %d", len(dirents))
}
dirents, err = ioutil.ReadDir(filepath.Join(nodes[1].StateDir, "wal"))
if err != nil {
return err
}
var walCount int
for _, f := range dirents {
if strings.HasSuffix(f.Name(), ".wal") {
walCount++
}
}
if walCount != 2 {
return fmt.Errorf("expected 2 WAL files, found %d", walCount)
}
return nil
}))
raftutils.TeardownCluster(t, nodes)
// Repeat this test, but trigger the snapshot after the WAL has rotated
proposals++
nodes, clockSource = raftutils.NewRaftCluster(t, tc, &api.RaftConfig{SnapshotInterval: uint64(proposals + extraLogEntries), LogEntriesForSlowFollowers: 0})
defer raftutils.TeardownCluster(t, nodes)
for i := 0; i != proposals; i++ {
_, err := proposeLargeValue(t, nodes[1], DefaultProposalTime, fmt.Sprintf("id%d", i))
assert.NoError(t, err, "failed to propose value")
}
time.Sleep(250 * time.Millisecond)
// This time only one WAL file should be saved.
assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
dirents, err := ioutil.ReadDir(filepath.Join(nodes[1].StateDir, "snap"))
if err != nil {
return err
}
if len(dirents) != 1 {
return fmt.Errorf("expected 1 snapshot, found %d", len(dirents))
}
dirents, err = ioutil.ReadDir(filepath.Join(nodes[1].StateDir, "wal"))
if err != nil {
return err
}
var walCount int
for _, f := range dirents {
if strings.HasSuffix(f.Name(), ".wal") {
walCount++
}
}
if walCount != 1 {
return fmt.Errorf("expected 1 WAL file, found %d", walCount)
}
return nil
}))
// Restart the whole cluster
for _, node := range nodes {
node.Server.Stop()
node.Shutdown()
}
raftutils.AdvanceTicks(clockSource, 5)
i := 0
for k, node := range nodes {
nodes[k] = raftutils.RestartNode(t, clockSource, node, false)
i++
}
raftutils.WaitForCluster(t, clockSource, nodes)
// Is the data intact after restart?
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) != proposals {
err = fmt.Errorf("expected %d nodes, got %d", proposals, len(allNodes))
return
}
})
return err
}))
}
// It should still be possible to propose values
_, err := raftutils.ProposeValue(t, raftutils.Leader(nodes), DefaultProposalTime, "newnode")
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) != proposals+1 {
err = fmt.Errorf("expected %d nodes, got %d", proposals, len(allNodes))
return
}
})
return err
}))
}
}
func TestRaftSnapshotRestart(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", "id6", "id7"}
values := make([]*api.Node, len(nodeIDs))
// Propose 3 values
var err error
for i, nodeID := range nodeIDs[:3] {
values[i], err = raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeID)
assert.NoError(t, err, "failed to propose value")
}
// Take down node 3
nodes[3].Server.Stop()
nodes[3].Shutdown()
// Propose a 4th value before the snapshot
values[3], err = raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeIDs[3])
assert.NoError(t, err, "failed to propose value")
// Remaining nodes shouldn't have snapshot files yet
for _, node := range []*raftutils.TestNode{nodes[1], nodes[2]} {
dirents, err := ioutil.ReadDir(filepath.Join(node.StateDir, "snap"))
assert.NoError(t, err)
assert.Len(t, dirents, 0)
}
// Add a node to the cluster before the snapshot. This is the event
// that triggers the snapshot.
nodes[4] = raftutils.NewJoinNode(t, clockSource, nodes[1].Address, tc)
raftutils.WaitForCluster(t, clockSource, map[uint64]*raftutils.TestNode{1: nodes[1], 2: nodes[2], 4: nodes[4]})
// Remaining nodes should now have a snapshot file
for nodeIdx, node := range []*raftutils.TestNode{nodes[1], nodes[2]} {
assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
dirents, err := ioutil.ReadDir(filepath.Join(node.StateDir, "snap"))
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
}))
}
raftutils.CheckValuesOnNodes(t, clockSource, map[uint64]*raftutils.TestNode{1: nodes[1], 2: nodes[2]}, nodeIDs[:4], values[:4])
// Propose a 5th value
values[4], err = raftutils.ProposeValue(t, nodes[1], DefaultProposalTime, nodeIDs[4])
require.NoError(t, err)
// Add another node to the cluster
nodes[5] = raftutils.NewJoinNode(t, clockSource, nodes[1].Address, tc)
raftutils.WaitForCluster(t, clockSource, map[uint64]*raftutils.TestNode{1: nodes[1], 2: nodes[2], 4: nodes[4], 5: nodes[5]})
// New node should get a copy of the snapshot
assert.NoError(t, raftutils.PollFunc(clockSource, func() error {
dirents, err := ioutil.ReadDir(filepath.Join(nodes[5].StateDir, "snap"))
if err != nil {
return err
}
if len(dirents) != 1 {
return fmt.Errorf("expected 1 snapshot, found %d on new node", len(dirents))
}
return nil
}))
dirents, err := ioutil.ReadDir(filepath.Join(nodes[5].StateDir, "snap"))
assert.NoError(t, err)
assert.Len(t, dirents, 1)
raftutils.CheckValuesOnNodes(t, clockSource, map[uint64]*raftutils.TestNode{1: nodes[1], 2: nodes[2]}, nodeIDs[:5], values[:5])
// It should know about the other nodes, including the one that was just added
stripMembers := func(memberList map[uint64]*api.RaftMember) map[uint64]*api.RaftMember {
raftNodes := make(map[uint64]*api.RaftMember)
for k, v := range memberList {
raftNodes[k] = &api.RaftMember{
RaftID: v.RaftID,
Addr: v.Addr,
}
}
return raftNodes
}
assert.Equal(t, stripMembers(nodes[1].GetMemberlist()), stripMembers(nodes[4].GetMemberlist()))
// Restart node 3
nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)
raftutils.WaitForCluster(t, clockSource, nodes)
// Node 3 should know about other nodes, including the new one
assert.Len(t, nodes[3].GetMemberlist(), 5)
assert.Equal(t, stripMembers(nodes[1].GetMemberlist()), stripMembers(nodes[3].GetMemberlist()))
// Propose yet another value, to make sure the rejoined node is still
// receiving new logs
values[5], err = raftutils.ProposeValue(t, raftutils.Leader(nodes), DefaultProposalTime, nodeIDs[5])
require.NoError(t, err)
// All nodes should have all the data
raftutils.CheckValuesOnNodes(t, clockSource, nodes, nodeIDs[:6], values[:6])
// Restart node 3 again. It should load the snapshot.
nodes[3].Server.Stop()
nodes[3].Shutdown()
nodes[3] = raftutils.RestartNode(t, clockSource, nodes[3], false)
raftutils.WaitForCluster(t, clockSource, nodes)
assert.Len(t, nodes[3].GetMemberlist(), 5)
assert.Equal(t, stripMembers(nodes[1].GetMemberlist()), stripMembers(nodes[3].GetMemberlist()))
raftutils.CheckValuesOnNodes(t, clockSource, nodes, nodeIDs[:6], values[:6])
// Propose again. Just to check consensus after this latest restart.
values[6], err = raftutils.ProposeValue(t, raftutils.Leader(nodes), DefaultProposalTime, nodeIDs[6])
require.NoError(t, err)
raftutils.CheckValuesOnNodes(t, clockSource, nodes, nodeIDs, values)
}
