sequencer_failover_setup.go

package op_e2e

import (
	"context"
	"fmt"
	"math/rand"
	"net"
	"strings"
	"testing"
	"time"

	"github.com/ethereum/go-ethereum/common/hexutil"
	"github.com/ethereum/go-ethereum/log"
	"github.com/ethereum/go-ethereum/rpc"
	"github.com/stretchr/testify/require"

	bss "github.com/ethereum-optimism/optimism/op-batcher/batcher"
	batcherFlags "github.com/ethereum-optimism/optimism/op-batcher/flags"
	con "github.com/ethereum-optimism/optimism/op-conductor/conductor"
	"github.com/ethereum-optimism/optimism/op-conductor/consensus"
	conrpc "github.com/ethereum-optimism/optimism/op-conductor/rpc"
	"github.com/ethereum-optimism/optimism/op-e2e/e2eutils/wait"
	rollupNode "github.com/ethereum-optimism/optimism/op-node/node"
	"github.com/ethereum-optimism/optimism/op-node/rollup"
	"github.com/ethereum-optimism/optimism/op-node/rollup/derive"
	"github.com/ethereum-optimism/optimism/op-node/rollup/driver"
	"github.com/ethereum-optimism/optimism/op-node/rollup/sync"
	oplog "github.com/ethereum-optimism/optimism/op-service/log"
	"github.com/ethereum-optimism/optimism/op-service/retry"
	oprpc "github.com/ethereum-optimism/optimism/op-service/rpc"
	"github.com/ethereum-optimism/optimism/op-service/sources"
	"github.com/ethereum-optimism/optimism/op-service/testlog"
)

const (
	Sequencer1Name = "sequencer1"
	Sequencer2Name = "sequencer2"
	Sequencer3Name = "sequencer3"
	VerifierName   = "verifier"

	localhost = "127.0.0.1"

	maxSetupRetries = 5
)

var retryStrategy = &retry.FixedStrategy{Dur: 50 * time.Millisecond}

type conductor struct {
	service       *con.OpConductor
	client        conrpc.API
	consensusPort int
	rpcPort       int
}

func (c *conductor) ConsensusEndpoint() string {
	return fmt.Sprintf("%s:%d", localhost, c.consensusPort)
}

func (c *conductor) RPCEndpoint() string {
	return fmt.Sprintf("http://%s:%d", localhost, c.rpcPort)
}

func setupSequencerFailoverTest(t *testing.T) (*System, map[string]*conductor, func()) {
	InitParallel(t)
	ctx := context.Background()

	sys, conductors, err := retry.Do2(ctx, maxSetupRetries, retryStrategy, func() (*System, map[string]*conductor, error) {
		return setupHAInfra(t, ctx)
	})
	require.NoError(t, err, "Expected to successfully setup sequencers and conductors after retry")

	// form a cluster
	c1 := conductors[Sequencer1Name]
	c2 := conductors[Sequencer2Name]
	c3 := conductors[Sequencer3Name]

	require.NoError(t, waitForLeadership(t, c1))
	require.NoError(t, c1.client.AddServerAsVoter(ctx, Sequencer2Name, c2.ConsensusEndpoint(), 0))
	require.NoError(t, c1.client.AddServerAsVoter(ctx, Sequencer3Name, c3.ConsensusEndpoint(), 0))
	require.True(t, leader(t, ctx, c1))
	require.False(t, leader(t, ctx, c2))
	require.False(t, leader(t, ctx, c3))

	// start sequencing on leader
	lid, _ := findLeader(t, conductors)
	unsafeHead, err := sys.Clients[lid].BlockByNumber(ctx, nil)
	require.NoError(t, err)
	require.Equal(t, uint64(0), unsafeHead.NumberU64())
	require.NoError(t, sys.RollupClient(lid).StartSequencer(ctx, unsafeHead.Hash()))

	// 1 batcher that listens to all 3 sequencers, in started mode.
	setupBatcher(t, sys, conductors)

	// weirdly, batcher does not submit a batch until unsafe block 9.
	// It became normal after that and submits a batch every L1 block (2s) per configuration.
	// Since our health monitor checks on safe head progression, wait for batcher to become normal before proceeding.
	_, err = wait.ForNextSafeBlock(ctx, sys.Clients[Sequencer1Name])
	require.NoError(t, err)
	_, err = wait.ForNextSafeBlock(ctx, sys.Clients[Sequencer1Name])
	require.NoError(t, err)

	// make sure conductor reports all sequencers as healthy, this means they're syncing correctly.
	require.Eventually(t, func() bool {
		return healthy(t, ctx, c1) &&
			healthy(t, ctx, c2) &&
			healthy(t, ctx, c3)
	}, 50*time.Second, 500*time.Millisecond, "Expected sequencers to become healthy")

	// unpause all conductors
	require.NoError(t, c1.client.Resume(ctx))
	require.NoError(t, c2.client.Resume(ctx))
	require.NoError(t, c3.client.Resume(ctx))

	// final check, make sure everything is in the right place
	require.True(t, conductorResumed(t, ctx, c1))
	require.True(t, conductorResumed(t, ctx, c2))
	require.True(t, conductorResumed(t, ctx, c3))
	require.False(t, conductorStopped(t, ctx, c1))
	require.False(t, conductorStopped(t, ctx, c2))
	require.False(t, conductorStopped(t, ctx, c3))
	require.True(t, conductorActive(t, ctx, c1))
	require.True(t, conductorActive(t, ctx, c2))
	require.True(t, conductorActive(t, ctx, c3))

	require.True(t, sequencerActive(t, ctx, sys.RollupClient(Sequencer1Name)))
	require.False(t, sequencerActive(t, ctx, sys.RollupClient(Sequencer2Name)))
	require.False(t, sequencerActive(t, ctx, sys.RollupClient(Sequencer3Name)))

	require.True(t, healthy(t, ctx, c1))
	require.True(t, healthy(t, ctx, c2))
	require.True(t, healthy(t, ctx, c3))

	return sys, conductors, func() {
		sys.Close()
		for _, c := range conductors {
			_ = c.service.Stop(ctx)
		}
	}
}

func setupHAInfra(t *testing.T, ctx context.Context) (*System, map[string]*conductor, error) {
	startTime := time.Now()

	var sys *System
	var conductors map[string]*conductor
	var err error

	// clean up if setup fails due to port in use.
	defer func() {
		if err != nil {
			if sys != nil {
				sys.Close()
			}

			for _, c := range conductors {
				if c == nil || c.service == nil {
					// pass. Sometimes we can get nil in this map
				} else if serr := c.service.Stop(ctx); serr != nil {
					t.Log("Failed to stop conductor", "error", serr)
				}
			}
		}
		t.Logf("setupHAInfra took %s\n", time.Since(startTime))
	}()

	conductorRpcPorts := map[string]int{
		Sequencer1Name: findAvailablePort(t),
		Sequencer2Name: findAvailablePort(t),
		Sequencer3Name: findAvailablePort(t),
	}

	// 3 sequencers, 1 verifier, 1 active sequencer.
	cfg := sequencerFailoverSystemConfig(t, conductorRpcPorts)
	if sys, err = cfg.Start(t); err != nil {
		return nil, nil, err
	}

	// 3 conductors that connects to 1 sequencer each.
	conductors = make(map[string]*conductor)

	// initialize all conductors in paused mode
	conductorCfgs := []struct {
		name      string
		port      int
		bootstrap bool
	}{
		{Sequencer1Name, conductorRpcPorts[Sequencer1Name], true}, // one in bootstrap mode so that we can form a cluster.
		{Sequencer2Name, conductorRpcPorts[Sequencer2Name], false},
		{Sequencer3Name, conductorRpcPorts[Sequencer3Name], false},
	}
	for _, cfg := range conductorCfgs {
		cfg := cfg
		nodePRC := sys.RollupNodes[cfg.name].HTTPEndpoint()
		engineRPC := sys.EthInstances[cfg.name].HTTPEndpoint()
		if conductors[cfg.name], err = setupConductor(t, cfg.name, t.TempDir(), nodePRC, engineRPC, cfg.port, cfg.bootstrap, *sys.RollupConfig); err != nil {
			return nil, nil, err
		}
	}

	return sys, conductors, nil
}

func setupConductor(
	t *testing.T,
	serverID, dir, nodeRPC, engineRPC string,
	rpcPort int,
	bootstrap bool,
	rollupCfg rollup.Config,
) (*conductor, error) {
	consensusPort := findAvailablePort(t)
	cfg := con.Config{
		ConsensusAddr:  localhost,
		ConsensusPort:  consensusPort,
		RaftServerID:   serverID,
		RaftStorageDir: dir,
		RaftBootstrap:  bootstrap,
		NodeRPC:        nodeRPC,
		ExecutionRPC:   engineRPC,
		Paused:         true,
		HealthCheck: con.HealthCheckConfig{
			Interval:     1, // per test setup, l2 block time is 1s.
			MinPeerCount: 2, // per test setup, each sequencer has 2 peers
			// CI is unstable in terms of the delay between now and the head time
			// so we set the unsafe interval to 30s to avoid flakiness.
			// This is fine because there's a progression check within health monitor to check progression.
			UnsafeInterval: 30,
			SafeInterval:   30,
		},
		RollupCfg:      rollupCfg,
		RPCEnableProxy: true,
		LogConfig: oplog.CLIConfig{
			Level: log.LevelInfo,
			Color: false,
		},
		RPC: oprpc.CLIConfig{
			ListenAddr: localhost,
			ListenPort: rpcPort,
		},
	}

	ctx := context.Background()
	service, err := con.New(ctx, &cfg, testlog.Logger(t, log.LevelInfo), "0.0.1")
	if err != nil {
		return nil, err
	}

	err = service.Start(ctx)
	if err != nil {
		return nil, err
	}

	rawClient, err := rpc.DialContext(ctx, service.HTTPEndpoint())
	if err != nil {
		return nil, err
	}
	client := conrpc.NewAPIClient(rawClient)

	return &conductor{
		service:       service,
		client:        client,
		consensusPort: consensusPort,
		rpcPort:       rpcPort,
	}, nil
}

func setupBatcher(t *testing.T, sys *System, conductors map[string]*conductor) {
	// enable active sequencer follow mode.
	// in sequencer HA, all batcher / proposer requests will be proxied by conductor so that we can make sure
	// that requests are always handled by leader.
	l2EthRpc := strings.Join([]string{
		conductors[Sequencer1Name].RPCEndpoint(),
		conductors[Sequencer2Name].RPCEndpoint(),
		conductors[Sequencer3Name].RPCEndpoint(),
	}, ",")
	rollupRpc := strings.Join([]string{
		conductors[Sequencer1Name].RPCEndpoint(),
		conductors[Sequencer2Name].RPCEndpoint(),
		conductors[Sequencer3Name].RPCEndpoint(),
	}, ",")
	batcherCLIConfig := &bss.CLIConfig{
		L1EthRpc:               sys.EthInstances["l1"].WSEndpoint(),
		L2EthRpc:               l2EthRpc,
		RollupRpc:              rollupRpc,
		MaxPendingTransactions: 0,
		MaxChannelDuration:     1,
		MaxL1TxSize:            120_000,
		TargetNumFrames:        1,
		ApproxComprRatio:       0.4,
		SubSafetyMargin:        4,
		PollInterval:           1 * time.Second,
		TxMgrConfig:            newTxMgrConfig(sys.EthInstances["l1"].WSEndpoint(), sys.Cfg.Secrets.Batcher),
		LogConfig: oplog.CLIConfig{
			Level:  log.LevelDebug,
			Format: oplog.FormatText,
		},
		Stopped:                      false,
		BatchType:                    derive.SpanBatchType,
		DataAvailabilityType:         batcherFlags.CalldataType,
		ActiveSequencerCheckDuration: 0,
		CompressionAlgo:              derive.Zlib,
	}

	batcher, err := bss.BatcherServiceFromCLIConfig(context.Background(), "0.0.1", batcherCLIConfig, sys.Cfg.Loggers["batcher"])
	require.NoError(t, err)
	err = batcher.Start(context.Background())
	require.NoError(t, err)
	sys.BatchSubmitter = batcher
}

func sequencerFailoverSystemConfig(t *testing.T, ports map[string]int) SystemConfig {
	cfg := DefaultSystemConfig(t)
	delete(cfg.Nodes, "sequencer")
	cfg.Nodes[Sequencer1Name] = sequencerCfg(ports[Sequencer1Name])
	cfg.Nodes[Sequencer2Name] = sequencerCfg(ports[Sequencer2Name])
	cfg.Nodes[Sequencer3Name] = sequencerCfg(ports[Sequencer3Name])

	delete(cfg.Loggers, "sequencer")
	cfg.Loggers[Sequencer1Name] = testlog.Logger(t, log.LevelInfo).New("role", Sequencer1Name)
	cfg.Loggers[Sequencer2Name] = testlog.Logger(t, log.LevelInfo).New("role", Sequencer2Name)
	cfg.Loggers[Sequencer3Name] = testlog.Logger(t, log.LevelInfo).New("role", Sequencer3Name)

	cfg.P2PTopology = map[string][]string{
		Sequencer1Name: {Sequencer2Name, Sequencer3Name},
		Sequencer2Name: {Sequencer3Name, VerifierName},
		Sequencer3Name: {VerifierName, Sequencer1Name},
		VerifierName:   {Sequencer1Name, Sequencer2Name},
	}
	offset := hexutil.Uint64(0)
	cfg.DeployConfig.L2GenesisDeltaTimeOffset = &offset
	cfg.DeployConfig.L2GenesisEcotoneTimeOffset = &offset

	return cfg
}

func sequencerCfg(rpcPort int) *rollupNode.Config {
	return &rollupNode.Config{
		Driver: driver.Config{
			VerifierConfDepth:  0,
			SequencerConfDepth: 0,
			SequencerEnabled:   true,
			SequencerStopped:   true,
		},
		// Submitter PrivKey is set in system start for rollup nodes where sequencer = true
		RPC: rollupNode.RPCConfig{
			ListenAddr:  localhost,
			ListenPort:  0,
			EnableAdmin: true,
		},
		L1EpochPollInterval:         time.Second * 2,
		RuntimeConfigReloadInterval: time.Minute * 10,
		ConfigPersistence:           &rollupNode.DisabledConfigPersistence{},
		Sync:                        sync.Config{SyncMode: sync.CLSync},
		ConductorEnabled:            true,
		ConductorRpc:                fmt.Sprintf("http://%s:%d", localhost, rpcPort),
		ConductorRpcTimeout:         1 * time.Second,
	}
}

func waitForLeadership(t *testing.T, c *conductor) error {
	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	condition := func() (bool, error) {
		isLeader, err := c.client.Leader(ctx)
		if err != nil {
			return false, err
		}
		return isLeader, nil
	}

	return wait.For(ctx, 1*time.Second, condition)
}

func waitForLeadershipChange(t *testing.T, prev *conductor, prevID string, conductors map[string]*conductor, sys *System) string {
	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	condition := func() (bool, error) {
		isLeader, err := prev.client.Leader(ctx)
		if err != nil {
			return false, err
		}
		return !isLeader, nil
	}

	err := wait.For(ctx, 1*time.Second, condition)
	require.NoError(t, err)

	ensureOnlyOneLeader(t, sys, conductors)
	newLeader, err := prev.client.LeaderWithID(ctx)
	require.NoError(t, err)
	require.NotEmpty(t, newLeader.ID)
	require.NotEqual(t, prevID, newLeader.ID, "Expected a new leader")
	require.NoError(t, waitForSequencerStatusChange(t, sys.RollupClient(newLeader.ID), true))

	return newLeader.ID
}

func waitForSequencerStatusChange(t *testing.T, rollupClient *sources.RollupClient, active bool) error {
	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	condition := func() (bool, error) {
		isActive, err := rollupClient.SequencerActive(ctx)
		if err != nil {
			return false, err
		}
		return isActive == active, nil
	}

	return wait.For(ctx, 1*time.Second, condition)
}

func leader(t *testing.T, ctx context.Context, con *conductor) bool {
	leader, err := con.client.Leader(ctx)
	require.NoError(t, err)
	return leader
}

func healthy(t *testing.T, ctx context.Context, con *conductor) bool {
	healthy, err := con.client.SequencerHealthy(ctx)
	require.NoError(t, err)
	return healthy
}

func conductorActive(t *testing.T, ctx context.Context, con *conductor) bool {
	active, err := con.client.Active(ctx)
	require.NoError(t, err)
	return active
}

func conductorResumed(t *testing.T, ctx context.Context, con *conductor) bool {
	paused, err := con.client.Paused(ctx)
	require.NoError(t, err)
	return !paused
}

func conductorStopped(t *testing.T, ctx context.Context, con *conductor) bool {
	stopped, err := con.client.Stopped(ctx)
	require.NoError(t, err)
	return stopped
}

func sequencerActive(t *testing.T, ctx context.Context, rollupClient *sources.RollupClient) bool {
	active, err := rollupClient.SequencerActive(ctx)
	require.NoError(t, err)
	return active
}

func findAvailablePort(t *testing.T) int {
	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
	defer cancel()
	for {
		select {
		case <-ctx.Done():
			t.Error("Failed to find available port")
		default:
			// private / ephemeral ports are in the range 49152-65535
			port := rand.Intn(65535-49152) + 49152
			addr := fmt.Sprintf("127.0.0.1:%d", port)
			l, err := net.Listen("tcp", addr)
			if err == nil {
				l.Close() // Close the listener and return the port if it's available
				return port
			}
		}
	}
}

func findLeader(t *testing.T, conductors map[string]*conductor) (string, *conductor) {
	for id, con := range conductors {
		if leader(t, context.Background(), con) {
			return id, con
		}
	}
	return "", nil
}

func findFollower(t *testing.T, conductors map[string]*conductor) (string, *conductor) {
	for id, con := range conductors {
		if !leader(t, context.Background(), con) {
			return id, con
		}
	}
	return "", nil
}

func ensureOnlyOneLeader(t *testing.T, sys *System, conductors map[string]*conductor) {
	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	condition := func() (bool, error) {
		leaders := 0
		for name, con := range conductors {
			leader, err := con.client.Leader(ctx)
			if err != nil {
				continue
			}
			active, err := sys.RollupClient(name).SequencerActive(ctx)
			if err != nil {
				continue
			}

			if leader && active {
				leaders++
			}
		}
		return leaders == 1, nil
	}
	require.NoError(t, wait.For(ctx, 1*time.Second, condition))
}

func memberIDs(membership *consensus.ClusterMembership) []string {
	ids := make([]string, len(membership.Servers))
	for _, member := range membership.Servers {
		ids = append(ids, member.ID)
	}
	return ids
}