e2e: Migrate duplicate node id test from kurtosis

Author: maru-ava

tests/e2e/e2e.go

@@ -5,11 +5,14 @@
 package e2e
 
 import (
+	"os"
 	"time"
 
 	ginkgo "github.com/onsi/ginkgo/v2"
 	"github.com/stretchr/testify/require"
 
+	cfg "github.com/ava-labs/avalanchego/config"
+	"github.com/ava-labs/avalanchego/tests"
 	"github.com/ava-labs/avalanchego/tests/fixture"
 	"github.com/ava-labs/avalanchego/tests/fixture/testnet"
 	"github.com/ava-labs/avalanchego/tests/fixture/testnet/local"
@@ -24,6 +27,10 @@ const (
 	// Enough for test/custom networks.
 	DefaultConfirmTxTimeout = 20 * time.Second
 
+	// Defines default node bootstrap timeout.
+	// Enough for test/custom networks.
+	DefaultBootstrapTimeout = 20 * time.Second
+
 	DefaultTimeout = 2 * time.Minute
 )
 
@@ -48,3 +55,27 @@ func (te *TestEnvironment) AllocateTestKeys(count int) []*secp256k1.PrivateKey {
 	require.NoError(ginkgo.GinkgoT(), err)
 	return keys
 }
+
+// Add a temporary node that is only intended to be used by a single
+// test. Its ID and URI are not intended to be returned from The
+// Network instance to avoid accessibility from other tests.
+func AddTemporaryNode(network testnet.Network, flags testnet.FlagsMap) testnet.Node {
+	require := require.New(ginkgo.GinkgoT())
+
+	// A temporary location ensures the node won't be accessible from
+	// the Network instance.
+	nodeDir, err := os.MkdirTemp("", "")
+	require.NoError(err)
+
+	flags[cfg.DataDirKey] = nodeDir
+	node, err := network.AddNode(ginkgo.GinkgoWriter, flags)
+	require.NoError(err)
+
+	// Ensure node is stopped and its configuration removed
+	ginkgo.DeferCleanup(func() {
+		tests.Outf("Shutting down temporary node %s\n", node.GetID())
+		require.NoError(node.Remove())
+	})
+
+	return node
+}

tests/e2e/e2e_test.go

@@ -23,6 +23,7 @@ import (
 
 	// ensure test packages are scanned by ginkgo
 	_ "github.com/ava-labs/avalanchego/tests/e2e/banff"
+	_ "github.com/ava-labs/avalanchego/tests/e2e/faultinjection"
 	_ "github.com/ava-labs/avalanchego/tests/e2e/p"
 	_ "github.com/ava-labs/avalanchego/tests/e2e/static-handlers"
 	_ "github.com/ava-labs/avalanchego/tests/e2e/x/transfer"

tests/e2e/faultinjection/duplicate_node_id.go

@@ -0,0 +1,98 @@
+// Copyright (C) 2019-2023, Ava Labs, Inc. All rights reserved.
+// See the file LICENSE for licensing terms.
+
+package faultinjection
+
+import (
+	"context"
+	"time"
+
+	ginkgo "github.com/onsi/ginkgo/v2"
+	"github.com/stretchr/testify/require"
+
+	"github.com/ava-labs/avalanchego/api/info"
+	cfg "github.com/ava-labs/avalanchego/config"
+	"github.com/ava-labs/avalanchego/ids"
+	"github.com/ava-labs/avalanchego/tests/e2e"
+	"github.com/ava-labs/avalanchego/tests/fixture/testnet"
+	"github.com/ava-labs/avalanchego/utils/set"
+)
+
+var _ = ginkgo.Describe("Duplicate node handling", func() {
+	require := require.New(ginkgo.GinkgoT())
+
+	ginkgo.It("should ensure that a given Node ID (i.e. staking keypair) can be used at most once on a network", func() {
+		network := e2e.Env.GetNetwork()
+		nodes := network.GetNodes()
+
+		ginkgo.By("creating new node")
+		newNode1 := e2e.AddTemporaryNode(network, testnet.FlagsMap{})
+		ctx, cancel := context.WithTimeout(context.Background(), e2e.DefaultBootstrapTimeout)
+		defer cancel()
+		require.NoError(newNode1.WaitForHealth(ctx))
+
+		ginkgo.By("checking that the new node is connected to its peers")
+		checkConnectedPeers(nodes, newNode1)
+
+		ginkgo.By("creating a second new node with the same staking keypair as the first new node")
+		node1Flags := newNode1.GetConfig().Flags
+		node2Flags := testnet.FlagsMap{
+			cfg.StakingTLSKeyContentKey: node1Flags[cfg.StakingTLSKeyContentKey],
+			cfg.StakingCertContentKey:   node1Flags[cfg.StakingCertContentKey],
+		}
+		newNode2 := e2e.AddTemporaryNode(network, node2Flags)
+
+		ginkgo.By("checking that the second new node fails to become healthy within timeout")
+		require.Never(func() bool {
+			isHealthy, err := newNode2.IsHealthy(context.Background())
+			require.NoError(err)
+			return isHealthy
+		}, e2e.DefaultBootstrapTimeout, time.Second, "node2 was able to bootstrap")
+
+		ginkgo.By("stopping the first new node")
+		require.NoError(newNode1.Stop())
+
+		ginkgo.By("checking that the second new node becomes healthy within timeout")
+		require.Eventually(func() bool {
+			isHealthy, err := newNode2.IsHealthy(context.Background())
+			require.NoError(err)
+			return isHealthy
+		}, e2e.DefaultBootstrapTimeout, time.Second, "node2 was not able to bootstrap")
+
+		ginkgo.By("checking that the second new node is connected to its peers")
+		checkConnectedPeers(nodes, newNode2)
+	})
+})
+
+// Check that a new node is connected to existing nodes and vice versa
+func checkConnectedPeers(existingNodes []testnet.Node, newNode testnet.Node) {
+	require := require.New(ginkgo.GinkgoT())
+
+	// Collect the node ids of the new node's peers
+	infoClient := info.NewClient(newNode.GetProcessContext().URI)
+	peers, err := infoClient.Peers(context.Background())
+	require.NoError(err)
+	peerIDs := set.NewSet[ids.NodeID](len(existingNodes))
+	for _, peer := range peers {
+		peerIDs.Add(peer.ID)
+	}
+
+	newNodeID := newNode.GetID().String()
+	for _, existingNode := range existingNodes {
+		// Check that the existing node is a peer of the new node
+		require.True(peerIDs.Contains(existingNode.GetID()))
+
+		// Check that the new node is a peer
+		infoClient := info.NewClient(existingNode.GetProcessContext().URI)
+		peers, err := infoClient.Peers(context.Background())
+		require.NoError(err)
+		isPeer := false
+		for _, peer := range peers {
+			if peer.ID.String() == newNodeID {
+				isPeer = true
+				break
+			}
+		}
+		require.True(isPeer)
+	}
+}

tests/fixture/testnet/cmd/main.go

@@ -97,6 +97,32 @@ func main() {
 	}
 	rootCmd.AddCommand(stopNetworkCmd)
 
+	addNodeCmd := &cobra.Command{
+		Use:   "add-node",
+		Short: "Adds a node to a local network",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if len(networkDir) == 0 {
+				return fmt.Errorf("--network-dir or %s are required", local.NetworkDirEnvName)
+			}
+			network, err := local.ReadNetwork(networkDir)
+			if err != nil {
+				return err
+			}
+			node, err := network.AddLocalNode(os.Stdout, nil)
+			if err != nil {
+				return err
+			}
+			ctx, cancel := context.WithTimeout(context.Background(), local.DefaultNodeStartTimeout)
+			defer cancel()
+			if err := node.WaitForHealth(ctx); err != nil {
+				return err
+			}
+			fmt.Fprintf(os.Stdout, "Added node %q to network configured at: %s\n", node.NodeID, network.Dir)
+			return nil
+		},
+	}
+	rootCmd.AddCommand(addNodeCmd)
+
 	if err := rootCmd.Execute(); err != nil {
 		fmt.Fprintf(os.Stderr, "testnetctl failed: %v\n", err)
 		os.Exit(1)

tests/fixture/testnet/interfaces.go

@@ -4,6 +4,9 @@
 package testnet
 
 import (
+	"context"
+	"io"
+
 	"github.com/ava-labs/avalanchego/ids"
 	"github.com/ava-labs/avalanchego/node"
 )
@@ -12,11 +15,16 @@ import (
 type Network interface {
 	GetConfig() NetworkConfig
 	GetNodes() []Node
+	AddNode(w io.Writer, flags FlagsMap) (Node, error)
 }
 
 // Defines node capabilities supportable regardless of how a network is orchestrated.
 type Node interface {
 	GetID() ids.NodeID
 	GetConfig() NodeConfig
 	GetProcessContext() node.NodeProcessContext
+	IsHealthy(ctx context.Context) (bool, error)
+	WaitForHealth(ctx context.Context) error
+	Stop() error
+	Remove() error
 }

tests/fixture/testnet/local/config.go

@@ -18,6 +18,7 @@ const (
 	RootDirEnvName         = "TESTNETCTL_ROOT_DIR"
 
 	DefaultNetworkStartTimeout = 2 * time.Minute
+	DefaultNodeStartTimeout    = 30 * time.Second
 
 	DefaultNodeStartTimeout = 10 * time.Second
 

tests/fixture/testnet/local/network.go

@@ -13,6 +13,7 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 	"time"
 
 	"github.com/spf13/cast"
@@ -93,6 +94,22 @@ func (ln *LocalNetwork) GetNodes() []testnet.Node {
 	return nodes
 }
 
+// Adds a backend-agnostic node to the network
+func (ln *LocalNetwork) AddNode(w io.Writer, flags testnet.FlagsMap) (testnet.Node, error) {
+	if flags == nil {
+		flags = testnet.FlagsMap{}
+	}
+	node, err := ln.AddLocalNode(w, &LocalNode{
+		NodeConfig: testnet.NodeConfig{
+			Flags: flags,
+		},
+	})
+	if err != nil {
+		return nil, err
+	}
+	return node, nil
+}
+
 // Starts a new network stored under the provided root dir. Required
 // configuration will be defaulted if not provided.
 func StartNetwork(
@@ -656,3 +673,67 @@ func (ln *LocalNetwork) ReadAll() error {
 	}
 	return ln.ReadNodes()
 }
+
+func (ln *LocalNetwork) AddLocalNode(w io.Writer, node *LocalNode) (*LocalNode, error) {
+	// Assume network configuration has been written to disk and is current in memory
+
+	if err := ln.ReadNodes(); err != nil {
+		return nil, err
+	}
+
+	maxPort := 0
+	bootstrapIPs := []string{}
+	bootstrapIDs := []string{}
+	for _, node := range ln.Nodes {
+		if len(node.StakingAddress) == 0 {
+			// Node is not running
+			continue
+		}
+
+		bootstrapIPs = append(bootstrapIPs, node.StakingAddress)
+		bootstrapIDs = append(bootstrapIDs, node.NodeID.String())
+
+		// Find the maximum port if using static ports
+		if ln.UseStaticPorts {
+			addressParts := strings.Split(node.StakingAddress, ":")
+			rawPort := addressParts[len(addressParts)-1]
+			port, err := strconv.Atoi(rawPort)
+			if err != nil {
+				return nil, err
+			}
+			if port > maxPort {
+				maxPort = port
+			}
+		}
+	}
+
+	if len(bootstrapIDs) == 0 {
+		return nil, errors.New("failed to add node due to missing bootstrap nodes")
+	}
+
+	if node == nil {
+		node = NewLocalNode("")
+	}
+
+	if err := ln.PopulateNodeConfig(node); err != nil {
+		return nil, err
+	}
+
+	// Configure the provided node for this network
+	httpPort := 0
+	stakingPort := 0
+	if ln.UseStaticPorts {
+		httpPort = maxPort
+		stakingPort = maxPort + 1
+	}
+	node.SetNetworkingConfig(httpPort, stakingPort, bootstrapIDs, bootstrapIPs)
+
+	if err := node.WriteConfig(); err != nil {
+		return nil, err
+	}
+	if err := node.Start(w, ln.ExecPath); err != nil {
+		return nil, err
+	}
+
+	return node, nil
+}

tests/fixture/testnet/local/node.go

@@ -272,6 +272,22 @@ func (n *LocalNode) IsHealthy(ctx context.Context) (bool, error) {
 	return health.Healthy, nil
 }
 
+func (n *LocalNode) WaitForHealth(ctx context.Context) error {
+	for {
+		select {
+		case <-ctx.Done():
+			return errors.New("failed to see node health before timeout")
+		default:
+			if healthy, err := n.IsHealthy(ctx); err != nil {
+				return err
+			} else if healthy {
+				return nil
+			}
+		}
+		time.Sleep(200 * time.Millisecond)
+	}
+}
+
 func (n *LocalNode) WaitForProcessContext(ctx context.Context) error {
 	ctx, cancel := context.WithTimeout(ctx, DefaultNodeStartTimeout)
 	defer cancel()
@@ -293,3 +309,15 @@ func (n *LocalNode) WaitForProcessContext(ctx context.Context) error {
 	}
 	return nil
 }
+
+// Ensure the node is stopped and its configuration removed.
+func (n *LocalNode) Remove() error {
+	// Ensure the node is stopped before removing its configuration
+	if err := n.Stop(); err != nil {
+		return fmt.Errorf("failed to stop node %s: %w", n.NodeID, err)
+	}
+	if err := os.RemoveAll(n.GetDataDir()); err != nil {
+		return fmt.Errorf("failed to remove configuration for node %s: %w", n.NodeID, err)
+	}
+	return nil
+}