WIP Migrate duplicate test e2e from avalanche-testing

Author: maru-ava

tests/e2e/e2e.go

@@ -99,6 +99,17 @@ func (te *TestEnvironment) Setup(
 		return err
 	}
 
+	// Need to always configure avalanchego.  Even if using a
+	// persistent network, avalanchego is required for operations like
+	// node addition.
+	if avalancheGoExecPath != "" {
+		if _, err := os.Stat(avalancheGoExecPath); err != nil {
+			return fmt.Errorf("could not find avalanchego binary: %w", err)
+		}
+	}
+	te.avalancheGoExecPath = avalancheGoExecPath
+	te.avalancheGoLogLevel = avalancheGoLogLevel
+
 	// TODO(marun) Maybe lazy-load so that errors only fail dependent tests?
 	err = te.LoadKeys(testKeysFile)
 	if err != nil {
@@ -111,25 +122,13 @@ func (te *TestEnvironment) Setup(
 		tests.Outf("{{yellow}}Using a pre-existing network{{/}}\n")
 
 		// Read the URIs for the existing network so that tests can access the nodes.
-
 		err = te.refreshURIs()
 		if err != nil {
 			return err
 		}
 	} else {
 		te.clusterType = StandAlone
 
-		// Create a new network
-
-		if avalancheGoExecPath != "" {
-			if _, err := os.Stat(avalancheGoExecPath); err != nil {
-				return fmt.Errorf("could not find avalanchego binary: %w", err)
-			}
-		}
-
-		te.avalancheGoExecPath = avalancheGoExecPath
-		te.avalancheGoLogLevel = avalancheGoLogLevel
-
 		err := te.startCluster()
 		if err != nil {
 			return err
@@ -221,21 +220,30 @@ func (te *TestEnvironment) startCluster() error {
 
 	tests.Outf("{{green}}successfully started network: {{/}} %+v\n", resp.ClusterInfo.NodeNames)
 
-	ctx, cancel = context.WithTimeout(context.Background(), 2*time.Minute)
-	_, err = te.GetRunnerClient().Health(ctx)
-	cancel()
+	err = te.CheckHealth()
 	if err != nil {
-		return fmt.Errorf("could not check network health: %w", err)
+		return err
 	}
 
-	tests.Outf("{{green}}network reporting health{{/}}\n")
-
 	te.isNetworkPristine = true
 
 	err = te.refreshURIs()
 	return err
 }
 
+func (te *TestEnvironment) CheckHealth() error {
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
+	_, err := te.GetRunnerClient().Health(ctx)
+	cancel()
+	if err != nil {
+		return fmt.Errorf("could not check network health: %w", err)
+	}
+
+	tests.Outf("{{green}}network reporting health{{/}}\n")
+
+	return nil
+}
+
 func (te *TestEnvironment) refreshURIs() error {
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	uriSlice, err := te.GetRunnerClient().URIs(ctx)
@@ -365,3 +373,16 @@ func (te *TestEnvironment) Teardown() error {
 	tests.Outf("{{red}}shutting down network-runner client{{/}}\n")
 	return te.closeRunnerClient()
 }
+
+func (te *TestEnvironment) AddNode(name string, opts ...runner_sdk.OpOption) error {
+	ctx, cancel := context.WithTimeout(context.Background(), DefaultShutdownTimeout)
+	_, err := te.GetRunnerClient().AddNode(ctx, name, te.avalancheGoExecPath, opts...)
+	cancel()
+	if err != nil {
+		return err
+	}
+
+	tests.Outf("{{green}}added node %s{{/}}\n", name)
+
+	return nil
+}

tests/e2e/e2e_test.go

@@ -15,6 +15,7 @@ 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/ping"
 	_ "github.com/ava-labs/avalanchego/tests/e2e/static-handlers"

tests/e2e/faultinjection/duplicate_node_id.go

@@ -0,0 +1,62 @@
+// Copyright (C) 2019-2023, Ava Labs, Inc. All rights reserved.
+// See the file LICENSE for licensing terms.
+
+package faultinjection
+
+import (
+	"fmt"
+	"math/rand"
+	"strings"
+
+	ginkgo "github.com/onsi/ginkgo/v2"
+	"github.com/stretchr/testify/require"
+
+	runner_sdk "github.com/ava-labs/avalanche-network-runner-sdk"
+
+	"github.com/ava-labs/avalanchego/staking"
+	"github.com/ava-labs/avalanchego/tests/e2e"
+)
+
+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() {
+		// TODO(marun) Ensure reliable node removal on teardown
+
+		// Minimize the potential for node name collision to allow for
+		// iterating with a persistent network. The use of non-secure random
+		// value is not security critical so the lint is ignored.
+		//nolint:gosec
+		baseNodeName := fmt.Sprintf("e2e-duplicate-node-%08x-", rand.Uint32())
+		node1Name := baseNodeName + "1"
+
+		var stakingCert, stakingKey []byte
+		ginkgo.By("generating a staking keypair", func() {
+			var err error
+			stakingCert, stakingKey, err = staking.NewCertAndKeyBytes()
+			require.NoError(err)
+		})
+
+		ginkgo.By("creating the first node using the staking keypair", func() {
+			err := e2e.Env.AddNode(node1Name, runner_sdk.WithGlobalNodeConfig(fmt.Sprintf(
+				`{""--staking-tls-cert-contents": "%s", "--staking-tls-key-contents": "%s"}`,
+				// Escape newlines in the PEM-formatted keys to
+				// ensure compatibility with grpc.
+				strings.ReplaceAll(string(stakingCert), "\n", "\\n"),
+				strings.ReplaceAll(string(stakingKey), "\n", "\\n"),
+			)))
+			require.NoError(err)
+			err = e2e.Env.CheckHealth()
+			require.NoError(err)
+		})
+
+		// TODO(marun)
+		// - Boot badNode2 with the same Certs
+		// - Check the network peers are connected amongst the network with badNode1 (the first to go up)
+		// - Check that badNode2 can't bootstrap (invalid peer alias)
+		// - Remove badNode1
+		// - Check that badNode2 can now bootstrap
+		// - Check both nodes have/had the same nodeID
+		// - Check the network peers are connected amongst the network with badNode2
+	})
+})