feat: add liveness check for nodes in e2e tests and update node startup parameters

Author: randygrok

test/e2e/base_test.go

@@ -65,20 +65,22 @@ func TestBasic(t *testing.T) {
 	require.NoError(t, err, "failed to init aggregator", output)
 
 	// start aggregator
+	node1P2P := "/ip4/0.0.0.0/tcp/26656"
 	sut.ExecCmd(binaryPath,
 		"start",
 		"--home="+node1Home,
 		"--evnode.node.aggregator",
 		"--evnode.signer.passphrase_file="+passphraseFile,
 		"--evnode.node.block_time=5ms",
 		"--evnode.da.block_time=15ms",
+		"--evnode.p2p.listen_address="+node1P2P,
 		"--kv-endpoint=127.0.0.1:9090",
 	)
 
 	sut.AwaitNodeUp(t, "http://127.0.0.1:7331", 2*time.Second)
 
 	// Give aggregator more time before starting the next node
-	time.Sleep(1 * time.Second) // Increased wait time
+	time.Sleep(2 * time.Second)
 
 	// Init the second node (full node)
 	output, err = sut.RunCmd(binaryPath,
@@ -91,7 +93,7 @@ func TestBasic(t *testing.T) {
 	// Copy genesis file from aggregator to full node
 	MustCopyFile(t, filepath.Join(node1Home, "config", "genesis.json"), filepath.Join(node2Home, "config", "genesis.json"))
 
-	// Start the full node
+	// Start the full node - will discover aggregator via DHT
 	node2RPC := "127.0.0.1:7332"
 	node2P2P := "/ip4/0.0.0.0/tcp/7676"
 	sut.ExecCmd(
@@ -100,11 +102,13 @@ func TestBasic(t *testing.T) {
 		"--home="+node2Home,
 		"--evnode.log.level=debug",
 		"--evnode.p2p.listen_address="+node2P2P,
+		"--evnode.node.readiness_max_blocks_behind=100", // Allow more blocks behind during bootstrap
 		fmt.Sprintf("--evnode.rpc.address=%s", node2RPC),
 	)
 
-	sut.AwaitNodeUp(t, "http://"+node2RPC, 2*time.Second)
-	t.Logf("Full node (node 2) is up.")
+	// For local e2e tests, only check liveness (not full readiness with peers)
+	sut.AwaitNodeLive(t, "http://"+node2RPC, 10*time.Second)
+	t.Logf("Full node (node 2) is live.")
 
 	// when a client TX for state update is executed
 	const myKey = "foo"

test/e2e/sut_helper.go

@@ -115,6 +115,24 @@ func (s *SystemUnderTest) AwaitNodeUp(t *testing.T, rpcAddr string, timeout time
 	}, timeout, min(timeout/10, 200*time.Millisecond), "node is not up")
 }
 
+// AwaitNodeLive waits until a node is alive (liveness check only).
+// This only verifies the process is alive and responsive, not that it's ready to serve traffic.
+// Use this for local tests where nodes may not have peers configured.
+func (s *SystemUnderTest) AwaitNodeLive(t *testing.T, rpcAddr string, timeout time.Duration) {
+	t.Helper()
+	t.Logf("Await node is live: %s", rpcAddr)
+	ctx, done := context.WithTimeout(context.Background(), timeout)
+	defer done()
+	require.EventuallyWithT(t, func(t *assert.CollectT) {
+		c := client.NewClient(rpcAddr)
+		require.NotNil(t, c)
+
+		// Check liveness only: is the process alive?
+		_, err := c.GetHealth(ctx)
+		require.NoError(t, err, "liveness check failed")
+	}, timeout, min(timeout/10, 200*time.Millisecond), "node is not live")
+}
+
 // AwaitNBlocks waits until the node has produced at least `n` blocks.
 func (s *SystemUnderTest) AwaitNBlocks(t *testing.T, n uint64, rpcAddr string, timeout time.Duration) {
 	t.Helper()