Merge pull request #5556 from ethereum-optimism/felipe/lb-round-robin

proxyd: round-robin load balancing for consensus group

proxyd/backend.go

@@ -591,9 +591,17 @@ func (b *BackendGroup) Forward(ctx context.Context, rpcReqs []*RPCReq, isBatch b
 		return nil, nil
 	}
 
+	backends := b.Backends
+
+	// When `consensus_aware` is set to `true`, the backend group acts as a load balancer
+	// serving traffic from any backend that agrees in the consensus group
+	if b.Consensus != nil {
+		backends = b.loadBalancedConsensusGroup()
+	}
+
 	rpcRequestsTotal.Inc()
 
-	for _, back := range b.Backends {
+	for _, back := range backends {
 		res, err := back.Forward(ctx, rpcReqs, isBatch)
 		if errors.Is(err, ErrMethodNotWhitelisted) {
 			return nil, err
@@ -670,6 +678,40 @@ func (b *BackendGroup) ProxyWS(ctx context.Context, clientConn *websocket.Conn,
 	return nil, ErrNoBackends
 }
 
+func (b *BackendGroup) loadBalancedConsensusGroup() []*Backend {
+	cg := b.Consensus.GetConsensusGroup()
+
+	backendsHealthy := make([]*Backend, 0, len(cg))
+	backendsDegraded := make([]*Backend, 0, len(cg))
+	// separate into healthy, degraded and unhealthy backends
+	for _, be := range cg {
+		// unhealthy are filtered out and not attempted
+		if !be.IsHealthy() {
+			continue
+		}
+		if be.IsDegraded() {
+			backendsDegraded = append(backendsDegraded, be)
+			continue
+		}
+		backendsHealthy = append(backendsHealthy, be)
+	}
+
+	// shuffle both slices
+	r := rand.New(rand.NewSource(time.Now().UnixNano()))
+	r.Shuffle(len(backendsHealthy), func(i, j int) {
+		backendsHealthy[i], backendsHealthy[j] = backendsHealthy[j], backendsHealthy[i]
+	})
+	r.Shuffle(len(backendsDegraded), func(i, j int) {
+		backendsDegraded[i], backendsDegraded[j] = backendsDegraded[j], backendsDegraded[i]
+	})
+
+	// healthy are put into a priority position
+	// degraded backends are used as fallback
+	backendsHealthy = append(backendsHealthy, backendsDegraded...)
+
+	return backendsHealthy
+}
+
 func calcBackoff(i int) time.Duration {
 	jitter := float64(rand.Int63n(250))
 	ms := math.Min(math.Pow(2, float64(i))*1000+jitter, 3000)

proxyd/integration_tests/consensus_test.go

@@ -3,6 +3,7 @@ package integration_tests
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"os"
 	"path"
@@ -47,6 +48,7 @@ func TestConsensus(t *testing.T) {
 	ctx := context.Background()
 	svr, shutdown, err := proxyd.Start(config)
 	require.NoError(t, err)
+	client := NewProxydClient("http://127.0.0.1:8545")
 	defer shutdown()
 
 	bg := svr.BackendGroups["node"]
@@ -76,7 +78,6 @@ func TestConsensus(t *testing.T) {
 		h2.ResetOverrides()
 		bg.Consensus.Unban()
 
-		// advance latest on node2 to 0x2
 		h1.AddOverride(&ms.MethodTemplate{
 			Method:   "net_peerCount",
 			Block:    "",
@@ -355,6 +356,83 @@ func TestConsensus(t *testing.T) {
 		// should resolve to 0x1, the highest common ancestor
 		require.Equal(t, "0x1", bg.Consensus.GetConsensusBlockNumber().String())
 	})
+
+	t.Run("load balancing should hit both backends", func(t *testing.T) {
+		h1.ResetOverrides()
+		h2.ResetOverrides()
+		bg.Consensus.Unban()
+
+		for _, be := range bg.Backends {
+			bg.Consensus.UpdateBackend(ctx, be)
+		}
+		bg.Consensus.UpdateBackendGroupConsensus(ctx)
+
+		require.Equal(t, 2, len(bg.Consensus.GetConsensusGroup()))
+
+		node1.Reset()
+		node2.Reset()
+
+		require.Equal(t, 0, len(node1.Requests()))
+		require.Equal(t, 0, len(node2.Requests()))
+
+		// there is a random component to this test,
+		// since our round-robin implementation shuffles the ordering
+		// to achieve uniform distribution
+
+		// so we just make 100 requests per backend and expect the number of requests to be somewhat balanced
+		// i.e. each backend should be hit minimally by at least 50% of the requests
+		consensusGroup := bg.Consensus.GetConsensusGroup()
+
+		numberReqs := len(consensusGroup) * 100
+		for numberReqs > 0 {
+			_, statusCode, err := client.SendRPC("eth_getBlockByNumber", []interface{}{"0x1", false})
+			require.NoError(t, err)
+			require.Equal(t, 200, statusCode)
+			numberReqs--
+		}
+
+		msg := fmt.Sprintf("n1 %d, n2 %d", len(node1.Requests()), len(node2.Requests()))
+		require.GreaterOrEqual(t, len(node1.Requests()), 50, msg)
+		require.GreaterOrEqual(t, len(node2.Requests()), 50, msg)
+	})
+
+	t.Run("load balancing should not hit if node is not healthy", func(t *testing.T) {
+		h1.ResetOverrides()
+		h2.ResetOverrides()
+		bg.Consensus.Unban()
+
+		// node1 should not be serving any traffic
+		h1.AddOverride(&ms.MethodTemplate{
+			Method:   "net_peerCount",
+			Block:    "",
+			Response: buildPeerCountResponse(1),
+		})
+
+		for _, be := range bg.Backends {
+			bg.Consensus.UpdateBackend(ctx, be)
+		}
+		bg.Consensus.UpdateBackendGroupConsensus(ctx)
+
+		require.Equal(t, 1, len(bg.Consensus.GetConsensusGroup()))
+
+		node1.Reset()
+		node2.Reset()
+
+		require.Equal(t, 0, len(node1.Requests()))
+		require.Equal(t, 0, len(node2.Requests()))
+
+		numberReqs := 10
+		for numberReqs > 0 {
+			_, statusCode, err := client.SendRPC("eth_getBlockByNumber", []interface{}{"0x1", false})
+			require.NoError(t, err)
+			require.Equal(t, 200, statusCode)
+			numberReqs--
+		}
+
+		msg := fmt.Sprintf("n1 %d, n2 %d", len(node1.Requests()), len(node2.Requests()))
+		require.Equal(t, len(node1.Requests()), 0, msg)
+		require.Equal(t, len(node2.Requests()), 10, msg)
+	})
 }
 
 func backend(bg *proxyd.BackendGroup, name string) *proxyd.Backend {

proxyd/integration_tests/testdata/consensus.toml

 [server]
-rpc_port = 8080
+rpc_port = 8545
 
 [backend]
 response_timeout_seconds = 1

proxyd/integration_tests/testdata/consensus_responses.yml

+- method: eth_chainId
+  response: >
+    {
+      "jsonrpc": "2.0",
+      "id": 67,
+      "result": "hello",
+    }
 - method: net_peerCount
   response: >
     {