better moar tests

proxyd/backend.go

@@ -374,7 +374,6 @@ func (b *Backend) ForwardRPC(ctx context.Context, res *RPCRes, id string, method
 func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool) ([]*RPCRes, error) {
 	// we are concerned about network error rates, so we record 1 request independently of how many are in the batch
 	b.networkRequestsSlidingWindow.Incr()
-	RecordBackendNetworkRequestCountSlidingWindow(b, b.networkRequestsSlidingWindow.Count())
 
 	isSingleElementBatch := len(rpcReqs) == 1
 
@@ -391,7 +390,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 	httpReq, err := http.NewRequestWithContext(ctx, "POST", b.rpcURL, bytes.NewReader(body))
 	if err != nil {
 		b.networkErrorsSlidingWindow.Incr()
-		RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+		RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 		return nil, wrapErr(err, "error creating backend request")
 	}
 
@@ -413,7 +412,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 	httpRes, err := b.client.DoLimited(httpReq)
 	if err != nil {
 		b.networkErrorsSlidingWindow.Incr()
-		RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+		RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 		return nil, wrapErr(err, "error in backend request")
 	}
 
@@ -432,7 +431,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 	// Alchemy returns a 400 on bad JSONs, so handle that case
 	if httpRes.StatusCode != 200 && httpRes.StatusCode != 400 {
 		b.networkErrorsSlidingWindow.Incr()
-		RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+		RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 		return nil, fmt.Errorf("response code %d", httpRes.StatusCode)
 	}
 
@@ -440,7 +439,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 	resB, err := io.ReadAll(io.LimitReader(httpRes.Body, b.maxResponseSize))
 	if err != nil {
 		b.networkErrorsSlidingWindow.Incr()
-		RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+		RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 		return nil, wrapErr(err, "error reading response body")
 	}
 
@@ -458,18 +457,18 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 			// Infura may return a single JSON-RPC response if, for example, the batch contains a request for an unsupported method
 			if responseIsNotBatched(resB) {
 				b.networkErrorsSlidingWindow.Incr()
-				RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+				RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 				return nil, ErrBackendUnexpectedJSONRPC
 			}
 			b.networkErrorsSlidingWindow.Incr()
-			RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+			RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 			return nil, ErrBackendBadResponse
 		}
 	}
 
 	if len(rpcReqs) != len(res) {
 		b.networkErrorsSlidingWindow.Incr()
-		RecordBackendNetworkErrorCountSlidingWindow(b, b.networkErrorsSlidingWindow.Count())
+		RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 		return nil, ErrBackendUnexpectedJSONRPC
 	}
 
@@ -483,6 +482,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 	duration := time.Since(start)
 	b.latencySlidingWindow.Add(float64(duration))
 	RecordBackendNetworkLatencyAverageSlidingWindow(b, time.Duration(b.latencySlidingWindow.Avg()))
+	RecordBackendNetworkErrorRateSlidingWindow(b, b.ErrorRate())
 
 	sortBatchRPCResponse(rpcReqs, res)
 	return res, nil
@@ -490,11 +490,7 @@ func (b *Backend) doForward(ctx context.Context, rpcReqs []*RPCReq, isBatch bool
 
 // IsHealthy checks if the backend is able to serve traffic, based on dynamic parameters
 func (b *Backend) IsHealthy() bool {
-	errorRate := float64(0)
-	// avoid division-by-zero when the window is empty
-	if b.networkRequestsSlidingWindow.Sum() >= 10 {
-		errorRate = b.networkErrorsSlidingWindow.Sum() / b.networkRequestsSlidingWindow.Sum()
-	}
+	errorRate := b.ErrorRate()
 	avgLatency := time.Duration(b.latencySlidingWindow.Avg())
 	if errorRate >= b.maxErrorRateThreshold {
 		return false
@@ -505,6 +501,16 @@ func (b *Backend) IsHealthy() bool {
 	return true
 }
 
+// ErrorRate returns the instant error rate of the backend
+func (b *Backend) ErrorRate() (errorRate float64) {
+	// we only really start counting the error rate after a minimum of 10 requests
+	// this is to avoid false positives when the backend is just starting up
+	if b.networkRequestsSlidingWindow.Sum() >= 10 {
+		errorRate = b.networkErrorsSlidingWindow.Sum() / b.networkRequestsSlidingWindow.Sum()
+	}
+	return errorRate
+}
+
 // IsDegraded checks if the backend is serving traffic in a degraded state (i.e. used as a last resource)
 func (b *Backend) IsDegraded() bool {
 	avgLatency := time.Duration(b.latencySlidingWindow.Avg())

proxyd/consensus_poller.go

@@ -275,23 +275,42 @@ func (cp *ConsensusPoller) UpdateBackend(ctx context.Context, be *Backend) {
 		log.Warn("error updating backend", "name", be.Name, "err", err)
 	}
 
-	finalizedBlockNumber, _, err := cp.fetchBlock(ctx, be, "finalized")
+	safeBlockNumber, _, err := cp.fetchBlock(ctx, be, "safe")
 	if err != nil {
 		log.Warn("error updating backend", "name", be.Name, "err", err)
 	}
 
-	safeBlockNumber, _, err := cp.fetchBlock(ctx, be, "safe")
+	finalizedBlockNumber, _, err := cp.fetchBlock(ctx, be, "finalized")
 	if err != nil {
 		log.Warn("error updating backend", "name", be.Name, "err", err)
 	}
 
+	_, _, _, _, oldFinalized, oldSafe, _, _ := cp.getBackendState(be)
+	expectedBlockTags := cp.checkExpectedBlockTags(finalizedBlockNumber, oldFinalized, safeBlockNumber, oldSafe, latestBlockNumber)
+
 	changed, updateDelay := cp.setBackendState(be, peerCount, inSync,
 		latestBlockNumber, latestBlockHash,
 		finalizedBlockNumber, safeBlockNumber)
 
+	RecordBackendLatestBlock(be, latestBlockNumber)
+	RecordBackendSafeBlock(be, safeBlockNumber)
+	RecordBackendFinalizedBlock(be, finalizedBlockNumber)
+	RecordBackendUnexpectedBlockTags(be, !expectedBlockTags)
+	RecordConsensusBackendUpdateDelay(be, updateDelay)
+
+	if !expectedBlockTags {
+		log.Warn("backend banned - unexpected block tags",
+			"backend", be.Name,
+			"oldFinalized", oldFinalized,
+			"finalizedBlockNumber", finalizedBlockNumber,
+			"oldSafe", oldSafe,
+			"safeBlockNumber", safeBlockNumber,
+			"latestBlockNumber", latestBlockNumber,
+		)
+		cp.Ban(be)
+	}
+
 	if changed {
-		RecordBackendLatestBlock(be, latestBlockNumber)
-		RecordConsensusBackendUpdateDelay(be, updateDelay)
 		log.Debug("backend state updated",
 			"name", be.Name,
 			"peerCount", peerCount,
@@ -304,6 +323,19 @@ func (cp *ConsensusPoller) UpdateBackend(ctx context.Context, be *Backend) {
 	}
 }
 
+// checkExpectedBlockTags for unexpected conditions on block tags
+// - finalized block number should never decrease
+// - safe block number should never decrease
+// - finalized block should be < safe block < latest block
+func (cp *ConsensusPoller) checkExpectedBlockTags(currentFinalized hexutil.Uint64, oldFinalized hexutil.Uint64,
+	currentSafe hexutil.Uint64, oldSafe hexutil.Uint64,
+	currentLatest hexutil.Uint64) bool {
+	return currentFinalized >= oldFinalized &&
+		currentSafe >= oldSafe &&
+		currentFinalized <= currentSafe &&
+		currentSafe <= currentLatest
+}
+
 // UpdateBackendGroupConsensus resolves the current group consensus based on the state of the backends
 func (cp *ConsensusPoller) UpdateBackendGroupConsensus(ctx context.Context) {
 	var highestLatestBlock hexutil.Uint64
@@ -320,6 +352,9 @@ func (cp *ConsensusPoller) UpdateBackendGroupConsensus(ctx context.Context) {
 	for _, be := range cp.backendGroup.Backends {
 		peerCount, inSync, backendLatestBlockNumber, _, _, _, lastUpdate, _ := cp.getBackendState(be)
 
+		if cp.IsBanned(be) {
+			continue
+		}
 		if !be.skipPeerCountCheck && peerCount < cp.minPeerCount {
 			continue
 		}
@@ -339,6 +374,9 @@ func (cp *ConsensusPoller) UpdateBackendGroupConsensus(ctx context.Context) {
 	for _, be := range cp.backendGroup.Backends {
 		peerCount, inSync, backendLatestBlockNumber, backendLatestBlockHash, backendFinalizedBlockNumber, backendSafeBlockNumber, lastUpdate, _ := cp.getBackendState(be)
 
+		if cp.IsBanned(be) {
+			continue
+		}
 		if !be.skipPeerCountCheck && peerCount < cp.minPeerCount {
 			continue
 		}
@@ -451,13 +489,17 @@ func (cp *ConsensusPoller) UpdateBackendGroupConsensus(ctx context.Context) {
 	}
 
 	cp.tracker.SetLatestBlockNumber(proposedBlock)
-	cp.tracker.SetFinalizedBlockNumber(lowestFinalizedBlock)
 	cp.tracker.SetSafeBlockNumber(lowestSafeBlock)
+	cp.tracker.SetFinalizedBlockNumber(lowestFinalizedBlock)
+
 	cp.consensusGroupMux.Lock()
 	cp.consensusGroup = consensusBackends
 	cp.consensusGroupMux.Unlock()
 
 	RecordGroupConsensusLatestBlock(cp.backendGroup, proposedBlock)
+	RecordGroupConsensusSafeBlock(cp.backendGroup, lowestSafeBlock)
+	RecordGroupConsensusFinalizedBlock(cp.backendGroup, lowestFinalizedBlock)
+
 	RecordGroupConsensusCount(cp.backendGroup, len(consensusBackends))
 	RecordGroupConsensusFilteredCount(cp.backendGroup, len(filteredBackendsNames))
 	RecordGroupTotalCount(cp.backendGroup, len(cp.backendGroup.Backends))
@@ -481,13 +523,10 @@ func (cp *ConsensusPoller) Ban(be *Backend) {
 	bs.bannedUntil = time.Now().Add(cp.banPeriod)
 }
 
-// Unban remove any bans from the backends
-func (cp *ConsensusPoller) Unban() {
+// Reset remove any bans from the backends and reset their states
+func (cp *ConsensusPoller) Reset() {
 	for _, be := range cp.backendGroup.Backends {
-		bs := cp.backendState[be]
-		bs.backendStateMux.Lock()
-		bs.bannedUntil = time.Now().Add(-10 * time.Hour)
-		bs.backendStateMux.Unlock()
+		cp.backendState[be] = &backendState{}
 	}
 }
 

proxyd/integration_tests/testdata/consensus_responses.yml

@@ -26,63 +26,85 @@
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash1",
-        "number": "0x1"
+        "hash": "hash_0x101",
+        "number": "0x101"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x1
+  block: 0x101
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash1",
-        "number": "0x1"
+        "hash": "hash_0x101",
+        "number": "0x101"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x2
+  block: 0x102
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash2",
-        "number": "0x2"
+        "hash": "hash_0x102",
+        "number": "0x102"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x3
+  block: 0x103
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash3",
-        "number": "0x3"
+        "hash": "hash_0x103",
+        "number": "0x103"
       }
     }
 - method: eth_getBlockByNumber
-  block: finalized
+  block: 0x132
+  response: >
+    {
+      "jsonrpc": "2.0",
+      "id": 67,
+      "result": {
+        "hash": "hash_0x132",
+        "number": "0x132"
+      }
+    }
+- method: eth_getBlockByNumber
+  block: 0x133
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash_finalized",
-        "number": "0x555"
+        "hash": "hash_0x133",
+        "number": "0x133"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x555
+  block: 0x134
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash_finalized",
-        "number": "0x555"
+        "hash": "hash_0x134",
+        "number": "0x134"
+      }
+    }
+- method: eth_getBlockByNumber
+  block: 0x200
+  response: >
+    {
+      "jsonrpc": "2.0",
+      "id": 67,
+      "result": {
+        "hash": "hash_0x200",
+        "number": "0x200"
       }
     }
 - method: eth_getBlockByNumber
@@ -92,40 +114,40 @@
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash_safe",
-        "number": "0x551"
+        "hash": "hash_0xe1",
+        "number": "0xe1"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x555
+  block: 0xe1
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash_safe",
-        "number": "0x551"
+        "hash": "hash_0xe1",
+        "number": "0xe1"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x5
+  block: finalized
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash5",
-        "number": "0x5"
+        "hash": "hash_0xc1",
+        "number": "0xc1"
       }
     }
 - method: eth_getBlockByNumber
-  block: 0x20
+  block: 0xc1
   response: >
     {
       "jsonrpc": "2.0",
       "id": 67,
       "result": {
-        "hash": "hash20",
-        "number": "0x20"
+        "hash": "hash_0xc1",
+        "number": "0xc1"
       }
     }

proxyd/metrics.go

@@ -246,6 +246,22 @@ var (
 		"backend_group_name",
 	})
 
+	consensusSafeBlock = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: MetricsNamespace,
+		Name:      "group_consensus_safe_block",
+		Help:      "Consensus safe block",
+	}, []string{
+		"backend_group_name",
+	})
+
+	consensusFinalizedBlock = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: MetricsNamespace,
+		Name:      "group_consensus_finalized_block",
+		Help:      "Consensus finalized block",
+	}, []string{
+		"backend_group_name",
+	})
+
 	backendLatestBlockBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: MetricsNamespace,
 		Name:      "backend_latest_block",
@@ -254,6 +270,30 @@ var (
 		"backend_name",
 	})
 
+	backendSafeBlockBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: MetricsNamespace,
+		Name:      "backend_safe_block",
+		Help:      "Current safe block observed per backend",
+	}, []string{
+		"backend_name",
+	})
+
+	backendFinalizedBlockBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: MetricsNamespace,
+		Name:      "backend_finalized_block",
+		Help:      "Current finalized block observed per backend",
+	}, []string{
+		"backend_name",
+	})
+
+	backendUnexpectedBlockTagsBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: MetricsNamespace,
+		Name:      "backend_unexpected_block_tags",
+		Help:      "Bool gauge for unexpected block tags",
+	}, []string{
+		"backend_name",
+	})
+
 	consensusGroupCount = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: MetricsNamespace,
 		Name:      "group_consensus_count",
@@ -318,18 +358,10 @@ var (
 		"backend_name",
 	})
 
-	networkErrorCountBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
+	networkErrorRateBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: MetricsNamespace,
-		Name:      "backend_net_error_count",
-		Help:      "Network error count per backend",
-	}, []string{
-		"backend_name",
-	})
-
-	requestCountBackend = promauto.NewGaugeVec(prometheus.GaugeOpts{
-		Namespace: MetricsNamespace,
-		Name:      "backend_request_count",
-		Help:      "Request count per backend",
+		Name:      "backend_error_rate",
+		Help:      "Request error rate per backend",
 	}, []string{
 		"backend_name",
 	})
@@ -402,6 +434,14 @@ func RecordGroupConsensusLatestBlock(group *BackendGroup, blockNumber hexutil.Ui
 	consensusLatestBlock.WithLabelValues(group.Name).Set(float64(blockNumber))
 }
 
+func RecordGroupConsensusSafeBlock(group *BackendGroup, blockNumber hexutil.Uint64) {
+	consensusSafeBlock.WithLabelValues(group.Name).Set(float64(blockNumber))
+}
+
+func RecordGroupConsensusFinalizedBlock(group *BackendGroup, blockNumber hexutil.Uint64) {
+	consensusFinalizedBlock.WithLabelValues(group.Name).Set(float64(blockNumber))
+}
+
 func RecordGroupConsensusCount(group *BackendGroup, count int) {
 	consensusGroupCount.WithLabelValues(group.Name).Set(float64(count))
 }
@@ -418,12 +458,20 @@ func RecordBackendLatestBlock(b *Backend, blockNumber hexutil.Uint64) {
 	backendLatestBlockBackend.WithLabelValues(b.Name).Set(float64(blockNumber))
 }
 
+func RecordBackendSafeBlock(b *Backend, blockNumber hexutil.Uint64) {
+	backendSafeBlockBackend.WithLabelValues(b.Name).Set(float64(blockNumber))
+}
+
+func RecordBackendFinalizedBlock(b *Backend, blockNumber hexutil.Uint64) {
+	backendFinalizedBlockBackend.WithLabelValues(b.Name).Set(float64(blockNumber))
+}
+
+func RecordBackendUnexpectedBlockTags(b *Backend, unexpected bool) {
+	backendFinalizedBlockBackend.WithLabelValues(b.Name).Set(boolToFloat64(unexpected))
+}
+
 func RecordConsensusBackendBanned(b *Backend, banned bool) {
-	v := float64(0)
-	if banned {
-		v = float64(1)
-	}
-	consensusBannedBackends.WithLabelValues(b.Name).Set(v)
+	consensusBannedBackends.WithLabelValues(b.Name).Set(boolToFloat64(banned))
 }
 
 func RecordConsensusBackendPeerCount(b *Backend, peerCount uint64) {
@@ -431,11 +479,7 @@ func RecordConsensusBackendPeerCount(b *Backend, peerCount uint64) {
 }
 
 func RecordConsensusBackendInSync(b *Backend, inSync bool) {
-	v := float64(0)
-	if inSync {
-		v = float64(1)
-	}
-	consensusInSyncBackend.WithLabelValues(b.Name).Set(v)
+	consensusInSyncBackend.WithLabelValues(b.Name).Set(boolToFloat64(inSync))
 }
 
 func RecordConsensusBackendUpdateDelay(b *Backend, delay time.Duration) {
@@ -446,10 +490,13 @@ func RecordBackendNetworkLatencyAverageSlidingWindow(b *Backend, avgLatency time
 	avgLatencyBackend.WithLabelValues(b.Name).Set(float64(avgLatency.Milliseconds()))
 }
 
-func RecordBackendNetworkRequestCountSlidingWindow(b *Backend, count uint) {
-	requestCountBackend.WithLabelValues(b.Name).Set(float64(count))
+func RecordBackendNetworkErrorRateSlidingWindow(b *Backend, rate float64) {
+	networkErrorRateBackend.WithLabelValues(b.Name).Set(rate)
 }
 
-func RecordBackendNetworkErrorCountSlidingWindow(b *Backend, count uint) {
-	networkErrorCountBackend.WithLabelValues(b.Name).Set(float64(count))
+func boolToFloat64(b bool) float64 {
+	if b {
+		return 1
+	}
+	return 0
 }

proxyd/rewriter.go

@@ -10,8 +10,8 @@ import (
 
 type RewriteContext struct {
 	latest    hexutil.Uint64
-	finalized hexutil.Uint64
 	safe      hexutil.Uint64
+	finalized hexutil.Uint64
 }
 
 type RewriteResult uint8