@lodestar/beacon-node 1.45.0-dev.3100972235 → 1.45.0-dev.31b495a1d4

package/src/network/core/metrics.ts

@@ -127,6 +127,100 @@ export function createNetworkCoreMetrics(register: RegistryMetricCreator) {
         name: "lodestar_peer_manager_starved_bool",
         help: "Whether lodestar is starved of data while syncing",
       }),
+      /**
+       * Core selection/pruning phase of the heartbeat, split out of heartbeatDuration so it
+       * can be attributed on its own. Divide by peersEvaluated for time-per-peer.
+       */
+      prioritizePeersDuration: register.histogram({
+        name: "lodestar_peer_manager_prioritize_peers_seconds",
+        help: "prioritizePeers function duration in seconds, the core peer selection/pruning algorithm",
+        buckets: [0.0005, 0.001, 0.005, 0.01, 0.05, 0.1],
+      }),
+      /**
+       * Score decay + map prune over every tracked peer, split out of heartbeatDuration.
+       * This loop iterates the whole score store, so divide by scoreMapSize (not
+       * peersEvaluated) for time-per-peer.
+       */
+      scoreUpdateDuration: register.histogram({
+        name: "lodestar_peer_score_update_seconds",
+        help: "Peer score store update (decay + prune over all peers) duration in seconds",
+        buckets: [0.0005, 0.001, 0.005, 0.01, 0.05, 0.1],
+      }),
+      /**
+       * Denominator for prioritizePeersDuration: it scales with the number of connected
+       * healthy peers, so time-per-peer (prioritizePeersDuration / this) is the comparable
+       * figure. For scoreUpdateDuration use scoreMapSize instead.
+       */
+      peersEvaluated: register.histogram({
+        name: "lodestar_peer_manager_peers_evaluated_count",
+        help: "Number of connected healthy peers evaluated by prioritizePeers per heartbeat, denominator for prioritize_peers_seconds",
+        buckets: [0, 25, 50, 75, 100, 150, 200],
+      }),
+      /**
+       * Every peer the manager intends to disconnect. Covers both the top-of-heartbeat
+       * bad-score disconnects ("banned"/"score_too_low") and the prioritization reasons;
+       * the older peersRequestedToDisconnect gauge only recorded the latter. Counts intent,
+       * not completion (goodbye is fire-and-forget).
+       */
+      peersPruned: register.counter<{reason: string}>({
+        name: "lodestar_peer_manager_peers_pruned_total",
+        help: "Total peers the peer manager intends to disconnect, labeled by reason (incl. bad-score and prioritization reasons)",
+        labelNames: ["reason"],
+      }),
+      /**
+       * Actual peer count per active subnet, to check the min-peers-per-subnet invariant.
+       * Buckets straddle the target (TARGET_SUBNET_PEERS = 6).
+       */
+      peersPerActiveSubnet: register.histogram<{type: SubnetType}>({
+        name: "lodestar_peer_manager_peers_per_active_subnet",
+        help: "Histogram of connected peer count per active subnet, labeled by subnet type",
+        labelNames: ["type"],
+        buckets: [0, 2, 4, 6, 8, 12],
+      }),
+      /**
+       * Live outbound ratio, to check the OUTBOUND_PEERS_RATIO (10%) invariant (issue #2215).
+       * Denominator is connected healthy peers (banned/disconnected already removed).
+       */
+      outboundPeersRatio: register.gauge({
+        name: "lodestar_peer_manager_outbound_peers_ratio",
+        help: "Ratio of outbound peers to total connected healthy peers, verifies the outbound peers invariant",
+      }),
+      /**
+       * Score state crossings (Healthy/Disconnected/Banned). Sensitive to the decay formula,
+       * thresholds, and gossip-score weighting.
+       */
+      scoreStateTransitions: register.counter<{from: string; to: string}>({
+        name: "lodestar_peer_score_state_transitions_total",
+        help: "Total peer score state transitions, labeled by from and to state (Healthy/Disconnected/Banned)",
+        labelNames: ["from", "to"],
+      }),
+      /**
+       * Entry count of the score store (not bytes). Reflects the prune-to-MAX_ENTRIES /
+       * SCORE_THRESHOLD retention logic. Also the denominator for scoreUpdateDuration
+       * time-per-peer (that loop iterates the whole store, not just connected peers).
+       */
+      scoreMapSize: register.gauge({
+        name: "lodestar_peer_manager_score_map_size",
+        help: "Current number of entries in the peer score store",
+      }),
+      /**
+       * Entry count of the connectedPeers map (not bytes). Should track libp2p_peers; a
+       * persistent gap indicates the connection leak guarded by leakedConnectionsCount.
+       */
+      connectedPeersMapSize: register.gauge({
+        name: "lodestar_peer_manager_connected_peers_map_size",
+        help: "Current number of entries in the peer manager connectedPeers map",
+      }),
+      /**
+       * assertPeerRelevance outcome per Status exchange: "relevant", an irrelevant reason
+       * code, or "error". Fires on every status evaluation (inbound requests and our own
+       * requestStatus responses both route through onStatus).
+       */
+      relevanceCheck: register.counter<{result: string}>({
+        name: "lodestar_peer_relevance_check_total",
+        help: "Total peer relevance checks on Status, labeled by result (relevant or irrelevant reason code)",
+        labelNames: ["result"],
+      }),
     },
     leakedConnectionsCount: register.gauge({
       name: "lodestar_peer_manager_leaked_connections_count",

package/src/network/discv5/utils.ts

@@ -13,9 +13,17 @@ export enum ENRRelevance {
 }
 
 export function enrRelevance(enr: ENR, config: BeaconConfig, clock: IClock): ENRRelevance {
-  // We are not interested in peers that don't advertise at least one transport (tcp or quic)
-  const multiaddrTCP = enr.getLocationMultiaddr(ENRKey.tcp);
-  const multiaddrQUIC = enr.getLocationMultiaddr(ENRKey.quic);
+  // We are not interested in peers that don't advertise at least one transport (tcp or quic).
+  let multiaddrTCP: ReturnType<ENR["getLocationMultiaddr"]>;
+  let multiaddrQUIC: ReturnType<ENR["getLocationMultiaddr"]>;
+  // Malformed ENR endpoint fields can make multiaddr conversion throw.
+  // Treat endpoints that cannot be converted as unusable.
+  try {
+    multiaddrTCP = enr.getLocationMultiaddr(ENRKey.tcp);
+    multiaddrQUIC = enr.getLocationMultiaddr(ENRKey.quic);
+  } catch {
+    return ENRRelevance.no_transport;
+  }
   if (!multiaddrTCP && !multiaddrQUIC) {
     return ENRRelevance.no_transport;
   }

package/src/network/gossip/encoding.ts

@@ -52,6 +52,13 @@ export function fastMsgIdFn(rpcMsg: RPC.Message): string {
 }
 
 export function msgIdToStrFn(msgId: Uint8Array): string {
+  // Spec mandates a 20-byte gossipsub message-id, but control-message IDs from peers
+  // can be any length. Reject non-20-byte IDs to avoid a Buffer.set() RangeError on
+  // longer IDs and stale shared-buffer bytes on shorter ones.
+  // Ref: https://github.com/ethereum/consensus-specs/blob/v1.7.0-alpha.10/specs/phase0/p2p-interface.md?plain=1#L486
+  if (msgId.length !== 20) {
+    throw Error(`Expect msgId to be 20 bytes, got ${msgId.length}`);
+  }
   // this is the same logic to `toHex(msgId)` with better performance
   sharedMsgIdBuf.set(msgId);
   return `0x${sharedMsgIdBuf.toString("hex")}`;

package/src/network/gossip/gossipsub.ts

@@ -35,7 +35,7 @@ import {
   computeGossipPeerScoreParams,
   gossipScoreThresholds,
 } from "./scoringParameters.js";
-import {GossipTopicCache, getCoreTopicsAtFork, stringifyGossipTopic} from "./topic.js";
+import {GossipTopicCache, getAllowedTopics, getCoreTopicsAtFork, stringifyGossipTopic} from "./topic.js";
 
 /** As specified in https://github.com/ethereum/consensus-specs/blob/v1.1.10/specs/phase0/p2p-interface.md */
 const GOSSIPSUB_HEARTBEAT_INTERVAL = 0.7 * 1000;
@@ -140,6 +140,7 @@ export class Eth2Gossipsub {
     const gossipsubInstance = gossipsub({
       globalSignaturePolicy: StrictNoSign,
       allowPublishToZeroTopicPeers: allowPublishToZeroPeers,
+      allowedTopics: getAllowedTopics(networkConfig),
       D: gossipsubD ?? GOSSIP_D,
       Dlo: gossipsubDLow ?? GOSSIP_D_LOW,
       Dhi: gossipsubDHigh ?? GOSSIP_D_HIGH,

package/src/network/gossip/topic.ts

@@ -246,7 +246,7 @@ export function parseGossipTopic(forkDigestContext: ForkDigestContext, topicStr:
 export function getCoreTopicsAtFork(
   networkConfig: NetworkConfig,
   fork: ForkName,
-  opts: {subscribeAllSubnets?: boolean; disableLightClientServer?: boolean}
+  opts: {subscribeAllSubnets?: boolean; subscribeAllColumnSubnets?: boolean; disableLightClientServer?: boolean}
 ): GossipTopicTypeMap[keyof GossipTopicTypeMap][] {
   // Common topics for all forks
   const topics: GossipTopicTypeMap[keyof GossipTopicTypeMap][] = [
@@ -266,7 +266,7 @@ export function getCoreTopicsAtFork(
 
   // After fulu also track data_column_sidecar_{index}
   if (ForkSeq[fork] >= ForkSeq.fulu) {
-    topics.push(...getDataColumnSidecarTopics(networkConfig));
+    topics.push(...getDataColumnSidecarTopics(networkConfig, opts.subscribeAllColumnSubnets));
   }
 
   // After Deneb and before Fulu also track blob_sidecar_{subnet_id}
@@ -309,15 +309,39 @@ export function getCoreTopicsAtFork(
   return topics;
 }
 
+/**
+ * Build the complete set of valid gossip topic strings across every fork boundary.
+ */
+export function getAllowedTopics(networkConfig: NetworkConfig): Set<string> {
+  const {config} = networkConfig;
+  const allowedTopics = new Set<string>();
+
+  for (const boundary of config.forkBoundariesAscendingEpochOrder) {
+    const topics = getCoreTopicsAtFork(networkConfig, boundary.fork, {
+      subscribeAllSubnets: true,
+      subscribeAllColumnSubnets: true,
+      disableLightClientServer: false,
+    });
+    for (const topic of topics) {
+      allowedTopics.add(stringifyGossipTopic(config, {...topic, boundary}));
+    }
+  }
+
+  return allowedTopics;
+}
+
 /**
  * Pick data column subnets to subscribe to post-fulu.
  */
 export function getDataColumnSidecarTopics(
-  networkConfig: NetworkConfig
+  networkConfig: NetworkConfig,
+  subscribeAllColumnSubnets = false
 ): GossipTopicTypeMap[keyof GossipTopicTypeMap][] {
   const topics: GossipTopicTypeMap[keyof GossipTopicTypeMap][] = [];
 
-  const subnets = networkConfig.custodyConfig.sampledSubnets;
+  const subnets = subscribeAllColumnSubnets
+    ? Array.from({length: networkConfig.config.DATA_COLUMN_SIDECAR_SUBNET_COUNT}, (_, i) => i)
+    : networkConfig.custodyConfig.sampledSubnets;
   for (const subnet of subnets) {
     topics.push({type: GossipType.data_column_sidecar, subnet});
   }

package/src/network/peers/peerManager.ts

@@ -415,8 +415,10 @@ export class PeerManager {
       );
       if (irrelevantReasonType === null) {
         isIrrelevant = false;
+        this.metrics?.peerManager.relevanceCheck.inc({result: "relevant"});
       } else {
         isIrrelevant = true;
+        this.metrics?.peerManager.relevanceCheck.inc({result: irrelevantReasonType.code});
         this.logger.debug("Irrelevant peer", {
           peer: prettyPrintPeerId(peer),
           reason: renderIrrelevantPeerType(irrelevantReasonType),
@@ -424,6 +426,7 @@ export class PeerManager {
       }
     } catch (e) {
       this.logger.error("Irrelevant peer - unexpected error", {peer: prettyPrintPeerId(peer)}, e as Error);
+      this.metrics?.peerManager.relevanceCheck.inc({result: "error"});
       isIrrelevant = true;
     }
 
@@ -552,15 +555,18 @@ export class PeerManager {
     for (const peer of connectedPeers) {
       switch (this.peerRpcScores.getScoreState(peer)) {
         case ScoreState.Banned:
+          this.metrics?.peerManager.peersPruned.inc({reason: "banned"});
           void this.goodbyeAndDisconnect(peer, GoodByeReasonCode.BANNED);
           break;
         case ScoreState.Disconnected:
+          this.metrics?.peerManager.peersPruned.inc({reason: "score_too_low"});
           void this.goodbyeAndDisconnect(peer, GoodByeReasonCode.SCORE_TOO_LOW);
           break;
         case ScoreState.Healthy:
           connectedHealthyPeers.push(peer);
       }
     }
+    this.metrics?.peerManager.peersEvaluated.observe(connectedHealthyPeers.length);
 
     const status = this.statusCache.get();
     const starved =
@@ -572,6 +578,7 @@ export class PeerManager {
     this.metrics?.peerManager.starved.set(starved ? 1 : 0);
     const forkSeq = this.config.getForkSeq(this.clock.currentSlot);
 
+    const prioritizeTimer = this.metrics?.peerManager.prioritizePeersDuration.startTimer();
     const {peersToDisconnect, peersToConnect, attnetQueries, syncnetQueries, custodyGroupQueries} = prioritizePeers(
       connectedHealthyPeers.map((peer) => {
         const peerData = this.connectedPeers.get(peer.toString());
@@ -601,6 +608,7 @@ export class PeerManager {
       this.config,
       this.metrics
     );
+    prioritizeTimer?.();
 
     const queriesMerged: SubnetDiscvQueryMs[] = [];
     for (const {type, queries} of [
@@ -632,6 +640,7 @@ export class PeerManager {
     // disconnect first to have more slots before we dial new peers
     for (const [reason, peers] of peersToDisconnect) {
       this.metrics?.peersRequestedToDisconnect.inc({reason}, peers.length);
+      this.metrics?.peerManager.peersPruned.inc({reason}, peers.length);
       for (const peer of peers) {
         void this.goodbyeAndDisconnect(peer, GoodByeReasonCode.TOO_MANY_PEERS);
       }
@@ -659,6 +668,8 @@ export class PeerManager {
       }
     }
 
+    this.metrics?.peerManager.connectedPeersMapSize.set(this.connectedPeers.size);
+
     timer?.();
 
     this.logger.debug("peerManager heartbeat result", {

package/src/network/peers/score/score.ts

@@ -1,4 +1,5 @@
 import {GoodByeReasonCode} from "../../../constants/network.js";
+import {NetworkCoreMetrics} from "../../core/metrics.js";
 import {
   COOL_DOWN_BEFORE_DECAY_MS,
   DEFAULT_SCORE,
@@ -23,13 +24,15 @@ export class RealScore implements IPeerScore {
   /** The final score, computed from the above */
   private score: number;
   private lastUpdate: number;
+  private readonly metrics: NetworkCoreMetrics | null;
 
-  constructor() {
+  constructor(metrics: NetworkCoreMetrics | null = null) {
     this.lodestarScore = DEFAULT_SCORE;
     this.gossipScore = DEFAULT_SCORE;
     this.score = DEFAULT_SCORE;
     this.ignoreNegativeGossipScore = false;
     this.lastUpdate = Date.now();
+    this.metrics = metrics;
   }
 
   isCoolingDown(): boolean {
@@ -136,6 +139,10 @@ export class RealScore implements IPeerScore {
     this.recomputeScore();
     const newState = scoreToState(this.score);
 
+    if (prevState !== newState) {
+      this.metrics?.peerManager.scoreStateTransitions.inc({from: prevState, to: newState});
+    }
+
     if (prevState !== ScoreState.Banned && newState === ScoreState.Banned) {
       // ban this peer for at least BANNED_BEFORE_DECAY_MS seconds
       this.lastUpdate = Date.now() + COOL_DOWN_BEFORE_DECAY_MS;

package/src/network/peers/score/store.ts

@@ -32,7 +32,9 @@ export class PeerRpcScoreStore implements IPeerRpcScoreStore {
   constructor(opts: PeerRpcScoreOpts = {}, metrics: NetworkCoreMetrics | null = null, logger: Logger | null = null) {
     this.metrics = metrics;
     this.logger = logger;
-    this.scores = opts.disablePeerScoring ? new MapDef(() => new MaxScore()) : new MapDef(() => new RealScore());
+    this.scores = opts.disablePeerScoring
+      ? new MapDef(() => new MaxScore())
+      : new MapDef(() => new RealScore(this.metrics));
   }
 
   getScore(peer: PeerId): number {
@@ -75,6 +77,8 @@ export class PeerRpcScoreStore implements IPeerRpcScoreStore {
   }
 
   update(): void {
+    const timer = this.metrics?.peerManager.scoreUpdateDuration.startTimer();
+
     // Bound size of data structures
     pruneSetToMax(this.scores, MAX_ENTRIES);
 
@@ -86,6 +90,9 @@ export class PeerRpcScoreStore implements IPeerRpcScoreStore {
         this.scores.delete(peerIdStr);
       }
     }
+
+    this.metrics?.peerManager.scoreMapSize.set(this.scores.size);
+    timer?.();
   }
 
   updateGossipsubScore(peerId: PeerIdStr, newScore: number, ignore: boolean): void {

package/src/network/peers/utils/prioritizePeers.ts

@@ -7,6 +7,7 @@ import {MapDef} from "@lodestar/utils";
 import {shuffle} from "../../../util/shuffle.js";
 import {sortBy} from "../../../util/sortBy.js";
 import {NetworkCoreMetrics} from "../../core/metrics.js";
+import {SubnetType} from "../../metadata.js";
 import {RequestedSubnet} from "./subnetMap.js";
 
 /** Target number of peers we'd like to have connected to a given long-lived subnet */
@@ -189,6 +190,18 @@ export function prioritizePeers(
 
   const connectedPeerCount = connectedPeers.length;
 
+  // Track the outbound peers invariant (we keep at least OUTBOUND_PEERS_RATIO outbound peers).
+  // Report 0 when there are no peers so the gauge doesn't go stale at its last value.
+  if (metrics) {
+    let outboundPeerCount = 0;
+    for (const peer of connectedPeers) {
+      if (peer.direction === "outbound") {
+        outboundPeerCount++;
+      }
+    }
+    metrics.peerManager.outboundPeersRatio.set(connectedPeerCount > 0 ? outboundPeerCount / connectedPeerCount : 0);
+  }
+
   if (connectedPeerCount < targetPeers) {
     // Need more peers.
     // Instead of attempting to connect the exact amount necessary this will overshoot a little since the success
@@ -256,6 +269,7 @@ function requestSubnetPeers(
 
     for (const {subnet, toSlot} of activeAttnets) {
       const peersInSubnet = peersPerSubnet.get(subnet) ?? 0;
+      metrics?.peerManager.peersPerActiveSubnet.observe({type: SubnetType.attnets}, peersInSubnet);
       if (peersInSubnet < targetSubnetPeers) {
         // We need more peers
         attnetQueries.push({subnet, toSlot, maxPeersToDiscover: targetSubnetPeers - peersInSubnet});
@@ -282,6 +296,7 @@ function requestSubnetPeers(
 
     for (const {subnet, toSlot} of activeSyncnets) {
       const peersInSubnet = peersPerSubnet.get(subnet) ?? 0;
+      metrics?.peerManager.peersPerActiveSubnet.observe({type: SubnetType.syncnets}, peersInSubnet);
       if (peersInSubnet < targetSubnetPeers) {
         // We need more peers
         syncnetQueries.push({subnet, toSlot, maxPeersToDiscover: targetSubnetPeers - peersInSubnet});

package/src/network/processor/gossipHandlers.ts

@@ -165,6 +165,7 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
     // tracked in https://github.com/ChainSafe/lodestar/issues/7957
 
     const logCtx = {
+      slot,
       currentSlot: chain.clock.currentSlot,
       peerId: peerIdStr,
       delaySec,
@@ -184,20 +185,24 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       seenTimestampSec,
       peerIdStr,
     });
+
+    // Optimistically seed the payload-envelope cache too, mirroring seenBlockInputCache above.
+    // This ensures we have PayloadEnvelopeInput, even through "PARENT_UNKNOWN" error
+    // see https://github.com/ChainSafe/lodestar/issues/9475
+    if (isForkPostGloas(fork)) {
+      chain.seenPayloadEnvelopeInputCache.add({
+        blockRootHex,
+        block: signedBlock as SignedBeaconBlock<ForkPostGloas>,
+        forkName: fork,
+        sampledColumns: chain.custodyConfig.sampledColumns,
+        custodyColumns: chain.custodyConfig.custodyColumns,
+        timeCreatedSec: seenTimestampSec,
+      });
+    }
+
     try {
       const {skippedSlots} = await validateGossipBlock(config, chain, signedBlock, fork);
 
-      if (isForkPostGloas(fork)) {
-        chain.seenPayloadEnvelopeInputCache.add({
-          blockRootHex,
-          block: signedBlock as SignedBeaconBlock<ForkPostGloas>,
-          forkName: fork,
-          sampledColumns: chain.custodyConfig.sampledColumns,
-          custodyColumns: chain.custodyConfig.custodyColumns,
-          timeCreatedSec: seenTimestampSec,
-        });
-      }
-
       const blockInputMeta = blockInput.getLogMeta();
 
       const recvToValidation = Date.now() / 1000 - seenTimestampSec;
@@ -234,12 +239,22 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
           throw e;
         }
 
-        if (e.action === GossipAction.REJECT) {
-          chain.persistInvalidSszValue(forkTypes.SignedBeaconBlock, signedBlock, `gossip_reject_slot_${slot}`);
+        // IGNORE means the block is acceptable (e.g. FUTURE_SLOT, ALREADY_KNOWN), just not propagated.
+        // Keep the optimistically-added cache entries; they are pruned on finalization. Only REJECT
+        // (provably invalid) and unexpected errors prune below.
+        if (e.action === GossipAction.IGNORE) {
+          throw e;
         }
+
+        chain.persistInvalidSszValue(forkTypes.SignedBeaconBlock, signedBlock, `gossip_reject_slot_${slot}`);
       }
 
+      // REJECT or unexpected (non-BlockGossipError) error: drop the optimistically-added entries from
+      // both caches, keeping them consistent.
       chain.seenBlockInputCache.prune(blockRootHex);
+      if (isForkPostGloas(fork)) {
+        chain.seenPayloadEnvelopeInputCache.prune(blockRootHex);
+      }
       throw e;
     }
   }
@@ -1127,9 +1142,9 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       const delaySec = chain.clock.secFromSlot(slot, seenTimestampSec);
 
       logger.debug("Received gossip payload envelope", {
+        slot,
         currentSlot: chain.clock.currentSlot,
         peerId: peerIdStr,
-        slot,
         blockRoot: toRootHex(envelope.beaconBlockRoot),
         delaySec,
       });
@@ -1258,9 +1273,8 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       const signedProposerPreferences = sszDeserialize(topic, serializedData);
       await validateGossipProposerPreferences(chain, signedProposerPreferences);
 
-      chain.proposerPreferencesPool.add(signedProposerPreferences);
       chain.emitter.emit(routes.events.EventType.proposerPreferences, {
-        version: ForkName.gloas,
+        version: config.getForkName(signedProposerPreferences.message.proposalSlot),
         data: signedProposerPreferences,
       });
     },

package/src/sync/range/chain.ts

@@ -186,10 +186,6 @@ export class SyncChain {
     this.logger = logger;
     this.logId = `${syncType}-${nextChainId++}`;
 
-    if (metrics) {
-      metrics.syncRange.headSyncPeers.addCollect(() => this.scrapeMetrics(metrics));
-    }
-
     // Trigger event on parent class
     this.sync().then(
       () => fns.onEnd(null, this.target),
@@ -775,7 +771,10 @@ export class SyncChain {
     });
   }
 
-  private scrapeMetrics(metrics: Metrics): void {
+  /**
+   * Called by `RangeSync`'s to avoid collecting metrics of removed chains.
+   */
+  scrapeMetrics(metrics: Metrics): void {
     const syncPeersMetric =
       this.syncType === RangeSyncType.Finalized
         ? metrics.syncRange.finalizedSyncPeers

package/src/sync/range/range.ts

@@ -105,6 +105,16 @@ export class RangeSync extends (EventEmitter as {new (): RangeSyncEmitter}) {
 
     if (metrics) {
       metrics.syncStatus.addCollect(() => this.scrapeMetrics(metrics));
+      metrics.syncRange.headSyncPeers.addCollect(() => {
+        // Gauges retain their last set value, so a removed chain would keep reporting stale
+        // per-column peer counts (finalizedSyncPeers especially, as it is rarely recreated once
+        // synced)
+        metrics.syncRange.headSyncPeers.reset();
+        metrics.syncRange.finalizedSyncPeers.reset();
+        for (const syncChain of this.chains.values()) {
+          syncChain.scrapeMetrics(metrics);
+        }
+      });
     }
   }
 

package/src/sync/utils/downloadByRange.ts

@@ -142,6 +142,7 @@ export function cacheByRangeResponses({
     const blockRootHex = toRootHex(blockRoot);
 
     const existing = updatedBatchBlocks.get(block.message.slot);
+    let blockInput: IBlockInput;
     if (existing) {
       // In practice this code block shouldn't be reached because we shouldn't be refetching a block we already have, see Batch#getRequests.
       // Will throw if root hex does not match (meaning we are following the wrong chain)
@@ -155,8 +156,9 @@ export function cacheByRangeResponses({
         },
         {throwOnDuplicateAdd: false}
       );
+      blockInput = existing;
     } else {
-      const blockInput = cache.getByBlock({
+      blockInput = cache.getByBlock({
         block,
         blockRootHex,
         source,
@@ -165,6 +167,21 @@ export function cacheByRangeResponses({
       });
       updatedBatchBlocks.set(blockInput.slot, blockInput);
     }
+
+    // Seed seenPayloadEnvelopeInputCache the instant we have the block, before the blob loop (or any
+    // later step) can throw and abort the batch — otherwise a gloas block would sit in
+    // seenBlockInputCache but unseeded here, and payload-by-root sync would later throw "Missing
+    // PayloadEnvelopeInput for known block" (see issue #9306). add() is idempotent.
+    if (isForkPostGloas(blockInput.forkName)) {
+      seenPayloadEnvelopeInputCache.add({
+        blockRootHex: blockInput.blockRootHex,
+        block: blockInput.getBlock() as SignedBeaconBlock<ForkPostGloas>,
+        forkName: blockInput.forkName,
+        sampledColumns: custodyConfig.sampledColumns,
+        custodyColumns: custodyConfig.custodyColumns,
+        timeCreatedSec: seenTimestampSec,
+      });
+    }
   }
 
   for (const {blockRoot, blobSidecars} of responses.validatedBlobSidecars ?? []) {
@@ -205,22 +222,6 @@ export function cacheByRangeResponses({
     }
   }
 
-  // Seed seenPayloadEnvelopeInputCache for every gloas block in the batch, regardless of whether
-  // the peer returned its envelope. Without this, a block returned without its envelope would be
-  // imported with no cache entry, and later payload-by-root sync would throw
-  // "Missing PayloadEnvelopeInput for known block" (see issue #9306).
-  for (const blockInput of updatedBatchBlocks.values()) {
-    if (!blockInput.hasBlock() || !isForkPostGloas(blockInput.forkName)) continue;
-    seenPayloadEnvelopeInputCache.add({
-      blockRootHex: blockInput.blockRootHex,
-      block: blockInput.getBlock() as SignedBeaconBlock<ForkPostGloas>,
-      forkName: blockInput.forkName,
-      sampledColumns: custodyConfig.sampledColumns,
-      custodyColumns: custodyConfig.custodyColumns,
-      timeCreatedSec: seenTimestampSec,
-    });
-  }
-
   let payloadEnvelopes: Map<Slot, PayloadEnvelopeInput> | null =
     existingPayloadEnvelopes !== null ? new Map(existingPayloadEnvelopes) : null;
   if (downloadedPayloadEnvelopes !== null) {
@@ -229,8 +230,8 @@ export function cacheByRangeResponses({
       const envelopeBlockRootHex = toRootHex(envelope.message.beaconBlockRoot);
       const payloadInput = seenPayloadEnvelopeInputCache.get(envelopeBlockRootHex);
       if (payloadInput === undefined) {
-        // Unreachable given the loop above seeded an entry for every gloas block in the batch.
-        // for the parent block, it's populated at BeaconChain init
+        // Unreachable given the validatedBlocks loop above seeded an entry for every gloas block in
+        // the batch. for the parent block, it's populated at BeaconChain init
         throw new Error(`Missing PayloadEnvelopeInput for block ${envelopeBlockRootHex}`);
       }
 
@@ -980,6 +981,12 @@ export async function validateColumnsByRangeResponse(
           missingIndices: prettyPrintIndices(request.columns),
         })
       );
+      // Gloas: the bid commits to blobs even for payloads that get orphaned, whose columns are
+      // pruned and unservable. Skip instead of truncating + re-requesting forever; DA is enforced
+      // on envelope processing. Pre-Gloas commitments are canonical, so keep strict.
+      if (isForkPostGloas(forkName)) {
+        continue;
+      }
       break;
     }