@lodestar/beacon-node 1.50.0-dev.456b75a81e → 1.50.0-dev.4b00f7bf39

package/src/execution/engine/utils.ts

@@ -8,6 +8,7 @@ import {
   JsonRpcHttpClientEvent,
   JsonRpcHttpClientEventEmitter,
 } from "./jsonRpcHttpClient.js";
+import {EngineRestError, EngineRestResponseError} from "./restHttpClient.js";
 
 /** QUANTITY as defined in ethereum execution layer JSON RPC https://eth.wiki/json-rpc/API */
 export type QUANTITY = string;
@@ -191,12 +192,21 @@ function getExecutionEngineStateForPayloadError(
     return oldState;
   }
 
+  if (payloadError instanceof EngineRestError && (payloadError.status === 401 || payloadError.status === 403)) {
+    return ExecutionEngineState.AUTH_FAILED;
+  }
+
   // Originally this case was handled with {status: ExecutePayloadStatus.ELERROR}
-  if (payloadError instanceof HttpRpcError || payloadError instanceof ErrorJsonRpcResponse) {
+  if (
+    payloadError instanceof HttpRpcError ||
+    payloadError instanceof ErrorJsonRpcResponse ||
+    payloadError instanceof EngineRestError ||
+    payloadError instanceof EngineRestResponseError
+  ) {
     return ExecutionEngineState.SYNCING;
   }
 
-  if (payloadError && isFetchError(payloadError) && HTTP_FATAL_ERROR_CODES.includes(payloadError.code)) {
+  if (isFetchError(payloadError) && HTTP_FATAL_ERROR_CODES.includes(payloadError.code)) {
     return ExecutionEngineState.OFFLINE;
   }
 

package/src/metrics/metrics/lodestar.ts

@@ -20,6 +20,7 @@ import {InsertOutcome} from "../../chain/opPools/types.js";
 import {RegenCaller, RegenFnName} from "../../chain/regen/interface.js";
 import {ReprocessStatus} from "../../chain/reprocess.js";
 import {RejectReason} from "../../chain/seenCache/seenAttestationData.js";
+import type {ShufflingPromiseCancelReason} from "../../chain/shufflingCache.js";
 import {CacheItemType} from "../../chain/stateCache/types.js";
 import {OpSource} from "../../chain/validatorMonitor.js";
 import type {FlatFileStoreOperation} from "../../db/flatFileStore/metrics.js";
@@ -1176,6 +1177,11 @@ export function createLodestarMetrics(
       help: "The total result of calling notifyForkchoiceUpdate execution engine api",
       labelNames: ["result"],
     }),
+    engineApiRequests: register.counter<{transport: string}>({
+      name: "lodestar_execution_engine_api_requests_total",
+      help: "Total engine API requests by transport",
+      labelNames: ["transport"],
+    }),
     opPool: {
       aggregatedAttestationPool: {
         size: register.gauge({
@@ -1281,6 +1287,10 @@ export function createLodestarMetrics(
             name: "lodestar_oppool_aggregated_attestation_pool_packed_attestations_seen_committees_total",
             help: "Total number of committees for which all members are seen when producing packed attestations",
           }),
+          missingShufflingCommittees: register.gauge({
+            name: "lodestar_oppool_aggregated_attestation_pool_packed_attestations_missing_shuffling_committees_total",
+            help: "Total number of committees skipped due to missing shuffling when producing packed attestations",
+          }),
         },
       },
       attestationPool: {
@@ -1614,6 +1624,19 @@ export function createLodestarMetrics(
         name: "lodestar_shuffling_cache_size",
         help: "Shuffling cache size",
       }),
+      epochs: register.gauge({
+        name: "lodestar_shuffling_cache_epochs",
+        help: "Number of epochs in shuffling cache",
+      }),
+      prunedShufflings: register.counter({
+        name: "lodestar_shuffling_cache_pruned_shufflings_total",
+        help: "Total number of shufflings pruned from shuffling cache",
+      }),
+      cancelledPromises: register.counter<{reason: ShufflingPromiseCancelReason}>({
+        name: "lodestar_shuffling_cache_cancelled_promises_total",
+        help: "Total number of shuffling promises resolved with null",
+        labelNames: ["reason"],
+      }),
       insertPromiseCount: register.gauge({
         name: "lodestar_shuffling_cache_insert_promise_count",
         help: "Total number of times insertPromise is called",
@@ -2038,11 +2061,17 @@ export function createLodestarMetrics(
       }),
       streamTime: register.histogram<{routeId: string}>({
         name: "lodestar_execution_engine_http_client_stream_time_seconds",
-        help: "ExecutionEngineHttp client - streaming time by routeId",
+        help: "ExecutionEngineHttp client - response body consumption time, including failures and excluding parsing",
         labelNames: ["routeId"],
         // Provide max resolution on problematic values around 1 second
         buckets: [0.1, 0.5, 1, 2, 5, 15],
       }),
+      responseParseTime: register.histogram<{routeId: string; encoding: "json" | "ssz"}>({
+        name: "lodestar_execution_engine_http_client_response_parse_time_seconds",
+        help: "ExecutionEngineHttp client - JSON.parse or SSZ deserialize time, including failures, excluding body reads and subsequent value conversion",
+        labelNames: ["routeId", "encoding"],
+        buckets: [0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1],
+      }),
       requestErrors: register.gauge<{routeId: string}>({
         name: "lodestar_execution_engine_http_client_request_errors_total",
         help: "ExecutionEngineHttp client - total count of request errors",

package/src/metrics/metrics.ts

@@ -9,27 +9,42 @@ import {RegistryMetricCreator} from "./utils/registryMetricCreator.js";
 
 export type Metrics = BeaconMetrics &
   ForkChoiceMetrics &
-  BeaconStateTransitionMetrics &
-  LodestarMetrics & {register: RegistryMetricCreator; close: () => void};
+  LodestarMetrics & {
+    /** Null when we use native state transition metrics instead */
+    stateTransition: BeaconStateTransitionMetrics | null;
+    register: RegistryMetricCreator;
+    close: () => void;
+  };
+
+export type CreateMetricsOptions = {
+  /*
+   * Toggle inclusion of state transition metrics.
+   *
+   * False when native state transition is used, so we
+   * can grab metrics from the native implementation instead.
+   * */
+  includeStateTransitionMetrics?: boolean;
+  collectNodeMetrics?: boolean;
+};
 
 export function createMetrics(
   opts: MetricsOptions,
   genesisTime: number,
   externalRegistries: Registry[] = [],
-  {collectNodeMetrics = true}: {collectNodeMetrics?: boolean} = {}
+  createOpts: CreateMetricsOptions = {}
 ): Metrics {
   const register = new RegistryMetricCreator();
   const beacon = createBeaconMetrics(register);
   const forkChoice = getForkChoiceMetrics(register);
   const lodestar = createLodestarMetrics(register, opts.metadata, genesisTime);
-  const stateTransition = getMetrics(register);
+  const stateTransition = createOpts.includeStateTransitionMetrics === false ? null : getMetrics(register);
 
   const onUnhandledRejection = (_error: unknown): void => {
     lodestar.unhandledPromiseRejections.inc();
   };
   process.on("unhandledRejection", onUnhandledRejection);
 
-  const nodeJsMetricsClose = collectNodeMetrics ? collectNodeJSMetrics(register) : (): void => {};
+  const nodeJsMetricsClose = createOpts.collectNodeMetrics === false ? (): void => {} : collectNodeJSMetrics(register);
   const close = (): void => {
     process.removeListener("unhandledRejection", onUnhandledRejection);
     nodeJsMetricsClose();
@@ -46,7 +61,7 @@ export function createMetrics(
     ...beacon,
     ...forkChoice,
     ...lodestar,
-    ...stateTransition,
+    stateTransition,
     register,
     close,
   };

package/src/network/core/events.ts

@@ -1,21 +1,30 @@
 import EventEmitter from "node:events";
 import {ResponseIncoming, ResponseOutgoing} from "@lodestar/reqresp";
-import {AsyncIterableEventBus, IteratorEvent, RequestEvent} from "../../util/asyncIterableToEvents.js";
+import {
+  AsyncIterableEventBus,
+  IteratorEvent,
+  RequestCancelEvent,
+  RequestEvent,
+} from "../../util/asyncIterableToEvents.js";
 import {StrictEventEmitterSingleArg} from "../../util/strictEvents.js";
 import {EventDirection} from "../events.js";
 import {IncomingRequestArgs, OutgoingRequestArgs} from "../reqresp/types.js";
 
 export enum ReqRespBridgeEvent {
   outgoingRequest = "reqresp.outgoingRequest",
+  outgoingRequestCancel = "reqresp.outgoingRequestCancel",
   outgoingResponse = "reqresp.outgoingResponse",
   incomingRequest = "reqresp.incomingRequest",
+  incomingRequestCancel = "reqresp.incomingRequestCancel",
   incomingResponse = "reqresp.incomingResponse",
 }
 
 export type ReqRespBridgeEventData = {
   [ReqRespBridgeEvent.outgoingRequest]: RequestEvent<OutgoingRequestArgs>;
+  [ReqRespBridgeEvent.outgoingRequestCancel]: RequestCancelEvent;
   [ReqRespBridgeEvent.outgoingResponse]: IteratorEvent<ResponseOutgoing>;
   [ReqRespBridgeEvent.incomingRequest]: RequestEvent<IncomingRequestArgs>;
+  [ReqRespBridgeEvent.incomingRequestCancel]: RequestCancelEvent;
   [ReqRespBridgeEvent.incomingResponse]: IteratorEvent<ResponseIncoming>;
 };
 
@@ -26,8 +35,10 @@ export class ReqRespBridgeEventBus extends (EventEmitter as {new (): IReqRespBri
 // NOTE: If the same event is on this two arrays it can create an infinite cycle
 export const reqRespBridgeEventDirection: Record<ReqRespBridgeEvent, EventDirection> = {
   [ReqRespBridgeEvent.outgoingRequest]: EventDirection.mainToWorker,
+  [ReqRespBridgeEvent.outgoingRequestCancel]: EventDirection.mainToWorker,
   [ReqRespBridgeEvent.outgoingResponse]: EventDirection.mainToWorker,
   [ReqRespBridgeEvent.incomingRequest]: EventDirection.workerToMain,
+  [ReqRespBridgeEvent.incomingRequestCancel]: EventDirection.workerToMain,
   [ReqRespBridgeEvent.incomingResponse]: EventDirection.workerToMain,
 };
 
@@ -36,8 +47,10 @@ export function getReqRespBridgeReqEvents(
 ): AsyncIterableEventBus<OutgoingRequestArgs, ResponseIncoming> {
   return {
     emitRequest: (data) => events.emit(ReqRespBridgeEvent.outgoingRequest, data),
+    emitRequestCancel: (data) => events.emit(ReqRespBridgeEvent.outgoingRequestCancel, data),
     emitResponse: (data) => events.emit(ReqRespBridgeEvent.incomingResponse, data),
     onRequest: (cb) => events.on(ReqRespBridgeEvent.outgoingRequest, cb),
+    onRequestCancel: (cb) => events.on(ReqRespBridgeEvent.outgoingRequestCancel, cb),
     onResponse: (cb) => events.on(ReqRespBridgeEvent.incomingResponse, cb),
   };
 }
@@ -47,8 +60,10 @@ export function getReqRespBridgeRespEvents(
 ): AsyncIterableEventBus<IncomingRequestArgs, ResponseOutgoing> {
   return {
     emitRequest: (data) => events.emit(ReqRespBridgeEvent.incomingRequest, data),
+    emitRequestCancel: (data) => events.emit(ReqRespBridgeEvent.incomingRequestCancel, data),
     emitResponse: (data) => events.emit(ReqRespBridgeEvent.outgoingResponse, data),
     onRequest: (cb) => events.on(ReqRespBridgeEvent.incomingRequest, cb),
+    onRequestCancel: (cb) => events.on(ReqRespBridgeEvent.incomingRequestCancel, cb),
     onResponse: (cb) => events.on(ReqRespBridgeEvent.outgoingResponse, cb),
   };
 }

package/src/network/processor/gossipHandlers.ts

@@ -12,9 +12,10 @@ import {
   isForkPostElectra,
   isForkPostGloas,
 } from "@lodestar/params";
-import {computeTimeAtSlot} from "@lodestar/state-transition";
+import {computeTimeAtSlot, signedBlockToSignedHeader} from "@lodestar/state-transition";
 import {
   Root,
+  RootHex,
   SignedBeaconBlock,
   SingleAttestation,
   Slot,
@@ -78,6 +79,7 @@ import {
   validateGossipSyncCommittee,
   validateGossipVoluntaryExit,
   validateSyncCommitteeGossipContributionAndProof,
+  verifyBlockProposerSignature,
 } from "../../chain/validation/index.js";
 import {validateLightClientFinalityUpdate} from "../../chain/validation/lightClientFinalityUpdate.js";
 import {validateLightClientOptimisticUpdate} from "../../chain/validation/lightClientOptimisticUpdate.js";
@@ -148,37 +150,16 @@ export function getGossipHandlers(modules: ValidatorFnsModules, options: GossipH
 function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHandlerOpts): SequentialGossipHandlers {
   const {chain, config, metrics, logger, core} = modules;
 
-  async function validateBeaconBlock(
+  /**
+   * Add a gossip block to the seen caches. The proposer signature of the block MUST be verified.
+   */
+  function cacheSignatureVerifiedBlock(
     signedBlock: SignedBeaconBlock,
+    blockRootHex: RootHex,
     fork: ForkName,
     peerIdStr: string,
     seenTimestampSec: number
-  ): Promise<IBlockInput> {
-    const slot = signedBlock.message.slot;
-    const forkTypes = config.getForkTypes(slot);
-    const blockRootHex = toRootHex(forkTypes.BeaconBlock.hashTreeRoot(signedBlock.message));
-    const delaySec = chain.clock.secFromSlot(slot, seenTimestampSec);
-    const recvToValLatency = Date.now() / 1000 - seenTimestampSec;
-
-    // always set block to seen cache for all forks so that we don't need to download it
-    // TODO: validate block before adding to cache
-    // tracked in https://github.com/ChainSafe/lodestar/issues/7957
-
-    const logCtx = {
-      slot,
-      root: blockRootHex,
-      currentSlot: chain.clock.currentSlot,
-      peerId: peerIdStr,
-      delaySec,
-      recvToValLatency,
-    };
-
-    logger.debug("Received gossip block", {...logCtx});
-
-    // optimistically add gossip block to the seen cache
-    // if validation fails, we will NOT forward this gossip block to peers
-    //   - if PARENT_BLOCK_UNKNOWN error, blockInput will then be queued inside BlockInputSync. If the gossip block is really invalid, it will be pruned there
-    //   - if other validator errors, blockInput will stay in the seen cache and will be pruned on finalization
+  ): BlockInput {
     const blockInput = chain.seenBlockInputCache.getByBlock({
       block: signedBlock,
       blockRootHex,
@@ -187,8 +168,7 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       peerIdStr,
     });
 
-    // Optimistically seed the payload-envelope cache too, mirroring seenBlockInputCache above.
-    // This ensures we have PayloadEnvelopeInput, even through "PARENT_BLOCK_UNKNOWN" error
+    // Also seed the payload-envelope cache, so we have PayloadEnvelopeInput even through "PARENT_BLOCK_UNKNOWN" error
     // see https://github.com/ChainSafe/lodestar/issues/9475
     if (isForkPostGloas(fork)) {
       chain.seenPayloadEnvelopeInputCache.add({
@@ -202,8 +182,37 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       });
     }
 
+    return blockInput;
+  }
+
+  async function validateBeaconBlock(
+    signedBlock: SignedBeaconBlock,
+    fork: ForkName,
+    peerIdStr: string,
+    seenTimestampSec: number
+  ): Promise<IBlockInput> {
+    const slot = signedBlock.message.slot;
+    const forkTypes = config.getForkTypes(slot);
+    const blockRootHex = toRootHex(forkTypes.BeaconBlock.hashTreeRoot(signedBlock.message));
+    const delaySec = chain.clock.secFromSlot(slot, seenTimestampSec);
+    const recvToValLatency = Date.now() / 1000 - seenTimestampSec;
+
+    const logCtx = {
+      slot,
+      root: blockRootHex,
+      currentSlot: chain.clock.currentSlot,
+      peerId: peerIdStr,
+      delaySec,
+      recvToValLatency,
+    };
+
+    logger.debug("Received gossip block", {...logCtx});
+
+    chain.seenBlockInputCache.markValidatingBlock(blockRootHex);
+
     try {
       const {skippedSlots} = await validateGossipBlock(config, chain, signedBlock, fork);
+      const blockInput = cacheSignatureVerifiedBlock(signedBlock, blockRootHex, fork, peerIdStr, seenTimestampSec);
 
       const blockInputMeta = blockInput.getLogMeta();
 
@@ -229,54 +238,54 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
       if (e instanceof BlockGossipError) {
         logger.debug("Gossip block has error", {slot, root: blockRootHex, code: e.type.code});
         if (
-          (e.type.code === BlockErrorCode.PARENT_BLOCK_UNKNOWN ||
-            e.type.code === BlockErrorCode.PARENT_PAYLOAD_UNKNOWN) &&
-          blockInput
+          e.type.code === BlockErrorCode.PARENT_BLOCK_UNKNOWN ||
+          e.type.code === BlockErrorCode.PARENT_PAYLOAD_UNKNOWN
         ) {
+          // The block is retained for unknown parent sync without full validation. Authenticate it first, this
+          // does not need the parent state
+          const proposerIndex = signedBlock.message.proposerIndex;
+          await verifyBlockProposerSignature(chain, signedBlock, blockRootHex);
+          chain.seenBlockProposers.observeBlockRoot(
+            slot,
+            proposerIndex,
+            blockRootHex,
+            signedBlockToSignedHeader(config, signedBlock)
+          );
+
+          // Check again after the signature verification so concurrent proposals cannot both be retained
+          if (chain.seenBlockProposers.isKnown(slot, proposerIndex)) {
+            if (chain.seenBlockProposers.isRepeatProposal(slot, proposerIndex, blockRootHex)) {
+              throw new BlockGossipError(GossipAction.IGNORE, {
+                code: BlockErrorCode.REPEAT_PROPOSAL,
+                proposerIndex,
+                root: blockRootHex,
+              });
+            }
+            throw new BlockGossipError(GossipAction.IGNORE, {code: BlockErrorCode.ALREADY_KNOWN, root: blockRootHex});
+          }
+          chain.seenBlockProposers.add(slot, proposerIndex, blockRootHex);
+
+          const blockInput = cacheSignatureVerifiedBlock(signedBlock, blockRootHex, fork, peerIdStr, seenTimestampSec);
           chain.emitter.emit(ChainEvent.blockUnknownParent, {
             blockInput,
             peer: peerIdStr,
             source: BlockInputSource.gossip,
           });
-          // throw error (don't prune the blockInput)
           throw e;
         }
 
-        // 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), unexpected errors and repeat proposals that are not imported prune.
-        if (e.action === GossipAction.IGNORE) {
-          // Only a signature-verified sibling is imported by the beacon_block handler, any other repeat proposal
-          // is dropped from the caches and re-downloaded by sync if it ever becomes relevant. Its signature was not
-          // verified, so only its own entry is removed, never its claimed ancestors
-          if (
-            e.type.code === BlockErrorCode.REPEAT_PROPOSAL &&
-            !chain.seenBlockProposers.hasBlockRoot(slot, signedBlock.message.proposerIndex, blockRootHex)
-          ) {
-            chain.seenBlockInputCache.remove(blockRootHex);
-            if (isForkPostGloas(fork)) {
-              chain.seenPayloadEnvelopeInputCache.remove(blockRootHex);
-            }
-          }
-          throw e;
+        if (e.action === GossipAction.REJECT) {
+          chain.persistInvalidSszValue(
+            forkTypes.SignedBeaconBlock,
+            signedBlock,
+            `gossip_reject_slot_${slot}`,
+            blockRootHex
+          );
         }
-
-        chain.persistInvalidSszValue(
-          forkTypes.SignedBeaconBlock,
-          signedBlock,
-          `gossip_reject_slot_${slot}`,
-          blockRootHex
-        );
-      }
-
-      // REJECT or unexpected (non-BlockGossipError) error: drop the optimistically-added entries from
-      // both caches, keeping them consistent. The block may carry any parent root, so only its own entry is removed
-      chain.seenBlockInputCache.remove(blockRootHex);
-      if (isForkPostGloas(fork)) {
-        chain.seenPayloadEnvelopeInputCache.remove(blockRootHex);
       }
       throw e;
     } finally {
+      chain.seenBlockInputCache.unmarkValidatingBlock(blockRootHex);
       // The block received from the network may have established an equivocation, either by conflicting
       // with a previously observed block root (REPEAT_PROPOSAL) or with a root observed during validation
       const proposerIndex = signedBlock.message.proposerIndex;
@@ -752,18 +761,23 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
           e.type.code === BlockErrorCode.REPEAT_PROPOSAL &&
           // Only a signature-verified sibling recorded in the seen cache is imported. The cache holds at most two
           // roots per proposer and slot, which bounds full imports over gossip to one alternate
-          chain.seenBlockProposers.hasBlockRoot(signedBlock.message.slot, e.type.proposerIndex, e.type.root)
+          chain.seenBlockProposers.hasBlockRoot(signedBlock.message.slot, e.type.proposerIndex, e.type.root) &&
+          // A sibling with an unknown parent cannot be imported, sync fetches it if its branch becomes relevant
+          chain.forkChoice.getBlockHexDefaultStatus(toRootHex(signedBlock.message.parentRoot)) !== null
         ) {
-          // blockInput was optimistically seeded in validateBeaconBlock and retained on IGNORE
-          const blockInput = chain.seenBlockInputCache.get(e.type.root);
-          if (blockInput) {
-            // We're returning IGNORE (thrown below), so this block is not forwarded to peers. Unlike the
-            // happy path there is no rush to forward, so we don't need to wrap in callInNextEventLoop.
-            chain.serializedCache.set(signedBlock, serializedData);
-            // this is technically not a valid gossip block but gossip validation is a cheap subset of checks
-            // this runs the full state transition, so importing an equivocating-but-valid block here is safe.
-            handleValidBeaconBlock(blockInput, peerIdStr, seenTimestampSec);
-          }
+          const blockInput = cacheSignatureVerifiedBlock(
+            signedBlock,
+            e.type.root,
+            topic.boundary.fork,
+            peerIdStr,
+            seenTimestampSec
+          );
+          // We're returning IGNORE (thrown below), so this block is not forwarded to peers. Unlike the
+          // happy path there is no rush to forward, so we don't need to wrap in callInNextEventLoop.
+          chain.serializedCache.set(signedBlock, serializedData);
+          // this is technically not a valid gossip block but gossip validation is a cheap subset of checks
+          // this runs the full state transition, so importing an equivocating-but-valid block here is safe.
+          handleValidBeaconBlock(blockInput, peerIdStr, seenTimestampSec);
         }
         // rethrow so gossipValidatorFn maps IGNORE -> TopicValidatorResult.Ignore (message not forwarded)
         throw e;
@@ -1341,6 +1355,9 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
                 case PayloadErrorCode.ENVELOPE_VERIFICATION_ERROR:
                 case PayloadErrorCode.EXECUTION_ENGINE_INVALID:
                   core.reportPeer(peerIdStr, PeerAction.LowToleranceError, "BadGossipPayload");
+                  // The builder may have signed another envelope that is valid, keeping this one would make
+                  // by-root and range sync reuse it and never import the payload
+                  chain.seenPayloadEnvelopeInputCache.removeInvalid(payloadInput);
                   // Misbehaving peer, but could highlight an issue in another client
                   logLevel = LogLevel.warn;
                   break;
@@ -1378,7 +1395,7 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
           const delaySec = chain.clock.secFromSlot(payloadAttestationMessage.data.slot, seenTimestampSec);
           metrics?.gossipPayloadAttestationMessage.elapsedTimeTillReceived.observe({source: OpSource.gossip}, delaySec);
           const nextSlotProposer = chain.getHeadState().getBeaconProposer(payloadAttestationMessage.data.slot + 1);
-          const isNextSlotProposer = chain.beaconProposerCache.get(nextSlotProposer) !== undefined;
+          const isNextSlotProposer = chain.beaconProposerCache.has(nextSlotProposer);
 
           if (isNextSlotProposer) {
             try {

package/src/network/processor/gossipValidatorFn.ts

@@ -20,6 +20,7 @@ import {
   VoluntaryExitErrorCode,
 } from "../../chain/errors/index.js";
 import {Metrics} from "../../metrics/index.js";
+import {PeerIdStr} from "../../util/peerId.js";
 import {INetworkCore} from "../core/index.js";
 import {
   BatchGossipHandlerFn,
@@ -40,6 +41,8 @@ export type ValidatorFnModules = {
   core: INetworkCore;
 };
 
+export type OnFatalPeerFn = (peer: PeerIdStr) => void;
+
 type RejectPeerActionRule = {default: PeerAction; byCode?: Record<string, PeerAction>};
 
 /**
@@ -123,7 +126,8 @@ export function rejectPeerAction(type: GossipType, code: string): PeerAction {
  */
 export function getGossipValidatorBatchFn(
   gossipHandlers: GossipHandlers,
-  modules: ValidatorFnModules
+  modules: ValidatorFnModules,
+  onFatalPeer: OnFatalPeerFn
 ): GossipValidatorBatchFn {
   const {logger, metrics, core} = modules;
 
@@ -173,6 +177,7 @@ export function getGossipValidatorBatchFn(
             metrics?.networkProcessor.gossipAttestationRejectByReason.inc({reason: e.type.code});
             const peerAction = rejectPeerAction(type, e.type.code);
             core.reportPeer(propagationSource, peerAction, e.type.code);
+            if (peerAction === PeerAction.Fatal) onFatalPeer(propagationSource);
             logger.debug(
               `Gossip validation ${type} rejected`,
               {peer: propagationSource, clientAgent, clientVersion, peerAction},
@@ -208,7 +213,11 @@ export function getGossipValidatorBatchFn(
  *
  * @see getGossipHandlers for reasoning on why GossipHandlerFn are used for gossip validation.
  */
-export function getGossipValidatorFn(gossipHandlers: GossipHandlers, modules: ValidatorFnModules): GossipValidatorFn {
+export function getGossipValidatorFn(
+  gossipHandlers: GossipHandlers,
+  modules: ValidatorFnModules,
+  onFatalPeer: OnFatalPeerFn
+): GossipValidatorFn {
   const {logger, metrics, core} = modules;
 
   return async function gossipValidatorFn({
@@ -260,6 +269,7 @@ export function getGossipValidatorFn(gossipHandlers: GossipHandlers, modules: Va
           metrics?.networkProcessor.gossipValidationReject.inc({topic: type});
           const peerAction = rejectPeerAction(type, e.type.code);
           core.reportPeer(propagationSource, peerAction, e.type.code);
+          if (peerAction === PeerAction.Fatal) onFatalPeer(propagationSource);
           logger.debug(
             `Gossip validation ${type} rejected`,
             {peer: propagationSource, clientAgent, clientVersion, peerAction},

package/src/network/processor/index.ts

@@ -1,6 +1,6 @@
 import {TopicValidatorResult} from "@libp2p/gossipsub";
 import {routes} from "@lodestar/api";
-import {ForkSeq, NUMBER_OF_COLUMNS} from "@lodestar/params";
+import {ForkSeq, NUMBER_OF_COLUMNS, isForkPostGloas} from "@lodestar/params";
 import {computeStartSlotAtEpoch} from "@lodestar/state-transition";
 import {RootHex, Slot, SlotRootHex} from "@lodestar/types";
 import {Logger, MapDef, mapValues, sleep} from "@lodestar/utils";
@@ -108,6 +108,12 @@ type SearchedRootEntry = {
  */
 const DEFAULT_EARLIEST_PERMISSIBLE_SLOT_DISTANCE = 32;
 
+/**
+ * The network worker bans a Fatal peer only on its next PeerManager heartbeat (30s). Until then the main thread
+ * would keep validating that peer's messages, so drop them here. 3 slots (36s) outlasts one heartbeat.
+ */
+export const FATAL_PEER_EXPIRY_SLOTS = 3;
+
 type WorkOpts = {
   bypassQueue?: boolean;
 };
@@ -263,6 +269,8 @@ export class NetworkProcessor {
     () => new MapDef<RootHex, SearchedRootEntry>(() => ({}))
   );
   private bufferedRootsBySlot = new MapDef<Slot, Set<RootHex>>(() => new Set());
+  /** peer -> clock slot when it was Fatal'ed */
+  private readonly fatalPeers = new Map<PeerIdStr, Slot>();
 
   constructor(
     modules: NetworkProcessorModules,
@@ -277,10 +285,15 @@ export class NetworkProcessor {
     this.events = events;
     this.gossipQueues = createGossipQueues();
     this.gossipTopicConcurrency = mapValues(this.gossipQueues, () => 0);
-    this.gossipValidatorFn = getGossipValidatorFn(modules.gossipHandlers ?? getGossipHandlers(modules, opts), modules);
+    this.gossipValidatorFn = getGossipValidatorFn(
+      modules.gossipHandlers ?? getGossipHandlers(modules, opts),
+      modules,
+      this.onFatalPeer
+    );
     this.gossipValidatorBatchFn = getGossipValidatorBatchFn(
       modules.gossipHandlers ?? getGossipHandlers(modules, opts),
-      modules
+      modules,
+      this.onFatalPeer
     );
 
     events.on(NetworkEvent.pendingGossipsubMessage, this.onPendingGossipsubMessage);
@@ -414,8 +427,18 @@ export class NetworkProcessor {
     }
   }
 
+  private onFatalPeer = (peer: PeerIdStr): void => {
+    this.fatalPeers.set(peer, this.chain.clock.currentSlot);
+  };
+
   private onPendingGossipsubMessage = (message: PendingGossipsubMessage): void => {
     const topicType = message.topic.type;
+    if (this.fatalPeers.has(message.propagationSource)) {
+      this.metrics?.networkProcessor.gossipValidationError.inc({topic: topicType, error: GossipErrorCode.FATAL_PEER});
+      // No need to report the dropped job to gossip. It will be eventually pruned from the mcache
+      return;
+    }
+
     const extractBlockSlotRootFn = this.extractBlockSlotRootFns[topicType];
     if (extractBlockSlotRootFn === undefined) {
       // some messages don't have slot and root
@@ -505,9 +528,9 @@ export class NetworkProcessor {
       const parentRoot = getParentRootFromSignedBeaconBlockSerialized(message.msg.data);
       if (parentRoot) {
         let search: SearchUnknownRootTarget | null = null;
-        if (ForkSeq[fork] >= ForkSeq.gloas) {
+        if (isForkPostGloas(fork)) {
           // GLOAS: also check parent envelope, same logic as execution_payload_bid
-          const parentBlockHash = getParentBlockHashFromGloasSignedBeaconBlockSerialized(message.msg.data);
+          const parentBlockHash = getParentBlockHashFromGloasSignedBeaconBlockSerialized(message.msg.data, fork);
           if (parentBlockHash && !this.chain.forkChoice.getBlockHexAndBlockHash(parentRoot, parentBlockHash)) {
             const protoBlock = this.chain.forkChoice.getBlockHexDefaultStatus(parentRoot);
             if (protoBlock === null) {
@@ -898,6 +921,10 @@ export class NetworkProcessor {
     for (const slot of this.bufferedRootsBySlot.keys()) {
       if (slot <= minSlot) this.bufferedRootsBySlot.delete(slot);
     }
+
+    for (const [peer, slot] of this.fatalPeers) {
+      if (clockSlot - slot > FATAL_PEER_EXPIRY_SLOTS) this.fatalPeers.delete(peer);
+    }
   };
 
   private executeWork(): void {
@@ -953,6 +980,32 @@ export class NetworkProcessor {
   private async processPendingGossipsubMessage(
     messageOrArray: PendingGossipsubMessage | PendingGossipsubMessage[]
   ): Promise<void> {
+    // drop messages from peers Fatal'ed after these were queued or started awaiting a block/envelope
+    // No need to report the dropped jobs to gossip. They will be eventually pruned from the mcache
+    if (this.fatalPeers.size > 0) {
+      if (Array.isArray(messageOrArray)) {
+        const allowed: PendingGossipsubMessage[] = [];
+        for (const msg of messageOrArray) {
+          if (this.fatalPeers.has(msg.propagationSource)) {
+            this.metrics?.networkProcessor.gossipValidationError.inc({
+              topic: msg.topic.type,
+              error: GossipErrorCode.FATAL_PEER,
+            });
+          } else {
+            allowed.push(msg);
+          }
+        }
+        if (allowed.length === 0) return;
+        messageOrArray = allowed;
+      } else if (this.fatalPeers.has(messageOrArray.propagationSource)) {
+        this.metrics?.networkProcessor.gossipValidationError.inc({
+          topic: messageOrArray.topic.type,
+          error: GossipErrorCode.FATAL_PEER,
+        });
+        return;
+      }
+    }
+
     const nowSec = Date.now() / 1000;
     if (Array.isArray(messageOrArray)) {
       for (const msg of messageOrArray) {

package/src/network/reqresp/handlers/executionPayloadEnvelopesByRoot.ts

@@ -53,7 +53,7 @@ export async function* onExecutionPayloadEnvelopesByRoot(
     const batch = requests.slice(i, i + MAX_BODIES_PER_REQUEST);
     let envelopesBytes: (Uint8Array | null)[];
     try {
-      // By-root allows omission, so a mismatched envelope is left out rather than failing the response.
+      // By-root allows omission, so an envelope that cannot be rebuilt is left out rather than failing the response.
       envelopesBytes = await chain.getSerializedExecutionPayloadEnvelopes(batch, "omit");
     } catch (e) {
       if (e instanceof EnvelopeReconstructionError) {