@lodestar/beacon-node 1.51.0-dev.c99cf49add → 1.51.0-dev.d2440f7e84

package/src/api/impl/validator/index.ts

@@ -18,6 +18,7 @@ import {
   isForkPostElectra,
   isForkPostFulu,
   isForkPostGloas,
+  isForkPostHeze,
 } from "@lodestar/params";
 import {
   DataAvailabilityStatus,
@@ -28,6 +29,7 @@ import {
   computeStartSlotAtEpoch,
   computeTimeAtSlot,
   getCurrentSlot,
+  getInclusionListCommittee,
   isStatePostAltair,
   isStatePostGloas,
   proposerShufflingDecisionRoot,
@@ -68,6 +70,7 @@ import {
 } from "@lodestar/utils";
 import {MAX_BUILDER_BOOST_FACTOR} from "@lodestar/validator";
 import {BlockInputSource} from "../../../chain/blocks/blockInput/types.js";
+import {InclusionListSource} from "../../../chain/blocks/types.js";
 import {
   AttestationError,
   AttestationErrorCode,
@@ -83,11 +86,15 @@ import {PREPARE_NEXT_SLOT_BPS} from "../../../chain/prepareNextSlot.js";
 import {BlockType, ProduceFullDeneb, ProduceFullGloas} from "../../../chain/produceBlock/index.js";
 import {RegenCaller} from "../../../chain/regen/index.js";
 import {CheckpointHex} from "../../../chain/stateCache/types.js";
-import {validateApiAggregateAndProof} from "../../../chain/validation/index.js";
+import {
+  InclusionListValidationResult,
+  validateApiAggregateAndProof,
+  validateGossipInclusionList,
+} from "../../../chain/validation/index.js";
 import {validateGossipProposerPreferences} from "../../../chain/validation/proposerPreferences.js";
 import {validateSyncCommitteeGossipContributionAndProof} from "../../../chain/validation/syncCommitteeContributionAndProof.js";
 import {ZERO_HASH} from "../../../constants/index.js";
-import {BUILDER_BID_DEADLINE_MS, BuilderApiBid, decodeBuilderUrl} from "../../../execution/builder/apiClient.js";
+import {BuilderApiBid, decodeBuilderUrl} from "../../../execution/builder/apiClient.js";
 import {BuilderStatus, NoBidReceived} from "../../../execution/builder/http.js";
 import {getBuilderBidTotalGwei, validateBuilderApiExecutionPayloadBid} from "../../../execution/builder/validateBid.js";
 import {validateGossipFnRetryUnknownRoot} from "../../../network/processor/gossipHandlers.js";
@@ -116,10 +123,14 @@ import {
  * A cutoff of 1667 BPS (2 seconds for 12s slots) gives enough time and if there are unexpected delays
  * it ensures we publish in time as proposals after the attestation deadline will likely be orphaned
  * due to proposer boost reorg.
- *
- * TODO GLOAS: re-evaluate cutoff timing due to attestation deadline changes in gloas
  */
 const BLOCK_PRODUCTION_RACE_CUTOFF_BPS = 1667;
+/**
+ * Post-gloas cutoff, kept at 1667 BPS (2 seconds for 12s slots) for now.
+ *
+ * TODO GLOAS: revisit, the block propagates faster as it only commits to a bid and the attestation deadline changed
+ */
+export const BLOCK_PRODUCTION_RACE_CUTOFF_GLOAS_BPS = 1667;
 /** Rejection message of the bid block branch when there is no viable bid to commit to */
 const NO_BID_AVAILABLE = "No builder bid available";
 
@@ -247,13 +258,13 @@ export function getValidatorApi(
         genesisBlockRoot = state.computeAnchorCheckpoint().checkpoint.root;
       } else if (state.slot < SLOTS_PER_HISTORICAL_ROOT) {
         genesisBlockRoot = state.getBlockRootAtSlot(GENESIS_SLOT);
-      }
-
-      const blockRes = await chain.getCanonicalBlockAtSlot(GENESIS_SLOT);
-      if (blockRes) {
-        genesisBlockRoot = config
-          .getForkTypes(blockRes.block.message.slot)
-          .SignedBeaconBlock.hashTreeRoot(blockRes.block);
+      } else {
+        const blockRes = await chain.getCanonicalBlockAtSlot(GENESIS_SLOT);
+        if (blockRes) {
+          genesisBlockRoot = config
+            .getForkTypes(blockRes.block.message.slot)
+            .BeaconBlock.hashTreeRoot(blockRes.block.message);
+        }
       }
     }
 
@@ -998,33 +1009,50 @@ export function getValidatorApi(
         (!circuitBreakerActive &&
           chain.executionPayloadBidPool.getBestBid(slot, bidParentBlockHash, parentBlockRootHex) !== null);
 
-      // Select the p2p bid once builders had time to bid up, matching the deadline advertised
-      // on builder API bid requests, unless the circuit breaker is active
-      const p2pBidPromise: Promise<PooledExecutionPayloadBid | null> = circuitBreakerActive
-        ? Promise.resolve(null)
-        : sleep(Math.max(0, BUILDER_BID_DEADLINE_MS - chain.clock.msFromSlot(slot))).then(() => {
-            const p2pBid = chain.executionPayloadBidPool.getBestBid(slot, bidParentBlockHash, parentBlockRootHex);
-            // Discard p2p bids below the proposer's configured floor on the total payment.
-            // A p2p bid's total is just its value since gossip validation enforces executionPayment=0.
-            if (p2pBid !== null && BigInt(p2pBid.signedBid.message.value) < builderConfig.minBid) {
-              logger.info("Best p2p bid below configured minimum", {
-                slot,
-                bidValue: prettyGweiToEth(p2pBid.signedBid.message.value),
-                minBid: prettyGweiToEth(builderConfig.minBid),
-              });
-              return null;
-            }
-            return p2pBid;
-          });
+      // Calculate cutoff time based on start of the slot, ensures a slow local payload build does
+      // not delay the proposal when a builder bid block is available (and vice versa)
+      const cutoffMs = Math.max(
+        0,
+        config.getSlotComponentDurationMs(fork, BLOCK_PRODUCTION_RACE_CUTOFF_GLOAS_BPS) - chain.clock.msFromSlot(slot)
+      );
+
+      // Give p2p bids the same time to bid up as granted to builder API bid requests via the advertised
+      // deadline, the pool is read once the deadline passed and the bid requests settled. A late request
+      // does not wait past the race cutoff, a pooled bid is still the fastest fallback if the local block fails
+      const bidDeadlinePromise = circuitBreakerActive
+        ? Promise.resolve()
+        : sleep(Math.min(chain.builderApiClient.bidDeadline, cutoffMs));
 
       // Candidates are ranked by their boosted counted total payment, the p2p bid is governed
       // by the top-level factors and each builder API bid by its own entry. Ties prefer the
       // earliest received builder API bid, so the signed block can be routed back to its
       // builder directly.
       const bestBidPromise: Promise<BidCandidate | null> = (async () => {
-        const [builderApiBids, p2pBid] = await Promise.all([builderApiBidsPromise, p2pBidPromise]);
+        // Past the race cutoff a pooled p2p bid is used right away rather than waiting for pending
+        // builder API requests, the local block is selected at the cutoff anyway if it is ready
+        let [builderApiBids] = await Promise.all([
+          Promise.race([builderApiBidsPromise, sleep(cutoffMs).then(() => null)]),
+          bidDeadlinePromise,
+        ]);
         let parentExecutionRequestsPromise: Promise<gloas.ExecutionRequests> | null = null;
 
+        let p2pBid: PooledExecutionPayloadBid | null = circuitBreakerActive
+          ? null
+          : chain.executionPayloadBidPool.getBestBid(slot, bidParentBlockHash, parentBlockRootHex);
+        // Discard p2p bids below the proposer's configured floor on the total payment.
+        // A p2p bid's total is just its value since gossip validation enforces executionPayment=0.
+        if (p2pBid !== null && BigInt(p2pBid.signedBid.message.value) < builderConfig.minBid) {
+          logger.info("Best p2p bid below configured minimum", {
+            slot,
+            bidValue: prettyGweiToEth(p2pBid.signedBid.message.value),
+            minBid: prettyGweiToEth(builderConfig.minBid),
+          });
+          p2pBid = null;
+        }
+        if (builderApiBids === null) {
+          builderApiBids = p2pBid !== null ? [] : await builderApiBidsPromise;
+        }
+
         const candidates = (
           await Promise.all(
             builderApiBids.map(async ({url, entry, signedBid, receivedMs}): Promise<BidCandidate | null> => {
@@ -1111,13 +1139,6 @@ export function getValidatorApi(
         return fn().finally(() => t?.({source}));
       };
 
-      // Calculate cutoff time based on start of the slot, ensures a slow local payload build does
-      // not delay the proposal when a builder bid block is available (and vice versa)
-      const cutoffMs = Math.max(
-        0,
-        config.getSlotComponentDurationMs(fork, BLOCK_PRODUCTION_RACE_CUTOFF_BPS) - chain.clock.msFromSlot(slot)
-      );
-
       // use abort controller to stop waiting for the bid block if the engine block will be selected
       const controller = new AbortController();
 
@@ -1700,7 +1721,7 @@ export function getValidatorApi(
         }
       }
 
-      const dependentRoot = fromHex(state.getShufflingDecisionRoot(epoch)) || (await getGenesisBlockRoot(state));
+      const dependentRoot = fromHex(state.getShufflingDecisionRoot(epoch));
 
       return {
         data: duties,
@@ -1749,7 +1770,10 @@ export function getValidatorApi(
         }
       }
 
-      const dependentRoot = fromHex(state.getShufflingDecisionRoot(epoch)) || (await getGenesisBlockRoot(state));
+      // The fork epoch PTC is computed at the fork from effective balances updated at the end of the
+      // last pre-fork epoch, so its dependent root is one epoch later than for other epochs
+      const dependentRootEpoch = epoch === config.GLOAS_FORK_EPOCH ? epoch + 1 : epoch;
+      const dependentRoot = fromHex(state.getShufflingDecisionRoot(dependentRootEpoch));
 
       return {
         data: duties,
@@ -2198,6 +2222,121 @@ export function getValidatorApi(
       }
     },
 
+    async produceInclusionList({slot}) {
+      notWhileSyncing(chain, sync.state);
+
+      if (!isForkPostHeze(chain.config.getForkName(slot))) {
+        throw new ApiError(400, `Producing inclusion list for pre-heze slot: ${slot}`);
+      }
+
+      await waitForSlot(slot); // Must never request for a future slot > currentSlot
+
+      const timer = metrics?.getInclusionListV1RequestsDuration.startTimer();
+      const transactions = await chain.executionEngine.getInclusionList();
+      timer?.();
+      metrics?.getInclusionListV1Requests.inc();
+
+      return {data: transactions, meta: {version: chain.config.getForkName(slot)}};
+    },
+
+    async getInclusionListCommitteeDuties({epoch, indices}) {
+      notWhileSyncing(chain, sync.state);
+
+      if (indices.length === 0) {
+        throw new ApiError(400, "No validator to get inclusion list committee duties");
+      }
+
+      if (!isForkPostHeze(chain.config.getForkName(computeStartSlotAtEpoch(epoch)))) {
+        throw new ApiError(400, `Requesting pre-heze inclusion list committee duties epoch: ${epoch}`);
+      }
+
+      // May request for an epoch that's in the future
+      await waitForNextClosestEpoch();
+
+      if (epoch > chain.clock.currentEpoch + 1) {
+        throw new ApiError(400, "Cannot get duties for epoch more than one ahead");
+      }
+
+      const head = chain.forkChoice.getHead();
+      const state = await chain.getHeadStateAtCurrentEpoch(RegenCaller.getDuties);
+
+      const pubkeys = getPubkeysForIndices(state, indices);
+      const requestedIndices = new Set(indices);
+      const epochStartSlot = computeStartSlotAtEpoch(epoch);
+      const shuffling = state.getShufflingAtEpoch(epoch);
+
+      // Validators set this as `dependent_root` on their inclusion lists
+      const dependentRoot = fromHex(state.getShufflingDecisionRoot(epoch));
+
+      // A validator can be on several slots' committees in an epoch; report the earliest, as for attester duties
+      const dutySlotByValidator = new Map<ValidatorIndex, Slot>();
+      for (let i = 0; i < SLOTS_PER_EPOCH; i++) {
+        const slot = epochStartSlot + i;
+        for (const validatorIndex of getInclusionListCommittee(shuffling, slot).validatorIndices) {
+          if (requestedIndices.has(validatorIndex) && !dutySlotByValidator.has(validatorIndex)) {
+            dutySlotByValidator.set(validatorIndex, slot);
+          }
+        }
+      }
+
+      const duties: routes.validator.InclusionListDuty[] = [];
+      for (let i = 0; i < indices.length; i++) {
+        const validatorIndex = indices[i];
+        const slot = dutySlotByValidator.get(validatorIndex);
+        if (slot !== undefined) {
+          duties.push({pubkey: pubkeys[i], validatorIndex, slot, dependentRoot});
+        }
+      }
+
+      return {
+        data: duties,
+        meta: {
+          dependentRoot: toRootHex(dependentRoot),
+          executionOptimistic: isOptimisticBlock(head),
+        },
+      };
+    },
+
+    async publishInclusionList({signedInclusionList}) {
+      notWhileSyncing(chain, sync.state);
+
+      const {slot} = signedInclusionList.message;
+      if (!isForkPostHeze(chain.config.getForkName(slot))) {
+        throw new ApiError(400, `Publishing pre-heze inclusion list slot: ${slot}`);
+      }
+
+      // Timeliness is judged on receipt, as for gossip, so a list submitted in time stays timely if validation is slow
+      const secFromSlot = chain.clock.secFromSlot(slot, Date.now() / 1000);
+      metrics?.inclusionListsReceived.inc({source: InclusionListSource.api});
+      metrics?.inclusionListTransactionsReceived.inc(
+        {source: InclusionListSource.api},
+        signedInclusionList.message.transactions.length
+      );
+
+      const timer = metrics?.inclusionListsValidationTime.startTimer();
+      let validationResult: InclusionListValidationResult;
+      try {
+        validationResult = await validateGossipInclusionList(chain, signedInclusionList);
+      } finally {
+        timer?.({source: InclusionListSource.api});
+      }
+      const {committeeIndices, totalBytes} = validationResult;
+      metrics?.inclusionListsValid.inc({source: InclusionListSource.api});
+      metrics?.inclusionListsValidSize.inc(totalBytes);
+
+      const isTimely =
+        secFromSlot >= 0 && secFromSlot * 1000 < chain.config.getInclusionListDueMs(chain.config.getForkName(slot));
+      chain.inclusionListStore.process(signedInclusionList, committeeIndices, isTimely);
+
+      chain.emitter.emit(routes.events.EventType.inclusionList, {
+        version: chain.config.getForkName(slot),
+        data: signedInclusionList,
+      });
+
+      await network.publishInclusionList(signedInclusionList);
+      metrics?.inclusionListsPublished.inc();
+    },
+
     async getExecutionPayloadEnvelope({slot, beaconBlockRoot}) {
       const fork = config.getForkName(slot);
 

package/src/api/options.ts

@@ -2,6 +2,8 @@ import {BeaconRestApiServerOpts, beaconRestApiServerOpts} from "./rest/index.js"
 
 export type ApiOptions = {
   maxGindicesInProof?: number;
+  /** Broadcast execution payload bids submitted via the API without running gossip validation */
+  disableBidValidation?: boolean;
   rest: BeaconRestApiServerOpts;
   commit?: string;
   version?: string;
@@ -10,6 +12,7 @@ export type ApiOptions = {
 
 export const defaultApiOptions: ApiOptions = {
   maxGindicesInProof: 512,
+  disableBidValidation: false,
   rest: beaconRestApiServerOpts,
   version: "dev",
   private: false,

package/src/chain/ColumnReconstructionTracker.ts

@@ -1,7 +1,10 @@
 import {ChainForkConfig} from "@lodestar/config";
+import {NUMBER_OF_COLUMNS} from "@lodestar/params";
+import {RootHex} from "@lodestar/types";
 import {Logger, sleep} from "@lodestar/utils";
 import {Metrics} from "../metrics/metrics.js";
 import {DataColumnReconstructionCode, recoverDataColumnSidecars} from "../util/dataColumns.js";
+import {JobItemQueue} from "../util/queue/index.js";
 import {BlockInputColumns} from "./blocks/blockInput/index.js";
 import {PayloadEnvelopeInput} from "./blocks/payloadEnvelopeInput/index.js";
 import {ChainEventEmitter} from "./emitter.js";
@@ -16,15 +19,22 @@ const RECONSTRUCTION_DELAY_MIN_BPS = 667;
  */
 const RECONSTRUCTION_DELAY_MAX_BPS = 1000;
 
+const QUEUE_MAX_LENGTH = 16;
+
+export type ColumnReconstructionInput = BlockInputColumns | PayloadEnvelopeInput;
+
 export type ColumnReconstructionTrackerInit = {
   logger: Logger;
   emitter: ChainEventEmitter;
   metrics: Metrics | null;
   config: ChainForkConfig;
+  signal: AbortSignal;
 };
 
 /**
- * Tracks column reconstruction attempts to avoid duplicate and multiple in-flight calls
+ * Runs column reconstruction one block root at a time, in trigger order. A trigger for a root that is
+ * queued or in flight is a no-op. A root is removed once its attempt finishes so a later column can
+ * retry a failed attempt.
  */
 export class ColumnReconstructionTracker {
   logger: Logger;
@@ -32,62 +42,71 @@ export class ColumnReconstructionTracker {
   metrics: Metrics | null;
   config: ChainForkConfig;
 
-  /**
-   * Track last attempted block root
-   *
-   * This is sufficient to avoid duplicate calls since we only call this
-   * function when we see a new data column sidecar from gossip.
-   */
-  lastBlockRootHex: string | null = null;
-  /** Track if a reconstruction attempt is in-flight */
-  running = false;
+  private readonly signal: AbortSignal;
+  private readonly pending = new Set<RootHex>();
+  private readonly queue: JobItemQueue<[ColumnReconstructionInput, number], void>;
 
   constructor(init: ColumnReconstructionTrackerInit) {
     this.logger = init.logger;
     this.emitter = init.emitter;
     this.metrics = init.metrics;
     this.config = init.config;
+    this.signal = init.signal;
+    this.queue = new JobItemQueue(
+      this.reconstruct,
+      {maxLength: QUEUE_MAX_LENGTH, signal: init.signal},
+      init.metrics?.columnReconstructionQueue
+    );
   }
 
-  triggerColumnReconstruction(input: BlockInputColumns | PayloadEnvelopeInput): void {
-    if (this.running) {
+  triggerColumnReconstruction(input: ColumnReconstructionInput): void {
+    if (this.pending.has(input.blockRootHex)) {
       return;
     }
 
-    if (this.lastBlockRootHex === input.blockRootHex) {
+    if (input.getAllColumns().length < NUMBER_OF_COLUMNS / 2) {
       return;
     }
 
-    // We don't care about the outcome of this call,
-    // just that it has been triggered for this block root.
-    this.running = true;
-    this.lastBlockRootHex = input.blockRootHex;
+    this.pending.add(input.blockRootHex);
+    void this.enqueue(input);
+  }
+
+  private async enqueue(input: ColumnReconstructionInput): Promise<void> {
+    try {
+      await this.queue.push(input, Date.now());
+    } catch (e) {
+      this.logger.debug(
+        "Data column sidecar reconstruction dropped",
+        {slot: input.slot, root: input.blockRootHex},
+        e as Error
+      );
+    } finally {
+      this.pending.delete(input.blockRootHex);
+    }
+  }
+
+  private reconstruct = async (input: ColumnReconstructionInput, queuedAtMs: number): Promise<void> => {
     const fork = this.config.getForkName(input.slot);
     const minDelayMs = this.config.getSlotComponentDurationMs(fork, RECONSTRUCTION_DELAY_MIN_BPS);
     const maxDelayMs = this.config.getSlotComponentDurationMs(fork, RECONSTRUCTION_DELAY_MAX_BPS);
     const delay = minDelayMs + Math.random() * (maxDelayMs - minDelayMs);
-    sleep(delay)
-      .then(() => {
-        const logCtx = {slot: input.slot, root: input.blockRootHex};
-        this.logger.debug("Attempting data column sidecar reconstruction", logCtx);
-        recoverDataColumnSidecars(input, this.emitter, this.metrics)
-          .then((result) => {
-            this.metrics?.recoverDataColumnSidecars.reconstructionResult.inc({result});
-            this.logger.debug("Data column sidecar reconstruction complete", {...logCtx, result});
-          })
-          .catch((e) => {
-            this.metrics?.recoverDataColumnSidecars.reconstructionResult.inc({
-              result: DataColumnReconstructionCode.Failed,
-            });
-            this.logger.debug("Error during data column sidecar reconstruction", logCtx, e as Error);
-          })
-          .finally(() => {
-            this.logger.debug("Data column sidecar reconstruction attempt finished", logCtx);
-            this.running = false;
-          });
-      })
-      .catch((err) => {
-        this.logger.debug("ColumnReconstructionTracker unreachable error", {}, err);
-      });
-  }
+
+    // The delay gives gossip a chance to deliver the remaining columns, so time already spent in the
+    // queue behind another root counts towards it
+    await sleep(delay - (Date.now() - queuedAtMs), this.signal);
+
+    const logCtx = {slot: input.slot, root: input.blockRootHex};
+    this.logger.debug("Attempting data column sidecar reconstruction", logCtx);
+    try {
+      const result = await recoverDataColumnSidecars(input, this.emitter, this.metrics);
+      this.metrics?.recoverDataColumnSidecars.reconstructionResult.inc({result});
+      this.logger.debug("Data column sidecar reconstruction complete", {...logCtx, result});
+    } catch (e) {
+      this.metrics?.recoverDataColumnSidecars.reconstructionResult.inc({result: DataColumnReconstructionCode.Failed});
+      this.logger.debug("Error during data column sidecar reconstruction", logCtx, e as Error);
+    } finally {
+      this.logger.debug("Data column sidecar reconstruction attempt finished", logCtx);
+    }
+  };
 }

package/src/chain/GetBlobsTracker.ts

@@ -1,10 +1,9 @@
 import {ChainForkConfig} from "@lodestar/config";
 import {computeEpochAtSlot} from "@lodestar/state-transition";
-import {Logger} from "@lodestar/utils";
+import {Logger, pruneSetToMax} from "@lodestar/utils";
 import {BLOB_AND_PROOF_V2_RPC_BYTES} from "../execution/engine/types.js";
 import {IExecutionEngine} from "../execution/index.js";
 import {Metrics} from "../metrics/metrics.js";
-import {callInNextEventLoop} from "../util/eventLoop.js";
 import {
   DataColumnEngineResult,
   getBlobSidecarsFromExecution,
@@ -14,6 +13,11 @@ import {IBlockInput, isBlockInputBlobs} from "./blocks/blockInput/index.js";
 import {PayloadEnvelopeInput} from "./blocks/payloadEnvelopeInput/index.js";
 import {ChainEventEmitter} from "./emitter.js";
 
+/** A `null` answer only changes if the EL receives the transactions later, so retries are few and spaced out */
+export const MAX_GET_BLOBS_ATTEMPTS = 3;
+export const GET_BLOBS_RETRY_INTERVAL_MS = 1000;
+const MAX_TRACKED_BLOCK_ROOTS = 64;
+
 export type GetBlobsTrackerInit = {
   logger: Logger;
   executionEngine: IExecutionEngine;
@@ -32,6 +36,8 @@ export class GetBlobsTracker {
   metrics: Metrics | null;
   config: ChainForkConfig;
   activeReconstructions = new Set<string>();
+  failedAttempts = new Map<string, {count: number; lastAttemptMs: number; retryTimer?: NodeJS.Timeout}>();
+  pendingTriggers = new Set<string>();
   // Preallocate buffers for getBlobsV2 RPC calls
   // See https://github.com/ChainSafe/lodestar/pull/8282 for context
   blobsAndProofsBuffers: {buffers: Uint8Array[]; inUse: boolean}[] = [];
@@ -46,23 +52,55 @@ export class GetBlobsTracker {
 
   triggerGetBlobs(input: IBlockInput | PayloadEnvelopeInput): void {
     if (this.activeReconstructions.has(input.blockRootHex)) {
+      this.pendingTriggers.add(input.blockRootHex);
       return;
     }
 
+    const failed = this.failedAttempts.get(input.blockRootHex);
+    if (failed) {
+      if (failed.count >= MAX_GET_BLOBS_ATTEMPTS) {
+        return;
+      }
+      const waitMs = GET_BLOBS_RETRY_INTERVAL_MS - (Date.now() - failed.lastAttemptMs);
+      if (waitMs > 0) {
+        // Columns tend to arrive in one burst, so keep a single trailing retry instead of relying on a later trigger
+        failed.retryTimer ??= setTimeout(() => {
+          failed.retryTimer = undefined;
+          if (this.failedAttempts.get(input.blockRootHex) === failed && !input.hasAllData()) {
+            this.triggerGetBlobs(input);
+          }
+        }, waitMs).unref();
+        return;
+      }
+    }
+
+    // The request is sent right away, before block processing issues newPayload: ELs that drop a
+    // transaction's blobs once the payload is validated can only answer a request that gets there first
     if (!(input instanceof PayloadEnvelopeInput) && isBlockInputBlobs(input)) {
       // there is not preallocation for blob sidecars like there is for columns sidecars so no need to
       // store the index for the preallocated buffers
       this.activeReconstructions.add(input.blockRootHex);
-      callInNextEventLoop(() => {
-        const logCtx = {slot: input.slot, root: input.blockRootHex};
-        this.logger.verbose("Trigger getBlobsV1 for block", logCtx);
-        getBlobSidecarsFromExecution(this.config, this.executionEngine, this.metrics, this.emitter, input).finally(
-          () => {
-            this.logger.verbose("Completed getBlobsV1 for block", logCtx);
-            this.activeReconstructions.delete(input.blockRootHex);
+      const logCtx = {slot: input.slot, root: input.blockRootHex};
+      this.logger.verbose("Trigger getBlobsV1 for block", logCtx);
+      let requestFailed = false;
+      getBlobSidecarsFromExecution(this.config, this.executionEngine, this.metrics, this.emitter, input)
+        .then(() => {
+          // EL errors and misses are handled inside the helper, blobs still missing means the attempt failed
+          if (!input.hasAllData()) {
+            requestFailed = true;
+            this.recordFailedAttempt(input.blockRootHex);
           }
-        );
-      });
+        })
+        .catch((error) => {
+          this.logger.debug("Error during getBlobsV1 for block", logCtx, error as Error);
+          requestFailed = true;
+          this.recordFailedAttempt(input.blockRootHex);
+        })
+        .finally(() => {
+          this.logger.verbose("Completed getBlobsV1 for block", logCtx);
+          this.activeReconstructions.delete(input.blockRootHex);
+          this.retryPendingTrigger(input, requestFailed);
+        });
 
       return;
     }
@@ -83,34 +121,55 @@ export class GetBlobsTracker {
     }
     timer?.();
 
-    // We don't care about the outcome of this call,
-    // just that it has been triggered for this block root.
     this.activeReconstructions.add(input.blockRootHex);
     this.blobsAndProofsBuffers[freeIndex].inUse = true;
-    callInNextEventLoop(() => {
-      const logCtx = {slot: input.slot, root: input.blockRootHex};
-      this.logger.verbose("Trigger getBlobsV2 for block", logCtx);
-      getDataColumnSidecarsFromExecution(
-        this.config,
-        this.executionEngine,
-        this.emitter,
-        input,
-        this.metrics,
-        this.blobsAndProofsBuffers[freeIndex].buffers
-      )
-        .then((result) => {
-          this.logger.debug("getBlobsV2 result for block", {...logCtx, result});
-          this.metrics?.dataColumns.dataColumnEngineResult.inc({result});
-        })
-        .catch((error) => {
-          this.logger.debug("Error during getBlobsV2 for block", logCtx, error as Error);
-          this.metrics?.dataColumns.dataColumnEngineResult.inc({result: DataColumnEngineResult.Failed});
-        })
-        .finally(() => {
-          this.logger.verbose("Completed getBlobsV2 for block", logCtx);
-          this.activeReconstructions.delete(input.blockRootHex);
-          this.blobsAndProofsBuffers[freeIndex].inUse = false;
-        });
-    });
+    const logCtx = {slot: input.slot, root: input.blockRootHex};
+    this.logger.verbose("Trigger getBlobsV2 for block", logCtx);
+    let requestFailed = false;
+    getDataColumnSidecarsFromExecution(
+      this.config,
+      this.executionEngine,
+      this.emitter,
+      input,
+      this.metrics,
+      this.blobsAndProofsBuffers[freeIndex].buffers
+    )
+      .then((result) => {
+        this.logger.debug("getBlobsV2 result for block", {...logCtx, result});
+        this.metrics?.dataColumns.dataColumnEngineResult.inc({result});
+        if (result === DataColumnEngineResult.NullResponse) {
+          requestFailed = true;
+          this.recordFailedAttempt(input.blockRootHex);
+        }
+      })
+      .catch((error) => {
+        this.logger.debug("Error during getBlobsV2 for block", logCtx, error as Error);
+        this.metrics?.dataColumns.dataColumnEngineResult.inc({result: DataColumnEngineResult.Failed});
+        requestFailed = true;
+        this.recordFailedAttempt(input.blockRootHex);
+      })
+      .finally(() => {
+        this.logger.verbose("Completed getBlobsV2 for block", logCtx);
+        this.activeReconstructions.delete(input.blockRootHex);
+        this.blobsAndProofsBuffers[freeIndex].inUse = false;
+        this.retryPendingTrigger(input, requestFailed);
+      });
+  }
+
+  private retryPendingTrigger(input: IBlockInput | PayloadEnvelopeInput, requestFailed: boolean): void {
+    // Triggers dropped while the request was in flight re-enter once it failed, the attempt that was just
+    // recorded makes this arm the trailing retry instead of sending another request right away.
+    // Post-gloas only the bid triggers a request, so every failed attempt re-enters to arm its own retry
+    const hadPendingTrigger = this.pendingTriggers.delete(input.blockRootHex);
+    if (requestFailed && (hadPendingTrigger || input instanceof PayloadEnvelopeInput)) {
+      this.triggerGetBlobs(input);
+    }
+  }
+
+  private recordFailedAttempt(blockRootHex: string): void {
+    const failed = this.failedAttempts.get(blockRootHex);
+    clearTimeout(failed?.retryTimer);
+    this.failedAttempts.set(blockRootHex, {count: (failed?.count ?? 0) + 1, lastAttemptMs: Date.now()});
+    pruneSetToMax(this.failedAttempts, MAX_TRACKED_BLOCK_ROOTS);
   }
 }

package/src/chain/archiveStore/archiveStore.ts

@@ -197,6 +197,7 @@ export class ArchiveStore {
     //
     // `jobQueue.push()` can throw synchronously (queue aborted / full) or reject asynchronously;
     // awaiting it inside the deferred callback funnels both failure modes into the one catch.
+    // biome-ignore lint/nursery/noMisusedPromises: the callback catches all errors itself
     callInNextEventLoop(async () => {
       try {
         await this.jobQueue.push(finalized);