@lodestar/beacon-node 1.50.0-dev.70314c9caa → 1.50.0-dev.79bf6381f8

package/src/sync/range/batch.ts

@@ -773,6 +773,7 @@ export class Batch {
     // remove any downloaded blocks and re-attempt
     // TODO(fulu): need to remove the bad blocks from the SeenBlockInputCache
     this.state = {status: BatchStatus.AwaitingDownload, blocks: [], payloadEnvelopes: null};
+    this.resetRequests();
   }
 
   private onProcessingError(attempt: FailedAttempt): void {
@@ -784,6 +785,16 @@ export class Batch {
     // remove any downloaded blocks and re-attempt
     // TODO(fulu): need to remove the bad blocks from the SeenBlockInputCache
     this.state = {status: BatchStatus.AwaitingDownload, blocks: [], payloadEnvelopes: null};
+    this.resetRequests();
+  }
+
+  /**
+   * After a partial download `requests` only cover the data still missing from the downloaded blocks,
+   * e.g. a single column. Once those blocks are dropped, request the whole batch again, otherwise
+   * every peer fails the retry for lack of blocks to validate the data against.
+   */
+  private resetRequests(): void {
+    this.requests = this.getRequests([]);
   }
 
   /** Helper to construct typed BatchError. Stack traces are correct as the error is thrown above */

package/src/sync/range/chain.ts

@@ -225,6 +225,9 @@ export class SyncChain {
       Math.floor((localFinalizedEpoch - this.lastEpochWithProcessBlocks) / EPOCHS_PER_BATCH) * EPOCHS_PER_BATCH;
     this.advanceChain(lastEpochWithProcessBlocksAligned);
 
+    // stopSyncing() cleared the retry timer, but peers can still be in rate limit backoff
+    this.scheduleRateLimitBackoffRetry();
+
     // Potentially download new batches and process pending
     this.triggerBatchDownloader();
     this.triggerBatchProcessor();
@@ -417,8 +420,10 @@ export class SyncChain {
     this.rateLimitBackoffTimeout = setTimeout(
       () => {
         this.rateLimitBackoffTimeout = undefined;
-        this.triggerBatchDownloader();
+        // Must run before the downloader, it drops expired peers without triggering a download. If it ran
+        // after, a backoff expiring in between would leave the peer idle with no retry scheduled.
         this.scheduleRateLimitBackoffRetry();
+        this.triggerBatchDownloader();
       },
       Math.max(0, retryAt - now)
     );

package/src/sync/unknownBlock.ts

@@ -73,7 +73,11 @@ type AdvancePendingBlockResult =
   | "removed";
 
 class UnknownBlockRateLimitedError extends Error {
-  constructor(message: string) {
+  constructor(
+    message: string,
+    /** Earliest backoff expiry seen when the download was deferred, it is retried once that has passed */
+    readonly retryAt: number
+  ) {
     super(message);
     this.name = "UnknownBlockRateLimitedError";
   }
@@ -849,17 +853,20 @@ export class BlockInputSync {
     }
   };
 
-  private scheduleRateLimitBackoffRetry(): void {
+  /**
+   * Returns false if there is no backoff left to wait for, in which case no retry is scheduled.
+   */
+  private scheduleRateLimitBackoffRetry(): boolean {
     this.clearRateLimitBackoffTimer();
 
     if (!this.subscribedToNetworkEvents || (this.pendingBlocks.size === 0 && this.pendingPayloads.size === 0)) {
-      return;
+      return false;
     }
 
     const now = Date.now();
     const retryAt = this.peerBalancer.getNextRateLimitRetryAt();
     if (retryAt === null) {
-      return;
+      return false;
     }
 
     this.rateLimitBackoffTimeout = setTimeout(
@@ -870,6 +877,7 @@ export class BlockInputSync {
       },
       Math.max(0, retryAt - now)
     );
+    return true;
   }
 
   private clearRateLimitBackoffTimer(): void {
@@ -1081,7 +1089,11 @@ export class BlockInputSync {
           pendingBlock.status = PendingBlockInputStatus.pending;
         }
         this.logger.debug("Deferring unknown block download due to peer rate limit", logCtx, res.err);
-        this.scheduleRateLimitBackoffRetry();
+        if (!this.scheduleRateLimitBackoffRetry() || res.err.retryAt <= Date.now()) {
+          // The backoff this download waited for expired in the meantime, a timer is only armed if another
+          // peer is still in backoff and would retry it later than needed
+          this.triggerUnknownBlockSearch();
+        }
         return;
       }
 
@@ -1364,6 +1376,14 @@ export class BlockInputSync {
     if (!isPendingPayloadEnvelope(payload)) {
       payload.status = PendingPayloadInputStatus.pending;
     }
+    if (
+      res.err instanceof UnknownBlockRateLimitedError &&
+      (!this.scheduleRateLimitBackoffRetry() || res.err.retryAt <= Date.now())
+    ) {
+      // The backoff this download waited for expired in the meantime, a timer is only armed if another
+      // peer is still in backoff and would retry it later than needed
+      this.triggerUnknownBlockSearch();
+    }
   }
 
   private async processPayload(pendingPayload: PendingPayloadInput): Promise<void> {
@@ -1488,16 +1508,21 @@ export class BlockInputSync {
       const pendingColumns = payloadInput?.hasAllData()
         ? new Set<number>()
         : new Set(payloadInput?.getMissingSampledColumnMeta().missing ?? []);
+      // Same `now` for both calls, a backoff expiring in between would be reported as no peer available
+      const now = Date.now();
       // prefer peers that gossiped this payload root to us (#9923)
       const peerMeta = this.peerBalancer.bestPeerForPendingColumns(
         pendingColumns,
         excludedPeers,
-        cacheItem.peerIdStrings
+        cacheItem.peerIdStrings,
+        now
       );
       if (peerMeta === null) {
-        if (this.peerBalancer.getNextRateLimitRetryAt(pendingColumns, excludedPeers) !== null) {
+        const retryAt = this.peerBalancer.getNextRateLimitRetryAt(pendingColumns, excludedPeers, now);
+        if (retryAt !== null) {
           throw new UnknownBlockRateLimitedError(
-            `Error fetching payload by root slot=${slot} root=${rootHex} after ${i}: peers with needed columns are rate-limited`
+            `Error fetching payload by root slot=${slot} root=${rootHex} after ${i}: peers with needed columns are rate-limited`,
+            retryAt
           );
         }
 
@@ -1636,9 +1661,11 @@ export class BlockInputSync {
     );
     this.metrics?.blockInputSync.payloadFetchPeers.set({result: FetchResult.FailureMaxAttempts}, i - 1);
 
-    if (deferredByRateLimit && this.peerBalancer.getNextRateLimitRetryAt() !== null) {
+    const retryAt = deferredByRateLimit ? this.peerBalancer.getNextRateLimitRetryAt() : null;
+    if (retryAt !== null) {
       throw new UnknownBlockRateLimitedError(
-        `Error fetching payload with slot=${slot} root=${rootHex} after ${i - 1} attempts: peers are rate-limited`
+        `Error fetching payload with slot=${slot} root=${rootHex} after ${i - 1} attempts: peers are rate-limited`,
+        retryAt
       );
     }
     throw Error(`Error fetching payload with slot=${slot} root=${rootHex} after ${i - 1} attempts.`);
@@ -1730,16 +1757,21 @@ export class BlockInputSync {
         isPendingBlockInput(cacheItem) && isBlockInputColumns(cacheItem.blockInput)
           ? new Set(cacheItem.blockInput.getMissingSampledColumnMeta().missing)
           : defaultPendingColumns;
+      // Same `now` for both calls, a backoff expiring in between would be reported as no peer available
+      const now = Date.now();
       // prefer peers that gossiped this block root to us (#9923)
       const peerMeta = this.peerBalancer.bestPeerForPendingColumns(
         pendingColumns,
         excludedPeers,
-        cacheItem.peerIdStrings
+        cacheItem.peerIdStrings,
+        now
       );
       if (peerMeta === null) {
-        if (this.peerBalancer.getNextRateLimitRetryAt(pendingColumns, excludedPeers) !== null) {
+        const retryAt = this.peerBalancer.getNextRateLimitRetryAt(pendingColumns, excludedPeers, now);
+        if (retryAt !== null) {
           throw new UnknownBlockRateLimitedError(
-            `Error fetching UnknownBlockRoot slot=${slot} root=${rootHex} after ${i}: peers with needed columns are rate-limited`
+            `Error fetching UnknownBlockRoot slot=${slot} root=${rootHex} after ${i}: peers with needed columns are rate-limited`,
+            retryAt
           );
         }
 
@@ -1845,8 +1877,9 @@ export class BlockInputSync {
 
     const message = `Error fetching BlockInput with slot=${slot} root=${rootHex} after ${i - 1} attempts.`;
 
-    if (deferredByRateLimit && this.peerBalancer.getNextRateLimitRetryAt() !== null) {
-      throw new UnknownBlockRateLimitedError(`${message} Peers are rate-limited.`);
+    const retryAt = deferredByRateLimit ? this.peerBalancer.getNextRateLimitRetryAt() : null;
+    if (retryAt !== null) {
+      throw new UnknownBlockRateLimitedError(`${message} Peers are rate-limited.`, retryAt);
     }
 
     if (!isPendingBlockInput(cacheItem)) {
@@ -2037,8 +2070,11 @@ export class UnknownBlockPeerBalancer {
     this.rateLimitedUntilByPeer.set(peerId, rateLimitedUntilMs);
   }
 
-  getNextRateLimitRetryAt(pendingColumns?: Set<number>, excludedPeers?: Set<PeerIdStr>): number | null {
-    const now = Date.now();
+  getNextRateLimitRetryAt(
+    pendingColumns?: Set<number>,
+    excludedPeers?: Set<PeerIdStr>,
+    now = Date.now()
+  ): number | null {
     let retryAt: number | null = null;
 
     for (const [peerId, rateLimitedUntil] of this.rateLimitedUntilByPeer.entries()) {
@@ -2077,9 +2113,10 @@ export class UnknownBlockPeerBalancer {
   bestPeerForPendingColumns(
     pendingColumns: Set<number>,
     excludedPeers: Set<PeerIdStr>,
-    preferredPeers?: Set<PeerIdStr>
+    preferredPeers?: Set<PeerIdStr>,
+    now = Date.now()
   ): PeerSyncMeta | null {
-    const eligiblePeers = this.filterPeers(pendingColumns, excludedPeers);
+    const eligiblePeers = this.filterPeers(pendingColumns, excludedPeers, now);
     if (eligiblePeers.length === 0) {
       return null;
     }
@@ -2120,8 +2157,7 @@ export class UnknownBlockPeerBalancer {
     return totalActiveRequests;
   }
 
-  private filterPeers(pendingDataColumns: Set<number>, excludedPeers: Set<PeerIdStr>): PeerIdStr[] {
-    const now = Date.now();
+  private filterPeers(pendingDataColumns: Set<number>, excludedPeers: Set<PeerIdStr>, now: number): PeerIdStr[] {
     let maxColumnCount = 0;
     const considerPeers: {peerId: PeerIdStr; columnCount: number}[] = [];
     for (const [peerId, syncMeta] of this.peersMeta.entries()) {

package/src/util/execution.ts

@@ -266,14 +266,15 @@ type SlotEnvelopeBytes = {slot: Slot; envelopeBytes: Uint8Array};
 
 type RangeEntry = ArchivedEnvelope & {slot: Slot};
 
-/** What a body root mismatch means on a given serving path */
-export type ReconstructMismatchPolicy = "throw" | "omit";
+/** What an envelope that cannot be rebuilt (EL-unavailable body or body root mismatch) means on a given serving path */
+export type ReconstructMissPolicy = "throw" | "omit";
 
-export type RebuildMiss =
-  /** EL does not have the block, or has pruned its block access list */
-  | {slot: Slot; reason: "unavailable"}
-  /** An EL body does not hash to its stored root (local inconsistency) */
-  | {slot: Slot; reason: "mismatch"; error: EnvelopeReconstructionError};
+/**
+ * An archived envelope that could not be rebuilt. `"unavailable"`: the EL does not have the block or its
+ * block access list (BODY_UNAVAILABLE). `"mismatch"`: an EL body does not hash to its stored root, a local
+ * inconsistency (BODY_ROOT_MISMATCH). `error` carries the matching code for paths that surface it.
+ */
+export type RebuildMiss = {slot: Slot; reason: "unavailable" | "mismatch"; error: EnvelopeReconstructionError};
 
 /**
  * Stream finalized envelopes over [startSlot, endSlot) as serialized bytes, rebuilding header
@@ -405,7 +406,14 @@ async function reconstructEnvelopesBatch(
     // A zero-length block access list cannot be valid, RLP encodes an empty list as 0xc0
     if (body == null || body.withdrawals == null || body.blockAccessList == null || body.blockAccessList.length === 0) {
       metrics?.payloadEnvelopeReconstruction.envelopes.inc({result: "unavailable"});
-      return {slot, reason: "unavailable"};
+      return {
+        slot,
+        reason: "unavailable",
+        error: new EnvelopeReconstructionError(
+          {code: EnvelopeReconstructionErrorCode.BODY_UNAVAILABLE, slot},
+          `execution client cannot serve the payload body or block access list for archived envelope slot=${slot}`
+        ),
+      };
     }
     try {
       const envelope = signedHeaderEnvelopeToFull(headerEnvelope, {