@lodestar/beacon-node 1.49.0-dev.4d35c8b82e → 1.49.0-dev.57ffc39074

package/src/sync/unknownBlock.ts

@@ -5,7 +5,7 @@ import {ForkSeq, SLOTS_PER_EPOCH, isForkPostGloas} from "@lodestar/params";
 import {RequestError, RequestErrorCode} from "@lodestar/reqresp";
 import {computeTimeAtSlot} from "@lodestar/state-transition";
 import {RootHex, Slot, gloas} from "@lodestar/types";
-import {Logger, fromHex, prettyPrintIndices, pruneSetToMax, sleep, toRootHex} from "@lodestar/utils";
+import {Logger, fromHex, isErrorAborted, prettyPrintIndices, pruneSetToMax, sleep, toRootHex} from "@lodestar/utils";
 import {isBlockInputBlobs, isBlockInputColumns} from "../chain/blocks/blockInput/blockInput.js";
 import {BlockInputSource, IBlockInput} from "../chain/blocks/blockInput/types.js";
 import {PayloadError, PayloadErrorCode} from "../chain/blocks/importExecutionPayload.js";
@@ -18,6 +18,7 @@ import {Metrics} from "../metrics/index.js";
 import {INetwork, NetworkEvent, NetworkEventData, prettyPrintPeerIdStr} from "../network/index.js";
 import {PeerSyncMeta} from "../network/peers/peersData.js";
 import {MAX_PEERS_PER_ROOT} from "../network/processor/constants.js";
+import {ClockEvent} from "../util/clock.js";
 import {PeerIdStr} from "../util/peerId.js";
 import {shuffle} from "../util/shuffle.js";
 import {sortBy} from "../util/sortBy.js";
@@ -26,7 +27,10 @@ import {MAX_CONCURRENT_REQUESTS} from "./constants.js";
 import {SyncOptions} from "./options.js";
 import {
   BlockInputSyncCacheItem,
+  DeferredPayloadResult,
+  DownloadResult,
   DroppedItemReason,
+  FetchResult,
   PayloadSyncCacheItem,
   PendingBlockInput,
   PendingBlockInputStatus,
@@ -55,6 +59,11 @@ const MAX_PENDING_BLOCKS = 100;
  */
 const PRUNE_UNRESOLVED_SLOT_EPOCHS = 2;
 
+/**
+ * Poll pending payloads of the current slot this often, between PAYLOAD_DUE and the end of the slot.
+ */
+export const PAYLOAD_POLL_INTERVAL_MS = 1_000;
+
 type AdvancePendingBlockResult =
   | "ready"
   | "queued_block"
@@ -63,14 +72,6 @@ type AdvancePendingBlockResult =
   | "blocked"
   | "removed";
 
-enum FetchResult {
-  SuccessResolved = "success_resolved",
-  SuccessMissingParent = "success_missing_parent",
-  SuccessLate = "success_late",
-  FailureTriedAllPeers = "failure_tried_all_peers",
-  FailureMaxAttempts = "failure_max_attempts",
-}
-
 class UnknownBlockRateLimitedError extends Error {
   constructor(message: string) {
     super(message);
@@ -122,6 +123,11 @@ export class BlockInputSync {
   private subscribedToNetworkEvents = false;
   private peerBalancer: UnknownBlockPeerBalancer;
   private rateLimitBackoffTimeout: NodeJS.Timeout | undefined;
+  /**
+   * Search unknown payload starting at PAYLOAD_DUE_BPS.
+   */
+  private readonly deferredPayloadRoots = new Map<RootHex, {slot: Slot; source: BlockInputSource; peer?: PeerIdStr}>();
+  private payloadPollAbortController: AbortController | undefined;
 
   constructor(
     private readonly config: ChainForkConfig,
@@ -158,6 +164,7 @@ export class BlockInputSync {
       this.logger.verbose("BlockInputSync enabled.");
       this.chain.emitter.on(ChainEvent.unknownBlockRoot, this.onUnknownBlockRoot);
       this.chain.emitter.on(ChainEvent.unknownEnvelopeBlockRoot, this.onUnknownEnvelopeBlockRoot);
+      this.chain.emitter.on(ChainEvent.unknownEnvelopeBlockRootSlot, this.onUnknownEnvelopeBlockRootSlot);
       this.chain.emitter.on(ChainEvent.incompleteBlockInput, this.onIncompleteBlockInput);
       this.chain.emitter.on(ChainEvent.incompletePayloadEnvelope, this.onIncompletePayloadEnvelope);
       this.chain.emitter.on(ChainEvent.blockUnknownParent, this.onUnknownParent);
@@ -166,6 +173,7 @@ export class BlockInputSync {
       this.chain.emitter.on(routes.events.EventType.executionPayload, this.onPayloadImported);
       this.network.events.on(NetworkEvent.peerConnected, this.onPeerConnected);
       this.network.events.on(NetworkEvent.peerDisconnected, this.onPeerDisconnected);
+      this.chain.clock.on(ClockEvent.slot, this.onClockSlot);
 
       // Seed the balancer with peers that connected before we subscribed
       for (const peerId of this.network.getConnectedPeers()) {
@@ -177,14 +185,22 @@ export class BlockInputSync {
       }
 
       this.subscribedToNetworkEvents = true;
+      // we subscribe mid-slot, and a root can be deferred at any point in the slot, so start this
+      // slot's loop now instead of waiting for the next ClockEvent.slot
+      this.onClockSlot(this.chain.clock.currentSlot);
     }
   }
 
   unsubscribeFromNetwork(): void {
     this.logger.verbose("BlockInputSync disabled.");
     this.clearRateLimitBackoffTimer();
+    this.chain.clock.off(ClockEvent.slot, this.onClockSlot);
+    this.payloadPollAbortController?.abort();
+    this.payloadPollAbortController = undefined;
+    this.deferredPayloadRoots.clear();
     this.chain.emitter.off(ChainEvent.unknownBlockRoot, this.onUnknownBlockRoot);
     this.chain.emitter.off(ChainEvent.unknownEnvelopeBlockRoot, this.onUnknownEnvelopeBlockRoot);
+    this.chain.emitter.off(ChainEvent.unknownEnvelopeBlockRootSlot, this.onUnknownEnvelopeBlockRootSlot);
     this.chain.emitter.off(ChainEvent.incompleteBlockInput, this.onIncompleteBlockInput);
     this.chain.emitter.off(ChainEvent.incompletePayloadEnvelope, this.onIncompletePayloadEnvelope);
     this.chain.emitter.off(ChainEvent.blockUnknownParent, this.onUnknownParent);
@@ -248,6 +264,33 @@ export class BlockInputSync {
     }
   };
 
+  private onUnknownEnvelopeBlockRootSlot = (data: ChainEventData[ChainEvent.unknownEnvelopeBlockRootSlot]): void => {
+    try {
+      if (data.slot === this.chain.clock.currentSlot) {
+        // we receive the optimistic search, the poll loop searches it at PAYLOAD_DUE or on its next
+        // tick. If that loop already returned - every root it watched was imported - this root waits
+        // for the next slot, where onClockSlot moves it to pendingPayloads. That is not the normal
+        // scenario, and in a forking condition it only delays the search of the extra root
+        const deferred = this.deferredPayloadRoots.get(data.rootHex);
+        if (deferred === undefined) {
+          this.deferredPayloadRoots.set(data.rootHex, {slot: data.slot, source: data.source, peer: data.peer});
+        } else if (deferred.peer === undefined) {
+          deferred.peer = data.peer;
+        }
+        return;
+      }
+
+      const isNewRoot = this.addByPayloadRootHex(data.rootHex, data.peer, data.slot);
+      if (isNewRoot) {
+        this.triggerUnknownBlockSearch();
+        this.metrics?.blockInputSync.requests.inc({type: PendingBlockType.UNKNOWN_PAYLOAD_BLOCK_ROOT_SLOT});
+        this.metrics?.blockInputSync.payloadSource.inc({source: data.source});
+      }
+    } catch (e) {
+      this.logger.debug("Error handling unknownEnvelopeBlockRootSlot event", {}, e as Error);
+    }
+  };
+
   private onIncompletePayloadEnvelope = (data: ChainEventData[ChainEvent.incompletePayloadEnvelope]): void => {
     try {
       this.addByPayloadInput(data.payloadInput, data.peer);
@@ -791,6 +834,118 @@ export class BlockInputSync {
     }
   }
 
+  private onClockSlot = (slot: Slot): void => {
+    if (!isForkPostGloas(this.config.getForkName(slot))) {
+      return;
+    }
+    // end the previous slot's loop, wherever it is: its sleep is bound to this same controller
+    this.payloadPollAbortController?.abort();
+    this.payloadPollAbortController = new AbortController();
+    // entries from an earlier slot is not polled
+    for (const [rootHex, {slot: deferredSlot, source, peer}] of this.deferredPayloadRoots) {
+      if (deferredSlot < slot) {
+        if (!this.chain.forkChoice.hasPayloadHexUnsafe(rootHex)) {
+          this.addDeferredPayloadRoot(rootHex, deferredSlot, source, peer);
+        }
+        this.deferredPayloadRoots.delete(rootHex);
+      }
+    }
+    this.pollPayloadsAtSlot(slot, this.payloadPollAbortController.signal).catch((e) => {
+      if (!isErrorAborted(e)) {
+        this.logger.debug("Error polling payloads", {slot}, e as Error);
+      }
+    });
+  };
+
+  /**
+   * Poll unknown payloads of the current slot.
+   */
+  private async pollPayloadsAtSlot(slot: Slot, signal: AbortSignal): Promise<void> {
+    // fixed tick count so the loop cannot outlive its slot: for mainnet, it's 6 ticks
+    const tickCount = Math.floor(
+      (this.config.SLOT_DURATION_MS - this.config.getPayloadDueMs()) / PAYLOAD_POLL_INTERVAL_MS
+    );
+    // a root still missing on several ticks is searched on each of them, recorded once
+    const searchedRoots = new Set<RootHex>();
+    let polls = 0;
+
+    try {
+      await sleep(Math.max(0, this.config.getPayloadDueMs() - this.chain.clock.msFromSlot(slot)), signal);
+
+      for (polls = 0; polls < tickCount; polls++) {
+        let importedCount = 0;
+        for (const [rootHex, {slot: deferredSlot, source, peer}] of this.deferredPayloadRoots) {
+          if (this.chain.forkChoice.hasPayloadHexUnsafe(rootHex)) {
+            importedCount++;
+            continue;
+          }
+          // copy to the regular pendingPayloads so that we can search
+          this.addDeferredPayloadRoot(rootHex, deferredSlot, source, peer);
+          searchedRoots.add(rootHex);
+        }
+
+        // an empty deferredPayloadRoots means there is no optimistic search comes before PAYLOAD_DUE
+        // we have to poll until the payload of current slot is imported or slot end
+        if (this.deferredPayloadRoots.size > 0 && importedCount === this.deferredPayloadRoots.size) {
+          if (searchedRoots.size > 0) {
+            // the backup search was needed
+            this.metrics?.blockInputSync.deferredPayloadResult.inc({
+              result: DeferredPayloadResult.ImportedAfterSearch,
+            });
+            this.metrics?.blockInputSync.deferredPayloadPolls.observe(polls);
+            this.logger.verbose("Payload imported after unknown sync search", {
+              slot,
+              roots: Array.from(searchedRoots).join(","),
+              polls,
+            });
+          } else {
+            // this happens most of the time on devnet
+            // gossip delivered everything before PAYLOAD_DUE
+            this.metrics?.blockInputSync.deferredPayloadResult.inc({
+              result: DeferredPayloadResult.ImportedBeforeSearch,
+            });
+          }
+          return;
+        }
+
+        this.triggerUnknownBlockSearch();
+        if (polls < tickCount - 1) {
+          // no sleep after the last tick: the loop would otherwise end right at the slot boundary,
+          // where onClockSlot aborts it
+          await sleep(PAYLOAD_POLL_INTERVAL_MS, signal);
+        }
+      }
+
+      // reaching here means the early return never fired
+      if (this.deferredPayloadRoots.size > 0) {
+        this.metrics?.blockInputSync.deferredPayloadResult.inc({result: DeferredPayloadResult.Unresolved});
+        this.logger.debug("Deferred payloads still missing at end of slot", {
+          slot,
+          roots: Array.from(this.deferredPayloadRoots.keys()).join(","),
+        });
+      } else {
+        // no optimistic search and no payload import ended the loop, eg a skipped slot
+        this.metrics?.blockInputSync.deferredPayloadResult.inc({result: DeferredPayloadResult.ResolvedNoRoot});
+        this.logger.debug("No deferred payloads at slot", {
+          slot,
+        });
+      }
+    } catch (e) {
+      if (!isErrorAborted(e)) {
+        throw e;
+      }
+      this.logger.verbose("Payload poll aborted", {slot, polls, msIntoSlot: this.chain.clock.msFromSlot(slot)});
+    }
+  }
+
+  private addDeferredPayloadRoot(rootHex: RootHex, slot: Slot, source: BlockInputSource, peer?: PeerIdStr): void {
+    const isNewRoot = this.addByPayloadRootHex(rootHex, peer, slot);
+    if (isNewRoot) {
+      this.metrics?.blockInputSync.requests.inc({type: PendingBlockType.DEFERRED_PAYLOAD_BLOCK_ROOT});
+      this.metrics?.blockInputSync.payloadSource.inc({source});
+    }
+  }
+
   private async downloadBlock(block: BlockInputSyncCacheItem): Promise<void> {
     if (block.status !== PendingBlockInputStatus.pending) {
       return;
@@ -1125,15 +1280,25 @@ export class BlockInputSync {
         return;
       }
       this.pendingPayloads.set(resultRootHex, pendingPayload);
+      const result = isPendingPayloadEnvelope(pendingPayload)
+        ? DownloadResult.WaitingForBlock
+        : this.chain.forkChoice.hasPayloadHexUnsafe(resultRootHex)
+          ? DownloadResult.Late
+          : DownloadResult.Resolved;
+      this.metrics?.blockInputSync.downloadedPayloadsResult.inc({result});
 
       if (isPendingPayloadEnvelope(pendingPayload)) {
         await this.reconcilePayloadEnvelope(pendingPayload);
       } else if (pendingPayload.status === PendingPayloadInputStatus.downloaded) {
+        this.metrics?.blockInputSync.elapsedTimeTillPayloadReceived.observe(
+          Date.now() / 1000 - computeTimeAtSlot(this.config, pendingPayload.payloadInput.slot, this.chain.genesisTime)
+        );
         await this.processPayload(pendingPayload);
       }
       return;
     }
 
+    this.metrics?.blockInputSync.downloadedPayloadsResult.inc({result: DownloadResult.Failed});
     this.logger.debug("Ignoring unknown payload root after failed download", logCtx, res.err);
     if (!isPendingPayloadEnvelope(payload)) {
       payload.status = PendingPayloadInputStatus.pending;
@@ -1248,6 +1413,11 @@ export class BlockInputSync {
       : await this.chain.seenPayloadEnvelopeInputCache.getOrReload(rootHex);
     let envelope = payloadInput?.hasPayloadEnvelope() ? payloadInput.getPayloadEnvelope() : undefined;
 
+    const fetchStartSec = Date.now() / 1000;
+    if (typeof slot === "number") {
+      this.metrics?.blockInputSync.payloadFetchBegin.observe(this.chain.clock.secFromSlot(slot, fetchStartSec));
+    }
+
     let i = 0;
     let deferredByRateLimit = false;
     while (i++ < this.getMaxDownloadAttempts()) {
@@ -1271,6 +1441,11 @@ export class BlockInputSync {
           pendingColumns.size > 0
             ? `cannot find peer with needed columns=${prettyPrintIndices(Array.from(pendingColumns))}`
             : "no peer available to download pending payload";
+        this.metrics?.blockInputSync.payloadFetchTime.observe(
+          {result: FetchResult.FailureTriedAllPeers},
+          Date.now() / 1000 - fetchStartSec
+        );
+        this.metrics?.blockInputSync.payloadFetchPeers.set({result: FetchResult.FailureTriedAllPeers}, i);
         throw Error(`Error fetching payload by root slot=${slot} root=${rootHex} after ${i}: ${reason}`);
       }
 
@@ -1280,7 +1455,12 @@ export class BlockInputSync {
       try {
         if (!envelope) {
           envelope = await this.fetchExecutionPayloadEnvelope(peerId, blockRoot, rootHex);
+          const slotWasUnknown = typeof slot !== "number";
           slot = envelope.message.payload.slotNumber;
+          if (slotWasUnknown) {
+            // we were not able to observe the time into slot when starting the fetch, do it now
+            this.metrics?.blockInputSync.payloadFetchBegin.observe(this.chain.clock.secFromSlot(slot, fetchStartSec));
+          }
         }
 
         payloadInput ??= await this.chain.seenPayloadEnvelopeInputCache.getOrReload(rootHex);
@@ -1289,6 +1469,11 @@ export class BlockInputSync {
           // envelope and wait for the block body; reconcilePayloadEnvelope validates once the block lands.
           // payloadInput may be seeded from the block body during download, so a non-null payloadInput does not
           // imply the block is imported.
+          this.metrics?.blockInputSync.payloadFetchTime.observe(
+            {result: FetchResult.SuccessWaitingForBlock},
+            Date.now() / 1000 - fetchStartSec
+          );
+          this.metrics?.blockInputSync.payloadFetchPeers.set({result: FetchResult.SuccessWaitingForBlock}, i);
           return {
             status: PendingPayloadInputStatus.waitingForBlock,
             envelope,
@@ -1339,6 +1524,12 @@ export class BlockInputSync {
         });
 
         if (pendingPayload.status === PendingPayloadInputStatus.downloaded) {
+          // late = the payload got imported (via gossip) while we were downloading, same as the block path
+          const result = this.chain.forkChoice.hasPayloadHexUnsafe(rootHex)
+            ? FetchResult.SuccessLate
+            : FetchResult.SuccessResolved;
+          this.metrics?.blockInputSync.payloadFetchTime.observe({result}, Date.now() / 1000 - fetchStartSec);
+          this.metrics?.blockInputSync.payloadFetchPeers.set({result}, i);
           return pendingPayload;
         }
 
@@ -1373,12 +1564,17 @@ export class BlockInputSync {
       }
     }
 
+    this.metrics?.blockInputSync.payloadFetchTime.observe(
+      {result: FetchResult.FailureMaxAttempts},
+      Date.now() / 1000 - fetchStartSec
+    );
+    this.metrics?.blockInputSync.payloadFetchPeers.set({result: FetchResult.FailureMaxAttempts}, i - 1);
+
     if (deferredByRateLimit && this.peerBalancer.getNextRateLimitRetryAt() !== null) {
       throw new UnknownBlockRateLimitedError(
         `Error fetching payload with slot=${slot} root=${rootHex} after ${i - 1} attempts: peers are rate-limited`
       );
     }
-
     throw Error(`Error fetching payload with slot=${slot} root=${rootHex} after ${i - 1} attempts.`);
   }
 
@@ -1483,7 +1679,7 @@ export class BlockInputSync {
 
         // no more peer with needed columns to try, throw error
         const message = `Error fetching UnknownBlockRoot slot=${slot} root=${rootHex} after ${i}: cannot find peer with needed columns=${prettyPrintIndices(Array.from(pendingColumns))}`;
-        this.metrics?.blockInputSync.fetchTimeSec.observe(
+        this.metrics?.blockInputSync.fetchTime.observe(
           {result: FetchResult.FailureTriedAllPeers},
           Date.now() / 1000 - fetchStartSec
         );
@@ -1572,7 +1768,7 @@ export class BlockInputSync {
           : this.chain.forkChoice.hasBlockHex(cacheItem.blockInput.parentRootHex)
             ? FetchResult.SuccessResolved
             : FetchResult.SuccessMissingParent;
-        this.metrics?.blockInputSync.fetchTimeSec.observe({result}, Date.now() / 1000 - fetchStartSec);
+        this.metrics?.blockInputSync.fetchTime.observe({result}, Date.now() / 1000 - fetchStartSec);
         this.metrics?.blockInputSync.fetchPeers.set({result}, i);
         return cacheItem;
       }
@@ -1606,7 +1802,7 @@ export class BlockInputSync {
       }
     }
 
-    this.metrics?.blockInputSync.fetchTimeSec.observe(
+    this.metrics?.blockInputSync.fetchTime.observe(
       {result: FetchResult.FailureMaxAttempts},
       Date.now() / 1000 - fetchStartSec
     );

package/src/util/sszBytes.ts

@@ -1,6 +1,7 @@
 import {BitArray, deserializeUint8ArrayBitListFromBytes} from "@chainsafe/ssz";
 import {ChainForkConfig} from "@lodestar/config";
 import {
+  BYTES_PER_CELL,
   BYTES_PER_FIELD_ELEMENT,
   FIELD_ELEMENTS_PER_BLOB,
   ForkName,
@@ -551,6 +552,23 @@ export function getBeaconBlockRootFromDataColumnSidecarSerialized(data: Uint8Arr
   return "0x" + blockRootBuf.toString("hex");
 }
 
+// beaconBlockRoot is the last fixed field in the Gloas layout
+const DATA_COLUMN_SIDECAR_GLOAS_FIXED_SIZE = BEACON_BLOCK_ROOT_POSITION_IN_GLOAS_DATA_COLUMN_SIDECAR + ROOT_SIZE; // 56 bytes
+const KZG_PROOF_SIZE = 48; // Bytes48
+const DATA_COLUMN_SIDECAR_GLOAS_PER_BLOB_SIZE = BYTES_PER_CELL + KZG_PROOF_SIZE; // 2048 + 48 = 2096 bytes
+
+/**
+ * Gloas sidecar size bound uses the largest blob limit across the entire schedule, including future entries.
+ * https://github.com/ethereum/consensus-specs/pull/5613
+ */
+export function computeMaxGloasDataColumnSidecarSize(config: ChainForkConfig): number {
+  const maxBlobs = config.BLOB_SCHEDULE.reduce(
+    (max, entry) => Math.max(max, entry.MAX_BLOBS_PER_BLOCK),
+    config.MAX_BLOBS_PER_BLOCK_ELECTRA
+  );
+  return DATA_COLUMN_SIDECAR_GLOAS_FIXED_SIZE + maxBlobs * DATA_COLUMN_SIDECAR_GLOAS_PER_BLOB_SIZE;
+}
+
 /**
  * SignedExecutionPayloadEnvelope SSZ Layout:
  * ├─ 4 bytes: message offset (points to byte 100)