@lodestar/beacon-node 1.49.0-dev.52a50304a4 → 1.49.0-dev.60ba78ee68

package/src/chain/blocks/importExecutionPayload.ts

@@ -12,6 +12,7 @@ import {isQueueErrorAborted} from "../../util/queue/index.js";
 import {BeaconChain} from "../chain.js";
 import {RegenCaller} from "../regen/interface.js";
 import {PayloadEnvelopeInput} from "../seenCache/seenPayloadEnvelopeInput.js";
+import {PayloadEnvelopeInputSource} from "./payloadEnvelopeInput/index.js";
 import {ImportPayloadOpts} from "./types.js";
 import {
   verifyExecutionPayloadEnvelope,
@@ -172,6 +173,9 @@ export async function importExecutionPayload(
   }
 
   // 4a. Run EL and signature verification in parallel
+  const logCtx = {slot, root: blockRootHex, executionBlock: envelope.payload.blockNumber};
+  this.logger.debug("Call engine api newPayload", logCtx);
+  const verifyStartSec = Date.now() / 1000;
   const [execResult, signatureValid] = await Promise.all([
     this.executionEngine.notifyNewPayload(
       fork,
@@ -191,6 +195,8 @@ export async function importExecutionPayload(
           this.bls
         ),
   ]);
+  this.logger.debug("Receive engine api newPayload result", {...logCtx, status: execResult.status});
+  this.metrics?.engineNotifyNewPayloadResult.inc({result: execResult.status});
 
   // 4b. Check signature verification result
   if (!signatureValid) {
@@ -223,6 +229,14 @@ export async function importExecutionPayload(
       });
   }
 
+  const {source, seenTimestampSec} = payloadInput.getPayloadEnvelopeSource();
+  if (execResult.status === ExecutionPayloadStatus.VALID && source === PayloadEnvelopeInputSource.gossip) {
+    const nowSec = Date.now() / 1000;
+    this.metrics?.gossipExecutionPayloadEnvelope.executionPayload.recvToValidation.observe(nowSec - seenTimestampSec);
+    // same metric as pre-gloas, it tracks EL verification time regardless of fork
+    this.metrics?.gossipBlock.executionPayload.validationTime.observe(nowSec - verifyStartSec);
+  }
+
   // 5. Persist payload envelope to hot DB. Wait for write-queue space here to apply backpressure
   // on the import pipeline during sync, then perform the write asynchronously to avoid blocking.
   await this.unfinalizedPayloadEnvelopeWrites.waitForSpace();

package/src/chain/produceBlock/computeNewStateRoot.ts

@@ -1,4 +1,4 @@
-import {EMPTY_SIGNATURE, IBeaconStateView} from "@lodestar/state-transition";
+import {EMPTY_SIGNATURE, IBeaconStateView, StateHashTreeRootSource} from "@lodestar/state-transition";
 import {BeaconBlock, BlindedBeaconBlock, Gwei, Root} from "@lodestar/types";
 import {Metrics} from "../../metrics/index.js";
 
@@ -14,5 +14,12 @@ export function computeNewStateRoot(
 ): {newStateRoot: Root; proposerReward: Gwei; postState: IBeaconStateView} {
   // Set signature to zero to re-use stateTransition() function which requires the SignedBeaconBlock type
   const signedBlock = {message: block, signature: EMPTY_SIGNATURE};
-  return state.computeNewStateRoot({block: signedBlock}, {metrics});
+  const {newStateRoot, proposerReward, postState, hashTreeRootTime} = state.computeNewStateRoot(
+    {block: signedBlock},
+    {metrics}
+  );
+
+  metrics?.stateHashTreeRootTime.observe({source: StateHashTreeRootSource.computeNewStateRoot}, hashTreeRootTime);
+
+  return {newStateRoot, proposerReward, postState};
 }

package/src/metrics/metrics/lodestar.ts

@@ -1,5 +1,6 @@
 /** biome-ignore-all lint/suspicious/noTemplateCurlyInString: The metric templates requires to have `${}` in a normal string */
 import {NotReorgedReason} from "@lodestar/fork-choice";
+import {StateHashTreeRootSource} from "@lodestar/state-transition";
 import {ArchiveStoreTask} from "../../chain/archiveStore/archiveStore.js";
 import {FrequencyStateArchiveStep} from "../../chain/archiveStore/strategies/frequencyStateArchiveStrategy.js";
 import type {LateCanonicalBlockReason} from "../../chain/archiveStore/utils/archiveBlocks.js";
@@ -408,6 +409,15 @@ export function createLodestarMetrics(
       labelNames: ["caller"],
     }),
 
+    // this metrics stay here instead of @lodestar/state-transition in order to prepare
+    // for the native zig state-transition in the future
+    stateHashTreeRootTime: register.histogram<{source: StateHashTreeRootSource}>({
+      name: "lodestar_stfn_hash_tree_root_seconds",
+      help: "Time to compute the hash tree root of a post state in seconds",
+      buckets: [0.01, 0.05, 0.1, 0.2, 0.3, 0.5],
+      labelNames: ["source"],
+    }),
+
     // BLS verifier thread pool and queue
 
     bls: {
@@ -1001,6 +1011,13 @@ export function createLodestarMetrics(
         help: "Count of errors, by error type, while processing execution payload envelopes",
         labelNames: ["error"],
       }),
+      executionPayload: {
+        recvToValidation: register.histogram({
+          name: "lodestar_gossip_execution_payload_envelope_received_to_execution_payload_verification_seconds",
+          help: "Time elapsed between execution payload envelope received and execution payload verification",
+          buckets: [0.05, 0.1, 0.25, 0.5, 1, 2, 3, 4, 6, 8],
+        }),
+      },
     },
     gossipExecutionPayloadBid: {
       elapsedTimeTillReceived: register.histogram<{source: OpSource}>({

package/src/metrics/metrics.ts

@@ -12,7 +12,12 @@ export type Metrics = BeaconMetrics &
   BeaconStateTransitionMetrics &
   LodestarMetrics & {register: RegistryMetricCreator; close: () => void};
 
-export function createMetrics(opts: MetricsOptions, genesisTime: number, externalRegistries: Registry[] = []): Metrics {
+export function createMetrics(
+  opts: MetricsOptions,
+  genesisTime: number,
+  externalRegistries: Registry[] = [],
+  {collectNodeMetrics = true}: {collectNodeMetrics?: boolean} = {}
+): Metrics {
   const register = new RegistryMetricCreator();
   const beacon = createBeaconMetrics(register);
   const forkChoice = getForkChoiceMetrics(register);
@@ -24,7 +29,7 @@ export function createMetrics(opts: MetricsOptions, genesisTime: number, externa
   };
   process.on("unhandledRejection", onUnhandledRejection);
 
-  const nodeJsMetricsClose = collectNodeJSMetrics(register);
+  const nodeJsMetricsClose = collectNodeMetrics ? collectNodeJSMetrics(register) : (): void => {};
   const close = (): void => {
     process.removeListener("unhandledRejection", onUnhandledRejection);
     nodeJsMetricsClose();

package/src/sync/types.ts

@@ -105,8 +105,6 @@ export enum DroppedItemReason {
   invalidEnvelope = "invalid_envelope",
   /** a payload had an invalid signature (INVALID_SIGNATURE) */
   invalidSignature = "invalid_signature",
-  /** a block root was given up on after exhausting download attempts (data could not be fetched) */
-  unavailable = "unavailable",
   /** an unexpected / unhandled error code led to removal */
   unknown = "unknown",
 }

package/src/sync/unknownBlock.ts

@@ -227,8 +227,11 @@ export class BlockInputSync {
   private onUnknownBlockRoot = (data: ChainEventData[ChainEvent.unknownBlockRoot]): void => {
     try {
       const isNewRoot = this.addByRootHex(data.rootHex, data.peer);
-      if (isNewRoot) {
+      // A retained entry (an earlier download failed, #10018) is not new but still needs a retry pass.
+      if (this.pendingBlocks.get(data.rootHex)?.status === PendingBlockInputStatus.pending) {
         this.triggerUnknownBlockSearch();
+      }
+      if (isNewRoot) {
         this.metrics?.blockInputSync.requests.inc({type: PendingBlockType.UNKNOWN_BLOCK_ROOT});
         this.metrics?.blockInputSync.source.inc({source: data.source});
       }
@@ -393,12 +396,21 @@ export class BlockInputSync {
       }
     }
 
-    let deletedBlocks = 0;
+    let belowFinalizedBlocks = 0;
+    let agedOutBlocks = 0;
     for (const [rootHex, block] of this.pendingBlocks) {
       const slot = getBlockInputSyncCacheItemSlot(block);
-      if (typeof slot === "number" && slot < finalizedSlot) {
+      if (typeof slot !== "number") {
+        // we may not be able to download a root, hence there is no slot for it
+        if (nowSec - block.timeAddedSec > maxSecInCache) {
+          this.pendingBlocks.delete(rootHex);
+          agedOutBlocks++;
+        }
+        continue;
+      }
+      if (slot < finalizedSlot) {
         this.pendingBlocks.delete(rootHex);
-        deletedBlocks++;
+        belowFinalizedBlocks++;
       }
     }
 
@@ -411,16 +423,20 @@ export class BlockInputSync {
     if (agedOutPayloads > 0) {
       this.metrics?.blockInputSync.removedPayloads.inc({reason: DroppedItemReason.agedOut}, agedOutPayloads);
     }
-    if (deletedBlocks > 0) {
-      this.metrics?.blockInputSync.removedBlocks.inc({reason: DroppedItemReason.belowFinalized}, deletedBlocks);
+    if (belowFinalizedBlocks > 0) {
+      this.metrics?.blockInputSync.removedBlocks.inc({reason: DroppedItemReason.belowFinalized}, belowFinalizedBlocks);
+    }
+    if (agedOutBlocks > 0) {
+      this.metrics?.blockInputSync.removedBlocks.inc({reason: DroppedItemReason.agedOut}, agedOutBlocks);
     }
 
-    this.logger.debug("BlockInputSync.pruneFinalized dropped pre-finalized pending items", {
+    this.logger.debug("BlockInputSync.pruneFinalized dropped stale pending items", {
       finalizedSlot,
       finalizedRoot: checkpoint.rootHex,
       belowFinalizedPayloads,
       agedOutPayloads,
-      deletedBlocks,
+      belowFinalizedBlocks,
+      agedOutBlocks,
       unresolvedSlotPayloads,
     });
   };
@@ -973,7 +989,12 @@ export class BlockInputSync {
     if (!res.err) {
       this.metrics?.blockInputSync.downloadedBlocksSuccess.inc();
       const pending = res.result;
-      this.pendingBlocks.set(pending.blockInput.blockRootHex, pending);
+      // pruning may have deleted this entry while we were awaiting the network, do not resurrect it via the set below
+      if (!this.pendingBlocks.has(rootHex)) {
+        this.logger.verbose("Dropping downloaded block, entry pruned during fetch", logCtx);
+        return;
+      }
+      this.pendingBlocks.set(rootHex, pending);
       const blockSlot = pending.blockInput.slot;
       const finalizedSlot = this.chain.forkChoice.getFinalizedBlock().slot;
       const delaySec = Date.now() / 1000 - computeTimeAtSlot(this.config, blockSlot, this.chain.genesisTime);
@@ -1035,8 +1056,11 @@ export class BlockInputSync {
       }
 
       this.metrics?.blockInputSync.downloadedBlocksError.inc();
-      this.logger.debug("Ignoring unknown block root after many failed downloads", logCtx, res.err);
-      this.removeAndDownScoreAllDescendants(block, DroppedItemReason.unavailable);
+      this.logger.debug("Failed to download unknown block root, retained for retry", logCtx, res.err);
+      const pendingBlock = this.pendingBlocks.get(rootHex);
+      if (pendingBlock && pendingBlock.status === PendingBlockInputStatus.fetching) {
+        pendingBlock.status = PendingBlockInputStatus.pending;
+      }
     }
   }
 
@@ -1303,7 +1327,7 @@ export class BlockInputSync {
     }
 
     this.metrics?.blockInputSync.downloadedPayloadsResult.inc({result: DownloadResult.Failed});
-    this.logger.debug("Ignoring unknown payload root after failed download", logCtx, res.err);
+    this.logger.debug("Failed to download unknown payload root, retained for retry", logCtx, res.err);
     if (!isPendingPayloadEnvelope(payload)) {
       payload.status = PendingPayloadInputStatus.pending;
     }
@@ -1763,7 +1787,11 @@ export class BlockInputSync {
         this.peerBalancer.onRequestCompleted(peerId);
       }
 
-      this.pendingBlocks.set(getBlockInputSyncCacheItemRootHex(cacheItem), cacheItem);
+      const cacheItemRootHex = getBlockInputSyncCacheItemRootHex(cacheItem);
+      // pruning may have dropped it mid-fetch
+      if (this.pendingBlocks.has(cacheItemRootHex)) {
+        this.pendingBlocks.set(cacheItemRootHex, cacheItem);
+      }
 
       if (cacheItem.status === PendingBlockInputStatus.downloaded) {
         // download was successful, no need to go with another peer, return