@lodestar/beacon-node 1.50.0-dev.7f24e06bb6 → 1.50.0-dev.82f05c43ff

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

@@ -6,6 +6,7 @@ import {toRootHex} from "@lodestar/utils";
 import {EnvelopeReconstructionError} from "../../../chain/errors/index.js";
 import {IBeaconChain} from "../../../chain/index.js";
 import {IBeaconDb} from "../../../db/index.js";
+import {MAX_BODIES_PER_REQUEST} from "../../../util/execution.js";
 import {ExecutionPayloadEnvelopesByRootRequest} from "../../../util/types.js";
 import {prettyPrintPeerId} from "../../util.js";
 
@@ -19,10 +20,15 @@ export async function* onExecutionPayloadEnvelopesByRoot(
   // The gloas req/resp spec uses MIN_EPOCHS_FOR_BLOCK_REQUESTS to define the minimum range peers MUST serve.
   // Archival nodes may still serve older retained payloads to allow genesis sync.
 
-  // Resolve slots first so archived header envelopes are rebuilt in one EL batch, not one call per root
+  // Resolve slots first so archived envelopes can be reconstructed in EL batches.
+  // Duplicate roots are served once, since each archived envelope costs an EL fetch and a rebuild.
   const requests: {blockSlot: Slot; blockRootHex: RootHex}[] = [];
+  const seenRoots = new Set<RootHex>();
   for (const root of requestBody) {
     const rootHex = toRootHex(root);
+    if (seenRoots.has(rootHex)) continue;
+    seenRoots.add(rootHex);
+
     const block = chain.forkChoice.getBlockHexDefaultStatus(rootHex);
     // If the block is not in fork choice, it may be finalized. Attempt to find its slot in block archive
     const slot = block ? block.slot : await db.blockArchive.getSlotByRoot(root);
@@ -41,35 +47,43 @@ export async function* onExecutionPayloadEnvelopesByRoot(
     requests.push({blockSlot: slot, blockRootHex: rootHex});
   }
 
-  let envelopesBytes: (Uint8Array | null)[];
-  try {
-    // by-root allows omission, so a mismatched envelope is left out rather than failing the response
-    envelopesBytes = await chain.getSerializedExecutionPayloadEnvelopes(requests, "omit");
-  } catch (e) {
-    if (e instanceof EnvelopeReconstructionError) {
-      throw new ResponseError(
-        RespStatus.RESOURCE_UNAVAILABLE,
-        `Failed to reconstruct archived envelopes: ${e.message}`
-      );
+  // Yield between EL batches so the first response does not wait for the entire request.
+  let yielded = 0;
+  for (let i = 0; i < requests.length; i += MAX_BODIES_PER_REQUEST) {
+    const batch = requests.slice(i, i + MAX_BODIES_PER_REQUEST);
+    let envelopesBytes: (Uint8Array | null)[];
+    try {
+      // 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) {
+        // By-root permits partial responses, so an EL failure can end the response here.
+        if (yielded > 0) return;
+        throw new ResponseError(
+          RespStatus.RESOURCE_UNAVAILABLE,
+          `Failed to reconstruct archived envelopes: ${e.message}`
+        );
+      }
+      throw e;
     }
-    throw e;
-  }
 
-  for (let i = 0; i < requests.length; i++) {
-    const {blockSlot, blockRootHex} = requests[i];
-    const envelopeBytes = envelopesBytes[i];
-    if (envelopeBytes) {
-      yield {
-        data: envelopeBytes,
-        boundary: chain.config.getForkBoundaryAtEpoch(computeEpochAtSlot(blockSlot)),
-      };
-    } else {
-      chain.logger.debug("Cannot serve ExecutionPayloadEnvelopesByRoot: envelope not found", {
-        slot: blockSlot,
-        root: blockRootHex,
-        peer: prettyPrintPeerId(peerId),
-        client: peerClient,
-      });
+    for (let j = 0; j < batch.length; j++) {
+      const {blockSlot, blockRootHex} = batch[j];
+      const envelopeBytes = envelopesBytes[j];
+      if (envelopeBytes) {
+        yielded++;
+        yield {
+          data: envelopeBytes,
+          boundary: chain.config.getForkBoundaryAtEpoch(computeEpochAtSlot(blockSlot)),
+        };
+      } else {
+        chain.logger.debug("Cannot serve ExecutionPayloadEnvelopesByRoot: envelope not found", {
+          slot: blockSlot,
+          root: blockRootHex,
+          peer: prettyPrintPeerId(peerId),
+          client: peerClient,
+        });
+      }
     }
   }
 }

package/src/node/nodejs.ts

@@ -1,6 +1,12 @@
 import {setMaxListeners} from "node:events";
 import {PrivateKey} from "@libp2p/interface";
 import {Registry} from "prom-client";
+import {
+  init as initNativeMetrics,
+  registerLocalValidator as registerNativeLocalValidator,
+  scrapeMetrics as scrapeNativeMetrics,
+  unregisterLocalValidator as unregisterNativeLocalValidator,
+} from "@chainsafe/lodestar-z/metrics";
 import {type PubkeyCache} from "@chainsafe/lodestar-z/pubkeys";
 import {hasher} from "@chainsafe/persistent-merkle-tree";
 import {BeaconApiMethods} from "@lodestar/api/beacon/server";
@@ -8,6 +14,7 @@ import {BeaconConfig} from "@lodestar/config";
 import type {LoggerNode} from "@lodestar/logger/node";
 import {ZERO_HASH_HEX} from "@lodestar/params";
 import {IBeaconStateView, isStatePostBellatrix, isStatePostGloas} from "@lodestar/state-transition";
+import {Slot} from "@lodestar/types";
 import {sleep, toRootHex} from "@lodestar/utils";
 import {ProcessShutdownCallback} from "@lodestar/validator";
 import {BeaconRestApiServer, getApi} from "../api/index.js";
@@ -55,6 +62,7 @@ export type BeaconNodeInitModules = {
   peerStoreDir?: string;
   anchorState: IBeaconStateView;
   isAnchorStateFinalized: boolean;
+  earliestAvailableSlot: Slot;
   metricsRegistries?: Registry[];
 };
 
@@ -153,6 +161,7 @@ export class BeaconNode {
     peerStoreDir,
     anchorState,
     isAnchorStateFinalized,
+    earliestAvailableSlot,
     metricsRegistries = [],
   }: BeaconNodeInitModules): Promise<T> {
     if (hasher.name !== "hashtree") {
@@ -171,7 +180,13 @@ export class BeaconNode {
       // monitoring relies on metrics data
       opts.monitoring.endpoint
     ) {
-      metrics = createMetrics(opts.metrics, anchorState.genesisTime, metricsRegistries);
+      if (opts.chain.nativeStateTransition) {
+        initNativeMetrics();
+      }
+
+      metrics = createMetrics(opts.metrics, anchorState.genesisTime, metricsRegistries, {
+        includeStateTransitionMetrics: !opts.chain.nativeStateTransition,
+      });
       initBeaconMetrics(metrics, anchorState);
       // Since the db is instantiated before this, metrics must be injected manually afterwards
       db.setMetrics(metrics.db, metrics.flatFileStore);
@@ -185,7 +200,13 @@ export class BeaconNode {
             config,
             anchorState.genesisTime,
             logger.child({module: LoggerModule.vmon}),
-            opts.validatorMonitor
+            opts.validatorMonitor,
+            opts.chain.nativeStateTransition
+              ? {
+                  registerLocalValidator: registerNativeLocalValidator,
+                  unregisterLocalValidator: unregisterNativeLocalValidator,
+                }
+              : null
           )
         : null;
 
@@ -248,6 +269,7 @@ export class BeaconNode {
       validatorMonitor,
       anchorState,
       isAnchorStateFinalized,
+      earliestAvailableSlot,
       executionEngine: initializeExecutionEngine(executionEngineOpts, {
         metrics,
         signal,
@@ -303,7 +325,17 @@ export class BeaconNode {
     const metricsServer = opts.metrics.enabled
       ? await getHttpMetricsServer(opts.metrics, {
           register: (metrics as Metrics).register,
-          getOtherMetrics: async () => Promise.all([network.scrapeMetrics(), chain.archiveStore.scrapeMetrics()]),
+          getOtherMetrics: async () => {
+            const otherMetrics = await Promise.all([network.scrapeMetrics(), chain.archiveStore.scrapeMetrics()]);
+            if (opts.chain.nativeStateTransition) {
+              try {
+                otherMetrics.push(scrapeNativeMetrics());
+              } catch (e) {
+                logger.warn("Failed to scrape native state-transition metrics", {}, e as Error);
+              }
+            }
+            return otherMetrics;
+          },
           logger: logger.child({module: LoggerModule.metrics}),
         })
       : null;

package/src/sync/range/batch.ts

@@ -224,6 +224,7 @@ export class Batch {
     return {
       blockRoot: parentBlockRoot,
       blockRootHex: toRootHex(parentBlockRoot),
+      blockHash: this.latestBid.blockHash,
       kzgCommitments: this.latestBid.blobKzgCommitments,
     };
   }
@@ -773,6 +774,7 @@ 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 +786,16 @@ 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)
     );
@@ -581,6 +586,7 @@ export class SyncChain {
           case DownloadByRangeErrorCode.OUT_OF_RANGE_BLOCKS:
           case DownloadByRangeErrorCode.PARENT_ROOT_MISMATCH:
           case DownloadByRangeErrorCode.INVALID_ENVELOPE_BEACON_BLOCK_ROOT:
+          case DownloadByRangeErrorCode.INVALID_ENVELOPE_BLOCK_HASH:
           case DownloadByRangeErrorCode.INVALID_CHAIN_SEGMENT:
           case BlobSidecarErrorCode.INCLUSION_PROOF_INVALID:
           case BlobSidecarErrorCode.INVALID_KZG_PROOF_BATCH:

package/src/sync/range/range.ts

@@ -5,6 +5,7 @@ import {IBeaconStateViewGloas, computeStartSlotAtEpoch, isStatePostGloas} from "
 import {Epoch, Status, fulu} from "@lodestar/types";
 import {Logger, prettyPrintIndices, toRootHex} from "@lodestar/utils";
 import {IBlockInput} from "../../chain/blocks/blockInput/types.js";
+import {PayloadError, PayloadErrorCode} from "../../chain/blocks/importExecutionPayload.js";
 import {AttestationImportOpt, ImportBlockOpts} from "../../chain/blocks/index.js";
 import {assertLinearChainSegment} from "../../chain/blocks/utils/chainSegment.js";
 import {BlockError} from "../../chain/errors/index.js";
@@ -207,22 +208,50 @@ export class RangeSync extends (EventEmitter as {new (): RangeSyncEmitter}) {
       blsVerifyOnMainThread: false,
     };
 
-    if (this.opts?.disableProcessAsChainSegment) {
-      // Should only be used for debugging or testing
-      for (const block of blocks) {
-        await this.chain.processBlock(block, flags);
-        const payloadEnvelope = payloadEnvelopes?.get(block.slot);
-        if (payloadEnvelope?.hasPayloadEnvelope()) {
-          await this.chain.processExecutionPayload(payloadEnvelope);
+    try {
+      if (this.opts?.disableProcessAsChainSegment) {
+        // Should only be used for debugging or testing
+        for (const block of blocks) {
+          await this.chain.processBlock(block, flags);
+          const payloadEnvelope = payloadEnvelopes?.get(block.slot);
+          if (payloadEnvelope?.hasPayloadEnvelope()) {
+            await this.chain.processExecutionPayload(payloadEnvelope);
+          }
+        }
+      } else {
+        const {orphaned, skipped} = await this.chain.processChainSegment(blocks, payloadEnvelopes, flags);
+        // We log orphaned payloads to work with different clients
+        // in the future, consider applying penalties in certain conditions
+        // make sure the payload source was from range sync in that case
+        for (const {slot, payloadEnvelopeInput} of orphaned) {
+          const {peerIdStr, source} = payloadEnvelopeInput.getPayloadEnvelopeSource();
+          let client = "unknown";
+          if (peerIdStr !== undefined) {
+            try {
+              client = this.getConnectedPeerSyncMeta(peerIdStr).client;
+            } catch {
+              // peer disconnected since serving the envelope, keep "unknown"
+            }
+          }
+          this.logger.debug("Orphaned payload envelope in range sync batch", {
+            slot,
+            root: payloadEnvelopeInput.blockRootHex,
+            peer: peerIdStr ?? "unknown",
+            source,
+            client,
+            skipped,
+          });
         }
       }
-    } else {
-      const {orphaned, skipped} = await this.chain.processChainSegment(blocks, payloadEnvelopes, flags);
-      // We log orphaned payloads to work with different clients
-      // in the future, consider applying penalties in certain conditions
-      // make sure the payload source was from range sync in that case
-      for (const {slot, payloadEnvelopeInput} of orphaned) {
-        const {peerIdStr, source} = payloadEnvelopeInput.getPayloadEnvelopeSource();
+    } catch (e) {
+      if (
+        e instanceof PayloadError &&
+        (e.type.code === PayloadErrorCode.INVALID_SIGNATURE ||
+          e.type.code === PayloadErrorCode.ENVELOPE_VERIFICATION_ERROR ||
+          e.type.code === PayloadErrorCode.EXECUTION_ENGINE_INVALID)
+      ) {
+        const {payloadInput} = e;
+        const {peerIdStr, source} = payloadInput.getPayloadEnvelopeSource();
         let client = "unknown";
         if (peerIdStr !== undefined) {
           try {
@@ -231,15 +260,18 @@ export class RangeSync extends (EventEmitter as {new (): RangeSyncEmitter}) {
             // peer disconnected since serving the envelope, keep "unknown"
           }
         }
-        this.logger.debug("Orphaned payload envelope in range sync batch", {
-          slot,
-          root: payloadEnvelopeInput.blockRootHex,
+        this.logger.debug("Removing invalid payload envelope in range sync batch", {
+          slot: payloadInput.slot,
+          root: payloadInput.blockRootHex,
           peer: peerIdStr ?? "unknown",
           source,
           client,
-          skipped,
+          code: e.type.code,
         });
+        // The batch is downloaded again, it would otherwise reuse the cached envelope and fail the same way
+        this.chain.seenPayloadEnvelopeInputCache.removeInvalid(payloadInput);
       }
+      throw e;
     }
   };