@lodestar/beacon-node 1.50.0-dev.9c3f7bebe9 → 1.50.0-dev.9c7f3c203f

package/src/network/processor/gossipValidatorFn.ts

@@ -20,6 +20,7 @@ import {
   VoluntaryExitErrorCode,
 } from "../../chain/errors/index.js";
 import {Metrics} from "../../metrics/index.js";
+import {PeerIdStr} from "../../util/peerId.js";
 import {INetworkCore} from "../core/index.js";
 import {
   BatchGossipHandlerFn,
@@ -40,6 +41,8 @@ export type ValidatorFnModules = {
   core: INetworkCore;
 };
 
+export type OnFatalPeerFn = (peer: PeerIdStr) => void;
+
 type RejectPeerActionRule = {default: PeerAction; byCode?: Record<string, PeerAction>};
 
 /**
@@ -112,7 +115,7 @@ const gossipRejectPeerAction: Record<GossipType, RejectPeerActionRule> = {
   [GossipType.light_client_optimistic_update]: {default: PeerAction.HighToleranceError},
 };
 
-function rejectPeerAction(type: GossipType, code: string): PeerAction {
+export function rejectPeerAction(type: GossipType, code: string): PeerAction {
   const rule = gossipRejectPeerAction[type];
   return rule.byCode?.[code] ?? rule.default;
 }
@@ -123,7 +126,8 @@ function rejectPeerAction(type: GossipType, code: string): PeerAction {
  */
 export function getGossipValidatorBatchFn(
   gossipHandlers: GossipHandlers,
-  modules: ValidatorFnModules
+  modules: ValidatorFnModules,
+  onFatalPeer: OnFatalPeerFn
 ): GossipValidatorBatchFn {
   const {logger, metrics, core} = modules;
 
@@ -173,6 +177,7 @@ export function getGossipValidatorBatchFn(
             metrics?.networkProcessor.gossipAttestationRejectByReason.inc({reason: e.type.code});
             const peerAction = rejectPeerAction(type, e.type.code);
             core.reportPeer(propagationSource, peerAction, e.type.code);
+            if (peerAction === PeerAction.Fatal) onFatalPeer(propagationSource);
             logger.debug(
               `Gossip validation ${type} rejected`,
               {peer: propagationSource, clientAgent, clientVersion, peerAction},
@@ -208,7 +213,11 @@ export function getGossipValidatorBatchFn(
  *
  * @see getGossipHandlers for reasoning on why GossipHandlerFn are used for gossip validation.
  */
-export function getGossipValidatorFn(gossipHandlers: GossipHandlers, modules: ValidatorFnModules): GossipValidatorFn {
+export function getGossipValidatorFn(
+  gossipHandlers: GossipHandlers,
+  modules: ValidatorFnModules,
+  onFatalPeer: OnFatalPeerFn
+): GossipValidatorFn {
   const {logger, metrics, core} = modules;
 
   return async function gossipValidatorFn({
@@ -260,6 +269,7 @@ export function getGossipValidatorFn(gossipHandlers: GossipHandlers, modules: Va
           metrics?.networkProcessor.gossipValidationReject.inc({topic: type});
           const peerAction = rejectPeerAction(type, e.type.code);
           core.reportPeer(propagationSource, peerAction, e.type.code);
+          if (peerAction === PeerAction.Fatal) onFatalPeer(propagationSource);
           logger.debug(
             `Gossip validation ${type} rejected`,
             {peer: propagationSource, clientAgent, clientVersion, peerAction},

package/src/network/processor/index.ts

@@ -22,6 +22,7 @@ import {
   getParentRootFromSignedBeaconBlockSerialized,
   getPayloadPresentFromPayloadAttestationMessageSerialized,
 } from "../../util/sszBytes.js";
+import {INetworkCore} from "../core/index.js";
 import {NetworkEvent, NetworkEventBus} from "../events.js";
 import {
   GossipHandlers,
@@ -34,7 +35,12 @@ import {MAX_PEERS_PER_ROOT} from "./constants.js";
 import {createExtractBlockSlotRootFns} from "./extractSlotRootFns.js";
 import {GossipHandlerOpts, ValidatorFnsModules, getGossipHandlers} from "./gossipHandlers.js";
 import {createGossipQueues} from "./gossipQueues/index.js";
-import {ValidatorFnModules, getGossipValidatorBatchFn, getGossipValidatorFn} from "./gossipValidatorFn.js";
+import {
+  ValidatorFnModules,
+  getGossipValidatorBatchFn,
+  getGossipValidatorFn,
+  rejectPeerAction,
+} from "./gossipValidatorFn.js";
 import {PendingGossipsubMessage} from "./types.js";
 
 export * from "./types.js";
@@ -102,6 +108,12 @@ type SearchedRootEntry = {
  */
 const DEFAULT_EARLIEST_PERMISSIBLE_SLOT_DISTANCE = 32;
 
+/**
+ * The network worker bans a Fatal peer only on its next PeerManager heartbeat (30s). Until then the main thread
+ * would keep validating that peer's messages, so drop them here. 3 slots (36s) outlasts one heartbeat.
+ */
+export const FATAL_PEER_EXPIRY_SLOTS = 3;
+
 type WorkOpts = {
   bypassQueue?: boolean;
 };
@@ -232,6 +244,7 @@ type SearchUnknownRootTarget =
  */
 export class NetworkProcessor {
   private readonly chain: IBeaconChain;
+  private readonly core: INetworkCore;
   private readonly events: NetworkEventBus;
   private readonly logger: Logger;
   private readonly metrics: Metrics | null;
@@ -256,23 +269,31 @@ export class NetworkProcessor {
     () => new MapDef<RootHex, SearchedRootEntry>(() => ({}))
   );
   private bufferedRootsBySlot = new MapDef<Slot, Set<RootHex>>(() => new Set());
+  /** peer -> clock slot when it was Fatal'ed */
+  private readonly fatalPeers = new Map<PeerIdStr, Slot>();
 
   constructor(
     modules: NetworkProcessorModules,
     private readonly opts: NetworkProcessorOpts
   ) {
-    const {chain, events, logger, metrics} = modules;
+    const {chain, core, events, logger, metrics} = modules;
     this.chain = chain;
+    this.core = core;
     this.events = events;
     this.metrics = metrics;
     this.logger = logger;
     this.events = events;
     this.gossipQueues = createGossipQueues();
     this.gossipTopicConcurrency = mapValues(this.gossipQueues, () => 0);
-    this.gossipValidatorFn = getGossipValidatorFn(modules.gossipHandlers ?? getGossipHandlers(modules, opts), modules);
+    this.gossipValidatorFn = getGossipValidatorFn(
+      modules.gossipHandlers ?? getGossipHandlers(modules, opts),
+      modules,
+      this.onFatalPeer
+    );
     this.gossipValidatorBatchFn = getGossipValidatorBatchFn(
       modules.gossipHandlers ?? getGossipHandlers(modules, opts),
-      modules
+      modules,
+      this.onFatalPeer
     );
 
     events.on(NetworkEvent.pendingGossipsubMessage, this.onPendingGossipsubMessage);
@@ -406,19 +427,47 @@ export class NetworkProcessor {
     }
   }
 
+  private onFatalPeer = (peer: PeerIdStr): void => {
+    this.fatalPeers.set(peer, this.chain.clock.currentSlot);
+  };
+
   private onPendingGossipsubMessage = (message: PendingGossipsubMessage): void => {
     const topicType = message.topic.type;
+    if (this.fatalPeers.has(message.propagationSource)) {
+      this.metrics?.networkProcessor.gossipValidationError.inc({topic: topicType, error: GossipErrorCode.FATAL_PEER});
+      // No need to report the dropped job to gossip. It will be eventually pruned from the mcache
+      return;
+    }
+
     const extractBlockSlotRootFn = this.extractBlockSlotRootFns[topicType];
+    if (extractBlockSlotRootFn === undefined) {
+      // some messages don't have slot and root
+      this.pushPendingGossipsubMessageToQueue(message);
+      return;
+    }
 
     // 1st extract round: make sure slot is in range and if block root is not available
     // proactively search for it + queue the message
-    const slotRoot = extractBlockSlotRootFn
-      ? extractBlockSlotRootFn(message.msg.data, message.topic.boundary.fork)
-      : null;
+    const slotRoot = extractBlockSlotRootFn(message.msg.data, message.topic.boundary.fork);
     if (slotRoot === null) {
-      // some messages don't have slot and root
-      // if the msg.data is invalid, message will be rejected when deserializing data in later phase (gossipValidatorFn)
-      this.pushPendingGossipsubMessageToQueue(message);
+      // DOS protection: null means malformed bytes or slot >= 2^32, neither is forwarded by honest peers.
+      const {propagationSource, clientAgent, clientVersion} = message;
+      const code = GossipErrorCode.INVALID_SLOT;
+      const peerAction = rejectPeerAction(topicType, code);
+      this.metrics?.networkProcessor.gossipValidationReject.inc({topic: topicType});
+      this.core.reportPeer(propagationSource, peerAction, code);
+      this.logger.debug(`Gossip validation ${topicType} rejected due to null slot`, {
+        peer: propagationSource,
+        clientAgent,
+        clientVersion,
+        peerAction,
+        code,
+      });
+      this.events.emit(NetworkEvent.gossipMessageValidationResult, {
+        msgId: message.msgId,
+        propagationSource,
+        acceptance: TopicValidatorResult.Reject,
+      });
       return;
     }
 
@@ -872,6 +921,10 @@ export class NetworkProcessor {
     for (const slot of this.bufferedRootsBySlot.keys()) {
       if (slot <= minSlot) this.bufferedRootsBySlot.delete(slot);
     }
+
+    for (const [peer, slot] of this.fatalPeers) {
+      if (clockSlot - slot > FATAL_PEER_EXPIRY_SLOTS) this.fatalPeers.delete(peer);
+    }
   };
 
   private executeWork(): void {
@@ -927,6 +980,32 @@ export class NetworkProcessor {
   private async processPendingGossipsubMessage(
     messageOrArray: PendingGossipsubMessage | PendingGossipsubMessage[]
   ): Promise<void> {
+    // drop messages from peers Fatal'ed after these were queued or started awaiting a block/envelope
+    // No need to report the dropped jobs to gossip. They will be eventually pruned from the mcache
+    if (this.fatalPeers.size > 0) {
+      if (Array.isArray(messageOrArray)) {
+        const allowed: PendingGossipsubMessage[] = [];
+        for (const msg of messageOrArray) {
+          if (this.fatalPeers.has(msg.propagationSource)) {
+            this.metrics?.networkProcessor.gossipValidationError.inc({
+              topic: msg.topic.type,
+              error: GossipErrorCode.FATAL_PEER,
+            });
+          } else {
+            allowed.push(msg);
+          }
+        }
+        if (allowed.length === 0) return;
+        messageOrArray = allowed;
+      } else if (this.fatalPeers.has(messageOrArray.propagationSource)) {
+        this.metrics?.networkProcessor.gossipValidationError.inc({
+          topic: messageOrArray.topic.type,
+          error: GossipErrorCode.FATAL_PEER,
+        });
+        return;
+      }
+    }
+
     const nowSec = Date.now() / 1000;
     if (Array.isArray(messageOrArray)) {
       for (const msg of messageOrArray) {

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

@@ -53,7 +53,7 @@ export async function* onExecutionPayloadEnvelopesByRoot(
     const batch = requests.slice(i, i + MAX_BODIES_PER_REQUEST);
     let envelopesBytes: (Uint8Array | null)[];
     try {
-      // By-root allows omission, so a mismatched envelope is left out rather than failing the response.
+      // 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) {

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";
@@ -171,7 +177,13 @@ 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 +197,13 @@ export class BeaconNode {
             config,
             anchorState.genesisTime,
             logger.child({module: LoggerModule.vmon}),
-            opts.validatorMonitor
+            opts.validatorMonitor,
+            opts.chain.nativeStateTransition
+              ? {
+                  registerLocalValidator: registerNativeLocalValidator,
+                  unregisterLocalValidator: unregisterNativeLocalValidator,
+                }
+              : null
           )
         : null;
 
@@ -303,7 +321,17 @@ 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

@@ -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/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;
     }
   };