@lodestar/beacon-node 1.46.0-dev.e6f04acd82 → 1.46.0-dev.eabf12abae

package/src/network/core/networkCore.ts

@@ -70,6 +70,15 @@ export type BaseNetworkInit = {
   initialCustodyGroupCount: number;
 };
 
+/** Counts of what is still keeping this thread's event loop alive, e.g. `TCPWrap=3,Timeout=1` */
+function formatActiveResources(): string {
+  const counts = new Map<string, number>();
+  for (const resource of process.getActiveResourcesInfo()) {
+    counts.set(resource, (counts.get(resource) ?? 0) + 1);
+  }
+  return Array.from(counts, ([name, count]) => `${name}=${count}`).join(",");
+}
+
 /**
  * This class is meant to work both:
  * - In a libp2p worker
@@ -289,7 +298,10 @@ export class NetworkCore implements INetworkCore {
     this.attnetsService.close();
     this.syncnetsService.close();
     await this.libp2p.stop();
-    this.logger.debug("network lib2p closed");
+    // Diagnostic for the shutdown hang, this thread can spin in `Environment::CleanupHandles()` on
+    // a handle that never closes and `Worker.terminate()` then never resolves. Diffing this list
+    // between a clean and a stuck shutdown should narrow down which handle it is
+    this.logger.debug("network lib2p closed", {activeResources: formatActiveResources()});
 
     this.closed = true;
   }

package/src/network/core/networkCoreWorkerHandler.ts

@@ -10,6 +10,7 @@ import {BeaconConfig, chainConfigToJson} from "@lodestar/config";
 import type {LoggerNode} from "@lodestar/logger/node";
 import {ResponseIncoming, ResponseOutgoing} from "@lodestar/reqresp";
 import {Status} from "@lodestar/types";
+import {withTimeout} from "@lodestar/utils";
 import {Metrics} from "../../metrics/index.js";
 import {AsyncIterableBridgeCaller, AsyncIterableBridgeHandler} from "../../util/asyncIterableToEvents.js";
 import {PeerIdStr, peerIdFromString} from "../../util/peerId.js";
@@ -58,6 +59,10 @@ type WorkerNetworkCoreModules = WorkerNetworkCoreInitModules & {
 
 const NETWORK_WORKER_EXIT_TIMEOUT_MS = 1000;
 const NETWORK_WORKER_EXIT_RETRY_COUNT = 3;
+/** `getApi().close()` is an unbounded RPC into the worker and runs before `BeaconNode.close()`
+ * archives chain state to disk, so it is bounded here. Set well above the ~2-4s a close takes on
+ * mainnet, tripping it leaves libp2p half torn down */
+const NETWORK_CORE_CLOSE_TIMEOUT_MS = 5000;
 
 /**
  * NetworkCore implementation using a Worker thread
@@ -165,15 +170,25 @@ export class WorkerNetworkCore implements INetworkCore {
 
   async close(): Promise<void> {
     this.modules.logger.debug("closing network core running in network worker");
-    await this.getApi().close();
+    try {
+      await withTimeout(async () => this.getApi().close(), NETWORK_CORE_CLOSE_TIMEOUT_MS);
+    } catch (e) {
+      this.modules.logger.debug("Error closing network core", {}, e as Error);
+    }
     this.modules.logger.debug("terminating network worker");
-    await terminateWorkerThread({
+    const terminated = await terminateWorkerThread({
       worker: this.getApi(),
       retryCount: NETWORK_WORKER_EXIT_RETRY_COUNT,
       retryMs: NETWORK_WORKER_EXIT_TIMEOUT_MS,
       logger: this.modules.logger,
     });
-    this.modules.logger.debug("terminated network worker");
+    if (terminated) {
+      this.modules.logger.debug("terminated network worker");
+    } else {
+      // Nothing more to do, `process.exit()` joins the thread regardless so the process still waits
+      // on it. What matters is returning, so the caller can archive state and close the db
+      this.modules.logger.debug("Network worker did not terminate in time, continuing shutdown without it");
+    }
   }
 
   async test(): Promise<void> {

package/src/network/network.ts

@@ -147,6 +147,7 @@ export class Network implements INetwork {
     this.chain.emitter.on(ChainEvent.updateTargetCustodyGroupCount, this.onTargetGroupCountUpdated);
     this.chain.emitter.on(ChainEvent.publishDataColumns, this.onPublishDataColumns);
     this.chain.emitter.on(ChainEvent.publishBlobSidecars, this.onPublishBlobSidecars);
+    this.chain.emitter.on(ChainEvent.publishProposerSlashing, this.onPublishProposerSlashing);
     this.chain.emitter.on(ChainEvent.updateStatus, this.onUpdateStatus);
   }
 
@@ -244,6 +245,7 @@ export class Network implements INetwork {
     this.chain.emitter.off(ChainEvent.updateTargetCustodyGroupCount, this.onTargetGroupCountUpdated);
     this.chain.emitter.off(ChainEvent.publishDataColumns, this.onPublishDataColumns);
     this.chain.emitter.off(ChainEvent.publishBlobSidecars, this.onPublishBlobSidecars);
+    this.chain.emitter.off(ChainEvent.publishProposerSlashing, this.onPublishProposerSlashing);
     this.chain.emitter.off(ChainEvent.updateStatus, this.onUpdateStatus);
     await this.core.close();
 
@@ -455,7 +457,8 @@ export class Network implements INetwork {
 
     return this.publishGossip<GossipType.proposer_slashing>(
       {type: GossipType.proposer_slashing, boundary},
-      proposerSlashing
+      proposerSlashing,
+      {ignoreDuplicatePublishError: true}
     );
   }
 
@@ -877,6 +880,18 @@ export class Network implements INetwork {
     return promiseAllMaybeAsync(sidecars.map((sidecar) => () => this.publishBlobSidecar(sidecar)));
   };
 
+  private onPublishProposerSlashing = async (proposerSlashing: phase0.ProposerSlashing): Promise<void> => {
+    try {
+      await this.publishProposerSlashing(proposerSlashing);
+    } catch (e) {
+      this.logger.debug(
+        "Error publishing proposer slashing",
+        {proposerIndex: proposerSlashing.signedHeader1.message.proposerIndex},
+        e as Error
+      );
+    }
+  };
+
   private onUpdateStatus = async (): Promise<void> => {
     await this.core.updateStatus(this.chain.getStatus());
   };

package/src/network/processor/gossipHandlers.ts

@@ -257,6 +257,13 @@ function getSequentialHandlers(modules: ValidatorFnsModules, options: GossipHand
         chain.seenPayloadEnvelopeInputCache.prune(blockRootHex);
       }
       throw e;
+    } finally {
+      // The block received from the network may have established an equivocation, either by conflicting
+      // with a previously observed block root (REPEAT_PROPOSAL) or with a root observed during validation
+      const proposerIndex = signedBlock.message.proposerIndex;
+      if (chain.seenBlockProposers.isEquivocating(slot, proposerIndex)) {
+        chain.processProposerEquivocation(slot, proposerIndex);
+      }
     }
   }
 

package/src/network/reqresp/score.ts

@@ -33,9 +33,28 @@ export function onOutgoingReqRespError(e: RequestError, method: ReqRespMethod):
 
     case RequestErrorCode.DIAL_TIMEOUT:
     case RequestErrorCode.DIAL_ERROR:
-      return e.message.includes(multiStreamSelectErrorCodes.protocolSelectionFailed) && method === ReqRespMethod.Ping
-        ? PeerAction.Fatal
-        : PeerAction.LowToleranceError;
+      // `RequestError` renders `e.message` as just the error code (see `renderErrorMessage`), so the
+      // multistream "protocol selection failed" text is only on the wrapped inner error, which is
+      // available on `DIAL_ERROR` via `e.type.error` (`DIAL_TIMEOUT` carries no inner error).
+      if (
+        e.type.code === RequestErrorCode.DIAL_ERROR &&
+        e.type.error.message.includes(multiStreamSelectErrorCodes.protocolSelectionFailed)
+      ) {
+        // Peer does not support the protocol, a real incompatibility rather than a transient
+        // failure, so keep the stronger penalty (Fatal for Ping, as before).
+        return method === ReqRespMethod.Ping ? PeerAction.Fatal : PeerAction.LowToleranceError;
+      }
+      switch (method) {
+        // Ping and Status are liveness probes; their dial timeouts are dominated by transient
+        // network congestion rather than peer misbehavior, so penalize leniently to avoid
+        // self-inflicted peer starvation (https://github.com/ChainSafe/lodestar/issues/9562),
+        // while still applying some penalty so genuinely dead peers eventually free the slot.
+        case ReqRespMethod.Ping:
+        case ReqRespMethod.Status:
+          return PeerAction.HighToleranceError;
+        default:
+          return PeerAction.LowToleranceError;
+      }
     // TODO: Detect SSZDecodeError and return PeerAction.Fatal
 
     case RequestErrorCode.RESP_TIMEOUT:

package/src/util/clock.ts

@@ -94,10 +94,7 @@ export class Clock extends EventEmitter implements IClock {
   }
   /**
    * If it's too close to next slot given MAXIMUM_GOSSIP_CLOCK_DISPARITY, return currentSlot + 1.
-   * Otherwise return currentSlot
-   *
-   * Spec: phase0/p2p-interface.md - gossip validation uses `current_time + MAXIMUM_GOSSIP_CLOCK_DISPARITY < message_time`
-   * to reject future messages (strict `<`), so the boundary (exactly equal) is accepted, hence `<=` here.
+   * Spec: phase0/p2p-interface.md rejects future messages with strict `<`, hence `<=` here.
    */
   get currentSlotWithGossipDisparity(): Slot {
     const currentSlot = this.currentSlot;
@@ -123,9 +120,7 @@ export class Clock extends EventEmitter implements IClock {
 
   /**
    * Check if a slot is current slot given MAXIMUM_GOSSIP_CLOCK_DISPARITY.
-   *
-   * Uses `<=` for disparity checks because the spec rejects with strict `<`
-   * (phase0/p2p-interface.md), meaning the boundary (exactly equal) is accepted.
+   * Spec: phase0/p2p-interface.md rejects future messages with strict `<`, hence `<=`.
    */
   isCurrentSlotGivenGossipDisparity(slot: Slot): boolean {
     const currentSlot = this.currentSlot;

package/src/util/workerEvents.ts

@@ -111,6 +111,12 @@ export function wireEventsOnMainThread<EventData>(
   }
 }
 
+/**
+ * Terminate a worker thread, bounded to `retryCount * retryMs`.
+ *
+ * @returns `false` if it could not be terminated, the worker thread is then still running and the
+ * caller has to decide how to proceed without it.
+ */
 export async function terminateWorkerThread({
   worker,
   retryMs,
@@ -121,7 +127,7 @@ export async function terminateWorkerThread({
   retryMs: number;
   retryCount: number;
   logger?: Logger;
-}): Promise<void> {
+}): Promise<boolean> {
   const terminated = new Promise((resolve) => {
     Thread.events(worker).subscribe((event) => {
       if (event.type === "termination") {
@@ -131,13 +137,19 @@ export async function terminateWorkerThread({
   });
 
   for (let i = 0; i < retryCount; i++) {
-    await Thread.terminate(worker);
-    const result = await Promise.race([terminated, sleep(retryMs).then(() => false)]);
+    // `Worker.terminate()` resolves only once the worker thread exits, and the thread can spin
+    // forever in `Environment::CleanupHandles()` when a libuv handle on its loop never closes, so
+    // it has to be raced too, otherwise `retryCount * retryMs` never applies.
+    const result = await Promise.race([
+      Thread.terminate(worker).then(() => terminated),
+      sleep(retryMs).then(() => false),
+    ]);
 
-    if (result) return;
+    if (result) return true;
 
-    logger?.warn("Worker thread failed to terminate, retrying...");
+    logger?.debug("Worker thread failed to terminate, retrying...");
   }
 
-  throw new Error(`Worker thread failed to terminate in ${retryCount * retryMs}ms.`);
+  logger?.debug(`Worker thread failed to terminate in ${retryCount * retryMs}ms`);
+  return false;
 }