feat(beacon-node): network worker event latency metrics

dashboards/lodestar_networking.json

@@ -1,12 +1,12 @@
 {
   "__inputs": [
     {
-      "description": "",
-      "label": "Prometheus",
       "name": "DS_PROMETHEUS",
+      "type": "datasource",
+      "label": "Prometheus",
+      "description": "",
       "pluginId": "prometheus",
-      "pluginName": "Prometheus",
-      "type": "datasource"
+      "pluginName": "Prometheus"
     }
   ],
   "annotations": {
@@ -937,7 +937,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "(\n  sum(rate(\n    lodestar_network_worker_wire_events_on_worker_thread_latency_sum[$rate_interval]\n  )) \n  +\n  sum(rate(\n    lodestar_network_worker_wire_events_on_main_thread_latency_sum[$rate_interval]\n  ))\n)\n/\n(\n  sum(rate(\n    lodestar_network_worker_wire_events_on_worker_thread_latency_count[$rate_interval]\n  ))\n  +\n  sum(rate(\n    lodestar_network_worker_wire_events_on_main_thread_latency_count[$rate_interval]\n  ))\n)",
+          "expr": "(\n  sum(rate(\n    lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_sum[$rate_interval]\n  )) \n  +\n  sum(rate(\n    lodestar_network_worker_wire_events_on_main_thread_latency_seconds_sum[$rate_interval]\n  ))\n)\n/\n(\n  sum(rate(\n    lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_count[$rate_interval]\n  ))\n  +\n  sum(rate(\n    lodestar_network_worker_wire_events_on_main_thread_latency_seconds_count[$rate_interval]\n  ))\n)",
           "hide": false,
           "interval": "",
           "legendFormat": "Average",
@@ -951,7 +951,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "avg(rate(lodestar_network_worker_wire_events_on_worker_thread_latency_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_worker_thread_latency_count[$rate_interval]))",
+          "expr": "avg(rate(lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_count[$rate_interval]))",
           "hide": false,
           "interval": "",
           "legendFormat": "Worker to Main",
@@ -964,7 +964,7 @@
             "uid": "${DS_PROMETHEUS}"
           },
           "editorMode": "code",
-          "expr": "avg(rate(lodestar_network_worker_wire_events_on_main_thread_latency_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_count[$rate_interval]))",
+          "expr": "avg(rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_count[$rate_interval]))",
           "hide": false,
           "legendFormat": "Main to Worker",
           "range": true,
@@ -1051,7 +1051,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "avg(rate(lodestar_network_worker_wire_events_on_main_thread_latency_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_count[$rate_interval]))",
+          "expr": "avg(rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_count[$rate_interval]))",
           "hide": true,
           "interval": "",
           "legendFormat": "Average to Main",
@@ -1065,7 +1065,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "rate(lodestar_network_worker_wire_events_on_main_thread_latency_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_count[$rate_interval])",
+          "expr": "rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_main_thread_latency_seconds_count[$rate_interval])",
           "interval": "",
           "legendFormat": "{{eventName}}",
           "range": true,
@@ -1077,7 +1077,7 @@
             "uid": "${DS_PROMETHEUS}"
           },
           "editorMode": "code",
-          "expr": "rate(lodestar_network_worker_wire_events_on_worker_thread_latency_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_worker_thread_latency_count[$rate_interval])",
+          "expr": "rate(lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_sum[$rate_interval])/rate(lodestar_network_worker_wire_events_on_worker_thread_latency_seconds_count[$rate_interval])",
           "hide": false,
           "legendFormat": "{{eventName}}",
           "range": true,

packages/beacon-node/src/metrics/metrics/lodestar.ts

@@ -121,6 +121,12 @@ export function createLodestarMetrics(
         help: "Current count of pending items in reqRespBridgeReqCaller data structure",
       }),
     },
+    networkWorkerWireEventsOnMainThreadLatency: register.histogram<"eventName">({
+      name: "lodestar_network_worker_wire_events_on_main_thread_latency_seconds",
+      help: "Latency in seconds to transmit network events to main thread across worker port",
+      labelNames: ["eventName"],
+      buckets: [0.001, 0.003, 0.01, 0.03, 0.1],
+    }),
 
     regenQueue: {
       length: register.gauge({

packages/beacon-node/src/network/core/metrics.ts

@@ -333,12 +333,20 @@ export function createNetworkCoreMetrics(register: RegistryMetricCreator) {
   };
 }
 
+export type NetworkCoreWorkerMetrics = ReturnType<typeof getNetworkCoreWorkerMetrics>;
+
 // eslint-disable-next-line @typescript-eslint/explicit-function-return-type
 export function getNetworkCoreWorkerMetrics(register: RegistryMetricCreator) {
   return {
     reqRespBridgeRespCallerPending: register.gauge({
       name: "lodestar_network_worker_reqresp_bridge_caller_pending_count",
       help: "Current count of pending elements in respBridgeCaller",
     }),
+    networkWorkerWireEventsOnWorkerThreadLatency: register.histogram<"eventName">({
+      name: "lodestar_network_worker_wire_events_on_worker_thread_latency_seconds",
+      help: "Latency in seconds to transmit network events to worker thread across parent port",
+      labelNames: ["eventName"],
+      buckets: [0.001, 0.003, 0.01, 0.03, 0.1],
+    }),
   };
 }

packages/beacon-node/src/network/core/networkCoreWorker.ts

@@ -13,6 +13,9 @@ import {peerIdToString} from "../../util/peerId.js";
 import {profileNodeJS} from "../../util/profile.js";
 import {NetworkEventBus, NetworkEventData, networkEventDirection} from "../events.js";
 import {wireEventsOnWorkerThread} from "../../util/workerEvents.js";
+import {getNetworkCoreWorkerMetrics} from "./metrics.js";
+import {NetworkWorkerApi, NetworkWorkerData} from "./types.js";
+import {NetworkCore} from "./networkCore.js";
 import {
   NetworkWorkerThreadEventType,
   ReqRespBridgeEventBus,
@@ -21,9 +24,6 @@ import {
   getReqRespBridgeRespEvents,
   reqRespBridgeEventDirection,
 } from "./events.js";
-import {getNetworkCoreWorkerMetrics} from "./metrics.js";
-import {NetworkCore} from "./networkCore.js";
-import {NetworkWorkerApi, NetworkWorkerData} from "./types.js";
 
 // Cloned data from instantiation
 const workerData = worker.workerData as NetworkWorkerData;
@@ -83,9 +83,9 @@ new AsyncIterableBridgeHandler(getReqRespBridgeReqEvents(reqRespBridgeEventBus),
 );
 const reqRespBridgeRespCaller = new AsyncIterableBridgeCaller(getReqRespBridgeRespEvents(reqRespBridgeEventBus));
 
+const networkCoreWorkerMetrics = metricsRegister ? getNetworkCoreWorkerMetrics(metricsRegister) : null;
 // respBridgeCaller metrics
-if (metricsRegister) {
-  const networkCoreWorkerMetrics = getNetworkCoreWorkerMetrics(metricsRegister);
+if (networkCoreWorkerMetrics) {
   networkCoreWorkerMetrics.reqRespBridgeRespCallerPending.addCollect(() => {
     networkCoreWorkerMetrics.reqRespBridgeRespCallerPending.set(reqRespBridgeRespCaller.pendingCount);
   });
@@ -110,12 +110,14 @@ wireEventsOnWorkerThread<NetworkEventData>(
   NetworkWorkerThreadEventType.networkEvent,
   events,
   parentPort,
+  networkCoreWorkerMetrics,
   networkEventDirection
 );
 wireEventsOnWorkerThread<ReqRespBridgeEventData>(
   NetworkWorkerThreadEventType.reqRespBridgeEvents,
   reqRespBridgeEventBus,
   parentPort,
+  networkCoreWorkerMetrics,
   reqRespBridgeEventDirection
 );
 

packages/beacon-node/src/network/core/networkCoreWorkerHandler.ts

@@ -75,12 +75,14 @@ export class WorkerNetworkCore implements INetworkCore {
       NetworkWorkerThreadEventType.networkEvent,
       modules.events,
       modules.worker as unknown as worker_threads.Worker,
+      modules.metrics,
       networkEventDirection
     );
     wireEventsOnMainThread<ReqRespBridgeEventData>(
       NetworkWorkerThreadEventType.reqRespBridgeEvents,
       this.reqRespBridgeEventBus,
       modules.worker as unknown as worker_threads.Worker,
+      modules.metrics,
       reqRespBridgeEventDirection
     );
 

packages/beacon-node/src/util/workerEvents.ts

@@ -2,11 +2,16 @@ import {MessagePort, Worker} from "node:worker_threads";
 import {Thread} from "@chainsafe/threads";
 import {Logger} from "@lodestar/logger";
 import {sleep} from "@lodestar/utils";
+import {Metrics} from "../metrics/metrics.js";
+import {NetworkCoreWorkerMetrics} from "../network/core/metrics.js";
 import {StrictEventEmitterSingleArg} from "./strictEvents.js";
 
+const NANO_TO_SECOND_CONVERSION = 1e9;
+
 export type WorkerBridgeEvent<EventData> = {
   type: string;
   event: keyof EventData;
+  posted: [number, number];
   data: EventData[keyof EventData];
 };
 
@@ -27,6 +32,7 @@ export function wireEventsOnWorkerThread<EventData>(
   mainEventName: string,
   events: StrictEventEmitterSingleArg<EventData>,
   parentPort: MessagePort,
+  metrics: NetworkCoreWorkerMetrics | null,
   isWorkerToMain: {[K in keyof EventData]: EventDirection}
 ): void {
   // Subscribe to events from main thread
@@ -37,6 +43,12 @@ export function wireEventsOnWorkerThread<EventData>(
       // This check is not necessary but added for safety in case of improper implemented events
       isWorkerToMain[data.event] === EventDirection.mainToWorker
     ) {
+      const [sec, nanoSec] = process.hrtime(data.posted);
+      const networkWorkerLatency = sec + nanoSec / NANO_TO_SECOND_CONVERSION;
+      metrics?.networkWorkerWireEventsOnWorkerThreadLatency.observe(
+        {eventName: data.event as string},
+        networkWorkerLatency
+      );
       events.emit(data.event, data.data);
     }
   });
@@ -48,6 +60,7 @@ export function wireEventsOnWorkerThread<EventData>(
         const workerEvent: WorkerBridgeEvent<EventData> = {
           type: mainEventName,
           event: eventName,
+          posted: process.hrtime(),
           data,
         };
         parentPort.postMessage(workerEvent);
@@ -60,6 +73,7 @@ export function wireEventsOnMainThread<EventData>(
   mainEventName: string,
   events: StrictEventEmitterSingleArg<EventData>,
   worker: Pick<Worker, "on" | "postMessage">,
+  metrics: Metrics | null,
   isWorkerToMain: {[K in keyof EventData]: EventDirection}
 ): void {
   // Subscribe to events from main thread
@@ -70,6 +84,12 @@ export function wireEventsOnMainThread<EventData>(
       // This check is not necessary but added for safety in case of improper implemented events
       isWorkerToMain[data.event] === EventDirection.workerToMain
     ) {
+      const [sec, nanoSec] = process.hrtime(data.posted);
+      const networkWorkerLatency = sec + nanoSec / NANO_TO_SECOND_CONVERSION;
+      metrics?.networkWorkerWireEventsOnMainThreadLatency.observe(
+        {eventName: data.event as string},
+        networkWorkerLatency
+      );
       events.emit(data.event, data.data);
     }
   });
@@ -81,6 +101,7 @@ export function wireEventsOnMainThread<EventData>(
         const workerEvent: WorkerBridgeEvent<EventData> = {
           type: mainEventName,
           event: eventName,
+          posted: process.hrtime(),
           data,
         };
         worker.postMessage(workerEvent);