Add breakdown of fraud detection timings. #460

lib/evaluate.js

@@ -55,7 +55,7 @@ export const evaluate = async ({
 
   const started = Date.now()
 
-  const { committees } = await runFraudDetection({
+  const { committees, timings: fraudDetectionTimings } = await runFraudDetection({
     roundIndex,
     measurements,
     sparkRoundDetails,
@@ -143,6 +143,9 @@ export const evaluate = async ({
     point.intField('honest_measurements', measurementsToReward.length)
     point.intField('set_scores_duration_ms', setScoresDuration)
     point.intField('fraud_detection_duration_ms', fraudDetectionDuration)
+    for (const { influxField, duration } of fraudDetectionTimings) {
+      point.intField(`fraud_detection_timings_${influxField}_ms`, duration)
+    }
 
     for (const [type, count] of Object.entries(evaluationOutcomes)) {
       point.intField(`measurements_${type}`, count)
@@ -236,6 +239,7 @@ export const evaluate = async ({
  * @param {import('./typings.js').RoundDetails} args.sparkRoundDetails
  * @param {number} args.requiredCommitteeSize
  * @param {Pick<Console, 'log' | 'error'>} args.logger
+ * @returns {Promise<{committees: import('./committee.js').Committee[], timings: {influxField: string, duration: number}[]}>}
  */
 export const runFraudDetection = async ({
   roundIndex,
@@ -244,6 +248,29 @@ export const runFraudDetection = async ({
   requiredCommitteeSize,
   logger
 }) => {
+  const timings = {
+    taskBuilding: {
+      influxField: 'task_building',
+      start: null,
+      end: null
+    },
+    taskingEvaluation: {
+      influxField: 'tasking_evaluation',
+      start: null,
+      end: null
+    },
+    inetGroupsEvaluation: {
+      influxField: 'inet_groups_evaluation',
+      start: null,
+      end: null
+    },
+    majorityEvaluation: {
+      influxField: 'majority_evaluation',
+      start: null,
+      end: null
+    }
+  }
+
   const randomness = await pRetry(
     () => getRandomnessForSparkRound(Number(sparkRoundDetails.startEpoch)),
     {
@@ -254,17 +281,18 @@ export const runFraudDetection = async ({
     }
   )
 
-  const taskBuildingStarted = Date.now()
+  timings.taskBuilding.start = new Date()
   const tasksAllowedForStations = await getTasksAllowedForStations(
     sparkRoundDetails,
     randomness,
     measurements
   )
+  timings.taskBuilding.end = new Date()
 
   logger.log(
     'EVALUATE ROUND %s: built per-node task lists in %sms [Tasks=%s;TN=%s;Nodes=%s]',
     roundIndex,
-    Date.now() - taskBuildingStarted,
+    timings.taskBuilding.end - timings.taskBuilding.start,
     sparkRoundDetails.retrievalTasks.length,
     sparkRoundDetails.maxTasksPerNode,
     tasksAllowedForStations.size
@@ -274,6 +302,7 @@ export const runFraudDetection = async ({
   // 1. Filter out measurements not belonging to any valid task in this round
   //    or missing some of the required fields like `inet_group`
   //
+  timings.taskingEvaluation.start = new Date()
   for (const m of measurements) {
     // sanity checks to get nicer errors if we forget to set required fields in unit tests
     assert(typeof m.inet_group === 'string', 'missing inet_group')
@@ -294,10 +323,12 @@ export const runFraudDetection = async ({
       m.taskingEvaluation = 'TASK_WRONG_NODE'
     }
   }
+  timings.taskingEvaluation.end = new Date()
 
   //
   // 2. Accept only maxTasksPerNode measurements from each inet group
   //
+  timings.inetGroupsEvaluation.start = new Date()
   /** @type {Map<string, Measurement[]>} */
   const inetGroups = new Map()
   for (const m of measurements) {
@@ -364,10 +395,12 @@ export const runFraudDetection = async ({
       debug('  pa: %s h: %s task: %s - REWARD', m.participantAddress, h, taskId)
     }
   }
+  timings.inetGroupsEvaluation.end = new Date()
 
   //
   // 3. Group measurements to per-task committees and find majority result
   //
+  timings.majorityEvaluation.start = new Date()
 
   // PERFORMANCE: Avoid duplicating the array of measurements because there are
   // hundreds of thousands of them. All the function groupMeasurementsToCommittees
@@ -384,6 +417,7 @@ export const runFraudDetection = async ({
   for (const c of committees.values()) {
     c.evaluate({ requiredCommitteeSize })
   }
+  timings.majorityEvaluation.end = new Date()
 
   if (debug.enabled) {
     for (const m of measurements) {
@@ -399,7 +433,12 @@ export const runFraudDetection = async ({
     }
   }
 
-  return { committees: Array.from(committees.values()) }
+  return {
+    committees: Array.from(committees.values()),
+    timings: Object
+      .values(timings)
+      .map(({ influxField, start, end }) => ({ influxField, duration: end - start }))
+  }
 }
 
 /**