feat(performance): add recent success rates to model summary and enhance metrics retrieval

pkg/perf_metrics/metrics.go

@@ -133,20 +133,23 @@ func QuerySummaryAll(hours int, groups []string) (SummaryAllResult, error) {
 	startTs := endTs - int64(hours)*3600
 	allowedGroups := allowedGroupSet(groups)
 
-	rows, err := model.GetPerfMetricsSummaryAll(startTs, endTs, groups)
+	rows, err := model.GetPerfMetricsSummaryBucketsAll(startTs, endTs, groups)
 	if err != nil {
 		return SummaryAllResult{}, err
 	}
 
 	totals := map[string]counters{}
+	modelBuckets := map[string]map[int64]counters{}
 	for _, row := range rows {
-		totals[row.ModelName] = counters{
+		value := counters{
 			requestCount:   row.RequestCount,
 			successCount:   row.SuccessCount,
 			totalLatencyMs: row.TotalLatencyMs,
 			outputTokens:   row.OutputTokens,
 			generationMs:   row.GenerationMs,
 		}
+		mergeModelTotals(totals, row.ModelName, value)
+		mergeModelBucket(modelBuckets, row.ModelName, row.BucketTs, value)
 	}
 
 	hotBuckets.Range(func(key, value any) bool {
@@ -163,13 +166,8 @@ func QuerySummaryAll(hours int, groups []string) (SummaryAllResult, error) {
 		if snap.requestCount == 0 {
 			return true
 		}
-		cur := totals[k.model]
-		cur.requestCount += snap.requestCount
-		cur.successCount += snap.successCount
-		cur.totalLatencyMs += snap.totalLatencyMs
-		cur.outputTokens += snap.outputTokens
-		cur.generationMs += snap.generationMs
-		totals[k.model] = cur
+		mergeModelTotals(totals, k.model, snap)
+		mergeModelBucket(modelBuckets, k.model, k.bucketTs, snap)
 		return true
 	})
 
@@ -185,11 +183,12 @@ func QuerySummaryAll(hours int, groups []string) (SummaryAllResult, error) {
 			avgTps = float64(total.outputTokens) / (float64(total.generationMs) / 1000.0)
 		}
 		models = append(models, ModelSummary{
-			ModelName:    name,
-			AvgLatencyMs: avgLatency,
-			SuccessRate:  math.Round(successRate*100) / 100,
-			AvgTps:       math.Round(avgTps*100) / 100,
-			RequestCount: total.requestCount,
+			ModelName:          name,
+			AvgLatencyMs:       avgLatency,
+			SuccessRate:        math.Round(successRate*100) / 100,
+			AvgTps:             math.Round(avgTps*100) / 100,
+			RecentSuccessRates: recentSuccessRates(modelBuckets[name], 3),
+			RequestCount:       total.requestCount,
 		})
 	}
 	sort.Slice(models, func(i, j int) bool {
@@ -199,6 +198,60 @@ func QuerySummaryAll(hours int, groups []string) (SummaryAllResult, error) {
 	return SummaryAllResult{Models: models}, nil
 }
 
+func mergeModelTotals(totals map[string]counters, modelName string, value counters) {
+	if value.requestCount == 0 {
+		return
+	}
+	current := totals[modelName]
+	current.requestCount += value.requestCount
+	current.successCount += value.successCount
+	current.totalLatencyMs += value.totalLatencyMs
+	current.ttftSumMs += value.ttftSumMs
+	current.ttftCount += value.ttftCount
+	current.outputTokens += value.outputTokens
+	current.generationMs += value.generationMs
+	totals[modelName] = current
+}
+
+func mergeModelBucket(modelBuckets map[string]map[int64]counters, modelName string, bucketTs int64, value counters) {
+	if value.requestCount == 0 {
+		return
+	}
+	if _, ok := modelBuckets[modelName]; !ok {
+		modelBuckets[modelName] = map[int64]counters{}
+	}
+	current := modelBuckets[modelName][bucketTs]
+	current.requestCount += value.requestCount
+	current.successCount += value.successCount
+	current.totalLatencyMs += value.totalLatencyMs
+	current.ttftSumMs += value.ttftSumMs
+	current.ttftCount += value.ttftCount
+	current.outputTokens += value.outputTokens
+	current.generationMs += value.generationMs
+	modelBuckets[modelName][bucketTs] = current
+}
+
+func recentSuccessRates(buckets map[int64]counters, limit int) []float64 {
+	if len(buckets) == 0 || limit <= 0 {
+		return nil
+	}
+	timestamps := make([]int64, 0, len(buckets))
+	for ts := range buckets {
+		timestamps = append(timestamps, ts)
+	}
+	sort.Slice(timestamps, func(i, j int) bool {
+		return timestamps[i] < timestamps[j]
+	})
+	if len(timestamps) > limit {
+		timestamps = timestamps[len(timestamps)-limit:]
+	}
+	rates := make([]float64, 0, len(timestamps))
+	for _, ts := range timestamps {
+		rates = append(rates, math.Round(successRate(buckets[ts])*100)/100)
+	}
+	return rates
+}
+
 func allowedGroupSet(groups []string) map[string]struct{} {
 	if groups == nil {
 		return nil

pkg/perf_metrics/types.go

@@ -48,11 +48,12 @@ type QueryResult struct {
 }
 
 type ModelSummary struct {
-	ModelName    string  `json:"model_name"`
-	AvgLatencyMs int64   `json:"avg_latency_ms"`
-	SuccessRate  float64 `json:"success_rate"`
-	AvgTps       float64 `json:"avg_tps"`
-	RequestCount int64   `json:"-"`
+	ModelName          string    `json:"model_name"`
+	AvgLatencyMs       int64     `json:"avg_latency_ms"`
+	SuccessRate        float64   `json:"success_rate"`
+	AvgTps             float64   `json:"avg_tps"`
+	RecentSuccessRates []float64 `json:"recent_success_rates,omitempty"`
+	RequestCount       int64     `json:"-"`
 }
 
 type SummaryAllResult struct {