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

model/perf_metric.go

@@ -68,6 +68,16 @@ type PerfMetricSummary struct {
 	GenerationMs   int64  `json:"generation_ms"`
 }
 
+type PerfMetricSummaryBucket struct {
+	ModelName      string `json:"model_name"`
+	BucketTs       int64  `json:"bucket_ts"`
+	RequestCount   int64  `json:"request_count"`
+	SuccessCount   int64  `json:"success_count"`
+	TotalLatencyMs int64  `json:"total_latency_ms"`
+	OutputTokens   int64  `json:"output_tokens"`
+	GenerationMs   int64  `json:"generation_ms"`
+}
+
 func GetPerfMetricsSummaryAll(startTs int64, endTs int64, groups []string) ([]PerfMetricSummary, error) {
 	var summaries []PerfMetricSummary
 	query := DB.Model(&PerfMetric{}).
@@ -86,6 +96,25 @@ func GetPerfMetricsSummaryAll(startTs int64, endTs int64, groups []string) ([]Pe
 	return summaries, err
 }
 
+func GetPerfMetricsSummaryBucketsAll(startTs int64, endTs int64, groups []string) ([]PerfMetricSummaryBucket, error) {
+	var summaries []PerfMetricSummaryBucket
+	query := DB.Model(&PerfMetric{}).
+		Select("model_name, bucket_ts, SUM(request_count) as request_count, SUM(success_count) as success_count, SUM(total_latency_ms) as total_latency_ms, SUM(output_tokens) as output_tokens, SUM(generation_ms) as generation_ms").
+		Where("bucket_ts >= ? AND bucket_ts <= ?", startTs, endTs)
+	if groups != nil {
+		if len(groups) == 0 {
+			return summaries, nil
+		}
+		query = query.Where(commonGroupCol+" IN ?", groups)
+	}
+	err := query.
+		Group("model_name, bucket_ts").
+		Having("SUM(request_count) > 0").
+		Order("bucket_ts ASC").
+		Find(&summaries).Error
+	return summaries, err
+}
+
 func DeletePerfMetricsBefore(cutoffTs int64) error {
 	if cutoffTs <= 0 {
 		return nil

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
 	})
 
@@ -189,6 +187,7 @@ func QuerySummaryAll(hours int, groups []string) (SummaryAllResult, error) {
 			AvgLatencyMs:       avgLatency,
 			SuccessRate:        math.Round(successRate*100) / 100,
 			AvgTps:             math.Round(avgTps*100) / 100,
+			RecentSuccessRates: recentSuccessRates(modelBuckets[name], 3),
 			RequestCount:       total.requestCount,
 		})
 	}
@@ -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

@@ -52,6 +52,7 @@ type ModelSummary struct {
 	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:"-"`
 }
 

web/default/src/features/performance-metrics/types.ts

@@ -48,6 +48,7 @@ export type PerfModelSummary = {
   avg_latency_ms: number
   success_rate: number
   avg_tps: number
+  recent_success_rates?: number[]
   request_count?: number
 }
 

web/default/src/features/pricing/components/model-perf-badge.tsx

@@ -29,6 +29,7 @@ export type ModelPerfBadgeData = {
   avg_latency_ms: number
   success_rate: number
   avg_tps: number
+  recent_success_rates?: number[]
 }
 
 export interface ModelPerfBadgeProps extends React.HTMLAttributes<HTMLDivElement> {
@@ -50,7 +51,15 @@ export const ModelPerfBadge = memo(function ModelPerfBadge(
 
   const { avg_latency_ms, avg_tps, success_rate } = props.perf
 
-  const statusColor = getSuccessRateDotClass(success_rate)
+  const recentRates =
+    props.perf.recent_success_rates?.filter((rate) => Number.isFinite(rate)) ??
+    []
+  const statusRates =
+    recentRates.length > 0 ? recentRates.slice(-3) : [success_rate]
+  const statusBars = [
+    ...Array(Math.max(0, 3 - statusRates.length)).fill(null),
+    ...statusRates,
+  ].slice(-3)
 
   return (
     <div
@@ -83,9 +92,22 @@ export const ModelPerfBadge = memo(function ModelPerfBadge(
           {t('Status short')}
         </div>
         <div className='flex h-4 items-center justify-end gap-0.5'>
-          <span className='bg-muted-foreground/10 h-2 w-1 rounded-full' />
-          <span className='bg-muted-foreground/15 h-2.5 w-1 rounded-full' />
-          <span className={cn('h-3 w-1 rounded-full', statusColor)} />
+          {statusBars.map((rate, index) => (
+            <span
+              key={`${index}-${rate ?? 'empty'}`}
+              className={cn(
+                'w-1 rounded-full',
+                index === 0 && 'h-2',
+                index === 1 && 'h-2.5',
+                index === 2 && 'h-3',
+                rate == null
+                  ? index === 0
+                    ? 'bg-muted-foreground/10'
+                    : 'bg-muted-foreground/15'
+                  : getSuccessRateDotClass(rate)
+              )}
+            />
+          ))}
         </div>
       </div>
     </div>