Replace average scaling ratio with min scaling ratio for perf aggregated

Signed-off-by: Katarzyna Kujawa <katarzyna.kujawa@intel.com>
google · Jul 9, 2020 · 821ed4b · 821ed4b
1 parent d370486
commit 821ed4b
Show file tree

Hide file tree

Showing 4 changed files with 47 additions and 18 deletions.
diff --git a/docs/runtime_options.md b/docs/runtime_options.md
@@ -144,7 +144,7 @@ Core perf events can be exposed on Prometheus endpoint per CPU or aggregated by
 - `--disable_metrics="percpu"` - core perf events are aggregated
 - `--disable_metrics=""` - core perf events are exposed per CPU.
 
-Aggregated form of core perf events significantly decrease volume of data. For aggregated form of core perf events scaling ratio (`container_perf_metric_scaling ratio`) indicates average scaling ratio for specific event.
+Aggregated form of core perf events significantly decrease volume of data. For aggregated form of core perf events scaling ratio (`container_perf_metric_scaling ratio`) indicates the lowest value of scaling ratio for specific event to show the worst precision.
 
 ### Perf subsystem introduction
 

diff --git a/metrics/prometheus.go b/metrics/prometheus.go
@@ -1586,7 +1586,7 @@ func NewPrometheusCollector(i infoProvider, f ContainerLabelsFunc, includedMetri
 					valueType:   prometheus.GaugeValue,
 					extraLabels: []string{"cpu", "event"},
 					getValues: func(s *info.ContainerStats) metricValues {
-						return getAvgCoreScalingRatio(s)
+						return getMinCoreScalingRatio(s)
 					},
 				}}...)
 		}
@@ -1954,22 +1954,23 @@ func getAggregatedCorePerfEvents(s *info.ContainerStats) metricValues {
 	return values
 }
 
-func getAvgCoreScalingRatio(s *info.ContainerStats) metricValues {
+func getMinCoreScalingRatio(s *info.ContainerStats) metricValues {
 	values := make(metricValues, 0)
-	perfEventStatAgg := make(map[string][]float64)
-	// collect scaling ratios for event
+	perfEventStatMin := make(map[string]float64)
+	// search for minimal value of scalin ratio for specific event
 	for _, perfStat := range s.PerfStats {
-		perfEventStatAgg[perfStat.Name] = append(perfEventStatAgg[perfStat.Name], perfStat.ScalingRatio)
-	}
-	// calculate average scaling ratio
-	for perfEvent, perfScalingRatio := range perfEventStatAgg {
-		sumScalingRatio := 0.0
-		for _, scalingRatio := range perfScalingRatio {
-			sumScalingRatio += scalingRatio
+		if _, ok := perfEventStatMin[perfStat.Name]; !ok {
+			// found a new event
+			perfEventStatMin[perfStat.Name] = perfStat.ScalingRatio
+		} else if perfStat.ScalingRatio < perfEventStatMin[perfStat.Name] {
+			// found a lower value of scaling ration so replace the minimal value
+			perfEventStatMin[perfStat.Name] = perfStat.ScalingRatio
 		}
+	}
 
+	for perfEvent, perfScalingRatio := range perfEventStatMin {
 		values = append(values, metricValue{
-			value:     sumScalingRatio / float64(len(perfScalingRatio)),
+			value:     perfScalingRatio,
 			labels:    []string{"", perfEvent},
 			timestamp: s.Timestamp,
 		})

diff --git a/metrics/prometheus_test.go b/metrics/prometheus_test.go
@@ -170,6 +170,14 @@ func TestGetPerCpuCorePerfEvents(t *testing.T) {
 	}
 	metricVals := getPerCPUCorePerfEvents(containerStats)
 	assert.Equal(t, 4, len(metricVals))
+	values := []float64{}
+	for _, metric := range metricVals {
+		values = append(values, metric.value)
+	}
+	assert.Contains(t, values, 123.0)
+	assert.Contains(t, values, 456.0)
+	assert.Contains(t, values, 321.0)
+	assert.Contains(t, values, 789.0)
 }
 
 func TestGetPerCpuCoreScalingRatio(t *testing.T) {
@@ -204,9 +212,17 @@ func TestGetPerCpuCoreScalingRatio(t *testing.T) {
 	}
 	metricVals := getPerCPUCoreScalingRatio(containerStats)
 	assert.Equal(t, 4, len(metricVals))
+	values := []float64{}
+	for _, metric := range metricVals {
+		values = append(values, metric.value)
+	}
+	assert.Contains(t, values, 1.0)
+	assert.Contains(t, values, 0.5)
+	assert.Contains(t, values, 0.7)
+	assert.Contains(t, values, 0.3)
 }
 
-func TestGetCorePerfEventsAggregated(t *testing.T) {
+func TestGetAggCorePerfEvents(t *testing.T) {
 	containerStats := &info.ContainerStats{
 		Timestamp: time.Unix(1395066367, 0),
 		PerfStats: []info.PerfStat{
@@ -238,9 +254,15 @@ func TestGetCorePerfEventsAggregated(t *testing.T) {
 	}
 	metricVals := getAggregatedCorePerfEvents(containerStats)
 	assert.Equal(t, 2, len(metricVals))
+	values := []float64{}
+	for _, metric := range metricVals {
+		values = append(values, metric.value)
+	}
+	assert.Contains(t, values, 579.0)
+	assert.Contains(t, values, 1110.0)
 }
 
-func TestGetCoreScalingRatioAverage(t *testing.T) {
+func TestGetMinCoreScalingRatio(t *testing.T) {
 	containerStats := &info.ContainerStats{
 		Timestamp: time.Unix(1395066367, 0),
 		PerfStats: []info.PerfStat{
@@ -270,6 +292,12 @@ func TestGetCoreScalingRatioAverage(t *testing.T) {
 			},
 		},
 	}
-	metricVals := getAvgCoreScalingRatio(containerStats)
+	metricVals := getMinCoreScalingRatio(containerStats)
 	assert.Equal(t, 2, len(metricVals))
+	values := []float64{}
+	for _, metric := range metricVals {
+		values = append(values, metric.value)
+	}
+	assert.Contains(t, values, 0.5)
+	assert.Contains(t, values, 0.3)
 }
diff --git a/metrics/testdata/prometheus_metrics_perf_aggregated b/metrics/testdata/prometheus_metrics_perf_aggregated
@@ -6,8 +6,8 @@ cadvisor_version_info{cadvisorRevision="abcdef",cadvisorVersion="0.16.0",dockerV
 container_last_seen{container_env_foo_env="prod",container_label_foo_label="bar",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 1.395066363e+09 1395066363000
 # HELP container_perf_events_scaling_ratio Perf event metric scaling ratio.
 # TYPE container_perf_events_scaling_ratio gauge
-container_perf_events_scaling_ratio{container_env_foo_env="prod",container_label_foo_label="bar",cpu="",event="instructions",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 0.75 1395066363000
-container_perf_events_scaling_ratio{container_env_foo_env="prod",container_label_foo_label="bar",cpu="",event="instructions_retired",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 0.499999999995 1395066363000
+container_perf_events_scaling_ratio{container_env_foo_env="prod",container_label_foo_label="bar",cpu="",event="instructions",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 0.5 1395066363000
+container_perf_events_scaling_ratio{container_env_foo_env="prod",container_label_foo_label="bar",cpu="",event="instructions_retired",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 0.33333333333 1395066363000
 # HELP container_perf_events_total Perf event metric.
 # TYPE container_perf_events_total counter
 container_perf_events_total{container_env_foo_env="prod",container_label_foo_label="bar",cpu="",event="instructions",id="testcontainer",image="test",name="testcontaineralias",zone_name="hello"} 579 1395066363000