Add metric 'nv_util_eff' like nvtop

collectors/cpustatMetric.go

@@ -27,7 +27,6 @@ const CPUSTATFILE = `/proc/stat`
 
 type CpustatCollectorConfig struct {
 	ExcludeMetrics []string `json:"exclude_metrics,omitempty"`
-	excludeNumCPUs bool
 }
 
 type CpustatCollector struct {
@@ -80,7 +79,6 @@ func (m *CpustatCollector) Init(config json.RawMessage) error {
 			m.matches[match] = index
 		}
 	}
-	m.config.excludeNumCPUs = slices.Contains(m.config.ExcludeMetrics, "num_cpus")
 
 	// Check input file
 	file, err := os.Open(CPUSTATFILE)
@@ -97,13 +95,11 @@ func (m *CpustatCollector) Init(config json.RawMessage) error {
 		line := scanner.Text()
 		linefields := strings.Fields(line)
 		if strings.Compare(linefields[0], "cpu") == 0 {
-			// Kernel system statistics for all CPUs
 			m.olddata["cpu"] = make(map[string]int64)
 			for k, v := range m.matches {
 				m.olddata["cpu"][k], _ = strconv.ParseInt(linefields[v], 0, 64)
 			}
 		} else if strings.HasPrefix(linefields[0], "cpu") && strings.Compare(linefields[0], "cpu") != 0 {
-			// Kernel system statistics per CPU
 			cpustr := strings.TrimLeft(linefields[0], "cpu")
 			cpu, _ := strconv.Atoi(cpustr)
 			m.cputags[linefields[0]] = map[string]string{
@@ -195,11 +191,10 @@ func (m *CpustatCollector) Read(interval time.Duration, output chan lp.CCMessage
 		}
 	}
 
-	if !m.config.excludeNumCPUs {
-		if num_cpus_metric, err := lp.NewMetric("num_cpus", m.nodetags, m.meta, num_cpus, now); err == nil {
+	num_cpus_metric, err := lp.NewMetric("num_cpus", m.nodetags, m.meta, num_cpus, now)
+	if err == nil {
 		output <- num_cpus_metric
 	}
-	}
 
 	m.lastTimestamp = now
 }

collectors/memstatMetric.go

@@ -72,8 +72,7 @@ func getStats(filename string) map[string]MemstatStats {
 	for scanner.Scan() {
 		line := scanner.Text()
 		linefields := strings.Fields(line)
-		switch len(linefields) {
-		case 3:
+		if len(linefields) == 3 {
 			v, err := strconv.ParseFloat(linefields[1], 64)
 			if err == nil {
 				stats[strings.Trim(linefields[0], ":")] = MemstatStats{
@@ -81,7 +80,7 @@ func getStats(filename string) map[string]MemstatStats {
 					unit:  linefields[2],
 				}
 			}
-		case 5:
+		} else if len(linefields) == 5 {
 			v, err := strconv.ParseFloat(linefields[3], 64)
 			if err == nil {
 				cclog.ComponentDebug("getStats", strings.Trim(linefields[2], ":"), v, linefields[4])
@@ -107,10 +106,7 @@ func (m *MemstatCollector) Init(config json.RawMessage) error {
 			return fmt.Errorf("%s Init(): Error decoding JSON config: %w", m.name, err)
 		}
 	}
-	m.meta = map[string]string{
-		"source": m.name,
-		"group":  "Memory",
-	}
+	m.meta = map[string]string{"source": m.name, "group": "Memory"}
 	m.stats = make(map[string]int64)
 	m.matches = make(map[string]string)
 	m.tags = map[string]string{"type": "node"}
@@ -149,7 +145,7 @@ func (m *MemstatCollector) Init(config json.RawMessage) error {
 		"KernelStack":     "mem_kernelstack",
 	}
 	for k, v := range matches {
-		if !slices.Contains(m.config.ExcludeMetrics, v) {
+		if !slices.Contains(m.config.ExcludeMetrics, k) {
 			m.matches[k] = v
 		}
 	}
@@ -157,7 +153,7 @@ func (m *MemstatCollector) Init(config json.RawMessage) error {
 	if !slices.Contains(m.config.ExcludeMetrics, "mem_used") {
 		m.sendMemUsed = true
 	}
-	if len(m.matches) == 0 && !m.sendMemUsed {
+	if len(m.matches) == 0 {
 		return fmt.Errorf("%s Init(): no metrics to collect", m.name)
 	}
 	if err := m.setup(); err != nil {

collectors/nvidiaMetric.go

@@ -1115,6 +1115,31 @@ func readNVLinkStats(device *NvidiaCollectorDevice, output chan lp.CCMessage) er
 	return nil
 }
 
+func readEfficiency(device *NvidiaCollectorDevice, output chan lp.CCMessage) error {
+	if !device.excludeMetrics["nv_util_eff"] {
+		maxPower, ret := nvml.DeviceGetEnforcedPowerLimit(device.device)
+		if ret == nvml.SUCCESS {
+			curPower, ret := nvml.DeviceGetPowerUsage(device.device)
+			if ret == nvml.SUCCESS {
+				util, ret := nvml.DeviceGetUtilizationRates(device.device)
+				if ret == nvml.SUCCESS {
+					factor := float64(curPower) / float64(maxPower)
+					eff := uint32(float64(util.Gpu) * factor)
+					if eff > 100 {
+						eff = 100
+					}
+					y, err := lp.NewMetric("nv_util_eff", device.tags, device.meta, eff, time.Now())
+					if err == nil {
+						y.AddTag("unit", "percent")
+						output <- y
+					}
+				}
+			}
+		}
+	}
+	return nil
+}
+
 func (m *NvidiaCollector) Read(interval time.Duration, output chan lp.CCMessage) {
 	var err error
 	if !m.init {
@@ -1220,6 +1245,11 @@ func (m *NvidiaCollector) Read(interval time.Duration, output chan lp.CCMessage)
 		if err != nil {
 			cclog.ComponentDebug(m.name, "readNVLinkStats for device", name, "failed")
 		}
+
+		err = readEfficiency(device, output)
+		if err != nil {
+			cclog.ComponentDebug(m.name, "readEfficiency for device", name, "failed")
+		}
 	}
 
 	// Actual read loop over all attached Nvidia GPUs

collectors/nvidiaMetric.md

@@ -85,5 +85,6 @@ Metrics:
 * `nv_energy`
 * `nv_energy_abs`
 * `nv_average_power`
+* `nv_util_eff` (`nv_util` * (`nv_power_usage` / `nv_power_max_limit`))
 
 Some metrics add the additional sub type tag (`stype`) like the `nv_nvlink_*` metrics set `stype=nvlink,stype-id=<link_number>`.