support xpu-smi for intel stats (#755)

- Add skipCollection propery
- Ensure that sensors are initialized as an empty map

beszel/internal/agent/gpu.go

@@ -4,6 +4,7 @@ import (
 	"beszel/internal/entities/system"
 	"bufio"
 	"bytes"
+	"encoding/csv"
 	"encoding/json"
 	"fmt"
 	"os/exec"
@@ -21,11 +22,13 @@ const (
 	nvidiaSmiCmd  = "nvidia-smi"
 	rocmSmiCmd    = "rocm-smi"
 	tegraStatsCmd = "tegrastats"
+	xpuSmiCmd     = "xpu-smi"
 
 	// Polling intervals
 	nvidiaSmiInterval  = "4"    // in seconds
 	tegraStatsInterval = "3700" // in milliseconds
 	rocmSmiInterval    = 4300 * time.Millisecond
+	xpuSmiInterval     = 4
 
 	// Command retry and timeout constants
 	retryWaitTime     = 5 * time.Second
@@ -41,10 +44,11 @@ const (
 // GPUManager manages data collection for GPUs (either Nvidia or AMD)
 type GPUManager struct {
 	sync.Mutex
-	nvidiaSmi  bool
-	rocmSmi    bool
-	tegrastats bool
-	GpuDataMap map[string]*system.GPUData
+	nvidiaSmi   bool
+	rocmSmi     bool
+	tegrastats  bool
+	intelXpuSmi bool
+	GpuDataMap  map[string]*system.GPUData
 }
 
 // RocmSmiJson represents the JSON structure of rocm-smi output
@@ -160,6 +164,59 @@ func (gm *GPUManager) getJetsonParser() func(output []byte) bool {
 	}
 }
 
+func (gm *GPUManager) parseIntelData(output []byte) bool {
+	gm.Lock()
+	defer gm.Unlock()
+	reader := csv.NewReader(bytes.NewReader(output))
+	records, err := reader.ReadAll()
+	if err != nil {
+		slog.Warn("Failed to parse Intel GPU data", "err", err)
+		return false
+	}
+
+	header := []string{"Timestamp", "DeviceId", "GPU Power (W)", "GPU Frequency (MHz)", "GPU Memory Utilization (%)", "GPU Memory Used (MiB)"}
+	gpuData := &system.GPUData{Name: "GPU"}
+	gm.GpuDataMap["0"] = gpuData
+
+	for _, record := range records {
+		if strings.Join(record, ",") == strings.Join(header, ",") {
+			slog.Debug("Skipping header", "header", record)
+			continue
+		}
+		var memoryUtilization *float64
+		var memoryUsed *float64
+		for i, field := range header {
+			if field == "Timestamp" {
+				continue
+			}
+			stripped := strings.TrimSpace(record[i])
+			value, err := strconv.ParseFloat(stripped, 64)
+			if err != nil {
+				slog.Warn("Failed to parse field", "field", field, "value", stripped, "err", err)
+				continue
+			}
+
+			switch field {
+			case "GPU Power (W)":
+				gpuData.Power += value
+			case "GPU Frequency (MHz)":
+				gpuData.Usage += value
+			case "GPU Memory Utilization (%)":
+				memoryUtilization = &value
+			case "GPU Memory Used (MiB)":
+				memoryUsed = &value
+			}
+		}
+		if memoryUtilization != nil && memoryUsed != nil {
+			gpuData.MemoryUsed = *memoryUsed
+			gpuData.MemoryTotal = (*memoryUsed / *memoryUtilization) * 100 // convert to total memory
+		}
+	}
+	gpuData.Count++
+
+	return true
+}
+
 // parseNvidiaData parses the output of nvidia-smi and updates the GPUData map
 func (gm *GPUManager) parseNvidiaData(output []byte) bool {
 	gm.Lock()
@@ -278,10 +335,14 @@ func (gm *GPUManager) detectGPUs() error {
 		gm.tegrastats = true
 		gm.nvidiaSmi = false
 	}
-	if gm.nvidiaSmi || gm.rocmSmi || gm.tegrastats {
+	fmt.Println("Looking for gpus")
+	if _, err := exec.LookPath(xpuSmiCmd); err == nil {
+		gm.intelXpuSmi = true
+	}
+	if gm.nvidiaSmi || gm.rocmSmi || gm.tegrastats || gm.intelXpuSmi {
 		return nil
 	}
-	return fmt.Errorf("no GPU found - install nvidia-smi, rocm-smi, or tegrastats")
+	return fmt.Errorf("no GPU found - install nvidia-smi, rocm-smi, intel_gpu_top, or tegrastats")
 }
 
 // startCollector starts the appropriate GPU data collector based on the command
@@ -318,6 +379,10 @@ func (gm *GPUManager) startCollector(command string) {
 				time.Sleep(rocmSmiInterval)
 			}
 		}()
+	case xpuSmiCmd:
+		collector.cmdArgs = []string{"dump", "-d", "-1", "-m", "1,2,5,18", "-i", strconv.Itoa(xpuSmiInterval)}
+		collector.parse = gm.parseIntelData
+		go collector.start()
 	}
 }
 
@@ -338,6 +403,9 @@ func NewGPUManager() (*GPUManager, error) {
 	if gm.tegrastats {
 		gm.startCollector(tegraStatsCmd)
 	}
+	if gm.intelXpuSmi {
+		gm.startCollector(xpuSmiCmd)
+	}
 
 	return &gm, nil
 }

beszel/internal/agent/sensors.go

@@ -13,26 +13,31 @@ import (
 )
 
 type SensorConfig struct {
-	context       context.Context
-	sensors       map[string]struct{}
-	primarySensor string
-	isBlacklist   bool
-	hasWildcards  bool
+	context        context.Context
+	sensors        map[string]struct{}
+	primarySensor  string
+	isBlacklist    bool
+	hasWildcards   bool
+	skipCollection bool
 }
 
 func (a *Agent) newSensorConfig() *SensorConfig {
 	primarySensor, _ := GetEnv("PRIMARY_SENSOR")
 	sysSensors, _ := GetEnv("SYS_SENSORS")
-	sensors, _ := GetEnv("SENSORS")
+	sensorsEnvVal, sensorsSet := GetEnv("SENSORS")
+	skipCollection := sensorsSet && sensorsEnvVal == ""
 
-	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensors)
+	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, skipCollection)
 }
 
 // newSensorConfigWithEnv creates a SensorConfig with the provided environment variables
-func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensors string) *SensorConfig {
+// sensorsSet indicates if the SENSORS environment variable was explicitly set (even to empty string)
+func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal string, skipCollection bool) *SensorConfig {
 	config := &SensorConfig{
-		context:       context.Background(),
-		primarySensor: primarySensor,
+		context:        context.Background(),
+		primarySensor:  primarySensor,
+		skipCollection: skipCollection,
+		sensors:        make(map[string]struct{}),
 	}
 
 	// Set sensors context (allows overriding sys location for sensors)
@@ -43,22 +48,18 @@ func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensors string
 		)
 	}
 
-	// Set sensors whitelist
-	if sensors != "" {
-		// handle blacklist
-		if strings.HasPrefix(sensors, "-") {
-			config.isBlacklist = true
-			sensors = sensors[1:]
-		}
+	// handle blacklist
+	if strings.HasPrefix(sensorsEnvVal, "-") {
+		config.isBlacklist = true
+		sensorsEnvVal = sensorsEnvVal[1:]
+	}
 
-		config.sensors = make(map[string]struct{})
-		for sensor := range strings.SplitSeq(sensors, ",") {
-			sensor = strings.TrimSpace(sensor)
-			if sensor != "" {
-				config.sensors[sensor] = struct{}{}
-				if strings.Contains(sensor, "*") {
-					config.hasWildcards = true
-				}
+	for sensor := range strings.SplitSeq(sensorsEnvVal, ",") {
+		sensor = strings.TrimSpace(sensor)
+		if sensor != "" {
+			config.sensors[sensor] = struct{}{}
+			if strings.Contains(sensor, "*") {
+				config.hasWildcards = true
 			}
 		}
 	}
@@ -69,7 +70,7 @@ func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensors string
 // updateTemperatures updates the agent with the latest sensor temperatures
 func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 	// skip if sensors whitelist is set to empty string
-	if a.sensorConfig.sensors != nil && len(a.sensorConfig.sensors) == 0 {
+	if a.sensorConfig.skipCollection {
 		slog.Debug("Skipping temperature collection")
 		return
 	}
@@ -113,8 +114,8 @@ func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 
 // isValidSensor checks if a sensor is valid based on the sensor name and the sensor config
 func isValidSensor(sensorName string, config *SensorConfig) bool {
-	// If no sensors configuration, everything is valid
-	if config.sensors == nil {
+	// if no sensors configured, everything is valid
+	if len(config.sensors) == 0 {
 		return true
 	}
 
@@ -123,7 +124,7 @@ func isValidSensor(sensorName string, config *SensorConfig) bool {
 		return !config.isBlacklist
 	}
 
-	// If no wildcards, return false if blacklist, true if whitelist
+	// If no wildcards, return true if blacklist, false if whitelist
 	if !config.hasWildcards {
 		return config.isBlacklist
 	}

beszel/internal/agent/sensors_test.go

@@ -97,10 +97,13 @@ func TestIsValidSensor(t *testing.T) {
 			expectedValid: true,
 		},
 		{
-			name:       "Nil sensor config",
+			name:       "No sensors configured",
 			sensorName: "any_temp",
 			config: &SensorConfig{
-				sensors: nil,
+				sensors:        map[string]struct{}{},
+				isBlacklist:    false,
+				hasWildcards:   false,
+				skipCollection: false,
 			},
 			expectedValid: true,
 		},
@@ -162,6 +165,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 		primarySensor  string
 		sysSensors     string
 		sensors        string
+		skipCollection bool
 		expectedConfig *SensorConfig
 	}{
 		{
@@ -170,22 +174,38 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			sysSensors:    "",
 			sensors:       "",
 			expectedConfig: &SensorConfig{
-				context:       context.Background(),
-				primarySensor: "",
-				sensors:       nil,
-				isBlacklist:   false,
-				hasWildcards:  false,
+				context:        context.Background(),
+				primarySensor:  "",
+				sensors:        map[string]struct{}{},
+				isBlacklist:    false,
+				hasWildcards:   false,
+				skipCollection: false,
 			},
 		},
 		{
-			name:          "Primary sensor only",
+			name:           "Explicitly set to empty string",
+			primarySensor:  "",
+			sysSensors:     "",
+			sensors:        "",
+			skipCollection: true,
+			expectedConfig: &SensorConfig{
+				context:        context.Background(),
+				primarySensor:  "",
+				sensors:        map[string]struct{}{},
+				isBlacklist:    false,
+				hasWildcards:   false,
+				skipCollection: true,
+			},
+		},
+		{
+			name:          "Primary sensor only - should create sensor map",
 			primarySensor: "cpu_temp",
 			sysSensors:    "",
 			sensors:       "",
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
-				sensors:       nil,
+				sensors:       map[string]struct{}{},
 				isBlacklist:   false,
 				hasWildcards:  false,
 			},
@@ -238,6 +258,22 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 				hasWildcards: true,
 			},
 		},
+		{
+			name:          "Sensors with whitespace",
+			primarySensor: "cpu_temp",
+			sysSensors:    "",
+			sensors:       "cpu_*, gpu_temp",
+			expectedConfig: &SensorConfig{
+				context:       context.Background(),
+				primarySensor: "cpu_temp",
+				sensors: map[string]struct{}{
+					"cpu_*":    {},
+					"gpu_temp": {},
+				},
+				isBlacklist:  false,
+				hasWildcards: true,
+			},
+		},
 		{
 			name:          "With SYS_SENSORS path",
 			primarySensor: "cpu_temp",
@@ -256,7 +292,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors)
+			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.skipCollection)
 
 			// Check primary sensor
 			assert.Equal(t, tt.expectedConfig.primarySensor, result.primarySensor)

beszel/internal/agent/system.go

@@ -199,16 +199,24 @@ func (a *Agent) getSystemStats() system.Stats {
 			if systemStats.Temperatures == nil {
 				systemStats.Temperatures = make(map[string]float64, len(gpuData))
 			}
+			highestTemp := 0.0
 			for _, gpu := range gpuData {
 				if gpu.Temperature > 0 {
 					systemStats.Temperatures[gpu.Name] = gpu.Temperature
 					if a.sensorConfig.primarySensor == gpu.Name {
 						a.systemInfo.DashboardTemp = gpu.Temperature
 					}
+					if gpu.Temperature > highestTemp {
+						highestTemp = gpu.Temperature
+					}
 				}
 				// update high gpu percent for dashboard
 				a.systemInfo.GpuPct = max(a.systemInfo.GpuPct, gpu.Usage)
 			}
+			// use highest temp for dashboard temp if dashboard temp is unset
+			if a.systemInfo.DashboardTemp == 0 {
+				a.systemInfo.DashboardTemp = highestTemp
+			}
 		}
 	}
 

supplemental/scripts/install-agent-brew.sh

@@ -67,7 +67,7 @@ fi
 mkdir -p ~/.config/beszel ~/.cache/beszel
 
 echo "KEY=\"$KEY\"" >~/.config/beszel/beszel-agent.env
-echo "PORT=$PORT" >>~/.config/beszel/beszel-agent.env
+echo "LISTEN=$PORT" >>~/.config/beszel/beszel-agent.env
 
 brew tap henrygd/beszel
 brew install beszel-agent