feat/fix: add linear fan speed control based on temperature

some smaller fixes

Signed-off-by: Matthias Riegler <matthias.riegler@ankorstore.com>

Makefile

@@ -9,6 +9,10 @@ run:
 lint:
 	golangci-lint run
 
+.PHONY: test
+test:
+	go test ./... -v
+
 .PHONY: generate
 generate: buf
 	$(BUF) generate
@@ -16,6 +20,9 @@ generate: buf
 release:
 	goreleaser release --clean
 
+snapshot:
+	goreleaser release --snapshot --skip-publish --clean
+
 # Dependencies
 LOCALBIN ?= $(shell pwd)/bin
 $(LOCALBIN):

cmd/agent/main.go

@@ -5,6 +5,8 @@ import (
 	"fmt"
 	"net"
 	"net/http"
+	"net/http/pprof"
+	_ "net/http/pprof"
 	"os"
 	"os/signal"
 	"sync"
@@ -14,6 +16,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus/promhttp"
 	bladeapiv1alpha1 "github.com/xvzf/computeblade-agent/api/bladeapi/v1alpha1"
 	"github.com/xvzf/computeblade-agent/internal/agent"
+	"github.com/xvzf/computeblade-agent/pkg/fancontroller"
 	"github.com/xvzf/computeblade-agent/pkg/ledengine"
 	"github.com/xvzf/computeblade-agent/pkg/log"
 	"go.uber.org/zap"
@@ -24,7 +27,6 @@ func main() {
 	var wg sync.WaitGroup
 
 	// setup logger
-
 	zapLogger := zap.Must(zap.NewDevelopment()).With(zap.String("app", "computeblade-agent"))
 	_ = zap.ReplaceGlobals(zapLogger.With(zap.String("scope", "global")))
 	baseCtx := log.IntoContext(context.Background(), zapLogger)
@@ -33,11 +35,17 @@ func main() {
 	defer cancelCtx(context.Canceled)
 
 	computebladeAgent, err := agent.NewComputeBladeAgent(agent.ComputeBladeAgentConfig{
-		IdleLedColor:        ledengine.LedColorGreen(0.05),
-		IdentifyLedColor:    ledengine.LedColorPurple(0.05),
-		CriticalLedColor:    ledengine.LedColorRed(0.3),
-		StealthModeEnabled:  false,
-		DefaultFanSpeed:     40,
+		IdleLedColor:       ledengine.LedColorGreen(0.05),
+		IdentifyLedColor:   ledengine.LedColorPurple(0.05),
+		CriticalLedColor:   ledengine.LedColorRed(0.3),
+		StealthModeEnabled: false,
+		FanControllerConfig: fancontroller.FanControllerConfig{
+			Steps: []fancontroller.FanControllerStep{
+				{Temperature: 40, Speed: 40},
+				{Temperature: 55, Speed: 80},
+			},
+		},
+		FanUpdateInterval:   5 * time.Second,
 		CriticalTemperature: 60,
 	})
 	if err != nil {
@@ -100,15 +108,20 @@ func main() {
 	}()
 
 	// setup prometheus endpoint
-	promHandler := http.NewServeMux()
-	promHandler.Handle("/metrics", promhttp.Handler())
-	server := &http.Server{Addr: ":9666", Handler: promHandler}
+	instrumentationHandler := http.NewServeMux()
+	instrumentationHandler.Handle("/metrics", promhttp.Handler())
+	instrumentationHandler.HandleFunc("/debug/pprof/", pprof.Index)
+	instrumentationHandler.HandleFunc("/debug/pprof/cmdline", pprof.Cmdline)
+	instrumentationHandler.HandleFunc("/debug/pprof/profile", pprof.Profile)
+	instrumentationHandler.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
+	instrumentationHandler.HandleFunc("/debug/pprof/trace", pprof.Trace)
+	server := &http.Server{Addr: ":9666", Handler: instrumentationHandler}
 	wg.Add(1)
 	go func() {
 		defer wg.Done()
 		err := server.ListenAndServe()
 		if err != nil && err != http.ErrServerClosed {
-			log.FromContext(ctx).Error("Failed to start prometheus server", zap.Error(err))
+			log.FromContext(ctx).Error("Failed to start prometheus/pprof server", zap.Error(err))
 			cancelCtx(err)
 		}
 	}()
@@ -120,7 +133,7 @@ func main() {
 		defer cancel()
 		err := server.Shutdown(shutdownCtx)
 		if err != nil {
-			log.FromContext(ctx).Error("Failed to shutdown prometheus server", zap.Error(err))
+			log.FromContext(ctx).Error("Failed to shutdown prometheus/pprof server", zap.Error(err))
 		}
 	}()
 

internal/agent/agent.go

@@ -4,9 +4,11 @@ import (
 	"context"
 	"errors"
 	"sync"
+	"time"
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
+	"github.com/xvzf/computeblade-agent/pkg/fancontroller"
 	"github.com/xvzf/computeblade-agent/pkg/hal"
 	"github.com/xvzf/computeblade-agent/pkg/ledengine"
 	"github.com/xvzf/computeblade-agent/pkg/log"
@@ -60,7 +62,6 @@ func (e Event) String() string {
 }
 
 type ComputeBladeAgentConfig struct {
-
 	// IdleLedColor is the color of the edge LED when the blade is idle mode
 	IdleLedColor hal.LedColor
 	// IdentifyLedColor is the color of the edge LED when the blade is in identify mode
@@ -72,11 +73,11 @@ type ComputeBladeAgentConfig struct {
 	// StealthModeEnabled indicates whether stealth mode is enabled
 	StealthModeEnabled bool
 
-	// DefaultFanSpeed is the default fan speed in percent. Usually 40% is sufficient
-	DefaultFanSpeed uint
-
 	// Critical temperature of the compute blade (used to trigger critical mode)
 	CriticalTemperature uint
+
+	FanControllerConfig fancontroller.FanControllerConfig
+	FanUpdateInterval   time.Duration
 }
 
 // ComputeBladeAgent implements the core-logic of the agent. It is responsible for handling events and interfacing with the hardware.
@@ -102,6 +103,8 @@ type computeBladeAgentImpl struct {
 	edgeLedEngine ledengine.LedEngine
 	topLedEngine  ledengine.LedEngine
 
+	fanController fancontroller.FanController
+
 	eventChan chan Event
 }
 
@@ -132,13 +135,22 @@ func NewComputeBladeAgent(opts ComputeBladeAgentConfig) (ComputeBladeAgent, erro
 		return nil, err
 	}
 
+	fanController, err := fancontroller.NewLinearFanController(opts.FanControllerConfig)
+	if err != nil {
+		return nil, err
+	}
+
 	return &computeBladeAgentImpl{
 		opts:          opts,
 		blade:         blade,
 		edgeLedEngine: edgeLedEngine,
 		topLedEngine:  topLedEngine,
+		fanController: fanController,
 		state:         NewComputeBladeState(),
-		eventChan:     make(chan Event, 10), // backlog of 10 events. They should process fast but we e.g. don't want to miss button presses
+		eventChan: make(
+			chan Event,
+			10,
+		), // backlog of 10 events. They should process fast but we e.g. don't want to miss button presses
 	}, nil
 }
 
@@ -153,9 +165,6 @@ func (a *computeBladeAgentImpl) Run(origCtx context.Context) error {
 	a.state.RegisterEvent(NoopEvent)
 
 	// Set defaults
-	if err := a.blade.SetFanSpeed(uint8(a.opts.DefaultFanSpeed)); err != nil {
-		return err
-	}
 	if err := a.blade.SetStealthMode(a.opts.StealthModeEnabled); err != nil {
 		return err
 	}
@@ -206,6 +215,18 @@ func (a *computeBladeAgentImpl) Run(origCtx context.Context) error {
 		}
 	}()
 
+	// Start fan controller
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		log.FromContext(ctx).Info("Starting fan controller")
+		err := a.runFanController(ctx)
+		if err != nil && err != context.Canceled {
+			log.FromContext(ctx).Error("Fan Controller Failed", zap.Error(err))
+			cancelCtx(err)
+		}
+	}()
+
 	// Start event handler
 	wg.Add(1)
 	go func() {
@@ -295,7 +316,7 @@ func (a *computeBladeAgentImpl) handleCriticalActive(ctx context.Context) error
 	log.FromContext(ctx).Warn("Blade in critical state, setting fan speed to 100% and turning on LEDs")
 
 	// Set fan speed to 100%
-	setFanspeedError := a.blade.SetFanSpeed(100)
+	a.fanController.Override(&fancontroller.FanOverrideOpts{Speed: 100})
 
 	// Disable stealth mode (turn on LEDs)
 	setStealthModeError := a.blade.SetStealthMode(false)
@@ -305,15 +326,13 @@ func (a *computeBladeAgentImpl) handleCriticalActive(ctx context.Context) error
 		ledengine.NewSlowBlinkPattern(hal.LedColor{}, a.opts.CriticalLedColor),
 	)
 	// Combine errors, but don't stop execution flow for now
-	return errors.Join(setFanspeedError, setStealthModeError, setPatternTopLedErr)
+	return errors.Join(setStealthModeError, setPatternTopLedErr)
 }
 
 func (a *computeBladeAgentImpl) handleCriticalReset(ctx context.Context) error {
 	log.FromContext(ctx).Info("Critical state cleared, setting fan speed to default and restoring LEDs to default state")
-	// Set fan speed to 100%
-	if err := a.blade.SetFanSpeed(uint8(a.opts.DefaultFanSpeed)); err != nil {
-		return err
-	}
+	// Reset fan controller overrides
+	a.fanController.Override(nil)
 
 	// Reset stealth mode
 	if err := a.blade.SetStealthMode(a.opts.StealthModeEnabled); err != nil {
@@ -351,12 +370,41 @@ func (a *computeBladeAgentImpl) runEdgeLedEngine(ctx context.Context) error {
 	return a.edgeLedEngine.Run(ctx)
 }
 
+func (a *computeBladeAgentImpl) runFanController(ctx context.Context) error {
+	// Update fan speed periodically
+	ticker := time.NewTicker(a.opts.FanUpdateInterval)
+
+	for {
+
+		// Wait for the next tick
+		select {
+		case <-ctx.Done():
+			ticker.Stop()
+			return ctx.Err()
+		case <-ticker.C:
+		}
+
+		// Get temperature
+		temp, err := a.blade.GetTemperature()
+		if err != nil {
+			log.FromContext(ctx).Error("Failed to get temperature", zap.Error(err))
+			temp = 100 // set to a high value to trigger the maximum speed defined by the fan curve
+		}
+		// Derive fan speed from temperature
+		speed := a.fanController.GetFanSpeed(temp)
+		// Set fan speed
+		if err := a.blade.SetFanSpeed(speed); err != nil {
+			log.FromContext(ctx).Error("Failed to set fan speed", zap.Error(err))
+		}
+	}
+}
+
 // EmitEvent dispatches an event to the event handler
 func (a *computeBladeAgentImpl) EmitEvent(ctx context.Context, event Event) error {
 	select {
 	case a.eventChan <- event:
 		return nil
-	case <- ctx.Done():
+	case <-ctx.Done():
 		return ctx.Err()
 	}
 }
@@ -366,7 +414,8 @@ func (a *computeBladeAgentImpl) SetFanSpeed(_ context.Context, speed uint8) erro
 	if a.state.CriticalActive() {
 		return errors.New("cannot set fan speed while the blade is in a critical state")
 	}
-	return a.blade.SetFanSpeed(speed)
+	a.fanController.Override(&fancontroller.FanOverrideOpts{Speed: speed})
+	return nil
 }
 
 // SetStealthMode enables/disables the stealth mode

pkg/fancontroller/fancontroller.go

@@ -0,0 +1,88 @@
+package fancontroller
+
+import (
+	"fmt"
+	"sync"
+)
+
+type FanController interface {
+	Override(opts *FanOverrideOpts)
+	GetFanSpeed(temperature float64) uint8
+}
+
+type FanOverrideOpts struct {
+	Speed uint8
+}
+
+type FanControllerStep struct {
+	// Temperature is the temperature to react to
+	Temperature float64
+	// Speed is the fan speed in percent
+	Speed uint8
+}
+
+// FanController configures a fan controller for the computeblade
+type FanControllerConfig struct {
+	// Steps defines the temperature/speed steps for the fan controller
+	Steps []FanControllerStep
+}
+
+// FanController is a simple fan controller that reacts to temperature changes with a linear function
+type fanControllerLinear struct {
+	mu       sync.Mutex
+	overrideOpts *FanOverrideOpts
+	config   FanControllerConfig
+}
+
+// NewFanControllerLinear creates a new FanControllerLinear
+func NewLinearFanController(config FanControllerConfig) (FanController, error) {
+
+	// Validate config for a very simple linear fan controller
+	if len(config.Steps) != 2 {
+		return nil, fmt.Errorf("exactly two steps must be defined")
+	}
+	if config.Steps[0].Temperature > config.Steps[1].Temperature {
+		return nil, fmt.Errorf("step 1 temperature must be lower than step 2 temperature")
+	}
+	if config.Steps[0].Speed > config.Steps[1].Speed {
+		return nil, fmt.Errorf("step 1 speed must be lower than step 2 speed")
+	}
+	if config.Steps[0].Speed > 100 || config.Steps[1].Speed > 100 {
+		return nil, fmt.Errorf("speed must be between 0 and 100")
+	}
+
+	return &fanControllerLinear{
+		config: config,
+	}, nil
+}
+
+func (f *fanControllerLinear) Override(opts *FanOverrideOpts) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.overrideOpts = opts
+}
+
+// GetFanSpeed returns the fan speed in percent based on the current temperature
+func (f *fanControllerLinear) GetFanSpeed(temperature float64) uint8 {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	if f.overrideOpts != nil {
+		return f.overrideOpts.Speed
+	}
+
+	if temperature <= f.config.Steps[0].Temperature {
+		return f.config.Steps[0].Speed
+	}
+	if temperature >= f.config.Steps[1].Temperature {
+		return f.config.Steps[1].Speed
+	}
+
+	// Calculate slope
+	slope := float64(f.config.Steps[1].Speed-f.config.Steps[0].Speed) / (f.config.Steps[1].Temperature - f.config.Steps[0].Temperature)
+
+	// Calculate speed
+	speed := float64(f.config.Steps[0].Speed) + slope*(temperature-f.config.Steps[0].Temperature)
+
+	return uint8(speed)
+}

pkg/fancontroller/fancontroller_test.go

@@ -0,0 +1,147 @@
+// fancontroller_test.go
+package fancontroller_test
+
+import (
+	"testing"
+
+	"github.com/xvzf/computeblade-agent/pkg/fancontroller"
+)
+
+func TestFanControllerLinear_GetFanSpeed(t *testing.T) {
+	t.Parallel()
+
+	config := fancontroller.FanControllerConfig{
+		Steps: []fancontroller.FanControllerStep{
+			{Temperature: 20, Speed: 30},
+			{Temperature: 30, Speed: 60},
+		},
+	}
+
+	controller, err := fancontroller.NewLinearFanController(config)
+	if err != nil {
+		t.Fatalf("Failed to create fan controller: %v", err)
+	}
+
+	testCases := []struct {
+		temperature float64
+		expected    uint8
+	}{
+		{15, 30}, // Should use the minimum speed
+		{25, 45}, // Should calculate speed based on linear function
+		{35, 60}, // Should use the maximum speed
+	}
+
+	for _, tc := range testCases {
+		expected := tc.expected
+		temperature := tc.temperature
+		t.Run("", func(t *testing.T) {
+			t.Parallel()
+			speed := controller.GetFanSpeed(temperature)
+			if speed != expected {
+				t.Errorf("For temperature %.2f, expected speed %d but got %d", temperature, expected, speed)
+			}
+		})
+	}
+}
+
+func TestFanControllerLinear_GetFanSpeedWithOverride(t *testing.T) {
+	t.Parallel()
+
+	config := fancontroller.FanControllerConfig{
+		Steps: []fancontroller.FanControllerStep{
+			{Temperature: 20, Speed: 30},
+			{Temperature: 30, Speed: 60},
+		},
+	}
+
+	controller, err := fancontroller.NewLinearFanController(config)
+	if err != nil {
+		t.Fatalf("Failed to create fan controller: %v", err)
+	}
+	controller.Override(&fancontroller.FanOverrideOpts{
+		Speed: 99,
+	})
+
+	testCases := []struct {
+		temperature float64
+		expected    uint8
+	}{
+		{15, 99},
+		{25, 99},
+		{35, 99},
+	}
+
+	for _, tc := range testCases {
+		expected := tc.expected
+		temperature := tc.temperature
+		t.Run("", func(t *testing.T) {
+			t.Parallel()
+			speed := controller.GetFanSpeed(temperature)
+			if speed != expected {
+				t.Errorf("For temperature %.2f, expected speed %d but got %d", temperature, expected, speed)
+			}
+		})
+	}
+}
+
+func TestFanControllerLinear_ConstructionErrors(t *testing.T) {
+	testCases := []struct {
+		name   string
+		config fancontroller.FanControllerConfig
+		errMsg string
+	}{
+		{
+			name: "InvalidStepCount",
+			config: fancontroller.FanControllerConfig{
+				Steps: []fancontroller.FanControllerStep{
+					{Temperature: 20, Speed: 30},
+				},
+			},
+			errMsg: "exactly two steps must be defined",
+		},
+		{
+			name: "InvalidStepTemperatures",
+			config: fancontroller.FanControllerConfig{
+				Steps: []fancontroller.FanControllerStep{
+					{Temperature: 30, Speed: 60},
+					{Temperature: 20, Speed: 30},
+				},
+			},
+			errMsg: "step 1 temperature must be lower than step 2 temperature",
+		},
+		{
+			name: "InvalidStepSpeeds",
+			config: fancontroller.FanControllerConfig{
+				Steps: []fancontroller.FanControllerStep{
+					{Temperature: 20, Speed: 60},
+					{Temperature: 30, Speed: 30},
+				},
+			},
+			errMsg: "step 1 speed must be lower than step 2 speed",
+		},
+		{
+			name: "InvalidSpeedRange",
+			config: fancontroller.FanControllerConfig{
+				Steps: []fancontroller.FanControllerStep{
+					{Temperature: 20, Speed: 10},
+					{Temperature: 30, Speed: 200},
+				},
+			},
+			errMsg: "speed must be between 0 and 100",
+		},
+	}
+
+	for _, tc := range testCases {
+		config := tc.config
+		expectedErrMsg := tc.errMsg
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			_, err := fancontroller.NewLinearFanController(config)
+			if err == nil {
+				t.Errorf("Expected error with message '%s', but got no error", expectedErrMsg)
+			} else if err.Error() != expectedErrMsg {
+				t.Errorf("Expected error message '%s', but got '%s'", expectedErrMsg, err.Error())
+			}
+		})
+	}
+}

pkg/hal/hal.go

@@ -22,6 +22,11 @@ var (
 		Name:      "fan_speed",
 		Help:      "Fan speed in RPM",
 	})
+	socTemperature = promauto.NewGauge(prometheus.GaugeOpts{
+		Namespace: "computeblade",
+		Name:      "soc_temperature",
+		Help:      "SoC temperature in °C",
+	})
 	computeModule = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: "computeblade",
 		Name:      "compute_modul_present",
@@ -103,6 +108,8 @@ type ComputeBladeHal interface {
 	SetLed(idx uint, color LedColor) error
 	// GetPowerStatus returns the current power status of the blade
 	GetPowerStatus() (PowerStatus, error)
+	// GetTemperature returns the current temperature of the SoC in °C
+	GetTemperature() (float64, error)
 	// GetEdgeButtonPressChan returns a channel emitting edge button press events
 	WaitForEdgeButtonPress(ctx context.Context) error
 }

pkg/hal/hal_bcm283x.go

@@ -6,7 +6,10 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
 	"os"
+	"strconv"
+	"strings"
 	"sync"
 	"syscall"
 	"time"
@@ -47,6 +50,8 @@ const (
 	bcm283xRegPwmclkCntrlBitEnable = 4
 
 	bcm283xDebounceInterval = 100 * time.Millisecond
+
+	bcm283xThermalZonePath = "/sys/class/thermal/thermal_zone0/temp"
 )
 
 type bcm283x struct {
@@ -114,17 +119,17 @@ func NewCm4Hal(opts ComputeBladeHalOpts) (ComputeBladeHal, error) {
 	}
 
 	bcm := &bcm283x{
-		devmem:              devmem,
-		gpioMem:             gpioMem,
-		gpioMem8:            gpioMem8,
-		pwmMem:              pwmMem,
-		pwmMem8:             pwmMem8,
-		clkMem:              clkMem,
-		clkMem8:             clkMem8,
-		gpioChip0:           gpioChip0,
-		opts:                opts,
+		devmem:                 devmem,
+		gpioMem:                gpioMem,
+		gpioMem8:               gpioMem8,
+		pwmMem:                 pwmMem,
+		pwmMem8:                pwmMem8,
+		clkMem:                 clkMem,
+		clkMem8:                clkMem8,
+		gpioChip0:              gpioChip0,
+		opts:                   opts,
 		edgeButtonDebounceChan: make(chan struct{}, 1),
-		edgeButtonWatchChan: make(chan struct{}),
+		edgeButtonWatchChan:    make(chan struct{}),
 	}
 
 	computeModule.WithLabelValues("cm4").Set(1)
@@ -181,7 +186,7 @@ func (bcm *bcm283x) handleEdgeButtonEdge(evt gpiod.LineEvent) {
 	case bcm.edgeButtonDebounceChan <- struct{}{}:
 		go func() {
 			// Manually debounce the button
-			defer <- bcm.edgeButtonDebounceChan
+			<-bcm.edgeButtonDebounceChan
 			time.Sleep(bcm283xDebounceInterval)
 			edgeButtonEventCount.Inc()
 			close(bcm.edgeButtonWatchChan)
@@ -440,3 +445,27 @@ func (bcm *bcm283x) updateLEDs() error {
 
 	return nil
 }
+
+// GetTemperature returns the current temperature of the SoC
+func (bcm *bcm283x) GetTemperature() (float64, error) {
+	// Read temperature
+
+	f, err := os.Open(bcm283xThermalZonePath)
+	if err != nil {
+		return -1, err
+	}
+	raw, err := io.ReadAll(f)
+	if err != nil {
+		return -1, err
+	}
+
+	cpuTemp, err := strconv.Atoi(strings.TrimSpace(string(raw)))
+	if err != nil {
+		return -1, err
+	}
+
+	temp := float64(cpuTemp) / 1000.0
+	socTemperature.Set(temp)
+
+	return temp, nil
+}

pkg/hal/hal_bcm283x_simulated.go

@@ -76,3 +76,8 @@ func (m *SimulatedHal) SetLed(idx uint, color LedColor) error {
 	m.logger.Info("SetLed", zap.Uint("idx", idx), zap.Any("color", color))
 	return nil
 }
+
+func (m *SimulatedHal) GetTemperature() (float64, error) {
+	m.logger.Info("GetTemperature")
+	return 42, nil
+}

pkg/hal/hal_mock.go

@@ -48,3 +48,8 @@ func (m *ComputeBladeHalMock) SetLed(idx uint, color LedColor) error {
 	args := m.Called(idx, color)
 	return args.Error(0)
 }
+
+func (m *ComputeBladeHalMock) GetTemperature() (float64, error) {
+	args := m.Called()
+	return args.Get(0).(float64), args.Error(1)
+}