feat: Add temperature collection to pulse-host-agent (related to #661)

Implements temperature monitoring in pulse-host-agent to support Docker-in-VM
deployments where the sensor proxy socket cannot cross VM boundaries.

Changes:
- Create internal/sensors package with local collection and parsing
- Add temperature collection to host agent (Linux only, best-effort)
- Support CPU package/core, NVMe, and GPU temperature sensors
- Update TEMPERATURE_MONITORING.md with Docker-in-VM setup instructions
- Update HOST_AGENT.md to document temperature feature

The host agent now automatically collects temperature data on Linux systems
with lm-sensors installed. This provides an alternative path for temperature
monitoring when running Pulse in a VM, avoiding the unix socket limitation.

Temperature collection is best-effort and fails gracefully if lm-sensors is
not available, ensuring other metrics continue to be reported.

Related to #661

docs/HOST_AGENT.md

@@ -2,12 +2,30 @@
 
 The Pulse host agent extends monitoring to standalone servers that do not expose
 Proxmox or Docker APIs. With it you can surface uptime, OS metadata, CPU load,
-memory/disk utilisation, and connection health for any Linux, macOS, or Windows
-machine alongside the rest of your infrastructure. Starting in v4.26.0 the
-installer handshakes with Pulse in real time so you can confirm registration
-from the UI and receive host-agent alerts alongside your existing
+memory/disk utilisation, temperature sensors, and connection health for any Linux,
+macOS, or Windows machine alongside the rest of your infrastructure. Starting in
+v4.26.0 the installer handshakes with Pulse in real time so you can confirm
+registration from the UI and receive host-agent alerts alongside your existing
 Docker/Proxmox notifications.
 
+## Temperature Monitoring
+
+The host agent automatically collects temperature data on Linux systems with lm-sensors installed:
+
+- **CPU Package Temperature**: Overall CPU temperature
+- **Per-Core Temperatures**: Individual CPU core readings
+- **NVMe Drive Temperatures**: SSD thermal data
+- **GPU Temperatures**: AMD and NVIDIA GPU sensors
+
+Temperature data appears in the **Servers** tab alongside other host metrics. This is particularly useful for monitoring Proxmox hosts when running Pulse in a VM (where the sensor proxy socket cannot cross VM boundaries).
+
+**Requirements:**
+- Linux operating system
+- lm-sensors package installed (`apt-get install lm-sensors`)
+- Sensors configured (`sensors-detect --auto`)
+
+Temperature collection is automatic and best-effort. If lm-sensors is not installed or sensors are unavailable, the agent continues reporting other metrics normally.
+
 ## Prerequisites
 
 - Pulse v4.26.0 or newer (host agent reporting shipped with `/api/agents/host/report`)

docs/TEMPERATURE_MONITORING.md

@@ -16,14 +16,40 @@ Pulse can display real-time CPU and NVMe temperatures directly in your dashboard
 
 > **Important:** Temperature monitoring setup differs by deployment type:
 > - **LXC containers:** Fully automatic via the setup script (Settings → Nodes → Setup Script)
-> - **Docker containers:** Requires manual proxy installation (see below)
+> - **Docker containers:** Requires manual proxy installation (see below) OR use pulse-host-agent
+> - **Docker in VM:** Use pulse-host-agent on the Proxmox host (see [Docker in VM Setup](#docker-in-vm-setup))
 > - **Native installs:** Direct SSH, no proxy needed
 >
 > **For automation (Ansible/Terraform/etc.):** Jump to [Automation-Friendly Installation](#automation-friendly-installation)
 
+## Docker in VM Setup
+
+**Running Pulse in Docker inside a VM on Proxmox?** The proxy socket cannot cross VM boundaries, so use pulse-host-agent instead.
+
+pulse-host-agent runs natively on your Proxmox host and reports temperatures back to Pulse over HTTPS. This works across VM boundaries without requiring socket mounts or SSH configuration.
+
+**Setup steps:**
+
+1. Install lm-sensors on your Proxmox host (if not already installed):
+   ```bash
+   apt-get update && apt-get install -y lm-sensors
+   sensors-detect --auto
+   ```
+
+2. Install pulse-host-agent on your Proxmox host:
+   ```bash
+   # Generate an API token in Pulse (Settings → Security → API Tokens) with host-agent:report scope
+   curl -fsSL http://your-pulse-vm:7655/install-host-agent.sh | \
+     bash -s -- --url http://your-pulse-vm:7655 --token YOUR_API_TOKEN
+   ```
+
+3. Verify temperatures appear in Pulse UI under the Servers tab
+
+The host agent will report CPU, NVMe, and GPU temperatures alongside other system metrics. No proxy installation or socket mounting needed.
+
 ## Quick Start for Docker Deployments
 
-**Running Pulse in Docker?** Temperature monitoring requires installing a small service on your Proxmox host that reads hardware sensors. The Pulse container connects to this service through a shared socket.
+**Running Pulse in Docker directly on Proxmox?** Temperature monitoring requires installing a small service on your Proxmox host that reads hardware sensors. The Pulse container connects to this service through a shared socket.
 
 **Why this is needed:** Docker containers cannot directly access hardware sensors. The proxy runs on your Proxmox host where it has access to sensor data, then shares that data with the Pulse container through a secure connection.
 

internal/hostagent/agent.go

@@ -13,6 +13,7 @@ import (
 	"time"
 
 	"github.com/rcourtman/pulse-go-rewrite/internal/hostmetrics"
+	"github.com/rcourtman/pulse-go-rewrite/internal/sensors"
 	agentshost "github.com/rcourtman/pulse-go-rewrite/pkg/agents/host"
 	"github.com/rs/zerolog"
 	gohost "github.com/shirou/gopsutil/v4/host"
@@ -220,6 +221,9 @@ func (a *Agent) buildReport(ctx context.Context) (agentshost.Report, error) {
 		return agentshost.Report{}, fmt.Errorf("collect metrics: %w", err)
 	}
 
+	// Collect temperature data (best effort - don't fail if unavailable)
+	sensorData := a.collectTemperatures(collectCtx)
+
 	report := agentshost.Report{
 		Agent: agentshost.AgentInfo{
 			ID:              a.agentID,
@@ -248,7 +252,7 @@ func (a *Agent) buildReport(ctx context.Context) (agentshost.Report, error) {
 		},
 		Disks:     append([]agentshost.Disk(nil), snapshot.Disks...),
 		Network:   append([]agentshost.NetworkInterface(nil), snapshot.Network...),
-		Sensors:   agentshost.Sensors{},
+		Sensors:   sensorData,
 		Tags:      append([]string(nil), a.cfg.Tags...),
 		Timestamp: time.Now().UTC(),
 	}
@@ -304,3 +308,64 @@ func isLoopback(flags []string) bool {
 	}
 	return false
 }
+
+// collectTemperatures attempts to collect temperature data from the local system.
+// Returns an empty Sensors struct if collection fails (best-effort).
+func (a *Agent) collectTemperatures(ctx context.Context) agentshost.Sensors {
+	// Only collect on Linux for now (lm-sensors is Linux-specific)
+	if a.platform != "linux" {
+		return agentshost.Sensors{}
+	}
+
+	// Collect sensor JSON output
+	jsonOutput, err := sensors.CollectLocal(ctx)
+	if err != nil {
+		a.logger.Debug().Err(err).Msg("Failed to collect sensor data (lm-sensors may not be installed)")
+		return agentshost.Sensors{}
+	}
+
+	// Parse the sensor output
+	tempData, err := sensors.Parse(jsonOutput)
+	if err != nil {
+		a.logger.Debug().Err(err).Msg("Failed to parse sensor data")
+		return agentshost.Sensors{}
+	}
+
+	if !tempData.Available {
+		a.logger.Debug().Msg("No temperature sensors available on this system")
+		return agentshost.Sensors{}
+	}
+
+	// Convert to host agent sensor format
+	result := agentshost.Sensors{
+		TemperatureCelsius: make(map[string]float64),
+	}
+
+	// Add CPU package temperature
+	if tempData.CPUPackage > 0 {
+		result.TemperatureCelsius["cpu_package"] = tempData.CPUPackage
+	}
+
+	// Add individual core temperatures
+	for coreName, temp := range tempData.Cores {
+		// Normalize core name (e.g., "Core 0" -> "cpu_core_0")
+		normalizedName := strings.ToLower(strings.ReplaceAll(coreName, " ", "_"))
+		result.TemperatureCelsius["cpu_"+normalizedName] = temp
+	}
+
+	// Add NVMe temperatures
+	for nvmeName, temp := range tempData.NVMe {
+		result.TemperatureCelsius[nvmeName] = temp
+	}
+
+	// Add GPU temperatures
+	for gpuName, temp := range tempData.GPU {
+		result.TemperatureCelsius[gpuName] = temp
+	}
+
+	a.logger.Debug().
+		Int("temperatureCount", len(result.TemperatureCelsius)).
+		Msg("Collected temperature data")
+
+	return result
+}

internal/sensors/collector.go

@@ -0,0 +1,47 @@
+package sensors
+
+import (
+	"context"
+	"fmt"
+	"os/exec"
+	"strings"
+	"time"
+)
+
+// CollectLocal reads sensor data from the local machine using lm-sensors.
+// Returns the raw JSON output from `sensors -j` or an error if sensors is not available.
+func CollectLocal(ctx context.Context) (string, error) {
+	// Check if sensors command exists
+	if _, err := exec.LookPath("sensors"); err != nil {
+		return "", fmt.Errorf("lm-sensors not installed: %w", err)
+	}
+
+	// Create context with timeout
+	cmdCtx, cancel := context.WithTimeout(ctx, 5*time.Second)
+	defer cancel()
+
+	// Run sensors -j command
+	// sensors exits non-zero when optional subfeatures fail; "|| true" keeps the JSON for parsing
+	cmd := exec.CommandContext(cmdCtx, "sh", "-c", "sensors -j 2>/dev/null || true")
+	output, err := cmd.Output()
+	if err != nil {
+		return "", fmt.Errorf("failed to execute sensors: %w", err)
+	}
+
+	outputStr := strings.TrimSpace(string(output))
+	if outputStr == "" || outputStr == "{}" {
+		// Try Raspberry Pi temperature method as fallback
+		cmd = exec.CommandContext(cmdCtx, "cat", "/sys/class/thermal/thermal_zone0/temp")
+		if rpiOutput, rpiErr := cmd.Output(); rpiErr == nil {
+			rpiTemp := strings.TrimSpace(string(rpiOutput))
+			if rpiTemp != "" {
+				// Convert to pseudo-sensors format for compatibility
+				// Raspberry Pi reports in millidegrees Celsius
+				return fmt.Sprintf(`{"cpu_thermal-virtual-0":{"temp1":{"temp1_input":%s}}}`, rpiTemp), nil
+			}
+		}
+		return "", fmt.Errorf("sensors returned empty output")
+	}
+
+	return outputStr, nil
+}

internal/sensors/parser.go

@@ -0,0 +1,248 @@
+package sensors
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"strings"
+
+	"github.com/rs/zerolog/log"
+)
+
+// TemperatureData contains parsed temperature readings from sensors
+type TemperatureData struct {
+	CPUPackage float64            // Overall CPU package temperature
+	CPUMax     float64            // Maximum CPU temperature
+	Cores      map[string]float64 // Per-core temperatures (e.g., "Core 0": 45.0)
+	NVMe       map[string]float64 // NVMe drive temperatures (e.g., "nvme0": 42.0)
+	GPU        map[string]float64 // GPU temperatures (e.g., "amdgpu-pci-0400": 55.0)
+	Available  bool               // Whether any temperature data was found
+}
+
+// Parse extracts temperature data from sensors -j JSON output
+func Parse(jsonStr string) (*TemperatureData, error) {
+	if strings.TrimSpace(jsonStr) == "" {
+		return nil, fmt.Errorf("empty sensors output")
+	}
+
+	var sensorsData map[string]interface{}
+	if err := json.Unmarshal([]byte(jsonStr), &sensorsData); err != nil {
+		return nil, fmt.Errorf("failed to parse sensors JSON: %w", err)
+	}
+
+	data := &TemperatureData{
+		Cores: make(map[string]float64),
+		NVMe:  make(map[string]float64),
+		GPU:   make(map[string]float64),
+	}
+
+	foundCPUChip := false
+
+	// Parse each sensor chip
+	for chipName, chipData := range sensorsData {
+		chipMap, ok := chipData.(map[string]interface{})
+		if !ok {
+			continue
+		}
+
+		chipLower := strings.ToLower(chipName)
+
+		// Handle CPU temperature sensors
+		if isCPUChip(chipLower) {
+			foundCPUChip = true
+			parseCPUTemps(chipMap, data)
+		}
+
+		// Handle NVMe temperature sensors
+		if strings.Contains(chipName, "nvme") {
+			parseNVMeTemps(chipName, chipMap, data)
+		}
+
+		// Handle GPU temperature sensors
+		if strings.Contains(chipLower, "amdgpu") || strings.Contains(chipLower, "nouveau") {
+			parseGPUTemps(chipName, chipMap, data)
+		}
+	}
+
+	// If we got CPU temps, calculate max from cores if package not available
+	if data.CPUPackage == 0 && len(data.Cores) > 0 {
+		for _, temp := range data.Cores {
+			if temp > data.CPUMax {
+				data.CPUMax = temp
+			}
+		}
+		// Use max core temp as package temp if not available
+		data.CPUPackage = data.CPUMax
+	}
+
+	data.Available = foundCPUChip || len(data.NVMe) > 0 || len(data.GPU) > 0
+
+	log.Debug().
+		Bool("available", data.Available).
+		Float64("cpuPackage", data.CPUPackage).
+		Float64("cpuMax", data.CPUMax).
+		Int("coreCount", len(data.Cores)).
+		Int("nvmeCount", len(data.NVMe)).
+		Int("gpuCount", len(data.GPU)).
+		Msg("Parsed temperature data")
+
+	return data, nil
+}
+
+func isCPUChip(chipLower string) bool {
+	cpuChips := []string{
+		"coretemp", "k10temp", "zenpower", "k8temp", "acpitz",
+		"it87", "nct6687", "nct6775", "nct6776", "nct6779",
+		"nct6791", "nct6792", "nct6793", "nct6795", "nct6796",
+		"nct6797", "nct6798", "w83627", "f71882",
+		"cpu_thermal", "rpitemp",
+	}
+
+	for _, chip := range cpuChips {
+		if strings.Contains(chipLower, chip) {
+			return true
+		}
+	}
+	return false
+}
+
+func parseCPUTemps(chipMap map[string]interface{}, data *TemperatureData) {
+	foundPackageTemp := false
+	var chipletTemps []float64
+
+	for sensorName, sensorData := range chipMap {
+		sensorMap, ok := sensorData.(map[string]interface{})
+		if !ok {
+			continue
+		}
+
+		sensorNameLower := strings.ToLower(sensorName)
+
+		// Look for Package id (Intel) or Tdie/Tctl (AMD)
+		if strings.Contains(sensorName, "Package id") ||
+			strings.Contains(sensorName, "Tdie") ||
+			strings.Contains(sensorNameLower, "tctl") {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) {
+				data.CPUPackage = tempVal
+				foundPackageTemp = true
+				if tempVal > data.CPUMax {
+					data.CPUMax = tempVal
+				}
+			}
+		}
+
+		// Look for AMD chiplet temperatures
+		if strings.HasPrefix(sensorName, "Tccd") {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) && tempVal > 0 {
+				chipletTemps = append(chipletTemps, tempVal)
+				if tempVal > data.CPUMax {
+					data.CPUMax = tempVal
+				}
+			}
+		}
+
+		// Look for SuperIO chip CPU temperature fields
+		if strings.Contains(sensorNameLower, "cputin") ||
+			strings.Contains(sensorNameLower, "cpu temperature") ||
+			(strings.Contains(sensorNameLower, "temp") && strings.Contains(sensorNameLower, "cpu")) {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) && tempVal > 0 {
+				if !foundPackageTemp {
+					data.CPUPackage = tempVal
+					foundPackageTemp = true
+				}
+				if tempVal > data.CPUMax {
+					data.CPUMax = tempVal
+				}
+			}
+		}
+
+		// Look for individual core temperatures
+		if strings.Contains(sensorName, "Core ") {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) {
+				data.Cores[sensorName] = tempVal
+				if tempVal > data.CPUMax {
+					data.CPUMax = tempVal
+				}
+			}
+		}
+	}
+
+	// If no package temp but we have chiplet temps, use highest chiplet
+	if !foundPackageTemp && len(chipletTemps) > 0 {
+		for _, temp := range chipletTemps {
+			if temp > data.CPUPackage {
+				data.CPUPackage = temp
+			}
+		}
+	}
+}
+
+func parseNVMeTemps(chipName string, chipMap map[string]interface{}, data *TemperatureData) {
+	for sensorName, sensorData := range chipMap {
+		sensorMap, ok := sensorData.(map[string]interface{})
+		if !ok {
+			continue
+		}
+
+		// Look for Composite temperature (main NVMe temp)
+		if strings.Contains(sensorName, "Composite") {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) {
+				data.NVMe[chipName] = tempVal
+				log.Debug().
+					Str("chip", chipName).
+					Float64("temp", tempVal).
+					Msg("Found NVMe temperature")
+			}
+		}
+	}
+}
+
+func parseGPUTemps(chipName string, chipMap map[string]interface{}, data *TemperatureData) {
+	for sensorName, sensorData := range chipMap {
+		sensorMap, ok := sensorData.(map[string]interface{})
+		if !ok {
+			continue
+		}
+
+		sensorNameLower := strings.ToLower(sensorName)
+
+		// Look for GPU temperature fields
+		if strings.Contains(sensorNameLower, "edge") ||
+			strings.Contains(sensorNameLower, "junction") ||
+			strings.Contains(sensorNameLower, "mem") ||
+			strings.Contains(sensorNameLower, "temp1") {
+			if tempVal := extractTempInput(sensorMap); !math.IsNaN(tempVal) {
+				// Use sensor name as key (e.g., "edge", "junction")
+				key := fmt.Sprintf("%s_%s", chipName, sensorName)
+				data.GPU[key] = tempVal
+				log.Debug().
+					Str("chip", chipName).
+					Str("sensor", sensorName).
+					Float64("temp", tempVal).
+					Msg("Found GPU temperature")
+			}
+		}
+	}
+}
+
+func extractTempInput(sensorMap map[string]interface{}) float64 {
+	// Look for temp*_input field (the actual temperature reading)
+	for key, value := range sensorMap {
+		if strings.HasSuffix(key, "_input") {
+			switch v := value.(type) {
+			case float64:
+				return v
+			case int:
+				return float64(v)
+			case string:
+				// Raspberry Pi reports in millidegrees as string
+				var milliTemp float64
+				if _, err := fmt.Sscanf(v, "%f", &milliTemp); err == nil {
+					// Convert from millidegrees to degrees
+					return milliTemp / 1000.0
+				}
+			}
+		}
+	}
+	return math.NaN()
+}