deploy-felhom-compose/controller/internal/monitor/healthcheck.go

package monitor

import (
	"fmt"
	"os/exec"
	"strings"
	"time"

	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
	"gitea.dooplex.hu/admin/felhom-controller/internal/system"
)

// HealthReport contains the results of a system health check.
type HealthReport struct {
	Status    string   // "ok", "warn", "fail"
	Issues    []string // critical problems
	Warnings  []string // non-critical warnings
	Info      []string // informational items
	Timestamp time.Time
}

// RunHealthCheck runs system checks and returns a diagnostic report.
func RunHealthCheck(cfg *config.Config, cpuCollector *system.CPUCollector) *HealthReport {
	report := &HealthReport{
		Status:    "ok",
		Timestamp: time.Now(),
	}

	sysInfo := system.GetInfo(cfg.Paths.HDDPath, cpuCollector)

	// 1. Disk usage (SSD)
	if sysInfo.DiskPercent > 0 {
		if sysInfo.DiskPercent >= float64(cfg.Monitoring.Thresholds.DiskCritPercent) {
			report.Issues = append(report.Issues, fmt.Sprintf("SSD disk usage critical: %.0f%%", sysInfo.DiskPercent))
		} else if sysInfo.DiskPercent >= float64(cfg.Monitoring.Thresholds.DiskWarnPercent) {
			report.Warnings = append(report.Warnings, fmt.Sprintf("SSD disk usage high: %.0f%%", sysInfo.DiskPercent))
		} else {
			report.Info = append(report.Info, fmt.Sprintf("SSD: %.0f%% used", sysInfo.DiskPercent))
		}
	}

	// HDD disk usage
	if sysInfo.HDDConfigured && sysInfo.HDDPercent > 0 {
		if sysInfo.HDDPercent >= float64(cfg.Monitoring.Thresholds.DiskCritPercent) {
			report.Issues = append(report.Issues, fmt.Sprintf("HDD disk usage critical: %.0f%%", sysInfo.HDDPercent))
		} else if sysInfo.HDDPercent >= float64(cfg.Monitoring.Thresholds.DiskWarnPercent) {
			report.Warnings = append(report.Warnings, fmt.Sprintf("HDD disk usage high: %.0f%%", sysInfo.HDDPercent))
		}
	}

	// 2. Memory usage
	if sysInfo.MemPercent > 0 {
		if sysInfo.MemPercent >= float64(cfg.Monitoring.Thresholds.MemoryWarnPercent) {
			report.Warnings = append(report.Warnings, fmt.Sprintf("Memory usage high: %.0f%%", sysInfo.MemPercent))
		} else {
			report.Info = append(report.Info, fmt.Sprintf("Memory: %.0f%% used", sysInfo.MemPercent))
		}
	}

	// 3. CPU usage
	if sysInfo.CPUPercent > 0 {
		if sysInfo.CPUPercent >= float64(cfg.Monitoring.Thresholds.CPUWarnPercent) {
			report.Warnings = append(report.Warnings, fmt.Sprintf("CPU usage high: %.0f%%", sysInfo.CPUPercent))
		} else {
			report.Info = append(report.Info, fmt.Sprintf("CPU: %.0f%%", sysInfo.CPUPercent))
		}
	}

	// 4. Temperature
	if sysInfo.TemperatureCelsius > 0 {
		if sysInfo.TemperatureCelsius >= float64(cfg.Monitoring.Thresholds.TemperatureWarnCelsius) {
			report.Warnings = append(report.Warnings, fmt.Sprintf("Temperature high: %.0f°C (%s)", sysInfo.TemperatureCelsius, sysInfo.TemperatureSource))
		} else {
			report.Info = append(report.Info, fmt.Sprintf("Temperature: %.0f°C", sysInfo.TemperatureCelsius))
		}
	}

	// 5. Docker health
	if err := checkDocker(); err != nil {
		report.Issues = append(report.Issues, fmt.Sprintf("Docker: %v", err))
	} else {
		report.Info = append(report.Info, "Docker: reachable")
	}

	// 6. Protected containers
	missingProtected := checkProtectedContainers(cfg.Stacks.Protected)
	for _, name := range missingProtected {
		report.Issues = append(report.Issues, fmt.Sprintf("Protected container not running: %s", name))
	}

	// Determine status
	if len(report.Issues) > 0 {
		report.Status = "fail"
	} else if len(report.Warnings) > 0 {
		report.Status = "warn"
	}

	return report
}

// FormatMessage returns a human-readable summary for healthcheck ping body.
func (r *HealthReport) FormatMessage() string {
	var sb strings.Builder

	sb.WriteString(fmt.Sprintf("Status: %s\n", strings.ToUpper(r.Status)))
	sb.WriteString(fmt.Sprintf("Time: %s\n\n", r.Timestamp.Format("2006-01-02 15:04:05")))

	if len(r.Issues) > 0 {
		sb.WriteString("ISSUES:\n")
		for _, issue := range r.Issues {
			sb.WriteString("  - " + issue + "\n")
		}
		sb.WriteString("\n")
	}

	if len(r.Warnings) > 0 {
		sb.WriteString("WARNINGS:\n")
		for _, w := range r.Warnings {
			sb.WriteString("  - " + w + "\n")
		}
		sb.WriteString("\n")
	}

	if len(r.Info) > 0 {
		sb.WriteString("INFO:\n")
		for _, info := range r.Info {
			sb.WriteString("  - " + info + "\n")
		}
	}

	return sb.String()
}

func checkDocker() error {
	cmd := exec.Command("docker", "info", "--format", "{{.ServerVersion}}")
	out, err := cmd.Output()
	if err != nil {
		return fmt.Errorf("docker not reachable: %v", err)
	}
	if len(strings.TrimSpace(string(out))) == 0 {
		return fmt.Errorf("docker returned empty version")
	}
	return nil
}

func checkProtectedContainers(protected []string) []string {
	var missing []string
	for _, name := range protected {
		cmd := exec.Command("docker", "inspect", "--format", "{{.State.Running}}", name)
		out, err := cmd.Output()
		if err != nil {
			missing = append(missing, name)
			continue
		}
		if strings.TrimSpace(string(out)) != "true" {
			missing = append(missing, name)
		}
	}
	return missing
}