AI-tianyan/internal/control/heartbeat.go

package control

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"log"
	"net/http"
	"os/exec"
	"strconv"
	"strings"
	"time"

	"tianyan-edge/internal/config"
)

type Heartbeat struct {
	cfg *config.Config
}

func NewHeartbeat(cfg *config.Config) *Heartbeat {
	return &Heartbeat{cfg: cfg}
}

func (h *Heartbeat) Run(ctx context.Context) {
	client := &http.Client{Timeout: 5 * time.Second}
	ticker := time.NewTicker(30 * time.Second)
	defer ticker.Stop()

	// Send first heartbeat immediately
	h.send(client)

	for {
		select {
		case <-ctx.Done():
			return
		case <-ticker.C:
			h.send(client)
		}
	}
}

func (h *Heartbeat) send(client *http.Client) {
	payload := map[string]any{
		"edge_id": h.cfg.EdgeID,
		"version": h.cfg.Version,
		"ts":      float64(time.Now().UnixMilli()) / 1000.0,
	}

	// Collect NPU stats if available
	if npu := collectNPUStats(); npu != nil {
		payload["npu"] = npu
	}

	body, _ := json.Marshal(payload)
	url := fmt.Sprintf("%s/api/v1/edge/heartbeat", h.cfg.CloudURL)
	req, _ := http.NewRequest(http.MethodPost, url, bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", "Bearer "+h.cfg.EdgeToken)
	resp, err := client.Do(req)
	if err != nil {
		log.Printf("heartbeat: %v", err)
		return
	}
	resp.Body.Close()
}

// NPUStats holds parsed NPU telemetry
type NPUStats struct {
	TempC      int `json:"temp_c"`
	UtilPct    int `json:"util_pct"`
	MemUsedMB  int `json:"mem_used_mb"`
	MemTotalMB int `json:"mem_total_mb"`
	MemPct     int `json:"mem_pct"`
}

// collectNPUStats runs npu-smi and parses the output
func collectNPUStats() *NPUStats {
	// Try npu-smi first (Ascend devices), fall back to the helper script
	var commonOut, memOut []byte
	var err error

	commonOut, err = exec.Command("/usr/local/sbin/npu-smi", "info", "-t", "common", "-i", "0").CombinedOutput()
	if err != nil {
		// Fallback to shell script
		commonOut, err = exec.Command("bash", "/opt/tianyan-edge/scripts/npu_info.sh").CombinedOutput()
		if err != nil {
			log.Printf("npu: monitoring unavailable: %v", err)
			return nil
		}
		// If script succeeded, it outputs Influx line protocol — parse that
		line := strings.TrimSpace(string(commonOut))
		return parseInfluxLine(line)
	}

	memOut, _ = exec.Command("/usr/local/sbin/npu-smi", "info", "-t", "memory", "-i", "0").CombinedOutput()

	return parseNpuSmiOutput(string(commonOut), string(memOut))
}

func parseNpuSmiOutput(common, memory string) *NPUStats {
	stats := &NPUStats{}

	for _, line := range strings.Split(common, "\n") {
		line = strings.TrimSpace(line)
		if strings.Contains(line, "Temperature") {
			if v := extractLastNumber(line); v != 0 {
				stats.TempC = v
			}
		}
		if strings.Contains(line, "Aicore Usage Rate") {
			if v := extractLastNumber(line); v != 0 {
				stats.UtilPct = v
			}
		}
		if strings.Contains(line, "Memory Usage Rate") {
			if v := extractLastNumber(line); v != 0 {
				stats.MemPct = v
			}
		}
	}

	for _, line := range strings.Split(memory, "\n") {
		line = strings.TrimSpace(line)
		if strings.Contains(line, "Capacity") {
			if v := extractLastNumber(line); v != 0 {
				stats.MemTotalMB = v
			}
		}
	}

	if stats.MemTotalMB > 0 && stats.MemPct > 0 {
		stats.MemUsedMB = stats.MemTotalMB * stats.MemPct / 100
	}

	return stats
}

func parseInfluxLine(line string) *NPUStats {
	// Format: npu_status,device=0 temp=X,utilization=Y,memory_used=Z,memory_total=W
	stats := &NPUStats{}
	parts := strings.Split(line, " ")
	if len(parts) < 2 {
		return nil
	}
	for _, field := range strings.Split(parts[1], ",") {
		kv := strings.SplitN(field, "=", 2)
		if len(kv) != 2 {
			continue
		}
		v, _ := strconv.Atoi(kv[1])
		switch kv[0] {
		case "temp":
			stats.TempC = v
		case "utilization":
			stats.UtilPct = v
		case "memory_used":
			stats.MemUsedMB = v
		case "memory_total":
			stats.MemTotalMB = v
		}
	}
	return stats
}

func extractLastNumber(line string) int {
	fields := strings.Fields(line)
	if len(fields) == 0 {
		return 0
	}
	// Last field might have units like "%" or "C"
	raw := fields[len(fields)-1]
	raw = strings.TrimRight(raw, "%C ")
	v, _ := strconv.Atoi(raw)
	return v
}