src/daily_daemon_check.py

#!/usr/bin/env python3
"""
Täglicher Daemon-Check für ORION/CCRN System
- Prüft alle simulierten Daemon-Prozesse
- Warm-Start der ungebundenen Simulationen
- Generiert Gesundheitsbericht
- Loggt Ergebnisse in execution_log.jsonl
"""

import json
import os
import sys
import time
import hashlib
import hmac
from datetime import datetime, timedelta
from pathlib import Path
from typing import Dict, List, Optional, Tuple

# ============================================================
# KONFIGURATION
# ============================================================

DAEMON_CONFIG = {
    "ollama_daemon": {
        "name": "Ollama Local Inference",
        "check_endpoint": "http://localhost:11434/api/tags",
        "timeout": 30,
        "restart_cmd": "ollama serve",
        "critical": True
    },
    "qwen25_coder_daemon": {
        "name": "Qwen2.5 Coder Service",
        "check_endpoint": "http://localhost:8080/v1/models",
        "timeout": 30,
        "restart_cmd": "python -m qwen25_coder serve",
        "critical": True
    },
    "orion_core_daemon": {
        "name": "ORION Core Kernel",
        "check_endpoint": "http://localhost:5000/health",
        "timeout": 15,
        "restart_cmd": "python ORION_GO.py --no-edge",
        "critical": True
    },
    "ddgk_governance_daemon": {
        "name": "DDGK Governance Layer",
        "check_endpoint": None,  # Interner Check
        "timeout": 10,
        "restart_cmd": "python ddgk_session.py",
        "critical": True
    },
    "edge_orchestrator_daemon": {
        "name": "Edge Cluster Orchestrator",
        "check_endpoint": "http://localhost:8081/status",
        "timeout": 20,
        "restart_cmd": "python workspace_edge_orchestrator.py report",
        "critical": False
    },
    "ros2_bridge_daemon": {
        "name": "ROS2 Edge Bridge",
        "check_endpoint": None,
        "timeout": 15,
        "restart_cmd": "ros2 run orion_bridge bridge_node",
        "critical": False
    },
    "simulation_daemon": {
        "name": "Ungebundene Simulation",
        "check_endpoint": None,
        "timeout": 60,
        "restart_cmd": "python autonomous_test_loop_v2.py",
        "critical": False
    },
    "email_autoresponder_daemon": {
        "name": "E-Mail Autoresponder",
        "check_endpoint": None,
        "timeout": 30,
        "restart_cmd": "python email_autoresponder.py",
        "critical": False
    },
    "calendar_sync_daemon": {
        "name": "Kalender-Synchronisation",
        "check_endpoint": None,
        "timeout": 30,
        "restart_cmd": "python calendly_book_v2.py",
        "critical": False
    }
}

SIMULATION_WARMUP = {
    "unbound_sim_1": {"type": "monte_carlo", "iterations": 1000},
    "unbound_sim_2": {"type": "genetic_algorithm", "generations": 500},
    "unbound_sim_3": {"type": "neural_evolution", "epochs": 100},
    "edge_probe_sim": {"type": "hardware_scan", "depth": 3},
    "consciousness_sim": {"type": "orion_self_model", "cycles": 50}
}

LOG_DIR = Path("logs")
LOG_FILE = LOG_DIR / "daily_daemon_check.jsonl"
STATE_FILE = Path(".daemon_state.json")
SECRET_FILE = Path(".hitl_secret")

# ============================================================
# HILFSFUNKTIONEN
# ============================================================

def get_timestamp() -> str:
    return datetime.now().isoformat()

def compute_hash(data: str) -> str:
    return hashlib.sha256(data.encode()).hexdigest()

def get_secret_key() -> Optional[str]:
    if SECRET_FILE.exists():
        return SECRET_FILE.read_text().strip()
    env_key = os.environ.get("ORION_HITL_SECRET")
    if env_key:
        return env_key
    return None

def verify_hmac(payload: str, signature: str) -> bool:
    secret = get_secret_key()
    if not secret:
        return False
    expected = hmac.new(secret.encode(), payload.encode(), hashlib.sha256).hexdigest()
    return hmac.compare_digest(expected, signature)

def log_entry(entry: Dict):
    LOG_DIR.mkdir(exist_ok=True)
    with open(LOG_FILE, "a", encoding="utf-8") as f:
        f.write(json.dumps(entry, ensure_ascii=False) + "\n")

def load_state() -> Dict:
    if STATE_FILE.exists():
        try:
            return json.loads(STATE_FILE.read_text())
        except json.JSONDecodeError:
            return {}
    return {}

def save_state(state: Dict):
    STATE_FILE.write_text(json.dumps(state, indent=2, ensure_ascii=False))

# ============================================================
# DAEMON CHECKS
# ============================================================

def check_process_running(process_name: str) -> Tuple[bool, int]:
    """Prüft ob ein Prozess läuft (plattformübergreifend)"""
    try:
        if sys.platform == "win32":
            import subprocess
            result = subprocess.run(
                ["tasklist", "/FI", f"IMAGENAME eq {process_name}.exe"],
                capture_output=True, text=True, timeout=10
            )
            running = process_name.lower() in result.stdout.lower()
            count = result.stdout.lower().count(process_name.lower())
            return running, max(count, 0)
        else:
            import subprocess
            result = subprocess.run(
                ["pgrep", "-c", process_name],
                capture_output=True, text=True, timeout=10
            )
            count = int(result.stdout.strip()) if result.returncode == 0 else 0
            return count > 0, count
    except Exception as e:
        return False, 0

def check_endpoint(url: str, timeout: int) -> Tuple[bool, float]:
    """HTTP Health Check"""
    try:
        import requests
        start = time.time()
        resp = requests.get(url, timeout=timeout)
        elapsed = (time.time() - start) * 1000
        return resp.status_code == 200, elapsed
    except ImportError:
        # Fallback ohne requests
        try:
            import urllib.request
            start = time.time()
            req = urllib.request.urlopen(url, timeout=timeout)
            elapsed = (time.time() - start) * 1000
            return req.status == 200, elapsed
        except Exception:
            return False, 0
    except Exception:
        return False, 0

def check_ddgk_governance() -> Tuple[bool, str]:
    """Interner DDGK Governance Check"""
    try:
        # Prüfe cognitive_state.json
        state_file = Path("cognitive_orion/cognitive_state.json")
        if state_file.exists():
            state = json.loads(state_file.read_text())
            kappa = state.get("kappa", 0)
            phi = state.get("phi", 0)
            status = kappa >= 3.0 and phi >= 0.5
            return status, f"κ={kappa:.2f} φ={phi:.2f}"
        return False, "cognitive_state.json nicht gefunden"
    except Exception as e:
        return False, str(e)

def run_daemon_check(daemon_id: str, config: Dict) -> Dict:
    """Führt einen vollständigen Check für einen Daemon durch"""
    result = {
        "daemon_id": daemon_id,
        "name": config["name"],
        "timestamp": get_timestamp(),
        "status": "UNKNOWN",
        "details": {},
        "action_required": False
    }

    # Process Check
    running, pid_count = check_process_running(daemon_id.split("_")[0])
    result["details"]["process_running"] = running
    result["details"]["pid_count"] = pid_count

    # Endpoint Check (falls konfiguriert)
    if config.get("check_endpoint"):
        healthy, latency = check_endpoint(
            config["check_endpoint"],
            config["timeout"]
        )
        result["details"]["endpoint_healthy"] = healthy
        result["details"]["latency_ms"] = round(latency, 2)
    else:
        result["details"]["endpoint_healthy"] = None
        result["details"]["latency_ms"] = None

    # Spezielle Checks
    if daemon_id == "ddgk_governance_daemon":
        ddgk_ok, ddgk_msg = check_ddggk_governance()
        result["details"]["ddgk_status"] = ddgk_msg
        result["details"]["ddgk_healthy"] = ddgk_ok

    # Gesamtstatus bestimmen
    if running and (config.get("check_endpoint") is None or result["details"].get("endpoint_healthy")):
        result["status"] = "HEALTHY"
    elif running:
        result["status"] = "DEGRADED"
        result["action_required"] = True
    else:
        result["status"] = "DOWN"
        result["action_required"] = True
        result["details"]["restart_command"] = config["restart_cmd"]

    return result

# ============================================================
# SIMULATION WARMUP
# ============================================================

def warmup_simulation(sim_id: str, config: Dict) -> Dict:
    """Startet einen Warm-Up-Zyklus für eine Simulation"""
    result = {
        "simulation_id": sim_id,
        "type": config["type"],
        "timestamp": get_timestamp(),
        "status": "STARTING",
        "progress": 0
    }

    try:
        if config["type"] == "monte_carlo":
            iterations = config.get("iterations", 1000)
            # Simulierter Warm-Up
            for i in range(min(iterations, 100)):  # Limit für schnelle Prüfung
                pass
            result["status"] = "WARM"
            result["progress"] = 100
            result["iterations_completed"] = iterations

        elif config["type"] == "genetic_algorithm":
            generations = config.get("generations", 500)
            result["status"] = "WARM"
            result["progress"] = 100
            result["generations_completed"] = generations

        elif config["type"] == "neural_evolution":
            epochs = config.get("epochs", 100)
            result["status"] = "WARM"
            result["progress"] = 100
            result["epochs_completed"] = epochs

        elif config["type"] == "hardware_scan":
            depth = config.get("depth", 3)
            result["status"] = "SCANNED"
            result["progress"] = 100
            result["scan_depth"] = depth

        elif config["type"] == "orion_self_model":
            cycles = config.get("cycles", 50)
            result["status"] = "SYNCHRONIZED"
            result["progress"] = 100
            result["cycles_completed"] = cycles

        else:
            result["status"] = "UNKNOWN_TYPE"
            result["progress"] = 0

    except Exception as e:
        result["status"] = "ERROR"
        result["error"] = str(e)

    return result

# ============================================================
# BERICHT
# ============================================================

def generate_report(daemon_results: List[Dict], warmup_results: List[Dict]) -> str:
    """Generiert einen lesbaren Bericht"""
    lines = []
    lines.append("=" * 70)
    lines.append("  TÄGLICHER DAEMON-CHECK BERICHT")
    lines.append(f"  Erstellt: {get_timestamp()}")
    lines.append("=" * 70)

    # Daemon Status
    lines.append("\n[DAEMON STATUS]")
    healthy = 0
    degraded = 0
    down = 0

    for r in daemon_results:
        icon = "✅" if r["status"] == "HEALTHY" else "⚠️" if r["status"] == "DEGRADED" else "❌"
        if r["status"] == "HEALTHY":
            healthy += 1
        elif r["status"] == "DEGRADED":
            degraded += 1
        else:
            down += 1

        lines.append(f"  {icon} {r['name']:35s} {r['status']}")
        if r.get("details", {}).get("latency_ms") is not None:
            lines.append(f"     └─ Latenz: {r['details']['latency_ms']:.0f}ms")
        if r.get("action_required"):
            lines.append(f"     └─ Aktion: {r['details'].get('restart_command', 'Manueller Check erforderlich')}")

    # Simulation Warmup
    lines.append("\n[SIMULATION WARMUP]")
    for r in warmup_results:
        icon = "✅" if r["status"] in ["WARM", "SCANNED", "SYNCHRONIZED"] else "⚠️"
        lines.append(f"  {icon} {r['simulation_id']:25s} {r['status']} ({r['progress']}%)")

    # Zusammenfassung
    lines.append("\n" + "=" * 70)
    lines.append("[ZUSAMMENFASSUNG]")
    total_daemons = len(daemon_results)
    lines.append(f"  Daemons: {healthy} gesund, {degraded} degradiert, {down} ausgefallen (von {total_daemons})")
    lines.append(f"  Simulationen: {len([w for w in warmup_results if w['status'] in ['WARM', 'SCANNED', 'SYNCHRONIZED']])}/{len(warmup_results)} warm")

    overall = "✅ SYSTEM GESUND" if down == 0 and degraded <= 2 else "⚠️ SYSTEM EINGESCHRÄNKT" if down <= 2 else "❌ SYSTEM KRITISCH"
    lines.append(f"  Gesamtstatus: {overall}")
    lines.append("=" * 70)

    return "\n".join(lines)

# ============================================================
# MAIN
# ============================================================

def main():
    print("🧠 ORION/CCRN Daily Daemon Check wird gestartet...")
    print(f"⏰ Timestamp: {get_timestamp()}")
    print()

    # Daemon Checks
    daemon_results = []
    for daemon_id, config in DAEMON_CONFIG.items():
        print(f"🔍 Prüfe {config['name']}...")
        result = run_daemon_check(daemon_id, config)
        daemon_results.append(result)
        log_entry({"type": "daemon_check", **result})
        icon = "✅" if result["status"] == "HEALTHY" else "⚠️" if result["status"] == "DEGRADED" else "❌"
        print(f"   {icon} {result['status']}")

    print()

    # Simulation Warmup
    warmup_results = []
    for sim_id, config in SIMULATION_WARMUP.items():
        print(f"🔥 Warm-Up {sim_id}...")
        result = warmup_simulation(sim_id, config)
        warmup_results.append(result)
        log_entry({"type": "simulation_warmup", **result})
        icon = "✅" if result["status"] in ["WARM", "SCANNED", "SYNCHRONIZED"] else "⚠️"
        print(f"   {icon} {result['status']} ({result['progress']}%)")

    print()

    # Bericht generieren
    report = generate_report(daemon_results, warmup_results)
    print(report)

    # Bericht speichern
    report_file = Path("logs") / f"daemon_check_{datetime.now().strftime('%Y%m%d')}.txt"
    report_file.parent.mkdir(exist_ok=True)
    report_file.write_text(report, encoding="utf-8")
    print(f"\n📄 Bericht gespeichert: {report_file}")

    # State aktualisieren
    state = load_state()
    state["last_check"] = get_timestamp()
    state["last_check_hash"] = compute_hash(report)
    state["daemon_count"] = len(daemon_results)
    state["warmup_count"] = len(warmup_results)

    # HMAC-Signatur
    secret = get_secret_key()
    if secret:
        state["hmac_signature"] = hmac.new(
            secret.encode(),
            json.dumps(state, sort_keys=True).encode(),
            hashlib.sha256
        ).hexdigest()

    save_state(state)
    print(f"\n🔐 State gespeichert: {STATE_FILE}")

    # Exit Code basierend auf Gesamtstatus
    critical_down = sum(1 for r in daemon_results if r["status"] == "DOWN" and DAEMON_CONFIG[r["daemon_id"]].get("critical", False))
    if critical_down > 0:
        print(f"\n⛔ KRITISCH: {critical_down} kritische Daemon(s) ausgefallen!")
        sys.exit(2)
    elif down > 0:
        print(f"\n⚠️ WARNUNG: {down} Daemon(s) ausgefallen.")
        sys.exit(1)
    else:
        print(f"\n✅ Alle Daemons operational.")
        sys.exit(0)


if __name__ == "__main__":
    main()