Autoparts-DB/manager/services/health_service.py

"""Health monitoring service for Nexus infrastructure."""
import subprocess
import shutil
import socket
import urllib.request
import urllib.error
import psycopg2
import redis
from config import (
    MASTER_DB_URL, REDIS_URL, POS_URL, DASHBOARD_URL, QUART_URL,
    TENANT_DB_URL_TEMPLATE, NEXUS_SERVER_HOST
)


def check_postgresql():
    """Check PostgreSQL connectivity."""
    try:
        conn = psycopg2.connect(MASTER_DB_URL, connect_timeout=5)
        cur = conn.cursor()
        cur.execute("SELECT version(), pg_database_size('nexus_autoparts')")
        version, size = cur.fetchone()
        cur.close()
        conn.close()
        return {
            "status": "ok",
            "version": version.split()[1] if version else "unknown",
            "master_size_mb": round(size / (1024 * 1024), 2)
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}


def check_redis():
    """Check Redis connectivity. May be unreachable if Redis only binds to localhost."""
    try:
        r = redis.from_url(REDIS_URL, socket_connect_timeout=3)
        info = r.info()
        return {
            "status": "ok",
            "version": info.get("redis_version", "unknown"),
            "used_memory_human": info.get("used_memory_human", "?"),
            "connected_clients": info.get("connected_clients", 0)
        }
    except redis.ConnectionError:
        return {
            "status": "warning",
            "error": "Redis unreachable. If manager runs on a separate VM, ensure Redis binds to 0.0.0.0 or a VPN interface, or that a tunnel is active."
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}


def check_http_service(name, url, timeout=5):
    """Generic HTTP health check."""
    try:
        req = urllib.request.Request(url, method="GET")
        req.add_header("User-Agent", "Nexus-Manager/1.0")
        with urllib.request.urlopen(req, timeout=timeout) as resp:
            return {
                "status": "ok",
                "http_status": resp.status,
                "latency_ms": None  # Could add timing later
            }
    except urllib.error.HTTPError as e:
        return {"status": "warning", "http_status": e.code, "error": str(e)}
    except Exception as e:
        return {"status": "error", "error": str(e)}


def check_disk_space(path="/"):
    """Check disk usage."""
    try:
        total, used, free = shutil.disk_usage(path)
        return {
            "status": "ok",
            "total_gb": round(total / (1024**3), 2),
            "used_gb": round(used / (1024**3), 2),
            "free_gb": round(free / (1024**3), 2),
            "percent_used": round((used / total) * 100, 1)
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}


def check_memory():
    """Check system memory via /proc/meminfo."""
    try:
        with open("/proc/meminfo") as f:
            meminfo = f.read()
        data = {}
        for line in meminfo.splitlines():
            if ":" in line:
                key, value = line.split(":", 1)
                data[key.strip()] = int(value.strip().split()[0])  # kB
        total = data.get("MemTotal", 0) / 1024 / 1024  # GB
        available = data.get("MemAvailable", data.get("MemFree", 0)) / 1024 / 1024
        used = total - available
        return {
            "status": "ok",
            "total_gb": round(total, 2),
            "used_gb": round(used, 2),
            "available_gb": round(available, 2),
            "percent_used": round((used / total) * 100, 1) if total else 0
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}


def check_systemd_service(service_name):
    """Check systemd service status."""
    try:
        result = subprocess.run(
            ["systemctl", "is-active", service_name],
            capture_output=True, text=True, timeout=5
        )
        active = result.stdout.strip() == "active"
        return {
            "status": "ok" if active else "warning",
            "active": active,
            "state": result.stdout.strip()
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}


def get_full_health_report():
    """Aggregate health report for all services."""
    return {
        "_meta": {
            "nexus_server_host": NEXUS_SERVER_HOST,
            "note": "disk/memory are local to this manager VM. PostgreSQL/HTTP checks target the remote Nexus server."
        },
        "postgresql": check_postgresql(),
        "redis": check_redis(),
        "pos": check_http_service("pos", POS_URL),
        "dashboard": check_http_service("dashboard", DASHBOARD_URL),
        "quart": check_http_service("quart", QUART_URL),
        "disk": check_disk_space(),
        "memory": check_memory(),
        "services": {
            "nexus": check_systemd_service("nexus.service"),
            "nexus-pos": check_systemd_service("nexus-pos.service"),
            "nexus-quart": check_systemd_service("nexus-quart.service"),
            "nexus-celery": check_systemd_service("nexus-celery.service"),
        }
    }


def get_tenant_health(db_name, timeout=5):
    """Check connectivity to a specific tenant database."""
    dsn = TENANT_DB_URL_TEMPLATE.format(db_name=db_name)
    try:
        conn = psycopg2.connect(dsn, connect_timeout=timeout)
        cur = conn.cursor()
        cur.execute("""
            SELECT
                (SELECT COUNT(*) FROM employees WHERE is_active = true) as employees,
                (SELECT COUNT(*) FROM inventory WHERE is_active = true) as inventory,
                (SELECT COUNT(*) FROM customers WHERE is_active = true) as customers,
                (SELECT COUNT(*) FROM sales WHERE created_at > NOW() - INTERVAL '30 days') as sales_30d,
                pg_database_size(current_database()) as db_size
        """)
        row = cur.fetchone()
        cur.close()
        conn.close()
        return {
            "status": "ok",
            "employees": row[0],
            "inventory": row[1],
            "customers": row[2],
            "sales_30d": row[3],
            "db_size_mb": round(row[4] / (1024 * 1024), 2)
        }
    except Exception as e:
        return {"status": "error", "error": str(e)}