feat: add health check endpoint and related schemas with tests

server/api_server/api_server_start.py

@@ -41,6 +41,7 @@ from .nettools_endpoint import (  # noqa: E402 [flake8 lint suppression]
 from .dbquery_endpoint import read_query, write_query, update_query, delete_query  # noqa: E402 [flake8 lint suppression]
 from .sync_endpoint import handle_sync_post, handle_sync_get  # noqa: E402 [flake8 lint suppression]
 from .logs_endpoint import clean_log  # noqa: E402 [flake8 lint suppression]
+from .health_endpoint import get_health_status  # noqa: E402 [flake8 lint suppression]
 from models.user_events_queue_instance import UserEventsQueueInstance  # noqa: E402 [flake8 lint suppression]
 
 from models.event_instance import EventInstance  # noqa: E402 [flake8 lint suppression]
@@ -86,6 +87,7 @@ from .openapi.schemas import (  # noqa: E402 [flake8 lint suppression]
     RecentEventsResponse, LastEventsResponse,
     NetworkTopologyResponse,
     InternetInfoResponse, NetworkInterfacesResponse,
+    HealthCheckResponse,
     CreateEventRequest, CreateSessionRequest,
     DeleteSessionRequest, CreateNotificationRequest,
     SyncPushRequest, SyncPullResponse,
@@ -1930,6 +1932,33 @@ def check_auth(payload=None):
     if request.method == "GET":
         return jsonify({"success": True, "message": "Authentication check successful"}), 200
 
+
+# --------------------------
+# Health endpoint
+# --------------------------
+@app.route("/health", methods=["GET"])
+@validate_request(
+    operation_id="check_health",
+    summary="System Health Check",
+    description="Retrieve system vitality metrics including database size, memory pressure, system load, disk usage, and CPU temperature.",
+    response_model=HealthCheckResponse,
+    tags=["system", "health"],
+    auth_callable=is_authorized
+)
+def check_health(payload=None):
+    """Get system health metrics for monitoring and diagnostics."""
+    try:
+        health_data = get_health_status()
+        return jsonify({"success": True, **health_data}), 200
+    except Exception as e:
+        mylog("none", [f"[health] Error retrieving health status: {e}"])
+        return jsonify({
+            "success": False,
+            "error": "Failed to retrieve health status",
+            "message": str(e)
+        }), 500
+
+
 # --------------------------
 # Background Server Start
 # --------------------------

server/api_server/health_endpoint.py

@@ -0,0 +1,147 @@
+"""Health check endpoint for NetAlertX system vitality monitoring."""
+
+import os
+import psutil
+from pathlib import Path
+
+from const import dbPath, dataPath
+from logger import mylog
+
+
+# ===============================================================================
+# Database Vitality
+# ===============================================================================
+
+def get_db_size_mb():
+    """
+    Calculate total database size in MB (app.db + app.db-wal).
+
+    Returns:
+        float: Size in MB, or 0 if database files don't exist.
+    """
+    try:
+        db_file = Path(dbPath)
+        wal_file = Path(f"{dbPath}-wal")
+
+        size_bytes = 0
+        if db_file.exists():
+            size_bytes += db_file.stat().st_size
+        if wal_file.exists():
+            size_bytes += wal_file.stat().st_size
+
+        return round(size_bytes / (1024 * 1024), 2)
+    except Exception as e:
+        mylog("verbose", [f"[health] Error calculating DB size: {e}"])
+        return 0.0
+
+
+# ===============================================================================
+# Memory Pressure
+# ===============================================================================
+
+def get_mem_usage_pct():
+    """
+    Calculate memory usage percentage (used / total * 100).
+
+    Returns:
+        int: Memory usage as integer percentage (0-100), or -1 on error.
+    """
+    try:
+        vm = psutil.virtual_memory()
+        pct = int((vm.used / vm.total) * 100)
+        return max(0, min(100, pct))  # Clamp to 0-100
+    except Exception as e:
+        mylog("verbose", [f"[health] Error calculating memory usage: {e}"])
+        return -1
+
+
+# ===============================================================================
+# System Stress
+# ===============================================================================
+
+def get_load_avg_1m():
+    """
+    Get 1-minute load average.
+
+    Returns:
+        float: 1-minute load average, or -1 on error.
+    """
+    try:
+        load_1m, _, _ = os.getloadavg()
+        return round(load_1m, 2)
+    except Exception as e:
+        mylog("verbose", [f"[health] Error getting load average: {e}"])
+        return -1.0
+
+
+# ===============================================================================
+# Disk Headroom
+# ===============================================================================
+
+def get_storage_pct():
+    """
+    Calculate disk usage percentage of /data mount.
+
+    Returns:
+        int: Disk usage as integer percentage (0-100), or -1 on error.
+    """
+    try:
+        stat = os.statvfs(dataPath)
+        total = stat.f_blocks * stat.f_frsize
+        used = (stat.f_blocks - stat.f_bfree) * stat.f_frsize
+        pct = int((used / total) * 100) if total > 0 else 0
+        return max(0, min(100, pct))  # Clamp to 0-100
+    except Exception as e:
+        mylog("verbose", [f"[health] Error calculating storage usage: {e}"])
+        return -1
+
+
+# ===============================================================================
+# Thermal Health
+# ===============================================================================
+
+def get_cpu_temp():
+    """
+    Get CPU temperature from hardware sensors if available.
+
+    Returns:
+        int: CPU temperature in Celsius, or None if unavailable.
+    """
+    try:
+        temps = psutil.sensors_temperatures()
+        if not temps:
+            return None
+
+        # Prefer 'coretemp' (Intel), fallback to first available
+        if "coretemp" in temps and temps["coretemp"]:
+            return int(temps["coretemp"][0].current)
+
+        # Fallback to first sensor with data
+        for sensor_type, readings in temps.items():
+            if readings:
+                return int(readings[0].current)
+
+        return None
+    except Exception as e:
+        mylog("verbose", [f"[health] Error reading CPU temperature: {e}"])
+        return None
+
+
+# ===============================================================================
+# Aggregator
+# ===============================================================================
+
+def get_health_status():
+    """
+    Collect all health metrics into a single dict.
+
+    Returns:
+        dict: Dictionary with all health metrics.
+    """
+    return {
+        "db_size_mb": get_db_size_mb(),
+        "mem_usage_pct": get_mem_usage_pct(),
+        "load_1m": get_load_avg_1m(),
+        "storage_pct": get_storage_pct(),
+        "cpu_temp": get_cpu_temp(),
+    }

server/api_server/openapi/schemas.py

@@ -651,6 +651,34 @@ class NetworkInterfacesResponse(BaseResponse):
     interfaces: Dict[str, Any] = Field(..., description="Details about network interfaces.")
 
 
+# =============================================================================
+# HEALTH CHECK SCHEMAS
+# =============================================================================
+
+
+class HealthCheckResponse(BaseResponse):
+    """System health check with vitality metrics."""
+    model_config = ConfigDict(
+        extra="allow",
+        json_schema_extra={
+            "examples": [{
+                "success": True,
+                "db_size_mb": 125.45,
+                "mem_usage_pct": 65,
+                "load_1m": 2.15,
+                "storage_pct": 42,
+                "cpu_temp": 58
+            }]
+        }
+    )
+
+    db_size_mb: float = Field(..., description="Database size in MB (app.db + app.db-wal)")
+    mem_usage_pct: int = Field(..., ge=0, le=100, description="Memory usage percentage (0-100)")
+    load_1m: float = Field(..., description="1-minute load average")
+    storage_pct: int = Field(..., ge=0, le=100, description="Disk usage percentage of /data mount (0-100)")
+    cpu_temp: Optional[int] = Field(None, description="CPU temperature in Celsius (nullable if unavailable)")
+
+
 # =============================================================================
 # EVENTS SCHEMAS
 # =============================================================================