Redesign health service + create backup service

2025-12-18 03:35:13 +07:00
parent e43e579329
commit 57bad3b4a8
16 changed files with 1486 additions and 98 deletions
--- a/status-service/database.py
+++ b/status-service/database.py
@@ -0,0 +1,261 @@
+"""SQLite database for storing metrics history."""
+import sqlite3
+from datetime import datetime, timedelta
+from pathlib import Path
+from typing import Optional
+import json
+
+
+DB_PATH = Path("/app/data/metrics.db")
+
+
+def get_connection() -> sqlite3.Connection:
+    """Get database connection."""
+    DB_PATH.parent.mkdir(parents=True, exist_ok=True)
+    conn = sqlite3.connect(str(DB_PATH))
+    conn.row_factory = sqlite3.Row
+    return conn
+
+
+def init_db():
+    """Initialize database tables."""
+    conn = get_connection()
+    cursor = conn.cursor()
+
+    # Metrics history table
+    cursor.execute("""
+        CREATE TABLE IF NOT EXISTS metrics (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            service_name TEXT NOT NULL,
+            status TEXT NOT NULL,
+            latency_ms REAL,
+            message TEXT,
+            checked_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+        )
+    """)
+
+    # Incidents table
+    cursor.execute("""
+        CREATE TABLE IF NOT EXISTS incidents (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            service_name TEXT NOT NULL,
+            status TEXT NOT NULL,
+            message TEXT,
+            started_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+            resolved_at TIMESTAMP,
+            notified BOOLEAN DEFAULT FALSE
+        )
+    """)
+
+    # SSL certificates table
+    cursor.execute("""
+        CREATE TABLE IF NOT EXISTS ssl_certificates (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            domain TEXT NOT NULL UNIQUE,
+            issuer TEXT,
+            expires_at TIMESTAMP,
+            days_until_expiry INTEGER,
+            checked_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+        )
+    """)
+
+    # Create indexes
+    cursor.execute("""
+        CREATE INDEX IF NOT EXISTS idx_metrics_service_time
+        ON metrics(service_name, checked_at DESC)
+    """)
+    cursor.execute("""
+        CREATE INDEX IF NOT EXISTS idx_incidents_service
+        ON incidents(service_name, started_at DESC)
+    """)
+
+    conn.commit()
+    conn.close()
+
+
+def save_metric(service_name: str, status: str, latency_ms: Optional[float], message: Optional[str]):
+    """Save a metric record."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute(
+        "INSERT INTO metrics (service_name, status, latency_ms, message) VALUES (?, ?, ?, ?)",
+        (service_name, status, latency_ms, message)
+    )
+    conn.commit()
+    conn.close()
+
+
+def get_latency_history(service_name: str, hours: int = 24) -> list[dict]:
+    """Get latency history for a service."""
+    conn = get_connection()
+    cursor = conn.cursor()
+
+    since = datetime.now() - timedelta(hours=hours)
+    cursor.execute("""
+        SELECT latency_ms, status, checked_at
+        FROM metrics
+        WHERE service_name = ? AND checked_at > ? AND latency_ms IS NOT NULL
+        ORDER BY checked_at ASC
+    """, (service_name, since.isoformat()))
+
+    rows = cursor.fetchall()
+    conn.close()
+
+    return [
+        {
+            "latency_ms": row["latency_ms"],
+            "status": row["status"],
+            "checked_at": row["checked_at"]
+        }
+        for row in rows
+    ]
+
+
+def get_uptime_stats(service_name: str, hours: int = 24) -> dict:
+    """Calculate uptime statistics for a service."""
+    conn = get_connection()
+    cursor = conn.cursor()
+
+    since = datetime.now() - timedelta(hours=hours)
+
+    cursor.execute("""
+        SELECT COUNT(*) as total,
+               SUM(CASE WHEN status = 'operational' THEN 1 ELSE 0 END) as successful
+        FROM metrics
+        WHERE service_name = ? AND checked_at > ?
+    """, (service_name, since.isoformat()))
+
+    row = cursor.fetchone()
+    conn.close()
+
+    total = row["total"] or 0
+    successful = row["successful"] or 0
+
+    return {
+        "total_checks": total,
+        "successful_checks": successful,
+        "uptime_percent": (successful / total * 100) if total > 0 else 100.0
+    }
+
+
+def get_avg_latency(service_name: str, hours: int = 24) -> Optional[float]:
+    """Get average latency for a service."""
+    conn = get_connection()
+    cursor = conn.cursor()
+
+    since = datetime.now() - timedelta(hours=hours)
+    cursor.execute("""
+        SELECT AVG(latency_ms) as avg_latency
+        FROM metrics
+        WHERE service_name = ? AND checked_at > ? AND latency_ms IS NOT NULL
+    """, (service_name, since.isoformat()))
+
+    row = cursor.fetchone()
+    conn.close()
+
+    return row["avg_latency"]
+
+
+def create_incident(service_name: str, status: str, message: Optional[str]) -> int:
+    """Create a new incident."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute(
+        "INSERT INTO incidents (service_name, status, message) VALUES (?, ?, ?)",
+        (service_name, status, message)
+    )
+    incident_id = cursor.lastrowid
+    conn.commit()
+    conn.close()
+    return incident_id
+
+
+def resolve_incident(service_name: str):
+    """Resolve open incidents for a service."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("""
+        UPDATE incidents
+        SET resolved_at = CURRENT_TIMESTAMP
+        WHERE service_name = ? AND resolved_at IS NULL
+    """, (service_name,))
+    conn.commit()
+    conn.close()
+
+
+def get_open_incident(service_name: str) -> Optional[dict]:
+    """Get open incident for a service."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("""
+        SELECT * FROM incidents
+        WHERE service_name = ? AND resolved_at IS NULL
+        ORDER BY started_at DESC LIMIT 1
+    """, (service_name,))
+    row = cursor.fetchone()
+    conn.close()
+
+    if row:
+        return dict(row)
+    return None
+
+
+def mark_incident_notified(incident_id: int):
+    """Mark incident as notified."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("UPDATE incidents SET notified = TRUE WHERE id = ?", (incident_id,))
+    conn.commit()
+    conn.close()
+
+
+def get_recent_incidents(limit: int = 10) -> list[dict]:
+    """Get recent incidents."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("""
+        SELECT * FROM incidents
+        ORDER BY started_at DESC
+        LIMIT ?
+    """, (limit,))
+    rows = cursor.fetchall()
+    conn.close()
+    return [dict(row) for row in rows]
+
+
+def save_ssl_info(domain: str, issuer: str, expires_at: datetime, days_until_expiry: int):
+    """Save SSL certificate info."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("""
+        INSERT OR REPLACE INTO ssl_certificates
+        (domain, issuer, expires_at, days_until_expiry, checked_at)
+        VALUES (?, ?, ?, ?, CURRENT_TIMESTAMP)
+    """, (domain, issuer, expires_at.isoformat(), days_until_expiry))
+    conn.commit()
+    conn.close()
+
+
+def get_ssl_info(domain: str) -> Optional[dict]:
+    """Get SSL certificate info."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cursor.execute("SELECT * FROM ssl_certificates WHERE domain = ?", (domain,))
+    row = cursor.fetchone()
+    conn.close()
+
+    if row:
+        return dict(row)
+    return None
+
+
+def cleanup_old_metrics(days: int = 7):
+    """Delete metrics older than specified days."""
+    conn = get_connection()
+    cursor = conn.cursor()
+    cutoff = datetime.now() - timedelta(days=days)
+    cursor.execute("DELETE FROM metrics WHERE checked_at < ?", (cutoff.isoformat(),))
+    deleted = cursor.rowcount
+    conn.commit()
+    conn.close()
+    return deleted