feat: add health monitoring endpoint

apps/health/checks.py

@@ -0,0 +1,80 @@
+from __future__ import annotations
+
+import importlib
+import os
+import time
+import uuid
+from pathlib import Path
+
+from django.core.cache import cache
+from django.db import connection
+
+BACKUP_MAX_AGE_SECONDS = 48 * 60 * 60
+
+
+def check_db() -> dict[str, float | str]:
+    started = time.perf_counter()
+    try:
+        with connection.cursor() as cursor:
+            cursor.execute("SELECT 1")
+    except Exception as exc:
+        return {"status": "fail", "detail": str(exc)}
+    return {"status": "ok", "latency_ms": (time.perf_counter() - started) * 1000}
+
+
+def check_cache() -> dict[str, float | str]:
+    cache_key = f"health:{uuid.uuid4().hex}"
+    probe_value = uuid.uuid4().hex
+    started = time.perf_counter()
+    try:
+        cache.set(cache_key, probe_value, timeout=5)
+        cached_value = cache.get(cache_key)
+        if cached_value != probe_value:
+            return {"status": "fail", "detail": "Cache probe returned unexpected value"}
+        cache.delete(cache_key)
+    except Exception as exc:
+        return {"status": "fail", "detail": str(exc)}
+    return {"status": "ok", "latency_ms": (time.perf_counter() - started) * 1000}
+
+
+def check_celery() -> dict[str, str]:
+    broker_url = os.environ.get("CELERY_BROKER_URL")
+    if not broker_url:
+        return {"status": "ok", "detail": "Celery not configured: CELERY_BROKER_URL is unset"}
+
+    try:
+        kombu = importlib.import_module("kombu")
+    except ImportError:
+        return {"status": "ok", "detail": "Celery broker check skipped: kombu is not installed"}
+
+    try:
+        with kombu.Connection(broker_url, connect_timeout=3) as broker_connection:
+            broker_connection.ensure_connection(max_retries=1)
+    except Exception as exc:
+        return {"status": "fail", "detail": str(exc)}
+    return {"status": "ok"}
+
+
+def check_backup() -> dict[str, str]:
+    backup_status_file = os.environ.get("BACKUP_STATUS_FILE")
+    if not backup_status_file:
+        return {"status": "fail", "detail": "Backup monitoring not configured: BACKUP_STATUS_FILE is unset"}
+
+    try:
+        raw_timestamp = Path(backup_status_file).read_text(encoding="utf-8").strip()
+    except FileNotFoundError:
+        return {"status": "fail", "detail": f"Backup status file not found: {backup_status_file}"}
+    except OSError as exc:
+        return {"status": "fail", "detail": str(exc)}
+
+    try:
+        last_backup_at = float(raw_timestamp)
+    except ValueError:
+        return {"status": "fail", "detail": "Invalid backup status file"}
+
+    age_seconds = time.time() - last_backup_at
+    if age_seconds > BACKUP_MAX_AGE_SECONDS:
+        age_hours = age_seconds / 3600
+        return {"status": "fail", "detail": f"Last backup is {age_hours:.1f} hours old (> 48 h)"}
+
+    return {"status": "ok"}