Fix Restic backup telemetry

collectors/backup.py

@@ -1,9 +1,12 @@
 #!/usr/bin/env python3
 
-import subprocess
 import json
+import logging
+import subprocess
+from datetime import datetime, timezone
 
 RESTIC_ENV = "/etc/restic/edge01.env"
+LOGGER = logging.getLogger(__name__)
 
 
 def collect():
@@ -16,17 +19,71 @@ def collect():
     )
 
     if result.returncode != 0:
+        LOGGER.error("Restic snapshots failed: %s", result.stderr.strip())
         return {
             "state": "offline",
-            "summary": "Restic error",
-            "detail": result.stderr.strip()
+            "status": "error",
+            "summary": "Backup error",
+            "detail": "Restic snapshot query failed; see collector logs",
         }
 
+    try:
         snapshots = json.loads(result.stdout)
+    except json.JSONDecodeError:
+        LOGGER.error("Restic snapshots returned invalid JSON")
+        return {
+            "state": "offline",
+            "status": "error",
+            "summary": "Backup error",
+            "detail": "Restic returned invalid snapshot data; see collector logs",
+        }
+
+    if not isinstance(snapshots, list):
+        LOGGER.error("Restic snapshots returned an unexpected JSON shape")
+        return {
+            "state": "offline",
+            "status": "error",
+            "summary": "Backup error",
+            "detail": "Restic returned an unexpected snapshot response; see collector logs",
+        }
+
+    if not snapshots:
+        return {
+            "state": "offline",
+            "status": "error",
+            "summary": "Backup error",
+            "detail": "Restic repository contains no snapshots",
+            "snapshot_count": 0,
+        }
+
+    try:
+        latest = max(
+            snapshots,
+            key=lambda snapshot: datetime.fromisoformat(snapshot["time"]),
+        )
+        latest_time = datetime.fromisoformat(latest["time"])
+    except (KeyError, TypeError, ValueError):
+        LOGGER.error("Restic snapshots contained an invalid timestamp")
+        return {
+            "state": "offline",
+            "status": "error",
+            "summary": "Backup error",
+            "detail": "Restic returned snapshots without valid timestamps; see collector logs",
+            "snapshot_count": len(snapshots),
+        }
+
+    age_seconds = max(0, int((datetime.now(timezone.utc) - latest_time).total_seconds()))
+    age_hours, remainder = divmod(age_seconds, 3600)
+    age_minutes = remainder // 60
+    age = f"{age_hours}h {age_minutes}m" if age_hours else f"{age_minutes}m"
 
     return {
         "state": "online",
+        "status": "healthy",
         "summary": "Healthy",
-        "detail": f"{len(snapshots)} snapshots",
-        "snapshot_count": len(snapshots)
+        "detail": f"{len(snapshots)} snapshots · latest {latest_time.isoformat()} · age {age}",
+        "snapshot_count": len(snapshots),
+        "latest_snapshot": latest_time.isoformat(),
+        "latest_snapshot_age": age,
+        "latest_snapshot_age_seconds": age_seconds,
     }