Add incidents export, queue alerts, and health summaries

services/selftest.py

@@ -5,10 +5,12 @@ from typing import Any
 
 from services.health import health
 from services.runner import run_cmd_full
+from services.incidents import log_incident
 
 
-async def run_selftest(cfg: dict[str, Any], docker_map: dict[str, str]) -> str:
+async def run_selftest(cfg: dict[str, Any], docker_map: dict[str, str]) -> tuple[str, bool]:
     lines = ["🧪 Self-test"]
+    ok = True
 
     # health
     try:
@@ -18,6 +20,7 @@ async def run_selftest(cfg: dict[str, Any], docker_map: dict[str, str]) -> str:
         lines.append(f"🟢 Health: {brief}")
     except Exception as e:
         lines.append(f"🔴 Health failed: {e}")
+        ok = False
 
     # restic snapshots check
     rc, out = await run_cmd_full(["restic", "snapshots", "--json"], use_restic_env=True, timeout=40)
@@ -35,8 +38,9 @@ async def run_selftest(cfg: dict[str, Any], docker_map: dict[str, str]) -> str:
             lines.append("🟡 Restic snapshots: invalid JSON")
     else:
         lines.append(f"🔴 Restic snapshots error: {out.strip() or rc}")
+        ok = False
 
-    return "\n".join(lines)
+    return "\n".join(lines), ok
 
 
 async def schedule_selftest(cfg: dict[str, Any], bot, admin_ids: list[int], docker_map: dict[str, str]):
@@ -58,9 +62,14 @@ async def schedule_selftest(cfg: dict[str, Any], bot, admin_ids: list[int], dock
         if run_at <= now:
             run_at += timedelta(days=1)
         await asyncio.sleep((run_at - now).total_seconds())
-        text = await run_selftest(cfg, docker_map)
+        text, ok = await run_selftest(cfg, docker_map)
         for chat_id in admin_ids:
             try:
                 await bot.send_message(chat_id, text)
             except Exception:
                 pass
+        if not ok:
+            try:
+                log_incident(cfg, "selftest failed", category="selftest")
+            except Exception:
+                pass