Add service watchdog for continuous Minecraft autonomy

app.py

@@ -528,6 +528,54 @@ def _stream(payload: dict[str, Any]) -> Iterator[str]:
 
 api = FastAPI(title="Programa Voyager OpenAI Endpoint", version="6.0.0")
 
+_supervisor_lock = threading.Lock()
+_supervisor_recoveries = 0
+_supervisor_last_action = "startup"
+_supervisor_last_check = 0.0
+
+
+def _service_supervisor() -> None:
+    global _supervisor_recoveries, _supervisor_last_action, _supervisor_last_check
+    backoff = 10
+    while True:
+        try:
+            time.sleep(30)
+            _supervisor_last_check = time.time()
+            mc = STACK.status()
+            processes = mc.get("processes") or {}
+            critical = ("minecraft", "viaproxy", "friends")
+            dead = [name for name in critical if name in processes and not processes[name].get("running")]
+            missing = [name for name in critical if mc.get("state") == "running" and name not in processes]
+            if mc.get("state") == "error" or dead or missing:
+                with _supervisor_lock:
+                    _supervisor_recoveries += 1
+                    _supervisor_last_action = f"minecraft recovery #{_supervisor_recoveries}: dead={dead} missing={missing} state={mc.get('state')}"
+                    VOYAGER.stop()
+                    STACK.stop()
+                    time.sleep(backoff)
+                    STACK.start_async()
+                    VOYAGER.start_async()
+                backoff = min(backoff * 2, 120)
+                continue
+
+            vs = VOYAGER.status()
+            if mc.get("state") == "running" and vs.get("state") in {"error", "exited", "stopped"} and not vs.get("running"):
+                with _supervisor_lock:
+                    _supervisor_recoveries += 1
+                    _supervisor_last_action = f"voyager recovery #{_supervisor_recoveries}: state={vs.get('state')}"
+                    VOYAGER.start_async()
+                backoff = min(backoff * 2, 120)
+                continue
+
+            if mc.get("state") == "running" and vs.get("running"):
+                backoff = 10
+                _supervisor_last_action = "healthy"
+        except Exception as exc:
+            _supervisor_last_action = f"supervisor exception: {type(exc).__name__}: {str(exc)[:160]}"
+            time.sleep(min(backoff, 120))
+            backoff = min(backoff * 2, 120)
+
+
 @api.on_event("startup")
 def _startup() -> None:
     if PRELOAD:
@@ -537,6 +585,7 @@ def _startup() -> None:
         STACK.start_async()
     if os.environ.get("VOYAGER_AUTOSTART", "1") == "1":
         VOYAGER.start_async()
+    threading.Thread(target=_service_supervisor, daemon=True, name="service-supervisor").start()
 
 @api.get("/")
 def root():
@@ -560,6 +609,9 @@ def health():
         "embedding_error": _embed_error or None,
         "backend": "cpu-only",
         "n_gpu_layers": 0,
+        "supervisor_recoveries": _supervisor_recoveries,
+        "supervisor_last_action": _supervisor_last_action,
+        "supervisor_last_check": _supervisor_last_check or None,
         "error": _model_error or None,
     }