Fix health check false positives, ClientDisconnect spam, and Stripe API burst

- Increase HEALTH_CHECK_TIMEOUT 10->30s, MAX_FAILURES 3->5 to prevent
  killing healthy servers during load spikes
- Handle ClientDisconnect in DiscordErrorMiddleware - skip Discord error
  reporting for normal client disconnections (was firing every minute)
- Add set_if_not_exists to SharedCache for atomic lock operations
- Use SETNX lock in background Stripe check to prevent race condition
  where multiple workers all start Stripe checks simultaneously
  (reduced 234 Stripe API calls to 78 per cache-miss cycle)

Author: Josh-XT

agixt/MagicalAuth.py

@@ -5667,6 +5667,13 @@ def _background_stripe_subscription_check(
             logging.debug(f"Stripe check cache hit for {user_email}")
             return
 
+        # Use a lock key to prevent multiple workers from checking simultaneously
+        lock_key = f"stripe_lock:{user_id}"
+        if not shared_cache.set_if_not_exists(lock_key, "1", ttl=30):
+            # Another worker is already checking, skip
+            logging.debug(f"Stripe check already in progress for {user_email}")
+            return
+
         stripe.api_key = api_key
 
         session = get_session()

agixt/SharedCache.py

@@ -201,6 +201,40 @@ def set(self, key: str, value: Any, ttl: int = 0) -> bool:
             }
         return True
 
+    def set_if_not_exists(self, key: str, value: Any, ttl: int = 0) -> bool:
+        """
+        Set a value only if the key does not already exist (atomic).
+        Returns True if the value was set, False if the key already existed.
+        """
+        full_key = self._make_key(key)
+
+        try:
+            serialized = json.dumps(value)
+        except (TypeError, ValueError):
+            return False
+
+        if self._redis is not None:
+            try:
+                if ttl > 0:
+                    result = self._redis.set(full_key, serialized, nx=True, ex=ttl)
+                else:
+                    result = self._redis.set(full_key, serialized, nx=True)
+                return bool(result)
+            except Exception as e:
+                logger.debug(f"SharedCache Redis setnx error: {e}")
+
+        # Local cache fallback
+        with self._local_cache_lock:
+            if full_key in self._local_cache:
+                entry = self._local_cache[full_key]
+                if entry["expires_at"] is None or time.time() <= entry["expires_at"]:
+                    return False
+            self._local_cache[full_key] = {
+                "value": value,
+                "expires_at": time.time() + ttl if ttl > 0 else None,
+            }
+        return True
+
     def delete(self, key: str) -> bool:
         """
         Delete a key from the cache.

agixt/middleware.py

@@ -570,6 +570,15 @@ async def dispatch(self, request: Request, call_next):
             response = await call_next(request)
             return response
         except Exception as e:
+            # Ignore ClientDisconnect - these are normal when clients close connections
+            from starlette.requests import ClientDisconnect
+
+            if isinstance(e, ClientDisconnect) or "ClientDisconnect" in str(type(e)):
+                self.logger.debug(
+                    f"Client disconnected during {request.method} {request.url.path}"
+                )
+                raise
+
             # Log the error locally
             self.logger.error(
                 f"Unhandled exception in {request.method} {request.url.path}: {e}"