Refactor backend inference and messages into focused services

api/adapters/README.md

@@ -0,0 +1,15 @@
+# Adapter Layer (Strangler Pattern)
+
+This package holds backward-compatibility shims while god-object modules are split into focused services.
+
+## Rules
+- New call sites should import extracted modules directly.
+- Legacy entry points can call extracted modules through adapter wrappers.
+- Use `deprecated_in_favor_of(...)` to log old API usage during migration.
+- Remove adapter modules only after all call sites are migrated and validated.
+
+## Typical Migration Flow
+1. Extract new module and add unit tests.
+2. Update legacy module to delegate through adapter.
+3. Migrate callers incrementally.
+4. Remove adapter when usage drops to zero.

api/adapters/__init__.py

@@ -0,0 +1,39 @@
+"""Compatibility adapters used during strangler-pattern refactors."""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from functools import wraps
+from typing import TypeVar, ParamSpec
+
+from logging_config import logger
+
+P = ParamSpec("P")
+R = TypeVar("R")
+
+
+def deprecated_in_favor_of(new_module: str) -> Callable[[Callable[P, R]], Callable[P, R]]:
+    """Log calls to legacy APIs that are being replaced.
+
+    Parameters
+    ----------
+    new_module:
+        The replacement module path or API identifier.
+    """
+
+    def decorator(func: Callable[P, R]) -> Callable[P, R]:
+        @wraps(func)
+        def wrapper(*args: P.args, **kwargs: P.kwargs) -> R:
+            logger.warning(
+                "deprecated_api_called",
+                old_api=f"{func.__module__}.{func.__name__}",
+                replacement=new_module,
+            )
+            return func(*args, **kwargs)
+
+        return wrapper
+
+    return decorator
+
+
+__all__ = ["deprecated_in_favor_of"]

api/auth.py

@@ -67,16 +67,19 @@ def invalidate_pro_cache(user_id: str) -> None:
         return
     with _PRO_STATE_CACHE_LOCK:
         _PRO_STATE_CACHE.pop(user_id, None)
+    async def _clear() -> None:
+        redis = await safe_redis_call(get_redis, operation="connect")
+        if redis is None:
+            return
+        await safe_redis_call(redis.delete, f"knowbear:user:is_pro:{user_id}", operation="delete")
+
     try:
-        async def _clear() -> None:
-            redis = await safe_redis_call(get_redis, operation="connect")
-            if redis is None:
-                return
-            await safe_redis_call(redis.delete, f"knowbear:user:is_pro:{user_id}", operation="delete")
-        asyncio.create_task(_clear())
-    except Exception:
+        task_coro = _clear()
+        asyncio.create_task(task_coro)
+    except Exception as exc:
+        task_coro.close()
         # Best-effort cache invalidation only.
-        pass
+        logger.debug("auth_invalidate_pro_cache_failed", user_id_hash=anonymize_user_id(user_id), error=str(exc))
 
 @lru_cache(maxsize=1)
 def get_supabase() -> Client | None:
@@ -287,8 +290,12 @@ async def check_is_pro(user_id: str, force_refresh: bool = False) -> bool:
                     _PRO_STATE_CACHE.move_to_end(user_id)
                     _prune_pro_cache_locked(now)
                 return is_pro
-        except Exception:
-            pass
+        except Exception as exc:
+            logger.debug(
+                "auth_pro_cache_redis_read_failed",
+                user_id_hash=anonymize_user_id(user_id),
+                error=str(exc),
+            )
 
     supabase = get_supabase_admin()
     if not supabase:
@@ -312,8 +319,12 @@ async def check_is_pro(user_id: str, force_refresh: bool = False) -> bool:
                     "1" if is_pro else "0",
                     operation="setex",
                 )
-        except Exception:
-            pass
+        except Exception as exc:
+            logger.debug(
+                "auth_pro_cache_redis_write_failed",
+                user_id_hash=anonymize_user_id(user_id),
+                error=str(exc),
+            )
         with _PRO_STATE_CACHE_LOCK:
             _prune_pro_cache_locked(now)
             _PRO_STATE_CACHE[user_id] = (is_pro, now + _pro_cache_ttl_seconds())

api/constants.py

@@ -0,0 +1,15 @@
+"""Centralized constants for quick-win technical debt cleanup."""
+
+REDIS_REST_CALL_TIMEOUT_SECONDS = 0.8
+UPSTASH_HTTP_TIMEOUT_SECONDS = 1.5
+UPSTASH_HTTP_CONNECT_TIMEOUT_SECONDS = 0.75
+
+MESSAGE_GATE_DEFAULT_TIMEOUT_SECONDS = 0.8
+STREAM_IDEMPOTENCY_TTL_MIN_SECONDS = 60
+STREAM_IDEMPOTENCY_TTL_MAX_SECONDS = 120
+STREAM_IDEMPOTENCY_STALE_MIN_SECONDS = 5
+
+RATE_LIMIT_HOURLY_WINDOW_MINUTES = 60
+RATE_LIMIT_HOURLY_WINDOW_SECONDS = RATE_LIMIT_HOURLY_WINDOW_MINUTES * 60
+
+PROVIDER_USAGE_TTL_SECONDS = 86400

api/requirements.txt

@@ -10,11 +10,8 @@ supabase>=2.3.4
 tenacity>=8.2.3
 orjson>=3.9.13
 structlog>=24.1.0
-fastapi-limiter>=0.1.6
-markdown>=3.5.2
 openai>=1.51.0
 sentry-sdk[fastapi]>=2.20.0
 slowapi>=0.1.9
-dodopayments[webhooks]>=1.92.0,<2
 standardwebhooks>=1.0.0
 tiktoken>=0.7.0