chore(config): ⚙️ enable conditional instantiation of API call persistence

This change introduces a global runtime flag to gate the creation of detailed, per-request log directories within the Gemini CLI provider plugin.

By default, the internal file logger for Gemini requests is now disabled, preventing unnecessary I/O operations and resource consumption from creating unique log files for every transaction. The system only initializes the logging infrastructure and interacts with the file system if `enable_request_logging` is explicitly set to true upon client initialization.

- Introduce `enable_request_logging` parameter in RotatingClient.
- Condition the file logger's directory creation on the new flag.
- Suppress verbose console debugging output that duplicates file logging.

src/proxy_app/main.py

@@ -282,9 +282,10 @@ async def lifespan(app: FastAPI):
         api_keys=api_keys,
         oauth_credentials=oauth_credentials, # Pass OAuth config
         configure_logging=True,
-        litellm_provider_params=litellm_provider_params, # [NEW]
+        litellm_provider_params=litellm_provider_params,
         ignore_models=ignore_models,
-        whitelist_models=whitelist_models
+        whitelist_models=whitelist_models,
+        enable_request_logging=ENABLE_REQUEST_LOGGING
     )
     client.background_refresher.start() # Start the background task
     app.state.rotating_client = client

src/rotator_library/client.py

@@ -47,9 +47,10 @@ class RotatingClient:
         configure_logging: bool = True,
         global_timeout: int = 30,
         abort_on_callback_error: bool = True,
-        litellm_provider_params: Optional[Dict[str, Any]] = None, # [NEW]
+        litellm_provider_params: Optional[Dict[str, Any]] = None,
         ignore_models: Optional[Dict[str, List[str]]] = None,
-        whitelist_models: Optional[Dict[str, List[str]]] = None
+        whitelist_models: Optional[Dict[str, List[str]]] = None,
+        enable_request_logging: bool = False
     ):
         os.environ["LITELLM_LOG"] = "ERROR"
         litellm.set_verbose = False
@@ -93,6 +94,7 @@ class RotatingClient:
         self.litellm_provider_params = litellm_provider_params or {}
         self.ignore_models = ignore_models or {}
         self.whitelist_models = whitelist_models or {}
+        self.enable_request_logging = enable_request_logging
 
     def _is_model_ignored(self, provider: str, model_id: str) -> bool:
         """
@@ -448,6 +450,7 @@ class RotatingClient:
                 if provider_plugin and provider_plugin.has_custom_logic():
                     lib_logger.debug(f"Provider '{provider}' has custom logic. Delegating call.")
                     litellm_kwargs["credential_identifier"] = current_cred
+                    litellm_kwargs["enable_request_logging"] = self.enable_request_logging
 
                     # Check body first for custom_reasoning_budget
                     if "custom_reasoning_budget" in kwargs:
@@ -668,6 +671,7 @@ class RotatingClient:
                     if provider_plugin and provider_plugin.has_custom_logic():
                         lib_logger.debug(f"Provider '{provider}' has custom logic. Delegating call.")
                         litellm_kwargs["credential_identifier"] = current_cred
+                        litellm_kwargs["enable_request_logging"] = self.enable_request_logging
                         
                         for attempt in range(self.max_retries):
                             try:

src/rotator_library/providers/gemini_cli_provider.py

@@ -23,7 +23,11 @@ GEMINI_CLI_LOGS_DIR = LOGS_DIR / "gemini_cli_logs"
 
 class _GeminiCliFileLogger:
     """A simple file logger for a single Gemini CLI transaction."""
-    def __init__(self, model_name: str):
+    def __init__(self, model_name: str, enabled: bool = True):
+        self.enabled = enabled
+        if not self.enabled:
+            return
+
         timestamp = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
         request_id = str(uuid.uuid4())
         # Sanitize model name for directory
@@ -31,7 +35,6 @@ class _GeminiCliFileLogger:
         self.log_dir = GEMINI_CLI_LOGS_DIR / f"{timestamp}_{safe_model_name}_{request_id}"
         try:
             self.log_dir.mkdir(parents=True, exist_ok=True)
-            self.enabled = True
         except Exception as e:
             lib_logger.error(f"Failed to create Gemini CLI log directory: {e}")
             self.enabled = False
@@ -519,6 +522,7 @@ class GeminiCliProvider(GeminiAuthBase, ProviderInterface):
     async def acompletion(self, client: httpx.AsyncClient, **kwargs) -> Union[litellm.ModelResponse, AsyncGenerator[litellm.ModelResponse, None]]:
         model = kwargs["model"]
         credential_path = kwargs.pop("credential_identifier")
+        enable_request_logging = kwargs.pop("enable_request_logging", False)
         
         async def do_call():
             # Get auth header once, it's needed for the request anyway
@@ -534,7 +538,10 @@ class GeminiCliProvider(GeminiAuthBase, ProviderInterface):
             model_name = model.split('/')[-1].replace(':thinking', '')
             
             # [NEW] Create a dedicated file logger for this request
-            file_logger = _GeminiCliFileLogger(model_name=model_name)
+            file_logger = _GeminiCliFileLogger(
+                model_name=model_name,
+                enabled=enable_request_logging
+            )
 
             gen_config = {
                 "maxOutputTokens": kwargs.get("max_tokens", 64000), # Increased default
@@ -578,7 +585,7 @@ class GeminiCliProvider(GeminiAuthBase, ProviderInterface):
                     request_payload["request"]["toolConfig"] = tool_config
             
             # Log the final payload for debugging and to the dedicated file
-            lib_logger.debug(f"Gemini CLI Request Payload: {json.dumps(request_payload, indent=2)}")
+            #lib_logger.debug(f"Gemini CLI Request Payload: {json.dumps(request_payload, indent=2)}")
             file_logger.log_request(request_payload)
             
             url = f"{CODE_ASSIST_ENDPOINT}:streamGenerateContent"