fix(guardrails/purview): raise_for_status on HTTP errors, cap scope cache, reuse executor

litellm/proxy/guardrails/guardrail_hooks/microsoft_purview/base.py

@@ -1,6 +1,7 @@
 import time
 import uuid
-from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
+from collections import OrderedDict
+from typing import TYPE_CHECKING, Any, Dict, List, MutableMapping, Optional, Tuple
 
 from litellm._logging import verbose_proxy_logger
 from litellm.llms.custom_httpx.http_handler import (
@@ -56,7 +57,9 @@ class PurviewGuardrailBase:
         self._token_cache: Optional[Tuple[str, float]] = None
 
         # Protection scope cache: user_id -> (etag, scope_response, fetched_at)
-        self._scope_cache: Dict[str, Tuple[str, Dict, float]] = {}
+        # Capped at 1000 entries (LRU eviction) to avoid unbounded growth.
+        self._scope_cache: MutableMapping[str, Tuple[str, Dict, float]] = OrderedDict()
+        self._scope_cache_maxsize = 1000
 
     # ------------------------------------------------------------------
     # OAuth2 token management
@@ -80,6 +83,7 @@ class PurviewGuardrailBase:
             data=data,
             headers={"Content-Type": "application/x-www-form-urlencoded"},
         )
+        response.raise_for_status()
         token_data = response.json()
         access_token = token_data["access_token"]
         expires_in = int(token_data.get("expires_in", 3599))
@@ -116,6 +120,7 @@ class PurviewGuardrailBase:
         response = await self.async_handler.post(
             url=url, headers=headers, json=json_body
         )
+        response.raise_for_status()
         response_json: Dict[str, Any] = response.json()
         response_headers = dict(response.headers)
         verbose_proxy_logger.debug("Purview Graph response: %s", response_json)
@@ -157,6 +162,9 @@ class PurviewGuardrailBase:
         etag = response_headers.get("etag", response_headers.get("ETag", ""))
 
         self._scope_cache[user_id] = (etag, response_json, now)
+        # Evict oldest entry when cache exceeds max size.
+        while len(self._scope_cache) > self._scope_cache_maxsize:
+            self._scope_cache.popitem(last=False)  # type: ignore[attr-defined]
         return etag, response_json
 
     # ------------------------------------------------------------------

litellm/proxy/guardrails/guardrail_hooks/microsoft_purview/purview_dlp.py

@@ -8,6 +8,7 @@ Supports three modes:
 """
 
 import asyncio
+from concurrent.futures import ThreadPoolExecutor
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type, Union
 
@@ -75,6 +76,7 @@ class MicrosoftPurviewDLPGuardrail(PurviewGuardrailBase, CustomGuardrail):
         )
         self._logging_only = logging_only
         self.guardrail_provider = "microsoft_purview"
+        self._executor = ThreadPoolExecutor(max_workers=1)
         verbose_proxy_logger.info(
             "Initialized Microsoft Purview DLP Guardrail: %s (logging_only=%s)",
             guardrail_name,
@@ -229,7 +231,6 @@ class MicrosoftPurviewDLPGuardrail(PurviewGuardrailBase, CustomGuardrail):
         self, kwargs: dict, result: Any, call_type: str
     ) -> Tuple[dict, Any]:
         """Sync wrapper for async_logging_hook (follows Presidio pattern)."""
-        from concurrent.futures import ThreadPoolExecutor
 
         def run_in_new_loop() -> Tuple[dict, Any]:
             new_loop = asyncio.new_event_loop()
@@ -246,9 +247,8 @@ class MicrosoftPurviewDLPGuardrail(PurviewGuardrailBase, CustomGuardrail):
 
         try:
             _ = asyncio.get_running_loop()
-            with ThreadPoolExecutor(max_workers=1) as executor:
-                future = executor.submit(run_in_new_loop)
-                return future.result()
+            future = self._executor.submit(run_in_new_loop)
+            return future.result()
         except RuntimeError:
             return run_in_new_loop()