Simplify solution

Remove result caching as its not the core part of the problem at hand.

litellm/proxy/guardrails/guardrail_hooks/presidio.py

@@ -121,9 +121,6 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
         # Loop-bound session cache for background threads
         self._loop_sessions: Dict[asyncio.AbstractEventLoop, aiohttp.ClientSession] = {}
 
-        # Result cache to avoid redundant network calls
-        self.pii_cache = DualCache()
-
         if mock_testing is True:  # for testing purposes only
             return
 
@@ -488,13 +485,6 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
         """
         Calls Presidio Analyze + Anonymize endpoints for PII Analysis + Masking
         """
-        # Cache check
-        cache_key = f"presidio:pii:{text}:{output_parse_pii}:{presidio_config}:{self.pii_entities_config}"
-        cached_result = self.pii_cache.get_cache(cache_key)
-        if cached_result is not None:
-            verbose_proxy_logger.debug("PII Cache Hit for text: %s", text)
-            return cached_result
-
         start_time = datetime.now()
         analyze_results: Optional[Union[List[PresidioAnalyzeResponseItem], Dict]] = None
         status: GuardrailStatus = "success"
@@ -532,7 +522,6 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
                     output_parse_pii=output_parse_pii,
                     masked_entity_count=masked_entity_count,
                 )
-                self.pii_cache.set_cache(cache_key, anonymized_text)
                 return anonymized_text
             return redacted_text["text"]
         except Exception as e:

tests/test_presidio_latency.py

@@ -97,43 +97,3 @@ async def test_optimization_presidio_avoid_recognizer_reloads_on_server():
 
     # Optimization Check: payload should NOT contain the redundant recognizers
     assert "ad_hoc_recognizers" not in payload
-
-@pytest.mark.asyncio
-async def test_optimization_presidio_pii_caching_skips_network_calls():
-    """
-    OPTIMIZATION VERIFICATION:
-    Verify that PII results are cached in local memory.
-    Identical text should return instantly from cache without making a 
-    network round-trip to the Presidio service.
-    """
-    presidio = _OPTIONAL_PresidioPIIMasking(
-        mock_testing=True,
-        presidio_analyzer_api_base="http://mock-analyzer",
-        presidio_anonymizer_api_base="http://mock-anonymizer"
-    )
-    
-    # Mock network calls
-    presidio.analyze_text = MagicMock(return_value=asyncio.Future())
-    presidio.analyze_text.return_value.set_result([])
-    presidio.anonymize_text = MagicMock(return_value=asyncio.Future())
-    presidio.anonymize_text.return_value.set_result("redacted text")
-    
-    # First call - Must hit the mock (Network)
-    result1 = await presidio.check_pii(
-        text="repetitive call center text",
-        output_parse_pii=False,
-        presidio_config=None,
-        request_data={}
-    )
-    assert result1 == "redacted text"
-    assert presidio.analyze_text.call_count == 1
-    
-    # Second call - Must hit the CACHE (No Network)
-    result2 = await presidio.check_pii(
-        text="repetitive call center text",
-        output_parse_pii=False,
-        presidio_config=None,
-        request_data={}
-    )
-    assert result2 == "redacted text"
-    assert presidio.analyze_text.call_count == 1 # Counter should NOT increase