fix(types): keep TagRateLimitEntry validation messages proxy-agnostic

litellm/types/router.py is imported by plain SDK users, not just the proxy;
the previous ValueError messages for limit/key_ttl_seconds explained the
proxy rate-limit hook's internal admission mechanics (atomic
check-and-increment, read-only tokens/dollars check, cache TTL rollover),
leaking implementation details across the SDK/proxy boundary. Move that
mechanistic reasoning into code comments for future maintainers and keep
the raised messages generic, per Greptile's finding on PR #38289.

Adds regression tests asserting the three affected validators reject their
invalid inputs without leaking proxy-internal enforcement jargon.

litellm/types/router.py

@@ -237,18 +237,15 @@ class TagRateLimitEntry(BaseModel):
         # defeats the entry; reject it at config load time instead.
         if math.isnan(self.limit):
             raise ValueError("limit must not be NaN")
+        # Positive infinity never rejects the checks that gate this limit; negative
+        # infinity always does. Both silently defeat the entry.
         if math.isinf(self.limit):
-            raise ValueError(
-                "limit must be finite -- positive infinity makes admission never reject (current + increment "
-                "> limit is always false), negative infinity makes it always reject every tagged request"
-            )
+            raise ValueError("limit must be finite")
+        # Zero or negative makes every check that gates this limit either always
+        # reject or never admit, silently blocking or admitting all matching traffic
+        # instead of the likely intended config.
         if self.limit <= 0:
-            raise ValueError(
-                "limit must be a positive number -- zero or negative makes the atomic requests/concurrency "
-                "check (current + increment > limit) reject every admission and the read-only tokens/dollars "
-                "check (current < limit) never admit, silently blocking all matching traffic instead of the "
-                "likely intended config"
-            )
+            raise ValueError("limit must be a positive number")
         return self
 
     @model_validator(mode="after")
@@ -261,11 +258,10 @@ class TagRateLimitEntry(BaseModel):
     def _validate_key_ttl_seconds(self) -> "TagRateLimitEntry":
         if self.key_ttl_seconds is not None and self.key_ttl_seconds <= 0:
             raise ValueError("key_ttl_seconds must be a positive integer when set")
+        # A shorter TTL than period_seconds expires the counter before its period
+        # elapses, letting it reset early and exceed the limit.
         if self.key_ttl_seconds is not None and self.key_ttl_seconds < self.period_seconds:
-            raise ValueError(
-                "key_ttl_seconds must be at least period_seconds when set -- a shorter TTL expires the "
-                "counter before its window rolls over, letting tagged traffic reset to zero and exceed the limit"
-            )
+            raise ValueError("key_ttl_seconds must be at least period_seconds when set")
         return self
 
     @model_validator(mode="after")

tests/test_litellm/types/test_router.py

@@ -5,6 +5,7 @@ from litellm.types.router import (
     Deployment,
     LiteLLM_Params,
     ModelInfo,
+    TagRateLimitEntry,
 )
 from litellm.types.utils import CustomPricingLiteLLMParams, MirroredPricingParams
 
@@ -89,3 +90,47 @@ def test_pricing_strings_are_coerced_to_float():
 def test_invalid_pricing_is_rejected():
     with pytest.raises(ValueError, match='validation error for ModelInfo'):
         ModelInfo(id="x", input_cost_per_token="free")
+
+
+# litellm/types/router.py is imported by plain SDK users, not just the proxy, so a
+# TagRateLimitEntry validation error should read as a generic config-validation
+# message and not describe the proxy rate-limit hook's internal admission mechanics.
+_INTERNAL_ENFORCEMENT_JARGON = (
+    "admission",
+    "tagged request",
+    "tagged traffic",
+    "check-and-increment",
+    "read-only",
+    "atomic",
+    "window rolls over",
+)
+
+
+def _assert_message_has_no_internal_jargon(excinfo: pytest.ExceptionInfo) -> None:
+    message = str(excinfo.value).lower()
+    leaked = [term for term in _INTERNAL_ENFORCEMENT_JARGON if term in message]
+    assert not leaked, f"validation message leaked internal enforcement jargon: {leaked}"
+
+
+def test_limit_infinite_rejected_without_internal_enforcement_jargon():
+    with pytest.raises(ValueError, match="validation error for TagRateLimitEntry") as excinfo:
+        TagRateLimitEntry(name="daily", tag_id="end_user_id", limit=float("inf"), period_seconds=60)
+    _assert_message_has_no_internal_jargon(excinfo)
+
+
+def test_limit_non_positive_rejected_without_internal_enforcement_jargon():
+    with pytest.raises(ValueError, match="validation error for TagRateLimitEntry") as excinfo:
+        TagRateLimitEntry(name="daily", tag_id="end_user_id", limit=0, period_seconds=60)
+    _assert_message_has_no_internal_jargon(excinfo)
+
+
+def test_key_ttl_seconds_shorter_than_period_rejected_without_internal_enforcement_jargon():
+    with pytest.raises(ValueError, match="validation error for TagRateLimitEntry") as excinfo:
+        TagRateLimitEntry(
+            name="daily",
+            tag_id="end_user_id",
+            limit=500,
+            period_seconds=86400,
+            key_ttl_seconds=60,
+        )
+    _assert_message_has_no_internal_jargon(excinfo)