potential fix for usage limits

backend/ee/onyx/server/tenant_usage_limits.py

@@ -32,12 +32,12 @@ def fetch_usage_limit_overrides() -> dict[str, TenantUsageLimitOverrides]:
         response = requests.get(url, headers=headers, timeout=30)
         response.raise_for_status()
 
-        data = response.json()
-        tenant_overrides = data.get("tenant_overrides", {})
+        tenant_overrides = response.json()
 
         # Parse each tenant's overrides
         result: dict[str, TenantUsageLimitOverrides] = {}
-        for tenant_id, override_data in tenant_overrides.items():
+        for override_data in tenant_overrides:
+            tenant_id = override_data["tenant_id"]
             try:
                 result[tenant_id] = TenantUsageLimitOverrides(**override_data)
             except Exception as e:
@@ -66,6 +66,9 @@ def load_usage_limit_overrides() -> dict[str, TenantUsageLimitOverrides]:
     logger.info("Loading tenant usage limit overrides from control plane...")
     overrides = fetch_usage_limit_overrides()
     _tenant_usage_limit_overrides = overrides
+    logger.debug(
+        "Loaded usage limit overrides for %s tenants: %s", len(overrides), overrides
+    )
 
     if overrides:
         logger.info(f"Loaded usage limit overrides for {len(overrides)} tenants")

backend/onyx/server/tenant_usage_limits.py

@@ -34,6 +34,8 @@ class TenantUsageLimitOverrides(BaseModel):
     - Field set to a positive integer: Use that specific limit
     """
 
+    tenant_id: str | None = None
+
     llm_cost_cents_trial: int | None = None
     llm_cost_cents_paid: int | None = None
     chunks_indexed_trial: int | None = None

backend/onyx/server/usage_limits.py

@@ -90,12 +90,20 @@ def _get_tenant_override(tenant_id: str, field_name: str) -> int | None:
             "onyx.server.tenant_usage_limits", "get_tenant_usage_limit_overrides"
         )
         overrides: TenantUsageLimitOverrides | None = get_overrides_fn(tenant_id)
+        logger.debug("Got tenant overrides for %s: %s", tenant_id, overrides)
         if overrides is not None:
             # Get the field value - None means not set, use default
-            return getattr(overrides, field_name, None)
+            fv = getattr(overrides, field_name, None)
+            logger.debug(
+                "Using tenant override for %s.%s: %s", tenant_id, field_name, fv
+            )
+            return fv
     except Exception:
-        # EE module not available or error - fall back to defaults
-        pass
+        logger.exception(
+            "Error getting tenant override for %s.%s falling back to defaults",
+            tenant_id,
+            field_name,
+        )
     return None
 
 
@@ -151,10 +159,18 @@ def get_limit_for_usage_type(
     """
 
     field_name, default_value = _FIELD_AND_DEFAULT[usage_type][is_trial]
+    logger.debug("Getting limit for %s.%s: %s", usage_type, is_trial, default_value)
     if tenant_id:
+        logger.debug("Getting tenant override for %s.%s", tenant_id, field_name)
         override = _get_tenant_override(tenant_id, field_name)
         if override is not None:
+            logger.debug(
+                "Using tenant override for %s.%s: %s", tenant_id, field_name, override
+            )
             return override
+    logger.debug(
+        "Using default value for %s.%s: %s", usage_type, is_trial, default_value
+    )
     return default_value
 
 
@@ -215,6 +231,7 @@ def check_usage_and_raise(
 
     is_trial = is_tenant_on_trial_fn(tenant_id)
     limit = get_limit_for_usage_type(usage_type, is_trial, tenant_id)
+    logger.debug("Checking usage limit for %s.%s: %s", usage_type, is_trial, limit)
 
     # NO_LIMIT means this tenant has unlimited usage for this type
     if limit == NO_LIMIT:

backend/tests/external_dependency_unit/llm/test_prompt_caching.py

@@ -145,6 +145,7 @@ def _validate_vertex_credentials_file(credentials_path: Path) -> None:
         ) from exc
 
 
+@pytest.mark.skip(reason="OpenAI prompt caching is unreliable")
 @pytest.mark.skipif(
     not os.environ.get("OPENAI_API_KEY"),
     reason="OpenAI API key not available",