[ENH]: Error if source_key set but no ef

chromadb/api/types.py

@@ -2044,6 +2044,17 @@ def _validate_single_sparse_vector_index(self, key: str) -> None:
                             f"Only one sparse vector index is allowed per collection."
                         )
 
+    def _validate_sparse_vector_config(self, config: SparseVectorIndexConfig) -> None:
+        """
+        Validate that if source_key is provided then ef is also provided since
+        there is no default embedding function. Raises ValueError otherwise.
+        """
+        if config.source_key is not None and config.embedding_function is None:
+            raise ValueError(
+                f"If source_key is provided then embedding_function must also be provided "
+                f"since there is no default embedding function. Config: {config}"
+            )
+
     def _set_index_for_key(self, key: str, config: IndexConfig, enabled: bool) -> None:
         """Set an index configuration for a specific key."""
         config_name = self._get_config_class_name(config)
@@ -2052,6 +2063,7 @@ def _set_index_for_key(self, key: str, config: IndexConfig, enabled: bool) -> No
         # Do this BEFORE creating the key entry
         if config_name == "SparseVectorIndexConfig" and enabled:
             self._validate_single_sparse_vector_index(key)
+            self._validate_sparse_vector_config(cast(SparseVectorIndexConfig, config))
 
         if key not in self.keys:
             self.keys[key] = ValueTypes()
@@ -2096,6 +2108,8 @@ def _enable_all_indexes_for_key(self, key: str) -> None:
         if key not in self.keys:
             self.keys[key] = ValueTypes()
 
+        self._validate_single_sparse_vector_index(key)
+
         # Enable all index types with default configs
         self.keys[key].string = StringValueType(
             fts_index=FtsIndexType(enabled=True, config=FtsIndexConfig()),

chromadb/test/api/test_schema.py

@@ -2337,3 +2337,23 @@ def test_config_source_key_validates_special_keys() -> None:
     # Regular keys (no #) are allowed
     config6 = SparseVectorIndexConfig(source_key="my_field")
     assert config6.source_key == "my_field"
+
+
+def test_sparse_vector_config_requires_ef_with_source_key() -> None:
+    """Test that SparseVectorIndexConfig raises ValueError when source_key is provided without embedding_function."""
+    schema = Schema()
+
+    # Attempt to create sparse vector index with source_key but no embedding_function
+    with pytest.raises(ValueError) as exc_info:
+        schema.create_index(
+            key="invalid_sparse",
+            config=SparseVectorIndexConfig(
+                source_key="text_field",
+                # No embedding_function provided - should raise ValueError
+            ),
+        )
+
+    # Verify the error message mentions both source_key and embedding_function
+    error_msg = str(exc_info.value)
+    assert "source_key" in error_msg.lower()
+    assert "embedding_function" in error_msg.lower()

chromadb/test/api/test_schema_e2e.py

@@ -2355,10 +2355,10 @@ def test_modify_collection_preserves_other_schema_fields(client: ClientAPI) -> N
     collection_refreshed = client.get_collection(collection_name)
     refreshed_schema = collection_refreshed.schema
     assert refreshed_schema is not None
-    
+
     # Verify vector index was updated on server
     assert refreshed_schema.defaults.float_list.vector_index.config.spann.search_nprobe == 128  # type: ignore
-    
+
     # Verify other value types are still intact on server
     assert refreshed_schema.defaults.string is not None
     assert refreshed_schema.defaults.string.string_inverted_index is not None

rust/types/src/validators.rs

@@ -281,6 +281,11 @@ pub fn validate_schema(schema: &Schema) -> Result<(), ValidationError> {
                             .into(),
                     ));
                 }
+                if svit.config.source_key.is_some() && svit.config.embedding_function.is_none() {
+                    return Err(ValidationError::new("schema").with_message(
+                        "If source_key is provided then embedding_function must also be provided since there is no default embedding function.".into(),
+                    ));
+                }
             }
             // Validate source_key for sparse vector index
             if let Some(source_key) = &svit.config.source_key {