[ENH]: Error if source_key set but no ef (#5751)

## Description of changes

_Summarize the changes made by this PR._

- Improvements & Bug fixes
- Errors if source_key is set but ef is not supplied. Both client and
server side
- New functionality
  - ...

## Test plan

_How are these changes tested?_
Added unit-test
- [x] Tests pass locally with `pytest` for python, `yarn test` for js,
`cargo test` for rust

## Migration plan
None

## Observability plan
None

## Documentation Changes
None

Author: sanketkedia

chromadb/api/types.py

@@ -2044,6 +2044,20 @@ 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 either embedding_function or bm25
+        must be provided since there is no default embedding function.
+        Raises ValueError otherwise.
+        """
+        if (config.source_key is not None
+                and config.embedding_function is None
+                and config.bm25 is not True):
+            raise ValueError(
+                f"If source_key is provided then either embedding_function or bm25 must 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 +2066,7 @@ 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 +2111,8 @@ 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

@@ -246,7 +246,7 @@ def test_chained_create_and_delete_operations(self) -> None:
         # 1. Create sparse vector index on "embeddings_key"
         # 2. Disable string inverted index on "text_key_1"
         # 3. Disable string inverted index on "text_key_2"
-        sparse_config = SparseVectorIndexConfig(source_key="raw_text")
+        sparse_config = SparseVectorIndexConfig(source_key="raw_text", bm25=True)
         string_config = StringInvertedIndexConfig()
 
         result = (
@@ -1268,7 +1268,7 @@ def test_multiple_index_types_on_same_key(self) -> None:
         schema = Schema()
 
         # Enable sparse vector on "multi_field"
-        sparse_config = SparseVectorIndexConfig(source_key="source")
+        sparse_config = SparseVectorIndexConfig(source_key="source", bm25=True)
         schema.create_index(config=sparse_config, key="multi_field")
 
         # Also enable string_inverted_index on the same key
@@ -1454,7 +1454,7 @@ def test_multiple_serialize_deserialize_roundtrips(self) -> None:
                 hnsw=hnsw_config
             )
             original.create_index(config=vector_config)
-            original.create_index(config=SparseVectorIndexConfig(source_key="text"), key="embeddings")
+            original.create_index(config=SparseVectorIndexConfig(source_key="text", bm25=True), key="embeddings")
             original.delete_index(config=StringInvertedIndexConfig(), key="tags")
 
             # First roundtrip
@@ -1514,7 +1514,7 @@ def test_many_keys_stress(self) -> None:
             key_name = f"field_{i}"
             if i == 0:
                 # Enable sparse vector on ONE key only
-                schema.create_index(config=SparseVectorIndexConfig(source_key=f"source_{i}"), key=key_name)
+                schema.create_index(config=SparseVectorIndexConfig(source_key=f"source_{i}", bm25=True), key=key_name)
             elif i % 2 == 1:
                 # Disable string inverted index
                 schema.delete_index(config=StringInvertedIndexConfig(), key=key_name)
@@ -1578,7 +1578,7 @@ def test_chained_operations(self) -> None:
 
         # Chain multiple operations
         result = (schema
-                  .create_index(config=SparseVectorIndexConfig(source_key="text"), key="field1")
+                  .create_index(config=SparseVectorIndexConfig(source_key="text", bm25=True), key="field1")
                   .delete_index(config=StringInvertedIndexConfig(), key="field2")
                   .delete_index(config=StringInvertedIndexConfig(), key="field3")
                   .delete_index(config=IntInvertedIndexConfig(), key="field4"))
@@ -1820,7 +1820,7 @@ def test_keys_have_independent_configs(self) -> None:
         schema = Schema()
 
         # Enable sparse vector on a key - it gets exactly what we specify
-        sparse_config = SparseVectorIndexConfig(source_key="default_source")
+        sparse_config = SparseVectorIndexConfig(source_key="default_source", bm25=True)
         schema.create_index(config=sparse_config, key="field1")
 
         # Verify field1 has the sparse vector with the specified source_key
@@ -1907,7 +1907,7 @@ def test_key_specific_overrides_are_independent(self) -> None:
         schema = Schema()
 
         # Create sparse vector on one key and string indexes on others
-        schema.create_index(config=SparseVectorIndexConfig(source_key="source_a"), key="key_a")
+        schema.create_index(config=SparseVectorIndexConfig(source_key="source_a", bm25=True), key="key_a")
         schema.create_index(config=StringInvertedIndexConfig(), key="key_b")
         schema.create_index(config=StringInvertedIndexConfig(), key="key_c")
 
@@ -1992,7 +1992,7 @@ def test_partial_override_fills_from_defaults(self) -> None:
         schema = Schema()
 
         # Enable sparse vector on a key
-        schema.create_index(config=SparseVectorIndexConfig(source_key="my_source"), key="multi_index_field")
+        schema.create_index(config=SparseVectorIndexConfig(source_key="my_source", bm25=True), key="multi_index_field")
 
         # This key now has sparse_vector overridden, but string, int, etc. should still follow global defaults
         field = schema.keys["multi_index_field"]
@@ -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

@@ -2359,10 +2359,10 @@ 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 {