Skip empty embeddings in custom DC

nl_server/embeddings.py

@@ -95,3 +95,8 @@ def vector_search(self, queries: List[str], top_k: int) -> SearchVarsResult:
 
     # Turn this into a map:
     return {k: v for k, v in zip(queries, results)}
+
+
+class NoEmbeddingsException(Exception):
+  """Custom exception raised when no embeddings are found in the embeddings csv."""
+  pass

nl_server/flask.py

@@ -58,8 +58,15 @@ def create_app():
     if not lib_utils.is_test_env():
       # Below is a safe check to ensure that the model and embedding is loaded.
       server_config = reg.server_config()
-      idx_type = server_config.default_indexes[0]
-      embeddings = reg.get_index(idx_type)
+
+      def _get_first_available_embeddings():
+        for idx in server_config.default_indexes:
+          embeddings = reg.get_index(idx)
+          if embeddings:
+            return (idx, embeddings)
+        raise ValueError('No embeddings found')
+
+      idx_type, embeddings = _get_first_available_embeddings()
       query = server_config.indexes[idx_type].healthcheck_query
       result = search.search_vars([embeddings], [query]).get(query)
       if not result or not result.svs:

nl_server/registry.py

@@ -23,6 +23,7 @@
 from nl_server.config import StoreType
 from nl_server.embeddings import Embeddings
 from nl_server.embeddings import EmbeddingsModel
+from nl_server.embeddings import NoEmbeddingsException
 from nl_server.model.attribute_model import AttributeModel
 from nl_server.model.create import create_embeddings_model
 from nl_server.ranking import RerankingModel
@@ -108,6 +109,15 @@ def _set_embeddings(self, idx_name: str, idx_info: IndexConfig):
           return
       elif idx_info.store_type == StoreType.VERTEXAI:
         store = VertexAIStore(idx_info)
+    except NoEmbeddingsException as e:
+      if not is_custom_dc():
+        raise e
+      # Some custom DCs may not have SVs or topics in which case no embeddings is a valid condition.
+      # We log a warning and skip it in that case.
+      logging.warning(
+          f'No embeddings found for the following index and will be skipped: {idx_info}'
+      )
+      store = None
     except Exception as e:
       logging.error(f'error loading index {idx_name}: {str(e)} ')
       raise e

nl_server/store/memory.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 """In-memory Embeddings store."""
 
+import csv
 import logging
 from typing import List
 
@@ -25,6 +26,7 @@
 from nl_server.embeddings import EmbeddingsMatch
 from nl_server.embeddings import EmbeddingsResult
 from nl_server.embeddings import EmbeddingsStore
+from nl_server.embeddings import NoEmbeddingsException
 from shared.lib.custom_dc_util import use_anonymous_gcs_client
 from shared.lib.gcs import is_gcs_path
 from shared.lib.gcs import maybe_download
@@ -53,6 +55,10 @@ def __init__(self, idx_info: MemoryIndexConfig) -> None:
           f'"embeddings_path" path must start with `/` or `gs://`: {idx_info.embeddings_path}'
       )
 
+    # Raise no embeddings exception if the embeddings path does not have any embeddings.
+    if _is_csv_empty_or_header_only(embeddings_path):
+      raise NoEmbeddingsException()
+
     self.dataset_embeddings: torch.Tensor = None
     self.dcids: List[str] = []
     self.sentences: List[str] = []
@@ -99,3 +105,27 @@ def vector_search(self, query_embeddings: torch.Tensor,
       results.append(matches)
 
     return results
+
+
+def _is_csv_empty_or_header_only(file_path):
+  """
+  Checks if a CSV file is empty or only contains the header row.
+
+  Args:
+    file_path: The path to the CSV file.
+
+  Returns:
+    True if the CSV file is empty or has only the header, False otherwise.
+  """
+  with open(file_path, 'r', newline='') as csvfile:
+    reader = csv.reader(csvfile)
+    try:
+      # Read the first row (header)
+      next(reader)
+      # Try reading the second row
+      next(reader)
+      # If no exception is raised, there are more rows than just the header
+      return False
+    except StopIteration:
+      # StopIteration is raised if there are no more rows to read
+      return True