fix: Avoid catching non-exception; skip retry for ValueErrors from VertexAI (#470)

The `VertexAIVectorizer` attempts to handle an error from Google that,
it turns out, is not an exception despite being in the Google exceptions
module. This PR removes this catch, and adjusts retry logic to capture
the correct set of exceptions for VertexAI use. In particular, this
fixes an issue where using an invalid model name would break with an
unclear error - this now breaks as expected, by reporting that the
provided model name was not found.

fixes #469

Author: vishal-bala

redisvl/utils/vectorize/vertexai.py

@@ -6,15 +6,11 @@
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 from tenacity.retry import retry_if_not_exception_type
 
-from redisvl.utils.utils import lazy_import
-
 if TYPE_CHECKING:
     from redisvl.extensions.cache.embeddings.embeddings import EmbeddingsCache
 
 from redisvl.utils.vectorize.base import BaseVectorizer
 
-InvalidArgument = lazy_import("google.api_core.exceptions.InvalidArgument")
-
 
 class VertexAIVectorizer(BaseVectorizer):
     """The VertexAIVectorizer uses Google's VertexAI embedding model
@@ -233,7 +229,7 @@ def _set_model_dims(self) -> int:
     @retry(
         wait=wait_random_exponential(min=1, max=60),
         stop=stop_after_attempt(6),
-        retry=retry_if_not_exception_type(TypeError),
+        retry=retry_if_not_exception_type((TypeError, ValueError)),
     )
     def _embed(self, content: Any, **kwargs) -> List[float]:
         """
@@ -286,15 +282,13 @@ def _embed(self, content: Any, **kwargs) -> List[float]:
             else:
                 return self._client.get_embeddings([content], **kwargs)[0].values
 
-        except InvalidArgument as e:
-            raise TypeError(f"Invalid input for embedding: {str(e)}") from e
         except Exception as e:
             raise ValueError(f"Embedding input failed: {e}")
 
     @retry(
         wait=wait_random_exponential(min=1, max=60),
         stop=stop_after_attempt(6),
-        retry=retry_if_not_exception_type(TypeError),
+        retry=retry_if_not_exception_type((TypeError, ValueError)),
     )
     def _embed_many(
         self, contents: List[str], batch_size: int = 10, **kwargs
@@ -311,7 +305,6 @@ def _embed_many(
             List[List[float]]: List of vector embeddings as lists of floats
 
         Raises:
-            TypeError: If contents is not a list of strings
             ValueError: If embedding fails
         """
         if self.is_multimodal:
@@ -329,8 +322,6 @@ def _embed_many(
                 response = self._client.get_embeddings(batch, **kwargs)
                 embeddings.extend([r.values for r in response])
             return embeddings
-        except InvalidArgument as e:
-            raise TypeError(f"Invalid input for embedding: {str(e)}") from e
         except Exception as e:
             raise ValueError(f"Embedding texts failed: {e}")
 

tests/integration/test_redis_cluster_support.py

@@ -89,6 +89,7 @@ def test_search_index_cluster_info(redis_cluster_url):
     finally:
         index.delete(drop=True)
 
+
 @pytest.mark.requires_cluster
 @pytest.mark.asyncio
 async def test_async_search_index_cluster_info(redis_cluster_url):
@@ -110,6 +111,7 @@ async def test_async_search_index_cluster_info(redis_cluster_url):
         await index.delete(drop=True)
         await client.aclose()
 
+
 @pytest.mark.requires_cluster
 @pytest.mark.asyncio
 async def test_async_search_index_client(redis_cluster_url):

tests/integration/test_search_index.py

@@ -304,6 +304,7 @@ def test_search_index_delete(index):
     assert not index.exists()
     assert index.name not in convert_bytes(index.client.execute_command("FT._LIST"))
 
+
 @pytest.mark.parametrize("num_docs", [0, 1, 5, 10, 2042])
 def test_search_index_clear(index, num_docs):
     index.create(overwrite=True, drop=True)