perf: Add init and cleanup of long lived resources (feast-dev#4642)

* rebase

Signed-off-by: Rob Howley <[email protected]>

* offline store init doesnt make sense

Signed-off-by: Rob Howley <[email protected]>

* dont init or close

Signed-off-by: Rob Howley <[email protected]>

* update test to handle event loop for dynamo case

Signed-off-by: Rob Howley <[email protected]>

* use run util complete

Signed-off-by: Rob Howley <[email protected]>

* fix: spelling sigh

Signed-off-by: Rob Howley <[email protected]>

* run integration test as async since that is default for read

Signed-off-by: Rob Howley <[email protected]>

* add pytest async to ci reqs

Signed-off-by: Rob Howley <[email protected]>

* be safe w cleanup in test fixture

Signed-off-by: Rob Howley <[email protected]>

* be safe w cleanup in test fixture

Signed-off-by: Rob Howley <[email protected]>

* update pytest ini

Signed-off-by: Rob Howley <[email protected]>

* not in a finally

Signed-off-by: Rob Howley <[email protected]>

* remove close

Signed-off-by: Rob Howley <[email protected]>

* test client is a lifespan aware context manager

Signed-off-by: Rob Howley <[email protected]>

---------

Signed-off-by: Rob Howley <[email protected]>

Author: robhowley

sdk/python/feast/feature_server.py

@@ -100,9 +100,11 @@ def async_refresh():
 
     @asynccontextmanager
     async def lifespan(app: FastAPI):
+        await store.initialize()
         async_refresh()
         yield
         stop_refresh()
+        await store.close()
 
     app = FastAPI(lifespan=lifespan)
 

sdk/python/feast/feature_store.py

@@ -2167,6 +2167,14 @@ def list_saved_datasets(
             self.project, allow_cache=allow_cache, tags=tags
         )
 
+    async def initialize(self) -> None:
+        """Initialize long-lived clients and/or resources needed for accessing datastores"""
+        await self._get_provider().initialize(self.config)
+
+    async def close(self) -> None:
+        """Cleanup any long-lived clients and/or resources"""
+        await self._get_provider().close()
+
 
 def _print_materialization_log(
     start_date, end_date, num_feature_views: int, online_store: str

sdk/python/feast/infra/online_stores/dynamodb.py

@@ -12,11 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import asyncio
+import contextlib
 import itertools
 import logging
 from datetime import datetime
 from typing import Any, Callable, Dict, List, Literal, Optional, Sequence, Tuple, Union
 
+from aiobotocore.config import AioConfig
 from pydantic import StrictBool, StrictStr
 
 from feast import Entity, FeatureView, utils
@@ -75,6 +77,9 @@ class DynamoDBOnlineStoreConfig(FeastConfigBaseModel):
     session_based_auth: bool = False
     """AWS session based client authentication"""
 
+    max_pool_connections: int = 10
+    """Max number of connections for async Dynamodb operations"""
+
 
 class DynamoDBOnlineStore(OnlineStore):
     """
@@ -87,7 +92,14 @@ class DynamoDBOnlineStore(OnlineStore):
 
     _dynamodb_client = None
     _dynamodb_resource = None
-    _aioboto_session = None
+
+    async def initialize(self, config: RepoConfig):
+        await _get_aiodynamodb_client(
+            config.online_store.region, config.online_store.max_pool_connections
+        )
+
+    async def close(self):
+        await _aiodynamodb_close()
 
     @property
     def async_supported(self) -> SupportedAsyncMethods:
@@ -326,15 +338,17 @@ def to_tbl_resp(raw_client_response):
             batches.append(batch)
             entity_id_batches.append(entity_id_batch)
 
-        async with self._get_aiodynamodb_client(online_config.region) as client:
-            response_batches = await asyncio.gather(
-                *[
-                    client.batch_get_item(
-                        RequestItems=entity_id_batch,
-                    )
-                    for entity_id_batch in entity_id_batches
-                ]
-            )
+        client = await _get_aiodynamodb_client(
+            online_config.region, online_config.max_pool_connections
+        )
+        response_batches = await asyncio.gather(
+            *[
+                client.batch_get_item(
+                    RequestItems=entity_id_batch,
+                )
+                for entity_id_batch in entity_id_batches
+            ]
+        )
 
         result_batches = []
         for batch, response in zip(batches, response_batches):
@@ -349,14 +363,6 @@ def to_tbl_resp(raw_client_response):
 
         return list(itertools.chain(*result_batches))
 
-    def _get_aioboto_session(self):
-        if self._aioboto_session is None:
-            self._aioboto_session = session.get_session()
-        return self._aioboto_session
-
-    def _get_aiodynamodb_client(self, region: str):
-        return self._get_aioboto_session().create_client("dynamodb", region_name=region)
-
     def _get_dynamodb_client(
         self,
         region: str,
@@ -489,6 +495,38 @@ def _to_client_batch_get_payload(online_config, table_name, batch):
         }
 
 
+_aioboto_session = None
+_aioboto_client = None
+
+
+def _get_aioboto_session():
+    global _aioboto_session
+    if _aioboto_session is None:
+        logger.debug("initializing the aiobotocore session")
+        _aioboto_session = session.get_session()
+    return _aioboto_session
+
+
+async def _get_aiodynamodb_client(region: str, max_pool_connections: int):
+    global _aioboto_client
+    if _aioboto_client is None:
+        logger.debug("initializing the aiobotocore dynamodb client")
+        client_context = _get_aioboto_session().create_client(
+            "dynamodb",
+            region_name=region,
+            config=AioConfig(max_pool_connections=max_pool_connections),
+        )
+        context_stack = contextlib.AsyncExitStack()
+        _aioboto_client = await context_stack.enter_async_context(client_context)
+    return _aioboto_client
+
+
+async def _aiodynamodb_close():
+    global _aioboto_client
+    if _aioboto_client:
+        await _aioboto_client.close()
+
+
 def _initialize_dynamodb_client(
     region: str,
     endpoint_url: Optional[str] = None,

sdk/python/feast/infra/online_stores/online_store.py

@@ -422,3 +422,9 @@ def retrieve_online_documents(
         raise NotImplementedError(
             f"Online store {self.__class__.__name__} does not support online retrieval"
         )
+
+    async def initialize(self, config: RepoConfig) -> None:
+        pass
+
+    async def close(self) -> None:
+        pass

sdk/python/feast/infra/passthrough_provider.py

@@ -518,3 +518,9 @@ def get_table_column_names_and_types_from_data_source(
         return self.offline_store.get_table_column_names_and_types_from_data_source(
             config=config, data_source=data_source
         )
+
+    async def initialize(self, config: RepoConfig) -> None:
+        await self.online_store.initialize(config)
+
+    async def close(self) -> None:
+        await self.online_store.close()

sdk/python/feast/infra/provider.py

@@ -476,6 +476,14 @@ def get_table_column_names_and_types_from_data_source(
         """
         pass
 
+    @abstractmethod
+    async def initialize(self, config: RepoConfig) -> None:
+        pass
+
+    @abstractmethod
+    async def close(self) -> None:
+        pass
+
 
 def get_provider(config: RepoConfig) -> Provider:
     if "." not in config.provider:

sdk/python/pytest.ini

@@ -1,4 +1,6 @@
 [pytest]
+asyncio_mode = auto
+
 markers =
     universal_offline_stores: mark a test as using all offline stores.
     universal_online_stores: mark a test as using all online stores.
@@ -7,6 +9,8 @@ env =
     IS_TEST=True
 
 filterwarnings =
+    error::_pytest.warning_types.PytestConfigWarning
+    error::_pytest.warning_types.PytestUnhandledCoroutineWarning
     ignore::DeprecationWarning:pyspark.sql.pandas.*:
     ignore::DeprecationWarning:pyspark.sql.connect.*:
     ignore::DeprecationWarning:httpx.*: