feast-dev
@@ -2,7 +2,7 @@
 
 ## Description
 
-This remote online store will let you interact with remote feature server. At this moment this only supports the read operation. You can use this online store and able retrieve online features `store.get_online_features`  from remote feature server. 
+This remote online store lets you interact with a remote feature server. You can use this online store to retrieve online features via `store.get_online_features()` from a remote feature server.
 
 ## Examples
 
@@ -25,6 +25,128 @@ auth:
 
 `cert` is an optional configuration to the public certificate path when the online server starts in TLS(SSL) mode. This may be needed if the online server is started with a self-signed certificate, typically this file ends with `*.crt`, `*.cer`, or `*.pem`.
 
+## Connection Pooling Configuration
+
+The remote online store uses HTTP connection pooling to improve performance by reusing TCP/TLS connections across multiple requests. This significantly reduces latency by avoiding the overhead of establishing new connections for each request.
+
+### Configuration Options
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `connection_pool_size` | int | 50 | Maximum number of connections to keep in the pool. Increase for high-concurrency workloads. |
+| `connection_idle_timeout` | int | 300 | Maximum time in seconds a session can be idle before being closed. Set to `0` to disable idle timeout. |
+| `connection_retries` | int | 3 | Number of retries for failed requests with exponential backoff. |
+
+### Example with Connection Pooling
+
+{% code title="feature_store.yaml" %}
+```yaml
+project: my-local-project
+registry: /remote/data/registry.db
+provider: local
+online_store:
+  type: remote
+  path: http://feast-feature-server:80
+
+  # Connection pooling configuration (optional)
+  connection_pool_size: 50        # Max connections in pool
+  connection_idle_timeout: 300    # Idle timeout in seconds (0 to disable)
+  connection_retries: 3           # Retry count with exponential backoff
+
+entity_key_serialization_version: 3
+auth:
+  type: no_auth
+```
+{% endcode %}
+
+### Use Cases
+
+**High-throughput workloads:**
+```yaml
+online_store:
+  type: remote
+  path: http://feast-server:80
+  connection_pool_size: 100       # More connections for high concurrency
+  connection_idle_timeout: 600    # 10 minutes idle timeout
+  connection_retries: 5           # More retries for resilience
+```
+
+**Long-running services:**
+```yaml
+online_store:
+  type: remote
+  path: http://feast-server:80
+  connection_idle_timeout: 0      # Never auto-close session
+```
+
+**Resource-constrained environments:**
+```yaml
+online_store:
+  type: remote
+  path: http://feast-server:80
+  connection_pool_size: 10        # Fewer connections to reduce memory
+  connection_idle_timeout: 60     # 1 minute timeout
+```
+
+### Performance Benefits
+
+Connection pooling provides significant latency improvements:
+
+- **Without pooling**: Each request requires a new TCP connection (~10-50ms) and TLS handshake (~30-100ms)
+- **With pooling**: Subsequent requests reuse existing connections, eliminating connection overhead
+
+This is especially beneficial for:
+- High-frequency feature retrieval in real-time inference pipelines
+- Batch processing with many sequential `get_online_features()` calls
+- Services with authentication enabled (reduces token refresh overhead)
+
+### Session Cleanup
+
+The HTTP session is automatically managed with idle timeout, but you can also explicitly close it when your application is shutting down or when you want to release resources.
+
+#### Using FeatureStore context manager (recommended)
+
+The recommended way to ensure proper cleanup is to use the `FeatureStore` as a context manager:
+
+```python
+from feast import FeatureStore
+
+# Session is automatically closed when exiting the context
+with FeatureStore(repo_path=".") as store:
+    features = store.get_online_features(
+        features=["driver_hourly_stats:conv_rate"],
+        entity_rows=[{"driver_id": 1001}]
+    )
+```
+
+#### Explicit cleanup
+
+You can also explicitly close the session by calling `close()` on the `FeatureStore`:
+
+```python
+from feast import FeatureStore
+
+store = FeatureStore(repo_path=".")
+try:
+    features = store.get_online_features(
+        features=["driver_hourly_stats:conv_rate"],
+        entity_rows=[{"driver_id": 1001}]
+    )
+finally:
+    store.close()  # Closes HTTP session and releases resources
+```
+
+#### Direct session management
+
+For advanced use cases, you can directly manage the HTTP session via `HttpSessionManager`:
+
+```python
+from feast.permissions.client.http_auth_requests_wrapper import HttpSessionManager
+
+# Close the cached HTTP session
+HttpSessionManager.close_session()
+```
+
 ## How to configure Authentication and Authorization
 Please refer the [page](./../../../docs/getting-started/concepts/permission.md) for more details on how to configure authentication and authorization.
 
@@ -23,6 +23,7 @@
 from feast import Entity, FeatureView, RepoConfig
 from feast.infra.online_stores.helpers import _to_naive_utc
 from feast.infra.online_stores.online_store import OnlineStore
+from feast.permissions.client.http_auth_requests_wrapper import HttpSessionManager
 from feast.protos.feast.types.EntityKey_pb2 import EntityKey as EntityKeyProto
 from feast.protos.feast.types.Value_pb2 import Value as ValueProto
 from feast.repo_config import FeastConfigBaseModel
@@ -51,6 +52,18 @@ class RemoteOnlineStoreConfig(FeastConfigBaseModel):
     """ str: Path to the public certificate when the online server starts in TLS(SSL) mode. This may be needed if the online server started with a self-signed certificate, typically this file ends with `*.crt`, `*.cer`, or `*.pem`.
     If type is 'remote', then this configuration is needed to connect to remote online server in TLS mode. """
 
+    # Connection pooling configuration
+    connection_pool_size: int = 50
+    """ int: Maximum number of connections to keep in the pool (default 50).
+    Increase for high-concurrency workloads. """
+
+    connection_idle_timeout: int = 300
+    """ int: Maximum time in seconds a session can be idle before being closed (default 300 = 5 minutes).
+    Set to 0 to disable idle timeout. """
+
+    connection_retries: int = 3
+    """ int: Number of retries for failed requests with exponential backoff (default 3). """
+
 
 class RemoteOnlineStore(OnlineStore):
     """
@@ -544,6 +557,20 @@ def teardown(
     ):
         pass
 
+    async def close(self) -> None:
+        """
+        Close the HTTP session and release connection pool resources.
+
+        This method is called automatically when FeatureStore.close() is invoked.
+        It cleans up the cached HTTP session used for connection pooling.
+
+        Note: Since the session is shared globally, calling close() will affect
+        all RemoteOnlineStore instances in the same process. This is typically
+        fine for SDK usage where there's usually one FeatureStore per process.
+        """
+        HttpSessionManager.close_session()
+        logger.debug("RemoteOnlineStore HTTP session closed")
+
 
 @rest_error_handling_decorator
 def get_remote_online_features(