feat(retries): added throttling mode

Author: Natan Tolparov

README.md

@@ -157,22 +157,22 @@ set_up_yc_api_endpoint(kz_region_endpoint)
 ```
 
 ### Retries
-If you want to retry SDK requests, build SDK with `retry_policy` field using `RetryPolicy` class:
+SDK provide built-in retry policy, that supports [exponential backoff and jitter](https://aws.amazon.com/ru/blogs/architecture/exponential-backoff-and-jitter/), and also [retry budget](https://github.com/grpc/proposal/blob/master/A6-client-retries.md#throttling-retry-attempts-and-hedged-rpcs). It's necessary to avoid retry amplification.
 
 ```python
 import grpc
 from yandexcloud import SDK, RetryPolicy
 
-sdk = SDK(retry_policy=RetryPolicy(max_attempts=2, status_codes=(grpc.StatusCode.UNAVAILABLE,)))
+sdk = SDK(retry_policy=RetryPolicy())
 ```
 
-It's **strongly recommended** to use default settings of RetryPolicy to avoid retry amplification:
-```python
-import grpc
-from yandexcloud import SDK, RetryPolicy
+SDK provide different modes for retry throttling policy:
+
+* `persistent` is suitable when you use SDK in any long-lived application, when SDK instance will live long enough for manage budget;
+* `temporary` is suitable when you use SDK in any short-lived application, e.g. scripts or CI/CD.
+
+By default, SDK will use temporary mode, but you can change it through `throttling_mode` argument.
 
-sdk = SDK(retry_policy=RetryPolicy())
-```
 
 ## Contributing
 ### Dependencies

tests/test_retry_policy.py

@@ -10,7 +10,7 @@
     NetworkServiceStub,
     add_NetworkServiceServicer_to_server,
 )
-from yandexcloud import SDK, RetryPolicy
+from yandexcloud import SDK, RetryPolicy, ThrottlingMode
 from yandexcloud._channels import Channels
 
 INSECURE_SERVICE_PORT = "50051"
@@ -27,7 +27,7 @@ def side_effect_unavailable(_, context):
 
 class VPCServiceMock:
     def __init__(self, fn):
-        self.Get = Mock(return_value=Network(id="12342314"))
+        self.Get = Mock(side_effect=fn)
         self.Create = Mock()
         self.Update = Mock()
         self.Delete = Mock()
@@ -65,7 +65,7 @@ def test_default_retries(mock_channel):
     server, service = grpc_server(side_effect_unavailable)
 
     sdk = SDK(
-        retry_policy=RetryPolicy(),
+        retry_policy=RetryPolicy(throttling_mode=ThrottlingMode.PERSISTENT),
         endpoint=f"localhost:{INSECURE_SERVICE_PORT}",
         endpoints={
             "vpc": SERVICE_ADDR + ":" + INSECURE_SERVICE_PORT,
@@ -98,7 +98,8 @@ def test_custom_retries(mock_channel):
         request = GetNetworkRequest(network_id="asdf")
         network_client.Get(request)
     except grpc.RpcError:
-        assert service.Get.call_count == 4
+        # because of temporary throttling mode by default
+        assert service.Get.call_count == 3
 
     server.stop(0)
 

uv.lock

@@ -8,7 +8,7 @@
     default_backoff,
 )
 from yandexcloud._retry_interceptor import RetryInterceptor
-from yandexcloud._retry_policy import RetryPolicy
+from yandexcloud._retry_policy import RetryPolicy, ThrottlingMode
 from yandexcloud._sdk import SDK
 
 __version__ = "0.336.0"

yandexcloud/__init__.py

@@ -1,14 +1,21 @@
 import json
+from enum import Enum
 from typing import Tuple
 
 import grpc
 
 
+class ThrottlingMode(Enum):
+    PERSISTENT = "persistent"
+    TEMPORARY = "temporary"
+
+
 class RetryPolicy:
     def __init__(
         self,
         max_attempts: int = 4,
         status_codes: Tuple["grpc.StatusCode"] = (grpc.StatusCode.UNAVAILABLE,),
+        throttling_mode: ThrottlingMode = ThrottlingMode.TEMPORARY,
     ):
         self.__policy = {
             "methodConfig": [
@@ -21,10 +28,14 @@ def __init__(
                         "backoffMultiplier": 2,
                         "retryableStatusCodes": [status.name for status in status_codes],
                     },
+                    "waitForReady": True,
                 }
             ],
-            "retryThrottling": {"maxTokens": 100, "tokenRatio": 0.1},
-            "waitForReady": True,
+            "retryThrottling": (
+                {"maxTokens": 100, "tokenRatio": 0.1}
+                if throttling_mode == ThrottlingMode.PERSISTENT
+                else {"maxTokens": 6, "tokenRatio": 0.1}
+            ),
         }
 
     def to_json(self) -> str: