fix: correct behaviour to avoid caching "INPROGRESS" records

aws_lambda_powertools/utilities/idempotency/persistence/base.py

@@ -243,7 +243,7 @@ def _save_to_cache(self, data_record: DataRecord):
     def _retrieve_from_cache(self, idempotency_key: str):
         cached_record = self._cache.get(idempotency_key)
         if cached_record:
-            if not cached_record.is_expired:
+            if all((not cached_record.is_expired, not cached_record.status == "INPROGRESS")):
                 return cached_record
             logger.debug(f"Removing expired local cache record for idempotency key: {idempotency_key}")
             self._delete_from_cache(idempotency_key)
@@ -305,11 +305,6 @@ def save_inprogress(self, event: Dict[str, Any]) -> None:
 
         self._put_record(data_record)
 
-        # This has to come after _put_record. If _put_record call raises ItemAlreadyExists we shouldn't populate the
-        # cache with an "INPROGRESS" record as we don't know the status in the data store at this point.
-        if self.use_local_cache:
-            self._save_to_cache(data_record)
-
     def delete_record(self, event: Dict[str, Any], exception: Exception):
         """
         Delete record from the persistence store
@@ -365,7 +360,9 @@ def get_record(self, event: Dict[str, Any]) -> DataRecord:
 
         record = self._get_record(idempotency_key)
 
-        if self.use_local_cache:
+        # We can't cache "INPROGRESS" records as we have no way to reflect updates that can happen outside of the
+        # execution environment
+        if self.use_local_cache and not record.status == STATUS_CONSTANTS["INPROGRESS"]:
             self._save_to_cache(data_record=record)
 
         self._validate_payload(event, record)

docs/utilities/idempotency.md

@@ -3,8 +3,11 @@ title: Idempotency
 description: Utility
 ---
 
-This utility provides a simple solution to convert your Lambda functions into idempotent operations which are safe to
-retry.
+!!! attention
+    **This utility is currently in beta**. Please open an [issue in GitHub](https://github.com/awslabs/aws-lambda-powertools-python/issues/new/choose) for any bugs or feature requests.
+
+The idempotency utility provides a simple solution to convert your Lambda functions into idempotent operations which
+are safe to retry.
 
 ## Terminology
 
@@ -31,31 +34,31 @@ storage layer, so you'll need to create a table first.
 > Example using AWS Serverless Application Model (SAM)
 
 === "template.yml"
-```yaml
-Resources:
-  HelloWorldFunction:
-  Type: AWS::Serverless::Function
-  Properties:
-    Runtime: python3.8
-    ...
-    Policies:
-      - DynamoDBCrudPolicy:
-          TableName: !Ref IdempotencyTable
-  IdempotencyTable:
-    Type: AWS::DynamoDB::Table
-    Properties:
-      AttributeDefinitions:
-        -   AttributeName: id
-            AttributeType: S
-      BillingMode: PAY_PER_REQUEST
-      KeySchema:
-        -   AttributeName: id
-            KeyType: HASH
-      TableName: "IdempotencyTable"
-      TimeToLiveSpecification:
-        AttributeName: expiration
-        Enabled: true
-```
+    ```yaml
+    Resources:
+      HelloWorldFunction:
+      Type: AWS::Serverless::Function
+      Properties:
+        Runtime: python3.8
+        ...
+        Policies:
+          - DynamoDBCrudPolicy:
+              TableName: !Ref IdempotencyTable
+      IdempotencyTable:
+        Type: AWS::DynamoDB::Table
+        Properties:
+          AttributeDefinitions:
+            -   AttributeName: id
+                AttributeType: S
+          BillingMode: PAY_PER_REQUEST
+          KeySchema:
+            -   AttributeName: id
+                KeyType: HASH
+          TableName: "IdempotencyTable"
+          TimeToLiveSpecification:
+            AttributeName: expiration
+            Enabled: true
+    ```
 
 !!! warning
     When using this utility with DynamoDB, your lambda responses must always be smaller than 400kb. Larger items cannot

tests/functional/idempotency/test_idempotency.py

@@ -133,6 +133,12 @@ def test_idempotent_lambda_in_progress_with_cache(
 
     stubber.add_client_error("put_item", "ConditionalCheckFailedException")
     stubber.add_response("get_item", ddb_response, expected_params)
+
+    stubber.add_client_error("put_item", "ConditionalCheckFailedException")
+    stubber.add_response("get_item", copy.deepcopy(ddb_response), copy.deepcopy(expected_params))
+
+    stubber.add_client_error("put_item", "ConditionalCheckFailedException")
+    stubber.add_response("get_item", copy.deepcopy(ddb_response), copy.deepcopy(expected_params))
     stubber.activate()
 
     @idempotent(persistence_store=persistence_store)
@@ -151,11 +157,8 @@ def lambda_handler(event, context):
     assert retrieve_from_cache_spy.call_count == 2 * loops
     retrieve_from_cache_spy.assert_called_with(idempotency_key=hashed_idempotency_key)
 
-    assert save_to_cache_spy.call_count == 1
-    first_call_args_data_record = save_to_cache_spy.call_args_list[0].kwargs["data_record"]
-    assert first_call_args_data_record.idempotency_key == hashed_idempotency_key
-    assert first_call_args_data_record.status == "INPROGRESS"
-    assert persistence_store._cache.get(hashed_idempotency_key)
+    save_to_cache_spy.assert_not_called()
+    assert persistence_store._cache.get(hashed_idempotency_key) is None
 
     stubber.assert_no_pending_responses()
     stubber.deactivate()
@@ -223,12 +226,10 @@ def lambda_handler(event, context):
 
     lambda_handler(lambda_apigw_event, {})
 
-    assert retrieve_from_cache_spy.call_count == 1
-    assert save_to_cache_spy.call_count == 2
-    first_call_args, second_call_args = save_to_cache_spy.call_args_list
-    assert first_call_args.args[0].status == "INPROGRESS"
-    assert second_call_args.args[0].status == "COMPLETED"
-    assert persistence_store._cache.get(hashed_idempotency_key)
+    retrieve_from_cache_spy.assert_called_once()
+    save_to_cache_spy.assert_called_once()
+    assert save_to_cache_spy.call_args[0][0].status == "COMPLETED"
+    assert persistence_store._cache.get(hashed_idempotency_key).status == "COMPLETED"
 
     # This lambda call should not call AWS API
     lambda_handler(lambda_apigw_event, {})