Fix Comprehend document classifier timing out after 20 minutes

providers/amazon/docs/changelog.rst

@@ -26,6 +26,14 @@
 Changelog
 ---------
 
+.. warning::
+  The default waiter timeout of ``ComprehendCreateDocumentClassifierOperator`` was raised from
+  20 minutes (``waiter_max_attempts=20``) to 60 minutes (``waiter_max_attempts=60``), because
+  document classifier training sometimes takes longer than 20 minutes. When the operator waits
+  for completion (the default, in both synchronous and deferrable mode), tasks that previously
+  failed with a waiter timeout around the 20-minute mark now keep waiting for up to an hour.
+  Pass ``waiter_max_attempts`` explicitly to restore the previous timeout.
+
 9.35.1
 ......
 

providers/amazon/src/airflow/providers/amazon/aws/operators/comprehend.py

@@ -238,7 +238,7 @@ class ComprehendCreateDocumentClassifierOperator(AwsBaseOperator[ComprehendHook]
 
     :param wait_for_completion: Whether to wait for job to stop. (default: True)
     :param waiter_delay: Time in seconds to wait between status checks. (default: 60)
-    :param waiter_max_attempts: Maximum number of attempts to check for job completion. (default: 20)
+    :param waiter_max_attempts: Maximum number of attempts to check for job completion. (default: 60)
     :param deferrable: If True, the operator will wait asynchronously for the job to stop.
         This implies waiting for completion. This mode requires aiobotocore module to be installed.
         (default: False)
@@ -285,7 +285,7 @@ def __init__(
         document_classifier_kwargs: dict[str, Any] | None = None,
         wait_for_completion: bool = True,
         waiter_delay: int = 60,
-        waiter_max_attempts: int = 20,
+        waiter_max_attempts: int = 60,
         deferrable: bool = conf.getboolean("operators", "default_deferrable", fallback=False),
         **kwargs,
     ):

providers/amazon/tests/system/amazon/aws/example_comprehend_document_classifier.py

@@ -106,7 +106,6 @@ def document_classifier_workflow():
         document_classifier_kwargs=document_classifier_kwargs,
     )
     # [END howto_operator_create_document_classifier]
-    create_document_classifier.wait_for_completion = False
 
     # [START howto_sensor_create_document_classifier]
     await_create_document_classifier = ComprehendCreateDocumentClassifierCompletedSensor(

providers/amazon/tests/unit/amazon/aws/operators/test_comprehend.py

@@ -243,6 +243,8 @@ def test_init(self):
         assert self.operator.language_code == "en"
         assert self.operator.document_classifier_kwargs == {"VersionName": "v1"}
         assert self.operator.fail_on_warnings is False
+        assert self.operator.waiter_delay == 60
+        assert self.operator.waiter_max_attempts == 60
 
     @mock.patch.object(ComprehendHook, "conn")
     def test_create_document_classifier(self, mock_conn):