Make RedshiftDeleteClusterOperator delete reliably during cluster transitions

docs/spelling_wordlist.txt

@@ -449,6 +449,7 @@ deferrable
 deidentify
 DeidentifyTemplate
 del
+deletable
 delim
 deliverability
 deltalake

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

@@ -27,6 +27,7 @@
 from airflow.providers.amazon.aws.hooks.redshift_cluster import RedshiftHook
 from airflow.providers.amazon.aws.operators.base_aws import AwsBaseOperator
 from airflow.providers.amazon.aws.triggers.redshift_cluster import (
+    RedshiftClusterSettledTrigger,
     RedshiftCreateClusterSnapshotTrigger,
     RedshiftCreateClusterTrigger,
     RedshiftDeleteClusterTrigger,
@@ -836,7 +837,9 @@ class RedshiftDeleteClusterOperator(AwsBaseOperator[RedshiftHook]):
         https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
     :param poll_interval: Time (in seconds) to wait between two consecutive calls to check cluster state
     :param deferrable: Run operator in the deferrable mode.
-    :param max_attempts: (Deferrable mode only) The maximum number of attempts to be made
+    :param max_attempts: The maximum number of attempts to be made. In deferrable mode this bounds the
+        async wait for a busy cluster to settle before the delete is re-issued; combined with
+        ``poll_interval`` the default gives a ~15 minute window, long enough to outlast a pause/resize.
     """
 
     template_fields: Sequence[str] = aws_template_fields(
@@ -864,15 +867,18 @@ def __init__(
         self.final_cluster_snapshot_identifier = final_cluster_snapshot_identifier
         self.wait_for_completion = wait_for_completion
         self.poll_interval = poll_interval
-        # These parameters are added to keep trying if there is a running operation in the cluster
-        # If there is a running operation in the cluster while trying to delete it, a InvalidClusterStateFault
-        # is thrown. In such case, retrying
-        self._attempts = 10
+        # Retry the delete while the cluster is mid-transition (InvalidClusterStateFault) until it
+        # settles into a deletable state.
+        self._attempts = 60
         self._attempt_interval = 15
         self.deferrable = deferrable
         self.max_attempts = max_attempts
 
     def execute(self, context: Context):
+        if self.deferrable:
+            self._delete_or_defer_until_settled()
+            return
+
         while self._attempts:
             try:
                 self.hook.delete_cluster(
@@ -897,36 +903,82 @@ def execute(self, context: Context):
                 else:
                     raise
 
-        if self.deferrable:
-            cluster_state = self.hook.cluster_status(cluster_identifier=self.cluster_identifier)
-            if cluster_state == "cluster_not_found":
-                self.log.info("Cluster deleted successfully")
-            elif cluster_state in ("creating", "modifying"):
-                raise AirflowException(
-                    f"Unable to delete cluster since cluster is currently in status: {cluster_state}"
-                )
-            else:
-                self.defer(
-                    timeout=timedelta(seconds=self.max_attempts * self.poll_interval + 60),
-                    trigger=RedshiftDeleteClusterTrigger(
-                        cluster_identifier=self.cluster_identifier,
-                        waiter_delay=self.poll_interval,
-                        waiter_max_attempts=self.max_attempts,
-                        aws_conn_id=self.aws_conn_id,
-                        region_name=self.region_name,
-                        verify=self.verify,
-                        botocore_config=self.botocore_config,
-                    ),
-                    method_name="execute_complete",
-                )
-
-        elif self.wait_for_completion:
+        if self.wait_for_completion:
             waiter = self.hook.conn.get_waiter("cluster_deleted")
             waiter.wait(
                 ClusterIdentifier=self.cluster_identifier,
                 WaiterConfig={"Delay": self.poll_interval, "MaxAttempts": self.max_attempts},
             )
 
+    def _delete_or_defer_until_settled(self) -> None:
+        """
+        Issue the delete once (deferrable mode), then defer.
+
+        If accepted, defer to :class:`RedshiftDeleteClusterTrigger` to await deletion. If the cluster is
+        busy (``InvalidClusterStateFault``), defer to :class:`RedshiftClusterSettledTrigger`; the
+        ``_retry_delete_when_settled`` callback re-issues the delete once it settles.
+        """
+        try:
+            self.hook.delete_cluster(
+                cluster_identifier=self.cluster_identifier,
+                skip_final_cluster_snapshot=self.skip_final_cluster_snapshot,
+                final_cluster_snapshot_identifier=self.final_cluster_snapshot_identifier,
+            )
+        except self.hook.conn.exceptions.InvalidClusterStateFault:
+            self.log.info(
+                "Cluster %s is busy; deferring until it settles into a deletable state.",
+                self.cluster_identifier,
+            )
+            self.defer(
+                timeout=timedelta(seconds=self.max_attempts * self.poll_interval + 60),
+                trigger=RedshiftClusterSettledTrigger(
+                    cluster_identifier=self.cluster_identifier,
+                    waiter_delay=self.poll_interval,
+                    waiter_max_attempts=self.max_attempts,
+                    aws_conn_id=self.aws_conn_id,
+                    region_name=self.region_name,
+                    verify=self.verify,
+                    botocore_config=self.botocore_config,
+                ),
+                method_name="_retry_delete_when_settled",
+            )
+            return
+
+        self._defer_until_deleted()
+
+    def _defer_until_deleted(self) -> None:
+        """Defer to the delete-completion waiter, short-circuiting if the cluster is already gone."""
+        cluster_state = self.hook.cluster_status(cluster_identifier=self.cluster_identifier)
+        if cluster_state == "cluster_not_found":
+            self.log.info("Cluster deleted successfully")
+            return
+        self.defer(
+            timeout=timedelta(seconds=self.max_attempts * self.poll_interval + 60),
+            trigger=RedshiftDeleteClusterTrigger(
+                cluster_identifier=self.cluster_identifier,
+                waiter_delay=self.poll_interval,
+                waiter_max_attempts=self.max_attempts,
+                aws_conn_id=self.aws_conn_id,
+                region_name=self.region_name,
+                verify=self.verify,
+                botocore_config=self.botocore_config,
+            ),
+            method_name="execute_complete",
+        )
+
+    def _retry_delete_when_settled(self, context: Context, event: dict[str, Any] | None = None) -> None:
+        """
+        Re-issue the delete once the cluster has settled, then defer until deletion completes.
+
+        Callback for :class:`RedshiftClusterSettledTrigger`. If the delete is still rejected because of a
+        race (the cluster re-entered a transitional state), defer to the settle-wait trigger again.
+        """
+        validated_event = validate_execute_complete_event(event)
+        if validated_event["status"] != "success":
+            raise AirflowException(f"Error waiting for cluster to become deletable: {validated_event}")
+
+        self._delete_or_defer_until_settled()
+
     def execute_complete(self, context: Context, event: dict[str, Any] | None = None) -> None:
         validated_event = validate_execute_complete_event(event)
 

providers/amazon/src/airflow/providers/amazon/aws/triggers/redshift_cluster.py

@@ -343,3 +343,56 @@ async def run(self) -> AsyncIterator[TriggerEvent]:
                 await asyncio.sleep(self.poke_interval)
         except Exception as e:
             yield TriggerEvent({"status": "error", "message": str(e)})
+
+
+class RedshiftClusterSettledTrigger(AwsBaseWaiterTrigger):
+    """
+    Wait until a Redshift cluster settles into a non-transitional (deletable) lifecycle.
+
+    A ``delete_cluster`` call is rejected with ``InvalidClusterStateFault`` while an operation is in
+    flight (e.g. a pause or resize). Because a busy cluster can settle into *different* terminal states
+    depending on the in-flight operation (a ``pausing`` cluster becomes ``paused``; a ``resizing`` cluster
+    becomes ``available``), this trigger uses the custom ``cluster_deletable`` waiter, which has one
+    ``success`` acceptor per deletable ``ClusterStatus`` value rather than a single target state.
+
+    :param cluster_identifier: unique identifier of a cluster
+    :param waiter_delay: The amount of time in seconds to wait between attempts.
+    :param waiter_max_attempts: The maximum number of attempts to be made.
+    :param aws_conn_id: The Airflow connection used for AWS credentials.
+    :param region_name: The AWS region where the cluster is. Used to build the hook.
+    :param verify: Whether or not to verify SSL certificates. Used to build the hook.
+    :param botocore_config: Configuration dictionary for the botocore client. Used to build the hook.
+    """
+
+    def __init__(
+        self,
+        *,
+        cluster_identifier: str,
+        aws_conn_id: str | None = "aws_default",
+        region_name: str | None = None,
+        waiter_delay: int = 30,
+        waiter_max_attempts: int = 30,
+        **kwargs,
+    ):
+        super().__init__(
+            serialized_fields={"cluster_identifier": cluster_identifier},
+            waiter_name="cluster_deletable",
+            waiter_args={"ClusterIdentifier": cluster_identifier},
+            failure_message="Error while waiting for the redshift cluster to become deletable",
+            status_message="Waiting for redshift cluster to settle into a deletable state",
+            status_queries=["Clusters[].ClusterStatus"],
+            return_value=None,
+            waiter_delay=waiter_delay,
+            waiter_max_attempts=waiter_max_attempts,
+            aws_conn_id=aws_conn_id,
+            region_name=region_name,
+            **kwargs,
+        )
+
+    def hook(self) -> AwsGenericHook:
+        return RedshiftHook(
+            aws_conn_id=self.aws_conn_id,
+            region_name=self.region_name,
+            verify=self.verify,
+            config=self.botocore_config,
+        )

providers/amazon/src/airflow/providers/amazon/aws/waiters/redshift.json

@@ -50,6 +50,61 @@
                     "state": "failure"
                 }
             ]
+        },
+        "cluster_deletable": {
+            "operation": "DescribeClusters",
+            "delay": 30,
+            "maxAttempts": 60,
+            "acceptors": [
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "available",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "paused",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "incompatible-hsm",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "incompatible-restore",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "incompatible-network",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "incompatible-parameters",
+                    "state": "success"
+                },
+                {
+                    "matcher": "pathAll",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "hardware-failure",
+                    "state": "success"
+                },
+                {
+                    "matcher": "error",
+                    "argument": "Clusters[].ClusterStatus",
+                    "expected": "ClusterNotFound",
+                    "state": "success"
+                }
+            ]
         }
     }
 }