Soft delete datasets that are no longer referenced in DAG schedules or task outlets

airflow/config_templates/config.yml

@@ -1946,11 +1946,12 @@
       type: string
       example: ~
       default: "30"
-    - name: deactivate_stale_dags_interval
+    - name: parsing_cleanup_interval
       description: |
         How often (in seconds) to check for stale DAGs (DAGs which are no longer present in
-        the expected files) which should be deactivated.
-      version_added: 2.2.5
+        the expected files) which should be deactivated, as well as datasets that are no longer
+        referenced and should be marked as orphaned.
+      version_added: 2.5.0
       type: integer
       example: ~
       default: "60"

airflow/config_templates/default_airflow.cfg

@@ -985,8 +985,9 @@ scheduler_idle_sleep_time = 1
 min_file_process_interval = 30
 
 # How often (in seconds) to check for stale DAGs (DAGs which are no longer present in
-# the expected files) which should be deactivated.
-deactivate_stale_dags_interval = 60
+# the expected files) which should be deactivated, as well as datasets that are no longer
+# referenced and should be marked as orphaned.
+parsing_cleanup_interval = 60
 
 # How often (in seconds) to scan the DAGs directory for new files. Default to 5 minutes.
 dag_dir_list_interval = 300

airflow/configuration.py

@@ -255,6 +255,7 @@ class AirflowConfigParser(ConfigParser):
                 "worker_pods_pending_timeout_batch_size",
             )
         },
+        ("scheduler", "parsing_cleanup_interval"): ("scheduler", "deactivate_stale_dags_interval", "2.5.0"),
     }
 
     # A mapping of old default values that we want to change and warn the user

airflow/dag_processing/manager.py

@@ -434,7 +434,7 @@ def __init__(
         # Last time we cleaned up DAGs which are no longer in files
         self.last_deactivate_stale_dags_time = timezone.make_aware(datetime.fromtimestamp(0))
         # How often to check for DAGs which are no longer in files
-        self.deactivate_stale_dags_interval = conf.getint("scheduler", "deactivate_stale_dags_interval")
+        self.parsing_cleanup_interval = conf.getint("scheduler", "parsing_cleanup_interval")
         # How long to wait before timing out a process to parse a DAG file
         self._processor_timeout = processor_timeout
         # How often to scan the DAGs directory for new files. Default to 5 minutes.
@@ -497,7 +497,7 @@ def _deactivate_stale_dags(self, session=None):
         """
         now = timezone.utcnow()
         elapsed_time_since_refresh = (now - self.last_deactivate_stale_dags_time).total_seconds()
-        if elapsed_time_since_refresh > self.deactivate_stale_dags_interval:
+        if elapsed_time_since_refresh > self.parsing_cleanup_interval:
             last_parsed = {
                 fp: self.get_last_finish_time(fp) for fp in self.file_paths if self.get_last_finish_time(fp)
             }

airflow/jobs/scheduler_job.py

@@ -30,7 +30,7 @@
 from pathlib import Path
 from typing import TYPE_CHECKING, Collection, DefaultDict, Iterator
 
-from sqlalchemy import func, not_, or_, text
+from sqlalchemy import and_, func, not_, or_, text
 from sqlalchemy.exc import OperationalError
 from sqlalchemy.orm import load_only, selectinload
 from sqlalchemy.orm.session import Session, make_transient
@@ -46,7 +46,13 @@
 from airflow.models.dag import DAG, DagModel
 from airflow.models.dagbag import DagBag
 from airflow.models.dagrun import DagRun
-from airflow.models.dataset import DagScheduleDatasetReference, DatasetDagRunQueue, DatasetEvent
+from airflow.models.dataset import (
+    DagScheduleDatasetReference,
+    DatasetDagRunQueue,
+    DatasetEvent,
+    DatasetModel,
+    TaskOutletDatasetReference,
+)
 from airflow.models.serialized_dag import SerializedDagModel
 from airflow.models.taskinstance import SimpleTaskInstance, TaskInstance, TaskInstanceKey
 from airflow.stats import Stats
@@ -854,9 +860,14 @@ def _run_scheduler_loop(self) -> None:
         )
         timers.call_regular_interval(60.0, self._update_dag_run_state_for_paused_dags)
 
+        timers.call_regular_interval(
+            conf.getfloat("scheduler", "parsing_cleanup_interval"),
+            self._orphan_unreferenced_datasets,
+        )
+
         if self._standalone_dag_processor:
             timers.call_regular_interval(
-                conf.getfloat("scheduler", "deactivate_stale_dags_interval", fallback=60.0),
+                conf.getfloat("scheduler", "parsing_cleanup_interval"),
                 self._cleanup_stale_dags,
             )
 
@@ -1574,3 +1585,33 @@ def _cleanup_stale_dags(self, session: Session = NEW_SESSION) -> None:
             dag.is_active = False
             SerializedDagModel.remove_dag(dag_id=dag.dag_id, session=session)
         session.flush()
+
+    @provide_session
+    def _orphan_unreferenced_datasets(self, session: Session = NEW_SESSION) -> None:
+        """
+        Detects datasets that are no longer referenced in any DAG schedule parameters or task outlets and
+        sets the dataset is_orphaned flag to True
+        """
+        orphaned_dataset_query = (
+            session.query(DatasetModel)
+            .join(
+                DagScheduleDatasetReference,
+                isouter=True,
+            )
+            .join(
+                TaskOutletDatasetReference,
+                isouter=True,
+            )
+            # MSSQL doesn't like it when we select a column that we haven't grouped by. All other DBs let us
+            # group by id and select all columns.
+            .group_by(DatasetModel if session.get_bind().dialect.name == "mssql" else DatasetModel.id)
+            .having(
+                and_(
+                    func.count(DagScheduleDatasetReference.dag_id) == 0,
+                    func.count(TaskOutletDatasetReference.dag_id) == 0,
+                )
+            )
+        )
+        for dataset in orphaned_dataset_query:
+            self.log.info("Orphaning unreferenced dataset '%s'", dataset.uri)
+            dataset.is_orphaned = expression.true()

airflow/migrations/versions/0122_2_5_0_add_is_orphaned_to_datasetmodel.py

@@ -0,0 +1,57 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+"""Add is_orphaned to DatasetModel
+
+Revision ID: 290244fb8b83
+Revises: 1986afd32c1b
+Create Date: 2022-11-22 00:12:53.432961
+
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "290244fb8b83"
+down_revision = "1986afd32c1b"
+branch_labels = None
+depends_on = None
+airflow_version = "2.5.0"
+
+
+def upgrade():
+    """Add is_orphaned to DatasetModel"""
+    with op.batch_alter_table("dataset") as batch_op:
+        batch_op.add_column(
+            sa.Column(
+                "is_orphaned",
+                sa.Boolean,
+                default=False,
+                nullable=False,
+                server_default="0",
+            )
+        )
+
+
+def downgrade():
+    """Remove is_orphaned from DatasetModel"""
+    with op.batch_alter_table("dataset") as batch_op:
+        batch_op.drop_column("is_orphaned", mssql_drop_default=True)

airflow/models/dag.py

@@ -2825,6 +2825,10 @@ def bulk_write_to_db(
         for dataset in all_datasets:
             stored_dataset = session.query(DatasetModel).filter(DatasetModel.uri == dataset.uri).first()
             if stored_dataset:
+                # Some datasets may have been previously unreferenced, and therefore orphaned by the
+                # scheduler. But if we're here, then we have found that dataset again in our DAGs, which
+                # means that it is no longer an orphan, so set is_orphaned to False.
+                stored_dataset.is_orphaned = expression.false()
                 stored_datasets[stored_dataset.uri] = stored_dataset
             else:
                 session.add(dataset)

airflow/models/dataset.py

@@ -21,6 +21,7 @@
 
 import sqlalchemy_jsonfield
 from sqlalchemy import (
+    Boolean,
     Column,
     ForeignKey,
     ForeignKeyConstraint,
@@ -64,6 +65,7 @@ class DatasetModel(Base):
     extra = Column(sqlalchemy_jsonfield.JSONField(json=json), nullable=False, default={})
     created_at = Column(UtcDateTime, default=timezone.utcnow, nullable=False)
     updated_at = Column(UtcDateTime, default=timezone.utcnow, onupdate=timezone.utcnow, nullable=False)
+    is_orphaned = Column(Boolean, default=False, nullable=False, server_default="0")
 
     consuming_dags = relationship("DagScheduleDatasetReference", back_populates="dataset")
     producing_tasks = relationship("TaskOutletDatasetReference", back_populates="dataset")

airflow/www/views.py

@@ -3565,7 +3565,7 @@ def next_run_datasets(self, dag_id):
                     ),
                     isouter=True,
                 )
-                .filter(DagScheduleDatasetReference.dag_id == dag_id)
+                .filter(DagScheduleDatasetReference.dag_id == dag_id, ~DatasetModel.is_orphaned)
                 .group_by(DatasetModel.id, DatasetModel.uri)
                 .order_by(DatasetModel.uri)
                 .all()
@@ -3688,7 +3688,7 @@ def datasets_summary(self):
             if has_event_filters:
                 count_query = count_query.join(DatasetEvent, DatasetEvent.dataset_id == DatasetModel.id)
 
-            filters = []
+            filters = [~DatasetModel.is_orphaned]
             if uri_pattern:
                 filters.append(DatasetModel.uri.ilike(f"%{uri_pattern}%"))
             if updated_after:

docs/apache-airflow/img/airflow_erd.sha256

@@ -1 +1 @@
-f529521071a6c9ae8bbd58d63cf1195fc1ec964308e7684569d0a36d26534def
+5bee32cf7239656360cfe7978e7f01d6ee3b9914d3f3fd89ed4ff747254dddf8