feat: cleanup of old indirect host audit records

awx/main/tasks/host_indirect.py

@@ -117,6 +117,15 @@ def save_indirect_host_entries_of_job(job: Job) -> None:
     job.event_queries_processed = True
 
 
+def cleanup_old_indirect_host_entries() -> None:
+    """
+    We assume that indirect host audit results older than one week have already been collected for analysis
+    and can be cleaned up
+    """
+    limit = now() - timedelta(days=settings.INDIRECT_HOST_AUDIT_RECORD_MAX_AGE_DAYS)
+    IndirectManagedNodeAudit.objects.filter(created__lt=limit).delete()
+
+
 @task(queue=get_task_queuename)
 def save_indirect_host_entries(job_id: int, wait_for_events: bool = True) -> None:
     try:
@@ -148,10 +157,15 @@ def save_indirect_host_entries(job_id: int, wait_for_events: bool = True) -> Non
 
 
 @task(queue=get_task_queuename)
-def save_indirect_host_entries_fallback() -> None:
+def cleanup_and_save_indirect_host_entries_fallback() -> None:
     if not flag_enabled("FEATURE_INDIRECT_NODE_COUNTING_ENABLED"):
         return
 
+    try:
+        cleanup_old_indirect_host_entries()
+    except Exception as e:
+        logger.error(f"error cleaning up indirect host audit records: {e}")
+
     job_ct = 0
     right_now_time = now()
     window_end = right_now_time - timedelta(seconds=settings.INDIRECT_HOST_QUERY_FALLBACK_MINUTES)

awx/main/tests/functional/tasks/test_host_indirect.py

@@ -5,10 +5,15 @@
 
 from django.utils.timezone import now, timedelta
 
-from awx.main.tasks.host_indirect import build_indirect_host_data, fetch_job_event_query, save_indirect_host_entries, save_indirect_host_entries_fallback
+from awx.main.tasks.host_indirect import (
+    build_indirect_host_data,
+    fetch_job_event_query,
+    save_indirect_host_entries,
+    cleanup_and_save_indirect_host_entries_fallback,
+)
 from awx.main.models.event_query import EventQuery
 from awx.main.models.indirect_managed_node_audit import IndirectManagedNodeAudit
-
+from awx.main.tests.functional.conftest import organization
 
 """These are unit tests, similar to test_indirect_host_counting in the live tests"""
 
@@ -39,12 +44,30 @@ def create_event_query(fqcn='demo.query'):
     return EventQuery.objects.create(fqcn=fqcn, collection_version='1.0.1', event_query=yaml.dump({module_name: TEST_JQ}, default_flow_style=False))
 
 
+def create_audit_record(name, job, organization, created=now()):
+    record = IndirectManagedNodeAudit.objects.create(name=name, job=job, organization=organization)
+    record.created = created
+    record.save()
+    return record
+
+
 @pytest.fixture
 def event_query():
     "This is ordinarily created by the artifacts callback"
     return create_event_query()
 
 
+@pytest.fixture
+def old_audit_record(bare_job, organization):
+    created_at = now() - timedelta(days=10)
+    return create_audit_record(name="old_job", job=bare_job, organization=organization, created=created_at)
+
+
+@pytest.fixture
+def new_audit_record(bare_job, organization):
+    return IndirectManagedNodeAudit.objects.create(name="new_job", job=bare_job, organization=organization)
+
+
 # ---- end fixtures ----
 
 
@@ -167,7 +190,7 @@ def test_events_not_fully_processed_no_op(bare_job):
 
     # Right away, the fallback processing will not run either
     with mock.patch.object(save_indirect_host_entries, 'delay') as mock_delay:
-        save_indirect_host_entries_fallback()
+        cleanup_and_save_indirect_host_entries_fallback()
     mock_delay.assert_not_called()
     bare_job.refresh_from_db()
     assert bare_job.event_queries_processed is False
@@ -178,7 +201,7 @@ def test_events_not_fully_processed_no_op(bare_job):
 
     # The fallback task will now process indirect host query data for this job
     with mock.patch.object(save_indirect_host_entries, 'delay') as mock_delay:
-        save_indirect_host_entries_fallback()
+        cleanup_and_save_indirect_host_entries_fallback()
     mock_delay.assert_called_once_with(bare_job.id, wait_for_events=True)
 
     # Test code to process anyway, events collected or not
@@ -190,3 +213,12 @@ def test_events_not_fully_processed_no_op(bare_job):
 @pytest.mark.django_db
 def test_job_id_does_not_exist():
     save_indirect_host_entries(10000001)
+
+
+@pytest.mark.django_db
+def test_cleanup_old_audit_records(old_audit_record, new_audit_record):
+    count_before_cleanup = IndirectManagedNodeAudit.objects.count()
+    assert count_before_cleanup == 2
+    cleanup_and_save_indirect_host_entries_fallback()
+    count_after_cleanup = IndirectManagedNodeAudit.objects.count()
+    assert count_after_cleanup == 1

awx/main/tests/unit/tasks/test_host_indirect_unit.py

@@ -47,7 +47,7 @@ def test_compare_equal_data(self, data):
             [['a', {'b': 'c'}], ['a', {'b': 'd'}]],
         ],
     )
-    def test_compar_different_data(self, data, other_data):
+    def test_compare_different_data(self, data, other_data):
         assert data != other_data  # sanity, otherwise why test this?
         assert get_hashable_form(data) != get_hashable_form(other_data)
         assert hash(get_hashable_form(data)) != hash(get_hashable_form(other_data))

awx/settings/defaults.py

@@ -441,7 +441,10 @@
     'cleanup_host_metrics': {'task': 'awx.main.tasks.host_metrics.cleanup_host_metrics', 'schedule': timedelta(hours=3, minutes=30)},
     'host_metric_summary_monthly': {'task': 'awx.main.tasks.host_metrics.host_metric_summary_monthly', 'schedule': timedelta(hours=4)},
     'periodic_resource_sync': {'task': 'awx.main.tasks.system.periodic_resource_sync', 'schedule': timedelta(minutes=15)},
-    'save_indirect_host_entries_fallback': {'task': 'awx.main.tasks.host_indirect.save_indirect_host_entries_fallback', 'schedule': timedelta(minutes=60)},
+    'cleanup_and_save_indirect_host_entries_fallback': {
+        'task': 'awx.main.tasks.host_indirect.save_indirect_host_entries_fallback',
+        'schedule': timedelta(minutes=60),
+    },
 }
 
 # Django Caching Configuration
@@ -1069,6 +1072,11 @@
 # If an error happens in event collection, give up after this time
 INDIRECT_HOST_QUERY_FALLBACK_GIVEUP_DAYS = 3
 
+# Maximum age for indirect host audit records
+# Older records will be cleaned up
+INDIRECT_HOST_AUDIT_RECORD_MAX_AGE_DAYS = 7
+
+
 # feature flags
 FLAGS = {'FEATURE_INDIRECT_NODE_COUNTING_ENABLED': [{'condition': 'boolean', 'value': False}]}