Files
awx/awx/main/management/commands/cleanup_jobs.py
Elijah DeLee bf6a5f6b21 AAP-36932 Batch-delete JobHostSummary before cleanup to avoid invalid memory alloc (#16563)
Django's cascade collector materializes all JobHostSummary IDs into a
single UPDATE ... IN (...) to SET_NULL on Host.last_job_host_summary.
With many jobs x hosts this exceeds PostgreSQL's max memory allocation
size for a single query (1GB).

Pre-delete JHS rows and clear Host FK references in chunks of 1000 job
IDs using raw SQL before Django's .delete() runs, so the cascade
collector finds nothing to collect.

Assisted-by: Claude Code via Google Vertex AI
2026-07-30 10:13:35 -04:00

512 lines
23 KiB
Python

# Copyright (c) 2015 Ansible, Inc.
# All Rights Reserved.
# Python
import datetime
import logging
import re
# Django
from django.apps import apps
from django.core.management.base import BaseCommand, CommandError
from django.db import transaction, connection
from django.db.models import Min, Max
from django.db.models.signals import pre_save, post_save, pre_delete, post_delete, m2m_changed
from django.utils.timezone import now
# AWX
from awx.main.models import Job, AdHocCommand, ProjectUpdate, InventoryUpdate, SystemJob, WorkflowJob, Notification
from awx.main.utils import unified_job_class_to_event_table_name
def partition_table_name(job_class, dt):
suffix = dt.replace(microsecond=0, second=0, minute=0).strftime('%Y%m%d_%H')
event_tbl_name = unified_job_class_to_event_table_name(job_class)
event_tbl_name += f'_{suffix}'
return event_tbl_name
def partition_name_dt(part_name):
"""
part_name examples:
main_jobevent_20210318_09
main_projectupdateevent_20210318_11
main_inventoryupdateevent_20210318_03
"""
if '_unpartitioned' in part_name:
return None
p = re.compile('([a-z]+)_([a-z]+)_([0-9]+)_([0-9][0-9])')
m = p.match(part_name)
if not m:
return m
dt_str = f"{m.group(3)}_{m.group(4)}"
dt = datetime.datetime.strptime(dt_str, '%Y%m%d_%H').replace(tzinfo=datetime.timezone.utc)
return dt
def dt_to_partition_name(tbl_name, dt):
return f"{tbl_name}_{dt.strftime('%Y%m%d_%H')}"
JHS_CHUNK_SIZE = 1000
def _pre_delete_job_host_summaries(job_pks, logger=None):
"""Pre-delete JobHostSummary rows and clear Host FK references in batches.
Django's cascade collector materializes all JHS IDs into a single
UPDATE ... IN (...) to SET_NULL on Host.last_job_host_summary.
With many jobs x hosts this exceeds PostgreSQL's 1GB alloc limit.
Doing it in chunks with raw SQL avoids that.
"""
if not job_pks:
return
# ANY(%s) is PostgreSQL-specific; AWX only supports PostgreSQL
with connection.cursor() as cursor:
for i in range(0, len(job_pks), JHS_CHUNK_SIZE):
chunk = list(job_pks[i : i + JHS_CHUNK_SIZE])
cursor.execute(
"UPDATE main_host SET last_job_host_summary_id = NULL"
" WHERE last_job_host_summary_id IN"
" (SELECT id FROM main_jobhostsummary WHERE job_id = ANY(%s))",
[chunk],
)
cursor.execute(
"DELETE FROM main_jobhostsummary WHERE job_id = ANY(%s)",
[chunk],
)
if logger:
logger.debug("Pre-deleted JobHostSummary chunk %d-%d of %d job PKs", i, i + len(chunk), len(job_pks))
class DeleteMeta:
def __init__(self, logger, job_class, cutoff, dry_run):
self.logger = logger
self.job_class = job_class
self.cutoff = cutoff
self.dry_run = dry_run
self.jobs_qs = None # Set in by find_jobs_to_delete()
self.parts_no_drop = set() # Set in identify_excluded_partitions()
self.parts_to_drop = set() # Set in find_partitions_to_drop()
self.jobs_pk_list = [] # Set in find_jobs_to_delete()
self.jobs_to_delete_count = 0 # Set in find_jobs_to_delete()
self.jobs_no_delete_count = 0 # Set in find_jobs_to_delete()
def find_jobs_to_delete(self):
self.jobs_qs = self.job_class.objects.filter(created__lt=self.cutoff).values_list('pk', 'status', 'created')
for pk, status, created in self.jobs_qs:
if status not in ['pending', 'waiting', 'running']:
self.jobs_to_delete_count += 1
self.jobs_pk_list.append(pk)
self.jobs_no_delete_count = (
self.job_class.objects.filter(created__gte=self.cutoff) | self.job_class.objects.filter(status__in=['pending', 'waiting', 'running'])
).count()
def identify_excluded_partitions(self):
part_drop = {}
for pk, status, created in self.jobs_qs:
part_key = partition_table_name(self.job_class, created)
if status in ['pending', 'waiting', 'running']:
part_drop[part_key] = False
else:
part_drop.setdefault(part_key, True)
# Note that parts_no_drop _may_ contain the names of partitions that don't exist
# This can happen when the cleanup of _unpartitioned_* logic leaves behind jobs with status pending, waiting, running. The find_jobs_to_delete() will
# pick these jobs up.
self.parts_no_drop = {k for k, v in part_drop.items() if v is False}
def delete_jobs(self):
if not self.dry_run:
if self.job_class is Job:
_pre_delete_job_host_summaries(self.jobs_pk_list, self.logger)
self.job_class.objects.filter(pk__in=self.jobs_pk_list).delete()
def find_partitions_to_drop(self):
tbl_name = unified_job_class_to_event_table_name(self.job_class)
with connection.cursor() as cursor:
query = "SELECT inhrelid::regclass::text AS child FROM pg_catalog.pg_inherits"
query += f" WHERE inhparent = '{tbl_name}'::regclass"
query += f" AND TO_TIMESTAMP(LTRIM(inhrelid::regclass::text, '{tbl_name}_'), 'YYYYMMDD_HH24') < '{self.cutoff}'"
query += " ORDER BY inhrelid::regclass::text"
cursor.execute(query)
partitions_from_db = [r[0] for r in cursor.fetchall()]
partitions_dt = [partition_name_dt(p) for p in partitions_from_db if not None]
partitions_dt = [p for p in partitions_dt if not None]
# convert datetime partition back to string partition
partitions_maybe_drop = {dt_to_partition_name(tbl_name, dt) for dt in partitions_dt}
# Do not drop partition if there is a job that will not be deleted pointing at it
self.parts_to_drop = partitions_maybe_drop - self.parts_no_drop
def drop_partitions(self):
if len(self.parts_to_drop) > 0:
parts_to_drop = list(self.parts_to_drop)
parts_to_drop.sort() # sort it to make reading it easier for humans
parts_to_drop_str = ','.join(parts_to_drop)
if self.dry_run:
self.logger.debug(f"Would drop event partition(s) {parts_to_drop_str}")
else:
self.logger.debug(f"Dropping event partition(s) {parts_to_drop_str}")
if not self.dry_run:
with connection.cursor() as cursor:
cursor.execute(f"DROP TABLE {parts_to_drop_str}")
else:
self.logger.debug("No event partitions to drop")
def delete(self):
self.find_jobs_to_delete()
self.identify_excluded_partitions()
self.find_partitions_to_drop()
self.drop_partitions()
self.delete_jobs()
return (self.jobs_no_delete_count, self.jobs_to_delete_count)
class Command(BaseCommand):
"""
Management command to cleanup old jobs and project updates.
"""
help = 'Remove old jobs, project and inventory updates from the database.'
def add_arguments(self, parser):
parser.add_argument('--days', dest='days', type=int, default=90, metavar='N', help='Remove jobs/updates executed more than N days ago. Defaults to 90.')
parser.add_argument('--dry-run', dest='dry_run', action='store_true', default=False, help='Dry run mode (show items that would be removed)')
parser.add_argument(
'--batch-size', dest='batch_size', type=int, default=100000, metavar='X', help='Remove jobs in batch of X jobs. Defaults to 100000.'
)
parser.add_argument('--jobs', dest='only_jobs', action='store_true', default=False, help='Remove jobs')
parser.add_argument('--ad-hoc-commands', dest='only_ad_hoc_commands', action='store_true', default=False, help='Remove ad hoc commands')
parser.add_argument('--project-updates', dest='only_project_updates', action='store_true', default=False, help='Remove project updates')
parser.add_argument('--inventory-updates', dest='only_inventory_updates', action='store_true', default=False, help='Remove inventory updates')
parser.add_argument('--management-jobs', default=False, action='store_true', dest='only_management_jobs', help='Remove management jobs')
parser.add_argument('--notifications', dest='only_notifications', action='store_true', default=False, help='Remove notifications')
parser.add_argument('--workflow-jobs', default=False, action='store_true', dest='only_workflow_jobs', help='Remove workflow jobs')
def init_logging(self):
log_levels = dict(enumerate([logging.ERROR, logging.INFO, logging.DEBUG, 0]))
self.logger = logging.getLogger('awx.main.commands.cleanup_jobs')
self.logger.setLevel(log_levels.get(self.verbosity, 0))
handler = logging.StreamHandler()
handler.setFormatter(logging.Formatter('%(message)s'))
self.logger.addHandler(handler)
self.logger.propagate = False
def cleanup(self, job_class):
delete_meta = DeleteMeta(self.logger, job_class, self.cutoff, self.dry_run)
skipped, deleted = delete_meta.delete()
return (delete_meta.jobs_no_delete_count, delete_meta.jobs_to_delete_count)
def cleanup_jobs_partition(self):
return self.cleanup(Job)
def cleanup_ad_hoc_commands_partition(self):
return self.cleanup(AdHocCommand)
def cleanup_project_updates_partition(self):
return self.cleanup(ProjectUpdate)
def cleanup_inventory_updates_partition(self):
return self.cleanup(InventoryUpdate)
def cleanup_management_jobs_partition(self):
return self.cleanup(SystemJob)
def cleanup_workflow_jobs_partition(self):
delete_meta = DeleteMeta(self.logger, WorkflowJob, self.cutoff, self.dry_run)
delete_meta.find_jobs_to_delete()
delete_meta.delete_jobs()
return (delete_meta.jobs_no_delete_count, delete_meta.jobs_to_delete_count)
def has_unpartitioned_table(self, model):
tblname = unified_job_class_to_event_table_name(model)
with connection.cursor() as cursor:
cursor.execute(f"SELECT 1 FROM pg_tables WHERE tablename = '_unpartitioned_{tblname}';")
row = cursor.fetchone()
if row is None:
return False
return True
def _delete_unpartitioned_table(self, model):
"If the unpartitioned table is no longer necessary, it will drop the table"
tblname = unified_job_class_to_event_table_name(model)
if not self.has_unpartitioned_table(model):
self.logger.debug(f'Table _unpartitioned_{tblname} does not exist, you are fully migrated.')
return
with connection.cursor() as cursor:
# same as UnpartitionedJobEvent.objects.aggregate(Max('created'))
cursor.execute(f'SELECT MAX("_unpartitioned_{tblname}"."created") FROM "_unpartitioned_{tblname}";')
row = cursor.fetchone()
last_created = row[0]
if last_created:
self.logger.info(f'Last event created in _unpartitioned_{tblname} was {last_created.isoformat()}')
else:
self.logger.info(f'Table _unpartitioned_{tblname} has no events in it')
if (last_created is None) or (last_created < self.cutoff):
self.logger.warning(
f'Dropping table _unpartitioned_{tblname} since no records are newer than {self.cutoff}\n'
'WARNING - this will happen in a separate transaction so a failure will not roll back prior cleanup'
)
with connection.cursor() as cursor:
cursor.execute(f'DROP TABLE _unpartitioned_{tblname};')
def _delete_unpartitioned_events(self, model, pk_list):
"If unpartitioned job events remain, it will cascade those from jobs in pk_list"
tblname = unified_job_class_to_event_table_name(model)
rel_name = model().event_parent_key
# Bail if the unpartitioned table does not exist anymore
if not self.has_unpartitioned_table(model):
return
# Table still exists, delete individual unpartitioned events
if pk_list:
with connection.cursor() as cursor:
self.logger.debug(f'Deleting {len(pk_list)} events from _unpartitioned_{tblname}, use a longer cleanup window to delete the table.')
pk_list_csv = ','.join(map(str, pk_list))
cursor.execute(f"DELETE FROM _unpartitioned_{tblname} WHERE {rel_name} IN ({pk_list_csv});")
def cleanup_jobs(self):
# Hack to avoid doing N+1 queries as each item in the Job query set does
# an individual query to get the underlying UnifiedJob.
Job.polymorphic_super_sub_accessors_replaced = True
skipped = (Job.objects.filter(created__gte=self.cutoff) | Job.objects.filter(status__in=['pending', 'waiting', 'running'])).count()
qs = Job.objects.select_related('unifiedjob_ptr').filter(created__lt=self.cutoff).exclude(status__in=['pending', 'waiting', 'running'])
if self.dry_run:
deleted = qs.count()
return skipped, deleted
deleted = 0
info = qs.aggregate(min=Min('id'), max=Max('id'))
if info['min'] is not None:
for start in range(info['min'], info['max'] + 1, self.batch_size):
qs_batch = qs.filter(id__gte=start, id__lte=start + self.batch_size)
pk_list = list(qs_batch.values_list('id', flat=True))
_pre_delete_job_host_summaries(pk_list, self.logger)
_, results = qs_batch.delete()
deleted += results['main.Job']
# Avoid dropping the job event table in case we have interacted with it already
self._delete_unpartitioned_events(Job, pk_list)
return skipped, deleted
def cleanup_ad_hoc_commands(self):
skipped, deleted = 0, 0
ad_hoc_commands = AdHocCommand.objects.filter(created__lt=self.cutoff)
pk_list = []
for ad_hoc_command in ad_hoc_commands.iterator():
ad_hoc_command_display = '"%s" (%d events)' % (str(ad_hoc_command), ad_hoc_command.ad_hoc_command_events.count())
if ad_hoc_command.status in ('pending', 'waiting', 'running'):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s ad hoc command %s', action_text, ad_hoc_command.status, ad_hoc_command_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, ad_hoc_command_display)
if not self.dry_run:
pk_list.append(ad_hoc_command.pk)
ad_hoc_command.delete()
deleted += 1
if not self.dry_run:
self._delete_unpartitioned_events(AdHocCommand, pk_list)
skipped += AdHocCommand.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def cleanup_project_updates(self):
skipped, deleted = 0, 0
project_updates = ProjectUpdate.objects.filter(created__lt=self.cutoff)
pk_list = []
for pu in project_updates.iterator():
pu_display = '"%s" (type %s)' % (str(pu), str(pu.launch_type))
if pu.status in ('pending', 'waiting', 'running'):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s project update %s', action_text, pu.status, pu_display)
skipped += 1
elif pu in (pu.project.current_update, pu.project.last_update) and pu.project.scm_type:
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s', action_text, pu_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, pu_display)
if not self.dry_run:
pk_list.append(pu.pk)
pu.delete()
deleted += 1
if not self.dry_run:
self._delete_unpartitioned_events(ProjectUpdate, pk_list)
skipped += ProjectUpdate.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def cleanup_inventory_updates(self):
skipped, deleted = 0, 0
inventory_updates = InventoryUpdate.objects.filter(created__lt=self.cutoff)
pk_list = []
for iu in inventory_updates.iterator():
iu_display = '"%s" (source %s)' % (str(iu), str(iu.source))
if iu.status in ('pending', 'waiting', 'running'):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s inventory update %s', action_text, iu.status, iu_display)
skipped += 1
elif iu in (iu.inventory_source.current_update, iu.inventory_source.last_update) and iu.inventory_source.source:
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s', action_text, iu_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, iu_display)
if not self.dry_run:
pk_list.append(iu.pk)
iu.delete()
deleted += 1
if not self.dry_run:
self._delete_unpartitioned_events(InventoryUpdate, pk_list)
skipped += InventoryUpdate.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def cleanup_management_jobs(self):
skipped, deleted = 0, 0
system_jobs = SystemJob.objects.filter(created__lt=self.cutoff)
pk_list = []
for sj in system_jobs.iterator():
sj_display = '"%s" (type %s)' % (str(sj), str(sj.job_type))
if sj.status in ('pending', 'waiting', 'running'):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s system_job %s', action_text, sj.status, sj_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, sj_display)
if not self.dry_run:
pk_list.append(sj.pk)
sj.delete()
deleted += 1
if not self.dry_run:
self._delete_unpartitioned_events(SystemJob, pk_list)
skipped += SystemJob.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def cleanup_workflow_jobs(self):
skipped, deleted = 0, 0
workflow_jobs = WorkflowJob.objects.filter(created__lt=self.cutoff)
for workflow_job in workflow_jobs.iterator():
workflow_job_display = '"{}" ({} nodes)'.format(str(workflow_job), workflow_job.workflow_nodes.count())
if workflow_job.status in ('pending', 'waiting', 'running'):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s job %s', action_text, workflow_job.status, workflow_job_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, workflow_job_display)
if not self.dry_run:
workflow_job.delete()
deleted += 1
skipped += WorkflowJob.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def cleanup_notifications(self):
skipped, deleted = 0, 0
notifications = Notification.objects.filter(created__lt=self.cutoff)
for notification in notifications.iterator():
notification_display = '"{}" (started {}, {} type, {} sent)'.format(
str(notification), str(notification.created), notification.notification_type, notification.notifications_sent
)
if notification.status in ('pending',):
action_text = 'would skip' if self.dry_run else 'skipping'
self.logger.debug('%s %s notification %s', action_text, notification.status, notification_display)
skipped += 1
else:
action_text = 'would delete' if self.dry_run else 'deleting'
self.logger.info('%s %s', action_text, notification_display)
if not self.dry_run:
notification.delete()
deleted += 1
skipped += Notification.objects.filter(created__gte=self.cutoff).count()
return skipped, deleted
def handle(self, *args, **options):
self.verbosity = int(options.get('verbosity', 1))
self.init_logging()
self.days = int(options.get('days', 90))
self.dry_run = bool(options.get('dry_run', False))
self.batch_size = int(options.get('batch_size', 100000))
try:
self.cutoff = now() - datetime.timedelta(days=self.days)
except OverflowError:
raise CommandError('--days specified is too large. Try something less than 99999 (about 270 years).')
model_names = ('jobs', 'ad_hoc_commands', 'project_updates', 'inventory_updates', 'management_jobs', 'workflow_jobs', 'notifications')
models_to_cleanup = set()
for m in model_names:
if options.get('only_%s' % m, False):
models_to_cleanup.add(m)
if not models_to_cleanup:
models_to_cleanup.update(model_names)
# Completely disconnect all signal handlers. This is very aggressive,
# but it will be ok since this command is run in its own process. The
# core of the logic is borrowed from Signal.disconnect().
for s in (pre_save, post_save, pre_delete, post_delete, m2m_changed):
with s.lock:
del s.receivers[:]
s.sender_receivers_cache.clear()
with transaction.atomic():
for m in models_to_cleanup:
skipped, deleted = getattr(self, 'cleanup_%s' % m)()
func = getattr(self, 'cleanup_%s_partition' % m, None)
if func:
skipped_partition, deleted_partition = func()
skipped += skipped_partition
deleted += deleted_partition
if self.dry_run:
self.logger.log(99, '%s: %d would be deleted, %d would be skipped.', m.replace('_', ' '), deleted, skipped)
else:
self.logger.log(99, '%s: %d deleted, %d skipped.', m.replace('_', ' '), deleted, skipped)
# Deleting unpartitioned tables cannot be done in same transaction as updates to related tables
if not self.dry_run:
with transaction.atomic():
for m in models_to_cleanup:
unified_job_class_name = m[:-1].title().replace('Management', 'System').replace('_', '')
unified_job_class = apps.get_model('main', unified_job_class_name)
try:
unified_job_class().event_class
except (NotImplementedError, AttributeError):
continue # no need to run this for models without events
self._delete_unpartitioned_table(unified_job_class)