PR#4032: implicit task refusals

Merges #4032
https://pagure.io/koji/pull-request/4032

https://pagure.io/koji/issue/4035
This commit is contained in:
Tomas Kopecek 2024-03-06 11:14:20 +01:00
commit 7d126abe91
2 changed files with 39 additions and 7 deletions

View file

@ -387,12 +387,14 @@ class TaskScheduler(object):
# TODO fix
if task['state'] == koji.TASK_STATES['ASSIGNED']:
# TODO check time since assigned
# if not taken within a timeout
# - if host not checking in, then make sure host marked unavail and free
# - if host *is* checking in, then treat as refusal and free
age = time.time() - min([r['create_ts'] for r in taskruns])
assign_ts = min([r['create_ts'] for r in taskruns])
age = time.time() - assign_ts
if age > self.assign_timeout:
# has the host checked in since we assigned?
if host['update_ts'] and host['update_ts'] > assign_ts:
# treat this as an implicit refusal
# possibly an older koji version on builder
set_refusal(host['id'], task['task_id'], msg='assignment timeout')
log_both('Task assignment timeout', task_id=task['task_id'],
host_id=host['id'])
kojihub.Task(task['task_id']).free()
@ -438,6 +440,9 @@ class TaskScheduler(object):
values={'host_ids': [h['id'] for h in hosts_to_mark]},
)
update.execute()
# also update our data
for host in hosts_to_mark:
host['ready'] = False
def get_active_runs(self):
runs = get_task_runs([["active", True]])

View file

@ -266,6 +266,7 @@ class TestCheckActiveRuns(BaseTest):
mock.patch('kojihub.Task.assign', new=my_assign).start()
self.log_db = mock.MagicMock()
mock.patch('kojihub.scheduler.log_db', new=self.log_db).start()
self.set_refusal = mock.patch('kojihub.scheduler.set_refusal').start()
def test_check_no_active(self):
self.assertEqual(self.sched.active_tasks, []) # set by init
@ -311,14 +312,40 @@ class TestCheckActiveRuns(BaseTest):
def test_check_assign_timeout(self):
# 'Task assignment timeout' case
create_ts = 0
create_ts = 1000
update_ts = 999 # host ts BEFORE assignment
now = 1000000
self.sched.active_tasks = [{'task_id': 99, 'host_id': 23, 'state': koji.TASK_STATES['ASSIGNED']}]
self.sched.hosts = {23: {'id': 23, 'name': 'test host 23'}}
self.sched.hosts = {23: {'id': 23, 'name': 'test host 23', 'update_ts': update_ts}}
self.sched.get_active_runs.return_value = {99: [{'create_ts': create_ts}]}
with mock.patch('time.time', return_value=now):
self.sched.check_active_tasks()
self.get_active_runs.assert_called_once()
self.set_refusal.assert_not_called()
self.log_db.assert_called_once_with('Task assignment timeout', 99, 23)
# we should free such tasks
self.assertEqual(self.frees, [99])
self.assertEqual(self.assigns, [])
self.assertEqual(len(self.updates), 1)
update = self.updates[0]
self.assertEqual(update.table, 'scheduler_task_runs')
def test_check_implicit_refusal(self):
# 'Task assignment timeout' case
create_ts = 1000
update_ts = 1001 # host ts AFTER assignment
now = 1000000
self.sched.active_tasks = [{'task_id': 99, 'host_id': 23, 'state': koji.TASK_STATES['ASSIGNED']}]
self.sched.hosts = {23: {'id': 23, 'name': 'test host 23', 'update_ts': update_ts}}
self.sched.get_active_runs.return_value = {99: [{'create_ts': create_ts}]}
with mock.patch('time.time', return_value=now):
self.sched.check_active_tasks()
self.get_active_runs.assert_called_once()
self.set_refusal.assert_called_once_with(23, 99, msg='assignment timeout')
self.log_db.assert_called_once_with('Task assignment timeout', 99, 23)
# we should free such tasks
self.assertEqual(self.frees, [99])