zware/awx
1
0
Fork 0
mirror of https://github.com/ZwareBear/awx.git synced 2026-05-02 12:51:48 -05:00
Code Issues Packages Projects Releases Wiki Activity

fail all jobs on an offline node

Browse Source
This commit is contained in:
Chris Meyers
2017-08-14 15:41:33 -04:00
parent 9314db646b
commit de82707581
3 changed files with 141 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
awx/main/tests/functional/task_management/test_scheduler.py
View File

@@ -8,6 +8,7 @@ from django.utils.timezone import now as tz_now
from awx.main.scheduler import TaskManager
from awx.main.models import (
Job,
Instance,
)
@@ -223,21 +224,33 @@ class TestReaper():
def all_jobs(self, mocker):
now = tz_now()
Instance.objects.create(hostname='host1', capacity=100)
Instance.objects.create(hostname='host2', capacity=100)
Instance.objects.create(hostname='host3_split', capacity=100)
Instance.objects.create(hostname='host4_offline', capacity=0)
j1 = Job.objects.create(status='pending', execution_node='host1')
j2 = Job.objects.create(status='waiting', celery_task_id='considered_j2', execution_node='host1')
j3 = Job.objects.create(status='waiting', celery_task_id='considered_j3', execution_node='host1')
j3.modified = now - timedelta(seconds=60)
j3.save(update_fields=['modified'])
j4 = Job.objects.create(status='running', celery_task_id='considered_j4', execution_node='host1')
j5 = Job.objects.create(status='waiting', celery_task_id='reapable_j5', execution_node='host2')
j5 = Job.objects.create(status='waiting', celery_task_id='reapable_j5', execution_node='host1')
j5.modified = now - timedelta(seconds=60)
j5.save(update_fields=['modified'])
j6 = Job.objects.create(status='waiting', celery_task_id='host2_j6', execution_node='host2_split')
j6 = Job.objects.create(status='waiting', celery_task_id='considered_j6', execution_node='host2')
j6.modified = now - timedelta(seconds=60)
j6.save(update_fields=['modified'])
j7 = Job.objects.create(status='running', celery_task_id='host2_j6', execution_node='host2_split')
j7 = Job.objects.create(status='running', celery_task_id='considered_j7', execution_node='host2')
j8 = Job.objects.create(status='running', celery_task_id='reapable_j7', execution_node='host2')
j9 = Job.objects.create(status='waiting', celery_task_id='host3_j8', execution_node='host3_split')
j9.modified = now - timedelta(seconds=60)
j9.save(update_fields=['modified'])
j10 = Job.objects.create(status='running', execution_node='host3_split')
js = [j1, j2, j3, j4, j5, j6, j7]
j11 = Job.objects.create(status='running', celery_task_id='host4_j11', execution_node='host4_offline')
js = [j1, j2, j3, j4, j5, j6, j7, j8, j9, j10, j11]
for j in js:
j.save = mocker.Mock(wraps=j.save)
j.websocket_emit_status = mocker.Mock()
@@ -245,11 +258,20 @@ class TestReaper():
@pytest.fixture
def considered_jobs(self, all_jobs):
return all_jobs[2:4] + [all_jobs[4]]
return all_jobs[2:7] + [all_jobs[10]]
@pytest.fixture
def running_tasks(self, all_jobs):
return {
'host1': all_jobs[2:5],
'host2': all_jobs[5:8],
'host3_split': all_jobs[8:10],
'host4_offline': [all_jobs[10]],
}
@pytest.fixture
def reapable_jobs(self, all_jobs):
return [all_jobs[4]]
return [all_jobs[4], all_jobs[7], all_jobs[10]]
@pytest.fixture
def unconsidered_jobs(self, all_jobs):
@@ -259,16 +281,16 @@ class TestReaper():
def active_tasks(self):
return ([], {
'host1': ['considered_j2', 'considered_j3', 'considered_j4',],
'host2_split': ['host2_j6', 'host2_j7'],
'host2': ['considered_j6', 'considered_j7'],
})
@pytest.mark.django_db
@mock.patch('awx.main.tasks._send_notification_templates')
@mock.patch.object(TaskManager, 'get_active_tasks', lambda self: ([], []))
def test_cleanup_inconsistent_task(self, notify, active_tasks, considered_jobs, reapable_jobs, mocker):
def test_cleanup_inconsistent_task(self, notify, active_tasks, considered_jobs, reapable_jobs, running_tasks, mocker):
tm = TaskManager()
#tm.get_running_tasks = mocker.Mock(return_value=considered_jobs)
tm.get_running_tasks = mocker.Mock(return_value=running_tasks)
tm.get_active_tasks = mocker.Mock(return_value=active_tasks)
tm.cleanup_inconsistent_celery_tasks()
@@ -277,15 +299,16 @@ class TestReaper():
if j not in reapable_jobs:
j.save.assert_not_called()
for reaped_job in reapable_jobs:
notify.assert_called_once_with(reaped_job, 'failed')
reaped_job.websocket_emit_status.assert_called_once_with('failed')
assert reaped_job.status == 'failed'
assert reaped_job.job_explanation == (
assert notify.call_count == 3
notify.assert_has_calls([mock.call(j, 'failed') for j in reapable_jobs], any_order=True)
for j in reapable_jobs:
j.websocket_emit_status.assert_called_once_with('failed')
assert j.status == 'failed'
assert j.job_explanation == (
'Task was marked as running in Tower but was not present in Celery, so it has been marked as failed.'
)
@pytest.mark.django_db
def test_get_running_tasks(self, all_jobs):
tm = TaskManager()
@@ -293,13 +316,18 @@ class TestReaper():
# Ensure the query grabs the expected jobs
execution_nodes_jobs = tm.get_running_tasks()
assert 'host1' in execution_nodes_jobs
assert 'host2_split' in execution_nodes_jobs
assert 'host2' in execution_nodes_jobs
assert 'host3_split' in execution_nodes_jobs
assert all_jobs[1] in execution_nodes_jobs['host1']
assert all_jobs[2] in execution_nodes_jobs['host1']
assert all_jobs[3] in execution_nodes_jobs['host1']
assert all_jobs[4] in execution_nodes_jobs['host1']
assert all_jobs[5] in execution_nodes_jobs['host2_split']
assert all_jobs[6] in execution_nodes_jobs['host2_split']
assert all_jobs[5] in execution_nodes_jobs['host2']
assert all_jobs[6] in execution_nodes_jobs['host2']
assert all_jobs[7] in execution_nodes_jobs['host2']
assert all_jobs[8] in execution_nodes_jobs['host3_split']
assert all_jobs[9] in execution_nodes_jobs['host3_split']
assert all_jobs[10] in execution_nodes_jobs['host4_offline']

52
awx/main/tests/unit/test_task_manager.py Normal file
View File

@@ -0,0 +1,52 @@
# Copyright (c) 2017 Ansible by Red Hat
# All Rights Reserved.
import mock
import pytest
from django.utils.timezone import now as tz_now
from django.db import DatabaseError
from awx.main.scheduler import TaskManager
from awx.main.models import (
Job,
Instance,
InstanceGroup,
)
from django.core.cache import cache
class TestCleanupInconsistentCeleryTasks():
@mock.patch.object(cache, 'get', return_value=None)
@mock.patch.object(TaskManager, 'get_active_tasks', return_value=([], {}))
@mock.patch.object(TaskManager, 'get_running_tasks', return_value={'host1': [Job(id=2), Job(id=3),]})
@mock.patch.object(InstanceGroup.objects, 'all', return_value=[])
@mock.patch.object(Instance.objects, 'get', side_effect=Instance.DoesNotExist)
@mock.patch('awx.main.scheduler.logger')
def test_instance_does_not_exist(self, logger_mock, *args):
logger_mock.error = mock.MagicMock(side_effect=RuntimeError("mocked"))
tm = TaskManager()
with pytest.raises(RuntimeError) as excinfo:
tm.cleanup_inconsistent_celery_tasks()
assert "mocked" in str(excinfo.value)
logger_mock.error.assert_called_once_with("Execution node Instance host1 not found in database. "
"The node is currently executing jobs ['None-2-new', "
"'None-3-new']")
@mock.patch.object(cache, 'get', return_value=None)
@mock.patch.object(TaskManager, 'get_active_tasks', return_value=([], {'host1': []}))
@mock.patch.object(InstanceGroup.objects, 'all', return_value=[])
@mock.patch.object(TaskManager, 'get_running_tasks')
@mock.patch('awx.main.scheduler.logger')
def test_save_failed(self, logger_mock, get_running_tasks, *args):
logger_mock.error = mock.MagicMock()
job = Job(id=2, modified=tz_now(), status='running', celery_task_id='blah', execution_node='host1')
job.websocket_emit_status = mock.MagicMock()
get_running_tasks.return_value = {'host1': [job]}
tm = TaskManager()
with mock.patch.object(job, 'save', side_effect=DatabaseError):
tm.cleanup_inconsistent_celery_tasks()
job.save.assert_called_once()
logger_mock.error.assert_called_once_with("Task job 2 (failed) DB error in marking failed. Job possibly deleted.")