summaryrefslogtreecommitdiffstats
path: root/roles/openshift_health_checker/test/disk_availability_test.py
diff options
context:
space:
mode:
Diffstat (limited to 'roles/openshift_health_checker/test/disk_availability_test.py')
-rw-r--r--roles/openshift_health_checker/test/disk_availability_test.py11
1 files changed, 4 insertions, 7 deletions
diff --git a/roles/openshift_health_checker/test/disk_availability_test.py b/roles/openshift_health_checker/test/disk_availability_test.py
index 945b9eafc..e98d02c58 100644
--- a/roles/openshift_health_checker/test/disk_availability_test.py
+++ b/roles/openshift_health_checker/test/disk_availability_test.py
@@ -17,7 +17,7 @@ def test_is_active(group_names, is_active):
task_vars = dict(
group_names=group_names,
)
- assert DiskAvailability.is_active(task_vars=task_vars) == is_active
+ assert DiskAvailability(None, task_vars).is_active() == is_active
@pytest.mark.parametrize('ansible_mounts,extra_words', [
@@ -30,10 +30,9 @@ def test_cannot_determine_available_disk(ansible_mounts, extra_words):
group_names=['masters'],
ansible_mounts=ansible_mounts,
)
- check = DiskAvailability(execute_module=fake_execute_module)
with pytest.raises(OpenShiftCheckException) as excinfo:
- check.run(tmp=None, task_vars=task_vars)
+ DiskAvailability(fake_execute_module, task_vars).run()
for word in 'determine disk availability'.split() + extra_words:
assert word in str(excinfo.value)
@@ -93,8 +92,7 @@ def test_succeeds_with_recommended_disk_space(group_names, configured_min, ansib
ansible_mounts=ansible_mounts,
)
- check = DiskAvailability(execute_module=fake_execute_module)
- result = check.run(tmp=None, task_vars=task_vars)
+ result = DiskAvailability(fake_execute_module, task_vars).run()
assert not result.get('failed', False)
@@ -168,8 +166,7 @@ def test_fails_with_insufficient_disk_space(group_names, configured_min, ansible
ansible_mounts=ansible_mounts,
)
- check = DiskAvailability(execute_module=fake_execute_module)
- result = check.run(tmp=None, task_vars=task_vars)
+ result = DiskAvailability(fake_execute_module, task_vars).run()
assert result['failed']
for word in 'below recommended'.split() + extra_words: