def test_service_check_ok(monkeypatch):
instance = {'prometheus_url': 'http://localhost:10252/metrics'}
instance_tags = []
check = KubeControllerManagerCheck(CHECK_NAME, {}, [instance])
monkeypatch.setattr(check, 'service_check', mock.Mock())
calls = [
mock.call('kube_controller_manager.up',
AgentCheck.OK,
tags=instance_tags),
mock.call('kube_controller_manager.up',
AgentCheck.CRITICAL,
tags=instance_tags,
message='health check failed'),
]
# successful health check
with mock.patch("requests.get",
return_value=mock.MagicMock(status_code=200)):
check._perform_service_check(instance)
# failed health check
raise_error = mock.Mock()
raise_error.side_effect = requests.HTTPError('health check failed')
with mock.patch("requests.get",
return_value=mock.MagicMock(raise_for_status=raise_error)):
check._perform_service_check(instance)
check.service_check.assert_has_calls(calls)
def test_check_metrics(aggregator, mock_metrics, mock_leader):
c = KubeControllerManagerCheck(CHECK_NAME, None, {}, [instance])
c.check(instance)
def assert_metric(name, **kwargs):
# Wrapper to keep assertions < 120 chars
aggregator.assert_metric(NAMESPACE + name, **kwargs)
assert_metric('.goroutines')
assert_metric('.threads')
assert_metric('.open_fds')
assert_metric('.client.http.requests')
assert_metric('.max_fds')
assert_metric('.nodes.evictions',
metric_type=aggregator.MONOTONIC_COUNT,
value=33,
tags=["zone:test"])
assert_metric('.nodes.count', value=5, tags=["zone:test"])
assert_metric('.nodes.unhealthy', value=1, tags=["zone:test"])
assert_metric('.rate_limiter.use',
value=1,
tags=["limiter:job_controller"])
assert_metric('.rate_limiter.use',
value=0,
tags=["limiter:daemon_controller"])
assert_metric('.queue.adds',
metric_type=aggregator.MONOTONIC_COUNT,
value=29,
tags=["queue:replicaset"])
assert_metric('.queue.depth',
metric_type=aggregator.GAUGE,
value=3,
tags=["queue:service"])
assert_metric('.queue.retries',
metric_type=aggregator.MONOTONIC_COUNT,
value=13,
tags=["queue:deployment"])
assert_metric('.queue.work_duration.sum',
value=255667,
tags=["queue:replicaset"])
assert_metric('.queue.work_duration.count',
value=29,
tags=["queue:replicaset"])
assert_metric('.queue.work_duration.quantile',
value=110,
tags=["queue:replicaset", "quantile:0.5"])
assert_metric('.queue.latency.sum',
value=423889,
tags=["queue:deployment"])
assert_metric('.queue.latency.count', value=29, tags=["queue:deployment"])
assert_metric('.queue.latency.quantile',
value=1005,
tags=["queue:deployment", "quantile:0.9"])
# Extra name from the instance
assert_metric('.rate_limiter.use',
value=0,
tags=["limiter:extra_controller"])
assert_metric('.queue.adds',
metric_type=aggregator.MONOTONIC_COUNT,
value=13,
tags=["queue:extra"])
assert_metric('.queue.depth',
metric_type=aggregator.GAUGE,
value=2,
tags=["queue:extra"])
assert_metric('.queue.retries',
metric_type=aggregator.MONOTONIC_COUNT,
value=55,
tags=["queue:extra"])
assert_metric('.queue.work_duration.sum',
value=45171,
tags=["queue:extra"])
assert_metric('.queue.work_duration.count', value=13, tags=["queue:extra"])
assert_metric('.queue.work_duration.quantile',
value=6,
tags=["queue:extra", "quantile:0.5"])
assert_metric('.queue.latency.sum', value=9309, tags=["queue:extra"])
assert_metric('.queue.latency.count', value=13, tags=["queue:extra"])
assert_metric('.queue.latency.quantile',
value=10,
tags=["queue:extra", "quantile:0.9"])
# Leader election mixin
expected_le_tags = [
"record_kind:endpoints", "record_name:kube-controller-manager",
"record_namespace:kube-system"
]
assert_metric('.leader_election.transitions',
value=3,
tags=expected_le_tags)
assert_metric('.leader_election.lease_duration',
value=15,
tags=expected_le_tags)
aggregator.assert_service_check(NAMESPACE + ".leader_election.status",
tags=expected_le_tags)
aggregator.assert_all_metrics_covered()
def test_check_metrics_without_deprecated(aggregator, mock_metrics,
mock_leader):
c = KubeControllerManagerCheck(CHECK_NAME, {}, [instance])
c.check(instance2)
generic_check_metrics(aggregator, False)