def test_process_nodes_status(self):
obj = json.loads(self.get_data_test_input("data/nodes_list.json"))
gauge = watchdog.gen_pai_node_gauge()
watchdog.process_nodes_status(gauge, obj)
self.assertTrue(len(gauge.samples) > 0)
def test_process_dlws_nodes_status_with_unscheduable(self):
obj = json.loads(
self.get_data_test_input(
"data/dlws_nodes_list_with_unschedulable.json"))
pod_info = collections.defaultdict(lambda: [])
pod_info["192.168.255.1"].append(watchdog.PodInfo("job1", False, 2))
gauges = watchdog.process_nodes_status(obj, pod_info,
watchdog.ClusterGPUInfo())
self.assertEqual(5, len(gauges))
self.assertEqual("k8s_node_count", gauges[0].name)
self.assertEqual(1, len(gauges[0].samples))
self.assertEqual("true", gauges[0].samples[0].labels["unschedulable"])
self.assertEqual("k8s_node_gpu_available", gauges[1].name)
self.assertEqual(1, len(gauges[1].samples))
self.assertEqual(0, gauges[1].samples[0].value)
self.assertEqual("k8s_node_preemptable_gpu_available", gauges[2].name)
self.assertEqual(1, len(gauges[2].samples))
self.assertEqual(0, gauges[2].samples[0].value)
self.assertEqual("k8s_node_gpu_total", gauges[3].name)
self.assertEqual(1, len(gauges[3].samples))
self.assertEqual(4, gauges[3].samples[0].value)
self.assertEqual("k8s_node_gpu_allocatable", gauges[4].name)
self.assertEqual(1, len(gauges[4].samples))
self.assertEqual(0, gauges[4].samples[0].value)
for gauge in gauges:
self.assertTrue(len(gauge.samples) > 0)
for gauge in gauges[1:]:
self.assertEqual("192.168.255.1",
gauge.samples[0].labels["host_ip"])
def test_process_dlws_nodes_status(self):
obj = json.loads(self.get_data_test_input("data/dlws_nodes_list.json"))
pod_info = collections.defaultdict(lambda: [])
pod_info["192.168.255.1"].append(watchdog.PodInfo("job1", 2))
gauges = watchdog.process_nodes_status(obj, pod_info)
self.assertTrue(len(gauges) == 4)
self.assertEqual("k8s_node_gpu_available", gauges[1].name)
self.assertEqual(1, len(gauges[1].samples))
self.assertEqual(2, gauges[1].samples[0].value)
self.assertEqual("k8s_node_gpu_total", gauges[2].name)
self.assertEqual(1, len(gauges[2].samples))
self.assertEqual(4, gauges[2].samples[0].value)
self.assertEqual("k8s_node_gpu_reserved", gauges[3].name)
self.assertEqual(1, len(gauges[3].samples))
self.assertEqual(0, gauges[3].samples[0].value)
for gauge in gauges:
self.assertTrue(len(gauge.samples) > 0)
for gauge in gauges[1:]:
self.assertEqual("192.168.255.1",
gauge.samples[0].labels["host_ip"])
def test_process_nodes_status(self):
obj = json.loads(self.get_data_test_input("data/nodes_list.json"))
gauges = watchdog.process_nodes_status(obj, {})
self.assertTrue(len(gauges) == 4)
for gauge in gauges:
self.assertTrue(len(gauge.samples) > 0)
def test_process_nodes_status(self):
obj = json.loads(self.get_data_test_input("data/nodes_list.json"))
gauges = watchdog.process_nodes_status(obj, {},
watchdog.ClusterGPUInfo())
self.assertEqual(5, len(gauges))
for gauge in gauges:
self.assertTrue(len(gauge.samples) > 0)