def test_get_queued_resources_with_no_resources(self):
"""Tests successfully getting queued resource information when all cluster resources is empty."""
resource_mgr.refresh_agent_resources([], now())
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
resource_mgr.update_all_cluster_resources()
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
queued_resources = resource_mgr.get_queued_resources()
self.assertDictEqual(
queued_resources, {
"cluster_resources": {},
"queue_lengths": {
'PENDING': 0,
'QUEUED': 3,
'RUNNING': 0
},
"total_resources": {
'PENDING': {},
'QUEUED': {
'cpus': 3.0,
'mem': 384.0
},
'RUNNING': {}
}
})
def test_max_resources(self):
"""Tests successfully calculating the max resources in a cluster"""
offer_1 = ResourceOffer(
'offer_1', self.agent_1.agent_id, self.framework_id,
NodeResources([Cpus(2.0), Mem(22048.0),
Disk(1024.0)]), now(), None)
offer_2 = ResourceOffer(
'offer_2', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(25.0), Mem(2048.0),
Disk(2048.0)]), now(), None)
offer_3 = ResourceOffer(
'offer_3', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(225.0),
Mem(1024.0),
Disk(22048.0)]), now(), None)
resource_mgr.add_new_offers([offer_1, offer_2, offer_3])
resource_mgr.refresh_agent_resources([], now())
max = resource_mgr.get_max_available_resources()
self.assertTrue(
max.is_equal(
NodeResources([Cpus(250.0),
Mem(22048.0),
Disk(24096.0)])))
def test_update_all_cluster_resources(self):
"""Tests successfully updating the all cluster resources database in a cluster"""
offer_1 = ResourceOffer(
'offer_1', self.agent_1.agent_id, self.framework_id,
NodeResources([Cpus(2.0), Mem(22048.0),
Disk(1024.0)]), now(), None)
offer_2 = ResourceOffer(
'offer_2', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(25.0), Mem(2048.0),
Disk(2048.0)]), now(), None)
offer_3 = ResourceOffer(
'offer_3', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(225.0),
Mem(1024.0),
Disk(22048.0)]), now(), None)
resource_mgr.add_new_offers([offer_1, offer_2, offer_3])
resource_mgr.refresh_agent_resources([], now())
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
resource_mgr.update_all_cluster_resources()
resource_db = ClusterResources.objects.first()
self.assertIsNotNone(resource_db)
self.assertEqual(resource_db.mem, 25120.0)
self.assertEqual(resource_db.gpus, 0.0)
self.assertEqual(resource_db.disk, 25120.0)
self.assertEqual(resource_db.cpus, 252.0)
def test_all_available_resources(self):
"""Tests successfully calculating the available resources in a cluster"""
offer_1 = ResourceOffer(
'offer_1', self.agent_1.agent_id, self.framework_id,
NodeResources([Cpus(2.0), Mem(22048.0),
Disk(1024.0)]), now(), None)
offer_2 = ResourceOffer(
'offer_2', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(25.0), Mem(2048.0),
Disk(2048.0)]), now(), None)
offer_3 = ResourceOffer(
'offer_3', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(225.0),
Mem(1024.0),
Disk(22048.0)]), now(), None)
resource_mgr.add_new_offers([offer_1, offer_2, offer_3])
resource_mgr.refresh_agent_resources([], now())
all_available_resources = resource_mgr.get_all_available_resources()
self.assertDictEqual(all_available_resources, {
'mem': 25120.0,
'gpus': 0.0,
'disk': 25120.0,
'cpus': 252.0
})
def test_get_queued_resources(self):
"""Tests successfully getting queued resource information"""
offer_1 = ResourceOffer(
'offer_1', self.agent_1.agent_id, self.framework_id,
NodeResources([Cpus(2.0), Mem(22048.0),
Disk(1024.0)]), now(), None)
offer_2 = ResourceOffer(
'offer_2', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(25.0), Mem(2048.0),
Disk(2048.0)]), now(), None)
offer_3 = ResourceOffer(
'offer_3', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(225.0),
Mem(1024.0),
Disk(22048.0)]), now(), None)
resource_mgr.add_new_offers([offer_1, offer_2, offer_3])
resource_mgr.refresh_agent_resources([], now())
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
resource_mgr.update_all_cluster_resources()
resource_db = ClusterResources.objects.first()
self.assertIsNotNone(resource_db)
self.assertEqual(resource_db.mem, 25120.0)
self.assertEqual(resource_db.gpus, 0.0)
self.assertEqual(resource_db.disk, 25120.0)
self.assertEqual(resource_db.cpus, 252.0)
queued_resources = resource_mgr.get_queued_resources()
self.assertDictEqual(
queued_resources, {
"cluster_resources": {
'cpus': 252,
'disk': 25120,
'gpus': 0,
'mem': 25120
},
"queue_lengths": {
'PENDING': 0,
'QUEUED': 3,
'RUNNING': 0
},
"total_resources": {
'PENDING': {},
'QUEUED': {
'cpus': 3.0,
'mem': 384.0
},
'RUNNING': {}
}
})
def test_update_all_cluster_resources_no_resources(self):
"""Tests updating the all cluster resources database when none are defined."""
resource_mgr.refresh_agent_resources([], now())
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
resource_mgr.update_all_cluster_resources()
resource_db = ClusterResources.objects.first()
self.assertIsNone(resource_db)
def setUp(self):
django.setup()
resource_mgr.clear()
self.agent_1 = Agent('agent_1', 'host_1')
self.agent_2 = Agent('agent_2', 'host_2')
self.framework_id = '1234'
offer_1 = ResourceOffer(
'offer_1', self.agent_1.agent_id, self.framework_id,
NodeResources([Cpus(2.0), Mem(1024.0),
Disk(1024.0)]), now(), None)
offer_2 = ResourceOffer(
'offer_2', self.agent_2.agent_id, self.framework_id,
NodeResources([Cpus(25.0), Mem(2048.0),
Disk(2048.0)]), now(), None)
resource_mgr.add_new_offers([offer_1, offer_2])
resource_mgr.refresh_agent_resources([], now())
def _prepare_nodes(self, tasks, running_job_exes, when):
"""Prepares the nodes to use for scheduling
:param tasks: The current current running
:type tasks: list
:param running_job_exes: The currently running job executions
:type running_job_exes: list
:param when: The current time
:type when: :class:`datetime.datetime`
:returns: The dict of scheduling nodes stored by node ID
:rtype: dict
"""
nodes = node_mgr.get_nodes()
# Group tasks by agent ID
tasks_by_agent_id = {} # {Agent ID: List of tasks}
for task in tasks:
if task.agent_id not in tasks_by_agent_id:
tasks_by_agent_id[task.agent_id] = [task]
else:
tasks_by_agent_id[task.agent_id].append(task)
# Group job executions by node ID
running_exes_by_node_id = {} # {Node ID: List of running job exes}
for running_job_exe in running_job_exes:
if running_job_exe.node_id not in running_exes_by_node_id:
running_exes_by_node_id[running_job_exe.node_id] = [
running_job_exe
]
else:
running_exes_by_node_id[running_job_exe.node_id].append(
running_job_exe)
agent_resources = resource_mgr.refresh_agent_resources(tasks, when)
scheduling_nodes = {} # {Node ID: SchedulingNode}
for node in nodes:
agent_id = node.agent_id # Grab agent ID once since it could change while we are scheduling
if agent_id in tasks_by_agent_id:
node_tasks = tasks_by_agent_id[agent_id]
else:
node_tasks = []
if node.id in running_exes_by_node_id:
node_exes = running_exes_by_node_id[node.id]
else:
node_exes = []
if agent_id in agent_resources:
resource_set = agent_resources[agent_id]
else:
resource_set = ResourceSet()
scheduling_node = SchedulingNode(agent_id, node, node_tasks,
node_exes, resource_set)
scheduling_nodes[scheduling_node.node_id] = scheduling_node
return scheduling_nodes
