def test_get_compute_nodes_info(pbsnodes_mocked_response, expected_output, mocker, test_datadir):
pbsnodes_output = read_text(test_datadir / pbsnodes_mocked_response)
mock = mocker.patch(
"common.schedulers.torque_commands.check_command_output", return_value=pbsnodes_output, autospec=True
)
nodes = get_compute_nodes_info(hostname_filter=["host1"])
mock.assert_called_with("/opt/torque/bin/pbsnodes -x host1", raise_on_error=False)
assert_that(nodes).is_equal_to(expected_output)
def get_busy_nodes():
nodes = get_compute_nodes_info()
busy_nodes = 0
for node in nodes.values():
# when a node is added it transitions from down,offline,MOM-list-not-sent -> down -> free
if node.jobs or (any(state in ["offline", "state-unknown"]
for state in node.state)
and "MOM-list-not-sent" not in node.state):
busy_nodes += 1
return busy_nodes
def get_busy_nodes():
nodes = get_compute_nodes_info()
logging.info("Found the following compute nodes:\n%s", nodes)
busy_nodes = 0
for node in nodes.values():
# when a node is added it transitions from down,offline,MOM-list-not-sent -> down -> free
if node.jobs or (any(state in ["state-unknown"]
for state in node.state)
and "MOM-list-not-sent" not in node.state):
busy_nodes += 1
return busy_nodes
def is_node_down():
"""Check if node is down according to scheduler"""
try:
hostname = check_command_output("hostname").strip()
node = get_compute_nodes_info(hostname_filter=[hostname]).get(hostname)
if node:
log.info("Node is in state: '{0}'".format(node.state))
if all(error_state not in node.state
for error_state in TORQUE_NODE_ERROR_STATES):
return False
else:
log.warning("Node is not attached to scheduler. Reporting as down")
except Exception as e:
log.error(
"Failed when checking if node is down with exception %s. Reporting node as down.",
e)
return True