fuel-qa/fuelweb_test/tests/test_cpu_pinning.py
Ksenia Demina f26c960e3a Add test for CPU allocation
It's verify that nailgun correctly handles unsuitable CPU values

Change-Id: Ie16297a2b7d6631e5a58b5aaa318b611b597c3cd
Implements: blueprint test-numa-cpu-pinning
(cherry picked from commit ac35383515)
2016-04-26 08:39:02 +00:00

584 lines
23 KiB
Python

# Copyright 2016 Mirantis, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
import random
from proboscis import asserts
from proboscis import test
# pylint: disable=import-error
from six.moves.urllib.error import HTTPError
# pylint: enable=import-error
from fuelweb_test.helpers.decorators import log_snapshot_after_test
from fuelweb_test.helpers import os_actions
from fuelweb_test.helpers import utils
from fuelweb_test import logger
from fuelweb_test import logwrap
from fuelweb_test import settings
from fuelweb_test.tests.base_test_case import SetupEnvironment
from fuelweb_test.tests.base_test_case import TestBasic
from gates_tests.helpers import exceptions
@test(groups=["numa_cpu_pinning"])
class NumaCpuPinning(TestBasic):
"""NumaCpuPinning."""
@staticmethod
@logwrap
def assert_entry_in_config(conf, conf_name, section, option, value):
"""Check entry of parameter with a proper value.
:param conf: a ConfigParser object
:param conf_name: a string of full file path
:param section: a string of section name in configuration file
:param option: a string of option name in configuration file
:param value: a string of value that has entry in configuration file
:return:
"""
current_value = conf.get(section, option)
asserts.assert_true(value in current_value,
'Expected that the option "{0}" contains value '
'"{1}" in config file "{2}", but actually has '
'value "{3}": FAIL'.format(option,
value,
conf_name,
current_value))
@staticmethod
@logwrap
def assert_quantity_in_config(conf, conf_name, section, option,
value):
"""Check number of parameters in option section.
:param conf: a ConfigParser object
:param conf_name: a string of full file path
:param section: a string of section name in configuration file
:param option: a string of option name in configuration file
:param value: an int number of values in specific option
:return:
"""
current_value = conf.get(section, option)
asserts.assert_equal(len(current_value.split(',')), value,
'Expected that the option "{0}" has "{1}"'
'values in config file {2} but actually has '
'value "{3}": FAIL'.format(option,
value,
conf_name,
current_value))
@logwrap
def create_pinned_instance(self, os_conn, cluster_id,
name, vcpus, hostname, meta):
"""Boot VM on specific compute with CPU pinning
:param os_conn: an object of connection to openstack services
:param cluster_id: an integer number of cluster id
:param name: a string name of flavor and aggregate
:param vcpus: an integer number of vcpus for flavor
:param hostname: a string fqdn name of compute
:param meta: a dict with metadata for aggregate
:return:
"""
aggregate_name = name + str(random.randint(0, 1000))
aggregate = os_conn.create_aggregate(aggregate_name,
metadata=meta,
hosts=[hostname])
extra_specs = {'aggregate_instance_extra_specs:pinned': 'true',
'hw:cpu_policy': 'dedicated'}
net_name = self.fuel_web.get_cluster_predefined_networks_name(
cluster_id)['private_net']
flavor_id = random.randint(10, 10000)
flavor = os_conn.create_flavor(name=name, ram=64, vcpus=vcpus, disk=1,
flavorid=flavor_id,
extra_specs=extra_specs)
server = os_conn.create_server_for_migration(neutron=True,
label=net_name,
flavor=flavor_id)
os_conn.verify_instance_status(server, 'ACTIVE')
os_conn.delete_instance(server)
os_conn.delete_flavor(flavor)
os_conn.delete_aggregate(aggregate, hosts=[hostname])
@test(depends_on=[SetupEnvironment.prepare_slaves_5],
groups=["numa_cpu_pinning",
"basic_env_for_numa_cpu_pinning"])
@log_snapshot_after_test
def basic_env_for_numa_cpu_pinning(self):
"""Basic environment for NUMA CPU pinning
Scenario:
1. Create cluster
2. Add 2 nodes with compute role
3. Add 3 nodes with controller role
4. Verify that quantity of NUMA is equal on node and in Fuel
Snapshot: basic_env_for_numa_cpu_pinning
"""
snapshot_name = 'basic_env_for_numa_cpu_pinning'
self.check_run(snapshot_name)
self.env.revert_snapshot("ready_with_5_slaves")
# TODO(kdemina) Use commomn function for variables asserts
if not settings.KVM_USE:
raise exceptions.FuelQAVariableNotSet(
'KVM_USE', 'true')
if int(settings.HARDWARE['slave_node_cpu']) < 6:
raise exceptions.FuelQAVariableNotSet(
'SLAVE_NODE_CPU', 6)
if int(settings.HARDWARE['numa_nodes']) < 2:
raise exceptions.FuelQAVariableNotSet(
'NUMA_NODES', 2)
if not settings.INTERFACES_DICT['eth0'] == 'ens3':
raise exceptions.FuelQAVariableNotSet(
'IFACE_0', 'ens3')
if not settings.INTERFACES_DICT['eth1'] == 'ens4':
raise exceptions.FuelQAVariableNotSet(
'IFACE_1', 'ens4')
if not settings.INTERFACES_DICT['eth2'] == 'ens5':
raise exceptions.FuelQAVariableNotSet(
'IFACE_2', 'ens5')
elif not settings.INTERFACES_DICT['eth3'] == 'ens6':
raise exceptions.FuelQAVariableNotSet(
'IFACE_3', 'ens6')
elif not settings.INTERFACES_DICT['eth4'] == 'ens7':
raise exceptions.FuelQAVariableNotSet(
'IFACE_4', 'ens7')
elif not settings.INTERFACES_DICT['eth5'] == 'ens8':
raise exceptions.FuelQAVariableNotSet(
'IFACE_5', 'ens8')
elif not settings.ACPI_ENABLE:
raise exceptions.FuelQAVariableNotSet(
'DRIVER_ENABLE_ACPI', 'true')
self.show_step(1)
cluster_id = self.fuel_web.create_cluster(
name=self.__class__.__name__,
mode=settings.DEPLOYMENT_MODE,
settings={
"net_provider": 'neutron',
"net_segment_type": settings.NEUTRON_SEGMENT_TYPE
}
)
self.show_step(2)
self.show_step(3)
self.fuel_web.update_nodes(
cluster_id,
{
'slave-01': ['compute'],
'slave-02': ['compute'],
'slave-03': ['controller'],
'slave-04': ['controller'],
'slave-05': ['controller']
})
self.show_step(4)
for node in ('slave-01', 'slave-02'):
target_node = self.fuel_web.get_nailgun_node_by_name(node)
numas_from_fuel = len(
target_node['meta']['numa_topology']['numa_nodes'])
numas_on_remote = utils.get_quantity_of_numa(target_node['ip'])
if not numas_on_remote:
# Fuel handle topology without NUMA as 1 NUMA node
asserts.assert_equal(numas_from_fuel, 1,
"No NUMA nodes on {0} "
"while Fuel shows it "
"has {1}".format(
target_node['ip'], numas_from_fuel))
raise AssertionError("No NUMA nodes on {0}".format(
target_node['ip']))
else:
asserts.assert_equal(numas_on_remote, numas_from_fuel,
"{0} NUMA nodes on {1} "
"while Fuel shows it "
"has {2}".format(
numas_on_remote, target_node['ip'],
numas_from_fuel))
logger.info("There is {0} NUMA nodes on node {1}".format(
numas_on_remote, target_node['ip']))
self.env.make_snapshot(snapshot_name, is_make=True)
@test(depends_on_groups=['basic_env_for_numa_cpu_pinning'],
groups=["cpu_pinning_on_two_compute"])
@log_snapshot_after_test
def cpu_pinning_on_two_compute(self):
"""Check different amount of pinned CPU
Scenario:
1. Revert snapshot "basic_env_for_numa_cpu_pinning"
2. Pin maximum CPU for the nova on the first compute
3. Pin minimun CPU for the nova on the second compute
4. Verify setting was successfully applied
5. Deploy cluster
6. Check new filters are enabled in nova.conf at controller
7. Check nova.conf contains pinned CPU at compute
8. Run OSTF
9. Boot VM with pinned CPU on the first compute
10. Boot VM with pinned CPU on the second compute
Snapshot: cpu_pinning_on_two_compute
"""
snapshot_name = 'cpu_pinning_on_two_compute'
self.check_run(snapshot_name)
self.show_step(1)
self.env.revert_snapshot("basic_env_for_numa_cpu_pinning")
cluster_id = self.fuel_web.get_last_created_cluster()
self.show_step(2)
first_compute = self.fuel_web.get_nailgun_node_by_name('slave-01')
first_compute_cpu = first_compute['meta']['cpu']['total']
first_config = self.fuel_web.client.get_node_attributes(
first_compute['id'])
first_config['cpu_pinning']['nova']['value'] = first_compute_cpu - 1
self.fuel_web.client.upload_node_attributes(
first_config, first_compute['id'])
self.show_step(3)
second_compute = self.fuel_web.get_nailgun_node_by_name('slave-02')
second_config = self.fuel_web.client.get_node_attributes(
second_compute['id'])
second_config['cpu_pinning']['nova']['value'] = 1
self.fuel_web.client.upload_node_attributes(
second_config, second_compute['id'])
self.show_step(4)
first_config = self.fuel_web.client.get_node_attributes(
first_compute['id'])
asserts.assert_equal(
first_config['cpu_pinning']['nova']['value'],
first_compute_cpu - 1,
"CPU pinning wasn't applied on '{0}': "
"Expected value '{1}', actual '{2}'"
.format(first_compute['ip'], first_compute_cpu - 1,
first_config['cpu_pinning']['nova']['value']))
second_config = self.fuel_web.client.get_node_attributes(
second_compute['id'])
asserts.assert_equal(
second_config['cpu_pinning']['nova']['value'],
1,
"CPU pinning wasn't applied on '{0}': "
"Expected value '{1}', actual '{2}'"
.format(second_compute['ip'], 1,
second_config['cpu_pinning']['nova']['value']))
self.show_step(5)
self.fuel_web.deploy_cluster_wait(cluster_id)
self.show_step(6)
controllers = self.fuel_web.get_nailgun_cluster_nodes_by_roles(
cluster_id,
roles=['controller'])
nova_conf_path = "/etc/nova/nova.conf"
for controller in controllers:
with self.ssh_manager.open_on_remote(
ip=controller['ip'],
path=nova_conf_path) as f:
nova_conf = utils.get_ini_config(f)
self.assert_entry_in_config(nova_conf,
nova_conf_path,
"DEFAULT",
"scheduler_default_filters",
"NUMATopologyFilter")
self.show_step(7)
with self.ssh_manager.open_on_remote(
ip=first_compute['ip'],
path=nova_conf_path) as f:
nova_conf = utils.get_ini_config(f)
self.assert_quantity_in_config(nova_conf,
nova_conf_path,
"DEFAULT",
"vcpu_pin_set",
first_compute_cpu - 1)
with self.ssh_manager.open_on_remote(
ip=second_compute['ip'],
path=nova_conf_path) as f:
nova_conf = utils.get_ini_config(f)
self.assert_quantity_in_config(nova_conf,
nova_conf_path,
"DEFAULT",
"vcpu_pin_set",
1)
self.show_step(8)
self.fuel_web.run_ostf(cluster_id=cluster_id)
self.show_step(9)
os_conn = os_actions.OpenStackActions(
self.fuel_web.get_public_vip(cluster_id))
meta = {'pinned': 'true'}
self.create_pinned_instance(os_conn=os_conn,
cluster_id=cluster_id,
name='cpu_3_',
vcpus=3,
hostname=first_compute['fqdn'],
meta=meta)
self.show_step(10)
self.create_pinned_instance(os_conn=os_conn,
cluster_id=cluster_id,
name='cpu_1_',
vcpus=1,
hostname=second_compute['fqdn'],
meta=meta)
self.env.make_snapshot(snapshot_name, is_make=True)
@test(depends_on_groups=['basic_env_for_numa_cpu_pinning'],
groups=["cpu_pinning_with_other_role"])
@log_snapshot_after_test
def cpu_pinning_with_other_role(self):
"""Check pinned CPU on compute,cinder node
Scenario:
1. Revert snapshot "basic_env_for_numa_cpu_pinning"
2. Add cinder role for compute nodes
3. Pin maximum CPU for the nova on the computes
4. Verify setting was successfully applied
5. Deploy cluster
6. Check new filters are enabled in nova.conf at controller
7. Check nova.conf contains pinned CPU at computes
8. Run OSTF
9. Boot VMs with pinned CPU on each compute, cinder node
Snapshot: cpu_pinning_with_other_role
"""
self.show_step(1)
self.env.revert_snapshot("basic_env_for_numa_cpu_pinning")
self.show_step(2)
cluster_id = self.fuel_web.get_last_created_cluster()
nodes = {'slave-01': ['compute', 'cinder'],
'slave-02': ['compute', 'cinder']}
self.fuel_web.update_nodes(cluster_id, nodes)
self.show_step(3)
target_nodes = self.fuel_web.get_nailgun_cluster_nodes_by_roles(
cluster_id, ['compute', 'cinder'], role_status='pending_roles')
for compute in target_nodes:
compute_cpu = compute['meta']['cpu']['total']
compute_config = self.fuel_web.client.get_node_attributes(
compute['id'])
compute_config['cpu_pinning']['nova']['value'] = compute_cpu - 1
self.fuel_web.client.upload_node_attributes(
compute_config, compute['id'])
self.show_step(4)
for compute in target_nodes:
compute_cpu = compute['meta']['cpu']['total']
compute_config = self.fuel_web.client.get_node_attributes(
compute['id'])
asserts.assert_equal(
compute_config['cpu_pinning']['nova']['value'],
compute_cpu - 1,
"CPU pinning wasn't applied on '{0}': "
"Expected value '{1}', actual '{2}'"
.format(compute['ip'], compute_cpu - 1,
compute_config['cpu_pinning']['nova']['value']))
self.show_step(5)
self.fuel_web.deploy_cluster_wait(cluster_id)
self.show_step(6)
controllers = self.fuel_web.get_nailgun_cluster_nodes_by_roles(
cluster_id,
roles=['controller'])
nova_conf_path = "/etc/nova/nova.conf"
for controller in controllers:
with self.ssh_manager.open_on_remote(
ip=controller['ip'],
path=nova_conf_path) as f:
nova_conf = utils.get_ini_config(f)
self.assert_entry_in_config(nova_conf,
nova_conf_path,
"DEFAULT",
"scheduler_default_filters",
"NUMATopologyFilter")
self.show_step(7)
for compute in target_nodes:
with self.ssh_manager.open_on_remote(
ip=compute['ip'],
path=nova_conf_path) as f:
nova_conf = utils.get_ini_config(f)
compute_cpu = compute['meta']['cpu']['total']
self.assert_quantity_in_config(nova_conf,
nova_conf_path,
"DEFAULT",
"vcpu_pin_set",
compute_cpu - 1)
self.show_step(8)
self.fuel_web.run_ostf(cluster_id=cluster_id)
self.show_step(9)
os_conn = os_actions.OpenStackActions(
self.fuel_web.get_public_vip(cluster_id))
meta = {'pinned': 'true'}
for compute in target_nodes:
self.create_pinned_instance(os_conn=os_conn,
cluster_id=cluster_id,
name='cpu_role_',
vcpus=2,
hostname=compute['fqdn'],
meta=meta)
self.env.make_snapshot("cpu_pinning_with_other_role")
@test(depends_on_groups=['cpu_pinning_on_two_compute'],
groups=["reboot_cpu_pinning_compute"])
@log_snapshot_after_test
def reboot_cpu_pinning_compute(self):
"""Check compute with pinned CPU after reboot
Scenario:
1. Revert snapshot "cpu_pinning_on_two_compute"
2. Reboot the first compute with CPU pinning
3. Run OSTF
4. Boot VM with pinned CPU on the first compute
5. Reboot the second compute with CPU pinning
6. Run OSTF
7. Boot VM with pinned CPU on the second compute
Snapshot: reboot_cpu_pinning_compute
"""
self.show_step(1)
self.env.revert_snapshot("cpu_pinning_on_two_compute")
cluster_id = self.fuel_web.get_last_created_cluster()
self.show_step(2)
first_compute = self.fuel_web.get_nailgun_node_by_name('slave-01')
second_compute = self.fuel_web.get_nailgun_node_by_name('slave-02')
self.fuel_web.warm_restart_nodes(
self.fuel_web.get_devops_nodes_by_nailgun_nodes([first_compute]))
self.show_step(3)
self.fuel_web.run_ostf(cluster_id=cluster_id)
self.show_step(4)
os_conn = os_actions.OpenStackActions(
self.fuel_web.get_public_vip(cluster_id))
meta = {'pinned': 'true'}
self.create_pinned_instance(os_conn=os_conn,
cluster_id=cluster_id,
name='cpu_3_',
vcpus=3,
hostname=first_compute['fqdn'],
meta=meta)
self.show_step(5)
self.fuel_web.warm_restart_nodes(
self.fuel_web.get_devops_nodes_by_nailgun_nodes([second_compute]))
self.show_step(6)
self.fuel_web.run_ostf(cluster_id=cluster_id)
self.show_step(7)
self.create_pinned_instance(os_conn=os_conn,
cluster_id=cluster_id,
name='cpu_1_',
vcpus=1,
hostname=second_compute['fqdn'],
meta=meta)
self.env.make_snapshot('reboot_cpu_pinning_compute')
@test(depends_on_groups=['basic_env_for_numa_cpu_pinning'],
groups=["cpu_pinning_allocation"])
@log_snapshot_after_test
def cpu_pinning_allocation(self):
"""Check errors for allocation of CPU for nova
Scenario:
1. Revert snapshot "basic_env_for_numa_cpu_pinning"
2. Pin all node CPU and twice of it for the nova
3. Check status code: 400
Snapshot: cpu_pinning_allocation
"""
self.show_step(1)
self.env.revert_snapshot("basic_env_for_numa_cpu_pinning")
cluster_id = self.fuel_web.get_last_created_cluster()
self.show_step(2)
self.show_step(3)
target_nodes = self.fuel_web.get_nailgun_cluster_nodes_by_roles(
cluster_id, ['compute'], role_status='pending_roles')
for compute in target_nodes:
compute_cpu = compute['meta']['cpu']['total']
compute_config = self.fuel_web.client.get_node_attributes(
compute['id'])
compute_config['cpu_pinning']['nova']['value'] = compute_cpu
try:
self.fuel_web.client.upload_node_attributes(
compute_config, compute['id'])
except HTTPError as e:
asserts.assert_equal(400, e.code)
else:
asserts.fail("Pinned all CPU on {0}, while expecting HTTP "
"error on CPU value {1}"
.format(compute['ip'], compute_cpu))
compute_config['cpu_pinning']['nova']['value'] = compute_cpu * 2
try:
self.fuel_web.client.upload_node_attributes(
compute_config, compute['id'])
except HTTPError as e:
asserts.assert_equal(400, e.code)
else:
asserts.fail("Pinned all CPU on {0}, while expecting HTTP "
"400 error on CPU value {1}"
.format(compute['ip'], compute_cpu * 2))
self.env.make_snapshot('cpu_pinning_allocation')