ubs全称是utilization-based scheduler，算是对scheduler filter的一种扩展实现。计算节点通过resource_tracker收集监控的资源，存入数据库的compute_nodes的metric字段。
这里演示根据ceph osd系统盘的数量来优先调度的场景(纯属demo)
环境： rdo L版
1、在nova setup.cfg中增加一条记录（rpm包装出来的是entry_points.txt）
vim /usr/lib/python2.7/site-packages/nova-4.0rc6-py2.7.egg-info/entry_points.txt [nova.compute.monitors.osd] virt_driver = nova.compute.monitors.osd.virt_driver:Monitor # compute.monitor下已经有cpu的实现了
2、加载osd monitor
vim /usr/lib/python2.7/site-packages/nova/compute/monitors/__init__.py class MonitorHandler(object): NAMESPACES = [ 'nova.compute.monitors.cpu', 'nova.compute.monitors.osd', # 增加osd monitor ]
3、
vim /usr/lib/python2.7/site-packages/nova/compute/monitors/base.py # 增加MonitorBase_OSD类、OSDMonitorBase类
@six.add_metaclass(abc.ABCMeta)
class MonitorBase_OSD(object):
"""Base class for all resource monitor plugins."""
def __init__(self, compute_manager):
self.compute_manager = compute_manager
self.source = None
@abc.abstractmethod
def get_metric(self, name):
"""Return a (value, timestamp) tuple for the supplied metric name.
:param name: The name/key for the metric to grab the value for.
"""
raise NotImplementedError('get_metric')
@abc.abstractmethod
def get_metric_names(self):
"""Get available metric names.
Get available metric names, which are represented by a set of keys
that can be used to check conflicts and duplications
:returns: set containing one or more values from
:py:attr: nova.objects.fields.MonitorMetricType.ALL
"""
raise NotImplementedError('get_metric_names')
def add_metrics_to_list(self, metrics_list):
"""Adds metric objects to a supplied list object.
:param metric_list: nova.objects.MonitorMetricList that the monitor
plugin should append nova.objects.MonitorMetric
objects to.
"""
metric_names = self.get_metric_names()
metrics = []
for name in metric_names:
value, timestamp = self.get_metric(name)
metric = objects.OSDMonitorMetric(name=name,
value=value,
timestamp=timestamp,
source=self.source)
metrics.append(metric)
metrics_list.objects.extend(metrics)
class OSDMonitorBase(MonitorBase_OSD):
"""Base class for all monitors that return CPU-related metrics."""
def get_metric_names(self):
return set([
fields.OSDMonitorMetricType.OSD_NUM,
])4、创建monitor osd目录
[root@node_172_16_214_111 ~(keystone_admin)]# ll /usr/lib/python2.7/site-packages/nova/compute/monitors/osd/
total 12
-rw-r--r--. 1 root root 0 Sep 17 07:35 __init__.py
-rw-r--r--. 1 root root 157 Sep 17 09:49 __init__.pyc
-rw-r--r--. 1 root root 2648 Sep 17 15:33 virt_driver.py
-rw-r--r--. 1 root root 2711 Sep 17 09:49 virt_driver.pyc
[root@node_172_16_214_111 ~(keystone_admin)]# cat /usr/lib/python2.7/site-packages/nova/compute/monitors/osd/virt_driver.py
"""
CPU monitor based on virt driver to retrieve CPU information
"""
import pyudev
import re
from oslo_config import cfg
from oslo_log import log as logging
from oslo_utils import timeutils
from nova.compute.monitors import base
from nova import exception
from nova.i18n import _LE
CONF = cfg.CONF
CONF.import_opt('compute_driver', 'nova.virt.driver')
LOG = logging.getLogger(__name__)
class Monitor(base.OSDMonitorBase):
"""CPU monitor that uses the virt driver's get_host_cpu_stats() call."""
def __init__(self, resource_tracker):
super(Monitor, self).__init__(resource_tracker)
self.source = CONF.compute_driver
self.driver = resource_tracker.driver
self._data = {}
self._cpu_stats = {}
def get_metric(self, name):
self._update_data()
return self._data[name], self._data["timestamp"]
def _update_data(self): # 采集数据的就是这个函数
# Don't allow to call this function so frequently (<= 1 sec)
now = timeutils.utcnow()
if self._data.get("timestamp") is not None:
delta = now - self._data.get("timestamp")
if delta.seconds <= 1:
return
self._data = {}
self._data["timestamp"] = now
# Extract node's CPU statistics.
try:
context = pyudev.Context()
pattern = re.compile('^/dev/vd[a-z]')
osd_devices = []
for device in context.list_devices(DEVTYPE='disk'):
major = device['MAJOR']
if major == '8' or re.search(pattern, device.device_node):
osd_devices.append(device.device_node)
self._data["osd.num"] = len(osd_devices)
except (NotImplementedError, TypeError, KeyError):
LOG.exception(_LE("Not all properties needed are implemented "
"in the compute driver"))
raise exception.ResourceMonitorError(
monitor=self.__class__.__name__)5、
vim /usr/lib/python2.7/site-packages/nova/objects/__init__.py
__import__('nova.objects.monitor_metric')
__import__('nova.objects.monitor_osd_metric') # 增加新object，跟数据库交互6、创建新的object类
[root@node_172_16_214_111 ~(keystone_admin)]# cat /usr/lib/python2.7/site-packages/nova/objects/monitor_osd_metric.py
from oslo_serialization import jsonutils
from oslo_utils import timeutils
from nova.objects import base
from nova.objects import fields
from nova import utils
# NOTE(jwcroppe): Used to determine which fields whose value we need to adjust
# (read: divide by 100.0) before sending information to the RPC notifier since
# these values were expected to be within the range [0, 1].
FIELDS_REQUIRING_CONVERSION = []
@base.NovaObjectRegistry.register
class OSDMonitorMetric(base.NovaObject):
# Version 1.0: Initial version
# Version 1.1: Added NUMA support
VERSION = '1.1'
fields = {
'name': fields.OSDMonitorMetricTypeField(nullable=False),
'value': fields.IntegerField(nullable=False),
'numa_membw_values': fields.DictOfIntegersField(nullable=True),
'timestamp': fields.DateTimeField(nullable=False),
# This will be the stevedore extension full class name
# for the plugin from which the metric originates.
'source': fields.StringField(nullable=False),
}
def obj_make_compatible(self, primitive, target_version):
super(OSDMonitorMetric, self).obj_make_compatible(primitive,
target_version)
target_version = utils.convert_version_to_tuple(target_version)
if target_version < (1, 1) and 'numa_nodes_values' in primitive:
del primitive['numa_membw_values']
# NOTE(jaypipes): This method exists to convert the object to the
# format expected by the RPC notifier for metrics events.
def to_dict(self):
dict_to_return = {
'name': self.name,
# NOTE(jaypipes): This is what jsonutils.dumps() does to
# datetime.datetime objects, which is what timestamp is in
# this object as well as the original simple dict metrics
'timestamp': timeutils.strtime(self.timestamp),
'source': self.source,
}
if self.obj_attr_is_set('value'):
if self.name in FIELDS_REQUIRING_CONVERSION:
dict_to_return['value'] = self.value / 100.0
else:
dict_to_return['value'] = self.value
elif self.obj_attr_is_set('numa_membw_values'):
dict_to_return['numa_membw_values'] = self.numa_membw_values
return dict_to_return
@base.NovaObjectRegistry.register
class OSDMonitorMetricList(base.ObjectListBase, base.NovaObject):
# Version 1.0: Initial version
# Version 1.1: MonitorMetric version 1.1
VERSION = '1.1'
fields = {
'objects': fields.ListOfObjectsField('OSDMonitorMetric'),
}
obj_relationships = {
'objects': [('1.0', '1.0'), ('1.1', '1.1')],
}
@classmethod
def from_json(cls, metrics):
"""Converts a legacy json object into a list of MonitorMetric objs
and finally returns of MonitorMetricList
:param metrics: a string of json serialized objects
:returns: a MonitorMetricList Object.
"""
metrics = jsonutils.loads(metrics) if metrics else []
metric_list = [
OSDMonitorMetric(**metric) for metric in metrics]
return OSDMonitorMetricList(objects=metric_list)
# NOTE(jaypipes): This method exists to convert the object to the
# format expected by the RPC notifier for metrics events.
def to_list(self):
return [m.to_dict() for m in self.objects]7、定义OSDMonitorMetricTypeField
vim /usr/lib/python2.7/site-packages/nova/objects/fields.py class OSDMonitorMetricTypeField(BaseEnumField): AUTO_TYPE = OSDMonitorMetricType() class OSDMonitorMetricType(Enum): OSD_NUM = "osd.num" ALL = ( OSD_NUM, ) def __init__(self): super(OSDMonitorMetricType, self).__init__( valid_values=OSDMonitorMetricType.ALL)
8、修改metrics获取方式
vim /usr/lib/python2.7/site-packages/nova/scheduler/host_manager.py def update_from_compute_node(self, compute): # self.metrics = objects.MonitorMetricList.from_json(compute.metrics) # 先注释掉cpu monitor self.metrics = objects.OSDMonitorMetricList.from_json(compute.metrics) vim /usr/lib/python2.7/site-packages/nova/compute/resource_tracker.py # 先注释掉cpu monitor def _get_host_metrics(self, context, nodename): """Get the metrics from monitors and notify information to message bus. """ # metrics = objects.MonitorMetricList() metrics = objects.OSDMonitorMetricList()
9、修改nova配置文件
vim /etc/nova/nova.conf [METRICS] # determine how metrics are weighed: score = -1.0 * cpu.percent # # the list format is weight_setting = name1=1.0, name2=-1.0 weight_setting = osd.num=5.0 [DEFAULT] compute_monitors=osd.virt_driver scheduler_weight_classes=nova.scheduler.weights.metrics.MetricsWeigher scheduler_host_subset_size = 1
scheduler知识点小记：
/usr/lib/python2.7/site-packages/nova/scheduler/driver.py # 默认host_manager driver是nova.scheduler.host_manager.HostManager
scheduler_driver_opts = [
cfg.StrOpt('scheduler_host_manager',
default='nova.scheduler.host_manager.HostManager',
help='The scheduler host manager class to use'),
]
/usr/lib/python2.7/site-packages/nova/scheduler/manager.py # 默认scheduler driver是nova.scheduler.filter_scheduler.FilterScheduler
scheduler_driver_opts = [
cfg.StrOpt('scheduler_driver',
default='nova.scheduler.filter_scheduler.FilterScheduler',
help='Default driver to use for the scheduler'),
cfg.IntOpt('scheduler_driver_task_period',
default=60,
help='How often (in seconds) to run periodic tasks in '
'the scheduler driver of your choice. '
'Please note this is likely to interact with the value '
'of service_down_time, but exactly how they interact '
'will depend on your choice of scheduler driver.'),
]
/usr/lib/python2.7/site-packages/nova/scheduler/filter_scheduler.py # 过滤出主机节点
def _schedule(self, context, request_spec, filter_properties):
"""Returns a list of hosts that meet the required specs,
ordered by their fitness.
"""
elevated = context.elevated()
instance_properties = request_spec['instance_properties']
# NOTE(danms): Instance here is still a dict, which is converted from
# an object. The pci_requests are a dict as well. Convert this when
# we get an object all the way to this path.
# TODO(sbauza): Will be fixed later by the RequestSpec object
pci_requests = instance_properties.get('pci_requests')
if pci_requests:
pci_requests = (
objects.InstancePCIRequests.from_request_spec_instance_props(
pci_requests))
instance_properties['pci_requests'] = pci_requests
instance_type = request_spec.get("instance_type", None)
update_group_hosts = filter_properties.get('group_updated', False)
config_options = self._get_configuration_options()
filter_properties.update({'context': context,
'request_spec': request_spec,
'config_options': config_options,
'instance_type': instance_type})
# Find our local list of acceptable hosts by repeatedly
# filtering and weighing our options. Each time we choose a
# host, we virtually consume resources on it so subsequent
# selections can adjust accordingly.
# Note: remember, we are using an iterator here. So only
# traverse this list once. This can bite you if the hosts
# are being scanned in a filter or weighing function.
hosts = self._get_all_host_states(elevated)
selected_hosts = []
num_instances = request_spec.get('num_instances', 1)
for num in range(num_instances):
# Filter local hosts based on requirements ...
hosts = self.host_manager.get_filtered_hosts(hosts, # 先经过filter driver过滤
filter_properties, index=num)
if not hosts:
# Can't get any more locally.
break
LOG.debug("Filtered %(hosts)s", {'hosts': hosts})
weighed_hosts = self.host_manager.get_weighed_hosts(hosts, # 然后再计算hosts权重列表(从大到小排列)
filter_properties)
LOG.debug("Weighed %(hosts)s", {'hosts': weighed_hosts})
scheduler_host_subset_size = CONF.scheduler_host_subset_size
if scheduler_host_subset_size > len(weighed_hosts):
scheduler_host_subset_size = len(weighed_hosts)
if scheduler_host_subset_size < 1:
scheduler_host_subset_size = 1
chosen_host = random.choice(
weighed_hosts[0:scheduler_host_subset_size])
LOG.debug("Selected host: %(host)s", {'host': chosen_host})
selected_hosts.append(chosen_host)
# Now consume the resources so the filter/weights
# will change for the next instance.
chosen_host.obj.consume_from_instance(instance_properties)
if update_group_hosts is True:
# NOTE(sbauza): Group details are serialized into a list now
# that they are populated by the conductor, we need to
# deserialize them
if isinstance(filter_properties['group_hosts'], list):
filter_properties['group_hosts'] = set(
filter_properties['group_hosts'])
filter_properties['group_hosts'].add(chosen_host.obj.host)
return selected_hosts参考链接
https://01.org/sites/default/files/utilization_based_scheduing_in_openstack_compute_nova-revision002.pdf
免责声明：本站发布的内容（图片、视频和文字）以原创、转载和分享为主，文章观点不代表本网站立场，如果涉及侵权请联系站长邮箱：is@yisu.com进行举报，并提供相关证据，一经查实，将立刻删除涉嫌侵权内容。
