Refactor scheduling weights.

This makes scheduling weights more plugin friendly and creates shared
code that can be used by the host scheduler as well as the future cells
scheduler.  Weighing classes can now be specified much like you can
specify scheduling host filters.

The new weights code reverses the old behavior where lower weights win.
Higher weights are now the winners.

The least_cost module and configs have been deprecated, but are still
supported for backwards compatibility.  The code has moved to
nova.scheduler.weights.least_cost and been modified to work with the new
loadable-class code.  If any of the least_cost related config options are
specified, this least_cost weigher will be used.

For those not overriding the default least_cost config values, the new
RamWeigher class will be used.  The default behavior of the RamWeigher
class is the same default behavior as the old least_cost module.

The new weights code introduces a new config option
'scheduler_weight_classes' which is used to specify which weigher classes
to use.  The default is 'all classes', but modified if least_cost
deprecated config options are used, as mentioned above.

The RamWeigher class introduces a new config option
'ram_weight_multiplier'.  The default of 1.0 causes weights equal to the
free memory in MB to be returned, thus hosts with more free memory are
preferred (causes equal spreading).  Changing this value to a negative
number such as -1.0 will cause reverse behavior (fill first).

DocImpact

Change-Id: I1e5e5039c299db02f7287f2d33299ebf0b9732ce
This commit is contained in:
Chris Behrens
2012-11-09 02:48:09 +00:00
parent 7d69c21189
commit 2c7eabc69d
2 changed files with 76 additions and 154 deletions

View File

@@ -1,118 +0,0 @@
# Copyright (c) 2011 OpenStack, LLC.
# All Rights Reserved.
#
# 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.
"""
Least Cost is an algorithm for choosing which host machines to
provision a set of resources to. The input is a WeightedHost object which
is decided upon by a set of objective-functions, called the 'cost-functions'.
The WeightedHost contains a combined weight for each cost-function.
The cost-function and weights are tabulated, and the host with the least cost
is then selected for provisioning.
"""
from nova import config
from nova import flags
from nova.openstack.common import cfg
from nova.openstack.common import log as logging
LOG = logging.getLogger(__name__)
least_cost_opts = [
cfg.ListOpt('least_cost_functions',
default=[
'nova.scheduler.least_cost.compute_fill_first_cost_fn'
],
help='Which cost functions the LeastCostScheduler should use'),
cfg.FloatOpt('noop_cost_fn_weight',
default=1.0,
help='How much weight to give the noop cost function'),
cfg.FloatOpt('compute_fill_first_cost_fn_weight',
default=-1.0,
help='How much weight to give the fill-first cost function. '
'A negative value will reverse behavior: '
'e.g. spread-first'),
]
CONF = config.CONF
CONF.register_opts(least_cost_opts)
# TODO(sirp): Once we have enough of these rules, we can break them out into a
# cost_functions.py file (perhaps in a least_cost_scheduler directory)
class WeightedHost(object):
"""Reduced set of information about a host that has been weighed.
This is an attempt to remove some of the ad-hoc dict structures
previously used."""
def __init__(self, weight, host_state=None):
self.weight = weight
self.host_state = host_state
def to_dict(self):
x = dict(weight=self.weight)
if self.host_state:
x['host'] = self.host_state.host
return x
def __repr__(self):
if self.host_state:
return "WeightedHost host: %s" % self.host_state.host
return "WeightedHost with no host_state"
def noop_cost_fn(host_state, weighing_properties):
"""Return a pre-weight cost of 1 for each host"""
return 1
def compute_fill_first_cost_fn(host_state, weighing_properties):
"""More free ram = higher weight. So servers with less free
ram will be preferred.
Note: the weight for this function in default configuration
is -1.0. With a -1.0 this function runs in reverse, so systems
with the most free memory will be preferred.
"""
return host_state.free_ram_mb
def weighted_sum(weighted_fns, host_states, weighing_properties):
"""Use the weighted-sum method to compute a score for an array of objects.
Normalize the results of the objective-functions so that the weights are
meaningful regardless of objective-function's range.
:param host_list: ``[(host, HostInfo()), ...]``
:param weighted_fns: list of weights and functions like::
[(weight, objective-functions), ...]
:param weighing_properties: an arbitrary dict of values that can
influence weights.
:returns: a single WeightedHost object which represents the best
candidate.
"""
min_score, best_host = None, None
for host_state in host_states:
score = sum(weight * fn(host_state, weighing_properties)
for weight, fn in weighted_fns)
if min_score is None or score < min_score:
min_score, best_host = score, host_state
return WeightedHost(min_score, host_state=best_host)

View File

@@ -1,4 +1,4 @@
# Copyright 2011 OpenStack LLC.
# Copyright 2011-2012 OpenStack LLC.
# All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
@@ -15,27 +15,51 @@
"""
Tests For Least Cost functions.
"""
from nova import config
from nova import context
from nova.scheduler import host_manager
from nova.scheduler import least_cost
from nova.openstack.common import cfg
from nova.scheduler import weights
from nova.scheduler.weights import least_cost
from nova import test
from nova.tests import matchers
from nova.tests.scheduler import fakes
def offset(hostinfo, options):
test_least_cost_opts = [
cfg.FloatOpt('compute_fake_weigher1_weight',
default=2.0,
help='How much weight to give the fake_weigher1 function'),
cfg.FloatOpt('compute_fake_weigher2_weight',
default=1.0,
help='How much weight to give the fake_weigher2 function'),
]
CONF = config.CONF
CONF.import_opt('least_cost_functions', 'nova.scheduler.weights.least_cost')
CONF.import_opt('compute_fill_first_cost_fn_weight',
'nova.scheduler.weights.least_cost')
CONF.register_opts(test_least_cost_opts)
def compute_fake_weigher1(hostinfo, options):
return hostinfo.free_ram_mb + 10000
def scale(hostinfo, options):
def compute_fake_weigher2(hostinfo, options):
return hostinfo.free_ram_mb * 2
class LeastCostTestCase(test.TestCase):
def setUp(self):
super(LeastCostTestCase, self).setUp()
self.flags(reserved_host_disk_mb=0, reserved_host_memory_mb=0)
self.host_manager = fakes.FakeHostManager()
self.weight_handler = weights.HostWeightHandler()
def _get_weighed_host(self, hosts, weight_properties=None):
weigher_classes = least_cost.get_least_cost_weighers()
if weight_properties is None:
weight_properties = {}
return self.weight_handler.get_weighed_objects(weigher_classes,
hosts, weight_properties)[0]
def _get_all_hosts(self):
ctxt = context.get_admin_context()
@@ -46,8 +70,39 @@ class LeastCostTestCase(test.TestCase):
self.mox.ResetAll()
return host_states
def test_weighted_sum_happy_day(self):
fn_tuples = [(1.0, offset), (1.0, scale)]
def test_default_of_spread_first(self):
# Default modifier is -1.0, so it turns out that hosts with
# the most free memory win
hostinfo_list = self._get_all_hosts()
# host1: free_ram_mb=512
# host2: free_ram_mb=1024
# host3: free_ram_mb=3072
# host4: free_ram_mb=8192
# so, host1 should win:
weighed_host = self._get_weighed_host(hostinfo_list)
self.assertEqual(weighed_host.weight, 8192)
self.assertEqual(weighed_host.obj.host, 'host4')
def test_filling_first(self):
self.flags(compute_fill_first_cost_fn_weight=1.0)
hostinfo_list = self._get_all_hosts()
# host1: free_ram_mb=-512
# host2: free_ram_mb=-1024
# host3: free_ram_mb=-3072
# host4: free_ram_mb=-8192
# so, host1 should win:
weighed_host = self._get_weighed_host(hostinfo_list)
self.assertEqual(weighed_host.weight, -512)
self.assertEqual(weighed_host.obj.host, 'host1')
def test_weighted_sum_provided_method(self):
fns = ['nova.tests.scheduler.test_least_cost.compute_fake_weigher1',
'nova.tests.scheduler.test_least_cost.compute_fake_weigher2']
self.flags(least_cost_functions=fns)
hostinfo_list = self._get_all_hosts()
# host1: free_ram_mb=512
@@ -59,18 +114,17 @@ class LeastCostTestCase(test.TestCase):
# [10512, 11024, 13072, 18192]
# [1024, 2048, 6144, 16384]
# adjusted [ 1.0 * x + 1.0 * y] =
# [11536, 13072, 19216, 34576]
# adjusted [ 2.0 * x + 1.0 * y] =
# [22048, 24096, 32288, 52768]
# so, host1 should win:
options = {}
weighted_host = least_cost.weighted_sum(fn_tuples, hostinfo_list,
options)
self.assertEqual(weighted_host.weight, 11536)
self.assertEqual(weighted_host.host_state.host, 'host1')
weighed_host = self._get_weighed_host(hostinfo_list)
self.assertEqual(weighed_host.weight, 52768)
self.assertEqual(weighed_host.obj.host, 'host4')
def test_weighted_sum_single_function(self):
fn_tuples = [(1.0, offset), ]
fns = ['nova.tests.scheduler.test_least_cost.compute_fake_weigher1']
self.flags(least_cost_functions=fns)
hostinfo_list = self._get_all_hosts()
# host1: free_ram_mb=0
@@ -80,24 +134,10 @@ class LeastCostTestCase(test.TestCase):
# [offset, ]=
# [10512, 11024, 13072, 18192]
# adjusted [ 2.0 * x ]=
# [21024, 22048, 26144, 36384]
# so, host1 should win:
options = {}
weighted_host = least_cost.weighted_sum(fn_tuples, hostinfo_list,
options)
self.assertEqual(weighted_host.weight, 10512)
self.assertEqual(weighted_host.host_state.host, 'host1')
class TestWeightedHost(test.TestCase):
def test_dict_conversion_without_host_state(self):
host = least_cost.WeightedHost('someweight')
expected = {'weight': 'someweight'}
self.assertThat(host.to_dict(), matchers.DictMatches(expected))
def test_dict_conversion_with_host_state(self):
host_state = host_manager.HostState('somehost', None)
host = least_cost.WeightedHost('someweight', host_state)
expected = {'weight': 'someweight',
'host': 'somehost'}
self.assertThat(host.to_dict(), matchers.DictMatches(expected))
weighed_host = self._get_weighed_host(hostinfo_list)
self.assertEqual(weighed_host.weight, 36384)
self.assertEqual(weighed_host.obj.host, 'host4')