9e7db3860c
This is a bit of a scheduler refactoring to support future scaling work as a part of blueprint scaling-zones. Also fixes bug 891971 (remove unused set_network_host in scheduler) With or without that blueprint work, this cleans a number of things up and paves the way for combining a lot of the schedulers by using this new 'HostManager' for filtering and weighing (future work :) On to the goodies: Introduces new HostManager, splitting code out from ZoneManager. Zone communication and management is handlded in the ZoneManager. Host filtering and weighing is handled in the HostManager. ZoneManager is removed from the SchedulerManager and direct calls to it from the SchedulerManager now occur via the scheduler driver. This simplifies a number of things. ZoneManager and HostManager classes to use are now flags. This allows one to extend the current classes and use them. HostManager uses a HostState class to keep info about hosts. This class needs to be extendable. Since it's very much tied to the HostManager, the HostState class to use is not a flag. It is, instead, a class variable in HostManager. Filtering functions now accept a single host to filter. This improves performance by not having to build a new array of hosts within every filter function. Filtering functions now accept a more generic 'filter_properties' dictionary which we can fill with information available for filtering. Adding additional data to this 'filter_properties' can be done by subclassing. Weighing functions now accept this 'filter_properties', also, although it's poorly named ATM. Will be cleaned up in a future branch when I move some weighing functions into the host manager. Filtering tests are true unit tests now. test_zones was moved from top level to under scheduler as zone_manager tests and refactored to be true unit tests. Host tests are true unit tests now. Other minor cleanups Change-Id: I0ef2acef6639b4500c400c18cf2c673cb80f0150
122 lines
4.4 KiB
Python
122 lines
4.4 KiB
Python
# 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 flags
|
|
from nova import log as logging
|
|
|
|
LOG = logging.getLogger('nova.scheduler.least_cost')
|
|
|
|
FLAGS = flags.FLAGS
|
|
flags.DEFINE_list('least_cost_functions',
|
|
['nova.scheduler.least_cost.compute_fill_first_cost_fn'],
|
|
'Which cost functions the LeastCostScheduler should use.')
|
|
|
|
|
|
# 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)
|
|
flags.DEFINE_float('noop_cost_fn_weight', 1.0,
|
|
'How much weight to give the noop cost function')
|
|
flags.DEFINE_float('compute_fill_first_cost_fn_weight', 1.0,
|
|
'How much weight to give the fill-first cost function')
|
|
|
|
|
|
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, blob=None, zone=None):
|
|
self.weight = weight
|
|
self.blob = blob
|
|
self.zone = zone
|
|
|
|
# Local members. These are not returned outside of the Zone.
|
|
self.host_state = host_state
|
|
|
|
def to_dict(self):
|
|
x = dict(weight=self.weight)
|
|
if self.blob:
|
|
x['blob'] = self.blob
|
|
if self.host_state:
|
|
x['host'] = self.host_state.host
|
|
if self.zone:
|
|
x['zone'] = self.zone
|
|
return x
|
|
|
|
|
|
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 will less free
|
|
ram 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.
|
|
|
|
host_list - [(host, HostInfo()), ...]
|
|
weighted_fns - list of weights and functions like:
|
|
[(weight, objective-functions), ...]
|
|
weighing_properties is an arbitrary dict of values that can influence
|
|
weights.
|
|
|
|
Returns a single WeightedHost object which represents the best
|
|
candidate.
|
|
"""
|
|
|
|
# Make a grid of functions results.
|
|
# One row per host. One column per function.
|
|
scores = []
|
|
for weight, fn in weighted_fns:
|
|
scores.append([fn(host_state, weighing_properties)
|
|
for host_state in host_states])
|
|
|
|
# Adjust the weights in the grid by the functions weight adjustment
|
|
# and sum them up to get a final list of weights.
|
|
adjusted_scores = []
|
|
for (weight, fn), row in zip(weighted_fns, scores):
|
|
adjusted_scores.append([weight * score for score in row])
|
|
|
|
# Now, sum down the columns to get the final score. Column per host.
|
|
final_scores = [0.0] * len(host_states)
|
|
for row in adjusted_scores:
|
|
for idx, col in enumerate(row):
|
|
final_scores[idx] += col
|
|
|
|
# Super-impose the host_state into the scores so
|
|
# we don't lose it when we sort.
|
|
final_scores = [(final_scores[idx], host_state)
|
|
for idx, host_state in enumerate(host_states)]
|
|
|
|
final_scores = sorted(final_scores)
|
|
weight, host_state = final_scores[0] # Lowest score is the winner!
|
|
return WeightedHost(weight, host_state=host_state)
|