deb-sahara/sahara/plugins/recommendations_utils.py
Vitaly Gridnev 3873a512f7 Disable autotune configs for scaling old clusters
Implemented option for disabling scaling for old clusters.
After autoconfiguration we will put 'auto-configured' flag into
cluster.extra field. That will allow us to disable autotune configs
for old created clusters.

Partial-Implements blueprint: recommend-configuration

Change-Id: I7d09ef20da51a170e9c021032c8bf13c34ae8d85
2015-08-27 15:56:14 +03:00

373 lines
15 KiB
Python

# Copyright (c) 2015 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 abc
from oslo_log import log as logging
import six
from sahara import conductor as cond
from sahara import context
from sahara.utils.openstack import nova
conductor = cond.API
LOG = logging.getLogger(__name__)
@six.add_metaclass(abc.ABCMeta)
class AutoConfigsProvider(object):
def __init__(self, mapper, plugin_configs, cluster, scaling):
"""This meta class provides general recommendation utils for cluster
configuration.
:param mapper: dictionary, that describes which cluster configs and
node_configs to configure. It should maps to following dicts:
node_configs to configure and cluster_configs to configure. This
dicts should contains abstract names of configs as keys and
tuple (correct_applicable_target, correct_name) as values. Such
representation allows to use same AutoConfigsProvider for plugins
with almost same configs and configuring principles.
:param plugin_configs: all plugins_configs for specified plugin
:param cluster: cluster which is required to configure
:param scaling: indicates that current cluster operation is scaling
"""
self.plugin_configs = plugin_configs
self.cluster = cluster
self.node_configs_to_update = mapper.get('node_configs', {})
self.cluster_configs_to_update = mapper.get('cluster_configs', {})
self.scaling = scaling
@abc.abstractmethod
def _get_recommended_node_configs(self, node_group):
"""Method calculates and returns recommended configs for node_group.
It's not required to update node_configs of node_group using the
conductor api in this method, because it will be done in the method
apply_node_configs.
:param node_group: NodeGroup Sahara resource.
:return: dictionary with calculated recommended configs for
node_group.
"""
pass
@abc.abstractmethod
def _get_recommended_cluster_configs(self):
"""Method calculates and returns recommended configs for cluster.
It's not required to update cluster_configs of cluster using the
conductor api in this method, because it will be done in the method
apply_cluster_configs.
:return: dictionary with calculated recommended configs for
cluster.
"""
pass
def _can_be_recommended(self, configs_list, node_group=None):
"""Method calculates and returns True, when it's possible to
automatically configure provided list of configs configs_list.
Otherwise, method should return False.
:param configs_list: list of configs which we want to configure
:param node_group: optional argument, which should be provided if
some config can be used in node_configs of some node_group
:return: True if all configs can be configured and False otherwise
"""
# cluster configs is Frozen Dict, so let's call to_dict()
cl_configs = self.cluster.cluster_configs.to_dict()
for ncfg in configs_list:
section, name = self._get_correct_section_and_name(ncfg)
if section in cl_configs and name in cl_configs[section]:
return False
if not node_group:
return True
cl_configs = node_group.node_configs.to_dict()
for ncfg in configs_list:
section, name = self._get_correct_section_and_name(ncfg)
if section in cl_configs and name in cl_configs[section]:
return False
return True
def _get_correct_section_and_name(self, config_name):
"""Calculates and returns correct applicable target and name from
abstract name of config.
:param config_name: abstract name of config.
:return: correct applicable target and name for config_name
"""
section, name = None, None
if config_name in self.cluster_configs_to_update:
section = self.cluster_configs_to_update[config_name][0]
name = self.cluster_configs_to_update[config_name][1]
elif config_name in self.node_configs_to_update:
section = self.node_configs_to_update[config_name][0]
name = self.node_configs_to_update[config_name][1]
return section, name
def _get_default_config_value(self, config_name):
"""Calculates and return default value of config from
abstract name of config.
:param config_name: abstract name of config.
:return: default config value for config_name.
"""
section, name = self._get_correct_section_and_name(config_name)
for config in self.plugin_configs:
if config.applicable_target == section and config.name == name:
return config.default_value
def _merge_configs(self, current_configs, proposed_configs):
"""Correctly merges old configs and new extra configs"""
result = {}
for (section, configs) in six.iteritems(proposed_configs):
cfg_values = {}
if section in current_configs:
cfg_values = (current_configs[section] if
current_configs[section] else {})
cfg_values.update(configs)
result.update({section: cfg_values})
for (section, configs) in six.iteritems(current_configs):
if section not in result:
result.update({section: configs})
return result
def _get_cluster_extra(self):
cluster = self.cluster
return cluster.extra.to_dict() if cluster.extra else {}
def finalize_autoconfiguration(self):
if not self.cluster.use_autoconfig:
return
cluster_extra = self._get_cluster_extra()
cluster_extra['auto-configured'] = True
conductor.cluster_update(
context.ctx(), self.cluster, {'extra': cluster_extra})
def apply_node_configs(self, node_group):
"""Method applies configs for node_group using conductor api,
which were calculated with recommend_node_configs method.
:param node_group: NodeGroup Sahara resource.
:return: None.
"""
if not node_group.use_autoconfig or not self.cluster.use_autoconfig:
return
to_update = self.node_configs_to_update
recommended_node_configs = self._get_recommended_node_configs(
node_group)
if not recommended_node_configs:
# Nothing to configure
return
current_dict = node_group.node_configs.to_dict()
configuration = {}
for ncfg in six.iterkeys(to_update):
if ncfg not in recommended_node_configs:
continue
n_section = to_update[ncfg][0]
n_name = to_update[ncfg][1]
proposed_config_value = recommended_node_configs[ncfg]
if n_section not in configuration:
configuration.update({n_section: {}})
configuration[n_section].update({n_name: proposed_config_value})
current_dict = self._merge_configs(current_dict, configuration)
conductor.node_group_update(context.ctx(), node_group,
{'node_configs': current_dict})
def apply_cluster_configs(self):
"""Method applies configs for cluster using conductor api, which were
calculated with recommend_cluster_configs method.
:return: None.
"""
cluster = self.cluster
if not cluster.use_autoconfig:
return
to_update = self.cluster_configs_to_update
recommended_cluster_configs = self._get_recommended_cluster_configs()
if not recommended_cluster_configs:
# Nothing to configure
return
current_dict = cluster.cluster_configs.to_dict()
configuration = {}
for ncfg in six.iterkeys(to_update):
if ncfg not in recommended_cluster_configs:
continue
n_section = to_update[ncfg][0]
n_name = to_update[ncfg][1]
proposed_config_value = recommended_cluster_configs[ncfg]
if n_section not in configuration:
configuration.update({n_section: {}})
configuration[n_section].update({n_name: proposed_config_value})
current_dict = self._merge_configs(current_dict, configuration)
conductor.cluster_update(context.ctx(), cluster,
{'cluster_configs': current_dict})
def apply_recommended_configs(self):
"""Method applies recommended configs for cluster and for all
node_groups using conductor api.
:return: None.
"""
if self.scaling:
# Validate cluster is not an old created cluster
cluster_extra = self._get_cluster_extra()
if 'auto-configured' not in cluster_extra:
# Don't configure
return
for ng in self.cluster.node_groups:
self.apply_node_configs(ng)
self.apply_cluster_configs()
configs = list(self.cluster_configs_to_update.keys())
configs.extend(list(self.node_configs_to_update.keys()))
LOG.debug("Following configs were auto-configured: {configs}".format(
configs=configs))
self.finalize_autoconfiguration()
class HadoopAutoConfigsProvider(AutoConfigsProvider):
def __init__(self, mapper, plugin_configs, cluster, scaling, hbase=False):
super(HadoopAutoConfigsProvider, self).__init__(
mapper, plugin_configs, cluster, scaling)
self.requested_flavors = {}
self.is_hbase_enabled = hbase
def _get_java_opts(self, value):
return "-Xmx%dm" % int(value)
def _transform_mb_to_gb(self, mb):
return mb / 1024.
def _transform_gb_to_mb(self, gb):
return gb * 1024.
def _get_min_size_of_container(self, ram):
if ram <= 4:
return 256
if ram <= 8:
return 512
if ram <= 24:
return 1024
return 2048
def _get_os_ram_recommendation(self, ram):
upper_bounds = [4, 8, 16, 24, 48, 64, 72, 96, 128, 256]
reserve_for_os = [1, 2, 2, 4, 6, 8, 8, 12, 24, 32]
for (upper, reserve) in zip(upper_bounds, reserve_for_os):
if ram <= upper:
return reserve
return 64
def _get_hbase_ram_recommendations(self, ram):
if not self.is_hbase_enabled:
return 0
upper_bounds = [4, 8, 16, 24, 48, 64, 72, 96, 128, 256]
reserve_for_hbase = [1, 1, 2, 4, 8, 8, 8, 16, 24, 32]
for (upper, reserve) in zip(upper_bounds, reserve_for_hbase):
if ram <= upper:
return reserve
return 64
def _get_node_group_data(self, node_group):
if node_group.flavor_id not in self.requested_flavors:
flavor = nova.get_flavor(id=node_group.flavor_id)
self.requested_flavors[node_group.flavor_id] = flavor
else:
flavor = self.requested_flavors[node_group.flavor_id]
cpu = flavor.vcpus
ram = flavor.ram
data = {}
# config recommendations was taken from Ambari code
os = self._get_os_ram_recommendation(self._transform_mb_to_gb(ram))
hbase = self._get_hbase_ram_recommendations(
self._transform_mb_to_gb(ram))
reserved_ram = self._transform_gb_to_mb(os + hbase)
min_container_size = self._get_min_size_of_container(
self._transform_mb_to_gb(ram))
# we use large amount of containers to allow users to run
# at least 4 jobs at same time on clusters based on small flavors
data["containers"] = int(max(
8, min(2 * cpu, ram / min_container_size)))
data["ramPerContainer"] = (ram - reserved_ram) / data["containers"]
data["ramPerContainer"] = max(data["ramPerContainer"],
min_container_size)
data["ramPerContainer"] = min(2048, int(data["ramPerContainer"]))
data["ramPerContainer"] = int(data["ramPerContainer"])
data["mapMemory"] = int(data["ramPerContainer"])
data["reduceMemory"] = int(2 * data["ramPerContainer"])
data["amMemory"] = int(min(data["mapMemory"], data["reduceMemory"]))
return data
def _get_recommended_node_configs(self, node_group):
"""Calculates recommended MapReduce and YARN configs for specified
node_group.
:param node_group: NodeGroup Sahara resource
:return: dictionary with recommended MapReduce and YARN configs
"""
configs_to_update = list(self.node_configs_to_update.keys())
if not self._can_be_recommended(configs_to_update, node_group):
return {}
data = self._get_node_group_data(node_group)
r = {}
r['yarn.nodemanager.resource.memory-mb'] = (data['containers'] *
data['ramPerContainer'])
r['yarn.scheduler.minimum-allocation-mb'] = data['ramPerContainer']
r['yarn.scheduler.maximum-allocation-mb'] = (data['containers'] *
data['ramPerContainer'])
r['yarn.nodemanager.vmem-check-enabled'] = "false"
r['yarn.app.mapreduce.am.resource.mb'] = data['amMemory']
r['yarn.app.mapreduce.am.command-opts'] = self._get_java_opts(
0.8 * data['amMemory'])
r['mapreduce.map.memory.mb'] = data['mapMemory']
r['mapreduce.reduce.memory.mb'] = data['reduceMemory']
r['mapreduce.map.java.opts'] = self._get_java_opts(
0.8 * data['mapMemory'])
r['mapreduce.reduce.java.opts'] = self._get_java_opts(
0.8 * data['reduceMemory'])
r['mapreduce.task.io.sort.mb'] = int(min(
0.4 * data['mapMemory'], 1024))
return r
def get_datanode_name(self):
return "datanode"
def _get_recommended_cluster_configs(self):
"""Method recommends dfs_replication for cluster.
:return: recommended value of dfs_replication.
"""
if not self._can_be_recommended(['dfs.replication']):
return {}
datanode_count = 0
datanode_proc_name = self.get_datanode_name()
for ng in self.cluster.node_groups:
if datanode_proc_name in ng.node_processes:
datanode_count += ng.count
replica = 'dfs.replication'
recommended_value = self._get_default_config_value(replica)
if recommended_value:
return {replica: min(recommended_value, datanode_count)}
else:
return {}