yardstick/network_services/nfvi/resource.py

   1 # Copyright (c) 2016-2017 Intel Corporation
   2 #
   3 # Licensed under the Apache License, Version 2.0 (the "License");
   4 # you may not use this file except in compliance with the License.
   5 # You may obtain a copy of the License at
   6 #
   7 #      http://www.apache.org/licenses/LICENSE-2.0
   8 #
   9 # Unless required by applicable law or agreed to in writing, software
  10 # distributed under the License is distributed on an "AS IS" BASIS,
  11 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12 # See the License for the specific language governing permissions and
  13 # limitations under the License.
  14 """ Resource collection definitions """
  15
  16 from __future__ import absolute_import
  17 from __future__ import print_function
  18
  19 import logging
  20 from itertools import chain
  21
  22 import jinja2
  23 import os
  24 import os.path
  25 import re
  26 import multiprocessing
  27 import pkg_resources
  28
  29 from oslo_config import cfg
  30 from oslo_utils.encodeutils import safe_decode
  31
  32 from yardstick import ssh
  33 from yardstick.common.task_template import finalize_for_yaml
  34 from yardstick.common.utils import validate_non_string_sequence
  35 from yardstick.network_services.nfvi.collectd import AmqpConsumer
  36 from yardstick.network_services.utils import get_nsb_option
  37
  38 LOG = logging.getLogger(__name__)
  39
  40 CONF = cfg.CONF
  41 ZMQ_OVS_PORT = 5567
  42 ZMQ_POLLING_TIME = 12000
  43 LIST_PLUGINS_ENABLED = ["amqp", "cpu", "cpufreq", "memory",
  44                         "hugepages"]
  45
  46
  47 class ResourceProfile(object):
  48     """
  49     This profile adds a resource at the beginning of the test session
  50     """
  51     COLLECTD_CONF = "collectd.conf"
  52     AMPQ_PORT = 5672
  53     DEFAULT_INTERVAL = 25
  54
  55     def __init__(self, mgmt, port_names=None, cores=None, plugins=None, interval=None):
  56         if plugins is None:
  57             self.plugins = {}
  58         else:
  59             self.plugins = plugins
  60         if interval is None:
  61             self.interval = self.DEFAULT_INTERVAL
  62         else:
  63             self.interval = interval
  64         self.enable = True
  65         self.cores = validate_non_string_sequence(cores, default=[])
  66         self._queue = multiprocessing.Queue()
  67         self.amqp_client = None
  68         self.port_names = validate_non_string_sequence(port_names, default=[])
  69
  70         # we need to save mgmt so we can connect to port 5672
  71         self.mgmt = mgmt
  72         self.connection = ssh.AutoConnectSSH.from_node(mgmt)
  73
  74     def check_if_sa_running(self, process):
  75         """ verify if system agent is running """
  76         err, pid, _ = self.connection.execute("pgrep -f %s" % process)
  77         return [err == 0, pid]
  78
  79     def run_collectd_amqp(self):
  80         """ run amqp consumer to collect the NFVi data """
  81         amqp_url = 'amqp://admin:admin@{}:{}/%2F'.format(self.mgmt['ip'], self.AMPQ_PORT)
  82         amqp = AmqpConsumer(amqp_url, self._queue)
  83         try:
  84             amqp.run()
  85         except (AttributeError, RuntimeError, KeyboardInterrupt):
  86             amqp.stop()
  87
  88     @classmethod
  89     def parse_simple_resource(cls, key, value):
  90         reskey = "/".join(rkey for rkey in key if "nsb_stats" not in rkey)
  91         return {reskey: value.split(":")[1]}
  92
  93     @classmethod
  94     def get_cpu_data(cls, res_key0, res_key1, value):
  95         """ Get cpu topology of the host """
  96         pattern = r"-(\d+)"
  97
  98         if 'cpufreq' in res_key0:
  99             metric, source = res_key0, res_key1
 100         else:
 101             metric, source = res_key1, res_key0
 102
 103         match = re.search(pattern, source, re.MULTILINE)
 104         if not match:
 105             return "error", "Invalid", "", ""
 106
 107         time, value = value.split(":")
 108         return str(match.group(1)), metric, value, time
 109
 110     @classmethod
 111     def parse_hugepages(cls, key, value):
 112         return cls.parse_simple_resource(key, value)
 113
 114     @classmethod
 115     def parse_dpdkstat(cls, key, value):
 116         return cls.parse_simple_resource(key, value)
 117
 118     @classmethod
 119     def parse_virt(cls, key, value):
 120         return cls.parse_simple_resource(key, value)
 121
 122     @classmethod
 123     def parse_ovs_stats(cls, key, value):
 124         return cls.parse_simple_resource(key, value)
 125
 126     @classmethod
 127     def parse_intel_pmu_stats(cls, key, value):
 128         return {''.join(str(v) for v in key): value.split(":")[1]}
 129
 130     def parse_collectd_result(self, metrics, core_list):
 131         """ convert collectd data into json"""
 132         result = {
 133             "cpu": {},
 134             "memory": {},
 135             "hugepages": {},
 136             "dpdkstat": {},
 137             "virt": {},
 138             "ovs_stats": {},
 139             "intel_pmu": {},
 140         }
 141         testcase = ""
 142
 143         # unicode decode
 144         decoded = ((safe_decode(k, 'utf-8'), safe_decode(v, 'utf-8')) for k, v in metrics.items())
 145         for key, value in decoded:
 146             key_split = key.split("/")
 147             res_key_iter = (key for key in key_split if "nsb_stats" not in key)
 148             res_key0 = next(res_key_iter)
 149             res_key1 = next(res_key_iter)
 150
 151             if "cpu" in res_key0 or "intel_rdt" in res_key0:
 152                 cpu_key, name, metric, testcase = \
 153                     self.get_cpu_data(res_key0, res_key1, value)
 154                 if cpu_key in core_list:
 155                     result["cpu"].setdefault(cpu_key, {}).update({name: metric})
 156
 157             elif "memory" in res_key0:
 158                 result["memory"].update({res_key1: value.split(":")[0]})
 159
 160             elif "hugepages" in res_key0:
 161                 result["hugepages"].update(self.parse_hugepages(key_split, value))
 162
 163             elif "dpdkstat" in res_key0:
 164                 result["dpdkstat"].update(self.parse_dpdkstat(key_split, value))
 165
 166             elif "virt" in res_key1:
 167                 result["virt"].update(self.parse_virt(key_split, value))
 168
 169             elif "ovs_stats" in res_key0:
 170                 result["ovs_stats"].update(self.parse_ovs_stats(key_split, value))
 171
 172             elif "intel_pmu-all" in res_key0:
 173                 result["intel_pmu"].update(self.parse_intel_pmu_stats(res_key1, value))
 174
 175         result["timestamp"] = testcase
 176
 177         return result
 178
 179     def amqp_process_for_nfvi_kpi(self):
 180         """ amqp collect and return nfvi kpis """
 181         if self.amqp_client is None and self.enable:
 182             self.amqp_client = \
 183                 multiprocessing.Process(target=self.run_collectd_amqp)
 184             self.amqp_client.start()
 185
 186     def amqp_collect_nfvi_kpi(self):
 187         """ amqp collect and return nfvi kpis """
 188         if not self.enable:
 189             return {}
 190
 191         metric = {}
 192         while not self._queue.empty():
 193             metric.update(self._queue.get())
 194         msg = self.parse_collectd_result(metric, self.cores)
 195         return msg
 196
 197     def _provide_config_file(self, config_file_path, nfvi_cfg, template_kwargs):
 198         template = pkg_resources.resource_string("yardstick.network_services.nfvi",
 199                                                  nfvi_cfg).decode('utf-8')
 200         cfg_content = jinja2.Template(template, trim_blocks=True, lstrip_blocks=True,
 201                                       finalize=finalize_for_yaml).render(
 202             **template_kwargs)
 203         # cfg_content = io.StringIO(template.format(**template_kwargs))
 204         cfg_file = os.path.join(config_file_path, nfvi_cfg)
 205         # must write as root, so use sudo
 206         self.connection.execute("cat | sudo tee {}".format(cfg_file), stdin=cfg_content)
 207
 208     def _prepare_collectd_conf(self, config_file_path):
 209         """ Prepare collectd conf """
 210
 211         kwargs = {
 212             "interval": self.interval,
 213             "loadplugins": set(chain(LIST_PLUGINS_ENABLED, self.plugins.keys())),
 214             # Optional fields PortName is descriptive only, use whatever is present
 215             "port_names": self.port_names,
 216             # "ovs_bridge_interfaces": ["br-int"],
 217             "plugins": self.plugins,
 218         }
 219         self._provide_config_file(config_file_path, self.COLLECTD_CONF, kwargs)
 220
 221     def _start_collectd(self, connection, bin_path):
 222         LOG.debug("Starting collectd to collect NFVi stats")
 223         connection.execute('sudo pkill -x -9 collectd')
 224         bin_path = get_nsb_option("bin_path")
 225         collectd_path = os.path.join(bin_path, "collectd", "sbin", "collectd")
 226         config_file_path = os.path.join(bin_path, "collectd", "etc")
 227         exit_status = connection.execute("which %s > /dev/null 2>&1" % collectd_path)[0]
 228         if exit_status != 0:
 229             LOG.warning("%s is not present disabling", collectd_path)
 230             # disable auto-provisioning because it requires Internet access
 231             # collectd_installer = os.path.join(bin_path, "collectd.sh")
 232             # provision_tool(connection, collectd)
 233             # http_proxy = os.environ.get('http_proxy', '')
 234             # https_proxy = os.environ.get('https_proxy', '')
 235             # connection.execute("sudo %s '%s' '%s'" % (
 236             #     collectd_installer, http_proxy, https_proxy))
 237             return
 238         LOG.debug("Starting collectd to collect NFVi stats")
 239         # ensure collectd.conf.d exists to avoid error/warning
 240         connection.execute("sudo mkdir -p /etc/collectd/collectd.conf.d")
 241         self._prepare_collectd_conf(config_file_path)
 242
 243         # Reset amqp queue
 244         LOG.debug("reset and setup amqp to collect data from collectd")
 245         connection.execute("sudo rm -rf /var/lib/rabbitmq/mnesia/rabbit*")
 246         connection.execute("sudo service rabbitmq-server start")
 247         connection.execute("sudo rabbitmqctl stop_app")
 248         connection.execute("sudo rabbitmqctl reset")
 249         connection.execute("sudo rabbitmqctl start_app")
 250         connection.execute("sudo service rabbitmq-server restart")
 251
 252         LOG.debug("Creating admin user for rabbitmq in order to collect data from collectd")
 253         connection.execute("sudo rabbitmqctl delete_user guest")
 254         connection.execute("sudo rabbitmqctl add_user admin admin")
 255         connection.execute("sudo rabbitmqctl authenticate_user admin admin")
 256         connection.execute("sudo rabbitmqctl set_permissions -p / admin '.*' '.*' '.*'")
 257
 258         LOG.debug("Start collectd service.....")
 259         connection.execute("sudo %s" % collectd_path)
 260         LOG.debug("Done")
 261
 262     def initiate_systemagent(self, bin_path):
 263         """ Start system agent for NFVi collection on host """
 264         if self.enable:
 265             try:
 266                 self._start_collectd(self.connection, bin_path)
 267             except Exception:
 268                 LOG.exception("Exception during collectd start")
 269                 raise
 270
 271     def start(self):
 272         """ start nfvi collection """
 273         if self.enable:
 274             LOG.debug("Start NVFi metric collection...")
 275
 276     def stop(self):
 277         """ stop nfvi collection """
 278         if not self.enable:
 279             return
 280
 281         agent = "collectd"
 282         LOG.debug("Stop resource monitor...")
 283
 284         if self.amqp_client is not None:
 285             self.amqp_client.terminate()
 286
 287         status, pid = self.check_if_sa_running(agent)
 288         if status == 0:
 289             return
 290
 291         self.connection.execute('sudo kill -9 %s' % pid)
 292         self.connection.execute('sudo pkill -9 %s' % agent)
 293         self.connection.execute('sudo service rabbitmq-server stop')
 294         self.connection.execute("sudo rabbitmqctl stop_app")