1 # Copyright (c) 2016-2017 Intel Corporation
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
7 # http://www.apache.org/licenses/LICENSE-2.0
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 """
16 from __future__ import absolute_import
17 from __future__ import print_function
20 from itertools import chain
27 import multiprocessing
30 from oslo_config import cfg
31 from oslo_utils.encodeutils import safe_decode
33 from yardstick import ssh
34 from yardstick.common.task_template import finalize_for_yaml
35 from yardstick.common.utils import validate_non_string_sequence
36 from yardstick.network_services.nfvi.collectd import AmqpConsumer
39 LOG = logging.getLogger(__name__)
43 ZMQ_POLLING_TIME = 12000
44 LIST_PLUGINS_ENABLED = ["amqp", "cpu", "cpufreq", "memory",
48 class ResourceProfile(object):
50 This profile adds a resource at the beginning of the test session
52 COLLECTD_CONF = "collectd.conf"
55 DEFAULT_TIMEOUT = 3600
56 OVS_SOCKET_PATH = "/usr/local/var/run/openvswitch/db.sock"
58 def __init__(self, mgmt, port_names=None, plugins=None, interval=None, timeout=None):
63 self.plugins = plugins
66 self.interval = self.DEFAULT_INTERVAL
68 self.interval = interval
71 self.timeout = self.DEFAULT_TIMEOUT
73 self.timeout = timeout
76 self._queue = multiprocessing.Queue()
77 self.amqp_client = None
78 self.port_names = validate_non_string_sequence(port_names, default=[])
80 # we need to save mgmt so we can connect to port 5672
82 self.connection = ssh.AutoConnectSSH.from_node(mgmt)
85 def make_from_node(cls, node, timeout):
86 # node dict works as mgmt dict
87 # don't need port names, there is no way we can
88 # tell what port is used on the compute node
89 collectd_options = node["collectd"]
90 plugins = collectd_options.get("plugins", {})
91 interval = collectd_options.get("interval")
93 return cls(node, plugins=plugins, interval=interval, timeout=timeout)
95 def check_if_sa_running(self, process):
96 """ verify if system agent is running """
98 err, pid, _ = self.connection.execute("pgrep -f %s" % process)
100 return err, pid.strip()
102 if e.errno in {errno.ECONNRESET}:
103 # if we can't connect to check, then we won't be able to connect to stop it
104 LOG.exception("can't connect to host to check collectd status")
108 def run_collectd_amqp(self):
109 """ run amqp consumer to collect the NFVi data """
110 amqp_url = 'amqp://admin:admin@{}:{}/%2F'.format(self.mgmt['ip'], self.AMPQ_PORT)
111 amqp = AmqpConsumer(amqp_url, self._queue)
114 except (AttributeError, RuntimeError, KeyboardInterrupt):
118 def parse_simple_resource(cls, key, value):
119 reskey = "/".join(rkey for rkey in key if "nsb_stats" not in rkey)
120 return {reskey: value.split(":")[1]}
123 def get_cpu_data(cls, res_key0, res_key1, value):
124 """ Get cpu topology of the host """
127 if 'cpufreq' in res_key0:
128 metric, source = res_key0, res_key1
130 metric, source = res_key1, res_key0
132 match = re.search(pattern, source, re.MULTILINE)
134 return "error", "Invalid", "", ""
136 time, value = value.split(":")
137 return str(match.group(1)), metric, value, time
140 def parse_hugepages(cls, key, value):
141 return cls.parse_simple_resource(key, value)
144 def parse_dpdkstat(cls, key, value):
145 return cls.parse_simple_resource(key, value)
148 def parse_virt(cls, key, value):
149 return cls.parse_simple_resource(key, value)
152 def parse_ovs_stats(cls, key, value):
153 return cls.parse_simple_resource(key, value)
156 def parse_intel_pmu_stats(cls, key, value):
157 return {''.join(str(v) for v in key): value.split(":")[1]}
159 def parse_collectd_result(self, metrics):
160 """ convert collectd data into json"""
172 decoded = ((safe_decode(k, 'utf-8'), safe_decode(v, 'utf-8')) for k, v in metrics.items())
173 for key, value in decoded:
174 key_split = key.split("/")
175 res_key_iter = (key for key in key_split if "nsb_stats" not in key)
176 res_key0 = next(res_key_iter)
177 res_key1 = next(res_key_iter)
179 if "cpu" in res_key0 or "intel_rdt" in res_key0 or "intel_pmu" in res_key0:
180 cpu_key, name, metric, testcase = \
181 self.get_cpu_data(res_key0, res_key1, value)
182 result["cpu"].setdefault(cpu_key, {}).update({name: metric})
184 elif "memory" in res_key0:
185 result["memory"].update({res_key1: value.split(":")[0]})
187 elif "hugepages" in res_key0:
188 result["hugepages"].update(self.parse_hugepages(key_split, value))
190 elif "dpdkstat" in res_key0:
191 result["dpdkstat"].update(self.parse_dpdkstat(key_split, value))
193 elif "virt" in res_key1:
194 result["virt"].update(self.parse_virt(key_split, value))
196 elif "ovs_stats" in res_key0:
197 result["ovs_stats"].update(self.parse_ovs_stats(key_split, value))
199 result["timestamp"] = testcase
203 def amqp_process_for_nfvi_kpi(self):
204 """ amqp collect and return nfvi kpis """
205 if self.amqp_client is None and self.enable:
206 self.amqp_client = multiprocessing.Process(
207 name="AmqpClient-{}-{}".format(self.mgmt['ip'], os.getpid()),
208 target=self.run_collectd_amqp)
209 self.amqp_client.start()
211 def amqp_collect_nfvi_kpi(self):
212 """ amqp collect and return nfvi kpis """
217 while not self._queue.empty():
218 metric.update(self._queue.get())
219 msg = self.parse_collectd_result(metric)
222 def _provide_config_file(self, config_file_path, nfvi_cfg, template_kwargs):
223 template = pkg_resources.resource_string("yardstick.network_services.nfvi",
224 nfvi_cfg).decode('utf-8')
225 cfg_content = jinja2.Template(template, trim_blocks=True, lstrip_blocks=True,
226 finalize=finalize_for_yaml).render(
228 # cfg_content = io.StringIO(template.format(**template_kwargs))
229 cfg_file = os.path.join(config_file_path, nfvi_cfg)
230 # must write as root, so use sudo
231 self.connection.execute("cat | sudo tee {}".format(cfg_file), stdin=cfg_content)
233 def _prepare_collectd_conf(self, config_file_path):
234 """ Prepare collectd conf """
237 "interval": self.interval,
238 "loadplugins": set(chain(LIST_PLUGINS_ENABLED, self.plugins.keys())),
239 # Optional fields PortName is descriptive only, use whatever is present
240 "port_names": self.port_names,
241 # "ovs_bridge_interfaces": ["br-int"],
242 "plugins": self.plugins,
244 self._provide_config_file(config_file_path, self.COLLECTD_CONF, kwargs)
246 def _setup_ovs_stats(self, connection):
248 socket_path = self.plugins["ovs_stats"].get("ovs_socket_path", self.OVS_SOCKET_PATH)
250 # ovs_stats is not a dict
251 socket_path = self.OVS_SOCKET_PATH
252 status = connection.execute("test -S {}".format(socket_path))[0]
254 LOG.error("cannot find OVS socket %s", socket_path)
256 def _start_collectd(self, connection, bin_path):
257 LOG.debug("Starting collectd to collect NFVi stats")
258 connection.execute('sudo pkill -x -9 collectd')
259 collectd_path = os.path.join(bin_path, "collectd", "sbin", "collectd")
260 config_file_path = os.path.join(bin_path, "collectd", "etc")
261 exit_status = connection.execute("which %s > /dev/null 2>&1" % collectd_path)[0]
263 LOG.warning("%s is not present disabling", collectd_path)
264 # disable auto-provisioning because it requires Internet access
265 # collectd_installer = os.path.join(bin_path, "collectd.sh")
266 # provision_tool(connection, collectd)
267 # http_proxy = os.environ.get('http_proxy', '')
268 # https_proxy = os.environ.get('https_proxy', '')
269 # connection.execute("sudo %s '%s' '%s'" % (
270 # collectd_installer, http_proxy, https_proxy))
272 if "ovs_stats" in self.plugins:
273 self._setup_ovs_stats(connection)
275 LOG.debug("Starting collectd to collect NFVi stats")
276 # ensure collectd.conf.d exists to avoid error/warning
277 connection.execute("sudo mkdir -p /etc/collectd/collectd.conf.d")
278 self._prepare_collectd_conf(config_file_path)
281 LOG.debug("reset and setup amqp to collect data from collectd")
282 connection.execute("sudo rm -rf /var/lib/rabbitmq/mnesia/rabbit*")
283 connection.execute("sudo service rabbitmq-server start")
284 connection.execute("sudo rabbitmqctl stop_app")
285 connection.execute("sudo rabbitmqctl reset")
286 connection.execute("sudo rabbitmqctl start_app")
287 connection.execute("sudo service rabbitmq-server restart")
289 LOG.debug("Creating admin user for rabbitmq in order to collect data from collectd")
290 connection.execute("sudo rabbitmqctl delete_user guest")
291 connection.execute("sudo rabbitmqctl add_user admin admin")
292 connection.execute("sudo rabbitmqctl authenticate_user admin admin")
293 connection.execute("sudo rabbitmqctl set_permissions -p / admin '.*' '.*' '.*'")
295 LOG.debug("Start collectd service..... %s second timeout", self.timeout)
296 # intel_pmu plug requires large numbers of files open, so try to set
297 # ulimit -n to a large value
298 connection.execute("sudo bash -c 'ulimit -n 1000000 ; %s'" % collectd_path,
299 timeout=self.timeout)
302 def initiate_systemagent(self, bin_path):
303 """ Start system agent for NFVi collection on host """
306 self._start_collectd(self.connection, bin_path)
308 LOG.exception("Exception during collectd start")
312 """ start nfvi collection """
314 LOG.debug("Start NVFi metric collection...")
317 """ stop nfvi collection """
322 LOG.debug("Stop resource monitor...")
324 if self.amqp_client is not None:
325 # we proper and try to join first
326 self.amqp_client.join(3)
327 self.amqp_client.terminate()
329 LOG.debug("Check if %s is running", agent)
330 status, pid = self.check_if_sa_running(agent)
331 LOG.debug("status %s pid %s", status, pid)
336 self.connection.execute('sudo kill -9 "%s"' % pid)
337 self.connection.execute('sudo pkill -9 "%s"' % agent)
338 self.connection.execute('sudo service rabbitmq-server stop')
339 self.connection.execute("sudo rabbitmqctl stop_app")