from oslo_utils.encodeutils import safe_decode
from yardstick import ssh
+from yardstick.common.exceptions import ResourceCommandError
from yardstick.common.task_template import finalize_for_yaml
from yardstick.common.utils import validate_non_string_sequence
from yardstick.network_services.nfvi.collectd import AmqpConsumer
+from yardstick.benchmark.contexts import heat
LOG = logging.getLogger(__name__)
DEFAULT_TIMEOUT = 3600
OVS_SOCKET_PATH = "/usr/local/var/run/openvswitch/db.sock"
- def __init__(self, mgmt, port_names=None, plugins=None, interval=None, timeout=None):
+ def __init__(self, mgmt, port_names=None, plugins=None,
+ interval=None, timeout=None, reset_mq_flag=True):
if plugins is None:
self.plugins = {}
# we need to save mgmt so we can connect to port 5672
self.mgmt = mgmt
self.connection = ssh.AutoConnectSSH.from_node(mgmt)
+ self._reset_mq_flag = reset_mq_flag
@classmethod
def make_from_node(cls, node, timeout):
plugins = collectd_options.get("plugins", {})
interval = collectd_options.get("interval")
- return cls(node, plugins=plugins, interval=interval, timeout=timeout)
+ reset_mq_flag = (False if node.get("ctx_type") == heat.HeatContext.__context_type__
+ else True)
+ return cls(node, plugins=plugins, interval=interval,
+ timeout=timeout, reset_mq_flag=reset_mq_flag)
def check_if_system_agent_running(self, process):
""" verify if system agent is running """
if not self.enable:
return {}
+ if self.check_if_system_agent_running("collectd")[0] != 0:
+ return {}
+
metric = {}
while not self._queue.empty():
metric.update(self._queue.get())
- msg = self.parse_collectd_result(metric)
- return msg
+
+ return self.parse_collectd_result(metric)
def _provide_config_file(self, config_file_path, nfvi_cfg, template_kwargs):
template = pkg_resources.resource_string("yardstick.network_services.nfvi",
if status != 0:
LOG.error("cannot find OVS socket %s", socket_path)
+ def _reset_rabbitmq(self, connection):
+ # Reset amqp queue
+ LOG.debug("reset and setup amqp to collect data from collectd")
+ # ensure collectd.conf.d exists to avoid error/warning
+ cmd_list = ["sudo mkdir -p /etc/collectd/collectd.conf.d",
+ "sudo service rabbitmq-server restart",
+ "sudo rabbitmqctl stop_app",
+ "sudo rabbitmqctl reset",
+ "sudo rabbitmqctl start_app",
+ "sudo rabbitmqctl add_user admin admin",
+ "sudo rabbitmqctl authenticate_user admin admin",
+ "sudo rabbitmqctl set_permissions -p / admin '.*' '.*' '.*'"
+ ]
+
+ for cmd in cmd_list:
+ exit_status, _, stderr = connection.execute(cmd)
+ if exit_status != 0:
+ raise ResourceCommandError(command=cmd, stderr=stderr)
+
+ def _check_rabbitmq_user(self, connection, user='admin'):
+ exit_status, stdout, _ = connection.execute("sudo rabbitmqctl list_users")
+ if exit_status == 0:
+ for line in stdout.split('\n')[1:]:
+ if line.split('\t')[0] == user:
+ return True
+
+ def _set_rabbitmq_admin_user(self, connection):
+ LOG.debug("add admin user to amqp")
+ cmd_list = ["sudo rabbitmqctl add_user admin admin",
+ "sudo rabbitmqctl authenticate_user admin admin",
+ "sudo rabbitmqctl set_permissions -p / admin '.*' '.*' '.*'"
+ ]
+
+ for cmd in cmd_list:
+ exit_status, stdout, stderr = connection.execute(cmd)
+ if exit_status != 0:
+ raise ResourceCommandError(command=cmd, stdout=stdout, stderr=stderr)
+
+ def _start_rabbitmq(self, connection):
+ if self._reset_mq_flag:
+ self._reset_rabbitmq(connection)
+ else:
+ if not self._check_rabbitmq_user(connection):
+ self._set_rabbitmq_admin_user(connection)
+
+ # check stdout for "sudo rabbitmqctl status" command
+ cmd = "sudo rabbitmqctl status"
+ _, stdout, stderr = connection.execute(cmd)
+ if not re.search("RabbitMQ", stdout):
+ LOG.error("rabbitmqctl status don't have RabbitMQ in running apps")
+ raise ResourceCommandError(command=cmd, stderr=stderr)
+
def _start_collectd(self, connection, bin_path):
LOG.debug("Starting collectd to collect NFVi stats")
- connection.execute('sudo pkill -x -9 collectd')
collectd_path = os.path.join(bin_path, "collectd", "sbin", "collectd")
config_file_path = os.path.join(bin_path, "collectd", "etc")
- exit_status = connection.execute("which %s > /dev/null 2>&1" % collectd_path)[0]
+ self._prepare_collectd_conf(config_file_path)
+
+ connection.execute('sudo pkill -x -9 collectd')
+ cmd = "which %s > /dev/null 2>&1" % collectd_path
+ exit_status, _, stderr = connection.execute(cmd)
if exit_status != 0:
- LOG.warning("%s is not present disabling", collectd_path)
- # disable auto-provisioning because it requires Internet access
- # collectd_installer = os.path.join(bin_path, "collectd.sh")
- # provision_tool(connection, collectd)
- # http_proxy = os.environ.get('http_proxy', '')
- # https_proxy = os.environ.get('https_proxy', '')
- # connection.execute("sudo %s '%s' '%s'" % (
- # collectd_installer, http_proxy, https_proxy))
- return
+ raise ResourceCommandError(command=cmd, stderr=stderr)
+
if "ovs_stats" in self.plugins:
self._setup_ovs_stats(connection)
LOG.debug("Starting collectd to collect NFVi stats")
- # ensure collectd.conf.d exists to avoid error/warning
- connection.execute("sudo mkdir -p /etc/collectd/collectd.conf.d")
- self._prepare_collectd_conf(config_file_path)
-
- # Reset amqp queue
- LOG.debug("reset and setup amqp to collect data from collectd")
- connection.execute("sudo rm -rf /var/lib/rabbitmq/mnesia/rabbit*")
- connection.execute("sudo service rabbitmq-server start")
- connection.execute("sudo rabbitmqctl stop_app")
- connection.execute("sudo rabbitmqctl reset")
- connection.execute("sudo rabbitmqctl start_app")
- connection.execute("sudo service rabbitmq-server restart")
-
- LOG.debug("Creating admin user for rabbitmq in order to collect data from collectd")
- connection.execute("sudo rabbitmqctl delete_user guest")
- connection.execute("sudo rabbitmqctl add_user admin admin")
- connection.execute("sudo rabbitmqctl authenticate_user admin admin")
- connection.execute("sudo rabbitmqctl set_permissions -p / admin '.*' '.*' '.*'")
-
LOG.debug("Start collectd service..... %s second timeout", self.timeout)
# intel_pmu plug requires large numbers of files open, so try to set
# ulimit -n to a large value
- connection.execute("sudo bash -c 'ulimit -n 1000000 ; %s'" % collectd_path,
- timeout=self.timeout)
+
+ cmd = "sudo bash -c 'ulimit -n 1000000 ; %s'" % collectd_path
+ exit_status, _, stderr = connection.execute(cmd, timeout=self.timeout)
+ if exit_status != 0:
+ raise ResourceCommandError(command=cmd, stderr=stderr)
+
LOG.debug("Done")
def initiate_systemagent(self, bin_path):
""" Start system agent for NFVi collection on host """
if self.enable:
try:
+ self._start_rabbitmq(self.connection)
self._start_collectd(self.connection, bin_path)
- except Exception:
- LOG.exception("Exception during collectd start")
+ except ResourceCommandError as e:
+ LOG.exception("Exception during collectd and rabbitmq start: %s", str(e))
raise
def start(self):
if pid:
self.connection.execute('sudo kill -9 "%s"' % pid)
self.connection.execute('sudo pkill -9 "%s"' % agent)
- self.connection.execute('sudo service rabbitmq-server stop')
- self.connection.execute("sudo rabbitmqctl stop_app")
+
+ if self._reset_mq_flag:
+ self.connection.execute('sudo service rabbitmq-server stop')
+ self.connection.execute("sudo rabbitmqctl stop_app")
Code Review / yardstick.git / blobdiff