X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=yardstick%2Fnetwork_services%2Fnfvi%2Fresource.py;h=0c0bf223a91a85ff8c302c76c74ae72c1138fb76;hb=b7f867ba873806a10972108fa19d26b2ae19fa58;hp=dea754d8baef577cd8aef7448c61168bd75dbc96;hpb=4f8b511428cf31f8a74ffab9b9e662e7a05a8a76;p=yardstick.git diff --git a/yardstick/network_services/nfvi/resource.py b/yardstick/network_services/nfvi/resource.py index dea754d8b..0c0bf223a 100644 --- a/yardstick/network_services/nfvi/resource.py +++ b/yardstick/network_services/nfvi/resource.py @@ -13,24 +13,21 @@ # limitations under the License. """ Resource collection definitions """ -from __future__ import absolute_import -from __future__ import print_function - -import logging -from itertools import chain - import errno -import jinja2 +from itertools import chain +import logging +import multiprocessing import os import os.path import re -import multiprocessing -import pkg_resources +import jinja2 +import pkg_resources from oslo_config import cfg 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 @@ -55,8 +52,7 @@ class ResourceProfile(object): DEFAULT_TIMEOUT = 3600 OVS_SOCKET_PATH = "/usr/local/var/run/openvswitch/db.sock" - def __init__(self, mgmt, port_names=None, cores=None, plugins=None, - interval=None, timeout=None): + def __init__(self, mgmt, port_names=None, plugins=None, interval=None, timeout=None): if plugins is None: self.plugins = {} @@ -82,7 +78,18 @@ class ResourceProfile(object): self.mgmt = mgmt self.connection = ssh.AutoConnectSSH.from_node(mgmt) - def check_if_sa_running(self, process): + @classmethod + def make_from_node(cls, node, timeout): + # node dict works as mgmt dict + # don't need port names, there is no way we can + # tell what port is used on the compute node + collectd_options = node["collectd"] + plugins = collectd_options.get("plugins", {}) + interval = collectd_options.get("interval") + + return cls(node, plugins=plugins, interval=interval, timeout=timeout) + + def check_if_system_agent_running(self, process): """ verify if system agent is running """ try: err, pid, _ = self.connection.execute("pgrep -f %s" % process) @@ -91,7 +98,7 @@ class ResourceProfile(object): except OSError as e: if e.errno in {errno.ECONNRESET}: # if we can't connect to check, then we won't be able to connect to stop it - LOG.exception("can't connect to host to check collectd status") + LOG.exception("Can't connect to host to check %s status", process) return 1, None raise @@ -233,19 +240,6 @@ class ResourceProfile(object): } self._provide_config_file(config_file_path, self.COLLECTD_CONF, kwargs) - def _setup_intel_pmu(self, connection, bin_path): - pmu_event_path = os.path.join(bin_path, "pmu_event.json") - try: - self.plugins["intel_pmu"]["pmu_event_path"] = pmu_event_path - except KeyError: - # if intel_pmu is not a dict, force it into a dict - self.plugins["intel_pmu"] = {"pmu_event_path": pmu_event_path} - LOG.debug("Downloading event list for pmu_stats plugin") - cmd = 'cd {0}; PMU_EVENTS_PATH={1} python event_download_local.py'.format( - bin_path, pmu_event_path) - cmd = "sudo bash -c '{}'".format(cmd) - connection.execute(cmd) - def _setup_ovs_stats(self, connection): try: socket_path = self.plugins["ovs_stats"].get("ovs_socket_path", self.OVS_SOCKET_PATH) @@ -256,47 +250,46 @@ class ResourceProfile(object): if status != 0: LOG.error("cannot find OVS socket %s", socket_path) + def _start_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, stdout, stderr = connection.execute(cmd) + if exit_status != 0: + raise ResourceCommandError(command=cmd, stderr=stderr) + + # 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") + self._prepare_collectd_conf(config_file_path) + + connection.execute('sudo pkill -x -9 collectd') exit_status = connection.execute("which %s > /dev/null 2>&1" % collectd_path)[0] 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 - if "intel_pmu" in self.plugins: - self._setup_intel_pmu(connection, bin_path) 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 @@ -308,9 +301,10 @@ class ResourceProfile(object): """ 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): @@ -332,7 +326,7 @@ class ResourceProfile(object): self.amqp_client.terminate() LOG.debug("Check if %s is running", agent) - status, pid = self.check_if_sa_running(agent) + status, pid = self.check_if_system_agent_running(agent) LOG.debug("status %s pid %s", status, pid) if status != 0: return