from subprocess import Popen, PIPE\r
import os\r
\r
+import paramiko\r
+import gevent\r
+\r
logging.basicConfig(level=logging.INFO)\r
\r
"""\r
class DCNetworkMonitor():\r
def __init__(self, net):\r
self.net = net\r
- # link to Ryu REST_API\r
- self.ip = '0.0.0.0'\r
- self.port = '8080'\r
- self.REST_api = 'http://{0}:{1}'.format(self.ip,self.port)\r
+\r
+ prometheus_ip = '0.0.0.0'\r
+ prometheus_port = '9090'\r
+ self.prometheus_REST_api = 'http://{0}:{1}'.format(prometheus_ip, prometheus_port)\r
+\r
+\r
\r
# helper variables to calculate the metrics\r
self.pushgateway = 'localhost:9091'\r
# supported Prometheus metrics\r
self.registry = CollectorRegistry()\r
self.prom_tx_packet_count = Gauge('sonemu_tx_count_packets', 'Total number of packets sent',\r
- ['vnf_name', 'vnf_interface'], registry=self.registry)\r
+ ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)\r
self.prom_rx_packet_count = Gauge('sonemu_rx_count_packets', 'Total number of packets received',\r
- ['vnf_name', 'vnf_interface'], registry=self.registry)\r
+ ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)\r
self.prom_tx_byte_count = Gauge('sonemu_tx_count_bytes', 'Total number of bytes sent',\r
- ['vnf_name', 'vnf_interface'], registry=self.registry)\r
+ ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)\r
self.prom_rx_byte_count = Gauge('sonemu_rx_count_bytes', 'Total number of bytes received',\r
- ['vnf_name', 'vnf_interface'], registry=self.registry)\r
+ ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)\r
\r
self.prom_metrics={'tx_packets':self.prom_tx_packet_count, 'rx_packets':self.prom_rx_packet_count,\r
'tx_bytes':self.prom_tx_byte_count,'rx_bytes':self.prom_rx_byte_count}\r
}\r
'''\r
self.monitor_lock = threading.Lock()\r
+ self.monitor_flow_lock = threading.Lock()\r
self.network_metrics = []\r
+ self.flow_metrics = []\r
\r
# start monitoring thread\r
self.start_monitoring = True\r
self.monitor_thread = threading.Thread(target=self.get_network_metrics)\r
self.monitor_thread.start()\r
\r
+ self.monitor_flow_thread = threading.Thread(target=self.get_flow_metrics)\r
+ self.monitor_flow_thread.start()\r
+\r
# helper tools\r
self.pushgateway_process = self.start_PushGateway()\r
self.prometheus_process = self.start_Prometheus()\r
self.cadvisor_process = self.start_cadvisor()\r
\r
+ # first set some parameters, before measurement can start\r
+ def setup_flow(self, vnf_name, vnf_interface=None, metric='tx_packets', cookie=0):\r
+\r
+ flow_metric = {}\r
+\r
+ # check if port is specified (vnf:port)\r
+ if vnf_interface is None:\r
+ # take first interface by default\r
+ connected_sw = self.net.DCNetwork_graph.neighbors(vnf_name)[0]\r
+ link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]\r
+ vnf_interface = link_dict[0]['src_port_id']\r
+\r
+ flow_metric['vnf_name'] = vnf_name\r
+ flow_metric['vnf_interface'] = vnf_interface\r
+\r
+ vnf_switch = None\r
+ for connected_sw in self.net.DCNetwork_graph.neighbors(vnf_name):\r
+ link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]\r
+ for link in link_dict:\r
+ # logging.info("{0},{1}".format(link_dict[link],vnf_interface))\r
+ if link_dict[link]['src_port_id'] == vnf_interface:\r
+ # found the right link and connected switch\r
+ # logging.info("{0},{1}".format(link_dict[link]['src_port_id'], vnf_source_interface))\r
+ vnf_switch = connected_sw\r
+ flow_metric['mon_port'] = link_dict[link]['dst_port_nr']\r
+ break\r
+\r
+ if not vnf_switch:\r
+ logging.exception("vnf switch of {0}:{1} not found!".format(vnf_name, vnf_interface))\r
+ return "vnf switch of {0}:{1} not found!".format(vnf_name, vnf_interface)\r
+\r
+ try:\r
+ # default port direction to monitor\r
+ if metric is None:\r
+ metric = 'tx_packets'\r
+\r
+ next_node = self.net.getNodeByName(vnf_switch)\r
+\r
+ if not isinstance(next_node, OVSSwitch):\r
+ logging.info("vnf: {0} is not connected to switch".format(vnf_name))\r
+ return\r
+\r
+ flow_metric['previous_measurement'] = 0\r
+ flow_metric['previous_monitor_time'] = 0\r
+\r
+ flow_metric['switch_dpid'] = int(str(next_node.dpid), 16)\r
+ flow_metric['metric_key'] = metric\r
+ flow_metric['cookie'] = cookie\r
+\r
+ self.monitor_flow_lock.acquire()\r
+ self.flow_metrics.append(flow_metric)\r
+ self.monitor_flow_lock.release()\r
+\r
+ logging.info('Started monitoring flow:{3} {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie))\r
+ return 'Started monitoring flow:{3} {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie)\r
+\r
+ except Exception as ex:\r
+ logging.exception("setup_metric error.")\r
+ return ex.message\r
+\r
+ def stop_flow(self, vnf_name, vnf_interface=None, metric=None, cookie=0):\r
+ for flow_dict in self.flow_metrics:\r
+ if flow_dict['vnf_name'] == vnf_name and flow_dict['vnf_interface'] == vnf_interface \\r
+ and flow_dict['metric_key'] == metric and flow_dict['cookie'] == cookie:\r
+\r
+ self.monitor_flow_lock.acquire()\r
+\r
+ self.flow_metrics.remove(flow_dict)\r
+\r
+ for collector in self.registry._collectors:\r
+ if (vnf_name, vnf_interface, cookie) in collector._metrics:\r
+ #logging.info('2 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames,\r
+ # collector._metrics))\r
+ collector.remove(vnf_name, vnf_interface, cookie)\r
+\r
+ delete_from_gateway(self.pushgateway, job='sonemu-SDNcontroller')\r
+\r
+ self.monitor_flow_lock.release()\r
+\r
+ logging.info('Stopped monitoring flow {3}: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie))\r
+ return 'Stopped monitoring flow {3}: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie)\r
+\r
+\r
# first set some parameters, before measurement can start\r
def setup_metric(self, vnf_name, vnf_interface=None, metric='tx_packets'):\r
\r
if link_dict[link]['src_port_id'] == vnf_interface:\r
# found the right link and connected switch\r
# logging.info("{0},{1}".format(link_dict[link]['src_port_id'], vnf_source_interface))\r
- network_metric['mon_port'] = link_dict[link]['dst_port']\r
+ network_metric['mon_port'] = link_dict[link]['dst_port_nr']\r
break\r
\r
if 'mon_port' not in network_metric:\r
logging.exception("setup_metric error.")\r
return ex.message\r
\r
- def stop_metric(self, vnf_name, vnf_interface, metric):\r
+ def stop_metric(self, vnf_name, vnf_interface=None, metric=None):\r
\r
for metric_dict in self.network_metrics:\r
+ #logging.info('start Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric_dict))\r
if metric_dict['vnf_name'] == vnf_name and metric_dict['vnf_interface'] == vnf_interface \\r
and metric_dict['metric_key'] == metric:\r
\r
#self.registry.unregister(self.prom_metrics[metric_dict['metric_key']])\r
\r
for collector in self.registry._collectors :\r
- logging.info('name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames, collector._metrics))\r
+ #logging.info('name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames, collector._metrics))\r
"""\r
INFO:root:name:sonemu_rx_count_packets\r
labels:('vnf_name', 'vnf_interface')\r
"""\r
logging.info('{0}'.format(collector._metrics.values()))\r
#if self.prom_metrics[metric_dict['metric_key']]\r
- if (vnf_name, vnf_interface) in collector._metrics:\r
+ if (vnf_name, vnf_interface, 'None') in collector._metrics:\r
logging.info('2 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames,\r
collector._metrics))\r
#collector._metrics = {}\r
- collector.remove(vnf_name, vnf_interface)\r
+ collector.remove(vnf_name, vnf_interface, 'None')\r
\r
# set values to NaN, prometheus api currently does not support removal of metrics\r
#self.prom_metrics[metric_dict['metric_key']].labels(vnf_name, vnf_interface).set(float('nan'))\r
logging.info('Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric))\r
return 'Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric)\r
\r
+ # delete everything from this vnf\r
+ elif metric_dict['vnf_name'] == vnf_name and vnf_interface is None and metric is None:\r
+ self.monitor_lock.acquire()\r
+ self.network_metrics.remove(metric_dict)\r
+ for collector in self.registry._collectors:\r
+ collector_dict = collector._metrics.copy()\r
+ for name, interface, id in collector_dict:\r
+ if name == vnf_name:\r
+ logging.info('3 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames,\r
+ collector._metrics))\r
+ collector.remove(name, interface, 'None')\r
+\r
+ delete_from_gateway(self.pushgateway, job='sonemu-SDNcontroller')\r
+ self.monitor_lock.release()\r
+ logging.info('Stopped monitoring vnf: {0}'.format(vnf_name))\r
+ return 'Stopped monitoring: {0}'.format(vnf_name)\r
+\r
\r
# get all metrics defined in the list and export it to Prometheus\r
+ def get_flow_metrics(self):\r
+ while self.start_monitoring:\r
+\r
+ self.monitor_flow_lock.acquire()\r
+\r
+ for flow_dict in self.flow_metrics:\r
+ data = {}\r
+\r
+ data['cookie'] = flow_dict['cookie']\r
+\r
+ if 'tx' in flow_dict['metric_key']:\r
+ data['match'] = {'in_port':flow_dict['mon_port']}\r
+ elif 'rx' in flow_dict['metric_key']:\r
+ data['out_port'] = flow_dict['mon_port']\r
+\r
+\r
+ # query Ryu\r
+ ret = self.net.ryu_REST('stats/flow', dpid=flow_dict['switch_dpid'], data=data)\r
+ flow_stat_dict = ast.literal_eval(ret)\r
+\r
+ #logging.info('received flow stat:{0} '.format(flow_stat_dict))\r
+ self.set_flow_metric(flow_dict, flow_stat_dict)\r
+\r
+ self.monitor_flow_lock.release()\r
+ time.sleep(1)\r
+\r
def get_network_metrics(self):\r
while self.start_monitoring:\r
\r
for dpid in dpid_set:\r
\r
# query Ryu\r
- ret = self.REST_cmd('stats/port', dpid)\r
+ ret = self.net.ryu_REST('stats/port', dpid=dpid)\r
port_stat_dict = ast.literal_eval(ret)\r
\r
metric_list = [metric_dict for metric_dict in self.network_metrics\r
\r
# set prometheus metric\r
self.prom_metrics[metric_dict['metric_key']].\\r
- labels({'vnf_name':vnf_name, 'vnf_interface':vnf_interface}).\\r
+ labels({'vnf_name': vnf_name, 'vnf_interface': vnf_interface, 'flow_id': None}).\\r
set(this_measurement)\r
#push_to_gateway(self.pushgateway, job='SDNcontroller',\r
# grouping_key={'metric':metric_dict['metric_key']}, registry=self.registry)\r
#logging.info('first measurement')\r
time.sleep(1)\r
self.monitor_lock.release()\r
+\r
metric_rate = self.get_network_metrics()\r
return metric_rate\r
+\r
else:\r
time_delta = (port_uptime - metric_dict['previous_monitor_time'])\r
metric_rate = (this_measurement - metric_dict['previous_measurement']) / float(time_delta)\r
- # logging.info('uptime:{2} delta:{0} rate:{1}'.format(time_delta,byte_rate,port_uptime))\r
+ #logging.info('metric: {0} rate:{1}'.format(metric_dict['metric_key'], metric_rate))\r
\r
metric_dict['previous_measurement'] = this_measurement\r
metric_dict['previous_monitor_time'] = port_uptime\r
logging.exception('metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface))\r
return 'metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface)\r
\r
+ def set_flow_metric(self, metric_dict, flow_stat_dict):\r
+ # vnf tx is the datacenter switch rx and vice-versa\r
+ #metric_key = self.switch_tx_rx(metric_dict['metric_key'])\r
+ metric_key = metric_dict['metric_key']\r
+ switch_dpid = metric_dict['switch_dpid']\r
+ vnf_name = metric_dict['vnf_name']\r
+ vnf_interface = metric_dict['vnf_interface']\r
+ previous_measurement = metric_dict['previous_measurement']\r
+ previous_monitor_time = metric_dict['previous_monitor_time']\r
+ cookie = metric_dict['cookie']\r
+\r
+ # TODO aggregate all found flow stats\r
+ flow_stat = flow_stat_dict[str(switch_dpid)][0]\r
+ if 'bytes' in metric_key:\r
+ counter = flow_stat['byte_count']\r
+ elif 'packet' in metric_key:\r
+ counter = flow_stat['packet_count']\r
+\r
+ flow_uptime = flow_stat['duration_sec'] + flow_stat['duration_nsec'] * 10 ** (-9)\r
\r
- def REST_cmd(self, prefix, dpid):\r
- url = self.REST_api + '/' + str(prefix) + '/' + str(dpid)\r
+ self.prom_metrics[metric_dict['metric_key']]. \\r
+ labels({'vnf_name': vnf_name, 'vnf_interface': vnf_interface, 'flow_id': cookie}). \\r
+ set(counter)\r
+ pushadd_to_gateway(self.pushgateway, job='sonemu-SDNcontroller', registry=self.registry)\r
+\r
+ #logging.exception('metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface))\r
+ #return 'metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface)\r
+\r
+ def query_Prometheus(self, query):\r
+ '''\r
+ escaped_chars='{}[]'\r
+ for old in escaped_chars:\r
+ new = '\{0}'.format(old)\r
+ query = query.replace(old, new)\r
+ '''\r
+ url = self.prometheus_REST_api + '/' + 'api/v1/query?query=' + query\r
+ #logging.info('query:{0}'.format(url))\r
req = urllib2.Request(url)\r
ret = urllib2.urlopen(req).read()\r
+ ret = ast.literal_eval(ret)\r
+ if ret['status'] == 'success':\r
+ #logging.info('return:{0}'.format(ret))\r
+ try:\r
+ ret = ret['data']['result'][0]['value']\r
+ except:\r
+ ret = None\r
+ else:\r
+ ret = None\r
return ret\r
\r
def start_Prometheus(self, port=9090):\r
"--rm",\r
"-p", "{0}:9090".format(port),\r
"-v", "{0}/prometheus.yml:/etc/prometheus/prometheus.yml".format(os.path.dirname(os.path.abspath(__file__))),\r
+ "-v", "{0}/profile.rules:/etc/prometheus/profile.rules".format(os.path.dirname(os.path.abspath(__file__))),\r
"--name", "prometheus",\r
"prom/prometheus"\r
]\r
# stop the monitoring thread\r
self.start_monitoring = False\r
self.monitor_thread.join()\r
+ self.monitor_flow_thread.join()\r
\r
if self.prometheus_process is not None:\r
logging.info('stopping prometheus container')\r
name]\r
Popen(cmd).wait()\r
\r
+ def profile(self, mgmt_ip, rate, input_ip, vnf_uuid ):\r
+\r
+ ssh = paramiko.SSHClient()\r
+ ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())\r
+ #ssh.connect(mgmt_ip, username='steven', password='test')\r
+ ssh.connect(mgmt_ip, username='root', password='root')\r
+\r
+ iperf_cmd = 'iperf -c {0} -u -l18 -b{1}M -t1000 &'.format(input_ip, rate)\r
+ if rate > 0:\r
+ stdin, stdout, stderr = ssh.exec_command(iperf_cmd)\r
+\r
+ start_time = time.time()\r
+ query_cpu = '(sum(rate(container_cpu_usage_seconds_total{{id="/docker/{0}"}}[{1}s])))'.format(vnf_uuid, 1)\r
+ while (time.time() - start_time) < 15:\r
+ data = self.query_Prometheus(query_cpu)\r
+ # logging.info('rate: {1} data:{0}'.format(data, rate))\r
+ gevent.sleep(0)\r
+ time.sleep(1)\r
+\r
+ query_cpu2 = '(sum(rate(container_cpu_usage_seconds_total{{id="/docker/{0}"}}[{1}s])))'.format(vnf_uuid, 8)\r
+ cpu_load = float(self.query_Prometheus(query_cpu2)[1])\r
+ output = 'rate: {1}Mbps; cpu_load: {0}%'.format(round(cpu_load * 100, 2), rate)\r
+ output_line = output\r
+ logging.info(output_line)\r
+\r
+ stop_iperf = 'pkill -9 iperf'\r
+ stdin, stdout, stderr = ssh.exec_command(stop_iperf)\r
+\r
+ return output_line\r
\r