src/emuvim/dcemulator/monitoring.py

"""
Copyright (c) 2015 SONATA-NFV
ALL RIGHTS RESERVED.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Neither the name of the SONATA-NFV [, ANY ADDITIONAL AFFILIATION]
nor the names of its contributors may be used to endorse or promote
products derived from this software without specific prior written
permission.

This work has been performed in the framework of the SONATA project,
funded by the European Commission under Grant number 671517 through
the Horizon 2020 and 5G-PPP programmes. The authors would like to
acknowledge the contributions of their colleagues of the SONATA
partner consortium (www.sonata-nfv.eu).
"""

import logging
from mininet.node import  OVSSwitch
import ast
import time
from prometheus_client import start_http_server, Summary, Histogram, Gauge, Counter, REGISTRY, CollectorRegistry, \
    pushadd_to_gateway, push_to_gateway, delete_from_gateway
import threading
from subprocess import Popen
import os


logging.basicConfig(level=logging.INFO)

"""
class to read openflow stats from the Ryu controller of the DCNetwork
"""

class DCNetworkMonitor():
    def __init__(self, net):
        self.net = net

        # TODO: these global variables should be part of a config file?
        '''
        # prometheus is started outside of son-emu
        prometheus_ip = '127.0.0.1'
        prometheus_port = '9090'
        self.prometheus_REST_api = 'http://{0}:{1}'.format(prometheus_ip, prometheus_port)
        '''
        # helper variables to calculate the metrics
        # pushgateway is started outside of son-emu and son-emu is started with net=host
        # so localhost:9091 works
        self.pushgateway = 'localhost:9091'
        # when sdk is started with docker-compose, we could use
        # self.pushgateway = 'pushgateway:9091'
        # Start up the server to expose the metrics to Prometheus
        #start_http_server(8000)

        # supported Prometheus metrics
        self.registry = CollectorRegistry()
        self.prom_tx_packet_count = Gauge('sonemu_tx_count_packets', 'Total number of packets sent',
                                          ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)
        self.prom_rx_packet_count = Gauge('sonemu_rx_count_packets', 'Total number of packets received',
                                          ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)
        self.prom_tx_byte_count = Gauge('sonemu_tx_count_bytes', 'Total number of bytes sent',
                                        ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)
        self.prom_rx_byte_count = Gauge('sonemu_rx_count_bytes', 'Total number of bytes received',
                                        ['vnf_name', 'vnf_interface', 'flow_id'], registry=self.registry)

        self.prom_metrics={'tx_packets':self.prom_tx_packet_count, 'rx_packets':self.prom_rx_packet_count,
                           'tx_bytes':self.prom_tx_byte_count,'rx_bytes':self.prom_rx_byte_count}

        # list of installed metrics to monitor
        # each entry can contain this data
        '''
        {
        switch_dpid = 0
        vnf_name = None
        vnf_interface = None
        previous_measurement = 0
        previous_monitor_time = 0
        metric_key = None
        mon_port = None
        }
        '''
        self.monitor_lock = threading.Lock()
        self.monitor_flow_lock = threading.Lock()
        self.network_metrics = []
        self.flow_metrics = []

        # start monitoring thread
        self.start_monitoring = True
        self.monitor_thread = threading.Thread(target=self.get_network_metrics)
        self.monitor_thread.start()

        self.monitor_flow_thread = threading.Thread(target=self.get_flow_metrics)
        self.monitor_flow_thread.start()

        # helper tools
        # Prometheus pushgateway and DB are started as external contianer, outside of son-emu
        #self.pushgateway_process = self.start_PushGateway()
        #self.prometheus_process = self.start_Prometheus()
        self.cadvisor_process = self.start_cadvisor()

    # first set some parameters, before measurement can start
    def setup_flow(self, vnf_name, vnf_interface=None, metric='tx_packets', cookie=0):

        flow_metric = {}

        # check if port is specified (vnf:port)
        if vnf_interface is None:
            # take first interface by default
            connected_sw = self.net.DCNetwork_graph.neighbors(vnf_name)[0]
            link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
            vnf_interface = link_dict[0]['src_port_id']

        flow_metric['vnf_name'] = vnf_name
        flow_metric['vnf_interface'] = vnf_interface

        vnf_switch = None
        for connected_sw in self.net.DCNetwork_graph.neighbors(vnf_name):
            link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
            for link in link_dict:
                if link_dict[link]['src_port_id'] == vnf_interface:
                    # found the right link and connected switch
                    vnf_switch = connected_sw
                    flow_metric['mon_port'] = link_dict[link]['dst_port_nr']
                    break

        if not vnf_switch:
            logging.exception("vnf switch of {0}:{1} not found!".format(vnf_name, vnf_interface))
            return "vnf switch of {0}:{1} not found!".format(vnf_name, vnf_interface)

        try:
            # default port direction to monitor
            if metric is None:
                metric = 'tx_packets'

            next_node = self.net.getNodeByName(vnf_switch)

            if not isinstance(next_node, OVSSwitch):
                logging.info("vnf: {0} is not connected to switch".format(vnf_name))
                return

            flow_metric['previous_measurement'] = 0
            flow_metric['previous_monitor_time'] = 0

            flow_metric['switch_dpid'] = int(str(next_node.dpid), 16)
            flow_metric['metric_key'] = metric
            flow_metric['cookie'] = cookie

            self.monitor_flow_lock.acquire()
            self.flow_metrics.append(flow_metric)
            self.monitor_flow_lock.release()

            logging.info('Started monitoring flow:{3} {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie))
            return 'Started monitoring flow:{3} {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie)

        except Exception as ex:
            logging.exception("setup_metric error.")
            return ex.message

    def stop_flow(self, vnf_name, vnf_interface=None, metric=None, cookie=0):
        for flow_dict in self.flow_metrics:
            if flow_dict['vnf_name'] == vnf_name and flow_dict['vnf_interface'] == vnf_interface \
                    and flow_dict['metric_key'] == metric and flow_dict['cookie'] == cookie:

                self.monitor_flow_lock.acquire()

                self.flow_metrics.remove(flow_dict)

                for collector in self.registry._collectors:
                    if (vnf_name, vnf_interface, cookie) in collector._metrics:
                        collector.remove(vnf_name, vnf_interface, cookie)

                delete_from_gateway(self.pushgateway, job='sonemu-SDNcontroller')

                self.monitor_flow_lock.release()

                logging.info('Stopped monitoring flow {3}: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie))
                return 'Stopped monitoring flow {3}: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric, cookie)


    # first set some parameters, before measurement can start
    def setup_metric(self, vnf_name, vnf_interface=None, metric='tx_packets'):

        network_metric = {}

        # check if port is specified (vnf:port)
        if vnf_interface is None:
            # take first interface by default
            connected_sw = self.net.DCNetwork_graph.neighbors(vnf_name)[0]
            link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
            vnf_interface = link_dict[0]['src_port_id']

        network_metric['vnf_name'] = vnf_name
        network_metric['vnf_interface'] = vnf_interface

        for connected_sw in self.net.DCNetwork_graph.neighbors(vnf_name):
            link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
            for link in link_dict:
                if link_dict[link]['src_port_id'] == vnf_interface:
                    # found the right link and connected switch
                    network_metric['mon_port'] = link_dict[link]['dst_port_nr']
                    break

        if 'mon_port' not in network_metric:
            logging.exception("vnf interface {0}:{1} not found!".format(vnf_name,vnf_interface))
            return "vnf interface {0}:{1} not found!".format(vnf_name,vnf_interface)

        try:
            # default port direction to monitor
            if metric is None:
                metric = 'tx_packets'

            vnf_switch = self.net.DCNetwork_graph.neighbors(str(vnf_name))

            if len(vnf_switch) > 1:
                logging.info("vnf: {0} has multiple ports".format(vnf_name))
                return
            elif len(vnf_switch) == 0:
                logging.info("vnf: {0} is not connected".format(vnf_name))
                return
            else:
                vnf_switch = vnf_switch[0]
            next_node = self.net.getNodeByName(vnf_switch)

            if not isinstance(next_node, OVSSwitch):
                logging.info("vnf: {0} is not connected to switch".format(vnf_name))
                return

            network_metric['previous_measurement'] = 0
            network_metric['previous_monitor_time'] = 0


            network_metric['switch_dpid'] = int(str(next_node.dpid), 16)
            network_metric['metric_key'] = metric

            self.monitor_lock.acquire()

            self.network_metrics.append(network_metric)
            self.monitor_lock.release()


            logging.info('Started monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric))
            return 'Started monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric)

        except Exception as ex:
            logging.exception("setup_metric error.")
            return ex.message

    def stop_metric(self, vnf_name, vnf_interface=None, metric=None):

        for metric_dict in self.network_metrics:
            if metric_dict['vnf_name'] == vnf_name and metric_dict['vnf_interface'] == vnf_interface \
                    and metric_dict['metric_key'] == metric:

                self.monitor_lock.acquire()

                self.network_metrics.remove(metric_dict)

                #this removes the complete metric, all labels...
                #REGISTRY.unregister(self.prom_metrics[metric_dict['metric_key']])
                #self.registry.unregister(self.prom_metrics[metric_dict['metric_key']])

                for collector in self.registry._collectors :

                    """
                    INFO:root:name:sonemu_rx_count_packets
                    labels:('vnf_name', 'vnf_interface')
                    metrics:{(u'tsrc', u'output'): < prometheus_client.core.Gauge
                    object
                    at
                    0x7f353447fd10 >}
                    """
                    logging.info('{0}'.format(collector._metrics.values()))

                    if (vnf_name, vnf_interface, 'None') in collector._metrics:
                        logging.info('2 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames,
                                                                              collector._metrics))
                        collector.remove(vnf_name, vnf_interface, 'None')

                # set values to NaN, prometheus api currently does not support removal of metrics
                #self.prom_metrics[metric_dict['metric_key']].labels(vnf_name, vnf_interface).set(float('nan'))

                # this removes the complete metric, all labels...
                # 1 single monitor job for all metrics of the SDN controller
                # we can only  remove from the pushgateway grouping keys(labels) which we have defined for the add_to_pushgateway
                # we can not specify labels from the metrics to be removed
                # if we need to remove the metrics seperatelty, we need to give them a separate grouping key, and probably a diffferent registry also
                delete_from_gateway(self.pushgateway, job='sonemu-SDNcontroller')

                self.monitor_lock.release()

                logging.info('Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric))
                return 'Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric)

            # delete everything from this vnf
            elif metric_dict['vnf_name'] == vnf_name and vnf_interface is None and metric is None:
                self.monitor_lock.acquire()
                self.network_metrics.remove(metric_dict)
                for collector in self.registry._collectors:
                    collector_dict = collector._metrics.copy()
                    for name, interface, id in collector_dict:
                        if name == vnf_name:
                            logging.info('3 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames,
                                                                           collector._metrics))
                            collector.remove(name, interface, 'None')

                delete_from_gateway(self.pushgateway, job='sonemu-SDNcontroller')
                self.monitor_lock.release()
                logging.info('Stopped monitoring vnf: {0}'.format(vnf_name))
                return 'Stopped monitoring: {0}'.format(vnf_name)


    # get all metrics defined in the list and export it to Prometheus
    def get_flow_metrics(self):
        while self.start_monitoring:

            self.monitor_flow_lock.acquire()

            for flow_dict in self.flow_metrics:
                data = {}

                data['cookie'] = flow_dict['cookie']

                if 'tx' in flow_dict['metric_key']:
                    data['match'] = {'in_port':flow_dict['mon_port']}
                elif 'rx' in flow_dict['metric_key']:
                    data['out_port'] = flow_dict['mon_port']


                # query Ryu
                ret = self.net.ryu_REST('stats/flow', dpid=flow_dict['switch_dpid'], data=data)
                flow_stat_dict = ast.literal_eval(ret)

                logging.debug('received flow stat:{0} '.format(flow_stat_dict))
                self.set_flow_metric(flow_dict, flow_stat_dict)

            self.monitor_flow_lock.release()
            time.sleep(1)

    def get_network_metrics(self):
        while self.start_monitoring:

            self.monitor_lock.acquire()

            # group metrics by dpid to optimize the rest api calls
            dpid_list = [metric_dict['switch_dpid'] for metric_dict in self.network_metrics]
            dpid_set = set(dpid_list)

            for dpid in dpid_set:

                # query Ryu
                ret = self.net.ryu_REST('stats/port', dpid=dpid)
                port_stat_dict = ast.literal_eval(ret)

                metric_list = [metric_dict for metric_dict in self.network_metrics
                               if int(metric_dict['switch_dpid'])==int(dpid)]

                for metric_dict in metric_list:
                    self.set_network_metric(metric_dict, port_stat_dict)

            self.monitor_lock.release()
            time.sleep(1)

    # add metric to the list to export to Prometheus, parse the Ryu port-stats reply
    def set_network_metric(self, metric_dict, port_stat_dict):
        # vnf tx is the datacenter switch rx and vice-versa
        metric_key = self.switch_tx_rx(metric_dict['metric_key'])
        switch_dpid = metric_dict['switch_dpid']
        vnf_name = metric_dict['vnf_name']
        vnf_interface = metric_dict['vnf_interface']
        previous_measurement = metric_dict['previous_measurement']
        previous_monitor_time = metric_dict['previous_monitor_time']
        mon_port = metric_dict['mon_port']

        for port_stat in port_stat_dict[str(switch_dpid)]:
            if int(port_stat['port_no']) == int(mon_port):
                port_uptime = port_stat['duration_sec'] + port_stat['duration_nsec'] * 10 ** (-9)
                this_measurement = int(port_stat[metric_key])

                # set prometheus metric
                self.prom_metrics[metric_dict['metric_key']].\
                    labels({'vnf_name': vnf_name, 'vnf_interface': vnf_interface, 'flow_id': None}).\
                    set(this_measurement)

                # 1 single monitor job for all metrics of the SDN controller
                pushadd_to_gateway(self.pushgateway, job='sonemu-SDNcontroller', registry=self.registry)

                # also the rate is calculated here, but not used for now
                # (rate can be easily queried from prometheus also)
                if previous_monitor_time <= 0 or previous_monitor_time >= port_uptime:
                    metric_dict['previous_measurement'] = int(port_stat[metric_key])
                    metric_dict['previous_monitor_time'] = port_uptime
                    # do first measurement
                    time.sleep(1)
                    self.monitor_lock.release()

                    metric_rate = self.get_network_metrics()
                    return metric_rate

                else:
                    time_delta = (port_uptime - metric_dict['previous_monitor_time'])
                    metric_rate = (this_measurement - metric_dict['previous_measurement']) / float(time_delta)

                metric_dict['previous_measurement'] = this_measurement
                metric_dict['previous_monitor_time'] = port_uptime
                return metric_rate

        logging.exception('metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface))
        return 'metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface)

    def set_flow_metric(self, metric_dict, flow_stat_dict):
        # vnf tx is the datacenter switch rx and vice-versa
        metric_key = metric_dict['metric_key']
        switch_dpid = metric_dict['switch_dpid']
        vnf_name = metric_dict['vnf_name']
        vnf_interface = metric_dict['vnf_interface']
        previous_measurement = metric_dict['previous_measurement']
        previous_monitor_time = metric_dict['previous_monitor_time']
        cookie = metric_dict['cookie']

        # TODO aggregate all found flow stats
        flow_stat = flow_stat_dict[str(switch_dpid)][0]
        if 'bytes' in  metric_key:
            counter = flow_stat['byte_count']
        elif 'packet' in metric_key:
            counter = flow_stat['packet_count']

        flow_uptime = flow_stat['duration_sec'] + flow_stat['duration_nsec'] * 10 ** (-9)

        self.prom_metrics[metric_dict['metric_key']]. \
            labels({'vnf_name': vnf_name, 'vnf_interface': vnf_interface, 'flow_id': cookie}). \
            set(counter)
        pushadd_to_gateway(self.pushgateway, job='sonemu-SDNcontroller', registry=self.registry)


    def start_Prometheus(self, port=9090):
        # prometheus.yml configuration file is located in the same directory as this file
        cmd = ["docker",
               "run",
               "--rm",
               "-p", "{0}:9090".format(port),
               "-v", "{0}/prometheus.yml:/etc/prometheus/prometheus.yml".format(os.path.dirname(os.path.abspath(__file__))),
               "-v", "{0}/profile.rules:/etc/prometheus/profile.rules".format(os.path.dirname(os.path.abspath(__file__))),
               "--name", "prometheus",
               "prom/prometheus"
               ]
        logging.info('Start Prometheus container {0}'.format(cmd))
        return Popen(cmd)

    def start_PushGateway(self, port=9091):
        cmd = ["docker",
               "run",
               "-d",
               "-p", "{0}:9091".format(port),
               "--name", "pushgateway",
               "prom/pushgateway"
               ]

        logging.info('Start Prometheus Push Gateway container {0}'.format(cmd))
        return Popen(cmd)

    def start_cadvisor(self, port=8090):
        cmd = ["docker",
               "run",
               "--rm",
               "--volume=/:/rootfs:ro",
               "--volume=/var/run:/var/run:rw",
               "--volume=/sys:/sys:ro",
               "--volume=/var/lib/docker/:/var/lib/docker:ro",
               "--publish={0}:8080".format(port),
               "--name=cadvisor",
               "google/cadvisor:latest"
               ]
        logging.info('Start cAdvisor container {0}'.format(cmd))
        return Popen(cmd)

    def stop(self):
        # stop the monitoring thread
        self.start_monitoring = False
        self.monitor_thread.join()
        self.monitor_flow_thread.join()

        '''
        if self.prometheus_process is not None:
            logging.info('stopping prometheus container')
            self.prometheus_process.terminate()
            self.prometheus_process.kill()
            self._stop_container('prometheus')

        if self.pushgateway_process is not None:
            logging.info('stopping pushgateway container')
            self.pushgateway_process.terminate()
            self.pushgateway_process.kill()
            self._stop_container('pushgateway')
        '''

        if self.cadvisor_process is not None:
            logging.info('stopping cadvisor container')
            self.cadvisor_process.terminate()
            self.cadvisor_process.kill()
            self._stop_container('cadvisor')

    def switch_tx_rx(self,metric=''):
        # when monitoring vnfs, the tx of the datacenter switch is actually the rx of the vnf
        # so we need to change the metric name to be consistent with the vnf rx or tx
        if 'tx' in metric:
            metric = metric.replace('tx','rx')
        elif 'rx' in metric:
            metric = metric.replace('rx','tx')

        return metric

    def _stop_container(self, name):
        cmd = ["docker",
               "stop",
               name]
        Popen(cmd).wait()

        cmd = ["docker",
               "rm",
               name]
        Popen(cmd).wait()