X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=src%2Femuvim%2Fdcemulator%2Fmonitoring.py;h=b3ec7ee1be65dd62e6c0def94e76966bd6c0220f;hb=edcbeeba7236e852e1ff5b02c3704e0749dd3be3;hp=d7450698f2f1a92d8649e36773652c888a13a72b;hpb=8675d09a5488f60a856145f92380db6066200484;p=osm%2Fvim-emu.git diff --git a/src/emuvim/dcemulator/monitoring.py b/src/emuvim/dcemulator/monitoring.py index d745069..b3ec7ee 100755 --- a/src/emuvim/dcemulator/monitoring.py +++ b/src/emuvim/dcemulator/monitoring.py @@ -5,10 +5,11 @@ import logging from mininet.node import OVSSwitch import ast import time -from prometheus_client import start_http_server, Summary, Histogram, Gauge, Counter +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 -from os import getcwd +from subprocess import Popen, PIPE +import os logging.basicConfig(level=logging.INFO) @@ -25,27 +26,19 @@ class DCNetworkMonitor(): self.REST_api = 'http://{0}:{1}'.format(self.ip,self.port) # helper variables to calculate the metrics - # TODO put these in a list to support multiple metrics simultaneously - self.switch_dpid = 0 - self.vnf_name = None - self.vnf_interface = None - self.previous_measurement = 0 - self.previous_monitor_time = 0 - self.metric_key = None - self.mon_port = None - - + self.pushgateway = 'localhost:9091' # Start up the server to expose the metrics to Prometheus. - start_http_server(8000) + #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']) + ['vnf_name', 'vnf_interface'], registry=self.registry) self.prom_rx_packet_count = Gauge('sonemu_rx_count_packets', 'Total number of packets received', - ['vnf_name', 'vnf_interface']) + ['vnf_name', 'vnf_interface'], registry=self.registry) self.prom_tx_byte_count = Gauge('sonemu_tx_count_bytes', 'Total number of bytes sent', - ['vnf_name', 'vnf_interface']) + ['vnf_name', 'vnf_interface'], registry=self.registry) self.prom_rx_byte_count = Gauge('sonemu_rx_count_bytes', 'Total number of bytes received', - ['vnf_name', 'vnf_interface']) + ['vnf_name', 'vnf_interface'], 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} @@ -63,16 +56,18 @@ class DCNetworkMonitor(): mon_port = None } ''' + self.monitor_lock = threading.Lock() self.network_metrics = [] # start monitoring thread + self.start_monitoring = True self.monitor_thread = threading.Thread(target=self.get_network_metrics) self.monitor_thread.start() # helper tools - self.prometheus_process = None - self.cAdvisor_process = None - + 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_metric(self, vnf_name, vnf_interface=None, metric='tx_packets'): @@ -88,8 +83,6 @@ class DCNetworkMonitor(): network_metric['vnf_name'] = vnf_name network_metric['vnf_interface'] = vnf_interface - #self.vnf_name = vnf_name - #self.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] @@ -99,7 +92,6 @@ class DCNetworkMonitor(): # found the right link and connected switch # logging.info("{0},{1}".format(link_dict[link]['src_port_id'], vnf_source_interface)) network_metric['mon_port'] = link_dict[link]['dst_port'] - # self.mon_port = link_dict[link]['dst_port'] break if 'mon_port' not in network_metric: @@ -129,27 +121,78 @@ class DCNetworkMonitor(): network_metric['previous_measurement'] = 0 network_metric['previous_monitor_time'] = 0 - #self.previous_measurement = 0 - #self.previous_monitor_time = 0 + network_metric['switch_dpid'] = int(str(next_node.dpid), 16) network_metric['metric_key'] = metric - #self.switch_dpid = int(str(next_node.dpid), 16) - #self.metric_key = '{0}_{1}'.format(direction, 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("get_rate error.") + logging.exception("setup_metric error.") return ex.message + def stop_metric(self, vnf_name, vnf_interface, metric): + + 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']]) - # get all metrics defined in the list + for collector in self.registry._collectors : + logging.info('name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames, collector._metrics)) + """ + 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 self.prom_metrics[metric_dict['metric_key']] + if (vnf_name, vnf_interface) in collector._metrics: + logging.info('2 name:{0} labels:{1} metrics:{2}'.format(collector._name, collector._labelnames, + collector._metrics)) + #collector._metrics = {} + collector.remove(vnf_name, vnf_interface) + + # 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) + + + # get all metrics defined in the list and export it to Prometheus def get_network_metrics(self): - while True: + 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) @@ -166,12 +209,13 @@ class DCNetworkMonitor(): for metric_dict in metric_list: self.set_network_metric(metric_dict, port_stat_dict) + self.monitor_lock.release() time.sleep(1) - # call this function repeatedly for streaming measurements + # 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): - - metric_key = metric_dict['metric_key'] + # 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'] @@ -186,7 +230,14 @@ class DCNetworkMonitor(): #logging.info('set prom packets:{0} {1}:{2}'.format(this_measurement, vnf_name, vnf_interface)) # set prometheus metric - self.prom_metrics[metric_key].labels(vnf_name, vnf_interface).set(this_measurement) + self.prom_metrics[metric_dict['metric_key']].\ + labels({'vnf_name':vnf_name, 'vnf_interface':vnf_interface}).\ + set(this_measurement) + #push_to_gateway(self.pushgateway, job='SDNcontroller', + # grouping_key={'metric':metric_dict['metric_key']}, registry=self.registry) + + # 1 single monitor job for all metrics of the SDN controller + pushadd_to_gateway(self.pushgateway, job='sonemu-SDNcontroller', registry=self.registry) if previous_monitor_time <= 0 or previous_monitor_time >= port_uptime: metric_dict['previous_measurement'] = int(port_stat[metric_key]) @@ -194,58 +245,24 @@ class DCNetworkMonitor(): # do first measurement #logging.info('first measurement') time.sleep(1) - byte_rate = self.get_network_metrics() - return byte_rate + self.monitor_lock.release() + + metric_rate = self.get_network_metrics() + return metric_rate + else: time_delta = (port_uptime - metric_dict['previous_monitor_time']) - byte_rate = (this_measurement - metric_dict['previous_measurement']) / float(time_delta) - # logging.info('uptime:{2} delta:{0} rate:{1}'.format(time_delta,byte_rate,port_uptime)) + metric_rate = (this_measurement - metric_dict['previous_measurement']) / float(time_delta) + logging.info('metric: {0} rate:{1}'.format(metric_dict['metric_key'], metric_rate)) metric_dict['previous_measurement'] = this_measurement metric_dict['previous_monitor_time'] = port_uptime - return byte_rate + 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) - # call this function repeatedly for streaming measurements - def get_rate(self, vnf_name, vnf_interface=None, direction='tx', metric='packets'): - - key = self.metric_key - - ret = self.REST_cmd('stats/port', self.switch_dpid) - port_stat_dict = ast.literal_eval(ret) - for port_stat in port_stat_dict[str(self.switch_dpid)]: - if port_stat['port_no'] == self.mon_port: - port_uptime = port_stat['duration_sec'] + port_stat['duration_nsec'] * 10 ** (-9) - this_measurement = int(port_stat[key]) - #logging.info('packets:{0}'.format(this_measurement)) - - # set prometheus metrics - if metric == 'packets': - self.prom_tx_packet_count.labels(self.vnf_name, self.vnf_interface).set(this_measurement) - elif metric == 'bytes': - self.prom_tx_byte_count.labels(self.vnf_name, self.vnf_interface).set(this_measurement) - - if self.previous_monitor_time <= 0 or self.previous_monitor_time >= port_uptime: - self.previous_measurement = int(port_stat[key]) - self.previous_monitor_time = port_uptime - # do first measurement - time.sleep(1) - byte_rate = self.get_rate(vnf_name, vnf_interface, direction, metric) - return byte_rate - else: - time_delta = (port_uptime - self.previous_monitor_time) - byte_rate = (this_measurement - self.previous_measurement) / float(time_delta) - #logging.info('uptime:{2} delta:{0} rate:{1}'.format(time_delta,byte_rate,port_uptime)) - - self.previous_measurement = this_measurement - self.previous_monitor_time = port_uptime - return byte_rate - - return ret - def REST_cmd(self, prefix, dpid): url = self.REST_api + '/' + str(prefix) + '/' + str(dpid) req = urllib2.Request(url) @@ -253,18 +270,32 @@ class DCNetworkMonitor(): return ret 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(getcwd()), + "-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) - self.prometheus_process = Popen(cmd) + def start_PushGateway(self, port=9091): + cmd = ["docker", + "run", + "-d", + "-p", "{0}:9091".format(port), + "--name", "pushgateway", + "prom/pushgateway" + ] - def start_cAdvisor(self, port=8090): + logging.info('Start Prometheus Push Gateway container {0}'.format(cmd)) + return Popen(cmd) + + def start_cadvisor(self, port=8090): cmd = ["docker", "run", "--rm", @@ -276,13 +307,51 @@ class DCNetworkMonitor(): "--name=cadvisor", "google/cadvisor:latest" ] - self.cAdvisor_process = Popen(cmd) + 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() + 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() + - if self.cAdvisor_process is not None: - self.cAdvisor_process.terminate() - self.cAdvisor_process.kill()