start cadvisor and prometheus docker container at startup
[osm/vim-emu.git] / src / emuvim / dcemulator / monitoring.py
1 __author__ = 'Administrator'
2
3 import urllib2
4 import logging
5 from mininet.node import OVSSwitch
6 import ast
7 import time
8 from prometheus_client import start_http_server, Summary, Histogram, Gauge, Counter
9 import threading
10 from subprocess import Popen, PIPE
11 import os
12
13 logging.basicConfig(level=logging.INFO)
14
15 """
16 class to read openflow stats from the Ryu controller of the DCNetwork
17 """
18
19 class DCNetworkMonitor():
20 def __init__(self, net):
21 self.net = net
22 # link to Ryu REST_API
23 self.ip = '0.0.0.0'
24 self.port = '8080'
25 self.REST_api = 'http://{0}:{1}'.format(self.ip,self.port)
26
27 # helper variables to calculate the metrics
28 # Start up the server to expose the metrics to Prometheus.
29 start_http_server(8000)
30 # supported Prometheus metrics
31 self.prom_tx_packet_count = Gauge('sonemu_tx_count_packets', 'Total number of packets sent',
32 ['vnf_name', 'vnf_interface'])
33 self.prom_rx_packet_count = Gauge('sonemu_rx_count_packets', 'Total number of packets received',
34 ['vnf_name', 'vnf_interface'])
35 self.prom_tx_byte_count = Gauge('sonemu_tx_count_bytes', 'Total number of bytes sent',
36 ['vnf_name', 'vnf_interface'])
37 self.prom_rx_byte_count = Gauge('sonemu_rx_count_bytes', 'Total number of bytes received',
38 ['vnf_name', 'vnf_interface'])
39
40 self.prom_metrics={'tx_packets':self.prom_tx_packet_count, 'rx_packets':self.prom_rx_packet_count,
41 'tx_bytes':self.prom_tx_byte_count,'rx_bytes':self.prom_rx_byte_count}
42
43 # list of installed metrics to monitor
44 # each entry can contain this data
45 '''
46 {
47 switch_dpid = 0
48 vnf_name = None
49 vnf_interface = None
50 previous_measurement = 0
51 previous_monitor_time = 0
52 metric_key = None
53 mon_port = None
54 }
55 '''
56 self.network_metrics = []
57
58 # start monitoring thread
59 self.start_monitoring = True
60 self.monitor_thread = threading.Thread(target=self.get_network_metrics)
61 self.monitor_thread.start()
62
63 # helper tools
64 self.prometheus_process = self.start_Prometheus()
65 self.cadvisor_process = self.start_cadvisor()
66
67 # first set some parameters, before measurement can start
68 def setup_metric(self, vnf_name, vnf_interface=None, metric='tx_packets'):
69
70 network_metric = {}
71
72 # check if port is specified (vnf:port)
73 if vnf_interface is None:
74 # take first interface by default
75 connected_sw = self.net.DCNetwork_graph.neighbors(vnf_name)[0]
76 link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
77 vnf_interface = link_dict[0]['src_port_id']
78
79 network_metric['vnf_name'] = vnf_name
80 network_metric['vnf_interface'] = vnf_interface
81
82 for connected_sw in self.net.DCNetwork_graph.neighbors(vnf_name):
83 link_dict = self.net.DCNetwork_graph[vnf_name][connected_sw]
84 for link in link_dict:
85 # logging.info("{0},{1}".format(link_dict[link],vnf_interface))
86 if link_dict[link]['src_port_id'] == vnf_interface:
87 # found the right link and connected switch
88 # logging.info("{0},{1}".format(link_dict[link]['src_port_id'], vnf_source_interface))
89 network_metric['mon_port'] = link_dict[link]['dst_port']
90 break
91
92 if 'mon_port' not in network_metric:
93 logging.exception("vnf interface {0}:{1} not found!".format(vnf_name,vnf_interface))
94 return "vnf interface {0}:{1} not found!".format(vnf_name,vnf_interface)
95
96 try:
97 # default port direction to monitor
98 if metric is None:
99 metric = 'tx_packets'
100
101 vnf_switch = self.net.DCNetwork_graph.neighbors(str(vnf_name))
102
103 if len(vnf_switch) > 1:
104 logging.info("vnf: {0} has multiple ports".format(vnf_name))
105 return
106 elif len(vnf_switch) == 0:
107 logging.info("vnf: {0} is not connected".format(vnf_name))
108 return
109 else:
110 vnf_switch = vnf_switch[0]
111 next_node = self.net.getNodeByName(vnf_switch)
112
113 if not isinstance(next_node, OVSSwitch):
114 logging.info("vnf: {0} is not connected to switch".format(vnf_name))
115 return
116
117 network_metric['previous_measurement'] = 0
118 network_metric['previous_monitor_time'] = 0
119
120
121 network_metric['switch_dpid'] = int(str(next_node.dpid), 16)
122 network_metric['metric_key'] = metric
123
124
125 self.network_metrics.append(network_metric)
126
127 logging.info('Started monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric))
128 return 'Started monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric)
129
130 except Exception as ex:
131 logging.exception("get_rate error.")
132 return ex.message
133
134 def remove_metric(self, vnf_name, vnf_interface, metric):
135 for metric_dict in self.network_metrics:
136 if metric_dict['vnf_name'] == vnf_name and metric_dict['vnf_interface'] == vnf_interface \
137 and metric_dict['metric'] == metric:
138 self.network_metrics.remove(metric_dict)
139 logging.info('Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric))
140 return 'Stopped monitoring: {2} on {0}:{1}'.format(vnf_name, vnf_interface, metric)
141
142
143 # get all metrics defined in the list and export it to Prometheus
144 def get_network_metrics(self):
145 while self.start_monitoring:
146 # group metrics by dpid to optimize the rest api calls
147 dpid_list = [metric_dict['switch_dpid'] for metric_dict in self.network_metrics]
148 dpid_set = set(dpid_list)
149
150 for dpid in dpid_set:
151
152 # query Ryu
153 ret = self.REST_cmd('stats/port', dpid)
154 port_stat_dict = ast.literal_eval(ret)
155
156 metric_list = [metric_dict for metric_dict in self.network_metrics
157 if int(metric_dict['switch_dpid'])==int(dpid)]
158 #logging.info('1set prom packets:{0} '.format(self.network_metrics))
159 for metric_dict in metric_list:
160 self.set_network_metric(metric_dict, port_stat_dict)
161
162 time.sleep(1)
163
164 # add metric to the list to export to Prometheus, parse the Ryu port-stats reply
165 def set_network_metric(self, metric_dict, port_stat_dict):
166 # vnf tx is the datacenter switch rx and vice-versa
167 metric_key = self.switch_tx_rx(metric_dict['metric_key'])
168 switch_dpid = metric_dict['switch_dpid']
169 vnf_name = metric_dict['vnf_name']
170 vnf_interface = metric_dict['vnf_interface']
171 previous_measurement = metric_dict['previous_measurement']
172 previous_monitor_time = metric_dict['previous_monitor_time']
173 mon_port = metric_dict['mon_port']
174
175 for port_stat in port_stat_dict[str(switch_dpid)]:
176 if int(port_stat['port_no']) == int(mon_port):
177 port_uptime = port_stat['duration_sec'] + port_stat['duration_nsec'] * 10 ** (-9)
178 this_measurement = int(port_stat[metric_key])
179 #logging.info('set prom packets:{0} {1}:{2}'.format(this_measurement, vnf_name, vnf_interface))
180
181 # set prometheus metric
182 self.prom_metrics[metric_key].labels(vnf_name, vnf_interface).set(this_measurement)
183
184 if previous_monitor_time <= 0 or previous_monitor_time >= port_uptime:
185 metric_dict['previous_measurement'] = int(port_stat[metric_key])
186 metric_dict['previous_monitor_time'] = port_uptime
187 # do first measurement
188 #logging.info('first measurement')
189 time.sleep(1)
190 byte_rate = self.get_network_metrics()
191 return byte_rate
192 else:
193 time_delta = (port_uptime - metric_dict['previous_monitor_time'])
194 byte_rate = (this_measurement - metric_dict['previous_measurement']) / float(time_delta)
195 # logging.info('uptime:{2} delta:{0} rate:{1}'.format(time_delta,byte_rate,port_uptime))
196
197 metric_dict['previous_measurement'] = this_measurement
198 metric_dict['previous_monitor_time'] = port_uptime
199 return byte_rate
200
201 logging.exception('metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface))
202 return 'metric {0} not found on {1}:{2}'.format(metric_key, vnf_name, vnf_interface)
203
204
205 def REST_cmd(self, prefix, dpid):
206 url = self.REST_api + '/' + str(prefix) + '/' + str(dpid)
207 req = urllib2.Request(url)
208 ret = urllib2.urlopen(req).read()
209 return ret
210
211 def start_Prometheus(self, port=9090):
212 # prometheus.yml configuration file is located in the same directory as this file
213 cmd = ["docker",
214 "run",
215 "--rm",
216 "-p", "{0}:9090".format(port),
217 "-v", "{0}/prometheus.yml:/etc/prometheus/prometheus.yml".format(os.path.dirname(os.path.abspath(__file__))),
218 "--name", "prometheus",
219 "prom/prometheus"
220 ]
221 logging.info('Start Prometheus container {0}'.format(cmd))
222 return Popen(cmd)
223
224 def start_cadvisor(self, port=8090):
225 cmd = ["docker",
226 "run",
227 "--rm",
228 "--volume=/:/rootfs:ro",
229 "--volume=/var/run:/var/run:rw",
230 "--volume=/sys:/sys:ro",
231 "--volume=/var/lib/docker/:/var/lib/docker:ro",
232 "--publish={0}:8080".format(port),
233 "--name=cadvisor",
234 "google/cadvisor:latest"
235 ]
236 logging.info('Start cAdvisor container {0}'.format(cmd))
237 return Popen(cmd)
238
239 def stop(self):
240 # stop the monitoring thread
241 self.start_monitoring = False
242 self.monitor_thread.join()
243
244 if self.prometheus_process is not None:
245 logging.info('stopping prometheus container')
246 self.prometheus_process.terminate()
247 self.prometheus_process.kill()
248 self._stop_container('prometheus')
249
250 if self.cadvisor_process is not None:
251 logging.info('stopping cadvisor container')
252 self.cadvisor_process.terminate()
253 self.cadvisor_process.kill()
254 self._stop_container('cadvisor')
255
256 def switch_tx_rx(self,metric=''):
257 # when monitoring vnfs, the tx of the datacenter switch is actually the rx of the vnf
258 # so we need to change the metric name to be consistent with the vnf rx or tx
259 if 'tx' in metric:
260 metric = metric.replace('tx','rx')
261 elif 'rx' in metric:
262 metric = metric.replace('rx','tx')
263
264 return metric
265
266 def _stop_container(self, name):
267 cmd = ["docker",
268 "stop",
269 name]
270 Popen(cmd).wait()
271
272 cmd = ["docker",
273 "rm",
274 name]
275 Popen(cmd).wait()
276
277