doctor_tests/main.py

   1 ##############################################################################
   2 # Copyright (c) 2017 ZTE Corporation and others.
   3 #
   4 # All rights reserved. This program and the accompanying materials
   5 # are made available under the terms of the Apache License, Version 2.0
   6 # which accompanies this distribution, and is available at
   7 # http://www.apache.org/licenses/LICENSE-2.0
   8 ##############################################################################
   9 import os
  10 from os.path import isfile, join
  11 import random
  12 import sys
  13 import time
  14
  15 from doctor_tests.alarm import Alarm
  16 from doctor_tests.common.constants import Host
  17 from doctor_tests.common.utils import match_rep_in_file
  18 from doctor_tests import config
  19 from doctor_tests.consumer import get_consumer
  20 from doctor_tests.identity_auth import get_identity_auth
  21 from doctor_tests.identity_auth import get_session
  22 from doctor_tests.image import Image
  23 from doctor_tests.instance import Instance
  24 from doctor_tests.inspector import get_inspector
  25 from doctor_tests.installer import get_installer
  26 import doctor_tests.logger as doctor_log
  27 from doctor_tests.network import Network
  28 from doctor_tests.monitor import get_monitor
  29 from doctor_tests.os_clients import nova_client
  30 from doctor_tests.profiler_poc import main as profiler_main
  31 from doctor_tests.scenario.common import calculate_notification_time
  32 from doctor_tests.scenario.network_failure import NetworkFault
  33 from doctor_tests.user import User
  34
  35
  36 LOG = doctor_log.Logger('doctor').getLogger()
  37
  38
  39 class DoctorTest(object):
  40
  41     def __init__(self, conf):
  42         self.conf = conf
  43         self.image = Image(self.conf, LOG)
  44         self.user = User(self.conf, LOG)
  45         self.network = Network(self.conf, LOG)
  46         self.instance = Instance(self.conf, LOG)
  47         self.alarm = Alarm(self.conf, LOG)
  48         self.installer = get_installer(self.conf, LOG)
  49         self.inspector = get_inspector(self.conf, LOG)
  50         self.monitor = get_monitor(self.conf,
  51                                    self.inspector.get_inspector_url(),
  52                                    LOG)
  53         self.consumer = get_consumer(self.conf, LOG)
  54         self.fault = NetworkFault(self.conf, self.installer, LOG)
  55         auth = get_identity_auth(project=self.conf.doctor_project)
  56         self.nova = nova_client(self.conf.nova_version,
  57                                 get_session(auth=auth))
  58         self.down_host = None
  59
  60     def setup(self):
  61         # prepare the cloud env
  62         self.installer.setup()
  63
  64         # preparing VM image...
  65         self.image.create()
  66
  67         # creating test user...
  68         self.user.create()
  69         self.user.update_quota()
  70
  71         # creating VM...
  72         self.network.create()
  73         self.instance.create()
  74         self.instance.wait_for_vm_launch()
  75
  76         # creating alarm...
  77         self.alarm.create()
  78
  79         # starting doctor sample components...
  80         self.inspector.start()
  81
  82         self.down_host = self.get_host_info_for_random_vm()
  83         self.monitor.start(self.down_host)
  84
  85         self.consumer.start()
  86
  87     def run(self):
  88         """run doctor test"""
  89         try:
  90             LOG.info('doctor test starting.......')
  91
  92             # prepare test env
  93             self.setup()
  94
  95             # wait for aodh alarms are updated in caches for event evaluator,
  96             # sleep time should be larger than event_alarm_cache_ttl(default 60)
  97             time.sleep(60)
  98
  99             # injecting host failure...
 100             # NOTE (umar) add INTERFACE_NAME logic to host injection
 101
 102             self.fault.start(self.down_host)
 103             time.sleep(10)
 104
 105             # verify the test results
 106             # NOTE (umar) copy remote monitor.log file when monitor=collectd
 107             self.check_host_status(self.down_host.name, 'down')
 108
 109             notification_time = calculate_notification_time(LOG.filename)
 110             if notification_time < 1 and notification_time > 0:
 111                 LOG.info('doctor test successfully, notification_time=%s' % notification_time)
 112             else:
 113                 LOG.error('doctor test failed, notification_time=%s' % notification_time)
 114                 sys.exit(1)
 115
 116             if self.conf.profiler_type:
 117                 LOG.info('doctor test begin to run profile.......')
 118                 self.collect_logs()
 119                 self.run_profiler()
 120         except Exception as e:
 121             LOG.error('doctor test failed, Exception=%s' % e)
 122             sys.exit(1)
 123         finally:
 124             self.cleanup()
 125
 126     def get_host_info_for_random_vm(self):
 127         num = random.randint(0, self.conf.instance_count - 1)
 128         vm_name = "%s%d" % (self.conf.instance_basename, num)
 129
 130         servers = \
 131             {getattr(server, 'name'): server
 132              for server in self.nova.servers.list()}
 133         server = servers.get(vm_name)
 134         if not server:
 135             raise \
 136                 Exception('Can not find instance: vm_name(%s)' % vm_name)
 137         host_name = server.__dict__.get('OS-EXT-SRV-ATTR:hypervisor_hostname')
 138         host_ip = self.installer.get_host_ip_from_hostname(host_name)
 139
 140         LOG.info('Get host info(name:%s, ip:%s) which vm(%s) launched at'
 141                  % (host_name, host_ip, vm_name))
 142         return Host(host_name, host_ip)
 143
 144     def check_host_status(self, hostname, state):
 145         service = self.nova.services.list(host=hostname, binary='nova-compute')
 146         host_state = service[0].__dict__.get('state')
 147         assert host_state == state
 148
 149     def unset_forced_down_hosts(self):
 150         if self.down_host:
 151             self.nova.services.force_down(self.down_host.name, 'nova-compute', False)
 152             time.sleep(2)
 153             self.check_host_status(self.down_host.name, 'up')
 154
 155     def collect_logs(self):
 156         self.fault.get_disable_network_log()
 157
 158     def run_profiler(self):
 159         test_dir = os.path.split(os.path.realpath(__file__))[0]
 160
 161         reg = '(?<=doctor set link down at )\d+.\d+'
 162         linkdown = float(match_rep_in_file(reg, LOG.filename).group(0))
 163
 164         reg = '(.* doctor mark vm.* error at )(\d+.\d+)'
 165         vmdown = float(match_rep_in_file(reg, LOG.filename).group(2))
 166
 167         reg = '(.* doctor mark host.* down at )(\d+.\d+)'
 168         hostdown = float(match_rep_in_file(reg, LOG.filename).group(2))
 169
 170         reg = '(?<=doctor monitor detected at )\d+.\d+'
 171         detected = float(match_rep_in_file(reg, LOG.filename).group(0))
 172
 173         reg = '(?<=doctor consumer notified at )\d+.\d+'
 174         notified = float(match_rep_in_file(reg, LOG.filename).group(0))
 175
 176         # TODO(yujunz) check the actual delay to verify time sync status
 177         # expected ~1s delay from $trigger to $linkdown
 178         relative_start = linkdown
 179         os.environ['DOCTOR_PROFILER_T00'] = str(int((linkdown - relative_start)*1000))
 180         os.environ['DOCTOR_PROFILER_T01'] = str(int((detected - relative_start) * 1000))
 181         os.environ['DOCTOR_PROFILER_T03'] = str(int((vmdown - relative_start) * 1000))
 182         os.environ['DOCTOR_PROFILER_T04'] = str(int((hostdown - relative_start) * 1000))
 183         os.environ['DOCTOR_PROFILER_T09'] = str(int((notified - relative_start) * 1000))
 184
 185         profiler_main(log=LOG)
 186
 187     def cleanup(self):
 188         self.unset_forced_down_hosts()
 189         self.inspector.stop()
 190         self.monitor.stop()
 191         self.consumer.stop()
 192         self.installer.cleanup()
 193         self.alarm.delete()
 194         self.instance.delete()
 195         self.network.delete()
 196         self.image.delete()
 197         self.fault.cleanup()
 198         self.user.delete()
 199
 200
 201 def main():
 202     """doctor main"""
 203     test_dir = os.path.split(os.path.realpath(__file__))[0]
 204     doctor_root_dir = os.path.dirname(test_dir)
 205
 206     config_file_dir = '{0}/{1}'.format(doctor_root_dir, 'etc/')
 207     config_files = [join(config_file_dir, f) for f in os.listdir(config_file_dir)
 208                     if isfile(join(config_file_dir, f))]
 209
 210     conf = config.prepare_conf(args=sys.argv[1:],
 211                                config_files=config_files)
 212
 213     doctor = DoctorTest(conf)
 214     doctor.run()