doctor_tests/scenario/maintenance.py

   1 ##############################################################################
   2 # Copyright (c) 2018 Nokia 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 datetime
  10 import json
  11 import requests
  12 import time
  13
  14 from doctor_tests.admin_tool import get_admin_tool
  15 from doctor_tests.app_manager import get_app_manager
  16 from doctor_tests.common.utils import get_doctor_test_root_dir
  17 from doctor_tests.identity_auth import get_identity_auth
  18 from doctor_tests.identity_auth import get_session
  19 from doctor_tests.inspector import get_inspector
  20 from doctor_tests.os_clients import keystone_client
  21 from doctor_tests.os_clients import neutron_client
  22 from doctor_tests.os_clients import nova_client
  23 from doctor_tests.stack import Stack
  24
  25
  26 class Maintenance(object):
  27
  28     def __init__(self, trasport_url, conf, log):
  29         self.conf = conf
  30         self.log = log
  31         self.keystone = keystone_client(
  32             self.conf.keystone_version, get_session())
  33         self.nova = nova_client(conf.nova_version, get_session())
  34         auth = get_identity_auth(project=self.conf.doctor_project)
  35         self.neutron = neutron_client(get_session(auth=auth))
  36         self.stack = Stack(self.conf, self.log)
  37         if self.conf.admin_tool.type == 'sample':
  38             self.admin_tool = get_admin_tool(trasport_url, self.conf, self.log)
  39             self.endpoint = 'maintenance'
  40         else:
  41             self.endpoint = 'v1/maintenance'
  42         self.app_manager = get_app_manager(self.stack, self.conf, self.log)
  43         self.inspector = get_inspector(self.conf, self.log, trasport_url)
  44
  45     def get_external_network(self):
  46         ext_net = None
  47         networks = self.neutron.list_networks()['networks']
  48         for network in networks:
  49             if network['router:external']:
  50                 ext_net = network['name']
  51                 break
  52         if ext_net is None:
  53             raise Exception("external network not defined")
  54         return ext_net
  55
  56     def setup_maintenance(self, user):
  57         # each hypervisor needs to have same amount of vcpus and they
  58         # need to be free before test
  59         hvisors = self.nova.hypervisors.list(detailed=True)
  60         prev_vcpus = 0
  61         prev_hostname = ''
  62         self.log.info('checking hypervisors.......')
  63         for hvisor in hvisors:
  64             vcpus = hvisor.__getattr__('vcpus')
  65             vcpus_used = hvisor.__getattr__('vcpus_used')
  66             hostname = hvisor.__getattr__('hypervisor_hostname')
  67             if vcpus < 2:
  68                 raise Exception('not enough vcpus (%d) on %s' %
  69                                 (vcpus, hostname))
  70             if vcpus_used > 0:
  71                 if self.conf.test_case == 'all':
  72                     # VCPU might not yet be free after fault_management test
  73                     self.log.info('%d vcpus used on %s, retry...'
  74                                   % (vcpus_used, hostname))
  75                     time.sleep(15)
  76                     hvisor = self.nova.hypervisors.get(hvisor.id)
  77                     vcpus_used = hvisor.__getattr__('vcpus_used')
  78                 if vcpus_used > 0:
  79                     raise Exception('%d vcpus used on %s'
  80                                     % (vcpus_used, hostname))
  81             if prev_vcpus != 0 and prev_vcpus != vcpus:
  82                 raise Exception('%d vcpus on %s does not match to'
  83                                 '%d on %s'
  84                                 % (vcpus, hostname,
  85                                    prev_vcpus, prev_hostname))
  86             prev_vcpus = vcpus
  87             prev_hostname = hostname
  88
  89         # maintenance flavor made so that 2 instances take whole node
  90         flavor_vcpus = int(vcpus / 2)
  91         compute_nodes = len(hvisors)
  92         amount_actstdby_instances = 2
  93         amount_noredundancy_instances = 2 * compute_nodes - 2
  94         self.log.info('testing %d computes with %d vcpus each'
  95                       % (compute_nodes, vcpus))
  96         self.log.info('testing %d actstdby and %d noredundancy instances'
  97                       % (amount_actstdby_instances,
  98                          amount_noredundancy_instances))
  99         max_instances = (amount_actstdby_instances +
 100                          amount_noredundancy_instances)
 101         max_cores = compute_nodes * vcpus
 102
 103         user.update_quota(max_instances, max_cores)
 104
 105         test_dir = get_doctor_test_root_dir()
 106         template_file = '{0}/{1}'.format(test_dir, 'maintenance_hot_tpl.yaml')
 107         files, template = self.stack.get_hot_tpl(template_file)
 108
 109         ext_net = self.get_external_network()
 110
 111         parameters = {'ext_net': ext_net,
 112                       'flavor_vcpus': flavor_vcpus,
 113                       'maint_image': self.conf.image_name,
 114                       'nonha_intances': amount_noredundancy_instances,
 115                       'ha_intances': amount_actstdby_instances}
 116
 117         self.log.info('creating maintenance stack.......')
 118         self.log.info('parameters: %s' % parameters)
 119
 120         self.stack.create('doctor_test_maintenance',
 121                           template,
 122                           parameters=parameters,
 123                           files=files)
 124
 125         if self.conf.admin_tool.type == 'sample':
 126             self.admin_tool.start()
 127         else:
 128             # TBD Now we expect Fenix is running in self.conf.admin_tool.port
 129             pass
 130         self.app_manager.start()
 131         self.inspector.start()
 132
 133     def start_maintenance(self):
 134         self.log.info('start maintenance.......')
 135         hvisors = self.nova.hypervisors.list(detailed=True)
 136         maintenance_hosts = list()
 137         for hvisor in hvisors:
 138             hostname = hvisor.__getattr__('hypervisor_hostname')
 139             maintenance_hosts.append(hostname)
 140
 141         url = ('http://%s:%s/%s' %
 142                (self.conf.admin_tool.ip,
 143                 self.conf.admin_tool.port,
 144                 self.endpoint))
 145         headers = {
 146             'Content-Type': 'application/json',
 147             'Accept': 'application/json'}
 148
 149         retries = 12
 150         while retries > 0:
 151             # let's start maintenance 20sec from now, so projects will have
 152             # time to ACK to it before that
 153             maintenance_at = (datetime.datetime.utcnow() +
 154                               datetime.timedelta(seconds=30)
 155                               ).strftime('%Y-%m-%d %H:%M:%S')
 156
 157             data = {'state': 'MAINTENANCE',
 158                     'maintenance_at': maintenance_at,
 159                     'metadata': {'openstack_version': 'Rocky'},
 160                     'workflow': 'default'}
 161
 162             if self.conf.admin_tool.type == 'sample':
 163                 data['hosts'] = maintenance_hosts
 164             else:
 165                 data['hosts'] = []
 166             try:
 167                 ret = requests.post(url, data=json.dumps(data),
 168                                     headers=headers)
 169             except:
 170                 if retries == 0:
 171                     raise Exception('admin tool did not respond in 120s')
 172                 else:
 173                     self.log.info('admin tool not ready, retry in 10s')
 174                 retries = retries - 1
 175                 time.sleep(10)
 176                 continue
 177             break
 178         if ret.status_code != 200:
 179             raise Exception(ret.text)
 180         return ret.json()['session_id']
 181
 182     def remove_maintenance_session(self, session_id):
 183         self.log.info('remove maintenance session %s.......' % session_id)
 184         url = ('http://%s:%s/%s/%s' %
 185                (self.conf.admin_tool.ip,
 186                 self.conf.admin_tool.port,
 187                 self.endpoint,
 188                 session_id))
 189
 190         headers = {
 191             'Content-Type': 'application/json',
 192             'Accept': 'application/json'}
 193
 194         ret = requests.delete(url, data=None, headers=headers)
 195         if ret.status_code != 200:
 196             raise Exception(ret.text)
 197
 198     def get_maintenance_state(self, session_id):
 199         url = ('http://%s:%s/%s/%s' %
 200                (self.conf.admin_tool.ip,
 201                 self.conf.admin_tool.port,
 202                 self.endpoint,
 203                 session_id))
 204
 205         headers = {
 206             'Content-Type': 'application/json',
 207             'Accept': 'application/json'}
 208         ret = requests.get(url, data=None, headers=headers)
 209         if ret.status_code != 200:
 210             raise Exception(ret.text)
 211         return ret.json()['state']
 212
 213     def wait_maintenance_complete(self, session_id):
 214         retries = 90
 215         state = None
 216         time.sleep(300)
 217         while (state not in ['MAINTENANCE_DONE', 'MAINTENANCE_FAILED'] and
 218                retries > 0):
 219             time.sleep(10)
 220             state = self.get_maintenance_state(session_id)
 221             retries = retries - 1
 222         self.remove_maintenance_session(session_id)
 223         self.log.info('maintenance %s ended with state %s' %
 224                       (session_id, state))
 225         if state == 'MAINTENANCE_FAILED':
 226             raise Exception('maintenance %s failed' % session_id)
 227         elif retries == 0:
 228             raise Exception('maintenance %s not completed within 20min' %
 229                             session_id)
 230
 231     def cleanup_maintenance(self):
 232         if self.conf.admin_tool.type == 'sample':
 233             self.admin_tool.stop()
 234         self.app_manager.stop()
 235         self.inspector.stop()
 236         self.log.info('stack delete start.......')
 237         self.stack.delete()