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)
  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                 raise Exception('%d vcpus used on %s'
  72                                 % (vcpus_used, hostname))
  73             if prev_vcpus != 0 and prev_vcpus != vcpus:
  74                 raise Exception('%d vcpus on %s does not match to'
  75                                 '%d on %s'
  76                                 % (vcpus, hostname,
  77                                    prev_vcpus, prev_hostname))
  78             prev_vcpus = vcpus
  79             prev_hostname = hostname
  80
  81         # maintenance flavor made so that 2 instances take whole node
  82         flavor_vcpus = int(vcpus / 2)
  83         compute_nodes = len(hvisors)
  84         amount_actstdby_instances = 2
  85         amount_noredundancy_instances = 2 * compute_nodes - 2
  86         self.log.info('testing %d computes with %d vcpus each'
  87                       % (compute_nodes, vcpus))
  88         self.log.info('testing %d actstdby and %d noredundancy instances'
  89                       % (amount_actstdby_instances,
  90                          amount_noredundancy_instances))
  91         max_instances = (amount_actstdby_instances +
  92                          amount_noredundancy_instances)
  93         max_cores = compute_nodes * vcpus
  94
  95         user.update_quota(max_instances, max_cores)
  96
  97         test_dir = get_doctor_test_root_dir()
  98         template_file = '{0}/{1}'.format(test_dir, 'maintenance_hot_tpl.yaml')
  99         files, template = self.stack.get_hot_tpl(template_file)
 100
 101         ext_net = self.get_external_network()
 102
 103         parameters = {'ext_net': ext_net,
 104                       'flavor_vcpus': flavor_vcpus,
 105                       'maint_image': self.conf.image_name,
 106                       'nonha_intances': amount_noredundancy_instances,
 107                       'ha_intances': amount_actstdby_instances}
 108
 109         self.log.info('creating maintenance stack.......')
 110         self.log.info('parameters: %s' % parameters)
 111
 112         self.stack.create('doctor_test_maintenance',
 113                           template,
 114                           parameters=parameters,
 115                           files=files)
 116
 117         if self.conf.admin_tool.type == 'sample':
 118             self.admin_tool.start()
 119         else:
 120             # TBD Now we expect Fenix is running in self.conf.admin_tool.port
 121             pass
 122         self.app_manager.start()
 123         self.inspector.start()
 124
 125     def start_maintenance(self):
 126         self.log.info('start maintenance.......')
 127         hvisors = self.nova.hypervisors.list(detailed=True)
 128         maintenance_hosts = list()
 129         for hvisor in hvisors:
 130             hostname = hvisor.__getattr__('hypervisor_hostname')
 131             maintenance_hosts.append(hostname)
 132
 133         url = ('http://%s:%s/%s' %
 134                (self.conf.admin_tool.ip,
 135                 self.conf.admin_tool.port,
 136                 self.endpoint))
 137
 138         # let's start maintenance 20sec from now, so projects will have
 139         # time to ACK to it before that
 140         maintenance_at = (datetime.datetime.utcnow() +
 141                           datetime.timedelta(seconds=30)
 142                           ).strftime('%Y-%m-%d %H:%M:%S')
 143         data = {'hosts': maintenance_hosts,
 144                 'state': 'MAINTENANCE',
 145                 'maintenance_at': maintenance_at,
 146                 'metadata': {'openstack_version': 'Rocky'},
 147                 'workflow': 'default'}
 148         headers = {
 149             'Content-Type': 'application/json',
 150             'Accept': 'application/json'}
 151
 152         ret = requests.post(url, data=json.dumps(data), headers=headers)
 153         if ret.status_code != 200:
 154             raise Exception(ret.text)
 155         return ret.json()['session_id']
 156
 157     def remove_maintenance_session(self, session_id):
 158         self.log.info('remove maintenance session %s.......' % session_id)
 159         url = ('http://%s:%s/%s/%s' %
 160                (self.conf.admin_tool.ip,
 161                 self.conf.admin_tool.port,
 162                 self.endpoint,
 163                 session_id))
 164
 165         headers = {
 166             'Content-Type': 'application/json',
 167             'Accept': 'application/json'}
 168
 169         ret = requests.delete(url, data=None, headers=headers)
 170         if ret.status_code != 200:
 171             raise Exception(ret.text)
 172
 173     def get_maintenance_state(self, session_id):
 174         url = ('http://%s:%s/%s/%s' %
 175                (self.conf.admin_tool.ip,
 176                 self.conf.admin_tool.port,
 177                 self.endpoint,
 178                 session_id))
 179
 180         headers = {
 181             'Content-Type': 'application/json',
 182             'Accept': 'application/json'}
 183         ret = requests.get(url, data=None, headers=headers)
 184         if ret.status_code != 200:
 185             raise Exception(ret.text)
 186         return ret.json()['state']
 187
 188     def wait_maintenance_complete(self, session_id):
 189         retries = 90
 190         state = None
 191         time.sleep(300)
 192         while (state not in ['MAINTENANCE_DONE', 'MAINTENANCE_FAILED'] and
 193                retries > 0):
 194             time.sleep(10)
 195             state = self.get_maintenance_state(session_id)
 196             retries = retries - 1
 197         self.remove_maintenance_session(session_id)
 198         self.log.info('maintenance %s ended with state %s' %
 199                       (session_id, state))
 200         if state == 'MAINTENANCE_FAILED':
 201             raise Exception('maintenance %s failed' % session_id)
 202         elif retries == 0:
 203             raise Exception('maintenance %s not completed within 20min' %
 204                             session_id)
 205
 206     def cleanup_maintenance(self):
 207         if self.conf.admin_tool.type == 'sample':
 208             self.admin_tool.stop()
 209         self.app_manager.stop()
 210         self.inspector.stop()
 211         self.log.info('stack delete start.......')
 212         self.stack.delete()