deploy/cloud/deployment.py

   1 ###############################################################################
   2 # Copyright (c) 2015 Ericsson AB and others.
   3 # szilard.cserey@ericsson.com
   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
  10 import time
  11 import re
  12 import json
  13
  14 from common import (
  15     N,
  16     exec_cmd,
  17     parse,
  18     err,
  19     log,
  20 )
  21
  22 SEARCH_TEXT = '(err)'
  23 LOG_FILE = '/var/log/puppet.log'
  24 GREP_LINES_OF_LEADING_CONTEXT = 100
  25 GREP_LINES_OF_TRAILING_CONTEXT = 100
  26 LIST_OF_CHAR_TO_BE_ESCAPED = ['[', ']', '"']
  27
  28
  29 class DeployNotStart(Exception):
  30     """Unable to start deployment"""
  31
  32
  33 class NodesGoOffline(Exception):
  34     """Nodes goes offline during deployment"""
  35
  36
  37 class Deployment(object):
  38
  39     def __init__(self, dea, yaml_config_dir, env_id, node_id_roles_dict,
  40                  no_health_check, deploy_timeout):
  41         self.dea = dea
  42         self.yaml_config_dir = yaml_config_dir
  43         self.env_id = env_id
  44         self.node_id_roles_dict = node_id_roles_dict
  45         self.no_health_check = no_health_check
  46         self.deploy_timeout = deploy_timeout
  47         self.pattern = re.compile(
  48             '\d\d\d\d-\d\d-\d\d\s\d\d:\d\d:\d\d')
  49
  50     def collect_error_logs(self):
  51         for node_id, roles_blade in self.node_id_roles_dict.iteritems():
  52             log_list = []
  53             cmd = ('ssh -q node-%s grep \'"%s"\' %s'
  54                    % (node_id, SEARCH_TEXT, LOG_FILE))
  55             results, _ = exec_cmd(cmd, False)
  56             for result in results.splitlines():
  57                 log_msg = ''
  58
  59                 sub_cmd = '"%s" %s' % (result, LOG_FILE)
  60                 for c in LIST_OF_CHAR_TO_BE_ESCAPED:
  61                     sub_cmd = sub_cmd.replace(c, '\%s' % c)
  62                 grep_cmd = ('grep -B%s %s'
  63                             % (GREP_LINES_OF_LEADING_CONTEXT, sub_cmd))
  64                 cmd = ('ssh -q node-%s "%s"' % (node_id, grep_cmd))
  65
  66                 details, _ = exec_cmd(cmd, False)
  67                 details_list = details.splitlines()
  68
  69                 found_prev_log = False
  70                 for i in range(len(details_list) - 2, -1, -1):
  71                     if self.pattern.match(details_list[i]):
  72                         found_prev_log = True
  73                         break
  74                 if found_prev_log:
  75                     log_msg += '\n'.join(details_list[i:-1]) + '\n'
  76
  77                 grep_cmd = ('grep -A%s %s'
  78                             % (GREP_LINES_OF_TRAILING_CONTEXT, sub_cmd))
  79                 cmd = ('ssh -q node-%s "%s"' % (node_id, grep_cmd))
  80
  81                 details, _ = exec_cmd(cmd, False)
  82                 details_list = details.splitlines()
  83
  84                 found_next_log = False
  85                 for i in range(1, len(details_list)):
  86                     if self.pattern.match(details_list[i]):
  87                         found_next_log = True
  88                         break
  89                 if found_next_log:
  90                     log_msg += '\n'.join(details_list[:i])
  91                 else:
  92                     log_msg += details
  93
  94                 if log_msg:
  95                     log_list.append(log_msg)
  96
  97             if log_list:
  98                 role = ('controller' if 'controller' in roles_blade[0]
  99                         else 'compute host')
 100                 log('_' * 40 + 'Errors in node-%s %s' % (node_id, role)
 101                     + '_' * 40)
 102                 for log_msg in log_list:
 103                     print(log_msg + '\n')
 104
 105     def run_deploy(self):
 106         SLEEP_TIME = 60
 107         abort_after = 60 * int(self.deploy_timeout)
 108         start = time.time()
 109
 110         log('Starting deployment of environment %s' % self.env_id)
 111         deploy_id = None
 112         ready = False
 113         timeout = False
 114
 115         attempts = 0
 116         while attempts < 3:
 117             try:
 118                 if time.time() > start + abort_after:
 119                     timeout = True
 120                     break
 121                 if not deploy_id:
 122                     deploy_id = self._start_deploy_task()
 123                 sts, prg, msg = self._deployment_status(deploy_id)
 124                 if sts == 'error':
 125                     log('Error during deployment: {}'.format(msg))
 126                     break
 127                 if sts == 'running':
 128                     log('Environment deployment progress: {}%'.format(prg))
 129                 elif sts == 'ready':
 130                     ready = True
 131                     break
 132                 time.sleep(SLEEP_TIME)
 133             except (DeployNotStart, NodesGoOffline) as e:
 134                 log(e)
 135                 attempts += 1
 136                 deploy_id = None
 137                 time.sleep(SLEEP_TIME * attempts)
 138
 139         if timeout:
 140             err('Deployment timed out, environment %s is not operational, '
 141                 'snapshot will not be performed'
 142                 % self.env_id)
 143         if ready:
 144             log('Environment %s successfully deployed'
 145                 % self.env_id)
 146         else:
 147             self.collect_error_logs()
 148             err('Deployment failed, environment %s is not operational'
 149                 % self.env_id, self.collect_logs)
 150
 151     def _start_deploy_task(self):
 152         out, _ = exec_cmd('fuel2 env deploy {}'.format(self.env_id), False)
 153         id = self._deployment_task_id(out)
 154         return id
 155
 156     def _deployment_task_id(self, response):
 157         response = str(response)
 158         if response.startswith('Deployment task with id'):
 159             for s in response.split():
 160                 if s.isdigit():
 161                     return int(s)
 162         raise DeployNotStart('Unable to start deployment: {}'.format(response))
 163
 164     def _deployment_status(self, id):
 165         task = self._task_fields(id)
 166         if task['status'] == 'error':
 167             if task['message'].endswith(
 168                     'offline. Remove them from environment and try again.'):
 169                 raise NodesGoOffline(task['message'])
 170         return task['status'], task['progress'], task['message']
 171
 172     def _task_fields(self, id):
 173         try:
 174             out, _ = exec_cmd('fuel2 task show {} -f json'.format(id), False)
 175             task_info = json.loads(out)
 176             properties = {}
 177             # for 9.0 this can be list of dicts or dict
 178             # see https://bugs.launchpad.net/fuel/+bug/1625518
 179             if isinstance(task_info, list):
 180                 for d in task_info:
 181                         properties.update({d['Field']: d['Value']})
 182             else:
 183                 return task_info
 184             return properties
 185         except ValueError as e:
 186             err('Unable to fetch task info: {}'.format(e))
 187
 188     def collect_logs(self):
 189         log('Cleaning out any previous deployment logs')
 190         exec_cmd('rm -f /var/log/remote/fuel-snapshot-*', False)
 191         exec_cmd('rm -f /root/deploy-*', False)
 192         log('Generating Fuel deploy snap-shot')
 193         if exec_cmd('fuel snapshot < /dev/null &> snapshot.log', False)[1] <> 0:
 194             log('Could not create a Fuel snapshot')
 195         else:
 196             exec_cmd('mv /root/fuel-snapshot* /var/log/remote/', False)
 197
 198         log('Collecting all Fuel Snapshot & deploy log files')
 199         r, _ = exec_cmd('tar -czhf /root/deploy-%s.log.tar.gz /var/log/remote' % time.strftime("%Y%m%d-%H%M%S"), False)
 200         log(r)
 201
 202     def verify_node_status(self):
 203         node_list = parse(exec_cmd('fuel --env %s node' % self.env_id))
 204         failed_nodes = []
 205         for node in node_list:
 206             if node[N['status']] != 'ready':
 207                 failed_nodes.append((node[N['id']], node[N['status']]))
 208
 209         if failed_nodes:
 210             summary = ''
 211             for node, status in failed_nodes:
 212                 summary += '[node %s, status %s]\n' % (node, status)
 213             err('Deployment failed: %s' % summary, self.collect_logs)
 214
 215     def health_check(self):
 216         log('Now running sanity and smoke health checks')
 217         r = exec_cmd('fuel health --env %s --check sanity,smoke --force' % self.env_id)
 218         log(r)
 219         if 'failure' in r:
 220             err('Healthcheck failed!', self.collect_logs)
 221
 222     def deploy(self):
 223         self.run_deploy()
 224         self.verify_node_status()
 225         if not self.no_health_check:
 226             self.health_check()
 227         self.collect_logs()