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