ansible-later/testenv/lib/python2.7/site-packages/ansible/plugins/action/reboot.py

412 lines
18 KiB
Python
Raw Normal View History

2019-04-23 11:04:27 +00:00
# Copyright: (c) 2016-2018, Matt Davis <mdavis@ansible.com>
# Copyright: (c) 2018, Sam Doran <sdoran@redhat.com>
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt)
from __future__ import (absolute_import, division, print_function)
__metaclass__ = type
import random
import time
from datetime import datetime, timedelta
from ansible.errors import AnsibleError, AnsibleConnectionFailure
from ansible.module_utils._text import to_native, to_text
from ansible.plugins.action import ActionBase
from ansible.utils.display import Display
display = Display()
try:
from __main__ import display
except ImportError:
from ansible.utils.display import Display
display = Display()
class TimedOutException(Exception):
pass
class ActionModule(ActionBase):
TRANSFERS_FILES = False
_VALID_ARGS = frozenset(('connect_timeout', 'msg', 'post_reboot_delay', 'pre_reboot_delay', 'test_command', 'reboot_timeout'))
DEFAULT_REBOOT_TIMEOUT = 600
DEFAULT_CONNECT_TIMEOUT = None
DEFAULT_PRE_REBOOT_DELAY = 0
DEFAULT_POST_REBOOT_DELAY = 0
DEFAULT_TEST_COMMAND = 'whoami'
DEFAULT_BOOT_TIME_COMMAND = 'cat /proc/sys/kernel/random/boot_id'
DEFAULT_REBOOT_MESSAGE = 'Reboot initiated by Ansible'
DEFAULT_SHUTDOWN_COMMAND = 'shutdown'
DEFAULT_SHUTDOWN_COMMAND_ARGS = '-r {delay_min} "{message}"'
DEFAULT_SUDOABLE = True
DEPRECATED_ARGS = {}
BOOT_TIME_COMMANDS = {
'freebsd': '/sbin/sysctl kern.boottime',
'openbsd': '/sbin/sysctl kern.boottime',
'macosx': 'who -b',
'solaris': 'who -b',
'sunos': 'who -b',
'vmkernel': 'grep booted /var/log/vmksummary.log | tail -n 1',
'aix': 'who -b',
}
SHUTDOWN_COMMANDS = {
'alpine': 'reboot',
'vmkernel': 'reboot',
}
SHUTDOWN_COMMAND_ARGS = {
'alpine': '',
'freebsd': '-r +{delay_sec}s "{message}"',
'linux': DEFAULT_SHUTDOWN_COMMAND_ARGS,
'macosx': '-r +{delay_min} "{message}"',
'openbsd': '-r +{delay_min} "{message}"',
'solaris': '-y -g {delay_sec} -i 6 "{message}"',
'sunos': '-y -g {delay_sec} -i 6 "{message}"',
'vmkernel': '-d {delay_sec}',
'aix': '-Fr',
}
TEST_COMMANDS = {
'solaris': 'who',
'vmkernel': 'who',
}
def __init__(self, *args, **kwargs):
super(ActionModule, self).__init__(*args, **kwargs)
@property
def pre_reboot_delay(self):
return self._check_delay('pre_reboot_delay', self.DEFAULT_PRE_REBOOT_DELAY)
@property
def post_reboot_delay(self):
return self._check_delay('post_reboot_delay', self.DEFAULT_POST_REBOOT_DELAY)
def _check_delay(self, key, default):
"""Ensure that the value is positive or zero"""
value = int(self._task.args.get(key, self._task.args.get(key + '_sec', default)))
if value < 0:
value = 0
return value
def _get_value_from_facts(self, variable_name, distribution, default_value):
"""Get dist+version specific args first, then distribution, then family, lastly use default"""
attr = getattr(self, variable_name)
value = attr.get(
distribution['name'] + distribution['version'],
attr.get(
distribution['name'],
attr.get(
distribution['family'],
getattr(self, default_value))))
return value
def get_shutdown_command_args(self, distribution):
args = self._get_value_from_facts('SHUTDOWN_COMMAND_ARGS', distribution, 'DEFAULT_SHUTDOWN_COMMAND_ARGS')
# Convert seconds to minutes. If less that 60, set it to 0.
delay_min = self.pre_reboot_delay // 60
reboot_message = self._task.args.get('msg', self.DEFAULT_REBOOT_MESSAGE)
return args.format(delay_sec=self.pre_reboot_delay, delay_min=delay_min, message=reboot_message)
def get_distribution(self, task_vars):
distribution = {}
display.debug('{action}: running setup module to get distribution'.format(action=self._task.action))
module_output = self._execute_module(
task_vars=task_vars,
module_name='setup',
module_args={'gather_subset': 'min'})
try:
if module_output.get('failed', False):
raise AnsibleError('Failed to determine system distribution. {0}, {1}'.format(
to_native(module_output['module_stdout']).strip(),
to_native(module_output['module_stderr']).strip()))
distribution['name'] = module_output['ansible_facts']['ansible_distribution'].lower()
distribution['version'] = to_text(module_output['ansible_facts']['ansible_distribution_version'].split('.')[0])
distribution['family'] = to_text(module_output['ansible_facts']['ansible_os_family'].lower())
display.debug("{action}: distribution: {dist}".format(action=self._task.action, dist=distribution))
return distribution
except KeyError as ke:
raise AnsibleError('Failed to get distribution information. Missing "{0}" in output.'.format(ke.args[0]))
def get_shutdown_command(self, task_vars, distribution):
shutdown_bin = self._get_value_from_facts('SHUTDOWN_COMMANDS', distribution, 'DEFAULT_SHUTDOWN_COMMAND')
display.debug('{action}: running find module to get path for "{command}"'.format(action=self._task.action, command=shutdown_bin))
find_result = self._execute_module(
task_vars=task_vars,
module_name='find',
module_args={
'paths': ['/sbin', '/usr/sbin', '/usr/local/sbin'],
'patterns': [shutdown_bin],
'file_type': 'any'
}
)
full_path = [x['path'] for x in find_result['files']]
if not full_path:
raise AnsibleError('Unable to find command "{0}" in system paths.'.format(shutdown_bin))
self._shutdown_command = full_path[0]
return self._shutdown_command
def deprecated_args(self):
for arg, version in self.DEPRECATED_ARGS.items():
if self._task.args.get(arg) is not None:
display.warning("Since Ansible {version}, {arg} is no longer a valid option for {action}".format(
version=version,
arg=arg,
action=self._task.action))
def get_system_boot_time(self, distribution):
boot_time_command = self._get_value_from_facts('BOOT_TIME_COMMANDS', distribution, 'DEFAULT_BOOT_TIME_COMMAND')
display.debug("{action}: getting boot time with command: '{command}'".format(action=self._task.action, command=boot_time_command))
command_result = self._low_level_execute_command(boot_time_command, sudoable=self.DEFAULT_SUDOABLE)
if command_result['rc'] != 0:
stdout = command_result['stdout']
stderr = command_result['stderr']
raise AnsibleError("{action}: failed to get host boot time info, rc: {rc}, stdout: {out}, stderr: {err}".format(
action=self._task.action,
rc=command_result['rc'],
out=to_native(stdout),
err=to_native(stderr)))
display.debug("{action}: last boot time: {boot}".format(action=self._task.action, boot=command_result['stdout'].strip()))
return command_result['stdout'].strip()
def check_boot_time(self, distribution, previous_boot_time):
display.vvv("{action}: attempting to get system boot time".format(action=self._task.action))
connect_timeout = self._task.args.get('connect_timeout', self._task.args.get('connect_timeout_sec', self.DEFAULT_CONNECT_TIMEOUT))
# override connection timeout from defaults to custom value
if connect_timeout:
try:
display.debug("{action}: setting connect_timeout to {value}".format(action=self._task.action, value=connect_timeout))
self._connection.set_option("connection_timeout", connect_timeout)
self._connection.reset()
except AttributeError:
display.warning("Connection plugin does not allow the connection timeout to be overridden")
# try and get boot time
try:
current_boot_time = self.get_system_boot_time(distribution)
except Exception as e:
raise e
# FreeBSD returns an empty string immediately before reboot so adding a length
# check to prevent prematurely assuming system has rebooted
if len(current_boot_time) == 0 or current_boot_time == previous_boot_time:
raise ValueError("boot time has not changed")
def run_test_command(self, distribution, **kwargs):
test_command = self._task.args.get('test_command', self._get_value_from_facts('TEST_COMMANDS', distribution, 'DEFAULT_TEST_COMMAND'))
display.vvv("{action}: attempting post-reboot test command".format(action=self._task.action))
display.debug("{action}: attempting post-reboot test command '{command}'".format(action=self._task.action, command=test_command))
try:
command_result = self._low_level_execute_command(test_command, sudoable=self.DEFAULT_SUDOABLE)
except Exception:
# may need to reset the connection in case another reboot occurred
# which has invalidated our connection
try:
self._connection.reset()
except AttributeError:
pass
raise
if command_result['rc'] != 0:
msg = 'Test command failed: {err} {out}'.format(
err=to_native(command_result['stderr']),
out=to_native(command_result['stdout']))
raise RuntimeError(msg)
display.vvv("{action}: system sucessfully rebooted".format(action=self._task.action))
def do_until_success_or_timeout(self, action, reboot_timeout, action_desc, distribution, action_kwargs=None):
max_end_time = datetime.utcnow() + timedelta(seconds=reboot_timeout)
if action_kwargs is None:
action_kwargs = {}
fail_count = 0
max_fail_sleep = 12
while datetime.utcnow() < max_end_time:
try:
action(distribution=distribution, **action_kwargs)
if action_desc:
display.debug('{action}: {desc} success'.format(action=self._task.action, desc=action_desc))
return
except Exception as e:
if isinstance(e, AnsibleConnectionFailure):
try:
self._connection.reset()
except AnsibleConnectionFailure:
pass
# Use exponential backoff with a max timout, plus a little bit of randomness
random_int = random.randint(0, 1000) / 1000
fail_sleep = 2 ** fail_count + random_int
if fail_sleep > max_fail_sleep:
fail_sleep = max_fail_sleep + random_int
if action_desc:
try:
error = to_text(e).splitlines()[-1]
except IndexError as e:
error = to_text(e)
display.debug("{action}: {desc} fail '{err}', retrying in {sleep:.4} seconds...".format(
action=self._task.action,
desc=action_desc,
err=error,
sleep=fail_sleep))
fail_count += 1
time.sleep(fail_sleep)
raise TimedOutException('Timed out waiting for {desc} (timeout={timeout})'.format(desc=action_desc, timeout=reboot_timeout))
def perform_reboot(self, task_vars, distribution):
result = {}
reboot_result = {}
shutdown_command = self.get_shutdown_command(task_vars, distribution)
shutdown_command_args = self.get_shutdown_command_args(distribution)
reboot_command = '{0} {1}'.format(shutdown_command, shutdown_command_args)
try:
display.vvv("{action}: rebooting server...".format(action=self._task.action))
display.debug("{action}: rebooting server with command '{command}'".format(action=self._task.action, command=reboot_command))
reboot_result = self._low_level_execute_command(reboot_command, sudoable=self.DEFAULT_SUDOABLE)
except AnsibleConnectionFailure as e:
# If the connection is closed too quickly due to the system being shutdown, carry on
display.debug('{action}: AnsibleConnectionFailure caught and handled: {error}'.format(action=self._task.action, error=to_native(e)))
reboot_result['rc'] = 0
result['start'] = datetime.utcnow()
if reboot_result['rc'] != 0:
result['failed'] = True
result['rebooted'] = False
result['msg'] = "Reboot command failed. Error was {stdout}, {stderr}".format(
stdout=to_native(reboot_result['stdout'].strip()),
stderr=to_native(reboot_result['stderr'].strip()))
return result
result['failed'] = False
return result
def validate_reboot(self, distribution, original_connection_timeout=None, action_kwargs=None):
display.vvv('{action}: validating reboot'.format(action=self._task.action))
result = {}
try:
# keep on checking system boot_time with short connection responses
reboot_timeout = int(self._task.args.get('reboot_timeout', self._task.args.get('reboot_timeout_sec', self.DEFAULT_REBOOT_TIMEOUT)))
self.do_until_success_or_timeout(
action=self.check_boot_time,
action_desc="last boot time check",
reboot_timeout=reboot_timeout,
distribution=distribution,
action_kwargs=action_kwargs)
# Get the connect_timeout set on the connection to compare to the original
try:
connect_timeout = self._connection.get_option('connection_timeout')
except AnsibleError:
pass
else:
if original_connection_timeout != connect_timeout:
try:
display.debug("{action}: setting connect_timeout back to original value of {value}".format(
action=self._task.action,
value=original_connection_timeout))
self._connection.set_option("connection_timeout", original_connection_timeout)
self._connection.reset()
except (AnsibleError, AttributeError) as e:
# reset the connection to clear the custom connection timeout
display.debug("{action}: failed to reset connection_timeout back to default: {error}".format(action=self._task.action,
error=to_text(e)))
# finally run test command to ensure everything is working
# FUTURE: add a stability check (system must remain up for N seconds) to deal with self-multi-reboot updates
self.do_until_success_or_timeout(
action=self.run_test_command,
action_desc="post-reboot test command",
reboot_timeout=reboot_timeout,
distribution=distribution,
action_kwargs=action_kwargs)
result['rebooted'] = True
result['changed'] = True
except TimedOutException as toex:
result['failed'] = True
result['rebooted'] = True
result['msg'] = to_text(toex)
return result
return result
def run(self, tmp=None, task_vars=None):
self._supports_check_mode = True
self._supports_async = True
# If running with local connection, fail so we don't reboot ourself
if self._connection.transport == 'local':
msg = 'Running {0} with local connection would reboot the control node.'.format(self._task.action)
return {'changed': False, 'elapsed': 0, 'rebooted': False, 'failed': True, 'msg': msg}
if self._play_context.check_mode:
return {'changed': True, 'elapsed': 0, 'rebooted': True}
if task_vars is None:
task_vars = {}
self.deprecated_args()
result = super(ActionModule, self).run(tmp, task_vars)
if result.get('skipped', False) or result.get('failed', False):
return result
distribution = self.get_distribution(task_vars)
# Get current boot time
try:
previous_boot_time = self.get_system_boot_time(distribution)
except Exception as e:
result['failed'] = True
result['reboot'] = False
result['msg'] = to_text(e)
return result
# Get the original connection_timeout option var so it can be reset after
original_connection_timeout = None
try:
original_connection_timeout = self._connection.get_option('connection_timeout')
display.debug("{action}: saving original connect_timeout of {timeout}".format(action=self._task.action, timeout=original_connection_timeout))
except AnsibleError:
display.debug("{action}: connect_timeout connection option has not been set".format(action=self._task.action))
# Initiate reboot
reboot_result = self.perform_reboot(task_vars, distribution)
if reboot_result['failed']:
result = reboot_result
elapsed = datetime.utcnow() - reboot_result['start']
result['elapsed'] = elapsed.seconds
return result
if self.post_reboot_delay != 0:
display.debug("{action}: waiting an additional {delay} seconds".format(action=self._task.action, delay=self.post_reboot_delay))
display.vvv("{action}: waiting an additional {delay} seconds".format(action=self._task.action, delay=self.post_reboot_delay))
time.sleep(self.post_reboot_delay)
# Make sure reboot was successful
result = self.validate_reboot(distribution, original_connection_timeout, action_kwargs={'previous_boot_time': previous_boot_time})
elapsed = datetime.utcnow() - reboot_result['start']
result['elapsed'] = elapsed.seconds
return result