def start(self):
"""
Start KazooClient and add connection listener.
"""
try:
self.kazoo = KazooClient(hosts=get_zk_conn_string(env=self._env),
timeout=60.0)
self.kazoo.add_listener(self._zk_listener)
self.kazoo.start()
logging.info("ZooKeeper client started against cluster <{0}>"
.format(get_zk_conn_string(env=self._env)))
except Exception as e:
logging.error(e)
def __init__(self, *args, **kwargs):
signal.signal(signal.SIGINT, self._handle_sig_event)
signal.signal(signal.SIGTERM, self._handle_sig_event)
self._prev_connection_state = None
self._zoo_keeper = KazooClient(hosts=get_zk_conn_string(kwargs.get('environment')),
timeout=60.0)
self._zoo_keeper.start()
self._configuration = Configuration(self._zoo_keeper, **kwargs)
self._bootstrap_zookeeper_paths()
self._task_server = TaskServer(self._configuration, self._zoo_keeper)
self._data_store = DataStore(self._configuration, self._zoo_keeper,
self._task_server)
self._web_server = WebServer(self._configuration, self._data_store,
self._task_server, self._zoo_keeper)
def __init__(self, port):
"""
Read config and spawn child processes.
:type port: int
"""
self._log = logging.getLogger('sent.daemon')
self._log.info('Creating Sentinel')
self._port = port
self.children = dict()
self._settings = None
self._system = get_system()
self._hostname = socket.getfqdn()
self._prev_state = None
self.listener_lock = Lock()
self.version = get_version()
self.task_client = None
self.zkclient = KazooClient(hosts=get_zk_conn_string(),
timeout=60.0,
handler=SequentialThreadingHandler(),
logger=logging.getLogger('kazoo.daemon'))
self.zkclient.add_listener(self._zk_listener)
# this will run self._reset_after_connection_loss
self.zkclient.start()
while not self._settings:
self._log.info('Waiting for settings.')
time.sleep(1)
self._tmp_dir = os.path.join(self._settings.get('zookeeper').get('temp_directory', '/'), 'ruok')
self.task_client = ZKTaskClient(self.children,
self.zkclient,
self._settings.get('zookeeper', {}).get('task'))
self._rest_server = tornado.httpserver.HTTPServer(RestServer(self.children,
self.version,
self._tmp_dir,
self._hostname,
self.zkclient))
signal.signal(signal.SIGINT, self._handle_sigint)
signal.signal(signal.SIGTERM, self._handle_sigint)
self._log.info('Created Sentinel')
def __init__(self, *args, **kwargs):
signal.signal(signal.SIGINT, self._handle_sig_event)
signal.signal(signal.SIGTERM, self._handle_sig_event)
self._prev_connection_state = None
self._zoo_keeper = KazooClient(hosts=get_zk_conn_string(
kwargs.get('environment')),
timeout=60.0)
self._zoo_keeper.start()
self._configuration = Configuration(self._zoo_keeper, **kwargs)
self._bootstrap_zookeeper_paths()
self._task_server = TaskServer(self._configuration, self._zoo_keeper)
self._data_store = DataStore(self._configuration, self._zoo_keeper,
self._task_server)
self._web_server = WebServer(self._configuration, self._data_store,
self._task_server, self._zoo_keeper)
def __init__(self, port):
"""
Read config and spawn child processes.
:type port: int
"""
self._log = logging.getLogger('sent.daemon')
self._log.info('Creating Sentinel')
self._port = port
self.children = dict()
self._settings = None
self._system = get_system()
self._hostname = socket.getfqdn()
self._prev_state = None
self.listener_lock = Lock()
self.version = get_version()
self.task_client = None
self.zkclient = KazooClient(hosts=get_zk_conn_string(),
timeout=60.0,
handler=SequentialThreadingHandler(),
logger=logging.getLogger('kazoo.daemon'))
self.zkclient.add_listener(self._zk_listener)
# this will run self._reset_after_connection_loss
self.zkclient.start()
while not self._settings:
self._log.info('Waiting for settings.')
time.sleep(1)
self._tmp_dir = os.path.join(
self._settings.get('zookeeper').get('temp_directory', '/'), 'ruok')
self.task_client = ZKTaskClient(
self.children, self.zkclient,
self._settings.get('zookeeper', {}).get('task'))
self._rest_server = tornado.httpserver.HTTPServer(
RestServer(self.children, self.version, self._tmp_dir,
self._hostname, self.zkclient))
signal.signal(signal.SIGINT, self._handle_sigint)
signal.signal(signal.SIGTERM, self._handle_sigint)
self._log.info('Created Sentinel')
def __init__(self, temp_path, timeout,
parent='None', acquire_lock=None, app_state=None):
"""
:type temp_path: str
:type timeout: int
:type parent: str
:type acquire_lock: zoom.agent.entities.thread_safe_object.ThreadSafeObject or None
:type app_state: zoom.agent.entities.thread_safe_object.ThreadSafeObject or None
"""
self._path = temp_path
self._timeout = timeout
self._parent = parent
self._thread = None
self._prev_state = None
self._zk = KazooClient(hosts=get_zk_conn_string(),
timeout=60.0)
self._zk.add_listener(self._zk_listener)
self._log = logging.getLogger('sent.{0}.sl'.format(parent))
self._counter = 0
self._acquire_lock = acquire_lock
self._app_state = app_state
def __init__(self, zookeeper, **kwargs):
"""
:type zookeeper: kazoo.client.KazooClient
"""
self._zookeeper = zookeeper
self._settings = kwargs
try:
data, stat = self._zookeeper.get(ZOOM_CONFIG)
config = json.loads(data)
# create 'logs' directory if it does not exist
if not os.path.exists("logs"):
os.makedirs("logs")
# initialize logging
logging_config = config.get('logging')
logging.config.dictConfig(logging_config)
# get system type
running_os = self._get_system()
self._host = socket.gethostname()
# web_server
web_server_settings = config.get('web_server')
self._port = self._get_setting('port',
web_server_settings.get('port'))
self._is_debug = web_server_settings.get('debug')
self._application_path = os.getcwd()
self._client_path = zk_path_join(
(os.path.normpath(os.getcwd() + os.sep + os.pardir)), 'client')
self._doc_path = zk_path_join(
(os.path.normpath(os.getcwd() + os.sep + os.pardir)), "doc")
self._html_path = zk_path_join(self._client_path, "views")
self._images_path = zk_path_join(self._client_path, "images")
self._pid = os.getpid()
self._environment = self._get_setting(
'environment', os.environ.get('EnvironmentToUse', 'Staging'))
# zookeeper
zookeeper_settings = config.get('zookeeper')
self._zookeeper_paths = zookeeper_settings
self._agent_configuration_path = zookeeper_settings.get(
'agent_configuration_path')
self._agent_state_path = zookeeper_settings.get('agent_state_path')
self._task_path = zookeeper_settings.get('task_path')
self._application_state_path = zookeeper_settings.get(
'application_state_path')
self._global_mode_path = zookeeper_settings.get('global_mode_path')
self._pillar_path = zookeeper_settings.get('pillar_path')
self._alert_path = zookeeper_settings.get('alert_path')
self._override_node = zookeeper_settings.get(
'override_node', '/spot/software/config/override')
self._temp_directory = zookeeper_settings.get('temp_directory')
self._zookeeper_host = get_zk_conn_string(self._environment)
# pagerduty
pagerduty_settings = config.get('pagerduty')
self._pagerduty_default_svc_key = pagerduty_settings.get(
'pagerduty_default_svc_key')
self._pagerduty_api_token = pagerduty_settings.get(
'pagerduty_api_token')
self._pagerduty_subdomain = pagerduty_settings.get(
'pagerduty_subdomain')
self._pagerduty_enabled_environments = pagerduty_settings.get(
'pagerduty_enabled_environments')
self._pagerduty_alert_footer = pagerduty_settings.get(
'pagerduty_footer', '')
# database
db_settings = config.get('database')
self._db_type = db_settings.get('db_type')
if running_os == PlatformType.WINDOWS:
self._sql_connection = db_settings.get(
'sql_connection_windows')
else:
self._sql_connection = db_settings.get('sql_connection')
# authentication
ad_settings = config.get('active_directory')
self._ldap_server = ad_settings.get('host')
self._ldap_port = ad_settings.get('port')
# environment specific
env_settings = config.get(self._environment.lower())
self._read_write_groups = env_settings.get('read_write_groups')
self._graphite_host = env_settings.get('graphite_host')
self._graphite_recheck = env_settings.get('graphite_recheck', '5m')
# chatops
chatops_settings = env_settings.get('chatops', {})
self._chatops_url = chatops_settings.get('url')
self._chatops_group = chatops_settings.get('group')
self._chatops_commands_to_chat = chatops_settings.get(
'commands_to_chat')
# message throttling
throttle_settings = config.get('message_throttle')
self._throttle_interval = throttle_settings.get('interval')
# salt
self._salt_settings = env_settings.get('saltREST')
except ValueError as e:
logging.error('Data at {0} is not valid JSON.'.format(ZOOM_CONFIG))
raise e
except NoNodeError as e:
logging.error('Config node missing: {}'.format(ZOOM_CONFIG))
raise e
except Exception as e:
logging.exception('An unhandled exception occurred.')
raise e
def __init__(self, zookeeper, **kwargs):
"""
:type zookeeper: :rtype: zoom.www.entities.zoo_keeper.ZooKeeper
"""
self._zookeeper = zookeeper
self._settings = kwargs
try:
data, stat = self._zookeeper.get(ZOOM_CONFIG)
config = json.loads(data)
# create 'logs' directory if it does not exist
if not os.path.exists("logs"):
os.makedirs("logs")
# initialize logging
logging_config = config.get('logging')
logging.config.dictConfig(logging_config)
# get system type
running_os = self._get_system()
self._host = socket.gethostname()
# web_server
web_server_settings = config.get('web_server')
self._port = self._get_setting('port', web_server_settings.get('port'))
self._is_debug = web_server_settings.get('debug')
self._application_path = os.getcwd()
self._client_path = zk_path_join((os.path.normpath(os.getcwd() + os.sep + os.pardir)), 'client')
self._doc_path = zk_path_join((os.path.normpath(os.getcwd() + os.sep + os.pardir)), "doc")
self._html_path = zk_path_join(self._client_path, "views")
self._images_path = zk_path_join(self._client_path, "images")
self._pid = os.getpid()
self._environment = self._get_setting('environment',
os.environ.get('EnvironmentToUse', 'Staging'))
# zookeeper
zookeeper_settings = config.get('zookeeper')
self._agent_configuration_path = zookeeper_settings.get('agent_configuration_path')
self._agent_state_path = zookeeper_settings.get('agent_state_path')
self._task_path = zookeeper_settings.get('task_path')
self._application_state_path = zookeeper_settings.get('application_state_path')
self._global_mode_path = zookeeper_settings.get('global_mode_path')
self._pillar_path = zookeeper_settings.get('pillar_path')
self._alert_path = zookeeper_settings.get('alert_path')
self._override_node = zookeeper_settings.get('override_node', '/spot/software/config/override')
self._zookeeper_host = get_zk_conn_string(self._environment)
# pagerduty
pagerduty_settings = config.get('pagerduty')
self._pagerduty_default_svc_key = pagerduty_settings.get('pagerduty_default_svc_key')
self._pagerduty_api_token = pagerduty_settings.get('pagerduty_api_token')
self._pagerduty_subdomain = pagerduty_settings.get('pagerduty_subdomain')
self._pagerduty_enabled_environments = pagerduty_settings.get('pagerduty_enabled_environments')
self._pagerduty_alert_footer = pagerduty_settings.get('pagerduty_footer', '')
# database
db_settings = config.get('database')
self._db_type = db_settings.get('db_type')
if running_os == PlatformType.WINDOWS:
self._sql_connection = db_settings.get('sql_connection_windows')
elif running_os == PlatformType.LINUX:
self._sql_connection = db_settings.get('sql_connection')
# authentication
ad_settings = config.get('active_directory')
self._ldap_server = ad_settings.get('host')
self._ldap_port = ad_settings.get('port')
# environment specific
env_settings = config.get(self._environment.lower())
self._read_write_groups = env_settings.get('read_write_groups')
self._graphite_host = env_settings.get('graphite_host')
self._graphite_recheck = env_settings.get('graphite_recheck', '5m')
# chatops
chatops_settings = env_settings.get('chatops', {})
self._chatops_url = chatops_settings.get('url')
self._chatops_group = chatops_settings.get('group')
self._chatops_commands_to_chat = chatops_settings.get('commands_to_chat')
# message throttling
throttle_settings = config.get('message_throttle')
self._throttle_interval = throttle_settings.get('interval')
# salt
self._salt_settings = env_settings.get('saltREST')
except ValueError as e:
logging.error('Data at {0} is not valid JSON.'.format(ZOOM_CONFIG))
raise e
except Exception as e:
logging.exception('An unhandled exception occurred.')
raise e
def __init__(self, config, settings, queue, system, application_type,
cancel_flag):
"""
:type config: dict (xml)
:type settings: dict
:type queue: zoom.agent.entities.unique_queue.UniqueQueue
:type system: zoom.common.types.PlatformType
:type application_type: zoom.common.types.ApplicationType
:type cancel_flag: zoom.agent.entities.thread_safe_object.ThreadSafeObject
"""
self.config = config
self._settings = settings
self.name = verify_attribute(self.config, 'id', none_allowed=False)
self._log = logging.getLogger('sent.{0}.app'.format(self.name))
# informational attributes
self._host = socket.getfqdn()
self._system = system
self._predicates = list()
self._running = True # used to manually stop the run loop
self._prev_state = None
self._actions = dict() # created in _reset_watches on zk connect
self._env = os.environ.get('EnvironmentToUse', 'Staging')
self._apptype = application_type
self._restart_on_crash = \
verify_attribute(self.config, 'restart_on_crash', none_allowed=True)
self._post_stop_sleep = verify_attribute(self.config, 'post_stop_sleep',
none_allowed=True, cast=int,
default=5)
# tool-like attributes
self.listener_lock = Lock()
self._action_queue = queue
self._mode = ApplicationMode(
ApplicationMode.MANUAL,
callback=self._update_agent_node_with_app_details)
self._state = ThreadSafeObject(
ApplicationState.OK,
callback=self._update_agent_node_with_app_details)
self._start_stop_time = '' # Default to empty string for comparison
self._login_user = '******' # Default to Zoom
self._user_set_in_react = False
self._run_check_mode = False
self._pd_svc_key = verify_attribute(config, 'pagerduty_service',
none_allowed=True)
restartmax = verify_attribute(config, 'restartmax', none_allowed=True,
cast=int, default=3)
self._rl = RestartLogic(
self.name,
restartmax,
count_callback=self._update_agent_node_with_app_details)
self._read_only = False
self._paths = self._init_paths(self.config, settings, application_type)
# clients
self.zkclient = KazooClient(
hosts=get_zk_conn_string(),
timeout=60.0,
handler=SequentialThreadingHandler(),
logger=logging.getLogger('kazoo.app.{0}'.format(self.name)))
self.zkclient.add_listener(self._zk_listener)
self._proc_client = self._init_proc_client(self.config,
application_type,
cancel_flag)
self._actions = self._init_actions(settings)
self._work_manager = self._init_work_manager(self._action_queue)
def __init__(self, config, settings, queue, system, application_type,
cancel_flag):
"""
:type config: dict (xml)
:type settings: dict
:type queue: zoom.agent.entities.unique_queue.UniqueQueue
:type system: zoom.common.types.PlatformType
:type application_type: zoom.common.types.ApplicationType
:type cancel_flag: zoom.agent.entities.thread_safe_object.ThreadSafeObject
"""
self.config = config
self._settings = settings
self.name = verify_attribute(self.config, 'id', none_allowed=False)
self._log = logging.getLogger('sent.{0}.app'.format(self.name))
# informational attributes
self._host = socket.getfqdn()
self._system = system
self._predicates = list()
self._running = True # used to manually stop the run loop
self._prev_state = None
self._actions = dict() # created in _reset_watches on zk connect
self._env = os.environ.get('EnvironmentToUse', 'Staging')
self._apptype = application_type
self._restart_on_crash = \
verify_attribute(self.config, 'restart_on_crash', none_allowed=True)
self._post_stop_sleep = verify_attribute(self.config,
'post_stop_sleep',
none_allowed=True,
cast=int,
default=5)
# tool-like attributes
self.listener_lock = Lock()
self._action_queue = queue
self._mode = ApplicationMode(
ApplicationMode.MANUAL,
callback=self._update_agent_node_with_app_details)
self._state = ThreadSafeObject(
ApplicationState.OK,
callback=self._update_agent_node_with_app_details)
self._start_stop_time = '' # Default to empty string for comparison
self._login_user = '******' # Default to Zoom
self._user_set_in_react = False
self._run_check_mode = False
self._pd_svc_key = verify_attribute(config,
'pagerduty_service',
none_allowed=True)
restartmax = verify_attribute(config,
'restartmax',
none_allowed=True,
cast=int,
default=3)
self._rl = RestartLogic(
self.name,
restartmax,
count_callback=self._update_agent_node_with_app_details)
self._read_only = False
self._paths = self._init_paths(self.config, settings, application_type)
# clients
self.zkclient = KazooClient(hosts=get_zk_conn_string(),
timeout=60.0,
handler=SequentialThreadingHandler(),
logger=logging.getLogger(
'kazoo.app.{0}'.format(self.name)))
self.zkclient.add_listener(self._zk_listener)
self._proc_client = self._init_proc_client(self.config,
application_type,
cancel_flag)
self._actions = self._init_actions(settings)
self._work_manager = self._init_work_manager(self._action_queue)
