def run_check(name, path=None):
"""
Test custom checks on Windows.
"""
# Read the config file
confd_path = path or os.path.join(get_confd_path(get_os()), "%s.yaml" % name)
try:
f = open(confd_path)
except IOError:
raise Exception("Unable to open configuration at %s" % confd_path)
config_str = f.read()
f.close()
# Run the check
check, instances = get_check(name, config_str)
if not instances:
raise Exception("YAML configuration returned no instances.")
for instance in instances:
check.check(instance)
if check.has_events():
print "Events:\n"
pprint(check.get_events(), indent=4)
print "Metrics:\n"
pprint(check.get_metrics(), indent=4)
def __init__(self, agentConfig, emitters, systemStats):
self.emit_duration = None
self.agentConfig = agentConfig
# system stats is generated by config.get_system_stats
self.agentConfig["system_stats"] = systemStats
# agent config is used during checks, system_stats can be accessed through the config
self.os = get_os()
self.plugins = None
self.emitters = emitters
self.metadata_interval = int(agentConfig.get("metadata_interval", 10 * 60))
self.metadata_start = time.time()
socket.setdefaulttimeout(15)
self.run_count = 0
self.continue_running = True
self.metadata_cache = None
self.initialized_checks_d = []
self.init_failed_checks_d = []
# Unix System Checks
self._unix_system_checks = {
"disk": u.Disk(log),
"io": u.IO(log),
"load": u.Load(log),
"memory": u.Memory(log),
"processes": u.Processes(log),
"cpu": u.Cpu(log),
}
# Win32 System `Checks
self._win32_system_checks = {
"disk": w32.Disk(log),
"io": w32.IO(log),
"proc": w32.Processes(log),
"memory": w32.Memory(log),
"network": w32.Network(log),
"cpu": w32.Cpu(log),
}
# Old-style metric checks
self._ganglia = Ganglia(log)
self._dogstream = Dogstreams.init(log, self.agentConfig)
self._ddforwarder = DdForwarder(log, self.agentConfig)
# Agent Metrics
self._agent_metrics = CollectorMetrics(log)
self._metrics_checks = []
# Custom metric checks
for module_spec in [s.strip() for s in self.agentConfig.get("custom_checks", "").split(",")]:
if len(module_spec) == 0:
continue
try:
self._metrics_checks.append(modules.load(module_spec, "Check")(log))
log.info("Registered custom check %s" % module_spec)
log.warning(
"Old format custom checks are deprecated. They should be moved to the checks.d interface as old custom checks will be removed in a next version"
)
except Exception, e:
log.exception("Unable to load custom check module %s" % module_spec)
def load_check(name, config, agentConfig):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for _, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
init_config = config.get('init_config', {})
instances = config.get('instances')
agentConfig['checksd_hostname'] = get_hostname(agentConfig)
# init the check class
try:
return check_class(name, init_config=init_config, agentConfig=agentConfig, instances=instances)
except Exception as e:
raise Exception("Check is using old API, {0}".format(e))
def load_check(name, config, agentConfig, is_sdk=False):
if not is_sdk:
checksd_path = get_checksd_path(get_os())
# find (in checksd_path) and load the check module
fd, filename, desc = imp.find_module(name, [checksd_path])
check_module = imp.load_module(name, fd, filename, desc)
else:
check_module = __import__("check")
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for _, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
init_config = config.get('init_config', {})
instances = config.get('instances')
agentConfig['checksd_hostname'] = get_hostname(agentConfig)
# init the check class
try:
return check_class(name, init_config=init_config, agentConfig=agentConfig, instances=instances)
except TypeError as e:
raise Exception("Check is using old API, {0}".format(e))
except Exception:
raise
def get_check(name, config_str):
from checks import AgentCheck
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
max_hierarchy_len = 0
for name, clsmember in classes:
if AgentCheck in clsmember.__bases__:
check_class = clsmember
break
class_hierarchy = inspect.getmro(clsmember)
if AgentCheck in class_hierarchy:
if len(class_hierarchy) > max_hierarchy_len:
max_hierarchy_len = len(class_hierarchy)
check_class = clsmember
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
agentConfig = {
'version': '0.1',
'api_key': 'tota'
}
return check_class.from_yaml(yaml_text=config_str, check_name=name,
agentConfig=agentConfig)
def run_check(name, path=None):
from tests.common import get_check
# Read the config file
confd_path = path or os.path.join(get_confd_path(get_os()), '%s.yaml' % name)
try:
f = open(confd_path)
except IOError:
raise Exception('Unable to open configuration at %s' % confd_path)
config_str = f.read()
f.close()
# Run the check
check, instances = get_check(name, config_str)
if not instances:
raise Exception('YAML configuration returned no instances.')
for instance in instances:
check.check(instance)
if check.has_events():
print "Events:\n"
pprint(check.get_events(), indent=4)
print "Metrics:\n"
pprint(check.get_metrics(), indent=4)
def get_config_path(cfg_path=None, os_name=None):
# Check if there's an override and if it exists
if cfg_path is not None and os.path.exists(cfg_path):
return cfg_path
# Check if there's a config stored in the current agent directory
try:
path = os.path.realpath(__file__)
path = os.path.dirname(path)
return _config_path(path)
except PathNotFound as e:
pass
if os_name is None:
os_name = get_os()
# Check for an OS-specific path, continue on not-found exceptions
bad_path = ''
try:
if os_name == 'windows':
return _windows_config_path()
elif os_name == 'mac':
return _mac_config_path()
else:
return _unix_config_path()
except PathNotFound as e:
if len(e.args) > 0:
bad_path = e.args[0]
# If all searches fail, exit the agent with an error
sys.stderr.write("Please supply a configuration file at %s or in the directory where "
"the Agent is currently deployed.\n" % bad_path)
sys.exit(3)
def load_check_directory(agentConfig, hostname):
''' Return the initialized checks from checks.d, and a mapping of checks that failed to
initialize. Only checks that have a configuration
file in conf.d will be returned. '''
from checks import AgentCheck, AGENT_METRICS_CHECK_NAME
initialized_checks = {}
init_failed_checks = {}
deprecated_checks = {}
agentConfig['checksd_hostname'] = hostname
deprecated_configs_enabled = [v for k,v in OLD_STYLE_PARAMETERS if len([l for l in agentConfig if l.startswith(k)]) > 0]
for deprecated_config in deprecated_configs_enabled:
msg = "Configuring %s in datadog.conf is not supported anymore. Please use conf.d" % deprecated_config
deprecated_checks[deprecated_config] = {'error': msg, 'traceback': None}
log.error(msg)
osname = get_os()
checks_paths = [glob.glob(os.path.join(agentConfig['additional_checksd'], '*.py'))]
try:
checksd_path = get_checksd_path(osname)
checks_paths.append(glob.glob(os.path.join(checksd_path, '*.py')))
except PathNotFound, e:
log.error(e.args[0])
sys.exit(3)
def load_check(agentConfig, hostname, checkname):
"""Same logic as load_check_directory except it loads one specific check"""
agentConfig['checksd_hostname'] = hostname
osname = get_os()
checks_places = get_checks_places(osname, agentConfig)
for config_path in _file_configs_paths(osname, agentConfig):
check_name = _conf_path_to_check_name(config_path)
if check_name == checkname:
conf_is_valid, check_config, invalid_check = _load_file_config(config_path, check_name, agentConfig)
if invalid_check and not conf_is_valid:
return invalid_check
# try to load the check and return the result
load_success, load_failure = load_check_from_places(check_config, check_name, checks_places, agentConfig)
return load_success.values()[0] or load_failure
# the check was not found, try with service discovery
for check_name, service_disco_check_config in _service_disco_configs(agentConfig).iteritems():
if check_name == checkname:
sd_init_config, sd_instances = service_disco_check_config
check_config = {'init_config': sd_init_config, 'instances': sd_instances}
# try to load the check and return the result
load_success, load_failure = load_check_from_places(check_config, check_name, checks_places, agentConfig)
return load_success.values()[0] or load_failure
return None
def load_check(name, config, agentConfig):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for name, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
init_config = config.get("init_config", None)
instances = config.get("instances")
# init the check class
try:
return check_class(name, init_config=init_config, agentConfig=agentConfig, instances=instances)
except:
# Backwards compatitiblity for old checks that don't support the
# instances argument.
c = check_class(name, init_config=init_config, agentConfig=agentConfig)
c.instances = instances
return c
def get_checksd_path(osname=None):
if not osname:
osname = get_os()
if osname == 'windows':
return _windows_checksd_path()
else:
return _unix_checksd_path()
def __init__(self, agentConfig, emitters, systemStats):
self.emit_duration = None
self.agentConfig = agentConfig
# system stats is generated by config.get_system_stats
self.agentConfig['system_stats'] = systemStats
# agent config is used during checks, system_stats can be accessed through the config
self.os = get_os()
self.plugins = None
self.emitters = emitters
self.metadata_interval = int(agentConfig.get('metadata_interval', 10 * 60))
self.metadata_start = time.time()
socket.setdefaulttimeout(15)
self.run_count = 0
self.continue_running = True
self.metadata_cache = None
self.checks_d = []
# Unix System Checks
self._unix_system_checks = {
'disk': u.Disk(log),
'io': u.IO(log),
'load': u.Load(log),
'memory': u.Memory(log),
'processes': u.Processes(log),
'cpu': u.Cpu(log)
}
# Win32 System `Checks
self._win32_system_checks = {
'disk': w32.Disk(log),
'io': w32.IO(log),
'proc': w32.Processes(log),
'memory': w32.Memory(log),
'network': w32.Network(log),
'cpu': w32.Cpu(log)
}
# Old-style metric checks
self._ganglia = Ganglia(log)
self._cassandra = Cassandra()
self._dogstream = Dogstreams.init(log, self.agentConfig)
self._ddforwarder = DdForwarder(log, self.agentConfig)
# Agent Metrics
self._agent_metrics = CollectorMetrics(log)
# Metric Checks
self._metrics_checks = [
Memcache(log),
]
# Custom metric checks
for module_spec in [s.strip() for s in self.agentConfig.get('custom_checks', '').split(',')]:
if len(module_spec) == 0: continue
try:
self._metrics_checks.append(modules.load(module_spec, 'Check')(log))
log.info("Registered custom check %s" % module_spec)
except Exception, e:
log.exception('Unable to load custom check module %s' % module_spec)
def init(config_path=None):
agentConfig = get_config(parse_args=False, cfg_path=config_path)
osname = get_os()
try:
confd_path = get_confd_path(osname)
except PathNotFound, e:
log.error("No conf.d folder found at '%s' or in the directory where"
"the Agent is currently deployed.\n" % e.args[0])
def initialize_logging(logger_name):
try:
logging_config = get_logging_config()
logging.basicConfig(
format=get_log_format(logger_name),
level=logging_config['log_level'] or logging.INFO,
)
log_file = logging_config.get('%s_log_file' % logger_name)
if log_file is not None and not logging_config['disable_file_logging']:
# make sure the log directory is writeable
# NOTE: the entire directory needs to be writable so that rotation works
if os.access(os.path.dirname(log_file), os.R_OK | os.W_OK):
file_handler = logging.handlers.RotatingFileHandler(log_file, maxBytes=LOGGING_MAX_BYTES, backupCount=1)
formatter = logging.Formatter(get_log_format(logger_name), get_log_date_format())
file_handler.setFormatter(formatter)
root_log = logging.getLogger()
root_log.addHandler(file_handler)
else:
sys.stderr.write("Log file is unwritable: '%s'\n" % log_file)
# set up syslog
if logging_config['log_to_syslog']:
try:
from logging.handlers import SysLogHandler
if logging_config['syslog_host'] is not None and logging_config['syslog_port'] is not None:
sys_log_addr = (logging_config['syslog_host'], logging_config['syslog_port'])
else:
sys_log_addr = "/dev/log"
# Special-case BSDs
if Platform.is_darwin():
sys_log_addr = "/var/run/syslog"
elif Platform.is_freebsd():
sys_log_addr = "/var/run/log"
handler = SysLogHandler(address=sys_log_addr, facility=SysLogHandler.LOG_DAEMON)
handler.setFormatter(logging.Formatter(get_syslog_format(logger_name), get_log_date_format()))
root_log = logging.getLogger()
root_log.addHandler(handler)
except Exception, e:
sys.stderr.write("Error setting up syslog: '%s'\n" % str(e))
traceback.print_exc()
# Setting up logging in the event viewer for windows
if get_os() == 'windows' and logging_config['log_to_event_viewer']:
try:
from logging.handlers import NTEventLogHandler
nt_event_handler = NTEventLogHandler(logger_name, get_win32service_file('windows', 'win32service.pyd'), 'Application')
nt_event_handler.setFormatter(logging.Formatter(get_syslog_format(logger_name), get_log_date_format()))
nt_event_handler.setLevel(logging.ERROR)
app_log = logging.getLogger(logger_name)
app_log.addHandler(nt_event_handler)
except Exception, e:
sys.stderr.write("Error setting up Event viewer logging: '%s'\n" % str(e))
traceback.print_exc()
def get_checksd_path(osname=None):
if not osname:
osname = get_os()
if osname == "windows":
return _windows_checksd_path()
elif osname == "mac":
return _mac_checksd_path()
else:
return _unix_checksd_path()
def initialize_logging(logger_name):
try:
if get_os() == 'windows':
logging.config.fileConfig(get_config_path())
else:
logging_config = get_logging_config()
logging.basicConfig(
format=get_log_format(logger_name),
level=logging_config['log_level'] or logging.INFO,
)
# set up file loggers
log_file = logging_config.get('%s_log_file' % logger_name)
if log_file is not None and not logging_config['disable_file_logging']:
# make sure the log directory is writeable
# NOTE: the entire directory needs to be writable so that rotation works
if os.access(os.path.dirname(log_file), os.R_OK | os.W_OK):
file_handler = logging.handlers.RotatingFileHandler(log_file, maxBytes=LOGGING_MAX_BYTES, backupCount=1)
file_handler.setFormatter(logging.Formatter(get_log_format(logger_name)))
root_log = logging.getLogger()
root_log.addHandler(file_handler)
else:
sys.stderr.write("Log file is unwritable: '%s'\n" % log_file)
# set up syslog
if logging_config['log_to_syslog']:
try:
from logging.handlers import SysLogHandler
if logging_config['syslog_host'] is not None and logging_config['syslog_port'] is not None:
sys_log_addr = (logging_config['syslog_host'], logging_config['syslog_port'])
else:
sys_log_addr = "/dev/log"
# Special-case macs
if sys.platform == 'darwin':
sys_log_addr = "/var/run/syslog"
handler = SysLogHandler(address=sys_log_addr, facility=SysLogHandler.LOG_DAEMON)
handler.setFormatter(logging.Formatter(get_syslog_format(logger_name)))
root_log = logging.getLogger()
root_log.addHandler(handler)
except Exception, e:
sys.stderr.write("Error setting up syslog: '%s'\n" % str(e))
traceback.print_exc()
except Exception, e:
sys.stderr.write("Couldn't initialize logging: %s\n" % str(e))
traceback.print_exc()
# if config fails entirely, enable basic stdout logging as a fallback
logging.basicConfig(
format=get_log_format(logger_name),
level=logging.INFO,
)
def get_confd_path(osname=None):
if not osname:
osname = get_os()
bad_path = ''
if osname == 'windows':
try:
return _windows_confd_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
def get_sdk_integrations_path(osname=None):
if not osname:
osname = get_os()
if osname in ['windows', 'mac']:
raise PathNotFound()
cur_path = os.path.dirname(os.path.realpath(__file__))
path = os.path.join(cur_path, '..', SDK_INTEGRATIONS_DIR)
if os.path.exists(path):
return path
raise PathNotFound(path)
def get_3rd_party_path(osname=None):
if not osname:
osname = get_os()
if osname in ['windows', 'mac']:
raise PathNotFound()
cur_path = os.path.dirname(os.path.realpath(__file__))
path = os.path.join(cur_path, '../3rd-party')
if os.path.exists(path):
return path
raise PathNotFound(path)
def load_class(check_name, class_name):
"""
Retrieve a class with the given name among the given check module.
"""
checksd_path = get_checksd_path(get_os())
check_module = __import__(check_name)
classes = inspect.getmembers(check_module, inspect.isclass)
for name, clsmember in classes:
if name == class_name:
return clsmember
raise Exception(u"Unable to import class {0} from the check module.".format(class_name))
def __init__(self, agentConfig, hostname):
multiprocessing.Process.__init__(self, name='jmxfetch')
self.config = agentConfig
self.is_enabled = True
self.hostname = hostname
osname = get_os()
try:
confd_path = get_confd_path(osname)
except PathNotFound, e:
log.error("No conf.d folder found at '%s' or in the directory where"
"the Agent is currently deployed.\n" % e.args[0])
def get_logging_config(cfg_path=None):
logging_config = {
'log_level': None,
'collector_log_file': '/var/log/datadog/collector.log',
'forwarder_log_file': '/var/log/datadog/forwarder.log',
'dogstatsd_log_file': '/var/log/datadog/dogstatsd.log',
'pup_log_file': '/var/log/datadog/pup.log',
'log_to_syslog': True,
'syslog_host': None,
'syslog_port': None,
}
config_path = get_config_path(cfg_path, os_name=get_os())
config = ConfigParser.ConfigParser()
config.readfp(skip_leading_wsp(open(config_path)))
if config.has_section('handlers') or config.has_section('loggers') or config.has_section('formatters'):
sys.stderr.write("Python logging config is no longer supported and will be ignored.\nTo configure logging, update the logging portion of 'datadog.conf' to match:\n 'https://github.com/DataDog/dd-agent/blob/master/datadog.conf.example'.\n")
for option in logging_config:
if config.has_option('Main', option):
logging_config[option] = config.get('Main', option)
levels = {
'CRITICAL': logging.CRITICAL,
'DEBUG': logging.DEBUG,
'ERROR': logging.ERROR,
'FATAL': logging.FATAL,
'INFO': logging.INFO,
'WARN': logging.WARN,
'WARNING': logging.WARNING,
}
if config.has_option('Main', 'log_level'):
logging_config['log_level'] = levels.get(config.get('Main', 'log_level'))
if config.has_option('Main', 'log_to_syslog'):
logging_config['log_to_syslog'] = config.get('Main', 'log_to_syslog').strip().lower() in ['yes', 'true', 1]
if config.has_option('Main', 'syslog_host'):
host = config.get('Main', 'syslog_host').strip()
if host:
logging_config['syslog_host'] = host
else:
logging_config['syslog_host'] = None
if config.has_option('Main', 'syslog_port'):
port = config.get('Main', 'syslog_port').strip()
try:
logging_config['syslog_port'] = int(port)
except:
logging_config['syslog_port'] = None
return logging_config
def get_confd_path(osname=None):
if not osname:
osname = get_os()
bad_path = ''
try:
if osname == 'windows':
return _windows_confd_path()
elif osname == 'mac':
return _mac_confd_path()
else:
return _unix_confd_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
def __init__(self, agentConfig, hostname):
multiprocessing.Process.__init__(self, name='jmxfetch')
self.config = agentConfig
self.hostname = hostname
try:
osname = get_os()
confd_path = get_confd_path(osname)
self.jmx_daemon = JMXFetch(confd_path, agentConfig)
self.jmx_daemon.configure()
self.is_enabled = self.jmx_daemon.should_run()
except PathNotFound:
self.is_enabled = False
def kill_subprocess(process_obj):
try:
process_obj.terminate()
except AttributeError:
# py < 2.6 doesn't support process.terminate()
if get_os() == 'windows':
import ctypes
PROCESS_TERMINATE = 1
handle = ctypes.windll.kernel32.OpenProcess(PROCESS_TERMINATE, False,
process_obj.pid)
ctypes.windll.kernel32.TerminateProcess(handle, -1)
ctypes.windll.kernel32.CloseHandle(handle)
else:
os.kill(process_obj.pid, signal.SIGKILL)
def get_config_path(cfg_path=None, os_name=None):
# Check if there's an override and if it exists
if cfg_path is not None and os.path.exists(cfg_path):
return cfg_path
if os_name is None:
os_name = get_os()
# Check for an OS-specific path, continue on not-found exceptions
bad_path = ''
if os_name == 'windows':
try:
return _windows_config_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
def get_check(name, config_str):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for name, clsmember in classes:
if AgentCheck in clsmember.__bases__:
check_class = clsmember
break
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
agentConfig = {"version": "0.1", "api_key": "tota"}
return check_class.from_yaml(yaml_text=config_str, check_name=name, agentConfig=agentConfig)
def load_check_directory(agentConfig):
''' Return the initialized checks from checks.d, and a mapping of checks that failed to
initialize. Only checks that have a configuration
file in conf.d will be returned. '''
from checks import AgentCheck
initialized_checks = {}
init_failed_checks = {}
osname = get_os()
checks_paths = [glob.glob(os.path.join(agentConfig['additional_checksd'], '*.py'))]
try:
checksd_path = get_checksd_path(osname)
checks_paths.append(glob.glob(os.path.join(checksd_path, '*.py')))
except PathNotFound, e:
log.error(e.args[0])
sys.exit(3)
def is_running(cls):
try:
pid = JMXFetch.pid_file.get_pid()
if pid is None:
return False
except Exception:
return False
if get_os() != 'windows':
try:
os.kill(pid, 0)
# os.kill(pid, 0) will throw an exception if pid is not running
# and won't do anything otherwise
# It doesn't work on windows as signal.CTRL_C_EVENT is 0, it would quit the process
return True
except Exception, e:
if "Errno 3" not in str(e):
log.debug("Couldn't determine if JMXFetch is running. We suppose it's not. %s" % str(e))
return False
def get_check_class(name):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for _, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
return check_class
def get_check_class(name):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for _, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
return check_class
def is_running(cls):
try:
pid = JMXFetch.pid_file.get_pid()
if pid is None:
return False
except Exception:
return False
if get_os() != 'windows':
try:
os.kill(pid, 0)
# os.kill(pid, 0) will throw an exception if pid is not running
# and won't do anything otherwise
# It doesn't work on windows as signal.CTRL_C_EVENT is 0, it would quit the process
return True
except Exception, e:
if "Errno 3" not in str(e):
log.debug("Couldn't determine if JMXFetch is running. We suppose it's not. %s" % str(e))
return False
def get_checks():
checks = {}
conf_d_directory = get_confd_path(get_os())
for filename in sorted(os.listdir(conf_d_directory)):
module_name, ext = osp.splitext(filename)
if filename.split('.')[0] in EXCLUDED_WINDOWS_CHECKS:
continue
if ext not in ('.yaml', '.example', '.disabled'):
continue
agent_check = AgentCheck(filename, ext, conf_d_directory)
if (agent_check.enabled or agent_check.module_name not in checks or
(not agent_check.is_example and not checks[agent_check.module_name].enabled)):
checks[agent_check.module_name] = agent_check
checks_list = checks.values()
checks_list.sort(key=lambda c: c.module_name)
return checks_list
def register(p_repo, p_isJSON):
[repo_type, name, url, package, key, repo_filter] = get_repo(p_repo, p_isJSON)
if util.get_os() in YUM_LIST:
repo_pkg = url + "/" + package
util.run_sudo("yum install -y " + repo_pkg, True, p_isJSON)
return 0
## APT ################
a_file = "/etc/apt/sources.list.d/pgdg.list"
a_repo = "http://apt.postgresql.org/pub/repos/apt/"
cmd = '"deb ' + a_repo + ' ' + p_repo + ' main" > ' + a_file
util.run_sudo("sh -c 'echo " + cmd + "'", True, p_isJSON)
a_key = "https://www.postgresql.org/media/keys/ACCC4CF8.asc"
cmd = "wget --quiet -O - " + a_key + " | sudo apt-key add -"
util.run_sudo(cmd, True, p_isJSON)
util.run_sudo("apt-get update", True, p_isJSON)
return 0
def get_config_path(cfg_path=None, os_name=None):
# Check if there's an override and if it exists
if cfg_path is not None and os.path.exists(cfg_path):
return cfg_path
if os_name is None:
os_name = get_os()
# Check for an OS-specific path, continue on not-found exceptions
bad_path = ''
try:
if os_name == 'windows':
return _windows_config_path()
elif os_name == 'mac':
return _mac_config_path()
else:
return _unix_config_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
def load_check_directory(agentConfig):
''' Return the initialized checks from checks.d, and a mapping of checks that failed to
initialize. Only checks that have a configuration
file in conf.d will be returned. '''
from checks import AgentCheck
initialized_checks = {}
init_failed_checks = {}
osname = get_os()
checks_paths = [
glob.glob(os.path.join(agentConfig['additional_checksd'], '*.py'))
]
try:
checksd_path = get_checksd_path(osname)
checks_paths.append(glob.glob(os.path.join(checksd_path, '*.py')))
except PathNotFound, e:
log.error(e.args[0])
sys.exit(3)
def get_check(name, config_str):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for name, clsmember in classes:
if AgentCheck in clsmember.__bases__:
check_class = clsmember
break
if check_class is None:
raise Exception("Unable to import check %s. Missing a class that inherits AgentCheck" % name)
agentConfig = {
'version': '0.1',
'api_key': 'tota'
}
return check_class.from_yaml(yaml_text=config_str, check_name=name,
agentConfig=agentConfig)
def load_check(name, config, agentConfig):
if not _is_sdk():
checksd_path = get_checksd_path(get_os())
# find (in checksd_path) and load the check module
fd, filename, desc = imp.find_module(name, [checksd_path])
check_module = imp.load_module(name, fd, filename, desc)
else:
check_module = __import__("check")
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for _, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
if check_class is None:
raise Exception(
"Unable to import check %s. Missing a class that inherits AgentCheck"
% name)
init_config = config.get('init_config', {})
instances = config.get('instances')
agentConfig['checksd_hostname'] = get_hostname(agentConfig)
# init the check class
try:
return check_class(name,
init_config=init_config,
agentConfig=agentConfig,
instances=instances)
except TypeError as e:
raise Exception("Check is using old API, {0}".format(e))
except Exception:
raise
def run_check(name, path=None):
confd_path = path or os.path.join(get_confd_path(get_os()),
'%s.yaml' % name)
try:
f = open(confd_path)
except IOError:
raise Exception('Unable to open configuration at %s' % confd_path)
config_str = f.read()
f.close()
check, instances = get_check(name, config_str)
if not instances:
raise Exception('YAML configuration returned no instances.')
for instance in instances:
check.check(instance)
if check.has_events():
print "Events:\n"
pprint(check.get_events(), indent=4)
print "Metrics:\n"
pprint(check.get_metrics(), indent=4)
def get_confd_path(osname=None):
try:
cur_path = os.path.dirname(os.path.realpath(__file__))
return _confd_path(cur_path)
except PathNotFound as e:
pass
if not osname:
osname = get_os()
bad_path = ''
try:
if osname == 'windows':
return _windows_confd_path()
elif osname == 'mac':
return _mac_confd_path()
else:
return _unix_confd_path()
except PathNotFound as e:
if len(e.args) > 0:
bad_path = e.args[0]
raise PathNotFound(bad_path)
def load_check(name, config, agentConfig):
checksd_path = get_checksd_path(get_os())
if checksd_path not in sys.path:
sys.path.append(checksd_path)
check_module = __import__(name)
check_class = None
classes = inspect.getmembers(check_module, inspect.isclass)
for name, clsmember in classes:
if clsmember == AgentCheck:
continue
if issubclass(clsmember, AgentCheck):
check_class = clsmember
if AgentCheck in clsmember.__bases__:
continue
else:
break
if check_class is None:
raise Exception(
"Unable to import check %s. Missing a class that inherits AgentCheck"
% name)
init_config = config.get('init_config', None)
instances = config.get('instances')
agentConfig['checksd_hostname'] = get_hostname(agentConfig)
# init the check class
try:
return check_class(name,
init_config=init_config,
agentConfig=agentConfig,
instances=instances)
except:
# Backwards compatitiblity for old checks that don't support the
# instances argument.
c = check_class(name, init_config=init_config, agentConfig=agentConfig)
c.instances = instances
return c
def build_ot_python_backend_executable(backend_exec_path):
print(script_tag + "Build procedure started.")
print(script_tag + "Checking for OS.")
os_type = util.get_os()
print(script_tag + "Building OT-App Backend for %s." % os_type)
print(script_tag + "Project directory is: %s" % project_root_dir)
print(script_tag + "Script working directory: %s" % os.getcwd())
# TODO: Check that app frontend assets have been moved to
print(script_tag + "Removing PyInstaller old temp directories.")
remove_pyinstaller_temps()
print(script_tag + "Running PyInstaller process.")
success = pyinstaller_build()
if not success:
print(script_tab + "Removing PyInstaller recent temp directories.")
remove_pyinstaller_temps()
raise SystemExit(script_tab + "Exiting as there was an error in the "
"PyInstaller execution.")
print(script_tag + "Removing old OT-App Backend executable directory.")
if os.path.isfile(backend_exec_path):
os.remove(backend_exec_path)
print(script_tag + "Moving executable folder to backend-dist.")
success = move_executable_folder(backend_exec_path)
if not success:
print(script_tab + "Removing PyInstaller recent temp directories.")
remove_pyinstaller_temps()
raise SystemExit(script_tab + "Exiting now as there was an error in "
"the PyInstaller execution.")
print(script_tag + "Removing PyInstaller recent temp directories.")
remove_pyinstaller_temps()
def load_check_directory(agentConfig, hostname):
''' Return the initialized checks from checks.d, and a mapping of checks that failed to
initialize. Only checks that have a configuration
file in conf.d will be returned. '''
from checks import AGENT_METRICS_CHECK_NAME
initialized_checks = {}
init_failed_checks = {}
deprecated_checks = {}
agentConfig['checksd_hostname'] = hostname
# the TRACE_CONFIG flag is used by the configcheck to trace config object loading and
# where they come from (service discovery, auto configuration or configuration file).
if agentConfig.get(TRACE_CONFIG):
configs_and_sources = {
# check_name: (config_source, config)
}
deprecated_configs_enabled = [v for k, v in OLD_STYLE_PARAMETERS if len([l for l in agentConfig if l.startswith(k)]) > 0]
for deprecated_config in deprecated_configs_enabled:
msg = "Configuring %s in datadog.conf is not supported anymore. Please use conf.d" % deprecated_config
deprecated_checks[deprecated_config] = {'error': msg, 'traceback': None}
log.error(msg)
osname = get_os()
checks_paths = get_checks_paths(agentConfig, osname)
# this can happen if check.d is not found
if checks_paths is None:
log.error('Check directory not found, exiting. The agent is likely misconfigured.')
sys.exit(3)
try:
confd_path = get_confd_path(osname)
except PathNotFound, e:
log.error("No conf.d folder found at '%s' or in the directory where "
"the Agent is currently deployed.\n" % e.args[0])
sys.exit(3)
def build_electron_app():
print(script_tag + "Running electron-builder process.")
platform_type = util.get_os()
process_args = [
which("build"),
os.path.join(project_root_dir, "app"),
"--{}".format(platform_type),
"--{}".format(util.get_arch()),
]
print(process_args)
if platform_type in {'mac'}:
electron_builder_process = subprocess.Popen(process_args)
elif platform_type in {'win', 'linux'}:
electron_builder_process = subprocess.Popen(process_args, shell=True)
electron_builder_process.communicate()
if electron_builder_process.returncode != 0:
raise SystemExit(script_tag + 'Failed to properly build electron app')
print(script_tab + 'electron-builder process completed successfully')
def validate_os(p_isJSON):
os = util.get_os()
if (os in YUM_LIST) or (os in APT_LIST):
return
util.exit_message("OS '" + os + "' not supported for this command.", 1,
p_isJSON)
def get_apt_repo_file_name():
rf_prefix = "/var/lib/apt/lists/apt.postgresql.org_pub_repos_apt_dists_"
rf_suffix = "_main_binary-amd64_Packages"
return (rf_prefix + util.get_os() + "-pgdg" + rf_suffix)
def get_config(parse_args=True, cfg_path=None, options=None):
if parse_args:
options, _ = get_parsed_args()
# General config
agentConfig = {
'check_freq': DEFAULT_CHECK_FREQUENCY,
'dogstatsd_port': 8125,
'dogstatsd_target': 'http://localhost:17123',
'graphite_listen_port': None,
'hostname': None,
'listen_port': None,
'tags': None,
'use_ec2_instance_id': False, # DEPRECATED
'version': get_version(),
'watchdog': True,
'additional_checksd': '/etc/sd-agent/checks.d/',
'bind_host': get_default_bind_host(),
'statsd_metric_namespace': None,
'utf8_decoding': False
}
if Platform.is_mac():
agentConfig['additional_checksd'] = '/usr/local/etc/sd-agent/checks.d/'
# Config handling
try:
# Find the right config file
path = os.path.realpath(__file__)
path = os.path.dirname(path)
config_path = get_config_path(cfg_path, os_name=get_os())
config = ConfigParser.ConfigParser()
config.readfp(skip_leading_wsp(open(config_path)))
# bulk import
for option in config.options('Main'):
agentConfig[option] = config.get('Main', option)
# Store developer mode setting in the agentConfig
if config.has_option('Main', 'developer_mode'):
agentConfig['developer_mode'] = _is_affirmative(
config.get('Main', 'developer_mode'))
# Allow an override with the --profile option
if options is not None and options.profile:
agentConfig['developer_mode'] = True
#
# Core config
#
# FIXME unnecessarily complex
if config.has_option('Main', 'sd_account'):
agentConfig['sd_account'] = config.get('Main', 'sd_account')
agentConfig['use_forwarder'] = False
if options is not None and options.use_forwarder:
listen_port = 17124
if config.has_option('Main', 'listen_port'):
listen_port = int(config.get('Main', 'listen_port'))
agentConfig['sd_url'] = "http://" + agentConfig[
'bind_host'] + ":" + str(listen_port)
agentConfig['use_forwarder'] = True
elif options is not None and not options.disable_sd and options.sd_url:
agentConfig['sd_url'] = options.sd_url
elif config.has_option('Main', 'sd_url'):
agentConfig['sd_url'] = config.get('Main', 'sd_url')
else:
# Default agent URL
agentConfig['sd_url'] = "https://" + agentConfig[
'sd_account'] + ".agent.serverdensity.io"
if agentConfig['sd_url'].endswith('/'):
agentConfig['sd_url'] = agentConfig['sd_url'][:-1]
# Extra checks.d path
# the linux directory is set by default
if config.has_option('Main', 'additional_checksd'):
agentConfig['additional_checksd'] = config.get(
'Main', 'additional_checksd')
elif get_os() == 'windows':
# default windows location
common_path = _windows_commondata_path()
agentConfig['additional_checksd'] = os.path.join(
common_path, 'ServerDensity', 'checks.d')
if config.has_option('Main', 'use_dogstatsd'):
agentConfig['use_dogstatsd'] = config.get(
'Main', 'use_dogstatsd').lower() in ("yes", "true")
else:
agentConfig['use_dogstatsd'] = True
# Service discovery
if config.has_option('Main', 'service_discovery_backend'):
try:
additional_config = extract_agent_config(config)
agentConfig.update(additional_config)
except:
log.error('Failed to load the agent configuration related to '
'service discovery. It will not be used.')
# Concerns only Windows
if config.has_option('Main', 'use_web_info_page'):
agentConfig['use_web_info_page'] = config.get(
'Main', 'use_web_info_page').lower() in ("yes", "true")
else:
agentConfig['use_web_info_page'] = True
# Which agent key to use
agentConfig['agent_key'] = config.get('Main', 'agent_key')
# local traffic only? Default to no
agentConfig['non_local_traffic'] = False
if config.has_option('Main', 'non_local_traffic'):
agentConfig['non_local_traffic'] = config.get(
'Main', 'non_local_traffic').lower() in ("yes", "true")
# DEPRECATED
if config.has_option('Main', 'use_ec2_instance_id'):
use_ec2_instance_id = config.get('Main', 'use_ec2_instance_id')
# translate yes into True, the rest into False
agentConfig['use_ec2_instance_id'] = (
use_ec2_instance_id.lower() == 'yes')
if config.has_option('Main', 'check_freq'):
try:
agentConfig['check_freq'] = int(
config.get('Main', 'check_freq'))
except Exception:
pass
# Custom histogram aggregate/percentile metrics
if config.has_option('Main', 'histogram_aggregates'):
agentConfig['histogram_aggregates'] = get_histogram_aggregates(
config.get('Main', 'histogram_aggregates'))
if config.has_option('Main', 'histogram_percentiles'):
agentConfig['histogram_percentiles'] = get_histogram_percentiles(
config.get('Main', 'histogram_percentiles'))
# Disable Watchdog (optionally)
if config.has_option('Main', 'watchdog'):
if config.get('Main', 'watchdog').lower() in ('no', 'false'):
agentConfig['watchdog'] = False
# Optional graphite listener
if config.has_option('Main', 'graphite_listen_port'):
agentConfig['graphite_listen_port'] = \
int(config.get('Main', 'graphite_listen_port'))
else:
agentConfig['graphite_listen_port'] = None
# Dogstatsd config
dogstatsd_defaults = {
'dogstatsd_port': 8125,
'dogstatsd_target':
'http://' + agentConfig['bind_host'] + ':17123',
}
for key, value in dogstatsd_defaults.iteritems():
if config.has_option('Main', key):
agentConfig[key] = config.get('Main', key)
else:
agentConfig[key] = value
# Create app:xxx tags based on monitored apps
agentConfig['create_dd_check_tags'] = config.has_option('Main', 'create_dd_check_tags') and \
_is_affirmative(config.get('Main', 'create_dd_check_tags'))
# Forwarding to external statsd server
if config.has_option('Main', 'statsd_forward_host'):
agentConfig['statsd_forward_host'] = config.get(
'Main', 'statsd_forward_host')
if config.has_option('Main', 'statsd_forward_port'):
agentConfig['statsd_forward_port'] = int(
config.get('Main', 'statsd_forward_port'))
# optionally send dogstatsd data directly to the agent.
if config.has_option('Main', 'dogstatsd_use_ddurl'):
if _is_affirmative(config.get('Main', 'dogstatsd_use_ddurl')):
agentConfig['dogstatsd_target'] = agentConfig['sd_url']
# Optional config
# FIXME not the prettiest code ever...
if config.has_option('Main', 'use_mount'):
agentConfig['use_mount'] = _is_affirmative(
config.get('Main', 'use_mount'))
if options is not None and options.autorestart:
agentConfig['autorestart'] = True
elif config.has_option('Main', 'autorestart'):
agentConfig['autorestart'] = _is_affirmative(
config.get('Main', 'autorestart'))
if config.has_option('Main', 'check_timings'):
agentConfig['check_timings'] = _is_affirmative(
config.get('Main', 'check_timings'))
if config.has_option('Main', 'exclude_process_args'):
agentConfig['exclude_process_args'] = _is_affirmative(
config.get('Main', 'exclude_process_args'))
try:
filter_device_re = config.get('Main', 'device_blacklist_re')
agentConfig['device_blacklist_re'] = re.compile(filter_device_re)
except ConfigParser.NoOptionError:
pass
if config.has_option("Main", "nagios_perf_cfg"):
agentConfig["nagios_perf_cfg"] = config.get(
"Main", "nagios_perf_cfg")
if config.has_option("Main", "use_curl_http_client"):
agentConfig["use_curl_http_client"] = _is_affirmative(
config.get("Main", "use_curl_http_client"))
else:
# Default to False as there are some issues with the curl client and ELB
agentConfig["use_curl_http_client"] = False
if config.has_section('WMI'):
agentConfig['WMI'] = {}
for key, value in config.items('WMI'):
agentConfig['WMI'][key] = value
if (config.has_option("Main", "limit_memory_consumption") and
config.get("Main", "limit_memory_consumption") is not None):
agentConfig["limit_memory_consumption"] = int(
config.get("Main", "limit_memory_consumption"))
else:
agentConfig["limit_memory_consumption"] = None
if config.has_option("Main", "skip_ssl_validation"):
agentConfig["skip_ssl_validation"] = _is_affirmative(
config.get("Main", "skip_ssl_validation"))
agentConfig["collect_instance_metadata"] = True
if config.has_option("Main", "collect_instance_metadata"):
agentConfig["collect_instance_metadata"] = _is_affirmative(
config.get("Main", "collect_instance_metadata"))
agentConfig["proxy_forbid_method_switch"] = False
if config.has_option("Main", "proxy_forbid_method_switch"):
agentConfig["proxy_forbid_method_switch"] = _is_affirmative(
config.get("Main", "proxy_forbid_method_switch"))
agentConfig["collect_ec2_tags"] = False
if config.has_option("Main", "collect_ec2_tags"):
agentConfig["collect_ec2_tags"] = _is_affirmative(
config.get("Main", "collect_ec2_tags"))
agentConfig["utf8_decoding"] = False
if config.has_option("Main", "utf8_decoding"):
agentConfig["utf8_decoding"] = _is_affirmative(
config.get("Main", "utf8_decoding"))
agentConfig["gce_updated_hostname"] = False
if config.has_option("Main", "gce_updated_hostname"):
agentConfig["gce_updated_hostname"] = _is_affirmative(
config.get("Main", "gce_updated_hostname"))
except ConfigParser.NoSectionError, e:
sys.stderr.write('Config file not found or incorrectly formatted.\n')
sys.exit(2)
def initialize_logging(logger_name):
try:
logging_config = get_logging_config()
logging.basicConfig(
format=get_log_format(logger_name),
level=logging_config['log_level'] or logging.INFO,
)
log_file = logging_config.get('%s_log_file' % logger_name)
if log_file is not None and not logging_config['disable_file_logging']:
# make sure the log directory is writeable
# NOTE: the entire directory needs to be writable so that rotation works
if os.access(os.path.dirname(log_file), os.R_OK | os.W_OK):
file_handler = logging.handlers.RotatingFileHandler(
log_file, maxBytes=LOGGING_MAX_BYTES, backupCount=1)
formatter = logging.Formatter(get_log_format(logger_name),
get_log_date_format())
file_handler.setFormatter(formatter)
root_log = logging.getLogger()
root_log.addHandler(file_handler)
else:
sys.stderr.write("Log file is unwritable: '%s'\n" % log_file)
# set up syslog
if logging_config['log_to_syslog']:
try:
from logging.handlers import SysLogHandler
if logging_config['syslog_host'] is not None and logging_config[
'syslog_port'] is not None:
sys_log_addr = (logging_config['syslog_host'],
logging_config['syslog_port'])
else:
sys_log_addr = "/dev/log"
# Special-case BSDs
if Platform.is_darwin():
sys_log_addr = "/var/run/syslog"
elif Platform.is_freebsd():
sys_log_addr = "/var/run/log"
handler = SysLogHandler(address=sys_log_addr,
facility=SysLogHandler.LOG_DAEMON)
handler.setFormatter(
logging.Formatter(get_syslog_format(logger_name),
get_log_date_format()))
root_log = logging.getLogger()
root_log.addHandler(handler)
except Exception, e:
sys.stderr.write("Error setting up syslog: '%s'\n" % str(e))
traceback.print_exc()
# Setting up logging in the event viewer for windows
if get_os() == 'windows' and logging_config['log_to_event_viewer']:
try:
from logging.handlers import NTEventLogHandler
nt_event_handler = NTEventLogHandler(
logger_name,
get_win32service_file('windows', 'win32service.pyd'),
'Application')
nt_event_handler.setFormatter(
logging.Formatter(get_syslog_format(logger_name),
get_log_date_format()))
nt_event_handler.setLevel(logging.ERROR)
app_log = logging.getLogger(logger_name)
app_log.addHandler(nt_event_handler)
except Exception, e:
sys.stderr.write(
"Error setting up Event viewer logging: '%s'\n" % str(e))
traceback.print_exc()
def get_logging_config(cfg_path=None):
system_os = get_os()
logging_config = {
'log_level': None,
'log_to_event_viewer': False,
'log_to_syslog': False,
'syslog_host': None,
'syslog_port': None,
}
if system_os == 'windows':
logging_config['windows_collector_log_file'] = os.path.join(
_windows_commondata_path(), 'ServerDensity', 'logs',
'collector.log')
logging_config['windows_forwarder_log_file'] = os.path.join(
_windows_commondata_path(), 'ServerDensity', 'logs',
'forwarder.log')
logging_config['windows_dogstatsd_log_file'] = os.path.join(
_windows_commondata_path(), 'ServerDensity', 'logs',
'dogstatsd.log')
logging_config['jmxfetch_log_file'] = os.path.join(
_windows_commondata_path(), 'ServerDensity', 'logs',
'jmxfetch.log')
else:
logging_config[
'collector_log_file'] = '/var/log/sd-agent/collector.log'
logging_config[
'forwarder_log_file'] = '/var/log/sd-agent/forwarder.log'
logging_config[
'dogstatsd_log_file'] = '/var/log/sd-agent/dogstatsd.log'
logging_config['jmxfetch_log_file'] = '/var/log/sd-agent/jmxfetch.log'
logging_config['go-metro_log_file'] = '/var/log/sd-agent/go-metro.log'
logging_config['log_to_syslog'] = True
config_path = get_config_path(cfg_path, os_name=system_os)
config = ConfigParser.ConfigParser()
config.readfp(skip_leading_wsp(open(config_path)))
if config.has_section('handlers') or config.has_section(
'loggers') or config.has_section('formatters'):
if system_os == 'windows':
config_example_file = "https://github.com/serverdensity/sd-agent/blob/master/packaging/sd-agent/win32/install_files/config_win32.conf"
else:
config_example_file = "https://github.com/serverdensity/sd-agent/blob/master/config.cfg.example"
sys.stderr.write(
"""Python logging config is no longer supported and will be ignored.
To configure logging, update the logging portion of 'config.cfg' to match:
'%s'.
""" % config_example_file)
for option in logging_config:
if config.has_option('Main', option):
logging_config[option] = config.get('Main', option)
levels = {
'CRITICAL': logging.CRITICAL,
'DEBUG': logging.DEBUG,
'ERROR': logging.ERROR,
'FATAL': logging.FATAL,
'INFO': logging.INFO,
'WARN': logging.WARN,
'WARNING': logging.WARNING,
}
if config.has_option('Main', 'log_level'):
logging_config['log_level'] = levels.get(
config.get('Main', 'log_level'))
if config.has_option('Main', 'log_to_syslog'):
logging_config['log_to_syslog'] = config.get(
'Main', 'log_to_syslog').strip().lower() in ['yes', 'true', 1]
if config.has_option('Main', 'log_to_event_viewer'):
logging_config['log_to_event_viewer'] = config.get(
'Main',
'log_to_event_viewer').strip().lower() in ['yes', 'true', 1]
if config.has_option('Main', 'syslog_host'):
host = config.get('Main', 'syslog_host').strip()
if host:
logging_config['syslog_host'] = host
else:
logging_config['syslog_host'] = None
if config.has_option('Main', 'syslog_port'):
port = config.get('Main', 'syslog_port').strip()
try:
logging_config['syslog_port'] = int(port)
except Exception:
logging_config['syslog_port'] = None
if config.has_option('Main', 'disable_file_logging'):
logging_config['disable_file_logging'] = config.get(
'Main',
'disable_file_logging').strip().lower() in ['yes', 'true', 1]
else:
logging_config['disable_file_logging'] = False
return logging_config
def get_log_format(logger_name):
if get_os() != 'windows':
return '%%(asctime)s | %%(levelname)s | sd.%s | %%(name)s(%%(filename)s:%%(lineno)s) | %%(message)s' % logger_name
return '%(asctime)s | %(levelname)s | %(name)s(%(filename)s:%(lineno)s) | %(message)s'
print "#" * 80
print "JMX tool to be used to help configuring your JMX checks."
print "See http://docs.datadoghq.com/integrations/java/ for more information"
print "#" * 80
print "\n"
print "You have to specify one of the following command:"
for command, desc in JMX_LIST_COMMANDS.iteritems():
print " - %s [OPTIONAL: LIST OF CHECKS]: %s" % (command,
desc)
print "Example: sudo /etc/init.d/datadog-agent jmx list_matching_attributes tomcat jmx solr"
print "\n"
else:
jmx_command = args[1]
checks_list = args[2:]
confd_directory = get_confd_path(get_os())
should_run = JMXFetch.init(confd_directory,
agentConfig,
get_logging_config(),
15,
jmx_command,
checks_list,
reporter="console")
if not should_run:
print "Couldn't find any valid JMX configuration in your conf.d directory: %s" % confd_directory
print "Have you enabled any JMX check ?"
print "If you think it's not normal please get in touch with Datadog Support"
return 0
def get_config_path(cfg_path=None, os_name=None):
# Check if there's an override and if it exists
if cfg_path is not None and os.path.exists(cfg_path):
return cfg_path
# Check if there's a config stored in the current agent directory
try:
path = os.path.realpath(__file__)
path = os.path.dirname(path)
return _config_path(path)
except PathNotFound, e:
pass
if os_name is None:
os_name = get_os()
# Check for an OS-specific path, continue on not-found exceptions
bad_path = ''
try:
if os_name == 'windows':
return _windows_config_path()
elif os_name == 'mac':
return _mac_config_path()
else:
return _unix_config_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
# If all searches fail, exit the agent with an error
def get_config(parse_args=True, cfg_path=None, options=None):
if parse_args:
options, _ = get_parsed_args()
# General config
agentConfig = {
'check_freq': DEFAULT_CHECK_FREQUENCY,
'dogstatsd_port': 8125,
'dogstatsd_target': 'http://localhost:17123',
'graphite_listen_port': None,
'hostname': None,
'listen_port': None,
'tags': None,
'use_ec2_instance_id': False, # DEPRECATED
'version': get_version(),
'watchdog': True,
'additional_checksd': '/etc/dd-agent/checks.d/',
'bind_host': get_default_bind_host(),
'statsd_metric_namespace': None,
'utf8_decoding': False
}
if Platform.is_mac():
agentConfig['additional_checksd'] = '/opt/datadog-agent/etc/checks.d'
# Config handling
try:
# Find the right config file
path = os.path.realpath(__file__)
path = os.path.dirname(path)
config_path = get_config_path(cfg_path, os_name=get_os())
config = ConfigParser.ConfigParser()
config.readfp(skip_leading_wsp(open(config_path)))
# bulk import
for option in config.options('Main'):
agentConfig[option] = config.get('Main', option)
# Store developer mode setting in the agentConfig
if config.has_option('Main', 'developer_mode'):
agentConfig['developer_mode'] = _is_affirmative(
config.get('Main', 'developer_mode'))
# Allow an override with the --profile option
if options is not None and options.profile:
agentConfig['developer_mode'] = True
#
# Core config
#ap
if not config.has_option('Main', 'api_key'):
log.warning(u"No API key was found. Aborting.")
sys.exit(2)
if not config.has_option('Main', 'dd_url'):
log.warning(u"No dd_url was found. Aborting.")
sys.exit(2)
# Endpoints
dd_url = clean_dd_url(config.get('Main', 'dd_url'))
api_key = config.get('Main', 'api_key').strip()
# For collector and dogstatsd
agentConfig['api_key'] = api_key
agentConfig['dd_url'] = dd_url
# multiple endpoints
if config.has_option('Main', 'other_dd_urls'):
other_dd_urls = map(clean_dd_url,
config.get('Main', 'other_dd_urls').split(','))
else:
other_dd_urls = []
if config.has_option('Main', 'other_api_keys'):
other_api_keys = map(
lambda x: x.strip(),
config.get('Main', 'other_api_keys').split(','))
else:
other_api_keys = []
# Forwarder endpoints logic
# endpoints is:
# {
# 'https://app.datadoghq.com': ['api_key_abc', 'api_key_def'],
# 'https://app.example.com': ['api_key_xyz']
# }
endpoints = {dd_url: [api_key]}
if len(other_dd_urls) == 0:
endpoints[dd_url] += other_api_keys
else:
assert len(other_dd_urls) == len(
other_api_keys), 'Please provide one api_key for each url'
for i, other_dd_url in enumerate(other_dd_urls):
endpoints[other_dd_url] = endpoints.get(
other_dd_url, []) + [other_api_keys[i]]
agentConfig['endpoints'] = endpoints
# Forwarder or not forwarder
agentConfig[
'use_forwarder'] = options is not None and options.use_forwarder
if agentConfig['use_forwarder']:
listen_port = 17123
if config.has_option('Main', 'listen_port'):
listen_port = int(config.get('Main', 'listen_port'))
agentConfig['dd_url'] = "http://{}:{}".format(
agentConfig['bind_host'], listen_port)
# FIXME: Legacy dd_url command line switch
elif options is not None and options.dd_url is not None:
agentConfig['dd_url'] = options.dd_url
# Forwarder timeout
agentConfig['forwarder_timeout'] = 20
if config.has_option('Main', 'forwarder_timeout'):
agentConfig['forwarder_timeout'] = int(
config.get('Main', 'forwarder_timeout'))
# Extra checks.d path
# the linux directory is set by default
if config.has_option('Main', 'additional_checksd'):
agentConfig['additional_checksd'] = config.get(
'Main', 'additional_checksd')
elif get_os() == 'windows':
# default windows location
common_path = _windows_commondata_path()
agentConfig['additional_checksd'] = os.path.join(
common_path, 'Datadog', 'checks.d')
if config.has_option('Main', 'use_dogstatsd'):
agentConfig['use_dogstatsd'] = config.get(
'Main', 'use_dogstatsd').lower() in ("yes", "true")
else:
agentConfig['use_dogstatsd'] = True
# Service discovery
if config.has_option('Main', 'service_discovery_backend'):
try:
additional_config = extract_agent_config(config)
agentConfig.update(additional_config)
except:
log.error('Failed to load the agent configuration related to '
'service discovery. It will not be used.')
# Concerns only Windows
if config.has_option('Main', 'use_web_info_page'):
agentConfig['use_web_info_page'] = config.get(
'Main', 'use_web_info_page').lower() in ("yes", "true")
else:
agentConfig['use_web_info_page'] = True
# local traffic only? Default to no
agentConfig['non_local_traffic'] = False
if config.has_option('Main', 'non_local_traffic'):
agentConfig['non_local_traffic'] = config.get(
'Main', 'non_local_traffic').lower() in ("yes", "true")
# DEPRECATED
if config.has_option('Main', 'use_ec2_instance_id'):
use_ec2_instance_id = config.get('Main', 'use_ec2_instance_id')
# translate yes into True, the rest into False
agentConfig['use_ec2_instance_id'] = (
use_ec2_instance_id.lower() == 'yes')
if config.has_option('Main', 'check_freq'):
try:
agentConfig['check_freq'] = int(
config.get('Main', 'check_freq'))
except Exception:
pass
# Custom histogram aggregate/percentile metrics
if config.has_option('Main', 'histogram_aggregates'):
agentConfig['histogram_aggregates'] = get_histogram_aggregates(
config.get('Main', 'histogram_aggregates'))
if config.has_option('Main', 'histogram_percentiles'):
agentConfig['histogram_percentiles'] = get_histogram_percentiles(
config.get('Main', 'histogram_percentiles'))
# Disable Watchdog (optionally)
if config.has_option('Main', 'watchdog'):
if config.get('Main', 'watchdog').lower() in ('no', 'false'):
agentConfig['watchdog'] = False
# Optional graphite listener
if config.has_option('Main', 'graphite_listen_port'):
agentConfig['graphite_listen_port'] = \
int(config.get('Main', 'graphite_listen_port'))
else:
agentConfig['graphite_listen_port'] = None
# Dogstatsd config
dogstatsd_defaults = {
'dogstatsd_port': 8125,
'dogstatsd_target':
'http://' + agentConfig['bind_host'] + ':17123',
}
for key, value in dogstatsd_defaults.iteritems():
if config.has_option('Main', key):
agentConfig[key] = config.get('Main', key)
else:
agentConfig[key] = value
# Create app:xxx tags based on monitored apps
agentConfig['create_dd_check_tags'] = config.has_option('Main', 'create_dd_check_tags') and \
_is_affirmative(config.get('Main', 'create_dd_check_tags'))
# Forwarding to external statsd server
if config.has_option('Main', 'statsd_forward_host'):
agentConfig['statsd_forward_host'] = config.get(
'Main', 'statsd_forward_host')
if config.has_option('Main', 'statsd_forward_port'):
agentConfig['statsd_forward_port'] = int(
config.get('Main', 'statsd_forward_port'))
# optionally send dogstatsd data directly to the agent.
if config.has_option('Main', 'dogstatsd_use_ddurl'):
if _is_affirmative(config.get('Main', 'dogstatsd_use_ddurl')):
agentConfig['dogstatsd_target'] = agentConfig['dd_url']
# Optional config
# FIXME not the prettiest code ever...
if config.has_option('Main', 'use_mount'):
agentConfig['use_mount'] = _is_affirmative(
config.get('Main', 'use_mount'))
if options is not None and options.autorestart:
agentConfig['autorestart'] = True
elif config.has_option('Main', 'autorestart'):
agentConfig['autorestart'] = _is_affirmative(
config.get('Main', 'autorestart'))
if config.has_option('Main', 'check_timings'):
agentConfig['check_timings'] = _is_affirmative(
config.get('Main', 'check_timings'))
if config.has_option('Main', 'exclude_process_args'):
agentConfig['exclude_process_args'] = _is_affirmative(
config.get('Main', 'exclude_process_args'))
try:
filter_device_re = config.get('Main', 'device_blacklist_re')
agentConfig['device_blacklist_re'] = re.compile(filter_device_re)
except ConfigParser.NoOptionError:
pass
if config.has_option('datadog', 'ddforwarder_log'):
agentConfig['has_datadog'] = True
# Dogstream config
if config.has_option("Main", "dogstream_log"):
# Older version, single log support
log_path = config.get("Main", "dogstream_log")
if config.has_option("Main", "dogstream_line_parser"):
agentConfig["dogstreams"] = ':'.join(
[log_path,
config.get("Main", "dogstream_line_parser")])
else:
agentConfig["dogstreams"] = log_path
elif config.has_option("Main", "dogstreams"):
agentConfig["dogstreams"] = config.get("Main", "dogstreams")
if config.has_option("Main", "nagios_perf_cfg"):
agentConfig["nagios_perf_cfg"] = config.get(
"Main", "nagios_perf_cfg")
if config.has_option("Main", "use_curl_http_client"):
agentConfig["use_curl_http_client"] = _is_affirmative(
config.get("Main", "use_curl_http_client"))
else:
# Default to False as there are some issues with the curl client and ELB
agentConfig["use_curl_http_client"] = False
if config.has_section('WMI'):
agentConfig['WMI'] = {}
for key, value in config.items('WMI'):
agentConfig['WMI'][key] = value
if (config.has_option("Main", "limit_memory_consumption") and
config.get("Main", "limit_memory_consumption") is not None):
agentConfig["limit_memory_consumption"] = int(
config.get("Main", "limit_memory_consumption"))
else:
agentConfig["limit_memory_consumption"] = None
if config.has_option("Main", "skip_ssl_validation"):
agentConfig["skip_ssl_validation"] = _is_affirmative(
config.get("Main", "skip_ssl_validation"))
agentConfig["collect_instance_metadata"] = True
if config.has_option("Main", "collect_instance_metadata"):
agentConfig["collect_instance_metadata"] = _is_affirmative(
config.get("Main", "collect_instance_metadata"))
agentConfig["proxy_forbid_method_switch"] = False
if config.has_option("Main", "proxy_forbid_method_switch"):
agentConfig["proxy_forbid_method_switch"] = _is_affirmative(
config.get("Main", "proxy_forbid_method_switch"))
agentConfig["collect_ec2_tags"] = False
if config.has_option("Main", "collect_ec2_tags"):
agentConfig["collect_ec2_tags"] = _is_affirmative(
config.get("Main", "collect_ec2_tags"))
agentConfig["utf8_decoding"] = False
if config.has_option("Main", "utf8_decoding"):
agentConfig["utf8_decoding"] = _is_affirmative(
config.get("Main", "utf8_decoding"))
agentConfig["gce_updated_hostname"] = False
if config.has_option("Main", "gce_updated_hostname"):
agentConfig["gce_updated_hostname"] = _is_affirmative(
config.get("Main", "gce_updated_hostname"))
except ConfigParser.NoSectionError as e:
sys.stderr.write('Config file not found or incorrectly formatted.\n')
sys.exit(2)
except ConfigParser.ParsingError as e:
sys.stderr.write('Config file not found or incorrectly formatted.\n')
sys.exit(2)
except ConfigParser.NoOptionError as e:
sys.stderr.write(
'There are some items missing from your config file, but nothing fatal [%s]'
% e)
# Storing proxy settings in the agentConfig
agentConfig['proxy_settings'] = get_proxy(agentConfig)
if agentConfig.get('ca_certs', None) is None:
agentConfig['ssl_certificate'] = get_ssl_certificate(
get_os(), 'datadog-cert.pem')
else:
agentConfig['ssl_certificate'] = agentConfig['ca_certs']
return agentConfig
def __init__(self, agentConfig, emitters, systemStats, hostname):
self.emit_duration = None
self.agentConfig = agentConfig
self.hostname = hostname
# system stats is generated by config.get_system_stats
self.agentConfig['system_stats'] = systemStats
# agent config is used during checks, system_stats can be accessed through the config
self.os = get_os()
self.plugins = None
self.emitters = emitters
self.check_timings = agentConfig.get('check_timings')
self.push_times = {
'host_metadata': {
'start': time.time(),
'interval': int(agentConfig.get('metadata_interval', 4 * 60 * 60))
},
'external_host_tags': {
'start': time.time() - 3 * 60, # Wait for the checks to init
'interval': int(agentConfig.get('external_host_tags', 5 * 60))
},
'agent_checks': {
'start': time.time(),
'interval': int(agentConfig.get('agent_checks_interval', 10 * 60))
},
'processes': {
'start': time.time(),
'interval': int(agentConfig.get('processes_interval', 60))
}
}
socket.setdefaulttimeout(15)
self.run_count = 0
self.continue_running = True
self.hostname_metadata_cache = None
self.initialized_checks_d = []
self.init_failed_checks_d = {}
# Unix System Checks
self._unix_system_checks = {
'io': u.IO(log),
'load': u.Load(log),
'memory': u.Memory(log),
'processes': u.Processes(log),
'cpu': u.Cpu(log),
'system': u.System(log)
}
# Win32 System `Checks
self._win32_system_checks = {
'io': w32.IO(log),
'proc': w32.Processes(log),
'memory': w32.Memory(log),
'network': w32.Network(log),
'cpu': w32.Cpu(log),
'system': w32.System(log)
}
# Old-style metric checks
self._ganglia = Ganglia(log)
self._dogstream = Dogstreams.init(log, self.agentConfig)
self._ddforwarder = DdForwarder(log, self.agentConfig)
# Agent performance metrics check
self._agent_metrics = None
self._metrics_checks = []
# Custom metric checks
for module_spec in [s.strip() for s in self.agentConfig.get('custom_checks', '').split(',')]:
if len(module_spec) == 0:
continue
try:
self._metrics_checks.append(modules.load(module_spec, 'Check')(log))
log.info("Registered custom check %s" % module_spec)
log.warning("Old format custom checks are deprecated. They should be moved to the checks.d interface as old custom checks will be removed in a next version")
except Exception:
log.exception('Unable to load custom check module %s' % module_spec)
def __init__(self, agentConfig, emitters, systemStats):
self.emit_duration = None
self.agentConfig = agentConfig
# system stats is generated by config.get_system_stats
self.agentConfig['system_stats'] = systemStats
# agent config is used during checks, system_stats can be accessed through the config
self.os = get_os()
self.plugins = None
self.emitters = emitters
self.metadata_interval = int(
agentConfig.get('metadata_interval', 10 * 60))
self.metadata_start = time.time()
socket.setdefaulttimeout(15)
self.run_count = 0
self.continue_running = True
self.metadata_cache = None
self.initialized_checks_d = []
self.init_failed_checks_d = []
# Unix System Checks
self._unix_system_checks = {
'disk': u.Disk(log),
'io': u.IO(log),
'load': u.Load(log),
'memory': u.Memory(log),
'processes': u.Processes(log),
'cpu': u.Cpu(log)
}
# Win32 System `Checks
self._win32_system_checks = {
'disk': w32.Disk(log),
'io': w32.IO(log),
'proc': w32.Processes(log),
'memory': w32.Memory(log),
'network': w32.Network(log),
'cpu': w32.Cpu(log)
}
# Old-style metric checks
self._ganglia = Ganglia(log)
self._cassandra = Cassandra()
self._dogstream = Dogstreams.init(log, self.agentConfig)
self._ddforwarder = DdForwarder(log, self.agentConfig)
# Agent Metrics
self._agent_metrics = CollectorMetrics(log)
self._metrics_checks = []
# Custom metric checks
for module_spec in [
s.strip()
for s in self.agentConfig.get('custom_checks', '').split(',')
]:
if len(module_spec) == 0: continue
try:
self._metrics_checks.append(
modules.load(module_spec, 'Check')(log))
log.info("Registered custom check %s" % module_spec)
except Exception, e:
log.exception('Unable to load custom check module %s' %
module_spec)
def get_config_path(cfg_path=None, os_name=None):
# Check if there's an override and if it exists
if cfg_path is not None and os.path.exists(cfg_path):
return cfg_path
# Check if there's a config stored in the current agent directory
try:
path = os.path.realpath(__file__)
path = os.path.dirname(path)
return _config_path(path)
except PathNotFound, e:
pass
if os_name is None:
os_name = get_os()
# Check for an OS-specific path, continue on not-found exceptions
bad_path = ''
try:
if os_name == 'windows':
return _windows_config_path()
elif os_name == 'mac':
return _mac_config_path()
else:
return _unix_config_path()
except PathNotFound, e:
if len(e.args) > 0:
bad_path = e.args[0]
# If all searches fail, exit the agent with an error
def list_packages(p_repo, p_SHOWDUPS, p_isJSON, p_isEXTRA):
if not is_repo(p_repo, p_isJSON):
util.exit_message(p_repo + " is not a valid REPO.", 1, p_isJSON)
[repo_type, name, url, package, key,
pkg_filter] = get_repo(p_repo, p_isJSON)
if not is_installed(p_repo):
util.exit_message(p_repo + " is not registered.", 1, p_isJSON)
options = ""
if p_SHOWDUPS:
options = "--showduplicates"
if util.get_os() in APT_LIST:
return list_apt_packages(p_repo, p_isJSON)
os = util.get_os()
if os == "el6":
cmd = "yum list all | grep " + p_repo
else:
cmd = "yum repo-pkgs " + p_repo + " list " + options
cmd = cmd + " | awk '"
## filter package list unless asked to show --extra or --test
kount = 0
if not p_isEXTRA:
for p in pkg_filter:
kount = kount + 1
ps = "/" + p.replace('.', '\.') + "/"
if kount > 1:
cmd = cmd + " || " + ps
else:
cmd = cmd + ps
cmd = "sudo " + cmd + " { print }' | awk '!/debug/ && !/docs/ { print }'"
outp = util.getoutput(cmd)
my_logger.info("\n$ " + cmd + "\n\n" + str(outp))
repoList = []
for line in outp.splitlines():
data = line.split()
if len(data) != 3:
continue
repoDict = {}
p1 = data[0].find('.')
pkg_nm = data[0][0:p1]
p2 = data[1].find('.rhel')
if p2 > 0:
pkg_ver = data[1][0:p2]
else:
pkg_ver = data[1]
status = ""
if data[2].startswith("@"):
status = "Installed"
repoDict['component'] = pkg_nm
repoDict['version'] = pkg_ver
repoDict['status'] = status
if pkg_nm > "":
repoList.append(repoDict)
keys = ['component', 'version', 'status']
headers = ['Component', 'Version', 'Status']
if p_isJSON:
print(json.dumps(repoList, sort_keys=True, indent=2))
else:
print(api.format_data_to_table(repoList, keys, headers))
return (0)
def load_check_directory(agentConfig, hostname):
''' Return the initialized checks from checks.d, and a mapping of checks that failed to
initialize. Only checks that have a configuration
file in conf.d will be returned. '''
from checks import AGENT_METRICS_CHECK_NAME
initialized_checks = {}
init_failed_checks = {}
deprecated_checks = {}
agentConfig['checksd_hostname'] = hostname
osname = get_os()
# the TRACE_CONFIG flag is used by the configcheck to trace config object loading and
# where they come from (service discovery, auto config or config file)
if agentConfig.get(TRACE_CONFIG):
configs_and_sources = {
# check_name: (config_source, config)
}
deprecated_checks.update(_deprecated_configs(agentConfig))
checks_places = get_checks_places(osname, agentConfig)
for config_path in _file_configs_paths(osname, agentConfig):
# '/etc/dd-agent/checks.d/my_check.py' -> 'my_check'
check_name = _conf_path_to_check_name(config_path)
conf_is_valid, check_config, invalid_check = _load_file_config(
config_path, check_name, agentConfig)
init_failed_checks.update(invalid_check)
if not conf_is_valid:
continue
if agentConfig.get(TRACE_CONFIG):
configs_and_sources[check_name] = (CONFIG_FROM_FILE, check_config)
# load the check
load_success, load_failure = load_check_from_places(
check_config, check_name, checks_places, agentConfig)
initialized_checks.update(load_success)
init_failed_checks.update(load_failure)
for check_name, service_disco_check_config in _service_disco_configs(
agentConfig).iteritems():
# ignore this config from service disco if the check has been loaded through a file config
if check_name in initialized_checks or check_name in init_failed_checks:
continue
# if TRACE_CONFIG is set, service_disco_check_config looks like:
# (config_src, (sd_init_config, sd_instances)) instead of
# (sd_init_config, sd_instances)
if agentConfig.get(TRACE_CONFIG):
sd_init_config, sd_instances = service_disco_check_config[1]
configs_and_sources[check_name] = (service_disco_check_config[0], {
'init_config': sd_init_config,
'instances': sd_instances
})
else:
sd_init_config, sd_instances = service_disco_check_config
check_config = {
'init_config': sd_init_config,
'instances': sd_instances
}
# load the check
load_success, load_failure = load_check_from_places(
check_config, check_name, checks_places, agentConfig)
initialized_checks.update(load_success)
init_failed_checks.update(load_failure)
init_failed_checks.update(deprecated_checks)
log.info('initialized checks.d checks: %s' % [
k for k in initialized_checks.keys() if k != AGENT_METRICS_CHECK_NAME
])
log.info('initialization failed checks.d checks: %s' %
init_failed_checks.keys())
if agentConfig.get(TRACE_CONFIG):
return configs_and_sources
return {
'initialized_checks': initialized_checks.values(),
'init_failed_checks': init_failed_checks,
}
def get_pgdg_base(p_ver, p_isJSON):
basedir = None
ver = p_ver
svcname = None
datadir = None
port = None
logdir = None
to_devnull = " >/dev/null 2>&1"
####### (1) PGHOME ##############################
if util.get_os() in YUM_LIST:
basedir = "/usr/pgsql-" + ver
else:
basedir = "/usr/lib/postgresql/" + ver
pgbin = basedir + "/bin/postgres"
if not os.path.isfile(pgbin):
print(" PGHOME could not be located at " + pgbin)
return ("1111111")
###### (2) Service Control File #################
if util.get_os() in APT_LIST:
svcname = "postgresql"
else:
svcname = "postgresql-" + ver
if util.is_systemd():
sysdir = util.get_systemd_dir()
svcfile = sysdir + "/" + svcname + ".service"
else:
sysdir = "/etc/init.d"
svcfile = sysdir + "/" + svcname
if not os.path.isfile(svcfile):
print("ERROR: ServiceFile not found (" + svcfile + ")")
return ("2222222")
###### (3) DATADIR ###############################
if util.get_os() in YUM_LIST:
datadir = "/var/lib/pgsql/" + ver + "/data"
else:
datadir = "/var/lib/postgresql/" + ver + "/main"
cmd = "sudo ls " + datadir
rc = os.system(cmd + to_devnull)
if rc != 0:
print("ERROR: DataDir not found (" + datadir + ")")
return ("3333333")
##### LOOK FOR PORT ####################
pidfile = datadir + "/postmaster.pid"
cmd = "sudo ls " + pidfile
rc = os.system(cmd + to_devnull)
if rc == 0:
cmd = "sudo cat " + pidfile + " | sed -n '4p'"
port = util.getoutput(cmd)
else:
port = "1"
##### LOOK FOR LOGDIR ##################
if util.get_os() in APT_LIST:
logdir = "/var/log/postgresql"
else:
logdir = datadir + "/pg_log"
cmd = "sudo ls " + logdir
rc = os.system(cmd + to_devnull)
if rc != 0:
logdir = ""
return ([basedir, ver, svcname, svcfile, datadir, port, logdir])
def main():
options, args = get_parsed_args()
agentConfig = get_config(options=options)
autorestart = agentConfig.get('autorestart', False)
COMMANDS = [
'start',
'stop',
'restart',
'foreground',
'status',
'info',
'check',
'configcheck',
'jmx',
]
if len(args) < 1:
sys.stderr.write("Usage: %s %s\n" % (sys.argv[0], "|".join(COMMANDS)))
return 2
command = args[0]
if command not in COMMANDS:
sys.stderr.write("Unknown command: %s\n" % command)
return 3
pid_file = PidFile('dd-agent')
if options.clean:
pid_file.clean()
agent = Agent(pid_file.get_path(), autorestart)
if command in START_COMMANDS:
log.info('Agent version %s' % get_version())
if 'start' == command:
log.info('Start daemon')
agent.start()
elif 'stop' == command:
log.info('Stop daemon')
agent.stop()
elif 'restart' == command:
log.info('Restart daemon')
agent.restart()
elif 'status' == command:
agent.status()
elif 'info' == command:
return agent.info(verbose=options.verbose)
elif 'foreground' == command:
logging.info('Running in foreground')
if autorestart:
# Set-up the supervisor callbacks and fork it.
logging.info('Running Agent with auto-restart ON')
def child_func():
agent.run()
def parent_func():
agent.start_event = False
AgentSupervisor.start(parent_func, child_func)
else:
# Run in the standard foreground.
agent.run(config=agentConfig)
elif 'check' == command:
check_name = args[1]
try:
import checks.collector
# Try the old-style check first
print getattr(checks.collector, check_name)(log).check(agentConfig)
except Exception:
# If not an old-style check, try checks.d
checks = load_check_directory(agentConfig)
for check in checks['initialized_checks']:
if check.name == check_name:
check.run()
print check.get_metrics()
print check.get_events()
if len(args) == 3 and args[2] == 'check_rate':
print "Running 2nd iteration to capture rate metrics"
time.sleep(1)
check.run()
print check.get_metrics()
print check.get_events()
elif 'configcheck' == command or 'configtest' == command:
osname = get_os()
all_valid = True
for conf_path in glob.glob(
os.path.join(get_confd_path(osname), "*.yaml")):
basename = os.path.basename(conf_path)
try:
check_yaml(conf_path)
except Exception, e:
all_valid = False
print "%s contains errors:\n %s" % (basename, e)
else:
print "%s is valid" % basename
if all_valid:
print "All yaml files passed. You can now run the Datadog agent."
return 0
else:
print(
"Fix the invalid yaml files above in order to start the Datadog agent. "
"A useful external tool for yaml parsing can be found at "
"http://yaml-online-parser.appspot.com/")
return 1
