def test_persistence():
i1 = InstanceStatus(1, STATUS_OK)
chk1 = CheckStatus("dummy", [i1], 1, 2)
c1 = CollectorStatus([chk1])
c1.persist()
c2 = CollectorStatus.load_latest_status()
nt.assert_equal(1, len(c2.check_statuses))
chk2 = c2.check_statuses[0]
assert chk2.name == chk1.name
assert chk2.status == chk2.status
assert chk2.metric_count == 1
assert chk2.event_count == 2
def run(self):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
agentConfig = self._set_agent_config_hostname(get_config())
systemStats = get_system_stats()
emitters = self._get_emitters(agentConfig)
self.collector = Collector(agentConfig, emitters, systemStats)
# Load the checks.d checks
checksd = load_check_directory(agentConfig)
# Configure the watchdog.
check_frequency = int(agentConfig['check_freq'])
watchdog = self._get_watchdog(check_frequency, agentConfig)
# Run the main loop.
while self.run_forever:
# Do the work.
self.collector.run(checksd=checksd)
# Only plan for the next loop if we will continue,
# otherwise just exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
time.sleep(check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except:
pass
# Explicitly kill the process, because it might be running
# as a daemon.
agent_logger.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, config=None):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
if not Platform.is_windows():
# A SIGUSR1 signals an exit with an autorestart
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
# Handle Keyboard Interrupt
signal.signal(signal.SIGINT, self._handle_sigterm)
# A SIGHUP signals a configuration reload
signal.signal(signal.SIGHUP, self._handle_sighup)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
if not config:
config = get_config(parse_args=True)
self._agentConfig = self._set_agent_config_hostname(config)
hostname = get_hostname(self._agentConfig)
systemStats = get_system_stats(proc_path=self._agentConfig.get(
'procfs_path', '/proc').rstrip('/'))
emitters = self._get_emitters()
# Initialize service discovery
if self._agentConfig.get('service_discovery'):
self.sd_backend = get_sd_backend(self._agentConfig)
if _is_affirmative(self._agentConfig.get('sd_jmx_enable', False)):
pipe_path = get_jmx_pipe_path()
if Platform.is_windows():
pipe_name = pipe_path.format(pipename=SD_PIPE_NAME)
else:
pipe_name = os.path.join(pipe_path, SD_PIPE_NAME)
if os.access(pipe_path, os.W_OK):
if not os.path.exists(pipe_name):
os.mkfifo(pipe_name)
self.sd_pipe = os.open(
pipe_name, os.O_RDWR) # RW to avoid blocking (will only W)
# Initialize Supervisor proxy
self.supervisor_proxy = self._get_supervisor_socket(
self._agentConfig)
else:
log.debug(
'Unable to create pipe in temporary directory. JMX service discovery disabled.'
)
# Load the checks.d checks
self._checksd = load_check_directory(self._agentConfig, hostname)
# Load JMX configs if available
if self._jmx_service_discovery_enabled:
jmx_sd_configs = generate_jmx_configs(self._agentConfig, hostname)
if jmx_sd_configs:
self._submit_jmx_service_discovery(jmx_sd_configs)
# Initialize the Collector
self.collector = Collector(self._agentConfig, emitters, systemStats,
hostname)
# In developer mode, the number of runs to be included in a single collector profile
try:
self.collector_profile_interval = int(
self._agentConfig.get('collector_profile_interval',
DEFAULT_COLLECTOR_PROFILE_INTERVAL))
except ValueError:
log.warn('collector_profile_interval is invalid. '
'Using default value instead (%s).' %
DEFAULT_COLLECTOR_PROFILE_INTERVAL)
self.collector_profile_interval = DEFAULT_COLLECTOR_PROFILE_INTERVAL
# Configure the watchdog.
self.check_frequency = int(self._agentConfig['check_freq'])
watchdog = self._get_watchdog(self.check_frequency)
# Initialize the auto-restarter
self.restart_interval = int(
self._agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
self.allow_profiling = self._agentConfig.get('allow_profiling', True)
profiled = False
collector_profiled_runs = 0
# Run the main loop.
while self.run_forever:
# Setup profiling if necessary
if self.allow_profiling and self.in_developer_mode and not profiled:
try:
profiler = AgentProfiler()
profiler.enable_profiling()
profiled = True
except Exception as e:
log.warn("Cannot enable profiler: %s" % str(e))
if self.reload_configs_flag:
if isinstance(self.reload_configs_flag, set):
self.reload_configs(
checks_to_reload=self.reload_configs_flag)
else:
self.reload_configs()
# Do the work. Pass `configs_reloaded` to let the collector know if it needs to
# look for the AgentMetrics check and pop it out.
self.collector.run(
checksd=self._checksd,
start_event=self.start_event,
configs_reloaded=True if self.reload_configs_flag else False)
self.reload_configs_flag = False
# Look for change in the config template store.
# The self.sd_backend.reload_check_configs flag is set
# to True if a config reload is needed.
if self._agentConfig.get('service_discovery') and self.sd_backend and \
not self.sd_backend.reload_check_configs:
try:
self.sd_backend.reload_check_configs = get_config_store(
self._agentConfig).crawl_config_template()
except Exception as e:
log.warn(
'Something went wrong while looking for config template changes: %s'
% str(e))
# Check if we should run service discovery
# The `reload_check_configs` flag can be set through the docker_daemon check or
# using ConfigStore.crawl_config_template
if self._agentConfig.get('service_discovery') and self.sd_backend and \
self.sd_backend.reload_check_configs:
self.reload_configs_flag = self.sd_backend.reload_check_configs
self.sd_backend.reload_check_configs = False
if profiled:
if collector_profiled_runs >= self.collector_profile_interval:
try:
profiler.disable_profiling()
profiled = False
collector_profiled_runs = 0
except Exception as e:
log.warn("Cannot disable profiler: %s" % str(e))
# Check if we should restart.
if self.autorestart and self._should_restart():
self._do_restart()
# Only plan for next loop if we will continue, otherwise exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
if profiled:
collector_profiled_runs += 1
log.debug("Sleeping for {0} seconds".format(
self.check_frequency))
time.sleep(self.check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except Exception:
pass
# Explicitly kill the process, because it might be running as a daemon.
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, checksd=None, start_event=True, configs_reloaded=False):
"""
Collect data from each check and submit their data.
"""
log.debug("Found {num_checks} checks".format(num_checks=len(checksd['initialized_checks'])))
timer = Timer()
if not Platform.is_windows():
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
if checksd:
self.initialized_checks_d = checksd['initialized_checks'] # is a list of AgentCheck instances
self.init_failed_checks_d = checksd['init_failed_checks'] # is of type {check_name: {error, traceback}}
payload = AgentPayload()
# Find the AgentMetrics check and pop it out
# This check must run at the end of the loop to collect info on agent performance
if not self._agent_metrics or configs_reloaded:
for check in self.initialized_checks_d:
if check.name == AGENT_METRICS_CHECK_NAME:
self._agent_metrics = check
self.initialized_checks_d.remove(check)
break
# Initialize payload
self._build_payload(payload)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
# Run the system checks. Checks will depend on the OS
if Platform.is_windows():
# Win32 system checks
try:
metrics.extend(self._win32_system_checks['memory'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['cpu'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['network'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['io'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['proc'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['system'].check(self.agentConfig))
except Exception:
log.exception('Unable to fetch Windows system metrics.')
else:
# Unix system checks
sys_checks = self._unix_system_checks
load = sys_checks['load'].check(self.agentConfig)
payload.update(load)
system = sys_checks['system'].check(self.agentConfig)
payload.update(system)
memory = sys_checks['memory'].check(self.agentConfig)
if memory:
memstats = {
'memPhysUsed': memory.get('physUsed'),
'memPhysPctUsable': memory.get('physPctUsable'),
'memPhysFree': memory.get('physFree'),
'memPhysTotal': memory.get('physTotal'),
'memPhysUsable': memory.get('physUsable'),
'memSwapUsed': memory.get('swapUsed'),
'memSwapFree': memory.get('swapFree'),
'memSwapPctFree': memory.get('swapPctFree'),
'memSwapTotal': memory.get('swapTotal'),
'memCached': memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared'),
'memSlab': memory.get('physSlab'),
'memPageTables': memory.get('physPageTables'),
'memSwapCached': memory.get('swapCached')
}
payload.update(memstats)
ioStats = sys_checks['io'].check(self.agentConfig)
if ioStats:
payload['ioStats'] = ioStats
processes = sys_checks['processes'].check(self.agentConfig)
payload.update({'processes': processes})
cpuStats = sys_checks['cpu'].check(self.agentConfig)
if cpuStats:
payload.update(cpuStats)
# Run old-style checks
gangliaData = self._ganglia.check(self.agentConfig)
dogstreamData = self._dogstream.check(self.agentConfig)
ddforwarderData = self._ddforwarder.check(self.agentConfig)
if gangliaData is not False and gangliaData is not None:
payload['ganglia'] = gangliaData
# dogstream
if dogstreamData:
dogstreamEvents = dogstreamData.get('dogstreamEvents', None)
if dogstreamEvents:
if 'dogstream' in payload['events']:
events['dogstream'].extend(dogstreamEvents)
else:
events['dogstream'] = dogstreamEvents
del dogstreamData['dogstreamEvents']
payload.update(dogstreamData)
# metrics about the forwarder
if ddforwarderData:
payload['datadog'] = ddforwarderData
# process collector of gohai (compliant with payload of legacy "resources checks")
if not Platform.is_windows() and self._should_send_additional_data('processes'):
gohai_processes = self._run_gohai_processes()
if gohai_processes:
try:
gohai_processes_json = json.loads(gohai_processes)
processes_payload = {
'snaps': [gohai_processes_json.get('processes')],
'format_version': 1
}
if self._is_first_run():
processes_payload['format_description'] = PROCESSES_FORMAT_DESCRIPTION
payload['resources'] = {
'processes': processes_payload,
'meta': {
'host': payload['internalHostname'],
}
}
except Exception:
log.exception("Error running gohai processes collection")
# newer-style checks (not checks.d style)
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
# checks.d checks
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log.info("Running check %s" % check.name)
instance_statuses = []
metric_count = 0
event_count = 0
service_check_count = 0
check_start_time = time.time()
check_stats = None
try:
# Run the check.
instance_statuses = check.run()
# Collect the metrics and events.
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
check_stats = check._get_internal_profiling_stats()
# Collect metadata
current_check_metadata = check.get_service_metadata()
# Save metrics & events for the payload.
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
# Save the status of the check.
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(
check.name, instance_statuses, metric_count,
event_count, service_check_count, service_metadata=current_check_metadata,
library_versions=check.get_library_info(),
source_type_name=check.SOURCE_TYPE_NAME or check.name,
check_stats=check_stats
)
# Service check for Agent checks failures
service_check_tags = ["check:%s" % check.name]
if check_status.status == STATUS_OK:
status = AgentCheck.OK
elif check_status.status == STATUS_ERROR:
status = AgentCheck.CRITICAL
check.service_check('datadog.agent.check_status', status, tags=service_check_tags)
# Collect the service checks and save them in the payload
current_check_service_checks = check.get_service_checks()
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
service_check_count = len(current_check_service_checks)
# Update the check status with the correct service_check_count
check_status.service_check_count = service_check_count
check_statuses.append(check_status)
check_run_time = time.time() - check_start_time
log.debug("Check %s ran in %.2f s" % (check.name, check_run_time))
# Intrument check run timings if enabled.
if self.check_timings:
metric = 'datadog.agent.check_run_time'
meta = {'tags': ["check:%s" % check.name]}
metrics.append((metric, time.time(), check_run_time, meta))
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name, None, None, None, None,
init_failed_error=info['error'],
init_failed_traceback=info['traceback'])
check_statuses.append(check_status)
# Add a service check for the agent
service_checks.append(create_service_check('datadog.agent.up', AgentCheck.OK,
hostname=self.hostname))
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
# Populate metadata
self._populate_payload_metadata(payload, check_statuses, start_event)
collect_duration = timer.step()
if self._agent_metrics:
metric_context = {
'collection_time': collect_duration,
'emit_time': self.emit_duration,
}
if not Platform.is_windows():
metric_context['cpu_time'] = time.clock() - cpu_clock
self._agent_metrics.set_metric_context(payload, metric_context)
self._agent_metrics.run()
agent_stats = self._agent_metrics.get_metrics()
payload['metrics'].extend(agent_stats)
if self.agentConfig.get('developer_mode'):
log.debug("\n Agent developer mode stats: \n {0}".format(
Collector._stats_for_display(agent_stats))
)
# Flush metadata for the Agent Metrics check. Otherwise they'll just accumulate and leak.
self._agent_metrics.get_service_metadata()
# Let's send our payload
emitter_statuses = payload.emit(log, self.agentConfig, self.emitters,
self.continue_running)
self.emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses,
self.hostname_metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info("First flushes done, next flushes will be logged every %s flushes." %
FLUSH_LOGGING_PERIOD)
else:
log.debug("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
return payload
def run(self, config=None):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
# A SIGUSR1 signals an exit with an autorestart
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
# Handle Keyboard Interrupt
signal.signal(signal.SIGINT, self._handle_sigterm)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
if not config:
config = get_config(parse_args=True)
agentConfig = self._set_agent_config_hostname(config)
systemStats = get_system_stats()
emitters = self._get_emitters(agentConfig)
# Load the checks.d checks
checksd = load_check_directory(agentConfig)
self.collector = Collector(agentConfig, emitters, systemStats)
# Configure the watchdog.
check_frequency = int(agentConfig['check_freq'])
watchdog = self._get_watchdog(check_frequency, agentConfig)
# Initialize the auto-restarter
self.restart_interval = int(
agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
# Run the main loop.
while self.run_forever:
# enable profiler if needed
profiled = False
if agentConfig.get(
'profile',
False) and agentConfig.get('profile').lower() == 'yes':
try:
import cProfile
profiler = cProfile.Profile()
profiled = True
profiler.enable()
log.debug("Agent profiling is enabled")
except Exception:
log.warn("Cannot enable profiler")
# Do the work.
self.collector.run(checksd=checksd, start_event=self.start_event)
# disable profiler and printout stats to stdout
if agentConfig.get('profile', False) and agentConfig.get(
'profile').lower() == 'yes' and profiled:
try:
profiler.disable()
import pstats
from cStringIO import StringIO
s = StringIO()
ps = pstats.Stats(profiler,
stream=s).sort_stats("cumulative")
ps.print_stats()
log.debug(s.getvalue())
except Exception:
log.warn("Cannot disable profiler")
# Check if we should restart.
if self.autorestart and self._should_restart():
self._do_restart()
# Only plan for the next loop if we will continue,
# otherwise just exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
time.sleep(check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except Exception:
pass
# Explicitly kill the process, because it might be running
# as a daemon.
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, checksd=None, start_event=True):
"""
Collect data from each check and submit their data.
"""
timer = Timer()
if self.os != 'windows':
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
payload = self._build_payload(start_event=start_event)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
if checksd:
self.initialized_checks_d = checksd['initialized_checks'] # is of type {check_name: check}
self.init_failed_checks_d = checksd['init_failed_checks'] # is of type {check_name: {error, traceback}}
# Run the system checks. Checks will depend on the OS
if self.os == 'windows':
# Win32 system checks
try:
metrics.extend(self._win32_system_checks['disk'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['memory'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['cpu'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['network'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['io'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['proc'].check(self.agentConfig))
except Exception:
log.exception('Unable to fetch Windows system metrics.')
else:
# Unix system checks
sys_checks = self._unix_system_checks
diskUsage = sys_checks['disk'].check(self.agentConfig)
if diskUsage and len(diskUsage) == 2:
payload["diskUsage"] = diskUsage[0]
payload["inodes"] = diskUsage[1]
load = sys_checks['load'].check(self.agentConfig)
payload.update(load)
memory = sys_checks['memory'].check(self.agentConfig)
if memory:
payload.update({
'memPhysUsed' : memory.get('physUsed'),
'memPhysPctUsable' : memory.get('physPctUsable'),
'memPhysFree' : memory.get('physFree'),
'memPhysTotal' : memory.get('physTotal'),
'memPhysUsable' : memory.get('physUsable'),
'memSwapUsed' : memory.get('swapUsed'),
'memSwapFree' : memory.get('swapFree'),
'memSwapPctFree' : memory.get('swapPctFree'),
'memSwapTotal' : memory.get('swapTotal'),
'memCached' : memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared')
})
ioStats = sys_checks['io'].check(self.agentConfig)
if ioStats:
payload['ioStats'] = ioStats
processes = sys_checks['processes'].check(self.agentConfig)
payload.update({'processes': processes})
cpuStats = sys_checks['cpu'].check(self.agentConfig)
if cpuStats:
payload.update(cpuStats)
# Run old-style checks
gangliaData = self._ganglia.check(self.agentConfig)
dogstreamData = self._dogstream.check(self.agentConfig)
ddforwarderData = self._ddforwarder.check(self.agentConfig)
if gangliaData is not False and gangliaData is not None:
payload['ganglia'] = gangliaData
# dogstream
if dogstreamData:
dogstreamEvents = dogstreamData.get('dogstreamEvents', None)
if dogstreamEvents:
if 'dogstream' in payload['events']:
events['dogstream'].extend(dogstreamEvents)
else:
events['dogstream'] = dogstreamEvents
del dogstreamData['dogstreamEvents']
payload.update(dogstreamData)
# metrics about the forwarder
if ddforwarderData:
payload['datadog'] = ddforwarderData
# Process the event checks.
for event_check in self._event_checks:
event_data = event_check.check(log, self.agentConfig)
if event_data:
events[event_check.key] = event_data
# Resources checks
if self.os != 'windows':
has_resource = False
for resources_check in self._resources_checks:
resources_check.check()
snaps = resources_check.pop_snapshots()
if snaps:
has_resource = True
res_value = { 'snaps': snaps,
'format_version': resources_check.get_format_version() }
res_format = resources_check.describe_format_if_needed()
if res_format is not None:
res_value['format_description'] = res_format
payload['resources'][resources_check.RESOURCE_KEY] = res_value
if has_resource:
payload['resources']['meta'] = {
'api_key': self.agentConfig['api_key'],
'host': payload['internalHostname'],
}
# newer-style checks (not checks.d style)
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
# checks.d checks
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log.info("Running check %s" % check.name)
instance_statuses = []
metric_count = 0
event_count = 0
try:
# Run the check.
instance_statuses = check.run()
# Collect the metrics and events.
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
current_check_service_checks = check.get_service_checks()
# Save them for the payload.
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
# Save the status of the check.
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
service_check_count = len(current_check_service_checks)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(check.name, instance_statuses, metric_count, event_count, service_check_count,
library_versions=check.get_library_info())
check_statuses.append(check_status)
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name, None, None, None, None,
init_failed_error=info['error'],
init_failed_traceback=info['traceback'])
check_statuses.append(check_status)
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
collect_duration = timer.step()
if self.os != 'windows':
payload['metrics'].extend(self._agent_metrics.check(payload, self.agentConfig,
collect_duration, self.emit_duration, time.clock() - cpu_clock))
else:
payload['metrics'].extend(self._agent_metrics.check(payload, self.agentConfig,
collect_duration, self.emit_duration))
emitter_statuses = self._emit(payload)
self.emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses, self.metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info("First flushes done, next flushes will be logged every %s flushes." % FLUSH_LOGGING_PERIOD)
else:
log.debug("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
payload['metrics'].extend(
self._agent_metrics.check(payload, self.agentConfig,
collect_duration, self.emit_duration,
time.clock() - cpu_clock))
else:
payload['metrics'].extend(
self._agent_metrics.check(payload, self.agentConfig,
collect_duration,
self.emit_duration))
emitter_statuses = self._emit(payload)
self.emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses,
self.metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info(
"First flushes done, next flushes will be logged every %s flushes."
% FLUSH_LOGGING_PERIOD)
else:
log.debug(
"Finished run #%s. Collection time: %ss. Emit time: %ss" %
def run(self, checksd=None, start_event=True, configs_reloaded=False):
"""
Collect data from each check and submit their data.
"""
log.debug("Found {num_checks} checks".format(
num_checks=len(checksd['initialized_checks'])))
timer = Timer()
if not Platform.is_windows():
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
if checksd:
self.initialized_checks_d = checksd[
'initialized_checks'] # is a list of AgentCheck instances
self.init_failed_checks_d = checksd[
'init_failed_checks'] # is of type {check_name: {error, traceback}}
payload = AgentPayload()
# Find the AgentMetrics check and pop it out
# This check must run at the end of the loop to collect info on agent performance
if not self._agent_metrics or configs_reloaded:
for check in self.initialized_checks_d:
if check.name == AGENT_METRICS_CHECK_NAME:
self._agent_metrics = check
self.initialized_checks_d.remove(check)
break
# Initialize payload
self._build_payload(payload)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
# Run the system checks. Checks will depend on the OS
if Platform.is_windows():
# Win32 system checks
for check_name in ['memory', 'cpu', 'io', 'proc', 'system']:
try:
metrics.extend(self._win32_system_checks[check_name].check(
self.agentConfig))
except Exception:
log.exception('Unable to get %s metrics', check_name)
else:
# Unix system checks
sys_checks = self._unix_system_checks
for check_name in ['load', 'system', 'cpu', 'file_handles']:
try:
result_check = sys_checks[check_name].check(
self.agentConfig)
if result_check:
payload.update(result_check)
except Exception:
log.exception('Unable to get %s metrics', check_name)
try:
memory = sys_checks['memory'].check(self.agentConfig)
except Exception:
log.exception('Unable to get memory metrics')
else:
if memory:
memstats = {
'memPhysUsed': memory.get('physUsed'),
'memPhysPctUsable': memory.get('physPctUsable'),
'memPhysFree': memory.get('physFree'),
'memPhysTotal': memory.get('physTotal'),
'memPhysUsable': memory.get('physUsable'),
'memSwapUsed': memory.get('swapUsed'),
'memSwapFree': memory.get('swapFree'),
'memSwapPctFree': memory.get('swapPctFree'),
'memSwapTotal': memory.get('swapTotal'),
'memCached': memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared'),
'memSlab': memory.get('physSlab'),
'memPageTables': memory.get('physPageTables'),
'memSwapCached': memory.get('swapCached')
}
payload.update(memstats)
try:
ioStats = sys_checks['io'].check(self.agentConfig)
except Exception:
log.exception('Unable to get io metrics')
else:
if ioStats:
payload['ioStats'] = ioStats
try:
processes = sys_checks['processes'].check(self.agentConfig)
except Exception:
log.exception('Unable to get processes metrics')
else:
payload.update({'processes': processes})
# Run old-style checks
if self._ganglia is not None:
payload['ganglia'] = self._ganglia.check(self.agentConfig)
if self._dogstream is not None:
# every 10 run (~2min) we reload the list of files watched by
# dogstream
if (self.run_count % 10) == 0:
log.info("reloading list of files watched by Dogstreams")
self._dogstream = Dogstreams.init(log, self.agentConfig)
dogstreamData = self._dogstream.check(self.agentConfig)
dogstreamEvents = dogstreamData.get('dogstreamEvents', None)
if dogstreamEvents:
if 'dogstream' in payload['events']:
events['dogstream'].extend(dogstreamEvents)
else:
events['dogstream'] = dogstreamEvents
del dogstreamData['dogstreamEvents']
payload.update(dogstreamData)
# process collector of gohai (compliant with payload of legacy "resources checks")
if not Platform.is_windows() and self._should_send_additional_data(
'processes'):
gohai_processes = self._run_gohai_processes()
if gohai_processes:
try:
gohai_processes_json = json.loads(gohai_processes)
processes_snaps = gohai_processes_json.get('processes')
if processes_snaps:
processes_payload = {'snaps': [processes_snaps]}
payload['resources'] = {
'processes': processes_payload,
'meta': {
'host': self.hostname,
}
}
except Exception:
log.exception("Error running gohai processes collection")
# newer-style checks (not checks.d style)
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
# Use `info` log level for some messages on the first run only, then `debug`
log_at_first_run = log.info if self._is_first_run() else log.debug
# checks.d checks
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log_at_first_run("Running check %s", check.name)
instance_statuses = []
metric_count = 0
event_count = 0
service_check_count = 0
check_start_time = time.time()
check_stats = None
try:
# Run the check.
instance_statuses = check.run()
# Collect the metrics and events.
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
check_stats = check._get_internal_profiling_stats()
# Collect metadata
current_check_metadata = check.get_service_metadata()
# Save metrics & events for the payload.
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
# Save the status of the check.
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(
check.name,
instance_statuses,
metric_count,
event_count,
service_check_count,
service_metadata=current_check_metadata,
library_versions=check.get_library_info(),
source_type_name=check.SOURCE_TYPE_NAME or check.name,
check_stats=check_stats,
check_version=check.check_version)
# Service check for Agent checks failures
service_check_tags = ["check:%s" % check.name]
if check_status.status == STATUS_OK:
status = AgentCheck.OK
elif check_status.status == STATUS_ERROR:
status = AgentCheck.CRITICAL
check.service_check('datadog.agent.check_status',
status,
tags=service_check_tags)
# Collect the service checks and save them in the payload
current_check_service_checks = check.get_service_checks()
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
# -1 because the user doesn't care about the service check for check failure
service_check_count = len(current_check_service_checks) - 1
# Update the check status with the correct service_check_count
check_status.service_check_count = service_check_count
check_statuses.append(check_status)
check_run_time = time.time() - check_start_time
log.debug("Check %s ran in %.2f s" % (check.name, check_run_time))
# Intrument check run timings if enabled.
if self.check_timings:
metric = 'datadog.agent.check_run_time'
meta = {'tags': ["check:%s" % check.name]}
metrics.append((metric, time.time(), check_run_time, meta))
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name,
None,
None,
None,
None,
check_version=info.get(
'version', 'unknown'),
init_failed_error=info['error'],
init_failed_traceback=info['traceback'])
check_statuses.append(check_status)
# Add a service check for the agent
service_checks.append(
create_service_check('datadog.agent.up',
AgentCheck.OK,
hostname=self.hostname))
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
# Populate metadata
self._populate_payload_metadata(payload, check_statuses, start_event)
collect_duration = timer.step()
if self._agent_metrics:
metric_context = {
'collection_time': collect_duration,
'emit_time': self.emit_duration,
}
if not Platform.is_windows():
metric_context['cpu_time'] = time.clock() - cpu_clock
self._agent_metrics.set_metric_context(payload, metric_context)
self._agent_metrics.run()
agent_stats = self._agent_metrics.get_metrics()
payload['metrics'].extend(agent_stats)
if self.agentConfig.get('developer_mode'):
log.debug("\n Agent developer mode stats: \n {0}".format(
Collector._stats_for_display(agent_stats)))
# Flush metadata for the Agent Metrics check. Otherwise they'll just accumulate and leak.
self._agent_metrics.get_service_metadata()
# Let's send our payload
emitter_statuses = payload.emit(log, self.agentConfig, self.emitters,
self.continue_running)
self.emit_duration = timer.step()
if self._is_first_run():
# This is not the exact payload sent to the backend as minor post
# processing is done, but this will give us a good idea of what is sent
# to the backend.
data = payload.payload # deep copy and merge of meta and metric data
data['apiKey'] = '*************************' + data.get(
'apiKey', '')[-5:]
# removing unused keys for the metadata payload
del data['metrics']
del data['events']
del data['service_checks']
if data.get('processes'):
data['processes'][
'apiKey'] = '*************************' + data[
'processes'].get('apiKey', '')[-5:]
log.debug("Metadata payload: %s", json.dumps(data))
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses,
self.hostname_metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info(
"First flushes done, next flushes will be logged every %s flushes."
% FLUSH_LOGGING_PERIOD)
else:
log.debug(
"Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(self.emit_duration, 2)))
return payload
def run(self, config=None):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
# A SIGUSR1 signals an exit with an autorestart
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
# Handle Keyboard Interrupt
signal.signal(signal.SIGINT, self._handle_sigterm)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
if not config:
config = get_config(parse_args=True)
agentConfig = self._set_agent_config_hostname(config)
systemStats = get_system_stats()
emitters = self._get_emitters(agentConfig)
# Load the checks.d checks
checksd = load_check_directory(agentConfig)
self.collector = Collector(agentConfig, emitters, systemStats)
# Configure the watchdog.
check_frequency = int(agentConfig['check_freq'])
watchdog = self._get_watchdog(check_frequency, agentConfig)
# Initialize the auto-restarter
self.restart_interval = int(
agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
# Run the main loop.
while self.run_forever:
# Do the work.
self.collector.run(checksd=checksd, start_event=self.start_event)
# Check if we should restart.
if self.autorestart and self._should_restart():
self._do_restart()
# Only plan for the next loop if we will continue,
# otherwise just exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
time.sleep(check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except:
pass
# Explicitly kill the process, because it might be running
# as a daemon.
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, config=None):
signal.signal(signal.SIGTERM, self._handle_sigterm)
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
signal.signal(signal.SIGINT, self._handle_sigterm)
signal.signal(signal.SIGHUP, self._handle_sighup)
CollectorStatus().persist()
if not config:
config = get_config(parse_args=True)
self._agentConfig = self._set_agent_config_hostname(config)
hostname = get_hostname(self._agentConfig)
systemStats = get_system_stats()
emitters = self._get_emitters()
self._checksd = load_check_directory(self._agentConfig, hostname)
self.collector = Collector(self._agentConfig, emitters, systemStats, hostname)
self.collector_profile_interval = self._agentConfig.get('collector_profile_interval',
DEFAULT_COLLECTOR_PROFILE_INTERVAL)
self.check_frequency = int(self._agentConfig['check_freq'])
watchmonitor = self._get_watchmonitor(self.check_frequency)
self.restart_interval = int(self._agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
profiled = False
collector_profiled_runs = 0
while self.run_forever:
log.debug("Found {num_checks} checks".format(num_checks=len(self._checksd['initialized_checks'])))
if self.in_developer_mode and not profiled:
try:
profiler = AgentProfiler()
profiler.enable_profiling()
profiled = True
except Exception as e:
log.warn("Cannot enable profiler: %s" % str(e))
self.collector.run(checksd=self._checksd,
start_event=self.start_event,
configs_reloaded=self.configs_reloaded)
if self.configs_reloaded:
self.configs_reloaded = False
if profiled:
if collector_profiled_runs >= self.collector_profile_interval:
try:
profiler.disable_profiling()
profiled = False
collector_profiled_runs = 0
except Exception as e:
log.warn("Cannot disable profiler: %s" % str(e))
if self.autorestart and self._should_restart():
self._do_restart()
if self.run_forever:
if watchmonitor:
watchmonitor.reset()
if profiled:
collector_profiled_runs += 1
log.debug("Sleeping for {0} seconds".format(self.check_frequency))
time.sleep(self.check_frequency)
try:
CollectorStatus.remove_latest_status()
except Exception:
pass
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, checksd=None, start_event=True, configs_reloaded=False):
log.debug("Found {num_checks} checks".format(num_checks=len(checksd['initialized_checks'])))
timer = Timer()
if not Platform.is_windows():
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
if checksd:
self.initialized_checks_d = checksd['initialized_checks']
self.init_failed_checks_d = checksd['init_failed_checks']
payload = AgentPayload()
if not self._agent_metrics or configs_reloaded:
for check in self.initialized_checks_d:
if check.name == AGENT_METRICS_CHECK_NAME:
self._agent_metrics = check
self.initialized_checks_d.remove(check)
break
self._build_payload(payload)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
if Platform.is_windows():
try:
metrics.extend(self._win32_system_checks['memory'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['cpu'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['network'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['io'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['proc'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['system'].check(self.agentConfig))
except Exception:
log.exception('Unable to fetch Windows system metrics.')
else:
sys_checks = self._unix_system_checks
load = sys_checks['load'].check(self.agentConfig)
payload.update(load)
system = sys_checks['system'].check(self.agentConfig)
payload.update(system)
memory = sys_checks['memory'].check(self.agentConfig)
if memory:
memstats = {
'memPhysUsed': memory.get('physUsed'),
'memPhysPctUsage': memory.get('physPctUsage'),
'memPhysFree': memory.get('physFree'),
'memPhysTotal': memory.get('physTotal'),
'memPhysUsable': memory.get('physUsable'),
'memSwapUsed': memory.get('swapUsed'),
'memSwapFree': memory.get('swapFree'),
'memSwapPctFree': memory.get('swapPctFree'),
'memSwapTotal': memory.get('swapTotal'),
'memCached': memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared'),
'memSlab': memory.get('physSlab'),
'memPageTables': memory.get('physPageTables'),
'memSwapCached': memory.get('swapCached')
}
payload.update(memstats)
ioStats = sys_checks['io'].check(self.agentConfig)
if ioStats:
payload['ioStats'] = ioStats
processes = sys_checks['processes'].check(self.agentConfig)
payload.update({'processes': processes})
cpuStats = sys_checks['cpu'].check(self.agentConfig)
if cpuStats:
payload.update(cpuStats)
gangliaData = self._ganglia.check(self.agentConfig)
monitorstreamData = self._monitorstream.check(self.agentConfig)
ddforwarderData = self._ddforwarder.check(self.agentConfig)
if gangliaData is not False and gangliaData is not None:
payload['ganglia'] = gangliaData
if monitorstreamData:
monitorstreamEvents = monitorstreamData.get('monitorstreamEvents', None)
if monitorstreamEvents:
if 'monitorstream' in payload['events']:
events['monitorstream'].extend(monitorstreamEvents)
else:
events['monitorstream'] = monitorstreamEvents
del monitorstreamData['monitorstreamEvents']
payload.update(monitorstreamData)
if ddforwarderData:
payload['datamonitor'] = ddforwarderData
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log.info("Running check %s" % check.name)
instance_statuses = []
metric_count = 0
event_count = 0
service_check_count = 0
check_start_time = time.time()
check_stats = None
try:
instance_statuses = check.run()
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
check_stats = check._get_internal_profiling_stats()
current_check_metadata = check.get_service_metadata()
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(
check.name, instance_statuses, metric_count,
event_count, service_check_count, service_metadata=current_check_metadata,
library_versions=check.get_library_info(),
source_type_name=check.SOURCE_TYPE_NAME or check.name,
check_stats=check_stats
)
service_check_tags = ["check:%s" % check.name]
if check_status.status == STATUS_OK:
status = AgentCheck.OK
elif check_status.status == STATUS_ERROR:
status = AgentCheck.CRITICAL
check.service_check('datamonitor.agent.check_status', status, tags=service_check_tags)
current_check_service_checks = check.get_service_checks()
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
service_check_count = len(current_check_service_checks)
check_status.service_check_count = service_check_count
check_statuses.append(check_status)
check_run_time = time.time() - check_start_time
log.debug("Check %s ran in %.2f s" % (check.name, check_run_time))
if self.check_timings:
metric = 'datamonitor.agent.check_run_time'
meta = {'tags': ["check:%s" % check.name]}
metrics.append((metric, time.time(), check_run_time, meta))
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name, None, None, None, None,
init_failed_error=info['error'],
init_failed_traceback=info['error'])
check_statuses.append(check_status)
service_checks.append(create_service_check('datamonitor.agent.up', AgentCheck.OK,
hostname=self.hostname))
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
self._populate_payload_metadata(payload, check_statuses, start_event)
collect_duration = timer.step()
if self._agent_metrics:
metric_context = {
'collection_time': collect_duration,
'emit_time': self.emit_duration,
}
if not Platform.is_windows():
metric_context['cpu_time'] = time.clock() - cpu_clock
self._agent_metrics.set_metric_context(payload, metric_context)
self._agent_metrics.run()
agent_stats = self._agent_metrics.get_metrics()
payload['metrics'].extend(agent_stats)
if self.agentConfig.get('developer_mode'):
log.debug("\n Agent developer mode stats: \n {0}".format(
Collector._stats_for_display(agent_stats))
)
self._agent_metrics.get_service_metadata()
emitter_statuses = payload.emit(log, self.agentConfig, self.emitters,
self.continue_running)
self.emit_duration = timer.step()
try:
CollectorStatus(check_statuses, emitter_statuses,
self.hostname_metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info("First flushes done, next flushes will be logged every %s flushes." %
FLUSH_LOGGING_PERIOD)
else:
log.debug("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
return payload
def run(self, config=None):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
# A SIGUSR1 signals an exit with an autorestart
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
# Handle Keyboard Interrupt
signal.signal(signal.SIGINT, self._handle_sigterm)
# A SIGHUP signals a configuration reload
signal.signal(signal.SIGHUP, self._handle_sighup)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
if not config:
config = get_config(parse_args=True)
self._agentConfig = self._set_agent_config_hostname(config)
hostname = get_hostname(self._agentConfig)
systemStats = get_system_stats(proc_path=self._agentConfig.get(
'procfs_path', '/proc').rstrip('/'))
emitters = self._get_emitters()
# Initialize service discovery
if self._agentConfig.get('service_discovery'):
self.sd_backend = get_sd_backend(self._agentConfig)
# Load the checks.d checks
self._checksd = load_check_directory(self._agentConfig, hostname)
# Initialize the Collector
self.collector = Collector(self._agentConfig, emitters, systemStats,
hostname)
# In developer mode, the number of runs to be included in a single collector profile
self.collector_profile_interval = self._agentConfig.get(
'collector_profile_interval', DEFAULT_COLLECTOR_PROFILE_INTERVAL)
# Configure the watchdog.
self.check_frequency = int(self._agentConfig['check_freq'])
watchdog = self._get_watchdog(self.check_frequency)
# Initialize the auto-restarter
self.restart_interval = int(
self._agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
profiled = False
collector_profiled_runs = 0
# Run the main loop.
while self.run_forever:
log.debug("Found {num_checks} checks".format(
num_checks=len(self._checksd['initialized_checks'])))
# Setup profiling if necessary
if self.in_developer_mode and not profiled:
try:
profiler = AgentProfiler()
profiler.enable_profiling()
profiled = True
except Exception as e:
log.warn("Cannot enable profiler: %s" % str(e))
# Do the work.
self.collector.run(checksd=self._checksd,
start_event=self.start_event,
configs_reloaded=self.configs_reloaded)
# This flag is used to know if the check configs have been reloaded at the current
# run of the agent yet or not. It's used by the collector to know if it needs to
# look for the AgentMetrics check and pop it out.
# See: https://github.com/DataDog/dd-agent/blob/5.6.x/checks/collector.py#L265-L272
self.configs_reloaded = False
# Look for change in the config template store.
# The self.sd_backend.reload_check_configs flag is set
# to True if a config reload is needed.
if self._agentConfig.get('service_discovery') and self.sd_backend and \
not self.sd_backend.reload_check_configs:
try:
self.sd_backend.reload_check_configs = get_config_store(
self._agentConfig).crawl_config_template()
except Exception as e:
log.warn(
'Something went wrong while looking for config template changes: %s'
% str(e))
# Check if we should run service discovery
# The `reload_check_configs` flag can be set through the docker_daemon check or
# using ConfigStore.crawl_config_template
if self._agentConfig.get('service_discovery') and self.sd_backend and \
self.sd_backend.reload_check_configs:
self.reload_configs()
self.configs_reloaded = True
self.sd_backend.reload_check_configs = False
if profiled:
if collector_profiled_runs >= self.collector_profile_interval:
try:
profiler.disable_profiling()
profiled = False
collector_profiled_runs = 0
except Exception as e:
log.warn("Cannot disable profiler: %s" % str(e))
# Check if we should restart.
if self.autorestart and self._should_restart():
self._do_restart()
# Only plan for next loop if we will continue, otherwise exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
if profiled:
collector_profiled_runs += 1
log.debug("Sleeping for {0} seconds".format(
self.check_frequency))
time.sleep(self.check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except Exception:
pass
# Explicitly kill the process, because it might be running as a daemon.
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, config=None):
"""Main loop of the collector"""
# Gracefully exit on sigterm.
signal.signal(signal.SIGTERM, self._handle_sigterm)
# A SIGUSR1 signals an exit with an autorestart
signal.signal(signal.SIGUSR1, self._handle_sigusr1)
# Handle Keyboard Interrupt
signal.signal(signal.SIGINT, self._handle_sigterm)
# Save the agent start-up stats.
CollectorStatus().persist()
# Intialize the collector.
if not config:
config = get_config(parse_args=True)
agentConfig = self._set_agent_config_hostname(config)
hostname = get_hostname(agentConfig)
systemStats = get_system_stats()
emitters = self._get_emitters(agentConfig)
# Load the checks.d checks
checksd = load_check_directory(agentConfig, hostname)
self.collector = Collector(agentConfig, emitters, systemStats,
hostname)
# In developer mode, the number of runs to be included in a single collector profile
collector_profile_interval = agentConfig.get(
'collector_profile_interval', DEFAULT_COLLECTOR_PROFILE_INTERVAL)
# Configure the watchdog.
check_frequency = int(agentConfig['check_freq'])
watchdog = self._get_watchdog(check_frequency, agentConfig)
# Initialize the auto-restarter
self.restart_interval = int(
agentConfig.get('restart_interval', RESTART_INTERVAL))
self.agent_start = time.time()
profiled = False
collector_profiled_runs = 0
# Run the main loop.
while self.run_forever:
# Setup profiling if necessary
if self.in_developer_mode and not profiled:
try:
profiler = AgentProfiler()
profiler.enable_profiling()
profiled = True
except Exception as e:
log.warn("Cannot enable profiler: %s" % str(e))
# Do the work.
self.collector.run(checksd=checksd, start_event=self.start_event)
if profiled:
if collector_profiled_runs >= collector_profile_interval:
try:
profiler.disable_profiling()
profiled = False
collector_profiled_runs = 0
except Exception as e:
log.warn("Cannot disable profiler: %s" % str(e))
# Check if we should restart.
if self.autorestart and self._should_restart():
self._do_restart()
# Only plan for the next loop if we will continue,
# otherwise just exit quickly.
if self.run_forever:
if watchdog:
watchdog.reset()
if profiled:
collector_profiled_runs += 1
time.sleep(check_frequency)
# Now clean-up.
try:
CollectorStatus.remove_latest_status()
except Exception:
pass
# Explicitly kill the process, because it might be running
# as a daemon.
log.info("Exiting. Bye bye.")
sys.exit(0)
def run(self, checksd=None, start_event=True):
"""
Collect data from each check and submit their data.
"""
timer = Timer()
if self.os != 'windows':
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
payload = self._build_payload(start_event=start_event)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
if checksd:
self.initialized_checks_d = checksd[
'initialized_checks'] # is of type {check_name: check}
self.init_failed_checks_d = checksd[
'init_failed_checks'] # is of type {check_name: {error, traceback}}
# Run the system checks. Checks will depend on the OS
if self.os == 'windows':
# Win32 system checks
try:
metrics.extend(self._win32_system_checks['disk'].check(
self.agentConfig))
metrics.extend(self._win32_system_checks['memory'].check(
self.agentConfig))
metrics.extend(self._win32_system_checks['cpu'].check(
self.agentConfig))
metrics.extend(self._win32_system_checks['network'].check(
self.agentConfig))
metrics.extend(self._win32_system_checks['io'].check(
self.agentConfig))
metrics.extend(self._win32_system_checks['proc'].check(
self.agentConfig))
except Exception:
log.exception('Unable to fetch Windows system metrics.')
else:
# Unix system checks
sys_checks = self._unix_system_checks
diskUsage = sys_checks['disk'].check(self.agentConfig)
if diskUsage and len(diskUsage) == 2:
payload["diskUsage"] = diskUsage[0]
payload["inodes"] = diskUsage[1]
load = sys_checks['load'].check(self.agentConfig)
payload.update(load)
system = sys_checks['system'].check(self.agentConfig)
payload.update(system)
memory = sys_checks['memory'].check(self.agentConfig)
if memory:
payload.update({
'memPhysUsed': memory.get('physUsed'),
'memPhysPctUsable': memory.get('physPctUsable'),
'memPhysFree': memory.get('physFree'),
'memPhysTotal': memory.get('physTotal'),
'memPhysUsable': memory.get('physUsable'),
'memSwapUsed': memory.get('swapUsed'),
'memSwapFree': memory.get('swapFree'),
'memSwapPctFree': memory.get('swapPctFree'),
'memSwapTotal': memory.get('swapTotal'),
'memCached': memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared')
})
ioStats = sys_checks['io'].check(self.agentConfig)
if ioStats:
payload['ioStats'] = ioStats
processes = sys_checks['processes'].check(self.agentConfig)
payload.update({'processes': processes})
cpuStats = sys_checks['cpu'].check(self.agentConfig)
if cpuStats:
payload.update(cpuStats)
# Run old-style checks
gangliaData = self._ganglia.check(self.agentConfig)
dogstreamData = self._dogstream.check(self.agentConfig)
ddforwarderData = self._ddforwarder.check(self.agentConfig)
if gangliaData is not False and gangliaData is not None:
payload['ganglia'] = gangliaData
# dogstream
if dogstreamData:
dogstreamEvents = dogstreamData.get('dogstreamEvents', None)
if dogstreamEvents:
if 'dogstream' in payload['events']:
events['dogstream'].extend(dogstreamEvents)
else:
events['dogstream'] = dogstreamEvents
del dogstreamData['dogstreamEvents']
payload.update(dogstreamData)
# metrics about the forwarder
if ddforwarderData:
payload['datadog'] = ddforwarderData
# Resources checks
if self.os != 'windows':
has_resource = False
for resources_check in self._resources_checks:
resources_check.check()
snaps = resources_check.pop_snapshots()
if snaps:
has_resource = True
res_value = {
'snaps': snaps,
'format_version': resources_check.get_format_version()
}
res_format = resources_check.describe_format_if_needed()
if res_format is not None:
res_value['format_description'] = res_format
payload['resources'][
resources_check.RESOURCE_KEY] = res_value
if has_resource:
payload['resources']['meta'] = {
'api_key': self.agentConfig['api_key'],
'host': payload['internalHostname'],
}
# newer-style checks (not checks.d style)
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
# checks.d checks
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log.info("Running check %s" % check.name)
instance_statuses = []
metric_count = 0
event_count = 0
service_check_count = 0
check_start_time = time.time()
try:
# Run the check.
instance_statuses = check.run()
# Collect the metrics and events.
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
# Save them for the payload.
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
# Save the status of the check.
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(
check.name,
instance_statuses,
metric_count,
event_count,
service_check_count,
library_versions=check.get_library_info(),
source_type_name=check.SOURCE_TYPE_NAME or check.name)
# Service check for Agent checks failures
service_check_tags = ["check:%s" % check.name]
if check_status.status == STATUS_OK:
status = AgentCheck.OK
elif check_status.status == STATUS_ERROR:
status = AgentCheck.CRITICAL
check.service_check('datadog.agent.check_status',
status,
tags=service_check_tags)
# Collect the service checks and save them in the payload
current_check_service_checks = check.get_service_checks()
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
service_check_count = len(current_check_service_checks)
# Update the check status with the correct service_check_count
check_status.service_check_count = service_check_count
check_statuses.append(check_status)
check_run_time = time.time() - check_start_time
log.debug("Check %s ran in %.2f s" % (check.name, check_run_time))
# Intrument check run timings if enabled.
if self.check_timings:
metric = 'datadog.agent.check_run_time'
meta = {'tags': ["check:%s" % check.name]}
metrics.append((metric, time.time(), check_run_time, meta))
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name,
None,
None,
None,
None,
init_failed_error=info['error'],
init_failed_traceback=info['traceback'])
check_statuses.append(check_status)
# Add a service check for the agent
service_checks.append(
create_service_check('datadog.agent.up',
AgentCheck.OK,
hostname=self.hostname))
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
if self._should_send_additional_data('agent_checks'):
# Add agent checks statuses and error/warning messages
agent_checks = []
for check in check_statuses:
if check.instance_statuses is not None:
for instance_status in check.instance_statuses:
agent_checks.append((
check.name,
check.source_type_name,
instance_status.instance_id,
instance_status.status,
# put error message or list of warning messages in the same field
# it will be handled by the UI
instance_status.error or instance_status.warnings
or ""))
else:
agent_checks.append(
(check.name, check.source_type_name, "initialization",
check.status, repr(check.init_failed_error)))
payload['agent_checks'] = agent_checks
payload['meta'] = self.metadata_cache # add hostname metadata
collect_duration = timer.step()
if self.os != 'windows':
payload['metrics'].extend(
self._agent_metrics.check(payload, self.agentConfig,
collect_duration, self.emit_duration,
time.clock() - cpu_clock))
else:
payload['metrics'].extend(
self._agent_metrics.check(payload, self.agentConfig,
collect_duration,
self.emit_duration))
emitter_statuses = self._emit(payload)
self.emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses,
self.metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info(
"First flushes done, next flushes will be logged every %s flushes."
% FLUSH_LOGGING_PERIOD)
else:
log.debug(
"Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(self.emit_duration, 2)))
return payload
logger.exception("Error running check %s" % check.name)
check_status = CheckStatus(check.name, instance_statuses,
metric_count, event_count)
check_statuses.append(check_status)
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
collect_duration = timer.step()
emitter_statuses = self._emit(payload)
emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses).persist()
except Exception:
logger.exception("Error persisting collector status")
logger.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration,
2), round(emit_duration, 2)))
def _emit(self, payload):
""" Send the payload via the emitters. """
statuses = []
for emitter in self.emitters:
# Don't try to send to an emitter if we're stopping/
if not self.continue_running:
return statuses
name = emitter.__name__
def run(self, checksd=None, start_event=True):
"""
Collect data from each check and submit their data.
"""
timer = Timer()
if self.os != 'windows':
cpu_clock = time.clock()
self.run_count += 1
log.debug("Starting collection run #%s" % self.run_count)
if checksd:
self.initialized_checks_d = checksd['initialized_checks'] # is a list of AgentCheck instances
self.init_failed_checks_d = checksd['init_failed_checks'] # is of type {check_name: {error, traceback}}
payload = AgentPayload()
# Find the AgentMetrics check and pop it out
# This check must run at the end of the loop to collect info on agent performance
if not self._agent_metrics:
for check in self.initialized_checks_d:
if check.name == AGENT_METRICS_CHECK_NAME:
self._agent_metrics = check
self.initialized_checks_d.remove(check)
break
# Initialize payload
self._build_payload(payload)
metrics = payload['metrics']
events = payload['events']
service_checks = payload['service_checks']
# Run the system checks. Checks will depend on the OS
if self.os == 'windows':
# Win32 system checks
try:
metrics.extend(self._win32_system_checks['memory'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['cpu'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['network'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['io'].check(self.agentConfig))
metrics.extend(self._win32_system_checks['proc'].check(self.agentConfig))
except Exception:
log.exception('Unable to fetch Windows system metrics.')
else:
sd_checks = self._server_density_checks
identifier = sd_checks['identifier'].check(self.agentConfig)
payload.update(identifier)
# SDv1 plugins
pluginsData = sd_checks['plugins'].check(self.agentConfig)
if pluginsData:
payload['plugins'] = pluginsData
# Unix system checks
sys_checks = self._unix_system_checks
load = sys_checks['load'].check(self.agentConfig)
payload.update(load)
system = sys_checks['system'].check(self.agentConfig)
payload.update(system)
memory = sys_checks['memory'].check(self.agentConfig)
if memory:
payload.update({
'memPhysUsed': memory.get('physUsed'),
'memPhysPctUsable': memory.get('physPctUsable'),
'memPhysFree': memory.get('physFree'),
'memPhysTotal': memory.get('physTotal'),
'memPhysUsable': memory.get('physUsable'),
'memSwapUsed': memory.get('swapUsed'),
'memSwapFree': memory.get('swapFree'),
'memSwapPctFree': memory.get('swapPctFree'),
'memSwapTotal': memory.get('swapTotal'),
'memCached': memory.get('physCached'),
'memBuffers': memory.get('physBuffers'),
'memShared': memory.get('physShared')
})
ioStats = sys_checks['io'].check(self.agentConfig)
if ioStats:
payload['ioStats'] = ioStats
processes = sys_checks['processes'].check(self.agentConfig)
payload.update({'processes': processes})
cpuStats = sys_checks['cpu'].check(self.agentConfig)
if cpuStats:
payload.update(cpuStats)
# Run old-style checks
gangliaData = self._ganglia.check(self.agentConfig)
if gangliaData is not False and gangliaData is not None:
payload['ganglia'] = gangliaData
# Resources checks
if self.os != 'windows':
has_resource = False
for resources_check in self._resources_checks:
resources_check.check()
snaps = resources_check.pop_snapshots()
if snaps:
has_resource = True
res_value = {
'snaps': snaps,
'format_version': resources_check.get_format_version()
}
res_format = resources_check.describe_format_if_needed()
if res_format is not None:
res_value['format_description'] = res_format
payload['resources'][resources_check.RESOURCE_KEY] = res_value
if has_resource:
payload['resources']['meta'] = {
'agent_key': self.agentConfig['agent_key'],
'host': payload['internalHostname'],
}
# newer-style checks (not checks.d style)
for metrics_check in self._metrics_checks:
res = metrics_check.check(self.agentConfig)
if res:
metrics.extend(res)
# checks.d checks
check_statuses = []
for check in self.initialized_checks_d:
if not self.continue_running:
return
log.info("Running check %s" % check.name)
instance_statuses = []
metric_count = 0
event_count = 0
service_check_count = 0
check_start_time = time.time()
check_stats = None
try:
# Run the check.
instance_statuses = check.run()
# Collect the metrics and events.
current_check_metrics = check.get_metrics()
current_check_events = check.get_events()
check_stats = check._get_internal_profiling_stats()
# Collect metadata
current_check_metadata = check.get_service_metadata()
# Save metrics & events for the payload.
metrics.extend(current_check_metrics)
if current_check_events:
if check.name not in events:
events[check.name] = current_check_events
else:
events[check.name] += current_check_events
# Save the status of the check.
metric_count = len(current_check_metrics)
event_count = len(current_check_events)
except Exception:
log.exception("Error running check %s" % check.name)
check_status = CheckStatus(
check.name, instance_statuses, metric_count,
event_count, service_check_count, service_metadata=current_check_metadata,
library_versions=check.get_library_info(),
source_type_name=check.SOURCE_TYPE_NAME or check.name,
check_stats=check_stats
)
# Service check for Agent checks failures
service_check_tags = ["check:%s" % check.name]
if check_status.status == STATUS_OK:
status = AgentCheck.OK
elif check_status.status == STATUS_ERROR:
status = AgentCheck.CRITICAL
check.service_check('sd.agent.check_status', status, tags=service_check_tags)
# Collect the service checks and save them in the payload
current_check_service_checks = check.get_service_checks()
if current_check_service_checks:
service_checks.extend(current_check_service_checks)
service_check_count = len(current_check_service_checks)
# Update the check status with the correct service_check_count
check_status.service_check_count = service_check_count
check_statuses.append(check_status)
check_run_time = time.time() - check_start_time
log.debug("Check %s ran in %.2f s" % (check.name, check_run_time))
# Intrument check run timings if enabled.
if self.check_timings:
metric = 'sd.agent.check_run_time'
meta = {'tags': ["check:%s" % check.name]}
metrics.append((metric, time.time(), check_run_time, meta))
for check_name, info in self.init_failed_checks_d.iteritems():
if not self.continue_running:
return
check_status = CheckStatus(check_name, None, None, None, None,
init_failed_error=info['error'],
init_failed_traceback=info['traceback'])
check_statuses.append(check_status)
# Add a service check for the agent
service_checks.append(create_service_check('sd.agent.up', AgentCheck.OK,
hostname=self.hostname))
# Store the metrics and events in the payload.
payload['metrics'] = metrics
payload['events'] = events
payload['service_checks'] = service_checks
# Populate metadata
self._populate_payload_metadata(payload, check_statuses, start_event)
collect_duration = timer.step()
if self.os != 'windows':
if self._agent_metrics is not None:
self._agent_metrics.set_metric_context(payload,
{
'collection_time': collect_duration,
'emit_time': self.emit_duration,
'cpu_time': time.clock() - cpu_clock
})
self._agent_metrics.run()
agent_stats = self._agent_metrics.get_metrics()
payload['metrics'].extend(agent_stats)
# Dump the metrics to log when in developer mode
if self.agentConfig.get('developer_mode', False):
log.info("\n AGENT STATS: \n {0}".format(Collector._stats_for_display(agent_stats)))
else:
if self._agent_metrics is not None:
self._agent_metrics.set_metric_context(payload,
{
'collection_time': collect_duration,
'emit_time': self.emit_duration,
})
self._agent_metrics.run()
agent_stats = self._agent_metrics.get_metrics()
payload['metrics'].extend(agent_stats)
# Dump the metrics to log when in developer mode
if self.agentConfig.get('developer_mode', False):
log.info("\n AGENT STATS: \n {0}".format(Collector._stats_for_display(agent_stats)))
# Let's send our payload
emitter_statuses = payload.emit(log, self.agentConfig, self.emitters,
self.continue_running)
self.emit_duration = timer.step()
# Persist the status of the collection run.
try:
CollectorStatus(check_statuses, emitter_statuses,
self.hostname_metadata_cache).persist()
except Exception:
log.exception("Error persisting collector status")
if self.run_count <= FLUSH_LOGGING_INITIAL or self.run_count % FLUSH_LOGGING_PERIOD == 0:
log.info("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
if self.run_count == FLUSH_LOGGING_INITIAL:
log.info("First flushes done, next flushes will be logged every %s flushes." %
FLUSH_LOGGING_PERIOD)
else:
log.debug("Finished run #%s. Collection time: %ss. Emit time: %ss" %
(self.run_count, round(collect_duration, 2), round(self.emit_duration, 2)))
return payload