def busy_run():
w = Watchdog(5)
w.reset()
x = 0
while True:
x = random()
return x
def busy_run():
w = Watchdog(5)
w.reset()
x = 0
while True:
x = random()
return x
def __init__(self,
port,
agent_config,
watchdog=True,
skip_ssl_validation=False,
use_simple_http_client=False):
self._port = int(port)
self._agentConfig = agent_config
self._metrics = {}
MetricTransaction.set_application(self)
MetricTransaction.set_endpoints(MonAPI(agent_config['Api']))
self._tr_manager = TransactionManager(MAX_WAIT_FOR_REPLAY,
MAX_QUEUE_SIZE, THROTTLING_DELAY)
MetricTransaction.set_tr_manager(self._tr_manager)
self._watchdog = None
self.skip_ssl_validation = skip_ssl_validation or agent_config.get(
'skip_ssl_validation', False)
self.use_simple_http_client = use_simple_http_client
if self.skip_ssl_validation:
log.info(
"Skipping SSL hostname validation, useful when using a transparent proxy"
)
if watchdog:
watchdog_timeout = TRANSACTION_FLUSH_INTERVAL * WATCHDOG_INTERVAL_MULTIPLIER
self._watchdog = Watchdog(watchdog_timeout,
max_mem_mb=agent_config.get(
'limit_memory_consumption', None))
def _get_watchdog(check_freq, agentConfig):
watchdog = None
if agentConfig.get("watchdog", True):
watchdog = Watchdog(check_freq * WATCHDOG_MULTIPLIER,
max_mem_mb=agentConfig.get('limit_memory_consumption', None))
watchdog.reset()
return watchdog
def __init__(self, interval, aggregator, api_host, use_watchdog=False, event_chunk_size=None):
threading.Thread.__init__(self)
self.interval = int(interval)
self.finished = threading.Event()
self.aggregator = aggregator
self.flush_count = 0
self.log_count = 0
self.watchdog = None
if use_watchdog:
from monagent.common.util import Watchdog
self.watchdog = Watchdog(WATCHDOG_TIMEOUT)
self.api_host = api_host
self.event_chunk_size = event_chunk_size or EVENT_CHUNK_SIZE
def use_lots_of_memory():
# Skip this step on travis
if os.environ.get('TRAVIS', False):
return
a = CollectorDaemon(12345, {})
a._watchdog = Watchdog(30, 50)
a._tr_manager = MemoryHogTxManager()
a.run()
def __init__(self, port, agent_config, watchdog=True, skip_ssl_validation=False,
use_simple_http_client=False):
self._port = int(port)
self._agentConfig = agent_config
self._metrics = {}
MetricTransaction.set_application(self)
MetricTransaction.set_endpoints(MonAPI(agent_config['Api']))
self._tr_manager = TransactionManager(MAX_WAIT_FOR_REPLAY, MAX_QUEUE_SIZE, THROTTLING_DELAY)
MetricTransaction.set_tr_manager(self._tr_manager)
self._watchdog = None
self.skip_ssl_validation = skip_ssl_validation or agent_config.get(
'skip_ssl_validation', False)
self.use_simple_http_client = use_simple_http_client
if self.skip_ssl_validation:
log.info("Skipping SSL hostname validation, useful when using a transparent proxy")
if watchdog:
watchdog_timeout = TRANSACTION_FLUSH_INTERVAL * WATCHDOG_INTERVAL_MULTIPLIER
self._watchdog = Watchdog(
watchdog_timeout, max_mem_mb=agent_config.get('limit_memory_consumption', None))
class Forwarder(tornado.web.Application):
def __init__(self, port, agent_config, watchdog=True, skip_ssl_validation=False,
use_simple_http_client=False):
self._port = int(port)
self._agentConfig = agent_config
self._metrics = {}
MetricTransaction.set_application(self)
MetricTransaction.set_endpoints(MonAPI(agent_config['Api']))
self._tr_manager = TransactionManager(MAX_WAIT_FOR_REPLAY, MAX_QUEUE_SIZE, THROTTLING_DELAY)
MetricTransaction.set_tr_manager(self._tr_manager)
self._watchdog = None
self.skip_ssl_validation = skip_ssl_validation or agent_config.get(
'skip_ssl_validation', False)
self.use_simple_http_client = use_simple_http_client
if self.skip_ssl_validation:
log.info("Skipping SSL hostname validation, useful when using a transparent proxy")
if watchdog:
watchdog_timeout = TRANSACTION_FLUSH_INTERVAL * WATCHDOG_INTERVAL_MULTIPLIER
self._watchdog = Watchdog(
watchdog_timeout, max_mem_mb=agent_config.get('limit_memory_consumption', None))
def _post_metrics(self):
if len(self._metrics) > 0:
MetricTransaction(self._metrics, headers={'Content-Type': 'application/json'})
self._metrics = {}
# todo why is the tornado logging method overridden? Perhaps ditch this.
def log_request(self, handler):
""" Override the tornado logging method.
If everything goes well, log level is DEBUG.
Otherwise it's WARNING or ERROR depending on the response code. """
if handler.get_status() < 400:
log_method = log.debug
elif handler.get_status() < 500:
log_method = log.warning
else:
log_method = log.error
request_time = 1000.0 * handler.request.request_time()
log_method("%d %s %.2fms", handler.get_status(),
handler._request_summary(), request_time)
def run(self):
handlers = [
(r"/intake/?", AgentInputHandler),
(r"/api/v1/series/?", AgentInputHandler),
(r"/status/?", StatusHandler),
]
settings = dict(
cookie_secret="12oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=",
xsrf_cookies=False,
debug=False,
log_function=self.log_request
)
non_local_traffic = self._agentConfig.get("non_local_traffic", False)
tornado.web.Application.__init__(self, handlers, **settings)
http_server = tornado.httpserver.HTTPServer(self)
try:
# non_local_traffic must be == True to match, not just some non-false value
if non_local_traffic is True:
http_server.listen(self._port)
else:
# localhost in lieu of 127.0.0.1 to support IPv6
try:
http_server.listen(self._port, address="localhost")
except gaierror:
log.warning(
"localhost seems undefined in your host file, using 127.0.0.1 instead")
http_server.listen(self._port, address="127.0.0.1")
except socket_error as e:
if "Errno 99" in str(e):
log.warning("IPv6 doesn't seem to be fully supported. Falling back to IPv4")
http_server.listen(self._port, address="127.0.0.1")
else:
raise
except socket_error as e:
log.exception(
"Socket error %s. Is another application listening on the same port ? Exiting", e)
sys.exit(1)
except Exception:
log.exception("Uncaught exception. Forwarder is exiting.")
sys.exit(1)
log.info("Listening on port %d" % self._port)
# Register callbacks
self.mloop = get_tornado_ioloop()
logging.getLogger().setLevel(get_logging_config()['log_level'] or logging.INFO)
def flush_trs():
if self._watchdog:
self._watchdog.reset()
self._post_metrics()
self._tr_manager.flush()
tr_sched = tornado.ioloop.PeriodicCallback(
flush_trs, TRANSACTION_FLUSH_INTERVAL, io_loop=self.mloop)
# Start everything
if self._watchdog:
self._watchdog.reset()
tr_sched.start()
self.mloop.start()
log.info("Stopped")
def stop(self):
self.mloop.stop()
def normal_run():
w = Watchdog(2)
w.reset()
for i in range(5):
time.sleep(1)
w.reset()
def hanging_net():
w = Watchdog(5)
w.reset()
x = url.urlopen("http://localhost:31834")
print "ERROR Net call returned", x
return True
class Reporter(threading.Thread):
"""
The reporter periodically sends the aggregated metrics to the
server.
"""
def __init__(self, interval, aggregator, api_host, use_watchdog=False, event_chunk_size=None):
threading.Thread.__init__(self)
self.interval = int(interval)
self.finished = threading.Event()
self.aggregator = aggregator
self.flush_count = 0
self.log_count = 0
self.watchdog = None
if use_watchdog:
from monagent.common.util import Watchdog
self.watchdog = Watchdog(WATCHDOG_TIMEOUT)
self.api_host = api_host
self.event_chunk_size = event_chunk_size or EVENT_CHUNK_SIZE
@staticmethod
def serialize_metrics(metrics):
return json.dumps({"series": metrics})
def stop(self):
log.info("Stopping reporter")
self.finished.set()
def run(self):
log.info("Reporting to %s every %ss" % (self.api_host, self.interval))
log.debug("Watchdog enabled: %s" % bool(self.watchdog))
# Persist a start-up message.
MonstatsdStatus().persist()
while not self.finished.isSet(): # Use camel case isSet for 2.4 support.
self.finished.wait(self.interval)
self.flush()
if self.watchdog:
self.watchdog.reset()
# Clean up the status messages.
log.debug("Stopped reporter")
MonstatsdStatus.remove_latest_status()
def flush(self):
try:
self.flush_count += 1
self.log_count += 1
metrics = self.aggregator.flush()
count = len(metrics)
if self.flush_count % FLUSH_LOGGING_PERIOD == 0:
self.log_count = 0
if count:
try:
http_emitter(metrics, log, self.api_host)
except Exception:
log.exception("Error running emitter.")
events = self.aggregator.flush_events()
event_count = len(events)
if event_count:
log.warn('Event received but events are not available in the mon api')
should_log = self.flush_count <= FLUSH_LOGGING_INITIAL or self.log_count <= FLUSH_LOGGING_COUNT
log_func = log.info
if not should_log:
log_func = log.debug
log_func(
"Flush #%s: flushed %s metric%s and %s event%s" %
(self.flush_count,
count,
plural(count),
event_count,
plural(event_count)))
if self.flush_count == FLUSH_LOGGING_INITIAL:
log.info(
"First flushes done, %s flushes will be logged every %s flushes." %
(FLUSH_LOGGING_COUNT, FLUSH_LOGGING_PERIOD))
# Persist a status message.
packet_count = self.aggregator.total_count
packets_per_second = self.aggregator.packets_per_second(self.interval)
MonstatsdStatus(
flush_count=self.flush_count,
packet_count=packet_count,
packets_per_second=packets_per_second,
metric_count=count,
event_count=event_count,
).persist()
except Exception:
log.exception("Error flushing metrics")
def fast_tornado():
a = CollectorDaemon(12345, {})
a._watchdog = Watchdog(6)
a._tr_manager = MockTxManager()
a.run()
def normal_run():
w = Watchdog(2)
w.reset()
for i in range(5):
time.sleep(1)
w.reset()
def hanging_net():
w = Watchdog(5)
w.reset()
x = url.urlopen("http://localhost:31834")
print("ERROR Net call returned", x)
return True
class Forwarder(tornado.web.Application):
def __init__(self,
port,
agent_config,
watchdog=True,
skip_ssl_validation=False,
use_simple_http_client=False):
self._port = int(port)
self._agentConfig = agent_config
self._metrics = {}
MetricTransaction.set_application(self)
MetricTransaction.set_endpoints(MonAPI(agent_config['Api']))
self._tr_manager = TransactionManager(MAX_WAIT_FOR_REPLAY,
MAX_QUEUE_SIZE, THROTTLING_DELAY)
MetricTransaction.set_tr_manager(self._tr_manager)
self._watchdog = None
self.skip_ssl_validation = skip_ssl_validation or agent_config.get(
'skip_ssl_validation', False)
self.use_simple_http_client = use_simple_http_client
if self.skip_ssl_validation:
log.info(
"Skipping SSL hostname validation, useful when using a transparent proxy"
)
if watchdog:
watchdog_timeout = TRANSACTION_FLUSH_INTERVAL * WATCHDOG_INTERVAL_MULTIPLIER
self._watchdog = Watchdog(watchdog_timeout,
max_mem_mb=agent_config.get(
'limit_memory_consumption', None))
def _post_metrics(self):
if len(self._metrics) > 0:
MetricTransaction(self._metrics,
headers={'Content-Type': 'application/json'})
self._metrics = {}
# todo why is the tornado logging method overridden? Perhaps ditch this.
def log_request(self, handler):
""" Override the tornado logging method.
If everything goes well, log level is DEBUG.
Otherwise it's WARNING or ERROR depending on the response code. """
if handler.get_status() < 400:
log_method = log.debug
elif handler.get_status() < 500:
log_method = log.warning
else:
log_method = log.error
request_time = 1000.0 * handler.request.request_time()
log_method("%d %s %.2fms", handler.get_status(),
handler._request_summary(), request_time)
def run(self):
handlers = [
(r"/intake/?", AgentInputHandler),
(r"/api/v1/series/?", AgentInputHandler),
(r"/status/?", StatusHandler),
]
settings = dict(
cookie_secret="12oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=",
xsrf_cookies=False,
debug=False,
log_function=self.log_request)
non_local_traffic = self._agentConfig.get("non_local_traffic", False)
tornado.web.Application.__init__(self, handlers, **settings)
http_server = tornado.httpserver.HTTPServer(self)
try:
# non_local_traffic must be == True to match, not just some non-false value
if non_local_traffic is True:
http_server.listen(self._port)
else:
# localhost in lieu of 127.0.0.1 to support IPv6
try:
http_server.listen(self._port, address="localhost")
except gaierror:
log.warning(
"localhost seems undefined in your host file, using 127.0.0.1 instead"
)
http_server.listen(self._port, address="127.0.0.1")
except socket_error as e:
if "Errno 99" in str(e):
log.warning(
"IPv6 doesn't seem to be fully supported. Falling back to IPv4"
)
http_server.listen(self._port, address="127.0.0.1")
else:
raise
except socket_error as e:
log.exception(
"Socket error %s. Is another application listening on the same port ? Exiting",
e)
sys.exit(1)
except Exception:
log.exception("Uncaught exception. Forwarder is exiting.")
sys.exit(1)
log.info("Listening on port %d" % self._port)
# Register callbacks
self.mloop = get_tornado_ioloop()
logging.getLogger().setLevel(get_logging_config()['log_level']
or logging.INFO)
def flush_trs():
if self._watchdog:
self._watchdog.reset()
self._post_metrics()
self._tr_manager.flush()
tr_sched = tornado.ioloop.PeriodicCallback(flush_trs,
TRANSACTION_FLUSH_INTERVAL,
io_loop=self.mloop)
# Start everything
if self._watchdog:
self._watchdog.reset()
tr_sched.start()
self.mloop.start()
log.info("Stopped")
def stop(self):
self.mloop.stop()