def start_events_device(
log_dir: Optional[Union[str, Path]] = None,
_registry: Optional[EventsProcessesRegistry] = None) -> None:
if _registry is None:
if log_dir is None:
raise RuntimeError(
"Should provide log_dir or instance of EventsProcessesRegistry"
)
_registry = create_default_events_process_registry(log_dir=log_dir)
start_events_main_device(_registry=_registry)
time.sleep(EVENTS_DEVICE_START_DELAY)
start_events_logger(_registry=_registry)
start_grafana_pipeline(_registry=_registry)
start_events_analyzer(_registry=_registry)
time.sleep(EVENTS_SUBSCRIBERS_START_DELAY)
# Default filters.
EventsSeverityChangerFilter(
new_severity=Severity.WARNING,
event_class=DatabaseLogEvent.DATABASE_ERROR,
regex=
r'.*workload prioritization - update_service_levels_from_distributed_data: an '
r'error occurred while retrieving configuration').publish()
DbEventsFilter(db_event=DatabaseLogEvent.BACKTRACE,
line='Rate-limit: supressed').publish()
DbEventsFilter(db_event=DatabaseLogEvent.BACKTRACE,
line='Rate-limit: suppressed').publish()
atexit.register(stop_events_device, _registry=_registry)
def start_events_device(
log_dir: Optional[Union[str, Path]] = None,
_registry: Optional[EventsProcessesRegistry] = None) -> None:
if _registry is None:
if log_dir is None:
raise RuntimeError(
"Should provide log_dir or instance of EventsProcessesRegistry"
)
_registry = create_default_events_process_registry(log_dir=log_dir)
start_events_main_device(_registry=_registry)
time.sleep(EVENTS_DEVICE_START_DELAY)
start_events_logger(_registry=_registry)
start_grafana_pipeline(_registry=_registry)
start_events_analyzer(_registry=_registry)
time.sleep(EVENTS_SUBSCRIBERS_START_DELAY)
# Default filters.
DbEventsFilter(db_event=DatabaseLogEvent.BACKTRACE,
line='Rate-limit: supressed').publish()
DbEventsFilter(db_event=DatabaseLogEvent.BACKTRACE,
line='Rate-limit: suppressed').publish()
atexit.register(stop_events_device, _registry=_registry)
def test_grafana(self):
start_grafana_pipeline(_registry=self.events_processes_registry)
grafana_annotator = get_events_process(EVENTS_GRAFANA_ANNOTATOR_ID, _registry=self.events_processes_registry)
grafana_aggregator = get_events_process(EVENTS_GRAFANA_AGGREGATOR_ID, _registry=self.events_processes_registry)
grafana_postman = get_grafana_postman(_registry=self.events_processes_registry)
time.sleep(EVENTS_SUBSCRIBERS_START_DELAY)
try:
self.assertIsInstance(grafana_annotator, GrafanaAnnotator)
self.assertTrue(grafana_annotator.is_alive())
self.assertEqual(grafana_annotator._registry, self.events_main_device._registry)
self.assertEqual(grafana_annotator._registry, self.events_processes_registry)
self.assertIsInstance(grafana_aggregator, GrafanaEventAggregator)
self.assertTrue(grafana_aggregator.is_alive())
self.assertEqual(grafana_aggregator._registry, self.events_main_device._registry)
self.assertEqual(grafana_aggregator._registry, self.events_processes_registry)
self.assertIsInstance(grafana_postman, GrafanaEventPostman)
self.assertTrue(grafana_postman.is_alive())
self.assertEqual(grafana_postman._registry, self.events_main_device._registry)
self.assertEqual(grafana_postman._registry, self.events_processes_registry)
grafana_aggregator.time_window = 1
set_grafana_url("http://localhost", _registry=self.events_processes_registry)
with unittest.mock.patch("requests.post") as mock:
for runs in range(1, 4):
with self.wait_for_n_events(grafana_annotator, count=10, timeout=1):
for _ in range(10):
self.events_main_device.publish_event(
ClusterHealthValidatorEvent.NodeStatus(severity=Severity.NORMAL))
time.sleep(1)
self.assertEqual(mock.call_count, runs * 5)
self.assertEqual(
mock.call_args.kwargs["json"]["tags"],
["ClusterHealthValidatorEvent", "NORMAL", "events", "NodeStatus"],
)
self.assertEqual(self.events_main_device.events_counter, grafana_annotator.events_counter)
self.assertEqual(grafana_annotator.events_counter, grafana_aggregator.events_counter)
self.assertLessEqual(grafana_postman.events_counter, grafana_aggregator.events_counter)
finally:
grafana_annotator.stop(timeout=1)
grafana_aggregator.stop(timeout=1)
grafana_postman.stop(timeout=1)