def run(self):
if self.rdt_enabled and not check_resctrl():
return
elif not self.rdt_enabled:
log.warning('Rdt disabled. Skipping collecting measurements '
'and resctrl synchronization')
else:
# Resctrl is enabled and available - cleanup previous runs.
cleanup_resctrl()
if not are_privileges_sufficient():
log.critical(
"Impossible to use perf_event_open. You need to: adjust "
"/proc/sys/kernel/perf_event_paranoid; or has CAP_DAC_OVERRIDE capability"
" set. You can run process as root too. See man 2 perf_event_open for "
"details.")
return
while True:
# Collect information about tasks running on node.
tasks = self.node.get_tasks()
# Keep sync of found tasks and internally managed containers.
self._sync_containers_state(tasks)
# Owca internal metrics.
internal_metrics = [
Metric(name='owca_up',
type=MetricType.COUNTER,
value=time.time()),
Metric(name='owca_tasks',
type=MetricType.GAUGE,
value=len(tasks)),
]
# Platform information
platform, platform_metrics, platform_labels = platforms.collect_platform_information(
)
# Common labels
common_labels = dict(platform_labels, **self.extra_labels)
# Update platform_metrics with common labels.
for metric in platform_metrics + internal_metrics:
metric.labels.update(common_labels)
# Build labeled tasks_metrics and task_metrics_values.
tasks_measurements: TasksMeasurements = {}
tasks_resources: TasksResources = {}
tasks_labels: TasksLabels = {}
tasks_metrics: List[Metric] = []
for task, container in self.containers.items():
# Single task data
task_measurements = container.get_measurements()
task_metrics = create_metrics(task_measurements)
# Prepare tasks labels based on Mesos tasks metadata labels and task id.
task_labels = {
sanitize_mesos_label(label_key): label_value
for label_key, label_value in task.labels.items()
}
task_labels['task_id'] = task.task_id
# Task scoped label decoration.
for task_metric in task_metrics:
task_metric.labels.update(common_labels)
task_metric.labels.update(task_labels)
# Aggregate over all tasks.
tasks_labels[task.task_id] = task_labels
tasks_measurements[task.task_id] = task_measurements
tasks_resources[task.task_id] = task.resources
tasks_metrics += task_metrics
self.metrics_storage.store(platform_metrics + tasks_metrics +
internal_metrics)
anomalies, extra_metrics = self.detector.detect(
platform, tasks_measurements, tasks_resources, tasks_labels)
log.debug('Anomalies detected: %d', len(anomalies))
# Note: anomaly metrics include metrics found in ContentionAnomaly.metrics.
anomaly_metrics = convert_anomalies_to_metrics(
anomalies, tasks_labels)
# Extra anomaly statistics
if len(anomalies):
self.anomaly_last_occurence = time.time()
self.anomaly_counter += len(anomalies)
statistics_metrics = [
Metric(name='anomaly_count',
type=MetricType.COUNTER,
value=self.anomaly_counter),
]
if self.anomaly_last_occurence:
statistics_metrics.extend([
Metric(name='anomaly_last_occurence',
type=MetricType.COUNTER,
value=self.anomaly_last_occurence),
])
# Update anomaly & extra metrics with common labels.
for metric in anomaly_metrics + extra_metrics + statistics_metrics:
metric.labels.update(common_labels)
self.anomalies_storage.store(anomaly_metrics + extra_metrics +
statistics_metrics)
if not self.wait_or_finish():
break
# cleanup
for container in self.containers.values():
container.cleanup()
def test_check_resctrl(*mock):
assert check_resctrl()