def to_runner_api_monitoring_info(self):
"""Returns a Metric with this value for use in a MonitoringInfo."""
return metrics_pb2.Metric(
distribution_data=metrics_pb2.DistributionData(
int_distribution_data=metrics_pb2.IntDistributionData(
count=self.count, sum=self.sum, min=self.min,
max=self.max)))
def pcollection_count_monitoring_infos(self, transform_id):
"""Returns the element count MonitoringInfo collected by this operation."""
if len(self.receivers) == 1:
# If there is exactly one output, we can unambiguously
# fix its name later, which we do.
# TODO(robertwb): Plumb the actual name here.
elem_count_mi = monitoring_infos.int64_counter(
monitoring_infos.ELEMENT_COUNT_URN,
self.receivers[0].opcounter.element_counter.value(),
ptransform=transform_id,
tag='ONLY_OUTPUT' if len(self.receivers) == 1 else str(None),
)
(unused_mean, sum, count, min,
max) = (self.receivers[0].opcounter.mean_byte_counter.value())
metric = metrics_pb2.Metric(
distribution_data=metrics_pb2.DistributionData(
int_distribution_data=metrics_pb2.IntDistributionData(
count=count, sum=sum, min=min, max=max)))
sampled_byte_count = monitoring_infos.int64_distribution(
monitoring_infos.SAMPLED_BYTE_SIZE_URN,
metric,
ptransform=transform_id,
tag='ONLY_OUTPUT' if len(self.receivers) == 1 else str(None),
)
return {
monitoring_infos.to_key(elem_count_mi): elem_count_mi,
monitoring_infos.to_key(sampled_byte_count): sampled_byte_count
}
return {}
def monitoring_infos(self, transform_id):
# type: (str) -> Dict[FrozenSet, metrics_pb2.MonitoringInfo]
infos = super(DoOperation, self).monitoring_infos(transform_id)
if self.tagged_receivers:
for tag, receiver in self.tagged_receivers.items():
mi = monitoring_infos.int64_counter(
monitoring_infos.ELEMENT_COUNT_URN,
receiver.opcounter.element_counter.value(),
ptransform=transform_id,
tag=str(tag))
infos[monitoring_infos.to_key(mi)] = mi
(unused_mean, sum, count, min,
max) = (receiver.opcounter.mean_byte_counter.value())
metric = metrics_pb2.Metric(
distribution_data=metrics_pb2.DistributionData(
int_distribution_data=metrics_pb2.IntDistributionData(
count=count, sum=sum, min=min, max=max)))
sampled_byte_count = monitoring_infos.int64_distribution(
monitoring_infos.SAMPLED_BYTE_SIZE_URN,
metric,
ptransform=transform_id,
tag=str(tag))
infos[monitoring_infos.to_key(
sampled_byte_count)] = sampled_byte_count
return infos
def to_runner_api_monitoring_info(self, name, transform_id):
from apache_beam.metrics import monitoring_infos
return monitoring_infos.int64_user_counter(
name.namespace,
name.name,
metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=self.get_cumulative())),
ptransform=transform_id)
def to_runner_api_monitoring_info(self):
"""Returns a Metric with this counter value for use in a MonitoringInfo."""
# TODO(ajamato): Update this code to be consistent with Gauges
# and Distributions. Since there is no CounterData class this method
# was added to CounterCell. Consider adding a CounterData class or
# removing the GaugeData and DistributionData classes.
return metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=self.get_cumulative()))
def distribution_combiner(metric_a, metric_b):
a_data = metric_a.distribution_data.int_distribution_data
b_data = metric_b.distribution_data.int_distribution_data
return metrics_pb2.Metric(distribution_data=metrics_pb2.DistributionData(
int_distribution_data=metrics_pb2.IntDistributionData(
count=a_data.count + b_data.count,
sum=a_data.sum + b_data.sum,
min=min(a_data.min, b_data.min),
max=max(a_data.max, b_data.max))))
def int64_gauge(urn, metric, ptransform=None, tag=None):
"""Return the gauge monitoring info for the URN, metric and labels.
Args:
urn: The URN of the monitoring info/metric.
metric: The metric proto field to use in the monitoring info.
ptransform: The ptransform/step name used as a label.
tag: The output tag name, used as a label.
"""
labels = create_labels(ptransform=ptransform, tag=tag)
if isinstance(metric, int):
metric = metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=metric))
return create_monitoring_info(urn, LATEST_INT64_TYPE, metric, labels)
def int64_counter(urn, metric, ptransform=None, tag=None):
# type: (...) -> metrics_pb2.MonitoringInfo
"""Return the counter monitoring info for the specifed URN, metric and labels.
Args:
urn: The URN of the monitoring info/metric.
metric: The metric proto field to use in the monitoring info.
Or an int value.
ptransform: The ptransform/step name used as a label.
tag: The output tag name, used as a label.
"""
labels = create_labels(ptransform=ptransform, tag=tag)
if isinstance(metric, int):
metric = metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=metric))
return create_monitoring_info(urn, SUM_INT64_TYPE, metric, labels)
def to_runner_api_monitoring_info(self):
"""Returns a Metric with this value for use in a MonitoringInfo."""
return metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=self.value))
def distribution_combiner(metric_a, metric_b):
a_data = metric_a.distribution_data.int_distribution_data
b_data = metric_b.distribution_data.int_distribution_data
return metrics_pb2.Metric(distribution_data=metrics_pb2.DistributionData(
int_distribution_data=metrics_pb2.IntDistributionData(
count=a_data.count + b_data.count,
sum=a_data.sum + b_data.sum,
min=min(a_data.min, b_data.min),
max=max(a_data.max, b_data.max))))
_KNOWN_COMBINERS = {
SUM_INT64_TYPE:
lambda a, b: metrics_pb2.Metric(counter_data=metrics_pb2.CounterData(
int64_value=a.counter_data.int64_value + b.counter_data.int64_value)),
DISTRIBUTION_INT64_TYPE:
distribution_combiner,
}
def max_timestamp(a, b):
if a.ToNanoseconds() > b.ToNanoseconds():
return a
else:
return b
def consolidate(metrics, key=to_key):
grouped = collections.defaultdict(list)
for metric in metrics: