Aggregate ValueRecorder to HistogramAggregator on Prometheus

[alexppg]

Describe your environment
Sorry if I use the wrong nomenclature, I'm beginning to understand how OT works.

I'm trying to use the view example changing the exporter to prometheus.

So far I've seen that:

• MinMaxSumCountAggregator is not supported yet, as this issue exists MinMaxSumCountAggregator export to Prometheus; map to summary + min & max? #1172
• HistogramAggregator is not supported yet, as it returns:

Traceback (most recent call last):
  File "example_microservice/test.py", line 98, in <module>
    print(generate_latest())
  File "/home/ap/.local/share/virtualenvs/example-microservice-5M-GZT7Z/lib/python3.6/site-packages/prometheus_client/exposition.py", line 137, in generate_latest
    output.append(sample_line(s))
  File "/home/ap/.local/share/virtualenvs/example-microservice-5M-GZT7Z/lib/python3.6/site-packages/prometheus_client/exposition.py", line 103, in sample_line
    line.name, labelstr, floatToGoString(line.value), timestamp)
  File "/home/ap/.local/share/virtualenvs/example-microservice-5M-GZT7Z/lib/python3.6/site-packages/prometheus_client/utils.py", line 9, in floatToGoString
    d = float(d)
TypeError: ("float() argument must be a string or a number, not 'collections.OrderedDict'", Metric(requests_size, size of requests, unknown, , [Sample(name='requests_size', labels={'test': 'value'}, value=OrderedDict([(20, 1), (40, 1), (60, 0), (80, 0), (100, 0), (inf, 1)]), timestamp=None, exemplar=None)]))

Steps to reproduce
Execute this modified version of the view.py example. To see the HistogramAggregator error, just comment the meter.register_view(counter_view2) line.

from opentelemetry import metrics
from opentelemetry.sdk.metrics import (
    MeterProvider,
    Counter,
    ValueRecorder,
)
from opentelemetry.exporter.prometheus import PrometheusMetricsExporter
from opentelemetry.sdk.metrics.export.aggregate import (
    HistogramAggregator,
    LastValueAggregator,
    MinMaxSumCountAggregator,
    SumAggregator,
)
from opentelemetry.sdk.metrics.view import View, ViewConfig
from prometheus_client import generate_latest
from time import sleep

# Use the meter type provided by the SDK package
metrics.set_meter_provider(MeterProvider())
meter = metrics.get_meter(__name__)
metrics.get_meter_provider().start_pipeline(meter, PrometheusMetricsExporter(), 1)

requests_counter = meter.create_metric(
    name="requests",
    description="number of requests",
    unit="1",
    value_type=int,
    metric_type=Counter,
)

requests_size = meter.create_metric(
    name="requests_size",
    description="size of requests",
    unit="1",
    value_type=int,
    metric_type=ValueRecorder,
)

# Views are used to define an aggregation type and label keys to aggregate by
# for a given metric

# Two views with the same metric and aggregation type but different label keys
# With ViewConfig.LABEL_KEYS, all labels but the ones defined in label_keys are
# dropped from the aggregation
counter_view1 = View(
    requests_counter,
    SumAggregator,
    label_keys=["environment"],
    view_config=ViewConfig.LABEL_KEYS,
)
counter_view2 = View(
    requests_counter,
    MinMaxSumCountAggregator,
    label_keys=["os_type"],
    view_config=ViewConfig.LABEL_KEYS,
)
# This view has ViewConfig set to UNGROUPED, meaning all recorded metrics take
# the labels directly without and consideration for label_keys
counter_view3 = View(
    requests_counter,
    LastValueAggregator,
    label_keys=["environment"],  # is not used due to ViewConfig.UNGROUPED
    view_config=ViewConfig.UNGROUPED,
)
# This view uses the HistogramAggregator which accepts an option config
# parameter to specify the bucket ranges
size_view = View(
    requests_size,
    HistogramAggregator,
    label_keys=["environment"],  # is not used due to ViewConfig.UNGROUPED
    aggregator_config={"bounds": [20, 40, 60, 80, 100]},
    view_config=ViewConfig.UNGROUPED,
)

# Register the views to the view manager to use the views. Views MUST be
# registered before recording metrics. Metrics that are recorded without
# views defined for them will use a default for that type of metric
meter.register_view(counter_view1)
meter.register_view(counter_view2)
meter.register_view(counter_view3)
meter.register_view(size_view)

# The views will evaluate the labels passed into the record and aggregate upon
# the unique labels that are generated
# view1 labels will evaluate to {"environment": "staging"}
# view2 labels will evaluate to {"os_type": linux}
# view3 labels will evaluate to {"environment": "staging", "os_type": "linux"}
requests_counter.add(100, {"environment": "staging", "os_type": "linux"})

# Since this is using the HistogramAggregator, the bucket counts will be reflected
# with each record
requests_size.record(25, {"test": "value"})
requests_size.record(-3, {"test": "value"})
requests_size.record(200, {"test": "value"})

sleep(2)

print(generate_latest())
input("...\n")

What is the expected behavior?
It should not return a traceback error and generate the metrics.

What is the actual behavior?
It does return a traceback error and doesn't generate the metrics.

[alexppg]

Just to update if anybody finds this issue, @lzchen answered me in the gitter channel. The Histogram aggregation is not implemented yet for prometheus. Also, there's some discussion about which should be the default aggregator for ValueRecord: open-telemetry/opentelemetry-specification#982