Add support for Delta temporality

[srikanthccv]

Fixes #2405
Fixes #1707

Enable using --prefer-delta CLI flag.

The delta temporality is low on memory, and programs with strict memory constraints use delta.

Corresponding collector PR SigNoz/signoz-otel-collector#121

Let's look at the example data for a metric http_requests_count with different temporalities.

Assume

• Collection is done every 5s
• Two counter measurements, one for each kind, are recorded in the same application

t	v(cumulative)	v(delta)
12.00	0	0
12.05	12	12
12.10	14	2
12.15	16	2
12.20	22	6
12.25	25	3
12.30	32	7
12.35	37	5
12.40	45	8
12.45	49	4
12.50	59	10
12.55	66	7

The request rate with a step interval of '15's

t	v(cumulative)	v(delta)
12.00	(0, 12, 14)	(0, 12, 2)
12.15	(16, 22, 25)	(2, 6, 3)
12.30	(32, 37, 45)	(7, 5, 8)
12.45	(49, 59, 66)	(4, 10, 7)

For the cumulative temporality, we take the highest value at the timestamp and calculate the running difference of the value
and for delta it will be sum of the values in the step interval group.

t	v(cumulative)	v(delta)
12.00	(14-0)/15 ~= 0.93	(0+12+2)/15 ~= 0.93
12.15	(25-14)/15 ~= 0.73	(2+6+3)/15 ~= 0.73
12.30	(45-25)/15 ~= 1.33	(7+5+8)/15 ~= 1.33
12.45	(66-45)/15 ~= 1.4	(4+10+7)/15 ~= 1.4

Use the following script to generate the above equivalent data.

app.py

import time
import random

from opentelemetry.metrics import get_meter_provider, set_meter_provider
from opentelemetry.sdk.metrics import Counter, MeterProvider
from opentelemetry.sdk.metrics.export import (
    AggregationTemporality,
    PeriodicExportingMetricReader,
)
from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter

temporality_cumulative = {Counter: AggregationTemporality.CUMULATIVE}
temporality_delta = {Counter: AggregationTemporality.DELTA}

# The metrics that are exported using this exporter will represent a cumulative value
exporter = OTLPMetricExporter(
    preferred_temporality=temporality_cumulative,
)

# The metrics that are exported using this exporter will represent a delta value
exporter2 = OTLPMetricExporter(
    preferred_temporality=temporality_delta,
)

# The PeriodicExportingMetricReader takes the preferred aggregation
# from the passed in exporter
reader = PeriodicExportingMetricReader(
    exporter,
    export_interval_millis=5_000,
)

# The PeriodicExportingMetricReader takes the preferred aggregation
# from the passed in exporter
reader2 = PeriodicExportingMetricReader(
    exporter2,
    export_interval_millis=5_000,
)

provider1 = MeterProvider(metric_readers=[reader])

meter1 = provider1.get_meter("preferred-temporality", "0.1.2")

counter_cumulative = meter1.create_counter("my-counter-cumulative")

provider2 = MeterProvider(metric_readers=[reader2])

meter2 = provider2.get_meter("preferred-temporality", "0.1.2")

counter_delta = meter2.create_counter("my-counter-delta")

indian_cities = ["Mumbai", "Delhi", "Bangalore", "Hyderabad", "Ahmedabad"]
american_cities = ["New York", "Los Angeles", "Chicago", "Houston", "Philadelphia"]

while True:
    labels = {"indian_city": random.choice(indian_cities), "american_city": random.choice(american_cities)}
    count = random.randint(1, 100)
    counter_cumulative.add(count, labels)
    counter_delta.add(count, labels)
    time.sleep(1)

requirements.txt

backoff==2.2.1
certifi==2022.12.7
charset-normalizer==3.1.0
Deprecated==1.2.13
googleapis-common-protos==1.58.0
grpcio==1.51.3
idna==3.4
opentelemetry-api==1.16.0
opentelemetry-distro==0.37b0
opentelemetry-exporter-otlp==1.16.0
opentelemetry-exporter-otlp-proto-grpc==1.16.0
opentelemetry-exporter-otlp-proto-http==1.16.0
opentelemetry-instrumentation==0.37b0
opentelemetry-proto==1.16.0
opentelemetry-sdk==1.16.0
opentelemetry-semantic-conventions==0.37b0
protobuf==4.22.1
requests==2.28.2
typing_extensions==4.5.0
urllib3==1.26.15
wrapt==1.15.0

Run the program for some time and try charting the SUM_RATE for both metrics.