[BENCHMARK] sync instrument recording benchmarks

[dbarker]

Contributes to #4317 and #4416

Adds sync instruments recording benchmarks covering all instrument types and aggregation types. Includes Exemplar filters and bound instruments when enabled.

Changes

• Disabled instrument benchmarks (Counter, Histogram, Gauge, and bound variants)
• Counter benchmarks
  • Drop aggregation
  • Sum aggregation (by attribute count, by cardinality, by thread count)
• Histogram benchmarks
  • Drop aggregation
  • Explicit bucket and Base2 aggregation (by attribute count, by cardinality, by thread count)
• Gauge benchmarks
  • Drop aggregation
  • Last value aggregation (by attribute count, by cardinality, by thread count)
• ExemplarFilter
  • Counter with AlwaysOff, AlwaysOn, and TraceBased filters
  • Histogram (Default and Base2Expo) with AlwaysOff and AlwaysOn filters
• Bound instruments (counter and histogram)
  • Drop aggregation
  • Sum aggregation (by thread count)
  • Histogram (default and base2) aggregation (by thread count)

For significant contributions please make sure you have completed the following items:

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added
• Changes in public API reviewed

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.09%. Comparing base (cd252b4) to head (cb8bd08).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4470   +/-   ##
=======================================
  Coverage   83.09%   83.09%           
=======================================
  Files         519      519           
  Lines       20253    20253           
=======================================
  Hits        16827    16827           
  Misses       3426     3426           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dbarker]

Some interesting findings from this benchmark that warrant a followup.

• Need to optimize the drop aggregation path to behave like a disabled metrics stream
• Need to investigate the performance regression for unbound instruments when building with ABIv2 and the bound instrument preview.

[dbarker]

There is an opportunity to reduce recording impact with drop aggregation to behave as if the stream was disabled. Currently the sync storage computes the attribute hash and gains locks when drop aggregation is set.

ABIv1: Disabled vs. Drop (by threads recording with one instrument)

Instrument	Disabled	Drop	Overhead vs. Disabled
Counter (threads:1)	0.223 ns	181 ns	812x
Counter (threads:2)	0.334 ns	336 ns	1006x
Counter (threads:4)	0.335 ns	548 ns	1636x
Histogram Explicit (threads:1)	0.215 ns	180 ns	837x
Histogram Explicit (threads:2)	0.215 ns	198 ns	921x
Histogram Explicit (threads:4)	0.227 ns	556 ns	2450x

[dbarker]

When building with ABIv2 and OPENTELEMETRY_HAVE_METRICS_BOUND_INSTRUMENTS_PREVIEW = ON there is a significant performance regression with unbound sync instrument recording compared to ABIv1.

Benchmark (threads:1)	ABIv1 no bound preview	ABIv2 bound preview ON	Delta
Counter Drop	181 ns	291 ns	+61%
Counter Sum	183 ns	293 ns	+60%
Histogram Drop	180 ns	288 ns	+60%
Histogram Explicit	192 ns	297 ns	+55%
Histogram Base2Expo	209 ns	300 ns	+44%

[dbarker]

The speed up for bound instruments compared to unbound is impressive at ~16-50x.

ABIv2: Unbound vs. Bound recording overhead (single thread) with nominal attributes (3)

Instrument	Unbound	Bound	Speedup
Counter Drop	291 ns	5.81 ns	50x
Counter Sum	293 ns	10.7 ns	27x
Histogram Drop	288 ns	6.19 ns	47x
Histogram Explicit Buckets	297 ns	11.8 ns	25x
Histogram Base2 Exponential	300 ns	19.3 ns	16x

[dbarker]

The attribute count has the highest impact during recording and cardinality is not a major factor for latency.

ABIv1 Counter Sum: Attribute count vs. Cardinality

Attribute count (single time series, varying key count per call)

Attribute count	0	1	10	128
Counter Sum	21.7 ns	63.7 ns	665 ns	11,495 ns
Histogram Explicit Buckets	25.4 ns	73.7 ns	662 ns	10,666 ns
Histogram Base2 Exponential	27.4 ns	72.1 ns	662 ns	10,619 ns

Cardinality (3 attributes, N distinct time series, cycling through all)

Active time series	10	500	2000	4000 (overflow)
Counter Sum	189 ns	187 ns	185 ns	195 ns
Histogram Explicit Buckets	202 ns	202 ns	205 ns	—
Histogram Base2 Exponential	207 ns	203 ns	206 ns	—