Remove Quantile aggregation, DDSketch aggregator; add Exact timestamps

[jmacd]

This is a cleanup of an area in the code that is not well exercised.

• Renames aggregator/array to aggregator/exact
• Add missing exact timestamps to the exact aggregator
• Remove Quantile and Distribution export aggregation interfaces
• Remove DDSketch aggregator; we expect this will return after more histogram types are available.

Before:

goos: darwin
goarch: amd64
pkg: go.opentelemetry.io/otel/sdk/metric
BenchmarkInt64ArrayAdd-8           	 3018822	       369 ns/op	     301 B/op	       2 allocs/op
BenchmarkInt64ArrayHandleAdd-8     	27656120	        55.6 ns/op	      46 B/op	       0 allocs/op
BenchmarkFloat64ArrayAdd-8         	 3263607	       423 ns/op	     298 B/op	       2 allocs/op
BenchmarkFloat64ArrayHandleAdd-8   	27069272	        56.4 ns/op	      47 B/op	       0 allocs/op

After:

BenchmarkInt64ExactAdd-8           	 1657836	       711 ns/op	     419 B/op	       2 allocs/op
BenchmarkInt64ExactHandleAdd-8     	 6865136	       208 ns/op	     188 B/op	       0 allocs/op
BenchmarkFloat64ExactAdd-8         	 2269024	       566 ns/op	     442 B/op	       2 allocs/op
BenchmarkFloat64ExactHandleAdd-8   	 5736513	       178 ns/op	     180 B/op	       0 allocs/op

Although this updates the exact aggregator, that code is still not used within this repository. A future PR may add support to the OTLP exporter to export raw point values and timestamps.

Closes #1417

[codecov]

Codecov Report

Merging #1412 (93feac1) into master (9c94941) will increase coverage by 0.0%.
The diff coverage is 88.1%.

@@           Coverage Diff           @@
##           master   #1412    +/-   ##
=======================================
  Coverage    78.8%   78.8%            
=======================================
  Files         131     130     -1     
  Lines        6753    6642   -111     
=======================================
- Hits         5326    5239    -87     
+ Misses       1167    1148    -19     
+ Partials      260     255     -5     

Impacted Files	Coverage Δ
exporters/metric/prometheus/prometheus.go	64.5% <0.0%> (+6.9%)	⬆️
exporters/stdout/config.go	70.0% <ø> (-6.0%)	⬇️
exporters/stdout/exporter.go	18.5% <0.0%> (-7.5%)	⬇️
exporters/stdout/metric.go	76.3% <ø> (-0.2%)	⬇️
sdk/export/metric/aggregation/aggregation.go	100.0% <ø> (ø)
sdk/metric/aggregator/aggregatortest/test.go	84.8% <ø> (-0.2%)	⬇️
sdk/metric/aggregator/minmaxsumcount/mmsc.go	96.0% <ø> (ø)
exporters/otlp/internal/transform/metric.go	80.5% <75.0%> (ø)
sdk/metric/processor/processortest/test.go	89.7% <93.3%> (+0.1%)	⬆️
sdk/metric/aggregator/exact/exact.go	94.4% <94.4%> (ø)
... and 3 more

[jmacd]

@open-telemetry/specs-metrics-approvers As I am trying to bring the OTel-Go metrics SDK into line with "Everything we will specify", this Quantile aggregation is something I'd like to remove. I was on the fence about adding timestamps and renaming the Array->Exact aggregator, but I think we want to keep an Exact aggregator that captures timestamps, so this is a small change in the forward direction.

(Note: I'm hoping to relieve the @open-telemetry/go-approvers of this responsibility.)

[krnowak]

The PR looks good, just found some nitpicks.

[jmacd]

@bogdandrutu Please review.

[bogdandrutu]

I would encourage to split this PR:

1. Changes to array/exact aggregator
2. Remove Quantile and Distribution export aggregation interfaces
3. Remove DDSketch aggregator; we expect this will return after more histogram types are available.

[jmacd]

@bogdandrutu Yes the three parts of this PR are logically independent and could be split in sequence. I think it's way more trouble than it's worth though.

[bogdandrutu]

@jmacd I am not the maintainer, so my comment is just a suggestion. If I was the maintainer I would have pushed for this :)

[MrAlias]

We talked about splitting this up in the SIG meeting yesterday and came to the decision that given the existing approvals we should progress as is. Though future PRs should follow the guidance you laid out @bogdandrutu.

@jmacd I think this can merge once the this is addressed.

[rmrf]

Where I can find the reason to remove quantile?

[rmrf]

Where I can find the reason to remove quantile?

I found out the reason: https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/datamodel.md#summary-legacy