feat(metrics): use MetricDescriptor to determine aggregator #989

[vmarchaud]

Currently the batcher interface define the following:

export abstract class Batcher {
  // [ ... ]
  abstract aggregatorFor(metricKind: MetricKind): Aggregator;
  // [ ... ]
}

Which is spec compliant but from a user perspective it only possible to use a specific aggregator for each kind of metricKind. Some might want to configure that on a more granular level using the metric name or labels for example.

The go implementation already use MetricDescriptor instead of MetricKind i believe for the same reason: https://github.com/open-telemetry/opentelemetry-go/blob/master/sdk/export/metric/metric.go#L100

More details here: #989 (comment)

[codecov-io]

Codecov Report

Merging #1014 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1014   +/-   ##
=======================================
  Coverage   95.00%   95.00%           
=======================================
  Files         212      212           
  Lines        8806     8806           
  Branches      796      796           
=======================================
  Hits         8366     8366           
  Misses        440      440           

Impacted Files	Coverage Δ
packages/opentelemetry-metrics/src/Metric.ts	96.87% <ø> (ø)
...ckages/opentelemetry-metrics/src/export/Batcher.ts	100.00% <100.00%> (ø)
packages/opentelemetry-metrics/test/Meter.test.ts	99.35% <100.00%> (ø)

[legendecas]

👍 PR-wise looks fine. But this doesn't explain the best practice to determine an aggregator for a metric too (depends on specific label keys or pattern of name doesn't seem to be cool). Maybe open-telemetry/oteps#89 is the right solution.

[vmarchaud]

@legendecas Will do a quick example in a seperate PR. You are right, this system would be replaced by the View API later on