View - allow to change aggregation type

[cijothomas]

Child of overall Views feature
This is to track supporting changing Aggregation type using Views. (eg: User wants to aggregate Gauge as Histogram etc.)

[cijothomas]

As per the most recent spec, there is no ability to change temporality on a per instrument basis, so this issue is only to specify the aggregation itself, along with anything configurable for a particular aggregation.

[cijothomas]

Removing this from the upcoming 1.4.0 milestone. This is less important, so far haven't seen any asks for this capability.

[cijothomas]

Already support the ability to configuration for the Aggregation itself (like pick tags, change boundaries for histogram etc.)

[OQO]

We would need this to correctly make it work for New Relic. In some cases counters should have last value aggregation instead of the default sum aggregation to make it work correctly when using OTLP for New relic.

See their Java example:
https://github.com/newrelic/newrelic-opentelemetry-examples/pull/89/files#diff-da355ef6d1092534a55829e95160ab8468884bdd521f9018feeaaa66aea6ac5bR82-R86

I have not found a way to do this in the .net SDK. Are there any workarounds to set the aggregations in the mean time?

(Please also note that according to the SDK setting aggregation should be supported in views as an optional parameter:
https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/sdk.md#view )

[github-actions]

This issue was marked stale due to lack of activity and will be closed in 7 days. Commenting will instruct the bot to automatically remove the label. This bot runs once per day.

[cijothomas]

Unstale. This is spec requirement.

[bjrara]

+1 on this feature request.

In my scenario, I need to configure different aggregation to change instrument type like ObservableUpDownCounter to ObservableGauge because runtime metrics like process.runtime.dotnet.gc.heap.size from OpenTelemetry.Instrumentation.Runtime are reported as cumulative, but the values are actually gauge.

In the CHANGELOG, it mentions

  If your backend system distinguishes between ObservableUpDownCounter and
  ObservableGauge, then you may need to adjust your queries. Systems like
  Prometheus are unaffected by this.

With Views supporting custom aggregation, we can workaround the problem within SDK.