OTLP native support

[songjiayang]

Is your feature request related to a problem? Please describe.

Currently, Cortex only supports Prometheus remote write protocol, with OTel is becoming more and more popular, more apps use OTel instrument SDK to export metrics but we can't push OTLP metrics to Cortex directly.

Describe the solution you'd like

Adding OTLPHttp exporter native support.

Describe alternatives you've considered

Although we can use OTel Collector to transform OTLP to Prometheus remote wirte protocol, which means we need extra maintenance an collector service.

[alanprot]

I think this makes lots of sense...

I just dont know if the OTLP data model can be translated to prometheus data model with no lost (histograms?)..

[friedrichg]

I just don't know if the OTLP data model can be translated to prometheus data model with no lost (histograms?)..

we can try to follow https://opentelemetry.io/docs/reference/specification/metrics/data-model/#otlp-metric-points-to-prometheus

There are Histograms and ExponentialHistograms. The latter is apparently a no go for prometheus.

But I think for experimental support we just do what's easy (counters, gauges, etc). Nobody is expecting full compatibility in our initial support.

[yeya24]

Based on Prometheus dev summit notes https://docs.google.com/document/d/11LC3wJcVk00l8w5P3oLQ-m3Y37iom6INAMEu2ZAGIIE/edit, Prometheus will support this natively soon and we can just reuse the handler. I recommend us to wait for upstream to not reinvent weels ourselves.

Goutham: Reconsider OTLP Ingest
The OpenTelemetry format is gaining adoption and I think we can make a lot of things simpler by adding an ingest layer similar to remote_write ingest to Prometheus.
Safeguards discussion
G: Behind flag not feature flag long term
Chris: gRPC is needed?
G: Yes, but not at first

CONSENSUS: We will implement this behind a feature flag according to the upstream OpenTelemetry Spec on OTLP → Prometheus datamodel mapping. We will have all appropriate safeguards in place before making it a first-class flag.

[yeya24]

FYI for the upstream OTLP PR prometheus/prometheus#11965.

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 60 days. It will be closed in 15 days if no further activity occurs. Thank you for your contributions.

[yeya24]

prometheus/prometheus#12571 Got merged and we can start integrating it.

I think Cortex cannot just reuse the handler as it is. We need to add some other code to ingest/convert otlp metrics to Cortex request format