-
Notifications
You must be signed in to change notification settings - Fork 893
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changing OTLP exporters to export cumulative metrics by default #731
Comments
Following the discussion at yesterday's Metric SIG, I believe our best step forward is:
|
@huyan0 will file an OTEP stating this in more detail. |
I created a OTEP PR #131 to describe our conclusion from Metrics SIG meetings |
Discussed during the 03/30/2021 #128 Metrics Data Model SIG Mtg, moving this out of the Metrics Data Model, tracking this in the Metrics SDK project. |
This is done, right? |
Not yet, I believe #2013 is going to nail it. |
What are you trying to achieve?
Currently, OTLP exporters in SDKs are pass-through and send delta metrics directly. This makes Prometheus Exporter in the collector exports delta counter values as "gauge". Because of this behavior of the SDK exporter, it is then up to the collector to perform aggregations before exporting to cumulative backends like Prometheus. This feature does not yet exist, but has been proposed in #1422 in the collector repository.
However, in the case of multiple collector instances behind a load balancer, each individual collector possibly cannot aggregate cumulative values correctly, as metric events could be distributed across collectors.
If OTLP exporters export cumulative by default, then it guarantees that events from the same metric or timeseries are aggregated correctly before sending to the collector. The collector could then correctly export the metric to cumulative backends.
Additional context
related issues: Collector #1422, #1255
cc @alolita @huyan0 @jmacd
The text was updated successfully, but these errors were encountered: