Recommend using a time-unbiased reservoir sampling algorithm for histograms#4678
Conversation
de93eba to
80f9b71
Compare
80f9b71 to
106246d
Compare
Co-authored-by: Tyler Yahn <[email protected]>
Co-authored-by: Tyler Yahn <[email protected]>
|
There is a concern from @tigrannajaryan that this is a breaking change. @jsuereth says there is language that allows us to make this change in the spec |
|
Feedback from @austinlparker: We should be careful about how we communicate this to users. Is there a way to make this opt-in for users? |
|
@carlosalberto requests that we leave this open for comment for a while. |
|
jmacd
left a comment
There was a problem hiding this comment.
Nice. This was a goal of mine in the original version, glad to see it improve.
I believe this is Jeffrey Vitter's "Algorithm R", for the record.
|
I might call this "time-unbiased" or "temporal uniform", not time weighted. |
|
This PR was marked stale due to lack of activity. It will be closed in 7 days. |
|
I think we are ready to merge this. |
### Metrics - `AlignedHistogramBucketExemplarReservoir` SHOULD use a time-weighted algorithm. ([#4678](#4678)) ### Profiles - Document the profiles signal. ([#4685](#4685)) ### Common - Extend the set of attribute value types to support more complex data structures. ([#4651](#4651)) --------- Co-authored-by: Armin Ruech <[email protected]>
Fixes #4675
Changes
Change the recommended algorithm for histogram reservoirs to be time-unbiased.
I've left the previous algorithm as an option to ensure this change is backwards-compatible.
Go prototype: open-telemetry/opentelemetry-go#7458
CHANGELOG.mdfile updated for non-trivial changes