Add the MeterProvider and related packages for review LS-29752 #172
Conversation
…launcher-go into alt_metrics_sdk/provider_review
lightstep/sdk/metric/periodic.go
Outdated
| } | ||
| } | ||
|
|
||
| // Shutdown stops the stops the export loop, canceling its Context, |
| (*p) = append(*p, empty) | ||
| } | ||
| return &(*p)[len(*p)-1] | ||
| } |
There was a problem hiding this comment.
Is there cause for concern here that the slices we're reallocating from will have a high-water size behavior?
There was a problem hiding this comment.
Yes, that could be a concern. The library takes steps to ensure that the slices are re-used in the same order each time, except for the Points array at the leaf of the structure, which is random-order due to a map iteration. This makes it so that if there's a large allocation here, at least it will be re-used by the same instrument.
If the SDK is cumulative, the allocation will always grow and never shrink. If delta temporality is used, then if a high-cardinality event happens and then stops happening, there will be unused memory here.
| (*p) = append(*p, empty) | ||
| } | ||
| return &(*p)[len(*p)-1] | ||
| } |
There was a problem hiding this comment.
Yes, that could be a concern. The library takes steps to ensure that the slices are re-used in the same order each time, except for the Points array at the leaf of the structure, which is random-order due to a map iteration. This makes it so that if there's a large allocation here, at least it will be re-used by the same instrument.
If the SDK is cumulative, the allocation will always grow and never shrink. If delta temporality is used, then if a high-cardinality event happens and then stops happening, there will be unused memory here.
Description:
The MeterProvider package contains the entry-point to the metrics SDK. This package is accompanied by a number of helper packages, including the basic instrument and number, as well as View configuration structs. Not much happens in this code, it mainly connects the other pieces.
Link to tracking Issue:
LS-29760 describes the overall goal; this is one part.
Testing:
End-to-end testing was done. The alt_metrics_sdk/development branch has a assembled copy of this code. Depends on various Lightstep-side PRs merging to see the exponential histograms work.