1898336 - Consider not storing or sending empty/0-count buckets in `timing_distribution` and `memory_distribution` (functional-bucket histogram) metrics

Status: RESOLVED FIXED

(Data Platform and Tools :: Glean: SDK, task, P1)

Description

[Chris H-C :chutten|PTO (back Jan 5)]

We decided back in the annals of time to always send a contiguous set of buckets for functionally-bucketed histogram metrics (like timing_distribution and memory_distribution) in the Glean SDK. This was to ensure that a given ping would contain the entire information necessary to determine the bucket ranges.

In the five or so years since then, we've never really needed it. And now, especially since February with the removal of GVST in bug 1877836, we have so many more _distribution metrics that we're seeing the effect of this decision scaled to the size of our largest user populations (bug 1887755).

Let's reconsider this choice.

This bug is about writing a proposal (by the process), gathering feedback and information, and eventually hopefully kicking off work to stop storing and sending all those buckets that have zeroes in them.

Comment 1

[Chris H-C :chutten|PTO (back Jan 5)]

Draft up for initial team review: https://docs.google.com/document/d/1b3jIfVLS_ZP78gS-iW_XERIKCbTAZlAnED7vI89Dl7o/edit

Comment 2

[Travis Long [:travis_]]

No comments, sounds good to me as it is proposed.

Comment 3

[BMO Github Automation]

Attachment: [mozilla/glean] Bug 1898336 - (Memory|Timing) Distribution: snapshot only contains non-zero buckets now (#2848)

Comment 4

[Chris H-C :chutten|PTO (back Jan 5)]

chutten merged PR [mozilla/glean]: Bug 1898336 - (Memory|Timing) Distribution: snapshot only contains non-zero buckets now (#2848) in 36d1bb1.

(Will need a Glean SDK release and vendoring for projects to see the benefits)