Closed Bug 1612954 Opened 6 years ago Closed 5 years ago

Report Android Build.DEVICE

Categories

(Data Platform and Tools :: Glean: SDK, enhancement, P3)

enhancement

Tracking

(Not tracked)

RESOLVED INCOMPLETE

People

(Reporter: tdsmith, Unassigned)

Details

(Whiteboard: [telemetry:glean-rs:backlog])

Mako does device model matching on Build.DEVICE (typically an opaque codename like "coral").

Glean reports device model as Build.MODEL, which is typically a user-facing product name like "Pixel 4".

Since there isn't a 1:1 correspondence between these, it would be nice if Glean reported the thing that Mako uses, which would help analysts debug experiment problems and do correct power analyses.

This doesn't feel urgent since having either is probably good-enough in practice, but it could avoid some confusion.

We're specifically reporting Build.MODEL since, at design time, we gathered from a-c's team that Build.MODEL is what should generally be reported to tell devices apart. Note that that value is what they've been reporting in Fennec and legacy service-telemetry as well

I think the real question is: does it make sense, for Mako, to target by "industrial design name"? (Build.DEVICE) Does a single design name map to multiple models?

For reference, I don't think there was any discussion on the Mako side with respect to that specific aspect. See this PR.

@Sebastian, do you know if Build.DEVICE can map to multiple Build.MODEL? Would it make more sense for glean to report the former instead of the latter?

Flags: needinfo?(s.kaspari)
Priority: -- → P3

@Sebastian, do you know if Build.DEVICE can map to multiple Build.MODEL?

Unfortunately I do not know.

I almost wonder if the opposite is possible. For example the "Galaxy S5" (which may be the MODEL?) has a lot of different variants (which may be the DEVICE?): SM-G900F, SM-G900I, SM-G900M, SM-G900T, SM-G900W8, SM-G900H, SM-G900FD, SM-G900P, SM-G900A, SC-04F, SM-G9008W, SM-G900L, SM-G900FQ, SM-G900K, SM-G900S, SCL23, SM-G900D, SM-G900MD, SM-G900V, SM-G900T3 (According to https://www.gsmarena.com/samsung_galaxy_s5-6033.php)

I think we can only make a good call here if we collect both temporarily and look at the real world data?

Would it make more sense for glean to report the former instead of the latter?

Good question. It may depend on the use case. MODEL is more human readable and often what frontend engineers care about ("Oh, this happens only on Pixel 4") - but for other use cases the exact variant may be important?

Flags: needinfo?(s.kaspari)

It looks like device and model have a many-to-many relationship: https://colab.research.google.com/drive/15B851JQz5LDfENzvmxG_I3YiLn_7LTRv

Most devices and most models relate to only one model or device, respectively, but either can correspond to dozens of the others. That doesn't necessarily mean that we'll ever care!

Whiteboard: [telemetry:glean-rs:m?] → [telemetry:glean-rs:backlog]
Status: NEW → RESOLVED
Closed: 5 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.