1416364 - Stub attribution in new-profile pings are not set in the first main ping

Status: RESOLVED INVALID

(Toolkit :: Telemetry, defect, P1)

Description

[Anthony Miyaguchi [:amiyaguchi]]

Bug 1400870 fixed a bug with attribution that was caused by overwriting the environment section with attribution. This patch went into effect for nightly 58 and beta 57. Bug 1410181 comment #4 validates the effects of the patch on new-profile pings. In particular, nearly 99% of new-profiles in beta 57 contain some form of attribution. [1] 

If you filter main_summary on clients containing attribution in new-profiles containing this patch, a large majority of the first session pings will still be missing attribution. [2] This value should be very low if not 0.

[1] https://sql.telemetry.mozilla.org/queries/48906/source#131795
[2] https://sql.telemetry.mozilla.org/queries/48937/source

Comment 1

[Anthony Miyaguchi [:amiyaguchi]]

It should be clear that there should not be information loss since main_summary can be joined against new_profile pings. However, this reduces the quality of the main_summary dataset and increases the computational complexity of analysis against attribution.

Comment 2

[Chris More [:cmore]]

Can someone confirm if this bug is actually a bug or intended behavior now that bug 140870 is resolved?

Comment 3

[Alessio Placitelli [:Dexter]]

(In reply to Chris More [:cmore] from comment #2)
> Can someone confirm if this bug is actually a bug or intended behavior now
> that bug 140870 is resolved?

I'm looking into that and will update the bug soon.

Comment 4

[Anthony Miyaguchi [:amiyaguchi]]

After review of the query, it seems like there were two conditions that were missing.

Instead of a RIGHT JOIN, an INNER JOIN should have been performed against new-profile pings. This lead to the inclusion of pings in new-profiles but not in main pings.

Secondly, the structure attribution needed to be checked for {NULL, NULL, NULL, NULL} in addition to being NULL itself. 

> attribution IS NOT NULL AND coalesce(attribution.source, attribution.medium, attribution.content, attribution.campaign) IS NOT NULL

With this, the number of clients with missing attribution drops to 6% on a 1% sample, which may not necessarily be related to the first main ping.


https://sql.telemetry.mozilla.org/queries/49808/source