Telemetry implications of per channel profiles

[Chris Hutten-Czapski]

I think it should be okay.

On a 1% sample of clients' pings from Monday (about 3.8M pings from 1.1M
clients) about 0.1% of clients reported in from more than one profile. (It
could be a little less than that given the slapdash method I'm using to
calculate this quickly)

If I expand to sample from pings received since the beginning of September
(133.8M pings from 3.3M clients), the proportion of clients reporting from
multiple channels rises to about 0.9%

(( Method: count distinct client ids and distinct client_id+channel pairs.
Overcounts the number of multi-channel clients if any switched channels,
tried a channel once and never did it again, or have used their profile on
more than two channels over the period ))

:chutten

On Thu, Oct 12, 2017 at 9:11 AM, Romain Testard <romain at mozilla.com> wrote:

> Hi all,
>
> We're planning on implementing dedicated profiles per channel (bug
> 1373244), hopefully in Firefox 59, to help address profile corruption
> issues our users can encounter when using a profile previously used on a
> later version.
> Once it gets implemented we'll have more profiles on telemetry and it
> probably impacts the interpretation we have of our telemetry data. A
> solution may be to implement a correlation ID between different profiles
> although current datasets and analysis would likely be impacted.
>
> If there are specific concerns about this please let me know since we’re
> still at scoping stage.
>
> Thanks,
> Romain
>
> _______________________________________________
> Fx-data-dev mailing list
> Fx-data-dev at mozilla.org
> https://mail.mozilla.org/listinfo/fx-data-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/fx-data-dev/attachments/20171013/e984589d/attachment.html>