973906 - Telemetry histograms by session

Status: RESOLVED FIXED

(Webtools Graveyard :: Telemetry Server, defect)

Description

[Benjamin Smedberg]

One of the very common requirements for UI measurements is to know what portion of the population is using a feature and in what amounts. A couple of examples:

* http://telemetry.mozilla.org/#nightly/30/PLUGINS_NOTIFICATION_SHOWN shows me how many times the plugins notification was shown across all session. But what I really need to know is the following:

** how many sessions didn't see the notification at all?
** how many sessions saw the notification once/twice
** how many sessions saw the notification many times

Similarly for the new "newtab page seen" measurement. Not only do I want to know how many times the new-tab page was seen at all, but I want to know the distribution: 50% of people never saw it, and 2% of people use it a lot, for example.

For right now I'd like to have help constructing custom mapreduce queries to answer these two questions, but also this is going to be a very common kind of request and it should be built into the default dashboard.

This is a followup to http://ask.mozilla.org/question/112/getting-percentages-from-telemetrymozillaorg/

Comment 1

[Mark Reid [:mreid]]

It sounds like what we need here is to show each reported state of a histogram, and how many times (or what proportion) that state was seen.

In general, this probably only makes sense for "boolean" and "flag" type histograms, since other types will probably have too many possible states to make comparison useful.

Comment 2

[Mark Reid [:mreid]]

Attachment: boolean_dist.py

Here is an example mapreduce job you can start from. As it stands, you would need to post-process the output depending on whether you want to look at the raw histogram data or the "interpreted" meanings of the histograms.

Comment 3

[Benjamin Smedberg]

https://github.com/mozilla/telemetry-server/blob/master/mapreduce/examples/distribution.py