Telemetry/Available Telemetry Datasets and their Applications: Difference between revisions

Jump to navigation Jump to search

Telemetry/Available Telemetry Datasets and their Applications (view source)

Revision as of 21:39, 7 November 2016

4,701 bytes removed ,  7 November 2016
Deprecate wiki based d10n
(Added crash aggregates)
(Deprecate wiki based d10n)
Line 1: Line 1:
#REDIRECT [[https://github.com/mozilla/telemetry-batch-view/blob/master/docs/choosing_a_dataset.md]]
=Data Set Documentation=
=Data Set Documentation=
This document describes a set of datasets which can be queried using re:dash/sql.telemetry.mozilla.org (s.t.m.o). In addition, they can be queried using a Spark cluster - see [https://wiki.mozilla.org/Telemetry/Custom_analysis_with_spark#How_can_I_load_parquet_datasets_in_a_Jupyter_notebook.3F these directions]. The Longditudinal dataset is also available natively within Spark, see the [https://github.com/mozilla/emr-bootstrap-spark/blob/master/examples/Longitudinal%20Dataset%20Tutorial.ipynb longitudinal tutorial].
==Longitudinal==
[[Telemetry/LongitudinalExamples|Complete documentation]]
{{longitudinal data intro}}
==Main Summary==
[https://github.com/mozilla/telemetry-batch-view/blob/master/docs/MainSummary.md Complete Documentation]
Like the longitudinal dataset, main summary summarizes [https://gecko.readthedocs.io/en/latest/toolkit/components/telemetry/telemetry/data/main-ping.html main pings]. Each row corresponds to a single ping. This table does no sampling and includes all desktop pings.
===Caveats===
Querying against main summary on SQL.t.m.o/re:dash can '''impact performance for other users''' and can '''take a while to complete''' (~30m for simple queries). Since main summary includes a row for every ping, there are a large number of records which can consume a lot of resources on the shared cluster.
Instead, we recommend using the Longitudinal dataset where possible if querying from re:dash/s.t.m.o. The longitudinal dataset samples to 1% of all data and organized the data by client_id. In the odd case where these queries are necessary, limit to a short submission_date_s3 range and ideally make use of the sample_id field. Ideally, users who require this dataset would use Spark.
==Cross Sectional==
The Cross Sectional dataset is a simplified version of the Longitudinal dataset.
The majority of Longitudinal columns contain array values with one element for each ping, which is difficult to work with in SQL. The Cross Sectional dataset '''replaces these array-valued columns with summary statistics'''. To give an example, the Longitudinal dataset will contain a column named "geo_country" where each row is an array of locales for one client (e.g. array<"en_US", "en_US", "en_GB">). Instead, the Cross Sectional dataset includes a column named "geo_country_mode" where each row contains a single string representing the mode (e.g. "en_US"). The Cross Sectional column is '''easier to work with''' in SQL and is more representative than just choosing a single value from the Longitudinal array.
Note that the Cross Sectional dataset is derived from the Longitudinal dataset, so the dataset is a '''1% sample of main pings'''
This dataset is sometimes abbreviated as the '''xsec dataset'''. You can find the current version of the code [https://github.com/mozilla/telemetry-batch-view/blob/master/src/main/scala/com/mozilla/telemetry/views/CrossSectionalView.scala here]. This dataset is under active development, please '''contact rharter@mozilla.com with any questions'''.
==Client Count==
The Client Count dataset is simply a count of clients in a time period, separated out into a set of dimensions.
This is useful for questions similar to: ''How many X type of users were there during Y?'' - where X is some dimensions, and Y is some dates. Examples of X are: E10s Enabled, Operating System Type, or Country. For a complete list of dimensions, see [https://github.com/mozilla/telemetry-batch-view/blob/master/src/main/scala/com/mozilla/telemetry/views/ClientCountView.scala#L22 here].
Client Count does not contain a traditional int count column, instead the counts are stored as a HyperLogLogs in the hll column. The count of the hll is found using  <code>cardinality(cast(hll AS HLL))</code>, and different hll's can be merged using  <code>merge(cast(hll AS HLL))</code>. An example can be found in the [https://sql.telemetry.mozilla.org/queries/81/source#129 Firefox ER Reporting].
===Caveats===
Currently there is no Python wrapper for the HyperLogLog library, so the client count dataset is unavailable in Spark.
==Crash Aggregates==
[https://github.com/mozilla/telemetry-batch-view/blob/master/docs/CrashAggregateView.md Complete Documentation]
The Crash Aggregates dataset compiles crash statistics over various dimensions for each day. Example dimensions include channel and country, example statistics include usage hours and plugin crashes. See the [https://github.com/mozilla/telemetry-batch-view/blob/master/docs/CrashAggregateView.md complete documentation] for all available dimensions and statistics.
This dataset is good for queries of the form ''How many crashes did X types of users get during time Y?'' and ''Which types of users crashed the most during time Y?''.


==Mobile Metrics==
This document now lives here:
The android_events, android_clients, android_addons, and mobile_clients tables are documented here:
https://github.com/mozilla/telemetry-batch-view/blob/master/docs/choosing_a_dataset.md
https://wiki.mozilla.org/Mobile/Metrics/Redash
Harter
54

edits

Retrieved from "https://wiki.mozilla.org/Special:MobileDiff/1153861"

Navigation menu