Build Reversion

Build Reversion.ipynb

Build Reversion.py

# coding: utf-8
---
title: "Longitudinal Dataset Tutorial"
authors:
- vitillo
tags:
- tutorial
- examples
- dataset
- longitudinal
created_at: 2016-03-10
updated_at: 2016-06-24
tldr: Tutorial of how to use the Longitudinal Dataset
---
# ### Longitudinal Dataset Tutorial

# The longitudinal dataset is logically organized as a table where rows represent profiles and columns the various metrics (e.g. startup time). Each field of the table contains a list of values, one per Telemetry submission received for that profile.
# 
# The dataset is going to be regenerated from scratch every week, this allows us to apply non backward compatible changes to the schema and not worry about merging procedures. 
# 
# The current version of the longitudinal dataset has been build with all main pings received from 1% of profiles across all channels after mid November, which is shortly after Unified Telemetry landed. Future version will store up to 180 days of data.

# In[1]:

import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import plotly.plotly as py

get_ipython().magic(u'pylab inline')


# In[2]:

sc.defaultParallelism


# The longitudinal dataset can be accessed as a Spark [DataFrame](https://spark.apache.org/docs/1.5.2/api/python/pyspark.sql.html#pyspark.sql.DataFrame), which is a distributed collection of data organized into named columns. It is conceptually equivalent to a table in a relational database or a data frame in R/Python.

# In[4]:

frame = sqlContext.sql("SELECT client_id, profile_subsession_counter, build, settings FROM longitudinal")


# Number of profiles:

# In[5]:

frame.count()


# The dataset contains all histograms but it doesn't yet include all metrics stored in the various sections of the pings. See the [code](https://github.com/vitillo/telemetry-batch-view/blob/longitudinal/src/main/scala/streams/Longitudinal.scala#L68) that generates the dataset for a complete list of supported metrics. More metrics are going to be included in future versions of the dataset, inclusion of specific metrics can be prioritized by filing a bug.

# ### Scalar metrics

# A Spark bug is slowing down the *first* and *take* methods on a dataframe. A way around that for now is to first convert the dataframe to a rdd and then invoke *first* or *take*, e.g.:

# In[6]:

first = frame.rdd.first()

#filter("normalized_channel = 'release'")\
#    .select("build",
#            "system", 
#            "gc_ms",
#            "fxa_configured",
#            "browser_set_default_always_check",
#            "browser_set_default_dialog_prompt_rawcount")


# As mentioned earlier on, each field of the dataframe is an array containing one value per submission per client. The submissions are chronologically sorted.

# In[9]:

first.build


# In[12]:

def get_version(v):
    try:
        return int(v.split(".")[0])
    except ValueError:
        return None


# In[13]:

print get_version("")
print get_version("12.3")
print get_version("12")
print get_version("abcd")


# In[26]:

count_total = sc.accumulator(0)
count_backwardsversion = sc.accumulator(0)
count_channelswitch = sc.accumulator(0)
count_backwardsversion_releaseonly = sc.accumulator(0)

def mapper(row):
    channel_switch = False
    release_only = True
    backwardsversion = False

    # sessions are sorted by subsessionStartDate and then profileSubsessionCounter, newest-first.
    last_version = 99
    last_channel = None

    for settings in row.settings:
        channel = settings.update.channel
        if channel != "release":
            release_only = False
        if last_channel is None:
            last_channel = channel
        elif last_channel != channel:
            channel_switch = True

    for build in row.build:
        version = get_version(build.version)
        if version is not None:
            if version > last_version:
                backwardsversion = True
            last_version = version

    count_total.add(1)
    if backwardsversion:
        count_backwardsversion.add(1)
        if release_only:
            count_backwardsversion_releaseonly.add(1)
    if channel_switch:
        count_channelswitch.add(1)
frame.rdd.foreach(mapper)

total = float(count_total.value)
print "users that switched channels at all: {:.2f}%".format(count_channelswitch.value / total * 100)
print "users that reverted to an older version: {:.2f}%".format(count_backwardsversion.value / total * 100)
print "users that reverted to an older version, staying on the release channel: {:.2f}".format(count_backwardsversion_releaseonly.value / total * 100)