Telemetry - Test Pilot KPI Validity Check

Telemetry - Test Pilot KPI Validity Check.ipynb

Telemetry - Test Pilot KPI Validity Check.py

# coding: utf-8

# # Telemetry - Telemetry KPI Validity Check
# 
# In this notebook we follow the standard for computing MAU and DAU as captured in [this notebook](https://gist.github.com/mreid-moz/8b2c2b1c6594d658ca5e).  This is to address [bug #1264049](https://bugzilla.mozilla.org/show_bug.cgi?id=1264049).  This notebook will create a csv that will be made available on [sql.telemetry.mozilla.org](https://sql.telemetry.mozilla.org).

# In[1]:

import boto3, botocore
from requests import get as fetch
from collections import defaultdict
from pyspark.sql import Row
import csv, os.path, json
from datetime import datetime as dt, timedelta, date
from pyspark.sql import SQLContext
from pyspark.sql.types import *
from pyspark.sql.functions import *
from boto3.s3.transfer import S3Transfer
from moztelemetry import get_pings, get_pings_properties, standards

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


# ## Define functions for the task

# In[47]:

def fmt(the_date, date_format="%Y%m%d"):
    return dt.strftime(the_date, date_format)

def convert_date(p):
    try:
        p['creationDate'] = dt.strptime(p.get('creationDate'), "%Y-%m-%dT%H:%M:%S.%fZ")
        if p['creationDate']:
            p['creationDate'] = standards.unix_time_nanos(p['creationDate'])
        return p
    except Exception as e:
        return e

def get_num_tests(p):
    result = len(p.get('payload/tests', None))
    if result is not None:
        p['num_tests'] = result
    return p

def get_tests(p):
    r = fetch('https://testpilot.firefox.com/api/experiments?format=json')
    active_tests = set([el['addon_id'] for el in dict(r.json())['results']])
    result = p.get('payload/tests', None)
    if result is not None:
        p['active_tests'] = [x for x in result if x in active_tests] # XXX Need to make sure they haven't uninstalled
    p['payload/tests'] = None
    return p


def process_pings(p):
    p = convert_date(p)
    p = get_num_tests(p)
    p = get_tests(p)
    return p

def should_be_updated(record,
        target_col="day",
        generated_col="generated_on",
        date_format="%Y%m%d"):
    target = dt.strptime(record[target_col], date_format)
    generated = dt.strptime(record[generated_col], date_format)
    
    # Don't regenerate data that was already updated today.
    today = fmt(dt.utcnow(), date_format)
    if record[generated_col] >= today:
        return False
    
    diff = generated - target
    return diff.days <= 10

def update_engagement_csv(dataset, old_filename, new_filename, 
                          cutoff_days=30, date_format="%Y%m%d"):
    cutoff_date = dt.utcnow() - timedelta(cutoff_days)
    cutoff = fmt(cutoff_date, date_format)
    print "Cutoff date: {}".format(cutoff)

    fields = ["day", "dau", "mau", "generated_on"]
    r = fetch('https://testpilot.firefox.com/api/experiments?format=json')
    active_tests = [el['addon_id'] for el in dict(r.json())['results']]

    active_test_names = [x.replace('mozilla.com', '').replace('@', '') for x in active_tests]
    fields.extend(["dau_" + x for x in active_test_names])
    fields.extend(["mau_" + x for x in active_test_names])

    should_write_header = True
    potential_updates = {}
    # Carry over rows we won't touch
    if os.path.exists(old_filename):
        with open(old_filename) as csv_old:
            reader = csv.DictReader(csv_old)
            with open(new_filename, "w") as csv_new:
                writer = csv.DictWriter(csv_new, fields)
                writer.writeheader()
                should_write_header = False
                for row in reader:
                    if row['day'] < cutoff:
                        writer.writerow(row)
                    else:
                        potential_updates[row['day']] = row

    with open(new_filename, "a") as csv_new:
        writer = csv.DictWriter(csv_new, fields)
        if should_write_header:
            writer.writeheader()

        for i in range(cutoff_days, 0, -1):
            target_day = fmt(dt.utcnow() - timedelta(i), date_format)
            if target_day in potential_updates and not should_be_updated(potential_updates[target_day]):
                # It's fine as-is.
                writer.writerow(potential_updates[target_day])
            else:
                # Update it.
                print "We should update data for {}".format(target_day)
                record = {"day": target_day, "generated_on": fmt(dt.utcnow(), date_format)}
                print "Starting dau {} at {}".format(target_day, dt.utcnow())
                record["dau"] = standards.dau(dataset, target_day)
                print "Finished dau {} at {}".format(target_day, dt.utcnow())
                print "Starting mau {} at {}".format(target_day, dt.utcnow())
                record["mau"] = standards.mau(dataset, target_day)
                print "Finished mau {} at {}".format(target_day, dt.utcnow())

                for el in active_tests:
                    key = el.replace('mozilla.com', '').replace("@", "")
                    test_df = dataset[array_contains(dataset.active_tests, el)]
                    print "Starting test dau {} at {}".format(target_day, dt.utcnow())
                    record['dau_' + key] = standards.dau(test_df, target_day)
                    print "Finished test dau {} at {}".format(target_day, dt.utcnow())
                    print "Starting test mau {} at {}".format(target_day, dt.utcnow())
                    record["mau_" + key] = standards.mau(test_df, target_day)
                    print "Finished test mau {} at {}".format(target_day, dt.utcnow())

                print record
#     test_count_dfs = defaultdict(pyspark.sql.DataFrame)
#     for el in active_tests:
#         test_count_dfs[el] = dataset[array_contains(dataset.active_tests, el)]

#    test_counts = dict.fromkeys(active_tests)
#    test_counts = {key: 0 for key in test_counts}

#    for k, v in dict(test_count_dfs).iteritems():
#        test_counts[k] = v.select('clientId').distinct().count()#standards.count_distinct_clientids(v)
        
#     print "We should update data for {}".format(target_day)
#     record = {"day": target_day, "generated_on": fmt(dt.utcnow(), date_format)}
#     print "Starting dau {} at {}".format(target_day, dt.utcnow())
#     record["dau"] = standards.dau(dataset, target_day)
#     print "Finished dau {} at {}".format(target_day, dt.utcnow())
#     print "Starting mau {} at {}".format(target_day, dt.utcnow())
#     record["mau"] = standards.mau(dataset, target_day)
#     print "Finished mau {} at {}".format(target_day, dt.utcnow())

#     for el in active_tests:
#         key = el.replace('mozilla.com', '').replace("@", "")
#         test_df = dataset[array_contains(dataset.active_tests, el)]
#         print "Starting dau {} at {}".format(target_day, dt.utcnow())
#         record['dau_' + key] = standards.dau(test_df, target_day)
#         print "Finished dau {} at {}".format(target_day, dt.utcnow())
#         print "Starting mau {} at {}".format(target_day, dt.utcnow())
#         record["mau_" + key] = standards.mau(test_df, target_day)
#         print "Finished mau {} at {}".format(target_day, dt.utcnow())
    
#     print record


# # Configure task variables

# In[3]:

data_bucket = "net-mozaws-prod-us-west-2-pipeline-analysis"
engagement_basename = "testpilot_engagement_ratio.csv"
new_engagement_basename = "testpilot_engagement_ratio.{}.csv".format(dt.strftime(dt.utcnow(), "%Y%m%d"))
s3path = "mreid/TxP_maudau"
data_version = "v1"

# Store the latest complete file in a subdir for hive / presto compatibility
engagement_key = "{}/{}/{}".format(s3path, data_version, engagement_basename)

new_engagement_key = "{}/{}".format(s3path, new_engagement_basename)

client = boto3.client('s3', 'us-west-2')
transfer = S3Transfer(client)


# # Retrieve list of current tests in the field

# In[4]:

r = fetch('https://testpilot.firefox.com/api/experiments?format=json')
active_tests = [el['addon_id'] for el in dict(r.json())['results']]
#active_tests = dict.fromkeys([el['addon_id'] for el in dict(r.json())['results']])
#active_tests = {key: 0 for key in active_tests}


# # Retrieve TxP pings and create DataFrame

# In[5]:

pings = get_pings(sc, doc_type="testpilot", app="Firefox")

subset = get_pings_properties(pings, ["clientId",
                                      "creationDate",
                                      "meta/submissionDate",
                                      "payload/tests"])

# Convert date string to appropriate format and count number of current tests
subset = subset.map(lambda p : process_pings(p))


# In[6]:

subset.take(3)


# In[10]:

sqlContext = SQLContext(sc)
rows = subset.map(lambda p: Row(clientId=p['clientId'], 
                                activityTimestamp=p['creationDate'], 
                                submission_date_s3=p['meta/submissionDate'],
                                num_tests=p['num_tests'],
                                active_tests=p['active_tests']))

dataset = sqlContext.createDataFrame(rows)
dataset.take(1)


# # Filter datasets for any tests and specific tests

# In[22]:

r = fetch('https://testpilot.firefox.com/api/experiments?format=json')
active_tests = [el['addon_id'] for el in dict(r.json())['results']]


# In[33]:

fields = ["day", "dau_txp", "mau_txp", "generated_on"]
r = fetch('https://testpilot.firefox.com/api/experiments?format=json')
active_tests = [el['addon_id'] for el in dict(r.json())['results']]
active_tests = [x.replace('mozilla.com', '') for x in active_tests]
active_tests = [x.replace('@', '') for x in active_tests]
fields.extend(["dau_" + x for x in active_tests])
fields.extend(["mau_" + x for x in active_tests])


# In[15]:

total_txp_gross = dataset.select('clientId').distinct().count()


# In[16]:

total_txp_net = dataset[dataset.num_tests > 0].select('clientId').distinct().count()


# In[17]:

test_count_dfs = defaultdict(pyspark.sql.DataFrame)
for el in active_tests:
    test_count_dfs[el] = dataset[array_contains(dataset.active_tests, el)]


# In[18]:

test_counts = dict.fromkeys(active_tests)
test_counts = {key: 0 for key in test_counts}

for k, v in dict(test_count_dfs).iteritems():
    test_counts[k] = v.select('clientId').distinct().count()#standards.count_distinct_clientids(v)    


# In[19]:

print "Clients with test pilot = {}".format(total_txp_gross)
print "Clients with at least 1 test = {}".format(total_txp_net)
for k, v in test_counts.iteritems():
    print "Clients with {test} = {count}".format(test=k, count=v)


# # Fetch existing data

# In[48]:

try:
    transfer.download_file(data_bucket, engagement_key, engagement_basename)
except botocore.exceptions.ClientError as e:
    # If the file wasn't there, that's ok. Otherwise, abort!
    if e.response['Error']['Code'] != "404":
        raise e
    else:
        print "Did not find an existing file at '{}'".format(engagement_key)

#%time update_engagement_csv(dataset_w_tests, engagement_basename, new_engagement_basename, cutoff_days=35)
get_ipython().magic(u'time update_engagement_csv(dataset, engagement_basename, new_engagement_basename, cutoff_days=35)')


# In[ ]:

## Upload the updated csv file to S3

# Update the day-specific file:
transfer.upload_file(new_engagement_basename, data_bucket, new_engagement_key)

# Update the "main" file
transfer.upload_file(new_engagement_basename, data_bucket, engagement_key)


# In[ ]: