slow_script_saturdays

slow_script_saturdays.ipynb

slow_script_saturdays.py

# coding: utf-8

# ### Slow Script Saturdays

# This is a small analysis of the `SLOW_SCRIPT_NOTICE_COUNT` histogram on two nightly builds: 20160903 and 20160910 (both Saturdays).

# In[1]:

import ujson as json
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import plotly.plotly as py
from plotly.graph_objs import *

from moztelemetry import get_pings, get_pings_properties, get_one_ping_per_client, get_clients_history, get_records

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


# In[2]:

sc.defaultParallelism


# ### 20160910

# In[11]:

pings = get_pings(sc, app="Firefox", channel="nightly", build_id=("20160910000000", "20160910999999"), fraction=1)


# ... and extract only the attributes we need from the Telemetry submissions:

# In[35]:

subset = get_pings_properties(pings, ["clientId",
                                      "environment/system/os/name",
                                      "payload/info/subsessionLength",
                                      "payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"])


# In[36]:

subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None)       .map(lambda p: (p["environment/system/os/name"], 1))       .countByKey()


# In[17]:

subset.count()


# In[19]:

usage = subset.map(lambda p: max(0, min(25, p["payload/info/subsessionLength"] / 3600))).reduce(lambda a,b: a+b)
usage


# In[21]:

notices = subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).map(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"]).reduce(lambda a,b: a+b)
notices


# In[38]:

reporting_pings = subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).count()
reporting_pings


# In[39]:

reporting_clients = subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).map(lambda p: p["clientId"]).distinct().count()
reporting_clients


# In[41]:

notices_per_ping = 1.0 * notices / subset.count()
notices_per_ping


# In[52]:

notices_per_reporting_client = 1.0 * notices / reporting_clients
notices_per_reporting_client


# In[53]:

notices_per_usage_hour = 1.0 * notices / usage
notices_per_usage_hour


# ### 20160903

# In[25]:

last_weeks_pings = get_pings(sc, app="Firefox", channel="nightly", build_id=("20160903000000", "20160903999999"), fraction=1)


# In[33]:

last_weeks_subset = get_pings_properties(last_weeks_pings, ["clientId",
                                      "environment/system/os/name",
                                      "payload/info/subsessionLength",
                                      "payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"])


# In[34]:

last_weeks_subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None)                  .map(lambda p: (p["environment/system/os/name"], 1))                  .countByKey()


# In[27]:

last_weeks_subset.count()


# In[28]:

last_weeks_usage = last_weeks_subset.map(lambda p: max(0, min(25, p["payload/info/subsessionLength"] / 3600))).reduce(lambda a,b: a+b)
last_weeks_usage


# In[29]:

last_weeks_notices = last_weeks_subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).map(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"]).reduce(lambda a,b: a+b)
last_weeks_notices


# In[42]:

last_weeks_reporting_pings = last_weeks_subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).count()
last_weeks_reporting_pings


# In[43]:

last_weeks_reporting_clients = last_weeks_subset.filter(lambda p: p["payload/histograms/SLOW_SCRIPT_NOTICE_COUNT"] is not None).map(lambda p: p["clientId"]).distinct().count()
last_weeks_reporting_clients


# In[46]:

last_weeks_notices_per_ping = 1.0 * last_weeks_notices / last_weeks_subset.count()
last_weeks_notices_per_ping


# In[51]:

last_weeks_notices_per_reporting_client = 1.0 * last_weeks_notices / last_weeks_reporting_clients
last_weeks_notices_per_reporting_client


# In[50]:

last_weeks_notices_per_usage_hour = 1.0 * last_weeks_notices / last_weeks_usage
last_weeks_notices_per_usage_hour


# ### Observations
# 
# 0903 and 0910's notices per usage hour metrics remained relatively stable (0.46 to 0.48), but everything else changed. 
# 
# Most alarming is the notices per reporting client measure which jumped from 10.7 to 31.1. This is consistent with going from a population where "many are having a poor time" to a population where "few are having a terrible time".