Missing child payloads

Missing child payloads.ipynb

Missing child payloads.py

# coding: utf-8

# ### Missing child payloads

# In[2]:

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')


# ### Get pings

# In[44]:

submission_dates = ("20160410", "20160420")
fraction = 0.1
channel = "aurora"
ss_pings = get_pings(sc,
                     app="Firefox",
                     channel=channel,
                     doc_type="saved_session",
                     schema="v4",
                     submission_date=submission_dates,
                     fraction=fraction)
main_pings = get_pings(sc,
                     app="Firefox",
                     channel=channel,
                     doc_type="main",
                     schema="v4",
                     submission_date=submission_dates,
                     fraction=fraction)


# In[45]:

properties = ["clientId",
              "environment/settings/e10sEnabled",
              "meta/reason",
              "payload/childPayloads",
              "payload/keyedHistograms/SUBPROCESS_ABNORMAL_ABORT/content",
              "payload/info/sessionLength"]
ss_props = get_pings_properties(ss_pings, properties)
main_props = get_pings_properties(main_pings, properties)

(ss_props.count(), main_props.count())


# We only want to look at pings with e10s enabled.

# In[46]:

def e10s_enabled(ping):
    return ping.get("environment/settings/e10sEnabled", False)


# In[47]:

ss = ss_props.filter(e10s_enabled)
main = main_props.filter(e10s_enabled)                 .filter(lambda p: p.get("meta/reason") == "shutdown")

counts = (ss.count(), main.count())
counts


# Now look for pings with missing child payloads.

# In[48]:

def missing_childPayloads(ping):
    return None == ping.get("payload/childPayloads", None)

ss_missing = ss.filter(missing_childPayloads)
main_missing = main.filter(missing_childPayloads)

missing_counts = (ss_missing.count(), main_missing.count())
missing_counts


# In[49]:

def relative(sub, full):
    return (round(float(sub[0]) / full[0], 4), round(float(sub[1]) / full[1], 4))
relative(missing_counts, counts)


# ### Does it correlate with crashes?

# Maybe missing child payloads correlate with content crashes?

# In[50]:

def has_content_crash(p):
    return 0 < p.get("payload/keyedHistograms/SUBPROCESS_ABNORMAL_ABORT/content", 0)

ss_missing_crash = ss_missing.filter(has_content_crash)
main_missing_crash = main_missing.filter(has_content_crash)

missing_crash_counts = (ss_missing_crash.count(), main_missing_crash.count())
relative(missing_crash_counts, missing_counts)


# So, yes, the bulk of the missing child payloads can be blamed on content crashes.
# If the content process crashes, it won't send its Telemetry data to the parent process on shutdown.
# This still leaves >30% of the pings unexplained though.

# ### Correlation to session lengths?

# Another reason for missing child payloads could simply be short session lengths.
# If a session is too short we might not have spawned child processes or not have finished the Telemetry initialization in the child process etc.
# Let's check that.

# In[51]:

def mapper(p):
    return p.get("payload/info/sessionLength", -1)

ss_other = ss_missing.filter(lambda p: not has_content_crash(p))                     .map(mapper)
main_other = main_missing.filter(lambda p: not has_content_crash(p))                         .map(mapper)


# In[52]:

ss_series = pd.Series(ss_other.collect())
main_series = pd.Series(main_other.collect())


# In[53]:

ss_series.describe(percentiles=[.25, .5, .75, .95, .99])


# In[54]:

main_series.describe(percentiles=[.25, .5, .75, .95, .99])


# So, most of the remaining missing child payloads are correlating with short session lengths.
# We can investigate that in more detail for improved data quality, but this doesn't seem to be a big problem right now.

# In[55]:

ss_series.plot(kind='hist', bins=10, title="Histogram of session length distributions in saved-session")


# In[56]:

main_series.plot(kind='hist', bins=10, title="Histogram of session length distributions in shutdown")