def test_trim_field(self):
uc = UpdateCount(addon_id=3615, count=1, date='2015-01-11')
self.command.trim_field(uc.versions) # Empty field.
assert not uc.versions
uc.versions = {'3.6': 123, '3.7': 321}
self.command.trim_field(uc.versions) # Small enough to fit in the db.
assert uc.versions == {'3.6': 123, '3.7': 321} # Unchanged.
very_long_key = 'x' * (2 ** 16)
uc.versions[very_long_key] = 1
self.command.trim_field(uc.versions) # Too big, must be trimmed.
assert uc.versions == {'3.6': 123, '3.7': 321} # Keep the most used.
uc.versions[very_long_key] = 1000 # Most used.
self.command.trim_field(uc.versions) # Too big, must be trimmed.
# Nothing left: least used removed, but still too big, so all the keys
# were removed.
assert uc.versions == {}
# Make sure we can store a very large field in the database.
long_key = 'x' * 65528 # This makes the dict barely fit in the db.
uc.versions[long_key] = 1
assert len(json.dumps(uc.versions)) == (2 ** 16) - 1
uc.save()
uc = UpdateCount.objects.get(pk=uc.pk) # Reload
# Fits in the database, so no truncation.
assert len(json.dumps(uc.versions)) == (2 ** 16) - 1
def test_trim_field(self):
uc = UpdateCount(addon_id=3615, count=1, date='2015-01-11')
self.command.trim_field(uc.versions) # Empty field.
assert not uc.versions
uc.versions = {'3.6': 123, '3.7': 321}
self.command.trim_field(uc.versions) # Small enough to fit in the db.
assert uc.versions == {'3.6': 123, '3.7': 321} # Unchanged.
very_long_key = 'x' * (2**16)
uc.versions[very_long_key] = 1
self.command.trim_field(uc.versions) # Too big, must be trimmed.
assert uc.versions == {'3.6': 123, '3.7': 321} # Keep the most used.
uc.versions[very_long_key] = 1000 # Most used.
self.command.trim_field(uc.versions) # Too big, must be trimmed.
# Nothing left: least used removed, but still too big, so all the keys
# were removed.
assert uc.versions == {}
# Make sure we can store a very large field in the database.
long_key = 'x' * 65528 # This makes the dict barely fit in the db.
uc.versions[long_key] = 1
assert len(json.dumps(uc.versions)) == (2**16) - 1
uc.save()
uc = UpdateCount.objects.get(pk=uc.pk) # Reload
# Fits in the database, so no truncation.
assert len(json.dumps(uc.versions)) == (2**16) - 1
def handle(self, *args, **options):
start = datetime.now() # Measure the time it takes to run the script.
day = options['date']
if not day:
day = (datetime.now() - timedelta(days=1)).strftime('%Y-%m-%d')
folder = args[0] if args else 'hive_results'
folder = path.join(settings.TMP_PATH, folder, day)
sep = options['separator']
groups = ('version', 'status', 'app', 'os', 'locale')
group_filepaths = []
# Make sure we're not trying to update with mismatched data.
for group in groups:
filepath = path.join(folder, 'update_counts_by_%s.hive' % group)
if get_date_from_file(filepath, sep) != day:
raise CommandError('%s file contains data for another day' %
filepath)
group_filepaths.append((group, filepath))
# First, make sure we don't have any existing counts for the same day,
# or it would just increment again the same data.
UpdateCount.objects.filter(date=day).delete()
# Memoize the addons and the UpdateCounts.
update_counts = {}
# Perf: preload all the addons once and for all.
# This builds a dict where each key (the addon guid we get from the
# hive query) has the addon_id as value.
guids_to_addon = (dict(Addon.objects.exclude(guid__isnull=True)
.exclude(type=amo.ADDON_PERSONA)
.values_list('guid', 'id')))
index = -1
for group, filepath in group_filepaths:
with codecs.open(filepath, encoding='utf8') as results_file:
for line in results_file:
index += 1
if index and (index % 1000000) == 0:
log.info('Processed %s lines' % index)
splitted = line[:-1].split(sep)
if ((group == 'app' and len(splitted) != 6)
or (group != 'app' and len(splitted) != 5)):
log.debug('Badly formatted row: %s' % line)
continue
if group == 'app':
day, addon_guid, app_id, app_ver, count, \
update_type = splitted
else:
day, addon_guid, data, count, update_type = splitted
addon_guid = addon_guid.strip()
if update_type:
update_type.strip()
# Old versions of Firefox don't provide the update type.
# All the following are "empty-like" values.
if update_type in ['0', 'NULL', 'None', '', '\N',
'%UPDATE_TYPE%']:
update_type = None
try:
count = int(count)
if update_type:
update_type = int(update_type)
except ValueError: # Badly formatted? Drop.
continue
# The following is magic that I don't understand. I've just
# been told that this is the way we can make sure a request
# is valid:
# > the lower bits for updateType (eg 112) should add to
# > 16, if not, ignore the request.
# > udpateType & 31 == 16 == valid request.
if update_type and update_type & 31 != 16:
log.debug("Update type doesn't add to 16: %s" %
update_type)
continue
# Does this addon exist?
if addon_guid and addon_guid in guids_to_addon:
addon_id = guids_to_addon[addon_guid]
else:
log.debug("Addon {guid} doesn't exist."
.format(guid=addon_guid.strip()))
continue
# Memoize the UpdateCount.
if addon_guid in update_counts:
uc = update_counts[addon_guid]
else:
uc = UpdateCount(date=day, addon_id=addon_id, count=0)
update_counts[addon_guid] = uc
# We can now fill the UpdateCount object.
if group == 'version':
# Take this count as the global number of daily users.
uc.count += count
uc.versions = update_inc(uc.versions, data, count)
elif group == 'status':
uc.statuses = update_inc(uc.statuses, data, count)
elif group == 'app':
# Applications is a dict of dicts, eg:
# {"{ec8030f7-c20a-464f-9b0e-13a3a9e97384}":
# {"10.0": 2, "21.0": 1, ....},
# "some other application guid": ...
# }
if uc.applications is None:
uc.applications = {}
app = uc.applications.get(app_id, {})
# Now overwrite this application's dict with
# incremented counts for its versions.
uc.applications.update(
{app_id: update_inc(app, app_ver, count)})
elif group == 'os':
uc.oses = update_inc(uc.oses, data, count)
elif group == 'locale':
# Drop incorrect locales sizes.
if len(data) > 10:
continue
# Collapse locales to `xx_yy` if possible.
data = data.strip().lower().replace('-', '_')
uc.locales = update_inc(uc.locales, data, count)
# Create in bulk: this is much faster.
UpdateCount.objects.bulk_create(update_counts.values(), 100)
log.info('Processed a total of %s lines' % (index + 1))
log.debug('Total processing time: %s' % (datetime.now() - start))
# Clean up files.
for _, filepath in group_filepaths:
log.debug('Deleting {path}'.format(path=filepath))
unlink(filepath)
def handle(self, *args, **options):
start = datetime.now() # Measure the time it takes to run the script.
day = options['date']
if not day:
day = (datetime.now() - timedelta(days=1)).strftime('%Y-%m-%d')
folder = args[0] if args else 'hive_results'
folder = path.join(settings.TMP_PATH, folder, day)
sep = options['separator']
groups = ('version', 'status', 'app', 'os', 'locale')
group_filepaths = []
# Make sure we're not trying to update with mismatched data.
for group in groups:
filepath = path.join(folder, 'update_counts_by_%s.hive' % group)
if get_date_from_file(filepath, sep) != day:
raise CommandError('%s file contains data for another day' %
filepath)
group_filepaths.append((group, filepath))
# First, make sure we don't have any existing counts for the same day,
# or it would just increment again the same data.
UpdateCount.objects.filter(date=day).delete()
# Memoize the addons and the UpdateCounts.
update_counts = {}
# Perf: preload all the addons once and for all.
# This builds a dict where each key (the addon guid we get from the
# hive query) has the addon_id as value.
guids_to_addon = (dict(Addon.objects.exclude(guid__isnull=True)
.exclude(type=amo.ADDON_PERSONA)
.values_list('guid', 'id')))
index = -1
for group, filepath in group_filepaths:
with codecs.open(filepath, encoding='utf8') as results_file:
for line in results_file:
index += 1
if index and (index % 1000000) == 0:
log.info('Processed %s lines' % index)
splitted = line[:-1].split(sep)
if ((group == 'app' and len(splitted) != 6)
or (group != 'app' and len(splitted) != 5)):
log.debug('Badly formatted row: %s' % line)
continue
if group == 'app':
day, addon_guid, app_id, app_ver, count, \
update_type = splitted
else:
day, addon_guid, data, count, update_type = splitted
addon_guid = addon_guid.strip()
if update_type:
update_type.strip()
# Old versions of Firefox don't provide the update type.
# All the following are "empty-like" values.
if update_type in ['0', 'NULL', 'None', '', '\N',
'%UPDATE_TYPE%']:
update_type = None
try:
count = int(count)
if update_type:
update_type = int(update_type)
except ValueError: # Badly formatted? Drop.
continue
# The following is magic that I don't understand. I've just
# been told that this is the way we can make sure a request
# is valid:
# > the lower bits for updateType (eg 112) should add to
# > 16, if not, ignore the request.
# > udpateType & 31 == 16 == valid request.
if update_type and update_type & 31 != 16:
log.debug("Update type doesn't add to 16: %s" %
update_type)
continue
# Does this addon exist?
if addon_guid and addon_guid in guids_to_addon:
addon_id = guids_to_addon[addon_guid]
else:
log.debug(u"Addon {guid} doesn't exist."
.format(guid=addon_guid.strip()))
continue
# Memoize the UpdateCount.
if addon_guid in update_counts:
uc = update_counts[addon_guid]
else:
uc = UpdateCount(date=day, addon_id=addon_id, count=0)
update_counts[addon_guid] = uc
# We can now fill the UpdateCount object.
if group == 'version':
# Take this count as the global number of daily users.
uc.count += count
uc.versions = update_inc(uc.versions, data, count)
elif group == 'status':
uc.statuses = update_inc(uc.statuses, data, count)
elif group == 'app':
# Applications is a dict of dicts, eg:
# {"{ec8030f7-c20a-464f-9b0e-13a3a9e97384}":
# {"10.0": 2, "21.0": 1, ....},
# "some other application guid": ...
# }
if uc.applications is None:
uc.applications = {}
app = uc.applications.get(app_id, {})
# Now overwrite this application's dict with
# incremented counts for its versions.
uc.applications.update(
{app_id: update_inc(app, app_ver, count)})
elif group == 'os':
uc.oses = update_inc(uc.oses, data, count)
elif group == 'locale':
# Drop incorrect locales sizes.
if len(data) > 10:
continue
uc.locales = update_inc(uc.locales, data, count)
# Create in bulk: this is much faster.
UpdateCount.objects.bulk_create(update_counts.values(), 100)
log.info('Processed a total of %s lines' % (index + 1))
log.debug('Total processing time: %s' % (datetime.now() - start))
# Clean up files.
for _, filepath in group_filepaths:
log.debug('Deleting {path}'.format(path=filepath))
unlink(filepath)