def execute_workflow(self, workflow):
# Let's do it.
self.stdout.write('Running all indexation tasks')
os.environ['FORCE_INDEXING'] = '1'
try:
workflow.apply_async()
if not getattr(settings, 'CELERY_TASK_ALWAYS_EAGER', False):
time.sleep(10) # give celeryd some time to flag the DB
while is_reindexing_amo():
self.stdout.write('.')
self.stdout.flush()
time.sleep(5)
finally:
del os.environ['FORCE_INDEXING']
self.stdout.write('\n')
# Let's return the /_aliases values.
aliases = ES.indices.get_alias()
aliases = json.dumps(aliases, sort_keys=True, indent=4)
summary = _SUMMARY % aliases
self.stdout.write(summary)
def _test_reindexation(self):
# Current indices with aliases.
old_indices = self.get_indices_aliases()
# This is to start a reindexation in the background.
class ReindexThread(threading.Thread):
def __init__(self):
self.stdout = StringIO.StringIO()
super(ReindexThread, self).__init__()
def run(self):
# We need to wait at least a second, to make sure the alias
# name is going to be different, since we already create an
# alias in setUpClass.
time.sleep(1)
management.call_command('reindex', stdout=self.stdout)
t = ReindexThread()
t.start()
# Wait for the reindex in the thread to flag the database.
# The database transaction isn't shared with the thread, so force the
# commit.
while t.is_alive() and not is_reindexing_amo():
connection._commit()
connection.clean_savepoints()
# We should still be able to search in the foreground while the reindex
# is being done in the background. We should also be able to index new
# documents, and they should not be lost.
old_addons_count = len(self.expected)
while t.is_alive() and len(self.expected) < old_addons_count + 3:
self.expected.append(addon_factory())
connection._commit()
connection.clean_savepoints()
self.refresh()
self.check_results(self.expected)
if len(self.expected) == old_addons_count:
raise AssertionError('Could not index objects in foreground while '
'reindexing in the background.')
t.join() # Wait for the thread to finish.
t.stdout.seek(0)
stdout = t.stdout.read()
assert 'Reindexation done' in stdout, stdout
# The reindexation is done, let's double check we have all our docs.
connection._commit()
connection.clean_savepoints()
self.refresh()
self.check_results(self.expected)
# New indices have been created, and aliases now point to them.
new_indices = self.get_indices_aliases()
assert len(new_indices)
assert old_indices != new_indices, (stdout, old_indices, new_indices)
self.check_settings(new_indices)
def _test_reindexation(self):
# Current indices with aliases.
old_indices = self.get_indices_aliases()
# This is to start a reindexation in the background.
class ReindexThread(threading.Thread):
def __init__(self):
self.stdout = StringIO.StringIO()
super(ReindexThread, self).__init__()
def run(self):
# We need to wait at least a second, to make sure the alias
# name is going to be different, since we already create an
# alias in setUpClass.
time.sleep(1)
management.call_command('reindex', stdout=self.stdout)
t = ReindexThread()
t.start()
# Wait for the reindex in the thread to flag the database.
# The database transaction isn't shared with the thread, so force the
# commit.
while t.is_alive() and not is_reindexing_amo():
connection._commit()
connection.clean_savepoints()
# We should still be able to search in the foreground while the reindex
# is being done in the background. We should also be able to index new
# documents, and they should not be lost.
old_addons_count = len(self.expected)
while t.is_alive() and len(self.expected) < old_addons_count + 3:
self.expected.append(addon_factory())
connection._commit()
connection.clean_savepoints()
self.refresh()
self.check_results(self.expected)
if len(self.expected) == old_addons_count:
raise AssertionError('Could not index objects in foreground while '
'reindexing in the background.')
t.join() # Wait for the thread to finish.
t.stdout.seek(0)
stdout = t.stdout.read()
assert 'Reindexation done' in stdout, stdout
# The reindexation is done, let's double check we have all our docs.
connection._commit()
connection.clean_savepoints()
self.refresh()
self.check_results(self.expected)
# New indices have been created, and aliases now point to them.
new_indices = self.get_indices_aliases()
assert len(new_indices)
assert old_indices != new_indices, (stdout, old_indices, new_indices)
self.check_settings(new_indices)
def setUp(self):
super(TestIndexCommand, self).setUp()
if is_reindexing_amo():
unflag_reindexing_amo()
self.url = reverse('search.search')
# We store previously existing indices in order to delete the ones
# created during this test run.
self.indices = self.es.indices.status()['indices'].keys()
def setUp(self):
super(TestIndexCommand, self).setUp()
if is_reindexing_amo():
unflag_reindexing_amo()
self.url = reverse('search.search')
# We store previously existing indices in order to delete the ones
# created during this test run.
self.indices = self.es.indices.stats()['indices'].keys()
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a task chain that creates new indexes over the old ones so the
search feature works while the indexation occurs.
"""
force = kwargs['force']
if is_reindexing_amo() and not force:
raise CommandError('Indexation already occurring - use --force to '
'bypass')
alias = settings.ES_INDEXES.get(kwargs['key'], None)
if alias is None:
raise CommandError(
'Invalid --key parameter. It should be one of: %s.' % (
self.accepted_keys()
)
)
self.stdout.write('Starting the reindexation for %s.' % alias)
if kwargs['wipe']:
skip_confirmation = kwargs['noinput']
confirm = ''
if not skip_confirmation:
confirm = input('Are you sure you want to wipe all AMO '
'Elasticsearch indexes? (yes/no): ')
while confirm not in ('yes', 'no'):
confirm = input('Please enter either "yes" or "no": ')
if (confirm == 'yes' or skip_confirmation):
unflag_database()
# Retrieve the actual index and delete it. That way whether or
# not this was an alias or an index (which is wrong, but
# happens if data was indexed before the first reindex was
# done) doesn't matter.
try:
index = next(iter(ES.indices.get(alias)))
ES.indices.delete(index)
except NotFoundError:
pass
else:
raise CommandError('Aborted.')
elif force:
unflag_database()
workflow = self.create_workflow(alias)
self.execute_workflow(workflow)
def setUp(self):
super(TestIndexCommand, self).setUp()
if is_reindexing_amo():
unflag_reindexing_amo()
self.url = reverse_ns('addon-search')
# We store previously existing indices in order to delete the ones
# created during this test run.
self.indices = self.es.indices.stats()['indices'].keys()
self.addons = []
self.expected = self.addons[:]
# Monkeypatch Celerys ".get()" inside async task error
# until https://github.com/celery/celery/issues/4661 (which isn't just
# about retries but a general regression that manifests only in
# eager-mode) fixed.
self.patch('celery.app.task.denied_join_result')
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a task chain that creates new indexes over the old ones so the
search feature works while the indexation occurs.
"""
force = kwargs['force']
if is_reindexing_amo() and not force:
raise CommandError('Indexation already occurring - use --force to '
'bypass')
if kwargs['data'] not in ('addons', 'stats'):
raise CommandError('--data should be "addons" or "stats".')
alias = get_alias(kwargs['data'])
self.stdout.write('Starting the reindexation')
if kwargs['wipe']:
skip_confirmation = kwargs['noinput']
confirm = ''
if not skip_confirmation:
confirm = input('Are you sure you want to wipe all AMO '
'Elasticsearch indexes? (yes/no): ')
while confirm not in ('yes', 'no'):
confirm = input('Please enter either "yes" or "no": ')
if (confirm == 'yes' or skip_confirmation):
unflag_database()
ES.indices.delete(alias, ignore=404)
else:
raise CommandError('Aborted.')
elif force:
unflag_database()
workflow = self.create_workflow(alias)
self.execute_workflow(workflow)
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a Tasktree that creates new indexes
over the old ones so the search feature
works while the indexation occurs.
"""
force = kwargs.get('force', False)
if is_reindexing_amo() and not force:
raise CommandError('Indexation already occuring - use --force to '
'bypass')
log('Starting the reindexation', stdout=self.stdout)
modules = get_modules(with_stats=kwargs.get('with_stats', False))
if kwargs.get('wipe', False):
skip_confirmation = kwargs.get('noinput', False)
confirm = ''
if not skip_confirmation:
confirm = raw_input('Are you sure you want to wipe all AMO '
'Elasticsearch indexes? (yes/no): ')
while confirm not in ('yes', 'no'):
confirm = raw_input('Please enter either "yes" or "no": ')
if (confirm == 'yes' or skip_confirmation):
unflag_database(stdout=self.stdout)
for index in set(modules.keys()):
ES.indices.delete(index, ignore=404)
else:
raise CommandError("Aborted.")
elif force:
unflag_database(stdout=self.stdout)
alias_actions = []
def add_alias_action(action, index, alias):
action = {action: {'index': index, 'alias': alias}}
if action in alias_actions:
return
alias_actions.append(action)
# Creating a task tree.
log('Building the task tree', stdout=self.stdout)
tree = TaskTree()
last_action = None
to_remove = []
# For each alias, we create a new time-stamped index.
for alias, module in modules.items():
old_index = None
olds = ES.indices.get_aliases(alias, ignore=404)
for old_index in olds:
# Mark the index to be removed later.
to_remove.append(old_index)
# Mark the alias to be removed from that index.
add_alias_action('remove', old_index, alias)
# Create a new index, using the alias name with a timestamp.
new_index = timestamp_index(alias)
# If old_index is None that could mean it's a full index.
# In that case we want to continue index in it.
if ES.indices.exists(alias):
old_index = alias
# Flag the database.
step1 = tree.add_task(flag_database,
args=[new_index, old_index, alias])
step2 = step1.add_task(create_new_index, args=[alias, new_index])
step3 = step2.add_task(index_data, args=[alias, new_index])
last_action = step3
# Adding new index to the alias.
add_alias_action('add', new_index, alias)
# Alias the new index and remove the old aliases, if any.
renaming_step = last_action.add_task(update_aliases,
args=[alias_actions])
# Unflag the database - there's no need to duplicate the
# indexing anymore.
delete = renaming_step.add_task(unflag_database)
# Delete the old indexes, if any.
if to_remove:
delete.add_task(delete_indexes, args=[to_remove])
# Let's do it.
log('Running all indexation tasks', stdout=self.stdout)
os.environ['FORCE_INDEXING'] = '1'
# This is a bit convoluted, and more complicated than simply providing
# the soft and hard time limits on the @task decorator. But we're not
# using the @task decorator here, but a decorator from celery_tasktree.
if not getattr(settings, 'CELERY_ALWAYS_EAGER', False):
control.time_limit(
'olympia.lib.es.management.commands.reindex.index_data',
soft=time_limits['soft'],
hard=time_limits['hard'])
try:
tree.apply_async()
if not getattr(settings, 'CELERY_ALWAYS_EAGER', False):
time.sleep(10) # give celeryd some time to flag the DB
while is_reindexing_amo():
sys.stdout.write('.')
sys.stdout.flush()
time.sleep(5)
finally:
del os.environ['FORCE_INDEXING']
sys.stdout.write('\n')
# Let's return the /_aliases values.
aliases = ES.indices.get_aliases()
aliases = json.dumps(aliases, sort_keys=True, indent=4)
summary = _SUMMARY % (len(modules), aliases)
log(summary, stdout=self.stdout)
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a Tasktree that creates new indexes
over the old ones so the search feature
works while the indexation occurs.
"""
force = kwargs.get("force", False)
if is_reindexing_amo() and not force:
raise CommandError("Indexation already occuring - use --force to " "bypass")
log("Starting the reindexation", stdout=self.stdout)
modules = ["addons"]
if kwargs.get("with_stats", False):
modules.append("stats")
if kwargs.get("wipe", False):
skip_confirmation = kwargs.get("noinput", False)
confirm = ""
if not skip_confirmation:
confirm = raw_input("Are you sure you want to wipe all AMO " "Elasticsearch indexes? (yes/no): ")
while confirm not in ("yes", "no"):
confirm = raw_input('Please enter either "yes" or "no": ')
if confirm == "yes" or skip_confirmation:
unflag_database(stdout=self.stdout)
for index in set(MODULES[m].get_alias() for m in modules):
ES.indices.delete(index, ignore=404)
else:
raise CommandError("Aborted.")
elif force:
unflag_database(stdout=self.stdout)
alias_actions = []
def add_alias_action(action, index, alias):
action = {action: {"index": index, "alias": alias}}
if action in alias_actions:
return
alias_actions.append(action)
# Creating a task tree.
log("Building the task tree", stdout=self.stdout)
tree = TaskTree()
last_action = None
to_remove = []
# For each index, we create a new time-stamped index.
for module in modules:
old_index = None
alias = MODULES[module].get_alias()
olds = ES.indices.get_aliases(alias, ignore=404)
for old_index in olds:
# Mark the index to be removed later.
to_remove.append(old_index)
# Mark the alias to be removed from that index.
add_alias_action("remove", old_index, alias)
# Create a new index, using the alias name with a timestamp.
new_index = timestamp_index(alias)
# If old_index is None that could mean it's a full index.
# In that case we want to continue index in it.
if ES.indices.exists(alias):
old_index = alias
# Flag the database.
step1 = tree.add_task(flag_database, args=[new_index, old_index, alias])
step2 = step1.add_task(create_new_index, args=[module, new_index])
step3 = step2.add_task(index_data, args=[module, new_index])
last_action = step3
# Adding new index to the alias.
add_alias_action("add", new_index, alias)
# Alias the new index and remove the old aliases, if any.
renaming_step = last_action.add_task(update_aliases, args=[alias_actions])
# Unflag the database - there's no need to duplicate the
# indexing anymore.
delete = renaming_step.add_task(unflag_database)
# Delete the old indexes, if any.
if to_remove:
delete.add_task(delete_indexes, args=[to_remove])
# Let's do it.
log("Running all indexation tasks", stdout=self.stdout)
os.environ["FORCE_INDEXING"] = "1"
# This is a bit convoluted, and more complicated than simply providing
# the soft and hard time limits on the @task decorator. But we're not
# using the @task decorator here, but a decorator from celery_tasktree.
if not getattr(settings, "CELERY_ALWAYS_EAGER", False):
control.time_limit(
"olympia.lib.es.management.commands.reindex.index_data",
soft=time_limits["soft"],
hard=time_limits["hard"],
)
try:
tree.apply_async()
if not getattr(settings, "CELERY_ALWAYS_EAGER", False):
time.sleep(10) # give celeryd some time to flag the DB
while is_reindexing_amo():
sys.stdout.write(".")
sys.stdout.flush()
time.sleep(5)
finally:
del os.environ["FORCE_INDEXING"]
sys.stdout.write("\n")
# Let's return the /_aliases values.
aliases = ES.indices.get_aliases()
aliases = json.dumps(aliases, sort_keys=True, indent=4)
summary = _SUMMARY % (len(modules), aliases)
log(summary, stdout=self.stdout)
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a task chain that creates new indexes
over the old ones so the search feature
works while the indexation occurs.
"""
force = kwargs.get('force', False)
if is_reindexing_amo() and not force:
raise CommandError('Indexation already occurring - use --force to '
'bypass')
self.stdout.write('Starting the reindexation')
modules = get_modules(with_stats=kwargs.get('with_stats', False))
if kwargs.get('wipe', False):
skip_confirmation = kwargs.get('noinput', False)
confirm = ''
if not skip_confirmation:
confirm = raw_input('Are you sure you want to wipe all AMO '
'Elasticsearch indexes? (yes/no): ')
while confirm not in ('yes', 'no'):
confirm = raw_input('Please enter either "yes" or "no": ')
if (confirm == 'yes' or skip_confirmation):
unflag_database()
for index in set(modules.keys()):
ES.indices.delete(index, ignore=404)
else:
raise CommandError("Aborted.")
elif force:
unflag_database()
alias_actions = []
def add_alias_action(action, index, alias):
action = {action: {'index': index, 'alias': alias}}
if action in alias_actions:
return
alias_actions.append(action)
# Creating a task chain.
self.stdout.write('Building the task chain')
to_remove = []
workflow = []
# For each alias, we create a new time-stamped index.
for alias, module in modules.items():
old_index = None
try:
olds = ES.indices.get_alias(alias)
for old_index in olds:
# Mark the index to be removed later.
to_remove.append(old_index)
# Mark the alias to be removed from that index.
add_alias_action('remove', old_index, alias)
except NotFoundError:
# If the alias dit not exist, ignore it, don't try to remove
# it.
pass
# Create a new index, using the alias name with a timestamp.
new_index = timestamp_index(alias)
# If old_index is None that could mean it's a full index.
# In that case we want to continue index in it.
if ES.indices.exists(alias):
old_index = alias
# Flag the database.
workflow.append(
flag_database.si(new_index, old_index, alias) |
create_new_index.si(alias, new_index) |
index_data.si(alias, new_index)
)
# Adding new index to the alias.
add_alias_action('add', new_index, alias)
workflow = group(workflow)
# Alias the new index and remove the old aliases, if any.
workflow |= update_aliases.si(alias_actions)
# Unflag the database - there's no need to duplicate the
# indexing anymore.
workflow |= unflag_database.si()
# Delete the old indexes, if any.
if to_remove:
workflow |= delete_indexes.si(to_remove)
# Let's do it.
self.stdout.write('Running all indexation tasks')
os.environ['FORCE_INDEXING'] = '1'
try:
workflow.apply_async()
if not getattr(settings, 'CELERY_ALWAYS_EAGER', False):
time.sleep(10) # give celeryd some time to flag the DB
while is_reindexing_amo():
self.stdout.write('.')
self.stdout.flush()
time.sleep(5)
finally:
del os.environ['FORCE_INDEXING']
self.stdout.write('\n')
# Let's return the /_aliases values.
aliases = ES.indices.get_alias()
aliases = json.dumps(aliases, sort_keys=True, indent=4)
summary = _SUMMARY % (len(modules), aliases)
self.stdout.write(summary)
def handle(self, *args, **kwargs):
"""Reindexing work.
Creates a Tasktree that creates new indexes
over the old ones so the search feature
works while the indexation occurs.
"""
force = kwargs.get('force', False)
if is_reindexing_amo() and not force:
raise CommandError('Indexation already occuring - use --force to '
'bypass')
log('Starting the reindexation', stdout=self.stdout)
modules = get_modules(with_stats=kwargs.get('with_stats', False))
if kwargs.get('wipe', False):
skip_confirmation = kwargs.get('noinput', False)
confirm = ''
if not skip_confirmation:
confirm = raw_input('Are you sure you want to wipe all AMO '
'Elasticsearch indexes? (yes/no): ')
while confirm not in ('yes', 'no'):
confirm = raw_input('Please enter either "yes" or "no": ')
if (confirm == 'yes' or skip_confirmation):
unflag_database(stdout=self.stdout)
for index in set(modules.keys()):
ES.indices.delete(index, ignore=404)
else:
raise CommandError("Aborted.")
elif force:
unflag_database(stdout=self.stdout)
alias_actions = []
def add_alias_action(action, index, alias):
action = {action: {'index': index, 'alias': alias}}
if action in alias_actions:
return
alias_actions.append(action)
# Creating a task tree.
log('Building the task tree', stdout=self.stdout)
tree = TaskTree()
last_action = None
to_remove = []
# For each alias, we create a new time-stamped index.
for alias, module in modules.items():
old_index = None
olds = ES.indices.get_aliases(alias, ignore=404)
for old_index in olds:
# Mark the index to be removed later.
to_remove.append(old_index)
# Mark the alias to be removed from that index.
add_alias_action('remove', old_index, alias)
# Create a new index, using the alias name with a timestamp.
new_index = timestamp_index(alias)
# If old_index is None that could mean it's a full index.
# In that case we want to continue index in it.
if ES.indices.exists(alias):
old_index = alias
# Flag the database.
step1 = tree.add_task(flag_database,
args=[new_index, old_index, alias])
step2 = step1.add_task(create_new_index,
args=[alias, new_index])
step3 = step2.add_task(index_data,
args=[alias, new_index])
last_action = step3
# Adding new index to the alias.
add_alias_action('add', new_index, alias)
# Alias the new index and remove the old aliases, if any.
renaming_step = last_action.add_task(update_aliases,
args=[alias_actions])
# Unflag the database - there's no need to duplicate the
# indexing anymore.
delete = renaming_step.add_task(unflag_database)
# Delete the old indexes, if any.
if to_remove:
delete.add_task(delete_indexes, args=[to_remove])
# Let's do it.
log('Running all indexation tasks', stdout=self.stdout)
os.environ['FORCE_INDEXING'] = '1'
# This is a bit convoluted, and more complicated than simply providing
# the soft and hard time limits on the @task decorator. But we're not
# using the @task decorator here, but a decorator from celery_tasktree.
if not getattr(settings, 'CELERY_ALWAYS_EAGER', False):
control.time_limit(
'olympia.lib.es.management.commands.reindex.index_data',
soft=time_limits['soft'],
hard=time_limits['hard'])
try:
tree.apply_async()
if not getattr(settings, 'CELERY_ALWAYS_EAGER', False):
time.sleep(10) # give celeryd some time to flag the DB
while is_reindexing_amo():
sys.stdout.write('.')
sys.stdout.flush()
time.sleep(5)
finally:
del os.environ['FORCE_INDEXING']
sys.stdout.write('\n')
# Let's return the /_aliases values.
aliases = ES.indices.get_aliases()
aliases = json.dumps(aliases, sort_keys=True, indent=4)
summary = _SUMMARY % (len(modules), aliases)
log(summary, stdout=self.stdout)
