openstack/swift
Code Issues Proposed changes
98d185905a
swift/test/probe/test_object_expirer.py
Samuel Merritt 745581ff2f Don't make async_pendings during object expiration 
After deleting an object, the object expirer deletes the corresponding
row from the expirer queue by making DELETE requests directly to the
container servers. The same thing happens after attempting to delete
an object, but failing because the object has already been deleted. If
the DELETE requests fail, then the expirer will encounter that row
again on its next pass and retry the DELETE at that time. Therefore,
it is not necessary for the object server to write an async_pending
for that queue row's deletion.

Currently, however, two of the object servers do write such
async_pendings. Given Rc container replicas, that's 2 * Rc updates
from async_pendings and another Rc from the object expirer
directly. Given a typical Rc of 3, that's 9 container updates per
expiring object.

This commit makes the object server write no async_pendings for DELETE
requests coming from the object expirer. This reduces the number of
container server requests to Rc (typically 3), all issued directly
from the object expirer.

Closes-Bug: 1076202
Change-Id: Icd63c80c73f864d2561e745c3154fbfda02bd0cc
2018-01-17 10:39:11 -08:00

374 lines
16 KiB
Python
Raw Blame History

#!/usr/bin/python -u
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import errno
import os
import random
import time
import uuid
import unittest
from swift.common.internal_client import InternalClient, UnexpectedResponse
from swift.common.manager import Manager
from swift.common.utils import Timestamp
from test.probe.common import ReplProbeTest, ENABLED_POLICIES
from test.probe.brain import BrainSplitter
from swiftclient import client
class TestObjectExpirer(ReplProbeTest):
def setUp(self):
self.expirer = Manager(['object-expirer'])
self.expirer.start()
err = self.expirer.stop()
if err:
raise unittest.SkipTest('Unable to verify object-expirer service')
conf_files = []
for server in self.expirer.servers:
conf_files.extend(server.conf_files())
conf_file = conf_files[0]
self.client = InternalClient(conf_file, 'probe-test', 3)
super(TestObjectExpirer, self).setUp()
self.container_name = 'container-%s' % uuid.uuid4()
self.object_name = 'object-%s' % uuid.uuid4()
self.brain = BrainSplitter(self.url, self.token, self.container_name,
self.object_name)
def _check_obj_in_container_listing(self):
for obj in self.client.iter_objects(self.account,
self.container_name):
if self.object_name == obj['name']:
return True
return False
@unittest.skipIf(len(ENABLED_POLICIES) < 2, "Need more than one policy")
def test_expirer_object_split_brain(self):
old_policy = random.choice(ENABLED_POLICIES)
wrong_policy = random.choice([p for p in ENABLED_POLICIES
if p != old_policy])
# create an expiring object and a container with the wrong policy
self.brain.stop_primary_half()
self.brain.put_container(int(old_policy))
self.brain.put_object(headers={'X-Delete-After': 2})
# get the object timestamp
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name,
headers={'X-Backend-Storage-Policy-Index': int(old_policy)})
create_timestamp = Timestamp(metadata['x-timestamp'])
self.brain.start_primary_half()
# get the expiring object updates in their queue, while we have all
# the servers up
Manager(['object-updater']).once()
self.brain.stop_handoff_half()
self.brain.put_container(int(wrong_policy))
# don't start handoff servers, only wrong policy is available
# make sure auto-created containers get in the account listing
Manager(['container-updater']).once()
# this guy should no-op since it's unable to expire the object
self.expirer.once()
self.brain.start_handoff_half()
self.get_to_final_state()
# validate object is expired
found_in_policy = None
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name,
acceptable_statuses=(4,),
headers={'X-Backend-Storage-Policy-Index': int(old_policy)})
self.assertIn('x-backend-timestamp', metadata)
self.assertEqual(Timestamp(metadata['x-backend-timestamp']),
create_timestamp)
# but it is still in the listing
self.assertTrue(self._check_obj_in_container_listing(),
msg='Did not find listing for %s' % self.object_name)
# clear proxy cache
client.post_container(self.url, self.token, self.container_name, {})
# run the expirer again after replication
self.expirer.once()
# object is not in the listing
self.assertFalse(self._check_obj_in_container_listing(),
msg='Found listing for %s' % self.object_name)
# and validate object is tombstoned
found_in_policy = None
for policy in ENABLED_POLICIES:
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name,
acceptable_statuses=(4,),
headers={'X-Backend-Storage-Policy-Index': int(policy)})
if 'x-backend-timestamp' in metadata:
if found_in_policy:
self.fail('found object in %s and also %s' %
(found_in_policy, policy))
found_in_policy = policy
self.assertIn('x-backend-timestamp', metadata)
self.assertGreater(Timestamp(metadata['x-backend-timestamp']),
create_timestamp)
def test_expirer_doesnt_make_async_pendings(self):
# The object expirer cleans up its own queue. The inner loop
# basically looks like this:
#
# for obj in stuff_to_delete:
# delete_the_object(obj)
# remove_the_queue_entry(obj)
#
# By default, upon receipt of a DELETE request for an expiring
# object, the object servers will create async_pending records to
# clean the expirer queue. Since the expirer cleans its own queue,
# this is unnecessary. The expirer can make requests in such a way
# tha the object server does not write out any async pendings; this
# test asserts that this is the case.
def gather_async_pendings(onodes):
async_pendings = []
for onode in onodes:
device_dir = self.device_dir('', onode)
for ap_pol_dir in os.listdir(device_dir):
if not ap_pol_dir.startswith('async_pending'):
# skip 'objects', 'containers', etc.
continue
async_pending_dir = os.path.join(device_dir, ap_pol_dir)
try:
ap_dirs = os.listdir(async_pending_dir)
except OSError as err:
if err.errno == errno.ENOENT:
pass
else:
raise
else:
for ap_dir in ap_dirs:
ap_dir_fullpath = os.path.join(
async_pending_dir, ap_dir)
async_pendings.extend([
os.path.join(ap_dir_fullpath, ent)
for ent in os.listdir(ap_dir_fullpath)])
return async_pendings
# Make an expiring object in each policy
for policy in ENABLED_POLICIES:
container_name = "expirer-test-%d" % policy.idx
container_headers = {'X-Storage-Policy': policy.name}
client.put_container(self.url, self.token, container_name,
headers=container_headers)
now = time.time()
delete_at = int(now + 2.0)
client.put_object(
self.url, self.token, container_name, "some-object",
headers={'X-Delete-At': str(delete_at),
'X-Timestamp': Timestamp(now).normal},
contents='dontcare')
time.sleep(2.0)
# make sure auto-created expirer-queue containers get in the account
# listing so the expirer can find them
Manager(['container-updater']).once()
# Make sure there's no async_pendings anywhere. Probe tests only run
# on single-node installs anyway, so this set should be small enough
# that an exhaustive check doesn't take too long.
all_obj_nodes = {}
for policy in ENABLED_POLICIES:
for dev in policy.object_ring.devs:
all_obj_nodes[dev['device']] = dev
pendings_before = gather_async_pendings(all_obj_nodes.values())
# expire the objects
Manager(['object-expirer']).once()
pendings_after = gather_async_pendings(all_obj_nodes.values())
self.assertEqual(pendings_after, pendings_before)
def test_expirer_object_should_not_be_expired(self):
# Current object-expirer checks the correctness via x-if-delete-at
# header that it can be deleted by expirer. If there are objects
# either which doesn't have x-delete-at header as metadata or which
# has different x-delete-at value from x-if-delete-at value,
# object-expirer's delete will fail as 412 PreconditionFailed.
# However, if some of the objects are in handoff nodes, the expirer
# can put the tombstone with the timestamp as same as x-delete-at and
# the object consistency will be resolved as the newer timestamp will
# be winner (in particular, overwritten case w/o x-delete-at). This
# test asserts such a situation that, at least, the overwriten object
# which have larger timestamp than the original expirered date should
# be safe.
def put_object(headers):
# use internal client to PUT objects so that X-Timestamp in headers
# is effective
headers['Content-Length'] = '0'
path = self.client.make_path(
self.account, self.container_name, self.object_name)
try:
self.client.make_request('PUT', path, headers, (2,))
except UnexpectedResponse as e:
self.fail(
'Expected 201 for PUT object but got %s' % e.resp.status)
obj_brain = BrainSplitter(self.url, self.token, self.container_name,
self.object_name, 'object', self.policy)
# T(obj_created) < T(obj_deleted with x-delete-at) < T(obj_recreated)
# < T(expirer_executed)
# Recreated obj should be appeared in any split brain case
obj_brain.put_container()
# T(obj_deleted with x-delete-at)
# object-server accepts req only if X-Delete-At is later than 'now'
# so here, T(obj_created) < T(obj_deleted with x-delete-at)
now = time.time()
delete_at = int(now + 2.0)
recreate_at = delete_at + 1.0
put_object(headers={'X-Delete-At': str(delete_at),
'X-Timestamp': Timestamp(now).normal})
# some object servers stopped to make a situation that the
# object-expirer can put tombstone in the primary nodes.
obj_brain.stop_primary_half()
# increment the X-Timestamp explicitly
# (will be T(obj_deleted with x-delete-at) < T(obj_recreated))
put_object(headers={'X-Object-Meta-Expired': 'False',
'X-Timestamp': Timestamp(recreate_at).normal})
# make sure auto-created containers get in the account listing
Manager(['container-updater']).once()
# sanity, the newer object is still there
try:
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name)
except UnexpectedResponse as e:
self.fail(
'Expected 200 for HEAD object but got %s' % e.resp.status)
self.assertIn('x-object-meta-expired', metadata)
# some object servers recovered
obj_brain.start_primary_half()
# sleep until after recreated_at
while time.time() <= recreate_at:
time.sleep(0.1)
# Now, expirer runs at the time after obj is recreated
self.expirer.once()
# verify that original object was deleted by expirer
obj_brain.stop_handoff_half()
try:
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name,
acceptable_statuses=(4,))
except UnexpectedResponse as e:
self.fail(
'Expected 404 for HEAD object but got %s' % e.resp.status)
obj_brain.start_handoff_half()
# and inconsistent state of objects is recovered by replicator
Manager(['object-replicator']).once()
# check if you can get recreated object
try:
metadata = self.client.get_object_metadata(
self.account, self.container_name, self.object_name)
except UnexpectedResponse as e:
self.fail(
'Expected 200 for HEAD object but got %s' % e.resp.status)
self.assertIn('x-object-meta-expired', metadata)
def _test_expirer_delete_outdated_object_version(self, object_exists):
# This test simulates a case where the expirer tries to delete
# an outdated version of an object.
# One case is where the expirer gets a 404, whereas the newest version
# of the object is offline.
# Another case is where the expirer gets a 412, since the old version
# of the object mismatches the expiration time sent by the expirer.
# In any of these cases, the expirer should retry deleting the object
# later, for as long as a reclaim age has not passed.
obj_brain = BrainSplitter(self.url, self.token, self.container_name,
self.object_name, 'object', self.policy)
obj_brain.put_container()
if object_exists:
obj_brain.put_object()
# currently, the object either doesn't exist, or does not have
# an expiration
# stop primary servers and put a newer version of the object, this
# time with an expiration. only the handoff servers will have
# the new version
obj_brain.stop_primary_half()
now = time.time()
delete_at = int(now + 2.0)
obj_brain.put_object({'X-Delete-At': str(delete_at)})
# make sure auto-created containers get in the account listing
Manager(['container-updater']).once()
# update object record in the container listing
Manager(['container-replicator']).once()
# take handoff servers down, and bring up the outdated primary servers
obj_brain.start_primary_half()
obj_brain.stop_handoff_half()
# wait until object expiration time
while time.time() <= delete_at:
time.sleep(0.1)
# run expirer against the outdated servers. it should fail since
# the outdated version does not match the expiration time
self.expirer.once()
# bring all servers up, and run replicator to update servers
obj_brain.start_handoff_half()
Manager(['object-replicator']).once()
# verify the deletion has failed by checking the container listing
self.assertTrue(self._check_obj_in_container_listing(),
msg='Did not find listing for %s' % self.object_name)
# run expirer again, delete should now succeed
self.expirer.once()
# verify the deletion by checking the container listing
self.assertFalse(self._check_obj_in_container_listing(),
msg='Found listing for %s' % self.object_name)
def test_expirer_delete_returns_outdated_404(self):
self._test_expirer_delete_outdated_object_version(object_exists=False)
def test_expirer_delete_returns_outdated_412(self):
self._test_expirer_delete_outdated_object_version(object_exists=True)
if __name__ == "__main__":
unittest.main()
View Git Blame Copy Permalink