def __init__(self, conf, **kwargs):
self.conf = conf
self.ns = conf['namespace']
self.logger = get_logger(conf)
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
self.rdir = RdirClient(conf, logger=self.logger, **kwargs)
self._cs = None
def setUp(self):
super(TestMeta2Indexing, self).setUp()
self.rdir_client = RdirClient(self.conf)
self.directory_client = DirectoryClient(self.conf)
self.container_client = ContainerClient(self.conf)
self.containers = [random_str(14) for _ in range(0, randint(1, 10))]
self.containers_svcs = {}
self.event_agent_name = 'event-agent-1'
def __init__(self, conf, rdir_client=None, **kwargs):
self.conf = conf
self.ns = conf['namespace']
self.logger = get_logger(conf)
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
if rdir_client:
self.rdir = rdir_client
else:
self.rdir = RdirClient(conf, logger=self.logger, **kwargs)
self._cs = None
self._pool_options = None
def __init__(self, volume_path, conf, pool_manager=None):
"""
Initializes an Indexing worker for indexing meta2 databases.
Possible values of conf relating to this worker are:
- interval: (int) in sec time between two full scans. Default: half an
hour.
- report_interval: (int) in sec, time between two reports: Default: 300
- scanned_per_second: (int) maximum number of indexed databases /s.
- try_removing_faulty_indexes : In the event where we encounter a
database that's not supposed to be handled by this volume, attempt
to remove it from this volume rdir index if it exists
WARNING: The decision is based off of a proxy response, that could
be affected by cache inconsistencies for example, use at your own
risk. Default: False
:param volume_path: The volume path to be indexed
:param conf: The configuration to be passed to the needed services
:param pool_manager: A connection pool manager. If none is given, a
new one with a default size of 10 will be created.
"""
self.logger = get_logger(conf)
self._stop = False
self.volume = volume_path
self.success_nb = 0
self.failed_nb = 0
self.full_scan_nb = 0
self.last_report_time = 0
self.last_scan_time = 0
self.last_index_time = 0
self.start_time = 0
self.indexed_since_last_report = 0
self.scans_interval = int_value(
conf.get('interval'), 1800)
self.report_interval = int_value(
conf.get('report_interval'), 300)
self.max_indexed_per_second = int_value(
conf.get('scanned_per_second'), 3000)
self.namespace, self.volume_id = check_volume_for_service_type(
self.volume, "meta2")
self.attempt_bad_index_removal = boolean_value(
conf.get('try_removing_faulty_indexes', False)
)
if not pool_manager:
pool_manager = get_pool_manager(pool_connections=10)
self.index_client = RdirClient(conf, logger=self.logger,
pool_manager=pool_manager)
self.dir_client = DirectoryClient(conf, logger=self.logger,
pool_manager=pool_manager)
def __init__(self, namespace, **kwargs):
ep_parts = ["http:/",
load_namespace_conf(namespace).get('proxy'),
"v3.0",
namespace,
"content"]
super(CheckMeta2, self).__init__(namespace, "meta2",
endpoint="/".join(ep_parts), **kwargs)
self.account = AccountClient({"namespace": self.ns})
self.container = ContainerClient({"namespace": self.ns})
self.directory = DirectoryClient({"namespace": self.ns})
self.reference = random_buffer('0123456789ABCDEF', 64)
def __init__(self, namespace, **kwargs):
"""
Initialize the object storage API.
:param namespace: name of the namespace to interract with
:type namespace: `str`
:keyword connection_timeout: connection timeout towards rawx services
:type connection_timeout: `float` seconds
:keyword read_timeout: timeout for rawx responses and data reads from
the caller (when uploading)
:type read_timeout: `float` seconds
:keyword write_timeout: timeout for rawx write requests
:type write_timeout: `float` seconds
"""
self.namespace = namespace
self.connection_timeout = utils.float_value(
kwargs.get("connection_timeout"), None)
self.read_timeout = utils.float_value(kwargs.get("read_timeout"), None)
self.write_timeout = utils.float_value(kwargs.get("write_timeout"),
None)
# FIXME: share session between all the clients
self.directory = DirectoryClient({"namespace": self.namespace},
**kwargs)
self.account = AccountClient({"namespace": self.namespace}, **kwargs)
self.container = ContainerClient({"namespace": self.namespace},
**kwargs)
def __init__(self, conf, directory_client=None, **kwargs):
super(RdirClient, self).__init__(service_type='rdir', **kwargs)
self.directory = directory_client or DirectoryClient(conf, **kwargs)
self.ns = conf['namespace']
self._addr_cache = dict()
self.conf = conf
self._cs = None
def make_client(instance):
from oio.directory.client import DirectoryClient
endpoint = instance.get_endpoint('directory')
client = DirectoryClient({"namespace": instance.namespace},
endpoint=endpoint)
return client
def __init__(self, namespace, logger=None, **kwargs):
"""
Initialize the object storage API.
:param namespace: name of the namespace to interract with
:type namespace: `str`
:keyword connection_timeout: connection timeout towards rawx services
:type connection_timeout: `float` seconds
:keyword read_timeout: timeout for rawx responses and data reads from
the caller (when uploading)
:type read_timeout: `float` seconds
:keyword write_timeout: timeout for rawx write requests
:type write_timeout: `float` seconds
:keyword pool_manager: a pooled connection manager that will be used
for all HTTP based APIs (except rawx)
:type pool_manager: `urllib3.PoolManager`
"""
self.namespace = namespace
conf = {"namespace": self.namespace}
self.logger = logger or get_logger(conf)
self.timeouts = {tok: float_value(tov, None)
for tok, tov in kwargs.items()
if tok in self.__class__.TIMEOUT_KEYS}
from oio.account.client import AccountClient
from oio.container.client import ContainerClient
from oio.directory.client import DirectoryClient
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
self.container = ContainerClient(conf, logger=self.logger, **kwargs)
# In AccountClient, "endpoint" is the account service, not the proxy
acct_kwargs = kwargs.copy()
acct_kwargs["proxy_endpoint"] = acct_kwargs.pop("endpoint", None)
self.account = AccountClient(conf, logger=self.logger, **acct_kwargs)
def run(args):
pool = get_pool_manager()
v = vars(args)
dirclient = DirectoryClient(v)
backend = AccountBackend(v)
for entry, _, _, partial in full_list(backend, prefix=args.prefix):
if partial:
if args.verbose:
print(":%s: partial, skip" % entry)
continue
try:
dirclient.show(account=ACCOUNT, reference=entry)
if args.verbose:
print("%s: OK" % entry)
continue
except NotFound:
pass
except Exception as exc:
print("Exception not managed for %s: %s" % (entry, str(exc)))
continue
print("%s: meta2 not found" % entry)
if args.dry_run:
continue
data = {"dtime": time(), "name": entry}
# post event to Account service
res = pool.request('POST',
HOST +
'/v1.0/account/container/update?id=%s' % ACCOUNT,
headers={'Content-Type': 'application/json'},
body=json.dumps(data))
if res.status_int / 100 != 2:
print(res.status)
class TestDirectoryAPI(BaseTestCase):
def setUp(self):
super(TestDirectoryAPI, self).setUp()
self.api = DirectoryClient({'namespace': self.ns}, endpoint=self.uri)
def _create(self, name, metadata=None):
return self.api.create(self.account, name, properties=metadata)
def _delete(self, name):
self.api.delete(self.account, name)
def _clean(self, name, clear=False):
if clear:
# must clean properties before
self.api.del_properties(self.account, name, [])
self._delete(name)
def _get_properties(self, name, properties=None):
return self.api.get_properties(self.account,
name,
properties=properties)
def _set_properties(self, name, properties=None):
return self.api.set_properties(self.account,
name,
properties=properties)
def test_list(self):
# get on unknown reference
name = random_str(32)
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
self._create(name)
# get on existing reference
res = self.api.list(self.account, name)
self.assertIsNot(res['dir'], None)
self.assertIsNot(res['srv'], None)
self.assertEqual(res['name'], name)
self.assertEqual(res['account'], self.account)
self._delete(name)
# get on deleted reference
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_show_by_cid(self):
name = random_str(32)
self._create(name)
res = self.api.list(cid=cid_from_name(self.account, name))
self.assertIsNotNone(res['dir'])
self.assertIsNotNone(res['srv'])
self.assertEqual(res['name'], name)
self.assertEqual(res['account'], self.account)
self._delete(name)
def test_create(self):
name = random_str(32)
res = self._create(name)
self.assertEqual(res, True)
# second create
res = self._create(name)
self.assertEqual(res, False)
# clean
self._delete(name)
def test_create_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
res = self._create(name, metadata)
self.assertEqual(res, True)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
def test_create_without_account(self):
account = random_str(32)
name = random_str(32)
account_client = AccountClient(self.conf)
self.assertRaises(exc.NotFound, account_client.account_show, account)
self.api.create(account, name)
time.sleep(0.5) # ensure account event have been processed
self.assertEqual(account_client.account_show(account)['id'], account)
# clean
self.api.delete(account, name)
account_client.account_delete(account)
def test_delete(self):
name = random_str(32)
# delete on unknown reference
self.assertRaises(exc.NotFound, self.api.delete, self.account, name)
res = self._create(name)
self.assertEqual(res, True)
# delete on existing reference
self._delete(name)
# verify deleted
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
# second delete
self.assertRaises(exc.NotFound, self.api.delete, self.account, name)
# verify deleted
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_get_properties(self):
name = random_str(32)
# get_properties on unknown reference
self.assertRaises(exc.NotFound, self.api.get_properties, self.account,
name)
res = self._create(name)
self.assertEqual(res, True)
# get_properties on existing reference
data = self.api.get_properties(self.account, name)
self.assertEqual(data['properties'], {})
# get_properties
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
self._set_properties(name, metadata)
data = self.api.get_properties(self.account, name)
self.assertEqual(data['properties'], metadata)
# get_properties specify key
key, old_val = metadata.popitem()
data = self.api.get_properties(self.account, name, [key])
self.assertEqual(data['properties'], {key: old_val})
# clean
self._clean(name, True)
# get_properties on deleted reference
self.assertRaises(exc.NotFound, self.api.get_properties, self.account,
name)
def test_set_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
# set_properties on unknown reference
self.assertRaises(exc.NotFound, self.api.set_properties, self.account,
name, metadata)
res = self._create(name)
self.assertEqual(res, True)
# set_properties on existing reference
self.api.set_properties(self.account, name, metadata)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# set_properties
key = random_str(32)
value = random_str(32)
metadata2 = {key: value}
self._set_properties(name, metadata2)
metadata.update(metadata2)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# set_properties overwrite key
key, _ = metadata.popitem()
value = random_str(32)
metadata3 = {key: value}
metadata.update(metadata3)
self.api.set_properties(self.account, name, metadata3)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# set_properties overwrite key with empty value
key = list(metadata.keys())[0]
metadata4 = {key: ''}
del metadata[key]
self.api.set_properties(self.account, name, metadata4)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
# set_properties on deleted reference
self.assertRaises(exc.NotFound, self.api.set_properties, self.account,
name, metadata)
def test_del_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
# del_properties on unknown reference
self.assertRaises(exc.NotFound, self.api.del_properties, self.account,
name, [])
res = self._create(name, metadata)
self.assertEqual(res, True)
key, _ = metadata.popitem()
# del_properties on existing reference
self.api.del_properties(self.account, name, [key])
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# del_properties on unknown key
key = random_str(32)
# We do not check if a property exists before deleting it
# self.assertRaises(
# exc.NotFound, self.api.del_properties, self.account, name,
# [key])
self.api.del_properties(self.account, name, [key])
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
# del_properties on deleted reference
self.assertRaises(exc.NotFound, self.api.set_properties, self.account,
name, metadata)
def test_list_services(self):
# list_services on unknown reference
name = random_str(32)
echo = 'echo'
self.assertRaises(exc.NotFound,
self.api.list,
self.account,
name,
service_type=echo)
self._create(name)
# list_services on existing reference
res = self.api.list(self.account, name, service_type=echo)
self.assertIsNot(res['dir'], None)
self.assertIsNot(res['srv'], None)
self.assertEqual(res['name'], name)
self.assertEqual(res['account'], self.account)
self._delete(name)
# get on deleted reference
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_link_rdir_to_zero_scored_rawx(self):
disp = RdirDispatcher({'namespace': self.ns},
pool_manager=self.http_pool)
# Register a service, with score locked to zero
new_rawx = self._srv('rawx', {'tag.loc': _fake_location})
new_rawx['score'] = 0
self._register_srv(new_rawx)
self._reload_proxy()
all_rawx = disp.assign_all_rawx()
all_rawx_keys = [x['addr'] for x in all_rawx]
self.assertIn(new_rawx['addr'], all_rawx_keys)
rdir_addr = disp.rdir._get_rdir_addr(new_rawx['addr'])
self.assertIsNotNone(rdir_addr)
try:
self.api.unlink(RDIR_ACCT, new_rawx['addr'], 'rdir')
self.api.delete(RDIR_ACCT, new_rawx['addr'])
# self._flush_cs('rawx')
except Exception:
pass
def test_link_rdir_unachievable_min_dist(self):
disp = RdirDispatcher({'namespace': self.ns},
pool_manager=self.http_pool)
# Register a service, with score locked to zero
new_rawx = self._srv('rawx', {'tag.loc': _fake_location})
new_rawx['score'] = 90
self._register_srv(new_rawx)
self._reload_proxy()
self.assertRaises(exc.OioException, disp.assign_all_rawx, min_dist=4)
all_rawx, _ = disp.get_assignments('rawx')
all_rawx_keys = [x['addr'] for x in all_rawx]
self.assertIn(new_rawx['addr'], all_rawx_keys)
self.assertRaises(exc.VolumeException, disp.rdir._get_rdir_addr,
new_rawx['addr'])
def _generate_services(self, types, score=50):
all_srvs = dict()
for type_, count in types.items():
srvs = [
self._srv(type_, {'tag.loc': 'whatever%d' % i})
for i in range(count)
]
for srv in srvs:
srv['score'] = score
srv['id'] = _make_id(self.ns, type_, srv['addr'])
all_srvs[type_] = srvs
return all_srvs
def _test_link_rdir_fail_to_force(self, side_effects, expected_exc):
disp = RdirDispatcher({'namespace': self.ns})
# Mock rdir and rawx services so we do not pollute following tests
all_srvs = self._generate_services({'rdir': 3, 'rawx': 3})
def _all_services(type_, *args, **kwargs):
"""Return all mocked services of specified type"""
return all_srvs[type_]
def _poll(*args, **kwargs):
"""Pick one mocked random service"""
return [random.choice(all_srvs['rdir'])]
disp.cs.all_services = Mock(side_effect=_all_services)
disp.cs.poll = Mock(side_effect=_poll)
# Mock the check method to avoid calling the proxy
disp.directory.list = Mock(side_effect=exc.NotFound)
# Mock the assignation methods so we can check the calls
disp._smart_link_rdir = \
Mock(wraps=disp._smart_link_rdir)
disp.directory.force = \
Mock(wraps=disp.directory.force,
side_effect=side_effects)
# Expect an exception since some assignations will fail
self.assertRaises(expected_exc, disp.assign_all_rawx, max_attempts=1)
# But ensure all calls have been made
link_calls = [
call(rawx['addr'],
ANY,
max_per_rdir=ANY,
max_attempts=1,
min_dist=ANY,
service_type='rawx',
reassign=False) for rawx in all_srvs['rawx']
]
disp._smart_link_rdir.assert_has_calls(link_calls)
force_calls = \
[call(RDIR_ACCT, rawx['addr'], 'rdir', ANY, autocreate=True,
replace=ANY)
for rawx in all_srvs['rawx']]
disp.directory.force.assert_has_calls(force_calls)
def test_link_rdir_fail_to_force_one(self):
"""
Verify that the failure of one 'force' operation does
not break the whole operation.
"""
self._test_link_rdir_fail_to_force(
[exc.ServiceBusy(message='Failed :(', status=503), None, None],
exc.ServiceBusy)
def test_link_rdir_fail_to_force_several(self):
"""
Verify that the failure of two 'force' operations does
not break the whole operation.
"""
self._test_link_rdir_fail_to_force([
exc.ServiceBusy(message='Failed :(', status=503),
exc.OioTimeout('Timeout :('), None
], exc.OioException)
def test_rdir_repartition(self):
# FIXME(FVE): this test will fail if run after self._flush_cs('rawx')
client = RdirDispatcher({'namespace': self.ns})
self._reload_proxy()
all_rawx = client.assign_all_rawx()
self.assertGreater(len(all_rawx), 0)
by_rdir = dict()
total = 0
for rawx in all_rawx:
count = by_rdir.get(rawx['rdir']['addr'], 0)
total += 1
by_rdir[rawx['rdir']['addr']] = count + 1
avg = total / float(len(by_rdir))
print("Ideal number of bases per rdir: ", avg)
print("Current repartition: ", by_rdir)
for count in by_rdir.values():
self.assertLessEqual(count, avg + 1)
class RdirDispatcher(object):
def __init__(self, conf, **kwargs):
self.conf = conf
self.ns = conf['namespace']
self.logger = get_logger(conf)
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
self.rdir = RdirClient(conf, logger=self.logger, **kwargs)
self._cs = None
@property
def cs(self):
if not self._cs:
self._cs = ConscienceClient(self.conf, logger=self.logger)
return self._cs
def get_assignation(self):
all_rawx = self.cs.all_services('rawx')
all_rdir = self.cs.all_services('rdir', True)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x for x in all_rdir}
for rawx in all_rawx:
try:
# Verify that there is no rdir linked
resp = self.directory.list(RDIR_ACCT,
rawx['addr'],
service_type='rdir')
rdir_host = _filter_rdir_host(resp)
try:
rawx['rdir'] = by_id[_make_id(self.ns, 'rdir', rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to rawx %s seems down",
rdir_host, rawx['addr'])
rawx['rdir'] = {"addr": rdir_host, "tags": dict()}
by_id[_make_id(self.ns, 'rdir', rdir_host)] = rawx['rdir']
except NotFound:
self.logger.info("No rdir linked to %s", rawx['addr'])
return all_rawx, all_rdir
def assign_all_rawx(self, max_per_rdir=None):
"""
Find a rdir service for all rawx that don't have one already.
:param max_per_rdir: maximum number or rawx services that an rdir
can be linked to
:type max_per_rdir: `int`
"""
all_rawx = self.cs.all_services('rawx')
all_rdir = self.cs.all_services('rdir', True)
if len(all_rdir) <= 0:
raise ServiceUnavailable("No rdir service found in %s" % self.ns)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x for x in all_rdir}
for rawx in all_rawx:
try:
# Verify that there is no rdir linked
resp = self.directory.list(RDIR_ACCT,
rawx['addr'],
service_type='rdir')
rdir_host = _filter_rdir_host(resp)
try:
rawx['rdir'] = by_id[_make_id(self.ns, 'rdir', rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to rawx %s seems down",
rdir_host, rawx['addr'])
except (NotFound, ClientException):
if rawx['score'] <= 0:
self.logger.warn(
"rawx %s has score %s, and thus cannot be"
" affected a rdir (load balancer "
"limitation)", rawx['addr'], rawx['score'])
continue
rdir = self._smart_link_rdir(rawx['addr'], all_rdir,
max_per_rdir)
n_bases = by_id[rdir]['tags'].get("stat.opened_db_count", 0)
by_id[rdir]['tags']["stat.opened_db_count"] = n_bases + 1
rawx['rdir'] = by_id[rdir]
return all_rawx
def _smart_link_rdir(self, volume_id, all_rdir, max_per_rdir=None):
"""
Force the load balancer to avoid services that already host more
bases than the average (or more than `max_per_rdir`)
while selecting rdir services.
"""
opened_db = [
x['tags']['stat.opened_db_count'] for x in all_rdir
if x['score'] > 0
]
if len(opened_db) <= 0:
raise ServiceUnavailable("No valid rdir service found in %s" %
self.ns)
if not max_per_rdir:
upper_limit = sum(opened_db) / float(len(opened_db))
else:
upper_limit = max_per_rdir - 1
avoids = [
_make_id(self.ns, "rdir", x['addr']) for x in all_rdir if
x['score'] > 0 and x['tags']['stat.opened_db_count'] > upper_limit
]
known = [_make_id(self.ns, "rawx", volume_id)]
try:
polled = self._poll_rdir(avoid=avoids, known=known)
except ClientException as exc:
if exc.status != 481 or max_per_rdir:
raise
# Retry without `avoids`, hoping the next iteration will rebalance
polled = self._poll_rdir(known=known)
forced = {
'host': polled['addr'],
'type': 'rdir',
'seq': 1,
'args': "",
'id': polled['id']
}
self.directory.force(RDIR_ACCT,
volume_id,
'rdir',
forced,
autocreate=True)
try:
self.rdir.create(volume_id)
except Exception as exc:
self.logger.warn("Failed to create database for %s on %s: %s",
volume_id, polled['addr'], exc)
return polled['id']
def _poll_rdir(self, avoid=None, known=None):
"""Call the special rdir service pool (created if missing)"""
try:
svcs = self.cs.poll('__rawx_rdir', avoid=avoid, known=known)
except ClientException as exc:
if exc.status != 400:
raise
self.cs.lb.create_pool('__rawx_rdir', ((1, 'rawx'), (1, 'rdir')))
svcs = self.cs.poll('__rawx_rdir', avoid=avoid, known=known)
for svc in svcs:
# FIXME: we should include the service type in a dedicated field
if 'rdir' in svc['id']:
return svc
raise ServerException("LB returned incoherent result: %s" % svcs)
class RdirClient(HttpApi):
"""
Client class for rdir services.
"""
base_url = {
'rawx': 'rdir',
'meta2': 'rdir/meta2',
}
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(service_type='rdir', **kwargs)
self.directory = DirectoryClient(conf, **kwargs)
self.ns = conf['namespace']
self._addr_cache = dict()
def _clear_cache(self, volume_id):
self._addr_cache.pop(volume_id, None)
def _get_rdir_addr(self, volume_id, reqid=None):
# Initial lookup in the cache
if volume_id in self._addr_cache:
return self._addr_cache[volume_id]
# Not cached, try a direct lookup
try:
headers = {REQID_HEADER: reqid or request_id()}
resp = self.directory.list(RDIR_ACCT, volume_id,
service_type='rdir',
headers=headers)
host = _filter_rdir_host(resp)
# Add the new service to the cache
self._addr_cache[volume_id] = host
return host
except NotFound:
raise VolumeException('No rdir assigned to volume %s' % volume_id)
def _make_uri(self, action, volume_id, reqid=None, service_type='rawx'):
rdir_host = self._get_rdir_addr(volume_id, reqid)
return 'http://%s/v1/%s/%s' % (rdir_host,
self.__class__.base_url[service_type],
action)
@ensure_headers
@ensure_request_id
def _rdir_request(self, volume, method, action, create=False, params=None,
service_type='rawx', **kwargs):
if params is None:
params = dict()
params['vol'] = volume
if create:
params['create'] = '1'
uri = self._make_uri(action, volume,
reqid=kwargs['headers'][REQID_HEADER],
service_type=service_type)
try:
resp, body = self._direct_request(method, uri, params=params,
**kwargs)
except OioNetworkException:
self._clear_cache(volume)
raise
return resp, body
def create(self, volume_id, service_type='rawx', **kwargs):
"""Create the database for `volume_id` on the appropriate rdir"""
self._rdir_request(volume_id, 'POST', 'create',
service_type=service_type, **kwargs)
def chunk_push(self, volume_id, container_id, content_id, chunk_id,
headers=None, **data):
"""Reference a chunk in the reverse directory"""
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
for key, value in data.iteritems():
body[key] = value
self._rdir_request(volume_id, 'POST', 'push', create=True,
json=body, headers=headers)
def chunk_delete(self, volume_id, container_id, content_id, chunk_id,
**kwargs):
"""Unreference a chunk from the reverse directory"""
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
self._rdir_request(volume_id, 'DELETE', 'delete',
json=body, **kwargs)
def chunk_fetch(self, volume, limit=1000, rebuild=False,
container_id=None, max_attempts=3,
start_after=None, shuffle=False, **kwargs):
"""
Fetch the list of chunks belonging to the specified volume.
:param volume: the volume to get chunks from
:type volume: `str`
:param limit: maximum number of results to return per request
to the rdir server.
:type limit: `int`
:param rebuild: fetch only the chunks that were there
before the last incident.
:type rebuild: `bool`
:keyword container_id: get only chunks belonging to
the specified container
:type container_id: `str`
:keyword start_after: fetch only chunk that appear after
this container ID
:type start_after: `str`
"""
req_body = {'limit': limit}
if rebuild:
req_body['rebuild'] = True
if container_id:
req_body['container_id'] = container_id
if start_after:
req_body['start_after'] = start_after
while True:
for i in range(max_attempts):
try:
_resp, resp_body = self._rdir_request(
volume, 'POST', 'fetch', json=req_body, **kwargs)
break
except OioNetworkException:
# Monotonic backoff
if i < max_attempts - 1:
sleep(i * 1.0)
continue
# Too many attempts
raise
truncated = _resp.headers.get(
HEADER_PREFIX + 'list-truncated')
if truncated is None:
# TODO(adu): Delete when it will no longer be used
if not resp_body:
break
truncated = True
req_body['start_after'] = resp_body[-1][0]
else:
truncated = true_value(truncated)
if truncated:
req_body['start_after'] = _resp.headers[
HEADER_PREFIX + 'list-marker']
if shuffle:
random.shuffle(resp_body)
for (key, value) in resp_body:
container, content, chunk = key.split('|')
yield container, content, chunk, value
if not truncated:
break
def admin_incident_set(self, volume, date, **kwargs):
body = {'date': int(float(date))}
self._rdir_request(volume, 'POST', 'admin/incident',
json=body, **kwargs)
def admin_incident_get(self, volume, **kwargs):
_resp, body = self._rdir_request(volume, 'GET',
'admin/incident', **kwargs)
return body.get('date')
def admin_lock(self, volume, who, **kwargs):
body = {'who': who}
self._rdir_request(volume, 'POST', 'admin/lock', json=body, **kwargs)
def admin_unlock(self, volume, **kwargs):
self._rdir_request(volume, 'POST', 'admin/unlock', **kwargs)
def admin_show(self, volume, **kwargs):
_resp, body = self._rdir_request(volume, 'GET', 'admin/show',
**kwargs)
return body
def admin_clear(self, volume, clear_all=False, before_incident=False,
repair=False, **kwargs):
params = {'all': clear_all, 'before_incident': before_incident,
'repair': repair}
_resp, resp_body = self._rdir_request(
volume, 'POST', 'admin/clear', params=params, **kwargs)
return resp_body
def status(self, volume, max=1000, prefix=None, marker=None,
max_attempts=3, **kwargs):
"""
Get the status of chunks belonging to the specified volume.
:param volume: the volume to get chunks from
:type volume: `str`
:param max: maximum number of results to return per request
to the rdir server.
:type max: `int`
:keyword prefix: get only chunks belonging to
the specified prefix
:type prefix: `str`
:keyword marker: fetch only chunk that appear after
this marker
:type marker: `str`
"""
req_params = {'max': max}
if prefix:
req_params['prefix'] = prefix
if marker:
req_params['marker'] = marker
chunks = dict()
containers = dict()
while True:
for i in range(max_attempts):
try:
_resp, resp_body = self._rdir_request(
volume, 'GET', 'status', params=req_params, **kwargs)
break
except OioNetworkException:
# Monotonic backoff
if i < max_attempts - 1:
sleep(i * 1.0)
continue
# Too many attempts
raise
for (key, value) in resp_body.get('chunk', dict()).items():
chunks[key] = chunks.get(key, 0) + value
for (cid, info) in resp_body.get('container', dict()).items():
for (key, value) in info.items():
containers[cid][key] = containers.setdefault(
cid, dict()).get(key, 0) + value
if not true_value(_resp.headers.get(
HEADER_PREFIX + 'list-truncated')):
break
req_params['marker'] = _resp.headers[HEADER_PREFIX + 'list-marker']
return {'chunk': chunks, 'container': containers}
def meta2_index_create(self, volume_id, **kwargs):
"""
Create a new meta2 rdir index.
:param volume_id: The meta2 volume.
"""
return self.create(volume_id, service_type='meta2', **kwargs)
def meta2_index_push(self, volume_id, container_url, container_id, mtime,
**kwargs):
"""
Add a newly created container to the list of containers handled
by the meta2 server in question.
:param volume_id: The meta2 volume.
:param container_url: The container path (NS/account/container)
:param container_id: The container ID.
:param mtime: The last time it was spotted on this volume.
:param headers: Optional headers to pass along to the request.
"""
body = {'container_url': container_url,
'container_id': container_id,
'mtime': int(mtime)}
for key, value in kwargs.iteritems():
body[key] = value
return self._rdir_request(volume=volume_id, method='POST',
action='push', create=True, json=body,
service_type='meta2',
**kwargs)
def _resolve_cid_to_path(self, cid):
"""
Resolves a container ID into a a container path.
:param cid: The container ID.
:return: NS/account/container path.
"""
resp = self.directory.list(cid=cid)
return '{0}/{1}/{2}'.format(
self.ns,
resp['account'],
resp['name']
)
def meta2_index_delete(self, volume_id, container_path=None,
container_id=None, **kwargs):
"""
Remove a meta2 record from the volume's index. Either the container ID
or the container path have to be given.
:param volume_id: The meta2 volume.
:param container_id: The container ID.
:param container_path: The container path
"""
if not container_path and container_id:
container_path = self._resolve_cid_to_path(container_id)
elif container_path and not container_id:
_tmp = container_path.rsplit("/")
container_id = cid_from_name(_tmp[1], _tmp[3])
elif not container_path and not container_id:
raise ValueError("At least the container ID or the container path "
"should be given.")
body = {'container_url': container_path,
'container_id': container_id}
for key, value in kwargs.iteritems():
body[key] = value
return self._rdir_request(volume=volume_id, method='POST',
action='delete', create=False, json=body,
service_type='meta2', **kwargs)
def meta2_index_fetch(self, volume_id, prefix=None, marker=None,
limit=4096, **kwargs):
"""
Fetch specific meta2 records, or a range of records.
:param volume_id: The meta2 volume.
:param prefix: The prefix all meta2 records should have.
:param marker: The container path from which the API will start the
listing. The marker will not be included in the result.
:param limit: The number of records to be returned. Capped at 4096
:return: A dictionary containing the following entries:
- records: A list containing the actual records.
- truncated: A boolean value representing whether there
are still records left that fulfill this query.
"""
params = {}
if prefix:
params['prefix'] = prefix
if marker:
# FIXME(ABO): Validate this one.
params['marker'] = marker
if limit:
params['limit'] = limit
_resp, body = self._rdir_request(volume=volume_id, method='POST',
action='fetch', json=params,
service_type='meta2', **kwargs)
return body
def meta2_index_fetch_all(self, volume_id, **kwargs):
"""
A wrapper around meta2_index_fetch that loops until no more records
are available, returning all the records in a certain volume's index.
WARNING: For testing purposes only
"""
return depaginate(
self.meta2_index_fetch,
volume_id=volume_id,
listing_key=lambda x: x['records'],
truncated_key=lambda x: x['truncated'],
# The following is only called when the list is truncated
# So we can assume there are records in the list
marker_key=lambda x: x['records'][-1]['container_url'],
**kwargs
)
class RdirClient(Client):
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.autocreate = true_value(conf.get('autocreate', True))
self.directory_client = DirectoryClient(conf)
# TODO keep rdir addr in local cache to avoid lookup requests
def _get_rdir_addr(self, volume_id):
try:
resp = self.directory_client.show(acct='_RDIR',
ref=volume_id,
srv_type='rdir')
except NotFound as e:
if self.autocreate:
self.directory_client.link('_RDIR',
volume_id,
'rdir',
autocreate=True)
resp = self.directory_client.show(acct='_RDIR',
ref=volume_id,
srv_type='rdir')
else:
raise e
for srv in resp['srv']:
if srv['type'] == 'rdir':
return srv['host']
raise ClientException("No rdir service found")
def _make_uri(self, action, volume_id):
rdir_host = self._get_rdir_addr(volume_id)
uri = 'http://%s/v1/%s/%s?vol=%s' % (rdir_host, self.ns, action,
volume_id)
return uri
def _rdir_request(self, volume, method, action, **kwargs):
uri = self._make_uri(action, volume)
resp, body = self._direct_request(method, uri, **kwargs)
return resp, body
def chunk_push(self, volume_id, container_id, content_id, chunk_id,
**data):
body = {
'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id
}
for key, value in data.iteritems():
body[key] = value
headers = {}
self._rdir_request(volume_id,
'POST',
'rdir/push',
json=body,
headers=headers)
def chunk_delete(self, volume_id, container_id, content_id, chunk_id):
body = {
'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id
}
self._rdir_request(volume_id, 'DELETE', 'rdir/delete', json=body)
def chunk_fetch(self, volume, limit=100, rebuild=False):
req_body = {'limit': limit}
if rebuild:
req_body['rebuild'] = True
while True:
resp, resp_body = self._rdir_request(volume,
'POST',
'rdir/fetch',
json=req_body)
resp.raise_for_status()
if len(resp_body) == 0:
break
for (key, value) in resp_body:
container, content, chunk = key.split('|')
yield container, content, chunk, value
req_body['start_after'] = key
def admin_incident_set(self, volume, date):
body = {'date': date}
self._rdir_request(volume, 'POST', 'rdir/admin/incident', json=body)
def admin_incident_get(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET',
'rdir/admin/incident')
return resp_body.get('date')
def admin_lock(self, volume, who):
body = {'who': who}
self._rdir_request(volume, 'POST', 'rdir/admin/lock', json=body)
def admin_unlock(self, volume):
self._rdir_request(volume, 'POST', 'rdir/admin/unlock')
def admin_show(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'rdir/admin/show')
return resp_body
def make_client(instance):
client = DirectoryClient({"namespace": instance.namespace},
endpoint=instance.get_endpoint('directory'))
return client
def reference(self):
if not self._reference:
self._reference = DirectoryClient(self.conf)
return self._reference
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(**kwargs)
self.directory = DirectoryClient(conf, **kwargs)
self.ns = conf['namespace']
self._addr_cache = dict()
def setUp(self):
super(TestDirectoryAPI, self).setUp()
self.api = DirectoryClient({'namespace': self.ns}, endpoint=self.uri)
class TestDirectoryAPI(BaseTestCase):
def setUp(self):
super(TestDirectoryAPI, self).setUp()
self.api = DirectoryClient({'namespace': self.ns}, endpoint=self.uri)
def _create(self, name, metadata=None):
return self.api.create(self.account, name, properties=metadata)
def _delete(self, name):
self.api.delete(self.account, name)
def _clean(self, name, clear=False):
if clear:
# must clean properties before
self.api.del_properties(self.account, name, [])
self._delete(name)
def _get_properties(self, name, properties=None):
return self.api.get_properties(
self.account, name, properties=properties)
def _set_properties(self, name, properties=None):
return self.api.set_properties(
self.account, name, properties=properties)
def test_list(self):
# get on unknown reference
name = random_str(32)
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
self._create(name)
# get on existing reference
res = self.api.list(self.account, name)
self.assertIsNot(res['dir'], None)
self.assertIsNot(res['srv'], None)
self._delete(name)
# get on deleted reference
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_create(self):
name = random_str(32)
res = self._create(name)
self.assertEqual(res, True)
# second create
res = self._create(name)
self.assertEqual(res, False)
# clean
self._delete(name)
def test_create_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
res = self._create(name, metadata)
self.assertEqual(res, True)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
def test_create_without_account(self):
account = random_str(32)
name = random_str(32)
account_client = AccountClient(self.conf)
self.assertRaises(exc.NotFound, account_client.account_show, account)
self.api.create(account, name)
time.sleep(0.5) # ensure account event have been processed
self.assertEqual(account_client.account_show(account)['id'],
account)
# clean
self.api.delete(account, name)
account_client.account_delete(account)
def test_delete(self):
name = random_str(32)
# delete on unknown reference
self.assertRaises(exc.NotFound, self.api.delete, self.account, name)
res = self._create(name)
self.assertEqual(res, True)
# delete on existing reference
self._delete(name)
# verify deleted
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
# second delete
self.assertRaises(exc.NotFound, self.api.delete, self.account, name)
# verify deleted
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_get_properties(self):
name = random_str(32)
# get_properties on unknown reference
self.assertRaises(
exc.NotFound, self.api.get_properties, self.account, name)
res = self._create(name)
self.assertEqual(res, True)
# get_properties on existing reference
data = self.api.get_properties(self.account, name)
self.assertEqual(data['properties'], {})
# get_properties
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
self._set_properties(name, metadata)
data = self.api.get_properties(self.account, name)
self.assertEqual(data['properties'], metadata)
# get_properties specify key
key = metadata.keys().pop(0)
data = self.api.get_properties(self.account, name, [key])
self.assertEqual(data['properties'], {key: metadata[key]})
# clean
self._clean(name, True)
# get_properties on deleted reference
self.assertRaises(
exc.NotFound, self.api.get_properties, self.account, name)
def test_set_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
# set_properties on unknown reference
self.assertRaises(
exc.NotFound, self.api.set_properties, self.account, name,
metadata)
res = self._create(name)
self.assertEqual(res, True)
# set_properties on existing reference
self.api.set_properties(self.account, name, metadata)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# set_properties
key = random_str(32)
value = random_str(32)
metadata2 = {key: value}
self._set_properties(name, metadata2)
metadata.update(metadata2)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# set_properties overwrite key
key = metadata.keys().pop(0)
value = random_str(32)
metadata3 = {key: value}
metadata.update(metadata3)
self.api.set_properties(self.account, name, metadata3)
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
# set_properties on deleted reference
self.assertRaises(
exc.NotFound, self.api.set_properties, self.account, name,
metadata)
def test_del_properties(self):
name = random_str(32)
metadata = {
random_str(32): random_str(32),
random_str(32): random_str(32),
}
# del_properties on unknown reference
self.assertRaises(
exc.NotFound, self.api.del_properties, self.account, name, [])
res = self._create(name, metadata)
self.assertEqual(res, True)
key = metadata.keys().pop()
del metadata[key]
# del_properties on existing reference
self.api.del_properties(self.account, name, [key])
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# del_properties on unknown key
key = random_str(32)
# We do not check if a property exists before deleting it
# self.assertRaises(
# exc.NotFound, self.api.del_properties, self.account, name,
# [key])
self.api.del_properties(self.account, name, [key])
data = self._get_properties(name)
self.assertEqual(data['properties'], metadata)
# clean
self._clean(name, True)
# del_properties on deleted reference
self.assertRaises(
exc.NotFound, self.api.set_properties, self.account, name,
metadata)
def test_list_services(self):
# list_services on unknown reference
name = random_str(32)
echo = 'echo'
self.assertRaises(
exc.NotFound, self.api.list, self.account, name,
service_type=echo)
self._create(name)
# list_services on existing reference
res = self.api.list(self.account, name, service_type=echo)
self.assertIsNot(res['dir'], None)
self.assertIsNot(res['srv'], None)
self._delete(name)
# get on deleted reference
self.assertRaises(exc.NotFound, self.api.list, self.account, name)
def test_rdir_linking_old(self):
"""
Tests that rdir services linked to rawx services
are not on the same locations
"""
self.skipTest('Deprecated way of linking rdir services')
self._reload_proxy()
cs = ConscienceClient({'namespace': self.ns})
rawx_list = cs.all_services('rawx')
rdir_dict = {x['addr']: x for x in cs.all_services('rdir')}
# Link the services
for rawx in rawx_list:
self.api.link('_RDIR_TEST', rawx['addr'], 'rdir',
autocreate=True)
# Do the checks
for rawx in rawx_list:
linked_rdir = self.api.list(
'_RDIR_TEST', rawx['addr'], service_type='rdir')['srv']
rdir = rdir_dict[linked_rdir[0]['host']]
rawx_loc = rawx['tags'].get('tag.loc')
rdir_loc = rdir['tags'].get('tag.loc')
self.assertNotEqual(rawx_loc, rdir_loc)
# Unlink the services
for rawx in rawx_list:
self.api.unlink('_RDIR_TEST', rawx['addr'], 'rdir')
self.api.delete('_RDIR_TEST', rawx['addr'])
def test_link_rdir_to_zero_scored_rawx(self):
client = RdirClient({'namespace': self.ns})
disp = RdirDispatcher({'namespace': self.ns})
# Register a service, with score locked to zero
new_rawx = self._srv('rawx', {'tag.loc': 'whatever'})
new_rawx['score'] = 0
self._register_srv(new_rawx)
self._reload_proxy()
all_rawx = disp.assign_all_rawx()
all_rawx_keys = [x['addr'] for x in all_rawx]
self.assertIn(new_rawx['addr'], all_rawx_keys)
rdir_addr = client._get_rdir_addr(new_rawx['addr'])
self.assertIsNotNone(rdir_addr)
try:
self.api.unlink('_RDIR', new_rawx['addr'], 'rdir')
self.api.delete('_RDIR', new_rawx['addr'])
# self._flush_cs('rawx')
except Exception:
pass
def test_rdir_repartition(self):
client = RdirDispatcher({'namespace': self.ns})
self._reload_proxy()
all_rawx = client.assign_all_rawx()
self.assertGreater(len(all_rawx), 0)
by_rdir = dict()
total = 0
for rawx in all_rawx:
count = by_rdir.get(rawx['rdir']['addr'], 0)
total += 1
by_rdir[rawx['rdir']['addr']] = count + 1
avg = total / float(len(by_rdir))
print("Ideal number of bases per rdir: ", avg)
print("Current repartition: ", by_rdir)
for count in by_rdir.itervalues():
self.assertLessEqual(count, avg + 1)
class RdirClient(Client):
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.autocreate = true_value(conf.get('autocreate', True))
self.directory_client = DirectoryClient(conf)
# TODO keep rdir addr in local cache to avoid lookup requests
def _get_rdir_addr(self, volume_id):
try:
resp = self.directory_client.show(acct='_RDIR', ref=volume_id,
srv_type='rdir')
except NotFound as e:
if self.autocreate:
self.directory_client.link('_RDIR', volume_id, 'rdir',
autocreate=True)
resp = self.directory_client.show(acct='_RDIR', ref=volume_id,
srv_type='rdir')
else:
raise e
for srv in resp['srv']:
if srv['type'] == 'rdir':
return srv['host']
raise ClientException("No rdir service found")
def _make_uri(self, action, volume_id):
rdir_host = self._get_rdir_addr(volume_id)
uri = 'http://%s/v1/%s/%s?vol=%s' % (
rdir_host, self.ns, action, volume_id)
return uri
def _rdir_request(self, volume, method, action, **kwargs):
uri = self._make_uri(action, volume)
resp, body = self._direct_request(method, uri, **kwargs)
return resp, body
def chunk_push(self, volume_id, container_id, content_id, chunk_id,
**data):
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
for key, value in data.iteritems():
body[key] = value
headers = {}
self._rdir_request(volume_id, 'POST', 'rdir/push',
json=body, headers=headers)
def chunk_delete(self, volume_id, container_id, content_id, chunk_id):
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
self._rdir_request(volume_id, 'DELETE', 'rdir/delete', json=body)
def chunk_fetch(self, volume, limit=100, rebuild=False):
req_body = {'limit': limit}
if rebuild:
req_body['rebuild'] = True
while True:
resp, resp_body = self._rdir_request(volume, 'POST', 'rdir/fetch',
json=req_body)
resp.raise_for_status()
if len(resp_body) == 0:
break
for (key, value) in resp_body:
container, content, chunk = key.split('|')
yield container, content, chunk, value
req_body['start_after'] = key
def admin_incident_set(self, volume, date):
body = {'date': date}
self._rdir_request(volume, 'POST', 'rdir/admin/incident', json=body)
def admin_incident_get(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET',
'rdir/admin/incident')
return resp_body.get('date')
def admin_lock(self, volume, who):
body = {'who': who}
self._rdir_request(volume, 'POST', 'rdir/admin/lock', json=body)
def admin_unlock(self, volume):
self._rdir_request(volume, 'POST', 'rdir/admin/unlock')
def admin_show(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'rdir/admin/show')
return resp_body
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.autocreate = true_value(conf.get('autocreate', True))
self.directory_client = DirectoryClient(conf)
class TestMeta2Indexing(BaseTestCase):
def setUp(self):
super(TestMeta2Indexing, self).setUp()
self.rdir_client = RdirClient(self.conf)
self.directory_client = DirectoryClient(self.conf)
self.container_client = ContainerClient(self.conf)
self.containers = [random_str(14) for _ in range(0, randint(1, 10))]
self.containers_svcs = {}
self.event_agent_name = 'event-agent-1'
def tearDown(self):
super(TestMeta2Indexing, self).tearDown()
self._containers_cleanup()
self._service(self.event_agent_name, 'start', wait=3)
def _containers_cleanup(self):
for container in self.containers:
self.container_client.container_delete(self.account, container)
for svc in self.containers_svcs[container]:
self.rdir_client.meta2_index_delete(
volume_id=svc['host'],
container_path="{0}/{1}/{2}".format(
self.ns, self.account, container),
container_id=cid_from_name(self.account, container))
def _filter_by_managing_svc(self, all_containers, svc_of_interest):
"""
Filters through the containers returning only those that have
svc_of_interest in their list of managing services.
"""
containers_list = []
for key in all_containers.keys():
if svc_of_interest in [x['host'] for x in all_containers[key]]:
containers_list.append(key)
return sorted(containers_list)
def test_volume_indexing_worker(self):
"""
Test steps:
- Generate a list of container names and create them
- Collect their respective meta2 servers
- For each meta2 server:
- Run a meta2 indexing worker
- List all rdir index records and match then with the
services we're expecting.
:return:
"""
self._service(self.event_agent_name, "stop", wait=3)
for container in self.containers:
self.container_client.container_create(account=self.account,
reference=container)
for container in self.containers:
self.containers_svcs[container] = [
x
for x in self.directory_client.list(account=self.account,
reference=container)['srv']
if x['type'] == 'meta2'
]
meta2_data_paths = {}
for svc in self.conf['services']['meta2']:
svc_host = svc.get('service_id', svc['addr'])
meta2_data_paths[svc_host] = svc['path']
distinct_meta2_servers = set()
for svc_list in self.containers_svcs.values():
for svc in svc_list:
distinct_meta2_servers.add(svc['host'])
for svc in distinct_meta2_servers:
expected_containers = self._filter_by_managing_svc(
self.containers_svcs, svc)
worker = Meta2IndexingWorker(meta2_data_paths[svc], self.conf)
worker.crawl_volume()
indexed_containers = sorted([
x['container_url'].split('/')[-1]
for x in self.rdir_client.meta2_index_fetch_all(volume_id=svc)
])
for cont in expected_containers:
self.assertIn(cont, indexed_containers)
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.autocreate = true_value(conf.get('autocreate', True))
self.directory_client = DirectoryClient(conf)
def directory(self):
if not self._directory:
self._directory = DirectoryClient(self.conf)
return self._directory
class RdirDispatcher(object):
def __init__(self, conf, rdir_client=None, **kwargs):
self.conf = conf
self.ns = conf['namespace']
self.logger = get_logger(conf)
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
if rdir_client:
self.rdir = rdir_client
else:
self.rdir = RdirClient(conf, logger=self.logger, **kwargs)
self._cs = None
self._pool_options = None
@property
def cs(self):
if not self._cs:
self._cs = ConscienceClient(self.conf, logger=self.logger,
pool_manager=self.rdir.pool_manager)
return self._cs
def get_assignments(self, service_type, **kwargs):
"""
Get rdir assignments for all services of the specified type.
:returns: a tuple with a list all services of the specified type,
and a list of all rdir services.
:rtype: `tuple<list<dict>,list<dict>>`
"""
all_services = self.cs.all_services(service_type, **kwargs)
all_rdir = self.cs.all_services('rdir', True, **kwargs)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x
for x in all_rdir}
for service in all_services:
try:
ref = service.get('tags', {}).get('tag.service_id')
resp = self.directory.list(RDIR_ACCT,
ref or service['addr'],
service_type='rdir',
**kwargs)
rdir_host = _filter_rdir_host(resp)
try:
service['rdir'] = by_id[
_make_id(self.ns, 'rdir', rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to %s %s seems down",
rdir_host, service_type,
service['addr'])
service['rdir'] = {"addr": rdir_host,
"tags": dict()}
loc_rdir = service['rdir']
by_id[_make_id(self.ns, 'rdir', rdir_host)] = loc_rdir
except NotFound:
self.logger.info("No rdir linked to %s",
service['addr'])
except OioException as exc:
self.logger.warn('Failed to get rdir linked to %s: %s',
service['addr'], exc)
return all_services, all_rdir
def assign_services(self, service_type,
max_per_rdir=None, min_dist=None, **kwargs):
"""
Assign an rdir service to all `service_type` servers that aren't
already assigned one.
:param max_per_rdir: Maximum number of services an rdir can handle.
:type max_per_rdir: `int`
:param min_dist: minimum required distance between any service and
its assigned rdir service.
:type min_dist: `int`
:returns: The list of `service_type` services that were assigned
rdir services.
"""
all_services = self.cs.all_services(service_type, **kwargs)
all_rdir = self.cs.all_services('rdir', True, **kwargs)
if len(all_rdir) <= 0:
raise ServiceUnavailable("No rdir service found in %s" % self.ns)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x
for x in all_rdir}
errors = list()
for provider in all_services:
provider_id = provider['tags'].get('tag.service_id',
provider['addr'])
try:
resp = self.directory.list(RDIR_ACCT, provider_id,
service_type='rdir', **kwargs)
rdir_host = _filter_rdir_host(resp)
try:
provider['rdir'] = by_id[_make_id(self.ns, 'rdir',
rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to %s %s seems down",
rdir_host, service_type,
provider_id)
except NotFound:
try:
rdir = self._smart_link_rdir(provider_id, all_rdir,
service_type=service_type,
max_per_rdir=max_per_rdir,
min_dist=min_dist,
**kwargs)
except OioException as exc:
self.logger.warn("Failed to link an rdir to %s %s: %s",
service_type, provider_id, exc)
errors.append((provider_id, exc))
continue
n_bases = by_id[rdir]['tags'].get("stat.opened_db_count", 0)
by_id[rdir]['tags']["stat.opened_db_count"] = n_bases + 1
provider['rdir'] = by_id[rdir]
except OioException as exc:
self.logger.warn("Failed to check rdir linked to %s %s "
"(thus won't try to make the link): %s",
service_type, provider_id, exc)
errors.append((provider_id, exc))
if errors:
# group_chunk_errors is flexible enough to accept service addresses
errors = group_chunk_errors(errors)
if len(errors) == 1:
err, addrs = errors.popitem()
oio_reraise(type(err), err, str(addrs))
else:
raise OioException('Several errors encountered: %s' %
errors)
return all_services
def assign_all_meta2(self, max_per_rdir=None, **kwargs):
"""
Assign an rdir service to all meta2 servers that aren't already
assigned one.
:param max_per_rdir: Maximum number of services an rdir can handle.
:type max_per_rdir: `int`
:returns: The list of meta2 that were assigned rdir services.
"""
return self.assign_services("meta2", max_per_rdir, **kwargs)
def assign_all_rawx(self, max_per_rdir=None, **kwargs):
"""
Find an rdir service for all rawx that don't have one already.
:param max_per_rdir: maximum number or rawx services that an rdir
can be linked to
:type max_per_rdir: `int`
"""
return self.assign_services("rawx", max_per_rdir, **kwargs)
def _smart_link_rdir(self, volume_id, all_rdir, max_per_rdir=None,
max_attempts=7, service_type='rawx', min_dist=None,
**kwargs):
"""
Force the load balancer to avoid services that already host more
bases than the average (or more than `max_per_rdir`)
while selecting rdir services.
"""
opened_db = [x['tags'].get('stat.opened_db_count', 0) for x in all_rdir
if x['score'] > 0]
if len(opened_db) <= 0:
raise ServiceUnavailable(
"No valid rdir service found in %s" % self.ns)
if not max_per_rdir:
upper_limit = sum(opened_db) / float(len(opened_db))
else:
upper_limit = max_per_rdir - 1
avoids = [_make_id(self.ns, "rdir", x['addr'])
for x in all_rdir
if x['score'] > 0 and
x['tags'].get('stat.opened_db_count', 0) > upper_limit]
known = [_make_id(self.ns, service_type, volume_id)]
try:
polled = self._poll_rdir(avoid=avoids, known=known,
min_dist=min_dist, **kwargs)
except ClientException as exc:
if exc.status != 481 or max_per_rdir:
raise
# Retry without `avoids`, hoping the next iteration will rebalance
polled = self._poll_rdir(known=known, min_dist=min_dist, **kwargs)
# Associate the rdir to the rawx
forced = {'host': polled['addr'], 'type': 'rdir',
'seq': 1, 'args': "", 'id': polled['id']}
for i in range(max_attempts):
try:
self.directory.force(RDIR_ACCT, volume_id, 'rdir',
forced, autocreate=True, **kwargs)
break
except ClientException as ex:
# Already done
done = (455,)
if ex.status in done:
break
if ex.message.startswith(
'META1 error: (SQLITE_CONSTRAINT) '
'UNIQUE constraint failed'):
self.logger.info(
"Ignored exception (already0): %s", ex)
break
if ex.message.startswith(
'META1 error: (SQLITE_CONSTRAINT) '
'columns cid, srvtype, seq are not unique'):
self.logger.info(
"Ignored exception (already1): %s", ex)
break
# Manage several unretriable errors
retry = (406, 450, 503, 504)
if ex.status >= 400 and ex.status not in retry:
raise
# Monotonic backoff (retriable and net erorrs)
if i < max_attempts - 1:
sleep(i * 1.0)
continue
# Too many attempts
raise
# Do the creation in the rdir itself
try:
self.rdir.create(volume_id, service_type=service_type, **kwargs)
except Exception as exc:
self.logger.warn("Failed to create database for %s on %s: %s",
volume_id, polled['addr'], exc)
return polled['id']
def _create_special_pool(self, options=None, force=False, **kwargs):
"""
Create the special pool for rdir services.
:param options: dictionary of custom options for the pool.
:param force: overwrite the pool if it exists already.
"""
self.cs.lb.create_pool(
'__rawx_rdir', ((1, JOKER_SVC_TARGET), (1, 'rdir')),
options=options, force=force, **kwargs)
def _poll_rdir(self, avoid=None, known=None, min_dist=None, **kwargs):
"""
Call the special rdir service pool (created if missing).
:param min_dist: minimum distance to ensure between the known
service and the selected rdir service.
"""
if not known or len(known) > 1:
raise ValueError('There should be exactly one "known" service')
options = dict()
if min_dist is not None:
options['min_dist'] = min_dist
if options != self._pool_options:
# Options have changed, overwrite the pool.
self._pool_options = options
self._create_special_pool(self._pool_options, force=True, **kwargs)
try:
svcs = self.cs.poll('__rawx_rdir', avoid=avoid, known=known,
**kwargs)
except ClientException as exc:
if exc.status != 400:
raise
self._create_special_pool(self._pool_options, **kwargs)
svcs = self.cs.poll('__rawx_rdir', avoid=avoid, known=known,
**kwargs)
for svc in svcs:
# FIXME: we should include the service type in a dedicated field
if 'rdir' in svc['id']:
return svc
raise ServerException("LB returned incoherent result: %s" % svcs)
class RdirClient(Client):
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.directory_client = DirectoryClient(conf, **kwargs)
def _lookup_rdir_host(self, resp):
host = None
for srv in resp.get('srv', {}):
if srv['type'] == 'rdir':
host = srv['host']
if not host:
raise ClientException("No rdir service found")
return host
def _link_rdir(self, volume_id):
self.directory_client.link(acct=RDIR_ACCT, ref=volume_id,
srv_type='rdir', autocreate=True)
return self.directory_client.show(
acct=RDIR_ACCT, ref=volume_id, srv_type='rdir')
# TODO keep rdir addr in local cache to avoid lookup requests
def _get_rdir_addr(self, volume_id, create=False):
resp = {}
try:
resp = self.directory_client.show(acct=RDIR_ACCT, ref=volume_id,
srv_type='rdir')
except NotFound:
if not create:
raise
try:
host = self._lookup_rdir_host(resp)
except ClientException:
# Reference exists but no rdir linked
if not create:
raise
resp = self._link_rdir(volume_id)
host = self._lookup_rdir_host(resp)
return host
def _make_uri(self, action, volume_id, create=False):
rdir_host = self._get_rdir_addr(volume_id, create=create)
uri = 'http://%s/v1/%s/%s' % (rdir_host, self.ns, action)
return uri
def _rdir_request(self, volume, method, action, create=False, **kwargs):
uri = self._make_uri(action, volume, create=create)
params = {'vol': volume}
if create:
params['create'] = '1'
resp, body = self._direct_request(method, uri, params=params, **kwargs)
return resp, body
def chunk_push(self, volume_id, container_id, content_id, chunk_id,
**data):
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
for key, value in data.iteritems():
body[key] = value
headers = {}
self._rdir_request(volume_id, 'POST', 'rdir/push', create=True,
json=body, headers=headers)
def chunk_delete(self, volume_id, container_id, content_id, chunk_id):
body = {'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id}
self._rdir_request(volume_id, 'DELETE', 'rdir/delete', json=body)
def chunk_fetch(self, volume, limit=100, rebuild=False):
req_body = {'limit': limit}
if rebuild:
req_body['rebuild'] = True
while True:
resp, resp_body = self._rdir_request(volume, 'POST', 'rdir/fetch',
json=req_body)
resp.raise_for_status()
if len(resp_body) == 0:
break
for (key, value) in resp_body:
container, content, chunk = key.split('|')
yield container, content, chunk, value
req_body['start_after'] = key
def admin_incident_set(self, volume, date):
body = {'date': date}
self._rdir_request(volume, 'POST', 'rdir/admin/incident', json=body)
def admin_incident_get(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET',
'rdir/admin/incident')
return resp_body.get('date')
def admin_lock(self, volume, who):
body = {'who': who}
self._rdir_request(volume, 'POST', 'rdir/admin/lock', json=body)
def admin_unlock(self, volume):
self._rdir_request(volume, 'POST', 'rdir/admin/unlock')
def admin_show(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'rdir/admin/show')
return resp_body
def status(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'rdir/status')
return resp_body
class RdirDispatcher(object):
def __init__(self, conf, **kwargs):
self.conf = conf
self.ns = conf['namespace']
self.logger = get_logger(conf)
self.directory = DirectoryClient(conf, logger=self.logger, **kwargs)
self.rdir = RdirClient(conf, logger=self.logger, **kwargs)
self._cs = None
@property
def cs(self):
if not self._cs:
self._cs = ConscienceClient(self.conf, logger=self.logger)
return self._cs
def get_assignation(self, **kwargs):
all_rawx = self.cs.all_services('rawx', **kwargs)
all_rdir = self.cs.all_services('rdir', True, **kwargs)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x for x in all_rdir}
for rawx in all_rawx:
try:
# Verify that there is no rdir linked
resp = self.directory.list(RDIR_ACCT,
rawx['addr'],
service_type='rdir',
**kwargs)
rdir_host = _filter_rdir_host(resp)
try:
rawx['rdir'] = by_id[_make_id(self.ns, 'rdir', rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to rawx %s seems down",
rdir_host, rawx['addr'])
rawx['rdir'] = {"addr": rdir_host, "tags": dict()}
by_id[_make_id(self.ns, 'rdir', rdir_host)] = rawx['rdir']
except NotFound:
self.logger.info("No rdir linked to %s", rawx['addr'])
return all_rawx, all_rdir
def assign_all_rawx(self, max_per_rdir=None, **kwargs):
"""
Find a rdir service for all rawx that don't have one already.
:param max_per_rdir: maximum number or rawx services that an rdir
can be linked to
:type max_per_rdir: `int`
"""
all_rawx = self.cs.all_services('rawx', **kwargs)
all_rdir = self.cs.all_services('rdir', True, **kwargs)
if len(all_rdir) <= 0:
raise ServiceUnavailable("No rdir service found in %s" % self.ns)
by_id = {_make_id(self.ns, 'rdir', x['addr']): x for x in all_rdir}
for rawx in all_rawx:
try:
# Verify that there is no rdir linked
resp = self.directory.list(RDIR_ACCT,
rawx['addr'],
service_type='rdir',
**kwargs)
rdir_host = _filter_rdir_host(resp)
try:
rawx['rdir'] = by_id[_make_id(self.ns, 'rdir', rdir_host)]
except KeyError:
self.logger.warn("rdir %s linked to rawx %s seems down",
rdir_host, rawx['addr'])
except (NotFound, ClientException):
rdir = self._smart_link_rdir(rawx['addr'], all_rdir,
max_per_rdir, **kwargs)
n_bases = by_id[rdir]['tags'].get("stat.opened_db_count", 0)
by_id[rdir]['tags']["stat.opened_db_count"] = n_bases + 1
rawx['rdir'] = by_id[rdir]
return all_rawx
def _smart_link_rdir(self,
volume_id,
all_rdir,
max_per_rdir=None,
**kwargs):
"""
Force the load balancer to avoid services that already host more
bases than the average (or more than `max_per_rdir`)
while selecting rdir services.
"""
opened_db = [
x['tags']['stat.opened_db_count'] for x in all_rdir
if x['score'] > 0
]
if len(opened_db) <= 0:
raise ServiceUnavailable("No valid rdir service found in %s" %
self.ns)
if not max_per_rdir:
upper_limit = sum(opened_db) / float(len(opened_db))
else:
upper_limit = max_per_rdir - 1
avoids = [
_make_id(self.ns, "rdir", x['addr']) for x in all_rdir if
x['score'] > 0 and x['tags']['stat.opened_db_count'] > upper_limit
]
known = [_make_id(self.ns, "rawx", volume_id)]
try:
polled = self._poll_rdir(avoid=avoids, known=known, **kwargs)
except ClientException as exc:
if exc.status != 481 or max_per_rdir:
raise
# Retry without `avoids`, hoping the next iteration will rebalance
polled = self._poll_rdir(known=known, **kwargs)
# Associate the rdir to the rawx
forced = {
'host': polled['addr'],
'type': 'rdir',
'seq': 1,
'args': "",
'id': polled['id']
}
max_attempts = 7
for i in range(max_attempts):
try:
self.directory.force(RDIR_ACCT,
volume_id,
'rdir',
forced,
autocreate=True,
**kwargs)
break
except ClientException as ex:
# Already done
done = (455, )
if ex.status in done:
break
if ex.message.startswith('META1 error: (SQLITE_CONSTRAINT) '
'UNIQUE constraint failed'):
self.logger.info("Ignored exception (already0): %s", ex)
break
if ex.message.startswith(
'META1 error: (SQLITE_CONSTRAINT) '
'columns cid, srvtype, seq are not unique'):
self.logger.info("Ignored exception (already1): %s", ex)
break
# Manage several unretriable errors
retry = (406, 450, 503, 504)
if ex.status >= 400 and ex.status not in retry:
raise
# Monotonic backoff (retriable and net erorrs)
if i < max_attempts - 1:
from time import sleep
sleep(i * 1.0)
continue
# Too many attempts
raise
# Do the creation in the rdir itself
try:
self.rdir.create(volume_id, **kwargs)
except Exception as exc:
self.logger.warn("Failed to create database for %s on %s: %s",
volume_id, polled['addr'], exc)
return polled['id']
def _poll_rdir(self, avoid=None, known=None, **kwargs):
"""Call the special rdir service pool (created if missing)"""
try:
svcs = self.cs.poll('__rawx_rdir',
avoid=avoid,
known=known,
**kwargs)
except ClientException as exc:
if exc.status != 400:
raise
self.cs.lb.create_pool('__rawx_rdir',
((1, JOKER_SVC_TARGET), (1, 'rdir')),
**kwargs)
svcs = self.cs.poll('__rawx_rdir',
avoid=avoid,
known=known,
**kwargs)
for svc in svcs:
# FIXME: we should include the service type in a dedicated field
if 'rdir' in svc['id']:
return svc
raise ServerException("LB returned incoherent result: %s" % svcs)
class CheckMeta2(CheckService):
account_name = "_meta2_probe"
def __init__(self, namespace, **kwargs):
ep_parts = ["http:/",
load_namespace_conf(namespace).get('proxy'),
"v3.0",
namespace,
"content"]
super(CheckMeta2, self).__init__(namespace, "meta2",
endpoint="/".join(ep_parts), **kwargs)
self.account = AccountClient({"namespace": self.ns})
self.container = ContainerClient({"namespace": self.ns})
self.directory = DirectoryClient({"namespace": self.ns})
self.reference = random_buffer('0123456789ABCDEF', 64)
def _get_params(self):
path = random_buffer('0123456789ABCDEF', 64)
return {'acct': self.account_name, 'ref': self.reference, 'path': path}
def _compare_chunks(self, chunks1, chunks2):
def light_chunks(chunks):
new_chunks = []
for chunk in chunks:
new_chunk = dict()
new_chunk["url"] = chunk["url"]
new_chunk["hash"] = chunk["hash"]
new_chunks.append(new_chunk)
return new_chunks
try:
chunks1 = light_chunks(chunks1)
chunks1.sort()
chunks2 = light_chunks(chunks2)
chunks2.sort()
return cmp(chunks1, chunks2) == 0
except TypeError:
return False
def _cycle(self, meta2_host):
self.directory.unlink(
account=self.account_name, reference=self.reference,
service_type=self.service_type)
service = {"host": meta2_host, "type": self.service_type, "args": "",
"seq": 1}
self.directory.force(
account=self.account_name, reference=self.reference,
service_type=self.service_type, services=service)
params = self._get_params()
global_success = True
_, body, success = self._request(
"GET", "/locate", params=params, expected_status=404)
global_success &= success
headers = {'X-oio-action-mode': 'autocreate'}
_, body, success = self._request(
"POST", "/prepare", params=params, headers=headers,
json={"size": "1024"}, expected_status=200)
global_success &= success
chunks = body
_, body, success = self._request(
"GET", "/locate", params=params, expected_status=404)
global_success &= success
headers = {"x-oio-content-meta-length": "1024"}
_, _, success = self._request(
"POST", "/create", params=params, headers=headers, json=chunks,
expected_status=204)
global_success &= success
_, body, success = self._request(
"GET", "/locate", params=params, expected_status=200)
global_success &= success
success = self._compare_chunks(chunks, body)
global_success &= success
_, _, success = self._request(
"POST", "/delete", params=params, expected_status=204)
global_success &= success
_, body, success = self._request(
"GET", "/locate", params=params, expected_status=404)
global_success &= success
return global_success
def run(self):
try:
self.container.container_create(account=self.account_name,
reference=self.reference)
super(CheckMeta2, self).run()
self.container.container_delete(account=self.account_name,
reference=self.reference)
sleep(1)
self.account.account_delete(self.account_name)
except Exception as exc:
print("Exception - " + str(exc))
class Meta2IndexingWorker(object):
"""
Indexing worker responsible for a single volume.
"""
def __init__(self, volume_path, conf, pool_manager=None):
"""
Initializes an Indexing worker for indexing meta2 databases.
Possible values of conf relating to this worker are:
- interval: (int) in sec time between two full scans. Default: half an
hour.
- report_interval: (int) in sec, time between two reports: Default: 300
- scanned_per_second: (int) maximum number of indexed databases /s.
- try_removing_faulty_indexes : In the event where we encounter a
database that's not supposed to be handled by this volume, attempt
to remove it from this volume rdir index if it exists
WARNING: The decision is based off of a proxy response, that could
be affected by cache inconsistencies for example, use at your own
risk. Default: False
:param volume_path: The volume path to be indexed
:param conf: The configuration to be passed to the needed services
:param pool_manager: A connection pool manager. If none is given, a
new one with a default size of 10 will be created.
"""
self.logger = get_logger(conf)
self._stop = False
self.volume = volume_path
self.success_nb = 0
self.failed_nb = 0
self.full_scan_nb = 0
self.last_report_time = 0
self.last_scan_time = 0
self.last_index_time = 0
self.start_time = 0
self.indexed_since_last_report = 0
self.scans_interval = int_value(
conf.get('interval'), 1800)
self.report_interval = int_value(
conf.get('report_interval'), 300)
self.max_indexed_per_second = int_value(
conf.get('scanned_per_second'), 3000)
self.namespace, self.volume_id = check_volume_for_service_type(
self.volume, "meta2")
self.attempt_bad_index_removal = boolean_value(
conf.get('try_removing_faulty_indexes'), False)
if not pool_manager:
pool_manager = get_pool_manager(pool_connections=10)
self.index_client = RdirClient(conf, logger=self.logger,
pool_manager=pool_manager)
self.dir_client = DirectoryClient(conf, logger=self.logger,
pool_manager=pool_manager)
def report(self, tag):
"""
Log the status of indexer
:param tag: One of three: starting, running, ended.
"""
total = self.success_nb + self.failed_nb
now = time.time()
elapsed = (now - self.start_time) or 0.00001
since_last_rprt = (now - self.last_report_time) or 0.00001
self.logger.info(
'volume_id=%(volume_id)s %(tag)s=%(current_time)s '
'elapsed=%(elapsed).02f '
'pass=%(pass)d '
'errors=%(errors)d '
'containers_indexed=%(total_indexed)d %(index_rate).2f/s',
{
'volume_id': self.volume_id,
'tag': tag,
'current_time': datetime.fromtimestamp(
int(now)).isoformat(),
'pass': self.full_scan_nb,
'errors': self.failed_nb,
'total_indexed': total,
'index_rate': self.indexed_since_last_report / since_last_rprt,
'elapsed': elapsed
}
)
self.last_report_time = now
self.indexed_since_last_report = 0
def warn(self, msg, container_id):
self.logger.warn(
'volume_id=%(volume_id)s container_id=%(container_id)s %(error)s',
{
'volume_id': self.volume_id,
'container_id': container_id,
'error': msg
}
)
def _attempt_index_removal(self, db_path, cid):
"""
Fail safe removal attempt.
"""
try:
self.index_client.meta2_index_delete(self.volume_id, db_path, cid)
except exc.OioException as exception:
self.warn(
container_id=cid,
msg="Unable to remove database from the volume "
"index : {0}".format(str(exception))
)
def index_meta2_database(self, db_id):
"""
Add a meta2 database to the rdir index. Fails if the database isn't
handled by the current volume.
:param db_id: The ContentID representing the reference to the database.
"""
if len(db_id) < STRLEN_REFERENCEID:
self.warn('Not a valid container ID', db_id)
return
try:
srvcs = self.dir_client.list(cid=db_id)
account, container = srvcs['account'], srvcs['name']
is_peer = self.volume_id in [x['host'] for x in srvcs['srv'] if
x['type'] == 'meta2']
container_id = db_id.rsplit(".")[0]
if six.PY2:
if isinstance(account, six.text_type):
account = account.encode('utf-8')
if isinstance(container, six.text_type):
container = container.encode('utf-8')
cont_url = "{0}/{1}/{2}".format(self.namespace, account, container)
if not is_peer:
self.warn("Trying to index a container that isn't handled by"
"this volume", db_id)
if self.attempt_bad_index_removal:
self._attempt_index_removal(cont_url, container_id)
return
self.index_client.meta2_index_push(volume_id=self.volume_id,
container_url=cont_url,
mtime=time.time(),
container_id=container_id)
self.success_nb += 1
except exc.OioException as exception:
self.failed_nb += 1
self.warn("Unable to to index container: %s" % str(exception),
db_id)
self.indexed_since_last_report += 1
def crawl_volume(self):
"""
Crawl the volume assigned to this worker, and index every database.
"""
paths = paths_gen(self.volume)
self.full_scan_nb += 1
self.success_nb = 0
self.failed_nb = 0
now = time.time()
self.last_report_time = now
self.report("starting")
for db_path in paths:
# Graceful exit, hopefully
if self._stop:
break
db_id = db_path.rsplit("/")[-1].rsplit(".")
if len(db_id) != 3:
self.warn("Malformed db file name !", db_path)
continue
db_id = ".".join(db_id[:2])
self.index_meta2_database(db_id)
self.last_index_time = ratelimit(
self.last_index_time,
self.max_indexed_per_second
)
now = time.time()
if now - self.last_report_time >= self.report_interval:
self.report("running")
self.report("ended")
def run(self):
"""
Main worker loop
"""
self.start_time = time.time()
while not self._stop:
try:
self.crawl_volume()
self.last_scan_time = time.time()
time.sleep(self.scans_interval)
except exc.OioException as exception:
self.logger.exception("ERROR during indexing meta2: %s",
exception)
def stop(self):
"""
Could be needed for eventually gracefully stopping.
"""
self._stop = True
class RdirClient(HttpApi):
"""
Client class for rdir services.
"""
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.directory = DirectoryClient(conf, **kwargs)
self._addr_cache = dict()
def _clear_cache(self, volume_id):
del self._addr_cache[volume_id]
def _get_rdir_addr(self, volume_id):
# Initial lookup in the cache
if volume_id in self._addr_cache:
return self._addr_cache[volume_id]
# Not cached, try a direct lookup
try:
resp = self.directory.list(RDIR_ACCT,
volume_id,
service_type='rdir')
host = _filter_rdir_host(resp)
# Add the new service to the cache
self._addr_cache[volume_id] = host
return host
except NotFound:
raise VolumeException('No rdir assigned to volume %s' % volume_id)
def _make_uri(self, action, volume_id):
rdir_host = self._get_rdir_addr(volume_id)
return 'http://%s/v1/rdir/%s' % (rdir_host, action)
def _rdir_request(self, volume, method, action, create=False, **kwargs):
params = {'vol': volume}
if create:
params['create'] = '1'
uri = self._make_uri(action, volume)
try:
resp, body = self._direct_request(method,
uri,
params=params,
**kwargs)
except OioNetworkException:
self._clear_cache(volume)
raise
return resp, body
def create(self, volume_id):
"""Create the database for `volume_id` on the appropriate rdir"""
self._rdir_request(volume_id, 'POST', 'create')
def chunk_push(self, volume_id, container_id, content_id, chunk_id,
**data):
"""Reference a chunk in the reverse directory"""
body = {
'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id
}
for key, value in data.iteritems():
body[key] = value
self._rdir_request(volume_id, 'POST', 'push', create=True, json=body)
def chunk_delete(self, volume_id, container_id, content_id, chunk_id):
"""Unreference a chunk from the reverse directory"""
body = {
'container_id': container_id,
'content_id': content_id,
'chunk_id': chunk_id
}
self._rdir_request(volume_id, 'DELETE', 'delete', json=body)
def chunk_fetch(self, volume, limit=100, rebuild=False, container_id=None):
"""
Fetch the list of chunks belonging to the specified volume.
:param volume: the volume to get chunks from
:type volume: `str`
:param limit: maximum number of results to return
:type limit: `int`
:param rebuild:
:type rebuild: `bool`
:keyword container_id: get only chunks belonging to
the specified container
:type container_id: `str`
"""
req_body = {'limit': limit}
if rebuild:
req_body['rebuild'] = True
if container_id:
req_body['container_id'] = container_id
while True:
resp, resp_body = self._rdir_request(volume,
'POST',
'fetch',
json=req_body)
if len(resp_body) == 0:
break
for (key, value) in resp_body:
container, content, chunk = key.split('|')
yield container, content, chunk, value
req_body['start_after'] = key
def admin_incident_set(self, volume, date):
body = {'date': int(float(date))}
self._rdir_request(volume, 'POST', 'admin/incident', json=body)
def admin_incident_get(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'admin/incident')
return resp_body.get('date')
def admin_lock(self, volume, who):
body = {'who': who}
self._rdir_request(volume, 'POST', 'admin/lock', json=body)
def admin_unlock(self, volume):
self._rdir_request(volume, 'POST', 'admin/unlock')
def admin_show(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'admin/show')
return resp_body
def admin_clear(self, volume, clear_all=False):
body = {'all': clear_all}
resp, resp_body = self._rdir_request(volume,
'POST',
'admin/clear',
json=body)
return resp_body
def status(self, volume):
resp, resp_body = self._rdir_request(volume, 'GET', 'status')
return resp_body
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.directory_client = DirectoryClient(conf, **kwargs)
def __init__(self, conf, **kwargs):
super(RdirClient, self).__init__(conf, **kwargs)
self.directory = DirectoryClient(conf, **kwargs)
self._addr_cache = dict()
