self.permission_hint()] + self.hints
if s is not None)
+ def stripped(self):
+ return "%s+%i" % (self.md5sum, self.size)
+
def _make_hex_prop(name, length):
# Build and return a new property with the given name that
# must be a hex string of the given length.
def cap_cache(self):
'''Cap the cache size to self.cache_max'''
- self._cache_lock.acquire()
- try:
+ with self._cache_lock:
# Select all slots except those where ready.is_set() and content is
# None (that means there was an error reading the block).
self._cache = [c for c in self._cache if not (c.ready.is_set() and c.content is None)]
del self._cache[i]
break
sm = sum([slot.size() for slot in self._cache])
- finally:
- self._cache_lock.release()
+
+ def _get(self, locator):
+ # Test if the locator is already in the cache
+ for i in xrange(0, len(self._cache)):
+ if self._cache[i].locator == locator:
+ n = self._cache[i]
+ if i != 0:
+ # move it to the front
+ del self._cache[i]
+ self._cache.insert(0, n)
+ return n
+ return None
+
+ def get(self, locator):
+ with self._cache_lock:
+ return self._get(locator)
def reserve_cache(self, locator):
'''Reserve a cache slot for the specified locator,
or return the existing slot.'''
- self._cache_lock.acquire()
- try:
- # Test if the locator is already in the cache
- for i in xrange(0, len(self._cache)):
- if self._cache[i].locator == locator:
- n = self._cache[i]
- if i != 0:
- # move it to the front
- del self._cache[i]
- self._cache.insert(0, n)
- return n, False
-
- # Add a new cache slot for the locator
- n = KeepBlockCache.CacheSlot(locator)
- self._cache.insert(0, n)
- return n, True
- finally:
- self._cache_lock.release()
+ with self._cache_lock:
+ n = self._get(locator)
+ if n:
+ return n, False
+ else:
+ # Add a new cache slot for the locator
+ n = KeepBlockCache.CacheSlot(locator)
+ self._cache.insert(0, n)
+ return n, True
class KeepClient(object):
HTTP_ERRORS = (requests.exceptions.RequestException,
socket.error, ssl.SSLError)
- def __init__(self, root, **headers):
+ def __init__(self, root, session, **headers):
self.root = root
self.last_result = None
self.success_flag = None
+ self.session = session
self.get_headers = {'Accept': 'application/octet-stream'}
self.get_headers.update(headers)
self.put_headers = headers
_logger.debug("Request: GET %s", url)
try:
with timer.Timer() as t:
- result = requests.get(url.encode('utf-8'),
+ result = self.session.get(url.encode('utf-8'),
headers=self.get_headers,
timeout=timeout)
except self.HTTP_ERRORS as e:
content = result.content
_logger.info("%s response: %s bytes in %s msec (%.3f MiB/sec)",
self.last_status(), len(content), t.msecs,
- (len(content)/(1024.0*1024))/t.secs)
+ (len(content)/(1024.0*1024))/t.secs if t.secs > 0 else 0)
if self.success_flag:
resp_md5 = hashlib.md5(content).hexdigest()
if resp_md5 == locator.md5sum:
url = self.root + hash_s
_logger.debug("Request: PUT %s", url)
try:
- result = requests.put(url.encode('utf-8'),
+ result = self.session.put(url.encode('utf-8'),
data=body,
headers=self.put_headers,
timeout=timeout)
def run_with_limiter(self, limiter):
if self.service.finished():
return
- _logger.debug("KeepWriterThread %s proceeding %s %s",
+ _logger.debug("KeepWriterThread %s proceeding %s+%i %s",
str(threading.current_thread()),
self.args['data_hash'],
+ len(self.args['data']),
self.args['service_root'])
self._success = bool(self.service.put(
self.args['data_hash'],
status = self.service.last_status()
if self._success:
result = self.service.last_result
- _logger.debug("KeepWriterThread %s succeeded %s %s",
+ _logger.debug("KeepWriterThread %s succeeded %s+%i %s",
str(threading.current_thread()),
self.args['data_hash'],
+ len(self.args['data']),
self.args['service_root'])
# Tick the 'done' counter for the number of replica
# reported stored by the server, for the case that
replicas_stored = int(result.headers['x-keep-replicas-stored'])
except (KeyError, ValueError):
replicas_stored = 1
- limiter.save_response(result.text.strip(), replicas_stored)
+ limiter.save_response(result.content.strip(), replicas_stored)
elif status is not None:
_logger.debug("Request fail: PUT %s => %s %s",
self.args['data_hash'], status,
- self.service.last_result.text)
+ self.service.last_result.content)
def __init__(self, api_client=None, proxy=None,
timeout=DEFAULT_TIMEOUT, proxy_timeout=DEFAULT_PROXY_TIMEOUT,
api_token=None, local_store=None, block_cache=None,
- num_retries=0):
+ num_retries=0, session=None):
"""Initialize a new KeepClient.
Arguments:
- * api_client: The API client to use to find Keep services. If not
+ :api_client:
+ The API client to use to find Keep services. If not
provided, KeepClient will build one from available Arvados
configuration.
- * proxy: If specified, this KeepClient will send requests to this
- Keep proxy. Otherwise, KeepClient will fall back to the setting
- of the ARVADOS_KEEP_PROXY configuration setting. If you want to
- ensure KeepClient does not use a proxy, pass in an empty string.
- * timeout: The timeout (in seconds) for HTTP requests to Keep
+
+ :proxy:
+ If specified, this KeepClient will send requests to this Keep
+ proxy. Otherwise, KeepClient will fall back to the setting of the
+ ARVADOS_KEEP_PROXY configuration setting. If you want to ensure
+ KeepClient does not use a proxy, pass in an empty string.
+
+ :timeout:
+ The timeout (in seconds) for HTTP requests to Keep
non-proxy servers. A tuple of two floats is interpreted as
(connection_timeout, read_timeout): see
http://docs.python-requests.org/en/latest/user/advanced/#timeouts.
Default: (2, 300).
- * proxy_timeout: The timeout (in seconds) for HTTP requests to
+
+ :proxy_timeout:
+ The timeout (in seconds) for HTTP requests to
Keep proxies. A tuple of two floats is interpreted as
(connection_timeout, read_timeout). Default: (20, 300).
- * api_token: If you're not using an API client, but only talking
+
+ :api_token:
+ If you're not using an API client, but only talking
directly to a Keep proxy, this parameter specifies an API token
to authenticate Keep requests. It is an error to specify both
api_client and api_token. If you specify neither, KeepClient
will use one available from the Arvados configuration.
- * local_store: If specified, this KeepClient will bypass Keep
+
+ :local_store:
+ If specified, this KeepClient will bypass Keep
services, and save data to the named directory. If unspecified,
KeepClient will fall back to the setting of the $KEEP_LOCAL_STORE
environment variable. If you want to ensure KeepClient does not
use local storage, pass in an empty string. This is primarily
intended to mock a server for testing.
- * num_retries: The default number of times to retry failed requests.
+
+ :num_retries:
+ The default number of times to retry failed requests.
This will be used as the default num_retries value when get() and
put() are called. Default 0.
+
+ :session:
+ The requests.Session object to use for get() and put() requests.
+ Will create one if not specified.
"""
self.lock = threading.Lock()
if proxy is None:
self.put = self.local_store_put
else:
self.num_retries = num_retries
+ self.session = session if session is not None else requests.Session()
if proxy:
if not proxy.endswith('/'):
proxy += '/'
local_roots = self.weighted_service_roots(md5_s, force_rebuild)
for root in local_roots:
if root not in roots_map:
- roots_map[root] = self.KeepService(root, **headers)
+ roots_map[root] = self.KeepService(root, self.session, **headers)
return local_roots
@staticmethod
else:
return None
+ def get_from_cache(self, loc):
+ """Fetch a block only if is in the cache, otherwise return None."""
+ slot = self.block_cache.get(loc)
+ if slot.ready.is_set():
+ return slot.get()
+ else:
+ return None
+
@retry.retry_method
def get(self, loc_s, num_retries=None):
"""Get data from Keep.
return ''.join(self.get(x) for x in loc_s.split(','))
locator = KeepLocator(loc_s)
expect_hash = locator.md5sum
-
slot, first = self.block_cache.reserve_cache(expect_hash)
if not first:
v = slot.get()
hint_roots = ['http://keep.{}.arvadosapi.com/'.format(hint[2:])
for hint in locator.hints if hint.startswith('K@')]
# Map root URLs their KeepService objects.
- roots_map = {root: self.KeepService(root) for root in hint_roots}
+ roots_map = {root: self.KeepService(root, self.session) for root in hint_roots}
blob = None
loop = retry.RetryLoop(num_retries, self._check_loop_result,
backoff_start=2)
exponential backoff. The default value is set when the
KeepClient is initialized.
"""
+
+ if isinstance(data, unicode):
+ data = data.encode("ascii")
+ elif not isinstance(data, str):
+ raise arvados.errors.ArgumentError("Argument 'data' to KeepClient.put must be type 'str'")
+
data_hash = hashlib.md5(data).hexdigest()
if copies < 1:
return data_hash
return ''
with open(os.path.join(self.local_store, locator.md5sum), 'r') as f:
return f.read()
+
+ def is_cached(self, locator):
+ return self.block_cache.reserve_cache(expect_hash)
'^d41d8cd98f00b204e9800998ecf8427e\+0',
('wrong locator from Keep.put(""): ' + blob_locator))
+ def test_unicode_must_be_ascii(self):
+ # If unicode type, must only consist of valid ASCII
+ foo_locator = self.keep_client.put(u'foo')
+ self.assertRegexpMatches(
+ foo_locator,
+ '^acbd18db4cc2f85cedef654fccc4a4d8\+3',
+ 'wrong md5 hash from Keep.put("foo"): ' + foo_locator)
+
+ with self.assertRaises(UnicodeEncodeError):
+ # Error if it is not ASCII
+ self.keep_client.put(u'\xe2')
+
+ with self.assertRaises(arvados.errors.ArgumentError):
+ # Must be a string type
+ self.keep_client.put({})
class KeepPermissionTestCase(run_test_server.TestCaseWithServers):
MAIN_SERVER = {}
def test_get_timeout(self):
api_client = self.mock_keep_services(count=1)
- keep_client = arvados.KeepClient(api_client=api_client)
force_timeout = [socket.timeout("timed out")]
- with mock.patch('requests.get', side_effect=force_timeout) as mock_request:
+ with tutil.mock_get(force_timeout) as mock_session:
+ keep_client = arvados.KeepClient(api_client=api_client)
with self.assertRaises(arvados.errors.KeepReadError):
keep_client.get('ffffffffffffffffffffffffffffffff')
- self.assertTrue(mock_request.called)
+ self.assertTrue(mock_session.return_value.get.called)
self.assertEqual(
arvados.KeepClient.DEFAULT_TIMEOUT,
- mock_request.call_args[1]['timeout'])
+ mock_session.return_value.get.call_args[1]['timeout'])
def test_put_timeout(self):
api_client = self.mock_keep_services(count=1)
- keep_client = arvados.KeepClient(api_client=api_client)
force_timeout = [socket.timeout("timed out")]
- with mock.patch('requests.put', side_effect=force_timeout) as mock_request:
+ with tutil.mock_put(force_timeout) as mock_session:
+ keep_client = arvados.KeepClient(api_client=api_client)
with self.assertRaises(arvados.errors.KeepWriteError):
keep_client.put('foo')
- self.assertTrue(mock_request.called)
+ self.assertTrue(mock_session.return_value.put.called)
self.assertEqual(
arvados.KeepClient.DEFAULT_TIMEOUT,
- mock_request.call_args[1]['timeout'])
+ mock_session.return_value.put.call_args[1]['timeout'])
def test_proxy_get_timeout(self):
# Force a timeout, verifying that the requests.get or
# requests.put method was called with the proxy_timeout
# setting rather than the default timeout.
api_client = self.mock_keep_services(service_type='proxy', count=1)
- keep_client = arvados.KeepClient(api_client=api_client)
force_timeout = [socket.timeout("timed out")]
- with mock.patch('requests.get', side_effect=force_timeout) as mock_request:
+ with tutil.mock_get(force_timeout) as mock_session:
+ keep_client = arvados.KeepClient(api_client=api_client)
with self.assertRaises(arvados.errors.KeepReadError):
keep_client.get('ffffffffffffffffffffffffffffffff')
- self.assertTrue(mock_request.called)
+ self.assertTrue(mock_session.return_value.get.called)
self.assertEqual(
arvados.KeepClient.DEFAULT_PROXY_TIMEOUT,
- mock_request.call_args[1]['timeout'])
+ mock_session.return_value.get.call_args[1]['timeout'])
def test_proxy_put_timeout(self):
# Force a timeout, verifying that the requests.get or
# requests.put method was called with the proxy_timeout
# setting rather than the default timeout.
api_client = self.mock_keep_services(service_type='proxy', count=1)
- keep_client = arvados.KeepClient(api_client=api_client)
force_timeout = [socket.timeout("timed out")]
- with mock.patch('requests.put', side_effect=force_timeout) as mock_request:
+ with tutil.mock_put(force_timeout) as mock_session:
+ keep_client = arvados.KeepClient(api_client=api_client)
with self.assertRaises(arvados.errors.KeepWriteError):
keep_client.put('foo')
- self.assertTrue(mock_request.called)
+ self.assertTrue(mock_session.return_value.put.called)
self.assertEqual(
arvados.KeepClient.DEFAULT_PROXY_TIMEOUT,
- mock_request.call_args[1]['timeout'])
+ mock_session.return_value.put.call_args[1]['timeout'])
def test_probe_order_reference_set(self):
# expected_order[i] is the probe order for
def check_errors_from_last_retry(self, verb, exc_class):
api_client = self.mock_keep_services(count=2)
- keep_client = arvados.KeepClient(api_client=api_client)
req_mock = getattr(tutil, 'mock_{}_responses'.format(verb))(
"retry error reporting test", 500, 500, 403, 403)
with req_mock, tutil.skip_sleep, \
self.assertRaises(exc_class) as err_check:
+ keep_client = arvados.KeepClient(api_client=api_client)
getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0',
num_retries=3)
self.assertEqual([403, 403], [
data = 'partial failure test'
data_loc = '{}+{}'.format(hashlib.md5(data).hexdigest(), len(data))
api_client = self.mock_keep_services(count=3)
- keep_client = arvados.KeepClient(api_client=api_client)
with tutil.mock_put_responses(data_loc, 200, 500, 500) as req_mock, \
self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
+ keep_client = arvados.KeepClient(api_client=api_client)
keep_client.put(data)
self.assertEqual(2, len(exc_check.exception.service_errors()))
self.check_success(locator=self.HINTED_LOCATOR)
def test_try_next_server_after_timeout(self):
- side_effects = [
- socket.timeout("timed out"),
- tutil.fake_requests_response(200, self.DEFAULT_EXPECT)]
- with mock.patch('requests.get',
- side_effect=iter(side_effects)):
+ with tutil.mock_get([
+ socket.timeout("timed out"),
+ tutil.fake_requests_response(200, self.DEFAULT_EXPECT)]):
self.check_success(locator=self.HINTED_LOCATOR)
def test_retry_data_with_wrong_checksum(self):
- side_effects = (tutil.fake_requests_response(200, s)
- for s in ['baddata', self.TEST_DATA])
- with mock.patch('requests.get', side_effect=side_effects):
+ with tutil.mock_get((tutil.fake_requests_response(200, s) for s in ['baddata', self.TEST_DATA])):
self.check_success(locator=self.HINTED_LOCATOR)
import calendar
import threading
import itertools
+ import ciso8601
from arvados.util import portable_data_hash_pattern, uuid_pattern, collection_uuid_pattern, group_uuid_pattern, user_uuid_pattern, link_uuid_pattern
# appear as underscores in the fuse mount.)
_disallowed_filename_characters = re.compile('[\x00/]')
-class SafeApi(object):
- """Threadsafe wrapper for API object.
-
- This stores and returns a different api object per thread, because
- httplib2 which underlies apiclient is not threadsafe.
- """
-
- def __init__(self, config):
- self.host = config.get('ARVADOS_API_HOST')
- self.api_token = config.get('ARVADOS_API_TOKEN')
- self.insecure = config.flag_is_true('ARVADOS_API_HOST_INSECURE')
- self.local = threading.local()
- self.block_cache = arvados.KeepBlockCache()
-
- def localapi(self):
- if 'api' not in self.local.__dict__:
- self.local.api = arvados.api(
- version='v1',
- host=self.host, token=self.api_token, insecure=self.insecure)
- return self.local.api
-
- def localkeep(self):
- if 'keep' not in self.local.__dict__:
- self.local.keep = arvados.KeepClient(api_client=self.localapi(), block_cache=self.block_cache)
- return self.local.keep
-
- def __getattr__(self, name):
- # Proxy nonexistent attributes to the local API client.
- try:
- return getattr(self.localapi(), name)
- except AttributeError:
- return super(SafeApi, self).__getattr__(name)
-
-
def convertTime(t):
"""Parse Arvados timestamp to unix time."""
if not t:
return 0
try:
- return calendar.timegm(time.strptime(t, "%Y-%m-%dT%H:%M:%SZ"))
+ return calendar.timegm(ciso8601.parse_datetime_unaware(t).timetuple())
except (TypeError, ValueError):
return 0
with llfuse.lock_released:
coll_reader = arvados.CollectionReader(
- self.collection_locator, self.api, self.api.localkeep(),
+ self.collection_locator, self.api, self.api.keep,
num_retries=self.num_retries)
new_collection_object = coll_reader.api_response() or {}
# If the Collection only exists in Keep, there will be no API
# arv-mount.
# The workaround is to implement it with the proper number of parameters,
# and then everything works out.
- def create(self, p1, p2, p3, p4, p5):
+ def create(self, inode_parent, name, mode, flags, ctx):
raise llfuse.FUSEError(errno.EROFS)
self.mounttmp = tempfile.mkdtemp()
run_test_server.run()
run_test_server.authorize_with("admin")
- self.api = fuse.SafeApi(arvados.config)
+ self.api = arvados.safeapi.ThreadSafeApiCache(arvados.config.settings())
def make_mount(self, root_class, **root_kwargs):
operations = fuse.Operations(os.getuid(), os.getgid())
# Double check that we can open and read objects in this folder as a file,
# and that its contents are what we expect.
- with open(os.path.join(
+ pipeline_template_path = os.path.join(
self.mounttmp,
'FUSE User',
'FUSE Test Project',
- 'pipeline template in FUSE project.pipelineTemplate')) as f:
+ 'pipeline template in FUSE project.pipelineTemplate')
+ with open(pipeline_template_path) as f:
j = json.load(f)
self.assertEqual("pipeline template in FUSE project", j['name'])
+ # check mtime on template
+ st = os.stat(pipeline_template_path)
+ self.assertEqual(st.st_mtime, 1397493304)
+
+ # check mtime on collection
+ st = os.stat(os.path.join(
+ self.mounttmp,
+ 'FUSE User',
+ 'collection #1 owned by FUSE'))
+ self.assertEqual(st.st_mtime, 1391448174)
+
class FuseHomeTest(MountTestBase):
def runTest(self):