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 += '/'
r['service_port'])
_logger.debug(str(self._keep_services))
- def _service_weight(self, hash, service_uuid):
+ def _service_weight(self, data_hash, service_uuid):
"""Compute the weight of a Keep service endpoint for a data
block with a known hash.
The weight is md5(h + u) where u is the last 15 characters of
the service endpoint's UUID.
"""
- return hashlib.md5(hash + service_uuid[-15:]).hexdigest()
+ return hashlib.md5(data_hash + service_uuid[-15:]).hexdigest()
- def weighted_service_roots(self, hash, force_rebuild=False):
+ def weighted_service_roots(self, data_hash, force_rebuild=False):
"""Return an array of Keep service endpoints, in the order in
which they should be probed when reading or writing data with
the given hash.
self.build_services_list(force_rebuild)
# Sort the available services by weight (heaviest first) for
- # this hash, and return their service_roots (base URIs) in
- # that order.
+ # this data_hash, and return their service_roots (base URIs)
+ # in that order.
sorted_roots = [
svc['_service_root'] for svc in sorted(
self._keep_services,
reverse=True,
- key=lambda svc: self._service_weight(hash, svc['uuid']))]
- _logger.debug(hash + ': ' + str(sorted_roots))
+ key=lambda svc: self._service_weight(data_hash, svc['uuid']))]
+ _logger.debug(data_hash + ': ' + str(sorted_roots))
return sorted_roots
def map_new_services(self, roots_map, md5_s, force_rebuild, **headers):
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)
if loop.success():
return blob
- # No servers fulfilled the request. Count how many responded
- # "not found;" if the ratio is high enough (currently 75%), report
- # Not Found; otherwise a generic error.
+ try:
+ all_roots = local_roots + hint_roots
+ except NameError:
+ # We never successfully fetched local_roots.
+ all_roots = hint_roots
# Q: Including 403 is necessary for the Keep tests to continue
# passing, but maybe they should expect KeepReadError instead?
- not_founds = sum(1 for ks in roots_map.values()
- if ks.last_status() in set([403, 404, 410]))
- if roots_map and ((float(not_founds) / len(roots_map)) >= .75):
- raise arvados.errors.NotFoundError(loc_s)
+ not_founds = sum(1 for key in all_roots
+ if roots_map[key].last_status() in {403, 404, 410})
+ service_errors = ((key, roots_map[key].last_result)
+ for key in all_roots)
+ if not roots_map:
+ raise arvados.errors.KeepReadError(
+ "failed to read {}: no Keep services available ({})".format(
+ loc_s, loop.last_result()))
+ elif not_founds == len(all_roots):
+ raise arvados.errors.NotFoundError(
+ "{} not found".format(loc_s), service_errors)
else:
- raise arvados.errors.KeepReadError(loc_s)
+ raise arvados.errors.KeepReadError(
+ "failed to read {}".format(loc_s), service_errors, label="service")
@retry.retry_method
def put(self, data, copies=2, num_retries=None):
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
if loop.success():
return thread_limiter.response()
- raise arvados.errors.KeepWriteError(
- "Write fail for %s: wanted %d but wrote %d" %
- (data_hash, copies, thread_limiter.done()))
-
- # Local storage methods need no-op num_retries arguments to keep
- # integration tests happy. With better isolation they could
- # probably be removed again.
- def local_store_put(self, data, num_retries=0):
+ if not roots_map:
+ raise arvados.errors.KeepWriteError(
+ "failed to write {}: no Keep services available ({})".format(
+ data_hash, loop.last_result()))
+ else:
+ service_errors = ((key, roots_map[key].last_result)
+ for key in local_roots
+ if not roots_map[key].success_flag)
+ raise arvados.errors.KeepWriteError(
+ "failed to write {} (wanted {} copies but wrote {})".format(
+ data_hash, copies, thread_limiter.done()), service_errors, label="service")
+
+ def local_store_put(self, data, copies=1, num_retries=None):
+ """A stub for put().
+
+ This method is used in place of the real put() method when
+ using local storage (see constructor's local_store argument).
+
+ copies and num_retries arguments are ignored: they are here
+ only for the sake of offering the same call signature as
+ put().
+
+ Data stored this way can be retrieved via local_store_get().
+ """
md5 = hashlib.md5(data).hexdigest()
locator = '%s+%d' % (md5, len(data))
with open(os.path.join(self.local_store, md5 + '.tmp'), 'w') as f:
os.path.join(self.local_store, md5))
return locator
- def local_store_get(self, loc_s, num_retries=0):
+ def local_store_get(self, loc_s, num_retries=None):
+ """Companion to local_store_put()."""
try:
locator = KeepLocator(loc_s)
except ValueError:
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)