+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
from __future__ import absolute_import
from __future__ import division
from future import standard_library
def __init__(self, root, user_agent_pool=queue.LifoQueue(),
upload_counter=None,
- download_counter=None, **headers):
+ download_counter=None,
+ headers={},
+ insecure=False):
self.root = root
self._user_agent_pool = user_agent_pool
self._result = {'error': None}
self.put_headers = headers
self.upload_counter = upload_counter
self.download_counter = download_counter
+ self.insecure = insecure
def usable(self):
"""Is it worth attempting a request?"""
'{}: {}'.format(k,v) for k,v in self.get_headers.items()])
curl.setopt(pycurl.WRITEFUNCTION, response_body.write)
curl.setopt(pycurl.HEADERFUNCTION, self._headerfunction)
+ if self.insecure:
+ curl.setopt(pycurl.SSL_VERIFYPEER, 0)
+ else:
+ curl.setopt(pycurl.CAINFO, arvados.util.ca_certs_path())
if method == "HEAD":
curl.setopt(pycurl.NOBODY, True)
- self._setcurltimeouts(curl, timeout)
+ self._setcurltimeouts(curl, timeout, method=="HEAD")
try:
curl.perform()
_logger.info("HEAD %s: %s bytes",
self._result['status_code'],
self._result.get('content-length'))
+ if self._result['headers'].get('x-keep-locator'):
+ # This is a response to a remote block copy request, return
+ # the local copy block locator.
+ return self._result['headers'].get('x-keep-locator')
return True
_logger.info("GET %s: %s bytes in %s msec (%.3f MiB/sec)",
'{}: {}'.format(k,v) for k,v in self.put_headers.items()])
curl.setopt(pycurl.WRITEFUNCTION, response_body.write)
curl.setopt(pycurl.HEADERFUNCTION, self._headerfunction)
+ if self.insecure:
+ curl.setopt(pycurl.SSL_VERIFYPEER, 0)
+ else:
+ curl.setopt(pycurl.CAINFO, arvados.util.ca_certs_path())
self._setcurltimeouts(curl, timeout)
try:
curl.perform()
self.upload_counter.add(len(body))
return True
- def _setcurltimeouts(self, curl, timeouts):
+ def _setcurltimeouts(self, curl, timeouts, ignore_bandwidth=False):
if not timeouts:
return
elif isinstance(timeouts, tuple):
conn_t, xfer_t = (timeouts, timeouts)
bandwidth_bps = KeepClient.DEFAULT_TIMEOUT[2]
curl.setopt(pycurl.CONNECTTIMEOUT_MS, int(conn_t*1000))
- curl.setopt(pycurl.LOW_SPEED_TIME, int(math.ceil(xfer_t)))
- curl.setopt(pycurl.LOW_SPEED_LIMIT, int(math.ceil(bandwidth_bps)))
+ if not ignore_bandwidth:
+ curl.setopt(pycurl.LOW_SPEED_TIME, int(math.ceil(xfer_t)))
+ curl.setopt(pycurl.LOW_SPEED_LIMIT, int(math.ceil(bandwidth_bps)))
def _headerfunction(self, header_line):
if isinstance(header_line, bytes):
self._lastheadername = name
self._headers[name] = value
# Returning None implies all bytes were written
-
+
class KeepWriterQueue(queue.Queue):
def __init__(self, copies):
self.successful_copies_lock = threading.Lock()
self.pending_tries = copies
self.pending_tries_notification = threading.Condition()
-
+
def write_success(self, response, replicas_nr):
with self.successful_copies_lock:
self.successful_copies += replicas_nr
self.response = response
with self.pending_tries_notification:
self.pending_tries_notification.notify_all()
-
+
def write_fail(self, ks):
with self.pending_tries_notification:
self.pending_tries += 1
self.pending_tries_notification.notify()
-
+
def pending_copies(self):
with self.successful_copies_lock:
return self.wanted_copies - self.successful_copies
for _ in range(num_threads):
w = KeepClient.KeepWriterThread(self.queue, data, data_hash, timeout)
self.workers.append(w)
-
+
def add_task(self, ks, service_root):
self.queue.put((ks, service_root))
self.total_task_nr += 1
-
+
def done(self):
return self.queue.successful_copies
-
+
def join(self):
# Start workers
for worker in self.workers:
worker.start()
# Wait for finished work
self.queue.join()
-
+
def response(self):
return self.queue.response
-
-
+
+
class KeepWriterThread(threading.Thread):
TaskFailed = RuntimeError()
if local_store is None:
local_store = os.environ.get('KEEP_LOCAL_STORE')
+ if api_client is None:
+ self.insecure = config.flag_is_true('ARVADOS_API_HOST_INSECURE')
+ else:
+ self.insecure = api_client.insecure
+
self.block_cache = block_cache if block_cache else KeepBlockCache()
self.timeout = timeout
self.proxy_timeout = proxy_timeout
if local_store:
self.local_store = local_store
+ self.head = self.local_store_head
self.get = self.local_store_get
self.put = self.local_store_put
else:
_logger.debug("{}: {}".format(locator, sorted_roots))
return sorted_roots
- def map_new_services(self, roots_map, locator, force_rebuild, need_writable, **headers):
+ def map_new_services(self, roots_map, locator, force_rebuild, need_writable, headers):
# roots_map is a dictionary, mapping Keep service root strings
# to KeepService objects. Poll for Keep services, and add any
# new ones to roots_map. Return the current list of local
root, self._user_agent_pool,
upload_counter=self.upload_counter,
download_counter=self.download_counter,
- **headers)
+ headers=headers,
+ insecure=self.insecure)
return local_roots
@staticmethod
else:
return None
+ def refresh_signature(self, loc):
+ """Ask Keep to get the remote block and return its local signature"""
+ now = datetime.datetime.utcnow().isoformat("T") + 'Z'
+ return self.head(loc, headers={'X-Keep-Signature': 'local, {}'.format(now)})
+
@retry.retry_method
- def head(self, loc_s, num_retries=None):
- return self._get_or_head(loc_s, method="HEAD", num_retries=num_retries)
+ def head(self, loc_s, **kwargs):
+ return self._get_or_head(loc_s, method="HEAD", **kwargs)
@retry.retry_method
- def get(self, loc_s, num_retries=None):
- return self._get_or_head(loc_s, method="GET", num_retries=num_retries)
+ def get(self, loc_s, **kwargs):
+ return self._get_or_head(loc_s, method="GET", **kwargs)
- def _get_or_head(self, loc_s, method="GET", num_retries=None):
+ def _get_or_head(self, loc_s, method="GET", num_retries=None, request_id=None, headers=None):
"""Get data from Keep.
This method fetches one or more blocks of data from Keep. It
self.get_counter.add(1)
- locator = KeepLocator(loc_s)
- if method == "GET":
- slot, first = self.block_cache.reserve_cache(locator.md5sum)
- if not first:
- self.hits_counter.add(1)
- v = slot.get()
- return v
-
- self.misses_counter.add(1)
-
- # If the locator has hints specifying a prefix (indicating a
- # remote keepproxy) or the UUID of a local gateway service,
- # read data from the indicated service(s) instead of the usual
- # list of local disk services.
- hint_roots = ['http://keep.{}.arvadosapi.com/'.format(hint[2:])
- for hint in locator.hints if hint.startswith('K@') and len(hint) == 7]
- hint_roots.extend([self._gateway_services[hint[2:]]['_service_root']
- for hint in locator.hints if (
- hint.startswith('K@') and
- len(hint) == 29 and
- self._gateway_services.get(hint[2:])
- )])
- # Map root URLs to their KeepService objects.
- roots_map = {
- root: self.KeepService(root, self._user_agent_pool,
- upload_counter=self.upload_counter,
- download_counter=self.download_counter)
- for root in hint_roots
- }
-
- # See #3147 for a discussion of the loop implementation. Highlights:
- # * Refresh the list of Keep services after each failure, in case
- # it's being updated.
- # * Retry until we succeed, we're out of retries, or every available
- # service has returned permanent failure.
- sorted_roots = []
- roots_map = {}
+ slot = None
blob = None
- loop = retry.RetryLoop(num_retries, self._check_loop_result,
- backoff_start=2)
- for tries_left in loop:
- try:
- sorted_roots = self.map_new_services(
- roots_map, locator,
- force_rebuild=(tries_left < num_retries),
- need_writable=False)
- except Exception as error:
- loop.save_result(error)
- continue
+ try:
+ locator = KeepLocator(loc_s)
+ if method == "GET":
+ slot, first = self.block_cache.reserve_cache(locator.md5sum)
+ if not first:
+ self.hits_counter.add(1)
+ blob = slot.get()
+ if blob is None:
+ raise arvados.errors.KeepReadError(
+ "failed to read {}".format(loc_s))
+ return blob
+
+ self.misses_counter.add(1)
+
+ if headers is None:
+ headers = {}
+ headers['X-Request-Id'] = (request_id or
+ (hasattr(self, 'api_client') and self.api_client.request_id) or
+ arvados.util.new_request_id())
+
+ # If the locator has hints specifying a prefix (indicating a
+ # remote keepproxy) or the UUID of a local gateway service,
+ # read data from the indicated service(s) instead of the usual
+ # list of local disk services.
+ hint_roots = ['http://keep.{}.arvadosapi.com/'.format(hint[2:])
+ for hint in locator.hints if hint.startswith('K@') and len(hint) == 7]
+ hint_roots.extend([self._gateway_services[hint[2:]]['_service_root']
+ for hint in locator.hints if (
+ hint.startswith('K@') and
+ len(hint) == 29 and
+ self._gateway_services.get(hint[2:])
+ )])
+ # Map root URLs to their KeepService objects.
+ roots_map = {
+ root: self.KeepService(root, self._user_agent_pool,
+ upload_counter=self.upload_counter,
+ download_counter=self.download_counter,
+ headers=headers,
+ insecure=self.insecure)
+ for root in hint_roots
+ }
+
+ # See #3147 for a discussion of the loop implementation. Highlights:
+ # * Refresh the list of Keep services after each failure, in case
+ # it's being updated.
+ # * Retry until we succeed, we're out of retries, or every available
+ # service has returned permanent failure.
+ sorted_roots = []
+ roots_map = {}
+ loop = retry.RetryLoop(num_retries, self._check_loop_result,
+ backoff_start=2)
+ for tries_left in loop:
+ try:
+ sorted_roots = self.map_new_services(
+ roots_map, locator,
+ force_rebuild=(tries_left < num_retries),
+ need_writable=False,
+ headers=headers)
+ except Exception as error:
+ loop.save_result(error)
+ continue
- # Query KeepService objects that haven't returned
- # permanent failure, in our specified shuffle order.
- services_to_try = [roots_map[root]
- for root in sorted_roots
- if roots_map[root].usable()]
- for keep_service in services_to_try:
- blob = keep_service.get(locator, method=method, timeout=self.current_timeout(num_retries-tries_left))
- if blob is not None:
- break
- loop.save_result((blob, len(services_to_try)))
-
- # Always cache the result, then return it if we succeeded.
- if method == "GET":
- slot.set(blob)
- self.block_cache.cap_cache()
- if loop.success():
- if method == "HEAD":
- return True
- else:
+ # Query KeepService objects that haven't returned
+ # permanent failure, in our specified shuffle order.
+ services_to_try = [roots_map[root]
+ for root in sorted_roots
+ if roots_map[root].usable()]
+ for keep_service in services_to_try:
+ blob = keep_service.get(locator, method=method, timeout=self.current_timeout(num_retries-tries_left))
+ if blob is not None:
+ break
+ loop.save_result((blob, len(services_to_try)))
+
+ # Always cache the result, then return it if we succeeded.
+ if loop.success():
return blob
+ finally:
+ if slot is not None:
+ slot.set(blob)
+ self.block_cache.cap_cache()
# Q: Including 403 is necessary for the Keep tests to continue
# passing, but maybe they should expect KeepReadError instead?
"{} not found".format(loc_s), service_errors)
else:
raise arvados.errors.KeepReadError(
- "failed to read {}".format(loc_s), service_errors, label="service")
+ "failed to read {} after {}".format(loc_s, loop.attempts_str()), service_errors, label="service")
@retry.retry_method
- def put(self, data, copies=2, num_retries=None):
+ def put(self, data, copies=2, num_retries=None, request_id=None):
"""Save data in Keep.
This method will get a list of Keep services from the API server, and
return loc_s
locator = KeepLocator(loc_s)
- headers = {}
- # Tell the proxy how many copies we want it to store
- headers['X-Keep-Desired-Replicas'] = str(copies)
+ headers = {
+ 'X-Request-Id': (request_id or
+ (hasattr(self, 'api_client') and self.api_client.request_id) or
+ arvados.util.new_request_id()),
+ 'X-Keep-Desired-Replicas': str(copies),
+ }
roots_map = {}
loop = retry.RetryLoop(num_retries, self._check_loop_result,
backoff_start=2)
try:
sorted_roots = self.map_new_services(
roots_map, locator,
- force_rebuild=(tries_left < num_retries), need_writable=True, **headers)
+ force_rebuild=(tries_left < num_retries),
+ need_writable=True,
+ headers=headers)
except Exception as error:
loop.save_result(error)
continue
- writer_pool = KeepClient.KeepWriterThreadPool(data=data,
+ writer_pool = KeepClient.KeepWriterThreadPool(data=data,
data_hash=data_hash,
copies=copies - done,
max_service_replicas=self.max_replicas_per_service,
for key in sorted_roots
if roots_map[key].last_result()['error'])
raise arvados.errors.KeepWriteError(
- "failed to write {} (wanted {} copies but wrote {})".format(
- data_hash, copies, writer_pool.done()), service_errors, label="service")
+ "failed to write {} after {} (wanted {} copies but wrote {})".format(
+ data_hash, loop.attempts_str(), copies, writer_pool.done()), service_errors, label="service")
def local_store_put(self, data, copies=1, num_retries=None):
"""A stub for put().
with open(os.path.join(self.local_store, locator.md5sum), 'rb') as f:
return f.read()
+ def local_store_head(self, loc_s, num_retries=None):
+ """Companion to local_store_put()."""
+ try:
+ locator = KeepLocator(loc_s)
+ except ValueError:
+ raise arvados.errors.NotFoundError(
+ "Invalid data locator: '%s'" % loc_s)
+ if locator.md5sum == config.EMPTY_BLOCK_LOCATOR.split('+')[0]:
+ return True
+ if os.path.exists(os.path.join(self.local_store, locator.md5sum)):
+ return True
+
def is_cached(self, locator):
return self.block_cache.reserve_cache(expect_hash)
projects / arvados.git / blobdiff