sdk/python/arvados/keep.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 from __future__ import absolute_import
   6 from __future__ import division
   7 import copy
   8 from future import standard_library
   9 from future.utils import native_str
  10 standard_library.install_aliases()
  11 from builtins import next
  12 from builtins import str
  13 from builtins import range
  14 from builtins import object
  15 import collections
  16 import datetime
  17 import hashlib
  18 import io
  19 import logging
  20 import math
  21 import os
  22 import pycurl
  23 import queue
  24 import re
  25 import socket
  26 import ssl
  27 import sys
  28 import threading
  29 from . import timer
  30 import urllib.parse
  31
  32 if sys.version_info >= (3, 0):
  33     from io import BytesIO
  34 else:
  35     from cStringIO import StringIO as BytesIO
  36
  37 import arvados
  38 import arvados.config as config
  39 import arvados.errors
  40 import arvados.retry as retry
  41 import arvados.util
  42
  43 _logger = logging.getLogger('arvados.keep')
  44 global_client_object = None
  45
  46
  47 # Monkey patch TCP constants when not available (apple). Values sourced from:
  48 # http://www.opensource.apple.com/source/xnu/xnu-2422.115.4/bsd/netinet/tcp.h
  49 if sys.platform == 'darwin':
  50     if not hasattr(socket, 'TCP_KEEPALIVE'):
  51         socket.TCP_KEEPALIVE = 0x010
  52     if not hasattr(socket, 'TCP_KEEPINTVL'):
  53         socket.TCP_KEEPINTVL = 0x101
  54     if not hasattr(socket, 'TCP_KEEPCNT'):
  55         socket.TCP_KEEPCNT = 0x102
  56
  57
  58 class KeepLocator(object):
  59     EPOCH_DATETIME = datetime.datetime.utcfromtimestamp(0)
  60     HINT_RE = re.compile(r'^[A-Z][A-Za-z0-9@_-]+$')
  61
  62     def __init__(self, locator_str):
  63         self.hints = []
  64         self._perm_sig = None
  65         self._perm_expiry = None
  66         pieces = iter(locator_str.split('+'))
  67         self.md5sum = next(pieces)
  68         try:
  69             self.size = int(next(pieces))
  70         except StopIteration:
  71             self.size = None
  72         for hint in pieces:
  73             if self.HINT_RE.match(hint) is None:
  74                 raise ValueError("invalid hint format: {}".format(hint))
  75             elif hint.startswith('A'):
  76                 self.parse_permission_hint(hint)
  77             else:
  78                 self.hints.append(hint)
  79
  80     def __str__(self):
  81         return '+'.join(
  82             native_str(s)
  83             for s in [self.md5sum, self.size,
  84                       self.permission_hint()] + self.hints
  85             if s is not None)
  86
  87     def stripped(self):
  88         if self.size is not None:
  89             return "%s+%i" % (self.md5sum, self.size)
  90         else:
  91             return self.md5sum
  92
  93     def _make_hex_prop(name, length):
  94         # Build and return a new property with the given name that
  95         # must be a hex string of the given length.
  96         data_name = '_{}'.format(name)
  97         def getter(self):
  98             return getattr(self, data_name)
  99         def setter(self, hex_str):
 100             if not arvados.util.is_hex(hex_str, length):
 101                 raise ValueError("{} is not a {}-digit hex string: {!r}".
 102                                  format(name, length, hex_str))
 103             setattr(self, data_name, hex_str)
 104         return property(getter, setter)
 105
 106     md5sum = _make_hex_prop('md5sum', 32)
 107     perm_sig = _make_hex_prop('perm_sig', 40)
 108
 109     @property
 110     def perm_expiry(self):
 111         return self._perm_expiry
 112
 113     @perm_expiry.setter
 114     def perm_expiry(self, value):
 115         if not arvados.util.is_hex(value, 1, 8):
 116             raise ValueError(
 117                 "permission timestamp must be a hex Unix timestamp: {}".
 118                 format(value))
 119         self._perm_expiry = datetime.datetime.utcfromtimestamp(int(value, 16))
 120
 121     def permission_hint(self):
 122         data = [self.perm_sig, self.perm_expiry]
 123         if None in data:
 124             return None
 125         data[1] = int((data[1] - self.EPOCH_DATETIME).total_seconds())
 126         return "A{}@{:08x}".format(*data)
 127
 128     def parse_permission_hint(self, s):
 129         try:
 130             self.perm_sig, self.perm_expiry = s[1:].split('@', 1)
 131         except IndexError:
 132             raise ValueError("bad permission hint {}".format(s))
 133
 134     def permission_expired(self, as_of_dt=None):
 135         if self.perm_expiry is None:
 136             return False
 137         elif as_of_dt is None:
 138             as_of_dt = datetime.datetime.now()
 139         return self.perm_expiry <= as_of_dt
 140
 141
 142 class Keep(object):
 143     """Simple interface to a global KeepClient object.
 144
 145     THIS CLASS IS DEPRECATED.  Please instantiate your own KeepClient with your
 146     own API client.  The global KeepClient will build an API client from the
 147     current Arvados configuration, which may not match the one you built.
 148     """
 149     _last_key = None
 150
 151     @classmethod
 152     def global_client_object(cls):
 153         global global_client_object
 154         # Previously, KeepClient would change its behavior at runtime based
 155         # on these configuration settings.  We simulate that behavior here
 156         # by checking the values and returning a new KeepClient if any of
 157         # them have changed.
 158         key = (config.get('ARVADOS_API_HOST'),
 159                config.get('ARVADOS_API_TOKEN'),
 160                config.flag_is_true('ARVADOS_API_HOST_INSECURE'),
 161                config.get('ARVADOS_KEEP_PROXY'),
 162                config.get('ARVADOS_EXTERNAL_CLIENT') == 'true',
 163                os.environ.get('KEEP_LOCAL_STORE'))
 164         if (global_client_object is None) or (cls._last_key != key):
 165             global_client_object = KeepClient()
 166             cls._last_key = key
 167         return global_client_object
 168
 169     @staticmethod
 170     def get(locator, **kwargs):
 171         return Keep.global_client_object().get(locator, **kwargs)
 172
 173     @staticmethod
 174     def put(data, **kwargs):
 175         return Keep.global_client_object().put(data, **kwargs)
 176
 177 class KeepBlockCache(object):
 178     # Default RAM cache is 256MiB
 179     def __init__(self, cache_max=(256 * 1024 * 1024)):
 180         self.cache_max = cache_max
 181         self._cache = []
 182         self._cache_lock = threading.Lock()
 183
 184     class CacheSlot(object):
 185         __slots__ = ("locator", "ready", "content")
 186
 187         def __init__(self, locator):
 188             self.locator = locator
 189             self.ready = threading.Event()
 190             self.content = None
 191
 192         def get(self):
 193             self.ready.wait()
 194             return self.content
 195
 196         def set(self, value):
 197             self.content = value
 198             self.ready.set()
 199
 200         def size(self):
 201             if self.content is None:
 202                 return 0
 203             else:
 204                 return len(self.content)
 205
 206     def cap_cache(self):
 207         '''Cap the cache size to self.cache_max'''
 208         with self._cache_lock:
 209             # Select all slots except those where ready.is_set() and content is
 210             # None (that means there was an error reading the block).
 211             self._cache = [c for c in self._cache if not (c.ready.is_set() and c.content is None)]
 212             sm = sum([slot.size() for slot in self._cache])
 213             while len(self._cache) > 0 and sm > self.cache_max:
 214                 for i in range(len(self._cache)-1, -1, -1):
 215                     if self._cache[i].ready.is_set():
 216                         del self._cache[i]
 217                         break
 218                 sm = sum([slot.size() for slot in self._cache])
 219
 220     def _get(self, locator):
 221         # Test if the locator is already in the cache
 222         for i in range(0, len(self._cache)):
 223             if self._cache[i].locator == locator:
 224                 n = self._cache[i]
 225                 if i != 0:
 226                     # move it to the front
 227                     del self._cache[i]
 228                     self._cache.insert(0, n)
 229                 return n
 230         return None
 231
 232     def get(self, locator):
 233         with self._cache_lock:
 234             return self._get(locator)
 235
 236     def reserve_cache(self, locator):
 237         '''Reserve a cache slot for the specified locator,
 238         or return the existing slot.'''
 239         with self._cache_lock:
 240             n = self._get(locator)
 241             if n:
 242                 return n, False
 243             else:
 244                 # Add a new cache slot for the locator
 245                 n = KeepBlockCache.CacheSlot(locator)
 246                 self._cache.insert(0, n)
 247                 return n, True
 248
 249 class Counter(object):
 250     def __init__(self, v=0):
 251         self._lk = threading.Lock()
 252         self._val = v
 253
 254     def add(self, v):
 255         with self._lk:
 256             self._val += v
 257
 258     def get(self):
 259         with self._lk:
 260             return self._val
 261
 262
 263 class KeepClient(object):
 264
 265     # Default Keep server connection timeout:  2 seconds
 266     # Default Keep server read timeout:       256 seconds
 267     # Default Keep server bandwidth minimum:  32768 bytes per second
 268     # Default Keep proxy connection timeout:  20 seconds
 269     # Default Keep proxy read timeout:        256 seconds
 270     # Default Keep proxy bandwidth minimum:   32768 bytes per second
 271     DEFAULT_TIMEOUT = (2, 256, 32768)
 272     DEFAULT_PROXY_TIMEOUT = (20, 256, 32768)
 273
 274
 275     class KeepService(object):
 276         """Make requests to a single Keep service, and track results.
 277
 278         A KeepService is intended to last long enough to perform one
 279         transaction (GET or PUT) against one Keep service. This can
 280         involve calling either get() or put() multiple times in order
 281         to retry after transient failures. However, calling both get()
 282         and put() on a single instance -- or using the same instance
 283         to access two different Keep services -- will not produce
 284         sensible behavior.
 285         """
 286
 287         HTTP_ERRORS = (
 288             socket.error,
 289             ssl.SSLError,
 290             arvados.errors.HttpError,
 291         )
 292
 293         def __init__(self, root, user_agent_pool=queue.LifoQueue(),
 294                      upload_counter=None,
 295                      download_counter=None,
 296                      headers={},
 297                      insecure=False):
 298             self.root = root
 299             self._user_agent_pool = user_agent_pool
 300             self._result = {'error': None}
 301             self._usable = True
 302             self._session = None
 303             self._socket = None
 304             self.get_headers = {'Accept': 'application/octet-stream'}
 305             self.get_headers.update(headers)
 306             self.put_headers = headers
 307             self.upload_counter = upload_counter
 308             self.download_counter = download_counter
 309             self.insecure = insecure
 310
 311         def usable(self):
 312             """Is it worth attempting a request?"""
 313             return self._usable
 314
 315         def finished(self):
 316             """Did the request succeed or encounter permanent failure?"""
 317             return self._result['error'] == False or not self._usable
 318
 319         def last_result(self):
 320             return self._result
 321
 322         def _get_user_agent(self):
 323             try:
 324                 return self._user_agent_pool.get(block=False)
 325             except queue.Empty:
 326                 return pycurl.Curl()
 327
 328         def _put_user_agent(self, ua):
 329             try:
 330                 ua.reset()
 331                 self._user_agent_pool.put(ua, block=False)
 332             except:
 333                 ua.close()
 334
 335         def _socket_open(self, *args, **kwargs):
 336             if len(args) + len(kwargs) == 2:
 337                 return self._socket_open_pycurl_7_21_5(*args, **kwargs)
 338             else:
 339                 return self._socket_open_pycurl_7_19_3(*args, **kwargs)
 340
 341         def _socket_open_pycurl_7_19_3(self, family, socktype, protocol, address=None):
 342             return self._socket_open_pycurl_7_21_5(
 343                 purpose=None,
 344                 address=collections.namedtuple(
 345                     'Address', ['family', 'socktype', 'protocol', 'addr'],
 346                 )(family, socktype, protocol, address))
 347
 348         def _socket_open_pycurl_7_21_5(self, purpose, address):
 349             """Because pycurl doesn't have CURLOPT_TCP_KEEPALIVE"""
 350             s = socket.socket(address.family, address.socktype, address.protocol)
 351             s.setsockopt(socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1)
 352             # Will throw invalid protocol error on mac. This test prevents that.
 353             if hasattr(socket, 'TCP_KEEPIDLE'):
 354                 s.setsockopt(socket.IPPROTO_TCP, socket.TCP_KEEPIDLE, 75)
 355             s.setsockopt(socket.IPPROTO_TCP, socket.TCP_KEEPINTVL, 75)
 356             self._socket = s
 357             return s
 358
 359         def get(self, locator, method="GET", timeout=None):
 360             # locator is a KeepLocator object.
 361             url = self.root + str(locator)
 362             _logger.debug("Request: %s %s", method, url)
 363             curl = self._get_user_agent()
 364             ok = None
 365             try:
 366                 with timer.Timer() as t:
 367                     self._headers = {}
 368                     response_body = BytesIO()
 369                     curl.setopt(pycurl.NOSIGNAL, 1)
 370                     curl.setopt(pycurl.OPENSOCKETFUNCTION,
 371                                 lambda *args, **kwargs: self._socket_open(*args, **kwargs))
 372                     curl.setopt(pycurl.URL, url.encode('utf-8'))
 373                     curl.setopt(pycurl.HTTPHEADER, [
 374                         '{}: {}'.format(k,v) for k,v in self.get_headers.items()])
 375                     curl.setopt(pycurl.WRITEFUNCTION, response_body.write)
 376                     curl.setopt(pycurl.HEADERFUNCTION, self._headerfunction)
 377                     if self.insecure:
 378                         curl.setopt(pycurl.SSL_VERIFYPEER, 0)
 379                     else:
 380                         curl.setopt(pycurl.CAINFO, arvados.util.ca_certs_path())
 381                     if method == "HEAD":
 382                         curl.setopt(pycurl.NOBODY, True)
 383                     self._setcurltimeouts(curl, timeout, method=="HEAD")
 384
 385                     try:
 386                         curl.perform()
 387                     except Exception as e:
 388                         raise arvados.errors.HttpError(0, str(e))
 389                     finally:
 390                         if self._socket:
 391                             self._socket.close()
 392                             self._socket = None
 393                     self._result = {
 394                         'status_code': curl.getinfo(pycurl.RESPONSE_CODE),
 395                         'body': response_body.getvalue(),
 396                         'headers': self._headers,
 397                         'error': False,
 398                     }
 399
 400                 ok = retry.check_http_response_success(self._result['status_code'])
 401                 if not ok:
 402                     self._result['error'] = arvados.errors.HttpError(
 403                         self._result['status_code'],
 404                         self._headers.get('x-status-line', 'Error'))
 405             except self.HTTP_ERRORS as e:
 406                 self._result = {
 407                     'error': e,
 408                 }
 409             self._usable = ok != False
 410             if self._result.get('status_code', None):
 411                 # The client worked well enough to get an HTTP status
 412                 # code, so presumably any problems are just on the
 413                 # server side and it's OK to reuse the client.
 414                 self._put_user_agent(curl)
 415             else:
 416                 # Don't return this client to the pool, in case it's
 417                 # broken.
 418                 curl.close()
 419             if not ok:
 420                 _logger.debug("Request fail: GET %s => %s: %s",
 421                               url, type(self._result['error']), str(self._result['error']))
 422                 return None
 423             if method == "HEAD":
 424                 _logger.info("HEAD %s: %s bytes",
 425                          self._result['status_code'],
 426                          self._result.get('content-length'))
 427                 if self._result['headers'].get('x-keep-locator'):
 428                     # This is a response to a remote block copy request, return
 429                     # the local copy block locator.
 430                     return self._result['headers'].get('x-keep-locator')
 431                 return True
 432
 433             _logger.info("GET %s: %s bytes in %s msec (%.3f MiB/sec)",
 434                          self._result['status_code'],
 435                          len(self._result['body']),
 436                          t.msecs,
 437                          1.0*len(self._result['body'])/2**20/t.secs if t.secs > 0 else 0)
 438
 439             if self.download_counter:
 440                 self.download_counter.add(len(self._result['body']))
 441             resp_md5 = hashlib.md5(self._result['body']).hexdigest()
 442             if resp_md5 != locator.md5sum:
 443                 _logger.warning("Checksum fail: md5(%s) = %s",
 444                                 url, resp_md5)
 445                 self._result['error'] = arvados.errors.HttpError(
 446                     0, 'Checksum fail')
 447                 return None
 448             return self._result['body']
 449
 450         def put(self, hash_s, body, timeout=None, headers={}):
 451             put_headers = copy.copy(self.put_headers)
 452             put_headers.update(headers)
 453             url = self.root + hash_s
 454             _logger.debug("Request: PUT %s", url)
 455             curl = self._get_user_agent()
 456             ok = None
 457             try:
 458                 with timer.Timer() as t:
 459                     self._headers = {}
 460                     body_reader = BytesIO(body)
 461                     response_body = BytesIO()
 462                     curl.setopt(pycurl.NOSIGNAL, 1)
 463                     curl.setopt(pycurl.OPENSOCKETFUNCTION,
 464                                 lambda *args, **kwargs: self._socket_open(*args, **kwargs))
 465                     curl.setopt(pycurl.URL, url.encode('utf-8'))
 466                     # Using UPLOAD tells cURL to wait for a "go ahead" from the
 467                     # Keep server (in the form of a HTTP/1.1 "100 Continue"
 468                     # response) instead of sending the request body immediately.
 469                     # This allows the server to reject the request if the request
 470                     # is invalid or the server is read-only, without waiting for
 471                     # the client to send the entire block.
 472                     curl.setopt(pycurl.UPLOAD, True)
 473                     curl.setopt(pycurl.INFILESIZE, len(body))
 474                     curl.setopt(pycurl.READFUNCTION, body_reader.read)
 475                     curl.setopt(pycurl.HTTPHEADER, [
 476                         '{}: {}'.format(k,v) for k,v in put_headers.items()])
 477                     curl.setopt(pycurl.WRITEFUNCTION, response_body.write)
 478                     curl.setopt(pycurl.HEADERFUNCTION, self._headerfunction)
 479                     if self.insecure:
 480                         curl.setopt(pycurl.SSL_VERIFYPEER, 0)
 481                     else:
 482                         curl.setopt(pycurl.CAINFO, arvados.util.ca_certs_path())
 483                     self._setcurltimeouts(curl, timeout)
 484                     try:
 485                         curl.perform()
 486                     except Exception as e:
 487                         raise arvados.errors.HttpError(0, str(e))
 488                     finally:
 489                         if self._socket:
 490                             self._socket.close()
 491                             self._socket = None
 492                     self._result = {
 493                         'status_code': curl.getinfo(pycurl.RESPONSE_CODE),
 494                         'body': response_body.getvalue().decode('utf-8'),
 495                         'headers': self._headers,
 496                         'error': False,
 497                     }
 498                 ok = retry.check_http_response_success(self._result['status_code'])
 499                 if not ok:
 500                     self._result['error'] = arvados.errors.HttpError(
 501                         self._result['status_code'],
 502                         self._headers.get('x-status-line', 'Error'))
 503             except self.HTTP_ERRORS as e:
 504                 self._result = {
 505                     'error': e,
 506                 }
 507             self._usable = ok != False # still usable if ok is True or None
 508             if self._result.get('status_code', None):
 509                 # Client is functional. See comment in get().
 510                 self._put_user_agent(curl)
 511             else:
 512                 curl.close()
 513             if not ok:
 514                 _logger.debug("Request fail: PUT %s => %s: %s",
 515                               url, type(self._result['error']), str(self._result['error']))
 516                 return False
 517             _logger.info("PUT %s: %s bytes in %s msec (%.3f MiB/sec)",
 518                          self._result['status_code'],
 519                          len(body),
 520                          t.msecs,
 521                          1.0*len(body)/2**20/t.secs if t.secs > 0 else 0)
 522             if self.upload_counter:
 523                 self.upload_counter.add(len(body))
 524             return True
 525
 526         def _setcurltimeouts(self, curl, timeouts, ignore_bandwidth=False):
 527             if not timeouts:
 528                 return
 529             elif isinstance(timeouts, tuple):
 530                 if len(timeouts) == 2:
 531                     conn_t, xfer_t = timeouts
 532                     bandwidth_bps = KeepClient.DEFAULT_TIMEOUT[2]
 533                 else:
 534                     conn_t, xfer_t, bandwidth_bps = timeouts
 535             else:
 536                 conn_t, xfer_t = (timeouts, timeouts)
 537                 bandwidth_bps = KeepClient.DEFAULT_TIMEOUT[2]
 538             curl.setopt(pycurl.CONNECTTIMEOUT_MS, int(conn_t*1000))
 539             if not ignore_bandwidth:
 540                 curl.setopt(pycurl.LOW_SPEED_TIME, int(math.ceil(xfer_t)))
 541                 curl.setopt(pycurl.LOW_SPEED_LIMIT, int(math.ceil(bandwidth_bps)))
 542
 543         def _headerfunction(self, header_line):
 544             if isinstance(header_line, bytes):
 545                 header_line = header_line.decode('iso-8859-1')
 546             if ':' in header_line:
 547                 name, value = header_line.split(':', 1)
 548                 name = name.strip().lower()
 549                 value = value.strip()
 550             elif self._headers:
 551                 name = self._lastheadername
 552                 value = self._headers[name] + ' ' + header_line.strip()
 553             elif header_line.startswith('HTTP/'):
 554                 name = 'x-status-line'
 555                 value = header_line
 556             else:
 557                 _logger.error("Unexpected header line: %s", header_line)
 558                 return
 559             self._lastheadername = name
 560             self._headers[name] = value
 561             # Returning None implies all bytes were written
 562
 563
 564     class KeepWriterQueue(queue.Queue):
 565         def __init__(self, copies, classes=[]):
 566             queue.Queue.__init__(self) # Old-style superclass
 567             self.wanted_copies = copies
 568             self.wanted_storage_classes = classes
 569             self.successful_copies = 0
 570             self.confirmed_storage_classes = {}
 571             self.response = None
 572             self.storage_classes_tracking = True
 573             self.queue_data_lock = threading.Lock()
 574             self.pending_tries = max(copies, len(classes))
 575             self.pending_tries_notification = threading.Condition()
 576
 577         def write_success(self, response, replicas_nr, classes_confirmed):
 578             with self.queue_data_lock:
 579                 self.successful_copies += replicas_nr
 580                 if classes_confirmed is None:
 581                     self.storage_classes_tracking = False
 582                 elif self.storage_classes_tracking:
 583                     for st_class, st_copies in classes_confirmed.items():
 584                         try:
 585                             self.confirmed_storage_classes[st_class] += st_copies
 586                         except KeyError:
 587                             self.confirmed_storage_classes[st_class] = st_copies
 588                 self.response = response
 589             with self.pending_tries_notification:
 590                 self.pending_tries_notification.notify_all()
 591
 592         def write_fail(self, ks):
 593             with self.pending_tries_notification:
 594                 self.pending_tries += 1
 595                 self.pending_tries_notification.notify()
 596
 597         def pending_copies(self):
 598             with self.queue_data_lock:
 599                 return self.wanted_copies - self.successful_copies
 600
 601         def satisfied_classes(self):
 602             with self.queue_data_lock:
 603                 if not self.storage_classes_tracking:
 604                     # Notifies disabled storage classes expectation to
 605                     # the outer loop.
 606                     return None
 607             return list(set(self.wanted_storage_classes) - set(self.pending_classes()))
 608
 609         def pending_classes(self):
 610             with self.queue_data_lock:
 611                 if self.wanted_storage_classes is None:
 612                     return []
 613                 unsatisfied_classes = copy.copy(self.wanted_storage_classes)
 614                 for st_class, st_copies in self.confirmed_storage_classes.items():
 615                     if st_class in unsatisfied_classes and st_copies >= self.wanted_copies:
 616                         unsatisfied_classes.remove(st_class)
 617                 return unsatisfied_classes
 618
 619         def get_next_task(self):
 620             with self.pending_tries_notification:
 621                 while True:
 622                     if self.pending_copies() < 1 and len(self.pending_classes()) == 0:
 623                         # This notify_all() is unnecessary --
 624                         # write_success() already called notify_all()
 625                         # when pending<1 became true, so it's not
 626                         # possible for any other thread to be in
 627                         # wait() now -- but it's cheap insurance
 628                         # against deadlock so we do it anyway:
 629                         self.pending_tries_notification.notify_all()
 630                         # Drain the queue and then raise Queue.Empty
 631                         while True:
 632                             self.get_nowait()
 633                             self.task_done()
 634                     elif self.pending_tries > 0:
 635                         service, service_root = self.get_nowait()
 636                         if service.finished():
 637                             self.task_done()
 638                             continue
 639                         self.pending_tries -= 1
 640                         return service, service_root
 641                     elif self.empty():
 642                         self.pending_tries_notification.notify_all()
 643                         raise queue.Empty
 644                     else:
 645                         self.pending_tries_notification.wait()
 646
 647
 648     class KeepWriterThreadPool(object):
 649         def __init__(self, data, data_hash, copies, max_service_replicas, timeout=None, classes=[]):
 650             self.total_task_nr = 0
 651             if (not max_service_replicas) or (max_service_replicas >= copies):
 652                 num_threads = 1
 653             else:
 654                 num_threads = int(math.ceil(1.0*copies/max_service_replicas))
 655             _logger.debug("Pool max threads is %d", num_threads)
 656             self.workers = []
 657             self.queue = KeepClient.KeepWriterQueue(copies, classes)
 658             # Create workers
 659             for _ in range(num_threads):
 660                 w = KeepClient.KeepWriterThread(self.queue, data, data_hash, timeout)
 661                 self.workers.append(w)
 662
 663         def add_task(self, ks, service_root):
 664             self.queue.put((ks, service_root))
 665             self.total_task_nr += 1
 666
 667         def done(self):
 668             return self.queue.successful_copies, self.queue.satisfied_classes()
 669
 670         def join(self):
 671             # Start workers
 672             for worker in self.workers:
 673                 worker.start()
 674             # Wait for finished work
 675             self.queue.join()
 676
 677         def response(self):
 678             return self.queue.response
 679
 680
 681     class KeepWriterThread(threading.Thread):
 682         class TaskFailed(RuntimeError): pass
 683
 684         def __init__(self, queue, data, data_hash, timeout=None):
 685             super(KeepClient.KeepWriterThread, self).__init__()
 686             self.timeout = timeout
 687             self.queue = queue
 688             self.data = data
 689             self.data_hash = data_hash
 690             self.daemon = True
 691
 692         def run(self):
 693             while True:
 694                 try:
 695                     service, service_root = self.queue.get_next_task()
 696                 except queue.Empty:
 697                     return
 698                 try:
 699                     locator, copies, classes = self.do_task(service, service_root)
 700                 except Exception as e:
 701                     if not isinstance(e, self.TaskFailed):
 702                         _logger.exception("Exception in KeepWriterThread")
 703                     self.queue.write_fail(service)
 704                 else:
 705                     self.queue.write_success(locator, copies, classes)
 706                 finally:
 707                     self.queue.task_done()
 708
 709         def do_task(self, service, service_root):
 710             classes = self.queue.pending_classes()
 711             headers = {}
 712             if len(classes) > 0:
 713                 headers['X-Keep-Storage-Classes'] = ', '.join(classes)
 714             success = bool(service.put(self.data_hash,
 715                                         self.data,
 716                                         timeout=self.timeout,
 717                                         headers=headers))
 718             result = service.last_result()
 719
 720             if not success:
 721                 if result.get('status_code', None):
 722                     _logger.debug("Request fail: PUT %s => %s %s",
 723                                   self.data_hash,
 724                                   result['status_code'],
 725                                   result['body'])
 726                 raise self.TaskFailed()
 727
 728             _logger.debug("KeepWriterThread %s succeeded %s+%i %s",
 729                           str(threading.current_thread()),
 730                           self.data_hash,
 731                           len(self.data),
 732                           service_root)
 733             try:
 734                 replicas_stored = int(result['headers']['x-keep-replicas-stored'])
 735             except (KeyError, ValueError):
 736                 replicas_stored = 1
 737
 738             classes_confirmed = {}
 739             try:
 740                 scch = result['headers']['x-keep-storage-classes-confirmed']
 741                 for confirmation in scch.replace(' ', '').split(','):
 742                     if '=' in confirmation:
 743                         stored_class, stored_copies = confirmation.split('=')[:2]
 744                         classes_confirmed[stored_class] = int(stored_copies)
 745             except (KeyError, ValueError):
 746                 # Storage classes confirmed header missing or corrupt
 747                 classes_confirmed = None
 748
 749             return result['body'].strip(), replicas_stored, classes_confirmed
 750
 751
 752     def __init__(self, api_client=None, proxy=None,
 753                  timeout=DEFAULT_TIMEOUT, proxy_timeout=DEFAULT_PROXY_TIMEOUT,
 754                  api_token=None, local_store=None, block_cache=None,
 755                  num_retries=0, session=None):
 756         """Initialize a new KeepClient.
 757
 758         Arguments:
 759         :api_client:
 760           The API client to use to find Keep services.  If not
 761           provided, KeepClient will build one from available Arvados
 762           configuration.
 763
 764         :proxy:
 765           If specified, this KeepClient will send requests to this Keep
 766           proxy.  Otherwise, KeepClient will fall back to the setting of the
 767           ARVADOS_KEEP_SERVICES or ARVADOS_KEEP_PROXY configuration settings.
 768           If you want to KeepClient does not use a proxy, pass in an empty
 769           string.
 770
 771         :timeout:
 772           The initial timeout (in seconds) for HTTP requests to Keep
 773           non-proxy servers.  A tuple of three floats is interpreted as
 774           (connection_timeout, read_timeout, minimum_bandwidth). A connection
 775           will be aborted if the average traffic rate falls below
 776           minimum_bandwidth bytes per second over an interval of read_timeout
 777           seconds. Because timeouts are often a result of transient server
 778           load, the actual connection timeout will be increased by a factor
 779           of two on each retry.
 780           Default: (2, 256, 32768).
 781
 782         :proxy_timeout:
 783           The initial timeout (in seconds) for HTTP requests to
 784           Keep proxies. A tuple of three floats is interpreted as
 785           (connection_timeout, read_timeout, minimum_bandwidth). The behavior
 786           described above for adjusting connection timeouts on retry also
 787           applies.
 788           Default: (20, 256, 32768).
 789
 790         :api_token:
 791           If you're not using an API client, but only talking
 792           directly to a Keep proxy, this parameter specifies an API token
 793           to authenticate Keep requests.  It is an error to specify both
 794           api_client and api_token.  If you specify neither, KeepClient
 795           will use one available from the Arvados configuration.
 796
 797         :local_store:
 798           If specified, this KeepClient will bypass Keep
 799           services, and save data to the named directory.  If unspecified,
 800           KeepClient will fall back to the setting of the $KEEP_LOCAL_STORE
 801           environment variable.  If you want to ensure KeepClient does not
 802           use local storage, pass in an empty string.  This is primarily
 803           intended to mock a server for testing.
 804
 805         :num_retries:
 806           The default number of times to retry failed requests.
 807           This will be used as the default num_retries value when get() and
 808           put() are called.  Default 0.
 809         """
 810         self.lock = threading.Lock()
 811         if proxy is None:
 812             if config.get('ARVADOS_KEEP_SERVICES'):
 813                 proxy = config.get('ARVADOS_KEEP_SERVICES')
 814             else:
 815                 proxy = config.get('ARVADOS_KEEP_PROXY')
 816         if api_token is None:
 817             if api_client is None:
 818                 api_token = config.get('ARVADOS_API_TOKEN')
 819             else:
 820                 api_token = api_client.api_token
 821         elif api_client is not None:
 822             raise ValueError(
 823                 "can't build KeepClient with both API client and token")
 824         if local_store is None:
 825             local_store = os.environ.get('KEEP_LOCAL_STORE')
 826
 827         if api_client is None:
 828             self.insecure = config.flag_is_true('ARVADOS_API_HOST_INSECURE')
 829         else:
 830             self.insecure = api_client.insecure
 831
 832         self.block_cache = block_cache if block_cache else KeepBlockCache()
 833         self.timeout = timeout
 834         self.proxy_timeout = proxy_timeout
 835         self._user_agent_pool = queue.LifoQueue()
 836         self.upload_counter = Counter()
 837         self.download_counter = Counter()
 838         self.put_counter = Counter()
 839         self.get_counter = Counter()
 840         self.hits_counter = Counter()
 841         self.misses_counter = Counter()
 842
 843         if local_store:
 844             self.local_store = local_store
 845             self.head = self.local_store_head
 846             self.get = self.local_store_get
 847             self.put = self.local_store_put
 848         else:
 849             self.num_retries = num_retries
 850             self.max_replicas_per_service = None
 851             if proxy:
 852                 proxy_uris = proxy.split()
 853                 for i in range(len(proxy_uris)):
 854                     if not proxy_uris[i].endswith('/'):
 855                         proxy_uris[i] += '/'
 856                     # URL validation
 857                     url = urllib.parse.urlparse(proxy_uris[i])
 858                     if not (url.scheme and url.netloc):
 859                         raise arvados.errors.ArgumentError("Invalid proxy URI: {}".format(proxy_uris[i]))
 860                 self.api_token = api_token
 861                 self._gateway_services = {}
 862                 self._keep_services = [{
 863                     'uuid': "00000-bi6l4-%015d" % idx,
 864                     'service_type': 'proxy',
 865                     '_service_root': uri,
 866                     } for idx, uri in enumerate(proxy_uris)]
 867                 self._writable_services = self._keep_services
 868                 self.using_proxy = True
 869                 self._static_services_list = True
 870             else:
 871                 # It's important to avoid instantiating an API client
 872                 # unless we actually need one, for testing's sake.
 873                 if api_client is None:
 874                     api_client = arvados.api('v1')
 875                 self.api_client = api_client
 876                 self.api_token = api_client.api_token
 877                 self._gateway_services = {}
 878                 self._keep_services = None
 879                 self._writable_services = None
 880                 self.using_proxy = None
 881                 self._static_services_list = False
 882
 883     def current_timeout(self, attempt_number):
 884         """Return the appropriate timeout to use for this client.
 885
 886         The proxy timeout setting if the backend service is currently a proxy,
 887         the regular timeout setting otherwise.  The `attempt_number` indicates
 888         how many times the operation has been tried already (starting from 0
 889         for the first try), and scales the connection timeout portion of the
 890         return value accordingly.
 891
 892         """
 893         # TODO(twp): the timeout should be a property of a
 894         # KeepService, not a KeepClient. See #4488.
 895         t = self.proxy_timeout if self.using_proxy else self.timeout
 896         if len(t) == 2:
 897             return (t[0] * (1 << attempt_number), t[1])
 898         else:
 899             return (t[0] * (1 << attempt_number), t[1], t[2])
 900     def _any_nondisk_services(self, service_list):
 901         return any(ks.get('service_type', 'disk') != 'disk'
 902                    for ks in service_list)
 903
 904     def build_services_list(self, force_rebuild=False):
 905         if (self._static_services_list or
 906               (self._keep_services and not force_rebuild)):
 907             return
 908         with self.lock:
 909             try:
 910                 keep_services = self.api_client.keep_services().accessible()
 911             except Exception:  # API server predates Keep services.
 912                 keep_services = self.api_client.keep_disks().list()
 913
 914             # Gateway services are only used when specified by UUID,
 915             # so there's nothing to gain by filtering them by
 916             # service_type.
 917             self._gateway_services = {ks['uuid']: ks for ks in
 918                                       keep_services.execute()['items']}
 919             if not self._gateway_services:
 920                 raise arvados.errors.NoKeepServersError()
 921
 922             # Precompute the base URI for each service.
 923             for r in self._gateway_services.values():
 924                 host = r['service_host']
 925                 if not host.startswith('[') and host.find(':') >= 0:
 926                     # IPv6 URIs must be formatted like http://[::1]:80/...
 927                     host = '[' + host + ']'
 928                 r['_service_root'] = "{}://{}:{:d}/".format(
 929                     'https' if r['service_ssl_flag'] else 'http',
 930                     host,
 931                     r['service_port'])
 932
 933             _logger.debug(str(self._gateway_services))
 934             self._keep_services = [
 935                 ks for ks in self._gateway_services.values()
 936                 if not ks.get('service_type', '').startswith('gateway:')]
 937             self._writable_services = [ks for ks in self._keep_services
 938                                        if not ks.get('read_only')]
 939
 940             # For disk type services, max_replicas_per_service is 1
 941             # It is unknown (unlimited) for other service types.
 942             if self._any_nondisk_services(self._writable_services):
 943                 self.max_replicas_per_service = None
 944             else:
 945                 self.max_replicas_per_service = 1
 946
 947     def _service_weight(self, data_hash, service_uuid):
 948         """Compute the weight of a Keep service endpoint for a data
 949         block with a known hash.
 950
 951         The weight is md5(h + u) where u is the last 15 characters of
 952         the service endpoint's UUID.
 953         """
 954         return hashlib.md5((data_hash + service_uuid[-15:]).encode()).hexdigest()
 955
 956     def weighted_service_roots(self, locator, force_rebuild=False, need_writable=False):
 957         """Return an array of Keep service endpoints, in the order in
 958         which they should be probed when reading or writing data with
 959         the given hash+hints.
 960         """
 961         self.build_services_list(force_rebuild)
 962
 963         sorted_roots = []
 964         # Use the services indicated by the given +K@... remote
 965         # service hints, if any are present and can be resolved to a
 966         # URI.
 967         for hint in locator.hints:
 968             if hint.startswith('K@'):
 969                 if len(hint) == 7:
 970                     sorted_roots.append(
 971                         "https://keep.{}.arvadosapi.com/".format(hint[2:]))
 972                 elif len(hint) == 29:
 973                     svc = self._gateway_services.get(hint[2:])
 974                     if svc:
 975                         sorted_roots.append(svc['_service_root'])
 976
 977         # Sort the available local services by weight (heaviest first)
 978         # for this locator, and return their service_roots (base URIs)
 979         # in that order.
 980         use_services = self._keep_services
 981         if need_writable:
 982             use_services = self._writable_services
 983         self.using_proxy = self._any_nondisk_services(use_services)
 984         sorted_roots.extend([
 985             svc['_service_root'] for svc in sorted(
 986                 use_services,
 987                 reverse=True,
 988                 key=lambda svc: self._service_weight(locator.md5sum, svc['uuid']))])
 989         _logger.debug("{}: {}".format(locator, sorted_roots))
 990         return sorted_roots
 991
 992     def map_new_services(self, roots_map, locator, force_rebuild, need_writable, headers):
 993         # roots_map is a dictionary, mapping Keep service root strings
 994         # to KeepService objects.  Poll for Keep services, and add any
 995         # new ones to roots_map.  Return the current list of local
 996         # root strings.
 997         headers.setdefault('Authorization', "OAuth2 %s" % (self.api_token,))
 998         local_roots = self.weighted_service_roots(locator, force_rebuild, need_writable)
 999         for root in local_roots:
1000             if root not in roots_map:
1001                 roots_map[root] = self.KeepService(
1002                     root, self._user_agent_pool,
1003                     upload_counter=self.upload_counter,
1004                     download_counter=self.download_counter,
1005                     headers=headers,
1006                     insecure=self.insecure)
1007         return local_roots
1008
1009     @staticmethod
1010     def _check_loop_result(result):
1011         # KeepClient RetryLoops should save results as a 2-tuple: the
1012         # actual result of the request, and the number of servers available
1013         # to receive the request this round.
1014         # This method returns True if there's a real result, False if
1015         # there are no more servers available, otherwise None.
1016         if isinstance(result, Exception):
1017             return None
1018         result, tried_server_count = result
1019         if (result is not None) and (result is not False):
1020             return True
1021         elif tried_server_count < 1:
1022             _logger.info("No more Keep services to try; giving up")
1023             return False
1024         else:
1025             return None
1026
1027     def get_from_cache(self, loc):
1028         """Fetch a block only if is in the cache, otherwise return None."""
1029         slot = self.block_cache.get(loc)
1030         if slot is not None and slot.ready.is_set():
1031             return slot.get()
1032         else:
1033             return None
1034
1035     def refresh_signature(self, loc):
1036         """Ask Keep to get the remote block and return its local signature"""
1037         now = datetime.datetime.utcnow().isoformat("T") + 'Z'
1038         return self.head(loc, headers={'X-Keep-Signature': 'local, {}'.format(now)})
1039
1040     @retry.retry_method
1041     def head(self, loc_s, **kwargs):
1042         return self._get_or_head(loc_s, method="HEAD", **kwargs)
1043
1044     @retry.retry_method
1045     def get(self, loc_s, **kwargs):
1046         return self._get_or_head(loc_s, method="GET", **kwargs)
1047
1048     def _get_or_head(self, loc_s, method="GET", num_retries=None, request_id=None, headers=None):
1049         """Get data from Keep.
1050
1051         This method fetches one or more blocks of data from Keep.  It
1052         sends a request each Keep service registered with the API
1053         server (or the proxy provided when this client was
1054         instantiated), then each service named in location hints, in
1055         sequence.  As soon as one service provides the data, it's
1056         returned.
1057
1058         Arguments:
1059         * loc_s: A string of one or more comma-separated locators to fetch.
1060           This method returns the concatenation of these blocks.
1061         * num_retries: The number of times to retry GET requests to
1062           *each* Keep server if it returns temporary failures, with
1063           exponential backoff.  Note that, in each loop, the method may try
1064           to fetch data from every available Keep service, along with any
1065           that are named in location hints in the locator.  The default value
1066           is set when the KeepClient is initialized.
1067         """
1068         if ',' in loc_s:
1069             return ''.join(self.get(x) for x in loc_s.split(','))
1070
1071         self.get_counter.add(1)
1072
1073         slot = None
1074         blob = None
1075         try:
1076             locator = KeepLocator(loc_s)
1077             if method == "GET":
1078                 slot, first = self.block_cache.reserve_cache(locator.md5sum)
1079                 if not first:
1080                     self.hits_counter.add(1)
1081                     blob = slot.get()
1082                     if blob is None:
1083                         raise arvados.errors.KeepReadError(
1084                             "failed to read {}".format(loc_s))
1085                     return blob
1086
1087             self.misses_counter.add(1)
1088
1089             if headers is None:
1090                 headers = {}
1091             headers['X-Request-Id'] = (request_id or
1092                                         (hasattr(self, 'api_client') and self.api_client.request_id) or
1093                                         arvados.util.new_request_id())
1094
1095             # If the locator has hints specifying a prefix (indicating a
1096             # remote keepproxy) or the UUID of a local gateway service,
1097             # read data from the indicated service(s) instead of the usual
1098             # list of local disk services.
1099             hint_roots = ['http://keep.{}.arvadosapi.com/'.format(hint[2:])
1100                           for hint in locator.hints if hint.startswith('K@') and len(hint) == 7]
1101             hint_roots.extend([self._gateway_services[hint[2:]]['_service_root']
1102                                for hint in locator.hints if (
1103                                        hint.startswith('K@') and
1104                                        len(hint) == 29 and
1105                                        self._gateway_services.get(hint[2:])
1106                                        )])
1107             # Map root URLs to their KeepService objects.
1108             roots_map = {
1109                 root: self.KeepService(root, self._user_agent_pool,
1110                                        upload_counter=self.upload_counter,
1111                                        download_counter=self.download_counter,
1112                                        headers=headers,
1113                                        insecure=self.insecure)
1114                 for root in hint_roots
1115             }
1116
1117             # See #3147 for a discussion of the loop implementation.  Highlights:
1118             # * Refresh the list of Keep services after each failure, in case
1119             #   it's being updated.
1120             # * Retry until we succeed, we're out of retries, or every available
1121             #   service has returned permanent failure.
1122             sorted_roots = []
1123             roots_map = {}
1124             loop = retry.RetryLoop(num_retries, self._check_loop_result,
1125                                    backoff_start=2)
1126             for tries_left in loop:
1127                 try:
1128                     sorted_roots = self.map_new_services(
1129                         roots_map, locator,
1130                         force_rebuild=(tries_left < num_retries),
1131                         need_writable=False,
1132                         headers=headers)
1133                 except Exception as error:
1134                     loop.save_result(error)
1135                     continue
1136
1137                 # Query KeepService objects that haven't returned
1138                 # permanent failure, in our specified shuffle order.
1139                 services_to_try = [roots_map[root]
1140                                    for root in sorted_roots
1141                                    if roots_map[root].usable()]
1142                 for keep_service in services_to_try:
1143                     blob = keep_service.get(locator, method=method, timeout=self.current_timeout(num_retries-tries_left))
1144                     if blob is not None:
1145                         break
1146                 loop.save_result((blob, len(services_to_try)))
1147
1148             # Always cache the result, then return it if we succeeded.
1149             if loop.success():
1150                 return blob
1151         finally:
1152             if slot is not None:
1153                 slot.set(blob)
1154                 self.block_cache.cap_cache()
1155
1156         # Q: Including 403 is necessary for the Keep tests to continue
1157         # passing, but maybe they should expect KeepReadError instead?
1158         not_founds = sum(1 for key in sorted_roots
1159                          if roots_map[key].last_result().get('status_code', None) in {403, 404, 410})
1160         service_errors = ((key, roots_map[key].last_result()['error'])
1161                           for key in sorted_roots)
1162         if not roots_map:
1163             raise arvados.errors.KeepReadError(
1164                 "failed to read {}: no Keep services available ({})".format(
1165                     loc_s, loop.last_result()))
1166         elif not_founds == len(sorted_roots):
1167             raise arvados.errors.NotFoundError(
1168                 "{} not found".format(loc_s), service_errors)
1169         else:
1170             raise arvados.errors.KeepReadError(
1171                 "failed to read {} after {}".format(loc_s, loop.attempts_str()), service_errors, label="service")
1172
1173     @retry.retry_method
1174     def put(self, data, copies=2, num_retries=None, request_id=None, classes=[]):
1175         """Save data in Keep.
1176
1177         This method will get a list of Keep services from the API server, and
1178         send the data to each one simultaneously in a new thread.  Once the
1179         uploads are finished, if enough copies are saved, this method returns
1180         the most recent HTTP response body.  If requests fail to upload
1181         enough copies, this method raises KeepWriteError.
1182
1183         Arguments:
1184         * data: The string of data to upload.
1185         * copies: The number of copies that the user requires be saved.
1186           Default 2.
1187         * num_retries: The number of times to retry PUT requests to
1188           *each* Keep server if it returns temporary failures, with
1189           exponential backoff.  The default value is set when the
1190           KeepClient is initialized.
1191         * classes: An optional list of storage class names where copies should
1192           be written.
1193         """
1194
1195         if not isinstance(data, bytes):
1196             data = data.encode()
1197
1198         self.put_counter.add(1)
1199
1200         data_hash = hashlib.md5(data).hexdigest()
1201         loc_s = data_hash + '+' + str(len(data))
1202         if copies < 1:
1203             return loc_s
1204         locator = KeepLocator(loc_s)
1205
1206         headers = {
1207             'X-Request-Id': (request_id or
1208                              (hasattr(self, 'api_client') and self.api_client.request_id) or
1209                              arvados.util.new_request_id()),
1210             'X-Keep-Desired-Replicas': str(copies),
1211         }
1212         roots_map = {}
1213         loop = retry.RetryLoop(num_retries, self._check_loop_result,
1214                                backoff_start=2)
1215         done_copies = 0
1216         done_classes = []
1217         for tries_left in loop:
1218             try:
1219                 sorted_roots = self.map_new_services(
1220                     roots_map, locator,
1221                     force_rebuild=(tries_left < num_retries),
1222                     need_writable=True,
1223                     headers=headers)
1224             except Exception as error:
1225                 loop.save_result(error)
1226                 continue
1227
1228             pending_classes = []
1229             if done_classes is not None:
1230                 pending_classes = list(set(classes) - set(done_classes))
1231             writer_pool = KeepClient.KeepWriterThreadPool(data=data,
1232                                                         data_hash=data_hash,
1233                                                         copies=copies - done_copies,
1234                                                         max_service_replicas=self.max_replicas_per_service,
1235                                                         timeout=self.current_timeout(num_retries - tries_left),
1236                                                         classes=pending_classes)
1237             for service_root, ks in [(root, roots_map[root])
1238                                      for root in sorted_roots]:
1239                 if ks.finished():
1240                     continue
1241                 writer_pool.add_task(ks, service_root)
1242             writer_pool.join()
1243             pool_copies, pool_classes = writer_pool.done()
1244             done_copies += pool_copies
1245             if (done_classes is not None) and (pool_classes is not None):
1246                 done_classes += pool_classes
1247                 loop.save_result(
1248                     (done_copies >= copies and set(done_classes) == set(classes),
1249                     writer_pool.total_task_nr))
1250             else:
1251                 # Old keepstore contacted without storage classes support:
1252                 # success is determined only by successful copies.
1253                 #
1254                 # Disable storage classes tracking from this point forward.
1255                 _logger.warning("X-Keep-Storage-Classes header not supported by the cluster")
1256                 done_classes = None
1257                 loop.save_result(
1258                     (done_copies >= copies, writer_pool.total_task_nr))
1259
1260         if loop.success():
1261             return writer_pool.response()
1262         if not roots_map:
1263             raise arvados.errors.KeepWriteError(
1264                 "failed to write {}: no Keep services available ({})".format(
1265                     data_hash, loop.last_result()))
1266         else:
1267             service_errors = ((key, roots_map[key].last_result()['error'])
1268                               for key in sorted_roots
1269                               if roots_map[key].last_result()['error'])
1270             raise arvados.errors.KeepWriteError(
1271                 "failed to write {} after {} (wanted {} copies but wrote {})".format(
1272                     data_hash, loop.attempts_str(), (copies, classes), writer_pool.done()), service_errors, label="service")
1273
1274     def local_store_put(self, data, copies=1, num_retries=None):
1275         """A stub for put().
1276
1277         This method is used in place of the real put() method when
1278         using local storage (see constructor's local_store argument).
1279
1280         copies and num_retries arguments are ignored: they are here
1281         only for the sake of offering the same call signature as
1282         put().
1283
1284         Data stored this way can be retrieved via local_store_get().
1285         """
1286         md5 = hashlib.md5(data).hexdigest()
1287         locator = '%s+%d' % (md5, len(data))
1288         with open(os.path.join(self.local_store, md5 + '.tmp'), 'wb') as f:
1289             f.write(data)
1290         os.rename(os.path.join(self.local_store, md5 + '.tmp'),
1291                   os.path.join(self.local_store, md5))
1292         return locator
1293
1294     def local_store_get(self, loc_s, num_retries=None):
1295         """Companion to local_store_put()."""
1296         try:
1297             locator = KeepLocator(loc_s)
1298         except ValueError:
1299             raise arvados.errors.NotFoundError(
1300                 "Invalid data locator: '%s'" % loc_s)
1301         if locator.md5sum == config.EMPTY_BLOCK_LOCATOR.split('+')[0]:
1302             return b''
1303         with open(os.path.join(self.local_store, locator.md5sum), 'rb') as f:
1304             return f.read()
1305
1306     def local_store_head(self, loc_s, num_retries=None):
1307         """Companion to local_store_put()."""
1308         try:
1309             locator = KeepLocator(loc_s)
1310         except ValueError:
1311             raise arvados.errors.NotFoundError(
1312                 "Invalid data locator: '%s'" % loc_s)
1313         if locator.md5sum == config.EMPTY_BLOCK_LOCATOR.split('+')[0]:
1314             return True
1315         if os.path.exists(os.path.join(self.local_store, locator.md5sum)):
1316             return True
1317
1318     def is_cached(self, locator):
1319         return self.block_cache.reserve_cache(expect_hash)