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