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                     if method == "HEAD":
 379                         curl.setopt(pycurl.NOBODY, True)
 380                     self._setcurltimeouts(curl, timeout)
 381
 382                     try:
 383                         curl.perform()
 384                     except Exception as e:
 385                         raise arvados.errors.HttpError(0, str(e))
 386                     finally:
 387                         if self._socket:
 388                             self._socket.close()
 389                             self._socket = None
 390                     self._result = {
 391                         'status_code': curl.getinfo(pycurl.RESPONSE_CODE),
 392                         'body': response_body.getvalue(),
 393                         'headers': self._headers,
 394                         'error': False,
 395                     }
 396
 397                 ok = retry.check_http_response_success(self._result['status_code'])
 398                 if not ok:
 399                     self._result['error'] = arvados.errors.HttpError(
 400                         self._result['status_code'],
 401                         self._headers.get('x-status-line', 'Error'))
 402             except self.HTTP_ERRORS as e:
 403                 self._result = {
 404                     'error': e,
 405                 }
 406             self._usable = ok != False
 407             if self._result.get('status_code', None):
 408                 # The client worked well enough to get an HTTP status
 409                 # code, so presumably any problems are just on the
 410                 # server side and it's OK to reuse the client.
 411                 self._put_user_agent(curl)
 412             else:
 413                 # Don't return this client to the pool, in case it's
 414                 # broken.
 415                 curl.close()
 416             if not ok:
 417                 _logger.debug("Request fail: GET %s => %s: %s",
 418                               url, type(self._result['error']), str(self._result['error']))
 419                 return None
 420             if method == "HEAD":
 421                 _logger.info("HEAD %s: %s bytes",
 422                          self._result['status_code'],
 423                          self._result.get('content-length'))
 424                 if self._result['headers'].get('x-keep-locator'):
 425                     # This is a response to a remote block copy request, return
 426                     # the local copy block locator.
 427                     return self._result['headers'].get('x-keep-locator')
 428                 return True
 429
 430             _logger.info("GET %s: %s bytes in %s msec (%.3f MiB/sec)",
 431                          self._result['status_code'],
 432                          len(self._result['body']),
 433                          t.msecs,
 434                          1.0*len(self._result['body'])/2**20/t.secs if t.secs > 0 else 0)
 435
 436             if self.download_counter:
 437                 self.download_counter.add(len(self._result['body']))
 438             resp_md5 = hashlib.md5(self._result['body']).hexdigest()
 439             if resp_md5 != locator.md5sum:
 440                 _logger.warning("Checksum fail: md5(%s) = %s",
 441                                 url, resp_md5)
 442                 self._result['error'] = arvados.errors.HttpError(
 443                     0, 'Checksum fail')
 444                 return None
 445             return self._result['body']
 446
 447         def put(self, hash_s, body, timeout=None):
 448             url = self.root + hash_s
 449             _logger.debug("Request: PUT %s", url)
 450             curl = self._get_user_agent()
 451             ok = None
 452             try:
 453                 with timer.Timer() as t:
 454                     self._headers = {}
 455                     body_reader = BytesIO(body)
 456                     response_body = BytesIO()
 457                     curl.setopt(pycurl.NOSIGNAL, 1)
 458                     curl.setopt(pycurl.OPENSOCKETFUNCTION,
 459                                 lambda *args, **kwargs: self._socket_open(*args, **kwargs))
 460                     curl.setopt(pycurl.URL, url.encode('utf-8'))
 461                     # Using UPLOAD tells cURL to wait for a "go ahead" from the
 462                     # Keep server (in the form of a HTTP/1.1 "100 Continue"
 463                     # response) instead of sending the request body immediately.
 464                     # This allows the server to reject the request if the request
 465                     # is invalid or the server is read-only, without waiting for
 466                     # the client to send the entire block.
 467                     curl.setopt(pycurl.UPLOAD, True)
 468                     curl.setopt(pycurl.INFILESIZE, len(body))
 469                     curl.setopt(pycurl.READFUNCTION, body_reader.read)
 470                     curl.setopt(pycurl.HTTPHEADER, [
 471                         '{}: {}'.format(k,v) for k,v in self.put_headers.items()])
 472                     curl.setopt(pycurl.WRITEFUNCTION, response_body.write)
 473                     curl.setopt(pycurl.HEADERFUNCTION, self._headerfunction)
 474                     if self.insecure:
 475                         curl.setopt(pycurl.SSL_VERIFYPEER, 0)
 476                     self._setcurltimeouts(curl, timeout)
 477                     try:
 478                         curl.perform()
 479                     except Exception as e:
 480                         raise arvados.errors.HttpError(0, str(e))
 481                     finally:
 482                         if self._socket:
 483                             self._socket.close()
 484                             self._socket = None
 485                     self._result = {
 486                         'status_code': curl.getinfo(pycurl.RESPONSE_CODE),
 487                         'body': response_body.getvalue().decode('utf-8'),
 488                         'headers': self._headers,
 489                         'error': False,
 490                     }
 491                 ok = retry.check_http_response_success(self._result['status_code'])
 492                 if not ok:
 493                     self._result['error'] = arvados.errors.HttpError(
 494                         self._result['status_code'],
 495                         self._headers.get('x-status-line', 'Error'))
 496             except self.HTTP_ERRORS as e:
 497                 self._result = {
 498                     'error': e,
 499                 }
 500             self._usable = ok != False # still usable if ok is True or None
 501             if self._result.get('status_code', None):
 502                 # Client is functional. See comment in get().
 503                 self._put_user_agent(curl)
 504             else:
 505                 curl.close()
 506             if not ok:
 507                 _logger.debug("Request fail: PUT %s => %s: %s",
 508                               url, type(self._result['error']), str(self._result['error']))
 509                 return False
 510             _logger.info("PUT %s: %s bytes in %s msec (%.3f MiB/sec)",
 511                          self._result['status_code'],
 512                          len(body),
 513                          t.msecs,
 514                          1.0*len(body)/2**20/t.secs if t.secs > 0 else 0)
 515             if self.upload_counter:
 516                 self.upload_counter.add(len(body))
 517             return True
 518
 519         def _setcurltimeouts(self, curl, timeouts):
 520             if not timeouts:
 521                 return
 522             elif isinstance(timeouts, tuple):
 523                 if len(timeouts) == 2:
 524                     conn_t, xfer_t = timeouts
 525                     bandwidth_bps = KeepClient.DEFAULT_TIMEOUT[2]
 526                 else:
 527                     conn_t, xfer_t, bandwidth_bps = timeouts
 528             else:
 529                 conn_t, xfer_t = (timeouts, timeouts)
 530                 bandwidth_bps = KeepClient.DEFAULT_TIMEOUT[2]
 531             curl.setopt(pycurl.CONNECTTIMEOUT_MS, int(conn_t*1000))
 532             curl.setopt(pycurl.LOW_SPEED_TIME, int(math.ceil(xfer_t)))
 533             curl.setopt(pycurl.LOW_SPEED_LIMIT, int(math.ceil(bandwidth_bps)))
 534
 535         def _headerfunction(self, header_line):
 536             if isinstance(header_line, bytes):
 537                 header_line = header_line.decode('iso-8859-1')
 538             if ':' in header_line:
 539                 name, value = header_line.split(':', 1)
 540                 name = name.strip().lower()
 541                 value = value.strip()
 542             elif self._headers:
 543                 name = self._lastheadername
 544                 value = self._headers[name] + ' ' + header_line.strip()
 545             elif header_line.startswith('HTTP/'):
 546                 name = 'x-status-line'
 547                 value = header_line
 548             else:
 549                 _logger.error("Unexpected header line: %s", header_line)
 550                 return
 551             self._lastheadername = name
 552             self._headers[name] = value
 553             # Returning None implies all bytes were written
 554
 555
 556     class KeepWriterQueue(queue.Queue):
 557         def __init__(self, copies):
 558             queue.Queue.__init__(self) # Old-style superclass
 559             self.wanted_copies = copies
 560             self.successful_copies = 0
 561             self.response = None
 562             self.successful_copies_lock = threading.Lock()
 563             self.pending_tries = copies
 564             self.pending_tries_notification = threading.Condition()
 565
 566         def write_success(self, response, replicas_nr):
 567             with self.successful_copies_lock:
 568                 self.successful_copies += replicas_nr
 569                 self.response = response
 570             with self.pending_tries_notification:
 571                 self.pending_tries_notification.notify_all()
 572
 573         def write_fail(self, ks):
 574             with self.pending_tries_notification:
 575                 self.pending_tries += 1
 576                 self.pending_tries_notification.notify()
 577
 578         def pending_copies(self):
 579             with self.successful_copies_lock:
 580                 return self.wanted_copies - self.successful_copies
 581
 582         def get_next_task(self):
 583             with self.pending_tries_notification:
 584                 while True:
 585                     if self.pending_copies() < 1:
 586                         # This notify_all() is unnecessary --
 587                         # write_success() already called notify_all()
 588                         # when pending<1 became true, so it's not
 589                         # possible for any other thread to be in
 590                         # wait() now -- but it's cheap insurance
 591                         # against deadlock so we do it anyway:
 592                         self.pending_tries_notification.notify_all()
 593                         # Drain the queue and then raise Queue.Empty
 594                         while True:
 595                             self.get_nowait()
 596                             self.task_done()
 597                     elif self.pending_tries > 0:
 598                         service, service_root = self.get_nowait()
 599                         if service.finished():
 600                             self.task_done()
 601                             continue
 602                         self.pending_tries -= 1
 603                         return service, service_root
 604                     elif self.empty():
 605                         self.pending_tries_notification.notify_all()
 606                         raise queue.Empty
 607                     else:
 608                         self.pending_tries_notification.wait()
 609
 610
 611     class KeepWriterThreadPool(object):
 612         def __init__(self, data, data_hash, copies, max_service_replicas, timeout=None):
 613             self.total_task_nr = 0
 614             self.wanted_copies = copies
 615             if (not max_service_replicas) or (max_service_replicas >= copies):
 616                 num_threads = 1
 617             else:
 618                 num_threads = int(math.ceil(1.0*copies/max_service_replicas))
 619             _logger.debug("Pool max threads is %d", num_threads)
 620             self.workers = []
 621             self.queue = KeepClient.KeepWriterQueue(copies)
 622             # Create workers
 623             for _ in range(num_threads):
 624                 w = KeepClient.KeepWriterThread(self.queue, data, data_hash, timeout)
 625                 self.workers.append(w)
 626
 627         def add_task(self, ks, service_root):
 628             self.queue.put((ks, service_root))
 629             self.total_task_nr += 1
 630
 631         def done(self):
 632             return self.queue.successful_copies
 633
 634         def join(self):
 635             # Start workers
 636             for worker in self.workers:
 637                 worker.start()
 638             # Wait for finished work
 639             self.queue.join()
 640
 641         def response(self):
 642             return self.queue.response
 643
 644
 645     class KeepWriterThread(threading.Thread):
 646         TaskFailed = RuntimeError()
 647
 648         def __init__(self, queue, data, data_hash, timeout=None):
 649             super(KeepClient.KeepWriterThread, self).__init__()
 650             self.timeout = timeout
 651             self.queue = queue
 652             self.data = data
 653             self.data_hash = data_hash
 654             self.daemon = True
 655
 656         def run(self):
 657             while True:
 658                 try:
 659                     service, service_root = self.queue.get_next_task()
 660                 except queue.Empty:
 661                     return
 662                 try:
 663                     locator, copies = self.do_task(service, service_root)
 664                 except Exception as e:
 665                     if e is not self.TaskFailed:
 666                         _logger.exception("Exception in KeepWriterThread")
 667                     self.queue.write_fail(service)
 668                 else:
 669                     self.queue.write_success(locator, copies)
 670                 finally:
 671                     self.queue.task_done()
 672
 673         def do_task(self, service, service_root):
 674             success = bool(service.put(self.data_hash,
 675                                         self.data,
 676                                         timeout=self.timeout))
 677             result = service.last_result()
 678
 679             if not success:
 680                 if result.get('status_code', None):
 681                     _logger.debug("Request fail: PUT %s => %s %s",
 682                                   self.data_hash,
 683                                   result['status_code'],
 684                                   result['body'])
 685                 raise self.TaskFailed
 686
 687             _logger.debug("KeepWriterThread %s succeeded %s+%i %s",
 688                           str(threading.current_thread()),
 689                           self.data_hash,
 690                           len(self.data),
 691                           service_root)
 692             try:
 693                 replicas_stored = int(result['headers']['x-keep-replicas-stored'])
 694             except (KeyError, ValueError):
 695                 replicas_stored = 1
 696
 697             return result['body'].strip(), replicas_stored
 698
 699
 700     def __init__(self, api_client=None, proxy=None,
 701                  timeout=DEFAULT_TIMEOUT, proxy_timeout=DEFAULT_PROXY_TIMEOUT,
 702                  api_token=None, local_store=None, block_cache=None,
 703                  num_retries=0, session=None):
 704         """Initialize a new KeepClient.
 705
 706         Arguments:
 707         :api_client:
 708           The API client to use to find Keep services.  If not
 709           provided, KeepClient will build one from available Arvados
 710           configuration.
 711
 712         :proxy:
 713           If specified, this KeepClient will send requests to this Keep
 714           proxy.  Otherwise, KeepClient will fall back to the setting of the
 715           ARVADOS_KEEP_SERVICES or ARVADOS_KEEP_PROXY configuration settings.
 716           If you want to KeepClient does not use a proxy, pass in an empty
 717           string.
 718
 719         :timeout:
 720           The initial timeout (in seconds) for HTTP requests to Keep
 721           non-proxy servers.  A tuple of three floats is interpreted as
 722           (connection_timeout, read_timeout, minimum_bandwidth). A connection
 723           will be aborted if the average traffic rate falls below
 724           minimum_bandwidth bytes per second over an interval of read_timeout
 725           seconds. Because timeouts are often a result of transient server
 726           load, the actual connection timeout will be increased by a factor
 727           of two on each retry.
 728           Default: (2, 256, 32768).
 729
 730         :proxy_timeout:
 731           The initial timeout (in seconds) for HTTP requests to
 732           Keep proxies. A tuple of three floats is interpreted as
 733           (connection_timeout, read_timeout, minimum_bandwidth). The behavior
 734           described above for adjusting connection timeouts on retry also
 735           applies.
 736           Default: (20, 256, 32768).
 737
 738         :api_token:
 739           If you're not using an API client, but only talking
 740           directly to a Keep proxy, this parameter specifies an API token
 741           to authenticate Keep requests.  It is an error to specify both
 742           api_client and api_token.  If you specify neither, KeepClient
 743           will use one available from the Arvados configuration.
 744
 745         :local_store:
 746           If specified, this KeepClient will bypass Keep
 747           services, and save data to the named directory.  If unspecified,
 748           KeepClient will fall back to the setting of the $KEEP_LOCAL_STORE
 749           environment variable.  If you want to ensure KeepClient does not
 750           use local storage, pass in an empty string.  This is primarily
 751           intended to mock a server for testing.
 752
 753         :num_retries:
 754           The default number of times to retry failed requests.
 755           This will be used as the default num_retries value when get() and
 756           put() are called.  Default 0.
 757         """
 758         self.lock = threading.Lock()
 759         if proxy is None:
 760             if config.get('ARVADOS_KEEP_SERVICES'):
 761                 proxy = config.get('ARVADOS_KEEP_SERVICES')
 762             else:
 763                 proxy = config.get('ARVADOS_KEEP_PROXY')
 764         if api_token is None:
 765             if api_client is None:
 766                 api_token = config.get('ARVADOS_API_TOKEN')
 767             else:
 768                 api_token = api_client.api_token
 769         elif api_client is not None:
 770             raise ValueError(
 771                 "can't build KeepClient with both API client and token")
 772         if local_store is None:
 773             local_store = os.environ.get('KEEP_LOCAL_STORE')
 774
 775         if api_client is None:
 776             self.insecure = config.flag_is_true('ARVADOS_API_HOST_INSECURE')
 777         else:
 778             self.insecure = api_client.insecure
 779
 780         self.block_cache = block_cache if block_cache else KeepBlockCache()
 781         self.timeout = timeout
 782         self.proxy_timeout = proxy_timeout
 783         self._user_agent_pool = queue.LifoQueue()
 784         self.upload_counter = Counter()
 785         self.download_counter = Counter()
 786         self.put_counter = Counter()
 787         self.get_counter = Counter()
 788         self.hits_counter = Counter()
 789         self.misses_counter = Counter()
 790
 791         if local_store:
 792             self.local_store = local_store
 793             self.get = self.local_store_get
 794             self.put = self.local_store_put
 795         else:
 796             self.num_retries = num_retries
 797             self.max_replicas_per_service = None
 798             if proxy:
 799                 proxy_uris = proxy.split()
 800                 for i in range(len(proxy_uris)):
 801                     if not proxy_uris[i].endswith('/'):
 802                         proxy_uris[i] += '/'
 803                     # URL validation
 804                     url = urllib.parse.urlparse(proxy_uris[i])
 805                     if not (url.scheme and url.netloc):
 806                         raise arvados.errors.ArgumentError("Invalid proxy URI: {}".format(proxy_uris[i]))
 807                 self.api_token = api_token
 808                 self._gateway_services = {}
 809                 self._keep_services = [{
 810                     'uuid': "00000-bi6l4-%015d" % idx,
 811                     'service_type': 'proxy',
 812                     '_service_root': uri,
 813                     } for idx, uri in enumerate(proxy_uris)]
 814                 self._writable_services = self._keep_services
 815                 self.using_proxy = True
 816                 self._static_services_list = True
 817             else:
 818                 # It's important to avoid instantiating an API client
 819                 # unless we actually need one, for testing's sake.
 820                 if api_client is None:
 821                     api_client = arvados.api('v1')
 822                 self.api_client = api_client
 823                 self.api_token = api_client.api_token
 824                 self._gateway_services = {}
 825                 self._keep_services = None
 826                 self._writable_services = None
 827                 self.using_proxy = None
 828                 self._static_services_list = False
 829
 830     def current_timeout(self, attempt_number):
 831         """Return the appropriate timeout to use for this client.
 832
 833         The proxy timeout setting if the backend service is currently a proxy,
 834         the regular timeout setting otherwise.  The `attempt_number` indicates
 835         how many times the operation has been tried already (starting from 0
 836         for the first try), and scales the connection timeout portion of the
 837         return value accordingly.
 838
 839         """
 840         # TODO(twp): the timeout should be a property of a
 841         # KeepService, not a KeepClient. See #4488.
 842         t = self.proxy_timeout if self.using_proxy else self.timeout
 843         if len(t) == 2:
 844             return (t[0] * (1 << attempt_number), t[1])
 845         else:
 846             return (t[0] * (1 << attempt_number), t[1], t[2])
 847     def _any_nondisk_services(self, service_list):
 848         return any(ks.get('service_type', 'disk') != 'disk'
 849                    for ks in service_list)
 850
 851     def build_services_list(self, force_rebuild=False):
 852         if (self._static_services_list or
 853               (self._keep_services and not force_rebuild)):
 854             return
 855         with self.lock:
 856             try:
 857                 keep_services = self.api_client.keep_services().accessible()
 858             except Exception:  # API server predates Keep services.
 859                 keep_services = self.api_client.keep_disks().list()
 860
 861             # Gateway services are only used when specified by UUID,
 862             # so there's nothing to gain by filtering them by
 863             # service_type.
 864             self._gateway_services = {ks['uuid']: ks for ks in
 865                                       keep_services.execute()['items']}
 866             if not self._gateway_services:
 867                 raise arvados.errors.NoKeepServersError()
 868
 869             # Precompute the base URI for each service.
 870             for r in self._gateway_services.values():
 871                 host = r['service_host']
 872                 if not host.startswith('[') and host.find(':') >= 0:
 873                     # IPv6 URIs must be formatted like http://[::1]:80/...
 874                     host = '[' + host + ']'
 875                 r['_service_root'] = "{}://{}:{:d}/".format(
 876                     'https' if r['service_ssl_flag'] else 'http',
 877                     host,
 878                     r['service_port'])
 879
 880             _logger.debug(str(self._gateway_services))
 881             self._keep_services = [
 882                 ks for ks in self._gateway_services.values()
 883                 if not ks.get('service_type', '').startswith('gateway:')]
 884             self._writable_services = [ks for ks in self._keep_services
 885                                        if not ks.get('read_only')]
 886
 887             # For disk type services, max_replicas_per_service is 1
 888             # It is unknown (unlimited) for other service types.
 889             if self._any_nondisk_services(self._writable_services):
 890                 self.max_replicas_per_service = None
 891             else:
 892                 self.max_replicas_per_service = 1
 893
 894     def _service_weight(self, data_hash, service_uuid):
 895         """Compute the weight of a Keep service endpoint for a data
 896         block with a known hash.
 897
 898         The weight is md5(h + u) where u is the last 15 characters of
 899         the service endpoint's UUID.
 900         """
 901         return hashlib.md5((data_hash + service_uuid[-15:]).encode()).hexdigest()
 902
 903     def weighted_service_roots(self, locator, force_rebuild=False, need_writable=False):
 904         """Return an array of Keep service endpoints, in the order in
 905         which they should be probed when reading or writing data with
 906         the given hash+hints.
 907         """
 908         self.build_services_list(force_rebuild)
 909
 910         sorted_roots = []
 911         # Use the services indicated by the given +K@... remote
 912         # service hints, if any are present and can be resolved to a
 913         # URI.
 914         for hint in locator.hints:
 915             if hint.startswith('K@'):
 916                 if len(hint) == 7:
 917                     sorted_roots.append(
 918                         "https://keep.{}.arvadosapi.com/".format(hint[2:]))
 919                 elif len(hint) == 29:
 920                     svc = self._gateway_services.get(hint[2:])
 921                     if svc:
 922                         sorted_roots.append(svc['_service_root'])
 923
 924         # Sort the available local services by weight (heaviest first)
 925         # for this locator, and return their service_roots (base URIs)
 926         # in that order.
 927         use_services = self._keep_services
 928         if need_writable:
 929             use_services = self._writable_services
 930         self.using_proxy = self._any_nondisk_services(use_services)
 931         sorted_roots.extend([
 932             svc['_service_root'] for svc in sorted(
 933                 use_services,
 934                 reverse=True,
 935                 key=lambda svc: self._service_weight(locator.md5sum, svc['uuid']))])
 936         _logger.debug("{}: {}".format(locator, sorted_roots))
 937         return sorted_roots
 938
 939     def map_new_services(self, roots_map, locator, force_rebuild, need_writable, headers):
 940         # roots_map is a dictionary, mapping Keep service root strings
 941         # to KeepService objects.  Poll for Keep services, and add any
 942         # new ones to roots_map.  Return the current list of local
 943         # root strings.
 944         headers.setdefault('Authorization', "OAuth2 %s" % (self.api_token,))
 945         local_roots = self.weighted_service_roots(locator, force_rebuild, need_writable)
 946         for root in local_roots:
 947             if root not in roots_map:
 948                 roots_map[root] = self.KeepService(
 949                     root, self._user_agent_pool,
 950                     upload_counter=self.upload_counter,
 951                     download_counter=self.download_counter,
 952                     headers=headers,
 953                     insecure=self.insecure)
 954         return local_roots
 955
 956     @staticmethod
 957     def _check_loop_result(result):
 958         # KeepClient RetryLoops should save results as a 2-tuple: the
 959         # actual result of the request, and the number of servers available
 960         # to receive the request this round.
 961         # This method returns True if there's a real result, False if
 962         # there are no more servers available, otherwise None.
 963         if isinstance(result, Exception):
 964             return None
 965         result, tried_server_count = result
 966         if (result is not None) and (result is not False):
 967             return True
 968         elif tried_server_count < 1:
 969             _logger.info("No more Keep services to try; giving up")
 970             return False
 971         else:
 972             return None
 973
 974     def get_from_cache(self, loc):
 975         """Fetch a block only if is in the cache, otherwise return None."""
 976         slot = self.block_cache.get(loc)
 977         if slot is not None and slot.ready.is_set():
 978             return slot.get()
 979         else:
 980             return None
 981
 982     def refresh_signature(self, loc):
 983         """Ask Keep to get the remote block and return its local signature"""
 984         now = datetime.datetime.utcnow().isoformat("T") + 'Z'
 985         return self.head(loc, headers={'X-Keep-Signature': 'local, {}'.format(now)})
 986
 987     @retry.retry_method
 988     def head(self, loc_s, **kwargs):
 989         return self._get_or_head(loc_s, method="HEAD", **kwargs)
 990
 991     @retry.retry_method
 992     def get(self, loc_s, **kwargs):
 993         return self._get_or_head(loc_s, method="GET", **kwargs)
 994
 995     def _get_or_head(self, loc_s, method="GET", num_retries=None, request_id=None, headers=None):
 996         """Get data from Keep.
 997
 998         This method fetches one or more blocks of data from Keep.  It
 999         sends a request each Keep service registered with the API
1000         server (or the proxy provided when this client was
1001         instantiated), then each service named in location hints, in
1002         sequence.  As soon as one service provides the data, it's
1003         returned.
1004
1005         Arguments:
1006         * loc_s: A string of one or more comma-separated locators to fetch.
1007           This method returns the concatenation of these blocks.
1008         * num_retries: The number of times to retry GET requests to
1009           *each* Keep server if it returns temporary failures, with
1010           exponential backoff.  Note that, in each loop, the method may try
1011           to fetch data from every available Keep service, along with any
1012           that are named in location hints in the locator.  The default value
1013           is set when the KeepClient is initialized.
1014         """
1015         if ',' in loc_s:
1016             return ''.join(self.get(x) for x in loc_s.split(','))
1017
1018         self.get_counter.add(1)
1019
1020         slot = None
1021         blob = None
1022         try:
1023             locator = KeepLocator(loc_s)
1024             if method == "GET":
1025                 slot, first = self.block_cache.reserve_cache(locator.md5sum)
1026                 if not first:
1027                     self.hits_counter.add(1)
1028                     blob = slot.get()
1029                     if blob is None:
1030                         raise arvados.errors.KeepReadError(
1031                             "failed to read {}".format(loc_s))
1032                     return blob
1033
1034             self.misses_counter.add(1)
1035
1036             if headers is None:
1037                 headers = {}
1038             headers['X-Request-Id'] = (request_id or
1039                                         (hasattr(self, 'api_client') and self.api_client.request_id) or
1040                                         arvados.util.new_request_id())
1041
1042             # If the locator has hints specifying a prefix (indicating a
1043             # remote keepproxy) or the UUID of a local gateway service,
1044             # read data from the indicated service(s) instead of the usual
1045             # list of local disk services.
1046             hint_roots = ['http://keep.{}.arvadosapi.com/'.format(hint[2:])
1047                           for hint in locator.hints if hint.startswith('K@') and len(hint) == 7]
1048             hint_roots.extend([self._gateway_services[hint[2:]]['_service_root']
1049                                for hint in locator.hints if (
1050                                        hint.startswith('K@') and
1051                                        len(hint) == 29 and
1052                                        self._gateway_services.get(hint[2:])
1053                                        )])
1054             # Map root URLs to their KeepService objects.
1055             roots_map = {
1056                 root: self.KeepService(root, self._user_agent_pool,
1057                                        upload_counter=self.upload_counter,
1058                                        download_counter=self.download_counter,
1059                                        headers=headers,
1060                                        insecure=self.insecure)
1061                 for root in hint_roots
1062             }
1063
1064             # See #3147 for a discussion of the loop implementation.  Highlights:
1065             # * Refresh the list of Keep services after each failure, in case
1066             #   it's being updated.
1067             # * Retry until we succeed, we're out of retries, or every available
1068             #   service has returned permanent failure.
1069             sorted_roots = []
1070             roots_map = {}
1071             loop = retry.RetryLoop(num_retries, self._check_loop_result,
1072                                    backoff_start=2)
1073             for tries_left in loop:
1074                 try:
1075                     sorted_roots = self.map_new_services(
1076                         roots_map, locator,
1077                         force_rebuild=(tries_left < num_retries),
1078                         need_writable=False,
1079                         headers=headers)
1080                 except Exception as error:
1081                     loop.save_result(error)
1082                     continue
1083
1084                 # Query KeepService objects that haven't returned
1085                 # permanent failure, in our specified shuffle order.
1086                 services_to_try = [roots_map[root]
1087                                    for root in sorted_roots
1088                                    if roots_map[root].usable()]
1089                 for keep_service in services_to_try:
1090                     blob = keep_service.get(locator, method=method, timeout=self.current_timeout(num_retries-tries_left))
1091                     if blob is not None:
1092                         break
1093                 loop.save_result((blob, len(services_to_try)))
1094
1095             # Always cache the result, then return it if we succeeded.
1096             if loop.success():
1097                 if method == "HEAD":
1098                     return blob or True
1099                 else:
1100                     return blob
1101         finally:
1102             if slot is not None:
1103                 slot.set(blob)
1104                 self.block_cache.cap_cache()
1105
1106         # Q: Including 403 is necessary for the Keep tests to continue
1107         # passing, but maybe they should expect KeepReadError instead?
1108         not_founds = sum(1 for key in sorted_roots
1109                          if roots_map[key].last_result().get('status_code', None) in {403, 404, 410})
1110         service_errors = ((key, roots_map[key].last_result()['error'])
1111                           for key in sorted_roots)
1112         if not roots_map:
1113             raise arvados.errors.KeepReadError(
1114                 "failed to read {}: no Keep services available ({})".format(
1115                     loc_s, loop.last_result()))
1116         elif not_founds == len(sorted_roots):
1117             raise arvados.errors.NotFoundError(
1118                 "{} not found".format(loc_s), service_errors)
1119         else:
1120             raise arvados.errors.KeepReadError(
1121                 "failed to read {}".format(loc_s), service_errors, label="service")
1122
1123     @retry.retry_method
1124     def put(self, data, copies=2, num_retries=None, request_id=None):
1125         """Save data in Keep.
1126
1127         This method will get a list of Keep services from the API server, and
1128         send the data to each one simultaneously in a new thread.  Once the
1129         uploads are finished, if enough copies are saved, this method returns
1130         the most recent HTTP response body.  If requests fail to upload
1131         enough copies, this method raises KeepWriteError.
1132
1133         Arguments:
1134         * data: The string of data to upload.
1135         * copies: The number of copies that the user requires be saved.
1136           Default 2.
1137         * num_retries: The number of times to retry PUT requests to
1138           *each* Keep server if it returns temporary failures, with
1139           exponential backoff.  The default value is set when the
1140           KeepClient is initialized.
1141         """
1142
1143         if not isinstance(data, bytes):
1144             data = data.encode()
1145
1146         self.put_counter.add(1)
1147
1148         data_hash = hashlib.md5(data).hexdigest()
1149         loc_s = data_hash + '+' + str(len(data))
1150         if copies < 1:
1151             return loc_s
1152         locator = KeepLocator(loc_s)
1153
1154         headers = {
1155             'X-Request-Id': (request_id or
1156                              (hasattr(self, 'api_client') and self.api_client.request_id) or
1157                              arvados.util.new_request_id()),
1158             'X-Keep-Desired-Replicas': str(copies),
1159         }
1160         roots_map = {}
1161         loop = retry.RetryLoop(num_retries, self._check_loop_result,
1162                                backoff_start=2)
1163         done = 0
1164         for tries_left in loop:
1165             try:
1166                 sorted_roots = self.map_new_services(
1167                     roots_map, locator,
1168                     force_rebuild=(tries_left < num_retries),
1169                     need_writable=True,
1170                     headers=headers)
1171             except Exception as error:
1172                 loop.save_result(error)
1173                 continue
1174
1175             writer_pool = KeepClient.KeepWriterThreadPool(data=data,
1176                                                         data_hash=data_hash,
1177                                                         copies=copies - done,
1178                                                         max_service_replicas=self.max_replicas_per_service,
1179                                                         timeout=self.current_timeout(num_retries - tries_left))
1180             for service_root, ks in [(root, roots_map[root])
1181                                      for root in sorted_roots]:
1182                 if ks.finished():
1183                     continue
1184                 writer_pool.add_task(ks, service_root)
1185             writer_pool.join()
1186             done += writer_pool.done()
1187             loop.save_result((done >= copies, writer_pool.total_task_nr))
1188
1189         if loop.success():
1190             return writer_pool.response()
1191         if not roots_map:
1192             raise arvados.errors.KeepWriteError(
1193                 "failed to write {}: no Keep services available ({})".format(
1194                     data_hash, loop.last_result()))
1195         else:
1196             service_errors = ((key, roots_map[key].last_result()['error'])
1197                               for key in sorted_roots
1198                               if roots_map[key].last_result()['error'])
1199             raise arvados.errors.KeepWriteError(
1200                 "failed to write {} (wanted {} copies but wrote {})".format(
1201                     data_hash, copies, writer_pool.done()), service_errors, label="service")
1202
1203     def local_store_put(self, data, copies=1, num_retries=None):
1204         """A stub for put().
1205
1206         This method is used in place of the real put() method when
1207         using local storage (see constructor's local_store argument).
1208
1209         copies and num_retries arguments are ignored: they are here
1210         only for the sake of offering the same call signature as
1211         put().
1212
1213         Data stored this way can be retrieved via local_store_get().
1214         """
1215         md5 = hashlib.md5(data).hexdigest()
1216         locator = '%s+%d' % (md5, len(data))
1217         with open(os.path.join(self.local_store, md5 + '.tmp'), 'wb') as f:
1218             f.write(data)
1219         os.rename(os.path.join(self.local_store, md5 + '.tmp'),
1220                   os.path.join(self.local_store, md5))
1221         return locator
1222
1223     def local_store_get(self, loc_s, num_retries=None):
1224         """Companion to local_store_put()."""
1225         try:
1226             locator = KeepLocator(loc_s)
1227         except ValueError:
1228             raise arvados.errors.NotFoundError(
1229                 "Invalid data locator: '%s'" % loc_s)
1230         if locator.md5sum == config.EMPTY_BLOCK_LOCATOR.split('+')[0]:
1231             return b''
1232         with open(os.path.join(self.local_store, locator.md5sum), 'rb') as f:
1233             return f.read()
1234
1235     def is_cached(self, locator):
1236         return self.block_cache.reserve_cache(expect_hash)