src/allmydata/mutable/publish.py

   1
   2
   3 import os, struct, time
   4 from itertools import count
   5 from zope.interface import implements
   6 from twisted.internet import defer
   7 from twisted.python import failure
   8 from allmydata.interfaces import IPublishStatus
   9 from allmydata.util import base32, hashutil, mathutil, idlib, log
  10 from allmydata import hashtree, codec
  11 from allmydata.storage.server import si_b2a
  12 from pycryptopp.cipher.aes import AES
  13 from foolscap.api import eventually, fireEventually
  14
  15 from allmydata.mutable.common import MODE_WRITE, MODE_CHECK, DictOfSets, \
  16      UncoordinatedWriteError, NotEnoughServersError
  17 from allmydata.mutable.servermap import ServerMap
  18 from allmydata.mutable.layout import pack_prefix, pack_share, unpack_header, pack_checkstring, \
  19      unpack_checkstring, SIGNED_PREFIX
  20
  21 class PublishStatus:
  22     implements(IPublishStatus)
  23     statusid_counter = count(0)
  24     def __init__(self):
  25         self.timings = {}
  26         self.timings["send_per_server"] = {}
  27         self.servermap = None
  28         self.problems = {}
  29         self.active = True
  30         self.storage_index = None
  31         self.helper = False
  32         self.encoding = ("?", "?")
  33         self.size = None
  34         self.status = "Not started"
  35         self.progress = 0.0
  36         self.counter = self.statusid_counter.next()
  37         self.started = time.time()
  38
  39     def add_per_server_time(self, peerid, elapsed):
  40         if peerid not in self.timings["send_per_server"]:
  41             self.timings["send_per_server"][peerid] = []
  42         self.timings["send_per_server"][peerid].append(elapsed)
  43
  44     def get_started(self):
  45         return self.started
  46     def get_storage_index(self):
  47         return self.storage_index
  48     def get_encoding(self):
  49         return self.encoding
  50     def using_helper(self):
  51         return self.helper
  52     def get_servermap(self):
  53         return self.servermap
  54     def get_size(self):
  55         return self.size
  56     def get_status(self):
  57         return self.status
  58     def get_progress(self):
  59         return self.progress
  60     def get_active(self):
  61         return self.active
  62     def get_counter(self):
  63         return self.counter
  64
  65     def set_storage_index(self, si):
  66         self.storage_index = si
  67     def set_helper(self, helper):
  68         self.helper = helper
  69     def set_servermap(self, servermap):
  70         self.servermap = servermap
  71     def set_encoding(self, k, n):
  72         self.encoding = (k, n)
  73     def set_size(self, size):
  74         self.size = size
  75     def set_status(self, status):
  76         self.status = status
  77     def set_progress(self, value):
  78         self.progress = value
  79     def set_active(self, value):
  80         self.active = value
  81
  82 class LoopLimitExceededError(Exception):
  83     pass
  84
  85 class Publish:
  86     """I represent a single act of publishing the mutable file to the grid. I
  87     will only publish my data if the servermap I am using still represents
  88     the current state of the world.
  89
  90     To make the initial publish, set servermap to None.
  91     """
  92
  93     def __init__(self, filenode, storage_broker, servermap):
  94         self._node = filenode
  95         self._storage_broker = storage_broker
  96         self._servermap = servermap
  97         self._storage_index = self._node.get_storage_index()
  98         self._log_prefix = prefix = si_b2a(self._storage_index)[:5]
  99         num = self.log("Publish(%s): starting" % prefix, parent=None)
 100         self._log_number = num
 101         self._running = True
 102         self._first_write_error = None
 103
 104         self._status = PublishStatus()
 105         self._status.set_storage_index(self._storage_index)
 106         self._status.set_helper(False)
 107         self._status.set_progress(0.0)
 108         self._status.set_active(True)
 109
 110     def get_status(self):
 111         return self._status
 112
 113     def log(self, *args, **kwargs):
 114         if 'parent' not in kwargs:
 115             kwargs['parent'] = self._log_number
 116         if "facility" not in kwargs:
 117             kwargs["facility"] = "tahoe.mutable.publish"
 118         return log.msg(*args, **kwargs)
 119
 120     def publish(self, newdata):
 121         """Publish the filenode's current contents.  Returns a Deferred that
 122         fires (with None) when the publish has done as much work as it's ever
 123         going to do, or errbacks with ConsistencyError if it detects a
 124         simultaneous write.
 125         """
 126
 127         # 1: generate shares (SDMF: files are small, so we can do it in RAM)
 128         # 2: perform peer selection, get candidate servers
 129         #  2a: send queries to n+epsilon servers, to determine current shares
 130         #  2b: based upon responses, create target map
 131         # 3: send slot_testv_and_readv_and_writev messages
 132         # 4: as responses return, update share-dispatch table
 133         # 4a: may need to run recovery algorithm
 134         # 5: when enough responses are back, we're done
 135
 136         self.log("starting publish, datalen is %s" % len(newdata))
 137         self._status.set_size(len(newdata))
 138         self._status.set_status("Started")
 139         self._started = time.time()
 140
 141         self.done_deferred = defer.Deferred()
 142
 143         self._writekey = self._node.get_writekey()
 144         assert self._writekey, "need write capability to publish"
 145
 146         # first, which servers will we publish to? We require that the
 147         # servermap was updated in MODE_WRITE, so we can depend upon the
 148         # peerlist computed by that process instead of computing our own.
 149         if self._servermap:
 150             assert self._servermap.last_update_mode in (MODE_WRITE, MODE_CHECK)
 151             # we will push a version that is one larger than anything present
 152             # in the grid, according to the servermap.
 153             self._new_seqnum = self._servermap.highest_seqnum() + 1
 154         else:
 155             # If we don't have a servermap, that's because we're doing the
 156             # initial publish
 157             self._new_seqnum = 1
 158             self._servermap = ServerMap()
 159         self._status.set_servermap(self._servermap)
 160
 161         self.log(format="new seqnum will be %(seqnum)d",
 162                  seqnum=self._new_seqnum, level=log.NOISY)
 163
 164         # having an up-to-date servermap (or using a filenode that was just
 165         # created for the first time) also guarantees that the following
 166         # fields are available
 167         self.readkey = self._node.get_readkey()
 168         self.required_shares = self._node.get_required_shares()
 169         assert self.required_shares is not None
 170         self.total_shares = self._node.get_total_shares()
 171         assert self.total_shares is not None
 172         self._status.set_encoding(self.required_shares, self.total_shares)
 173
 174         self._pubkey = self._node.get_pubkey()
 175         assert self._pubkey
 176         self._privkey = self._node.get_privkey()
 177         assert self._privkey
 178         self._encprivkey = self._node.get_encprivkey()
 179
 180         sb = self._storage_broker
 181         full_peerlist = sb.get_servers_for_index(self._storage_index)
 182         self.full_peerlist = full_peerlist # for use later, immutable
 183         self.bad_peers = set() # peerids who have errbacked/refused requests
 184
 185         self.newdata = newdata
 186         self.salt = os.urandom(16)
 187
 188         self.setup_encoding_parameters()
 189
 190         # if we experience any surprises (writes which were rejected because
 191         # our test vector did not match, or shares which we didn't expect to
 192         # see), we set this flag and report an UncoordinatedWriteError at the
 193         # end of the publish process.
 194         self.surprised = False
 195
 196         # as a failsafe, refuse to iterate through self.loop more than a
 197         # thousand times.
 198         self.looplimit = 1000
 199
 200         # we keep track of three tables. The first is our goal: which share
 201         # we want to see on which servers. This is initially populated by the
 202         # existing servermap.
 203         self.goal = set() # pairs of (peerid, shnum) tuples
 204
 205         # the second table is our list of outstanding queries: those which
 206         # are in flight and may or may not be delivered, accepted, or
 207         # acknowledged. Items are added to this table when the request is
 208         # sent, and removed when the response returns (or errbacks).
 209         self.outstanding = set() # (peerid, shnum) tuples
 210
 211         # the third is a table of successes: share which have actually been
 212         # placed. These are populated when responses come back with success.
 213         # When self.placed == self.goal, we're done.
 214         self.placed = set() # (peerid, shnum) tuples
 215
 216         # we also keep a mapping from peerid to RemoteReference. Each time we
 217         # pull a connection out of the full peerlist, we add it to this for
 218         # use later.
 219         self.connections = {}
 220
 221         self.bad_share_checkstrings = {}
 222
 223         # we use the servermap to populate the initial goal: this way we will
 224         # try to update each existing share in place.
 225         for (peerid, shnum) in self._servermap.servermap:
 226             self.goal.add( (peerid, shnum) )
 227             self.connections[peerid] = self._servermap.connections[peerid]
 228         # then we add in all the shares that were bad (corrupted, bad
 229         # signatures, etc). We want to replace these.
 230         for key, old_checkstring in self._servermap.bad_shares.items():
 231             (peerid, shnum) = key
 232             self.goal.add(key)
 233             self.bad_share_checkstrings[key] = old_checkstring
 234             self.connections[peerid] = self._servermap.connections[peerid]
 235
 236         # create the shares. We'll discard these as they are delivered. SDMF:
 237         # we're allowed to hold everything in memory.
 238
 239         self._status.timings["setup"] = time.time() - self._started
 240         d = self._encrypt_and_encode()
 241         d.addCallback(self._generate_shares)
 242         def _start_pushing(res):
 243             self._started_pushing = time.time()
 244             return res
 245         d.addCallback(_start_pushing)
 246         d.addCallback(self.loop) # trigger delivery
 247         d.addErrback(self._fatal_error)
 248
 249         return self.done_deferred
 250
 251     def setup_encoding_parameters(self):
 252         segment_size = len(self.newdata)
 253         # this must be a multiple of self.required_shares
 254         segment_size = mathutil.next_multiple(segment_size,
 255                                               self.required_shares)
 256         self.segment_size = segment_size
 257         if segment_size:
 258             self.num_segments = mathutil.div_ceil(len(self.newdata),
 259                                                   segment_size)
 260         else:
 261             self.num_segments = 0
 262         assert self.num_segments in [0, 1,] # SDMF restrictions
 263
 264     def _fatal_error(self, f):
 265         self.log("error during loop", failure=f, level=log.UNUSUAL)
 266         self._done(f)
 267
 268     def _update_status(self):
 269         self._status.set_status("Sending Shares: %d placed out of %d, "
 270                                 "%d messages outstanding" %
 271                                 (len(self.placed),
 272                                  len(self.goal),
 273                                  len(self.outstanding)))
 274         self._status.set_progress(1.0 * len(self.placed) / len(self.goal))
 275
 276     def loop(self, ignored=None):
 277         self.log("entering loop", level=log.NOISY)
 278         if not self._running:
 279             return
 280
 281         self.looplimit -= 1
 282         if self.looplimit <= 0:
 283             raise LoopLimitExceededError("loop limit exceeded")
 284
 285         if self.surprised:
 286             # don't send out any new shares, just wait for the outstanding
 287             # ones to be retired.
 288             self.log("currently surprised, so don't send any new shares",
 289                      level=log.NOISY)
 290         else:
 291             self.update_goal()
 292             # how far are we from our goal?
 293             needed = self.goal - self.placed - self.outstanding
 294             self._update_status()
 295
 296             if needed:
 297                 # we need to send out new shares
 298                 self.log(format="need to send %(needed)d new shares",
 299                          needed=len(needed), level=log.NOISY)
 300                 self._send_shares(needed)
 301                 return
 302
 303         if self.outstanding:
 304             # queries are still pending, keep waiting
 305             self.log(format="%(outstanding)d queries still outstanding",
 306                      outstanding=len(self.outstanding),
 307                      level=log.NOISY)
 308             return
 309
 310         # no queries outstanding, no placements needed: we're done
 311         self.log("no queries outstanding, no placements needed: done",
 312                  level=log.OPERATIONAL)
 313         now = time.time()
 314         elapsed = now - self._started_pushing
 315         self._status.timings["push"] = elapsed
 316         return self._done(None)
 317
 318     def log_goal(self, goal, message=""):
 319         logmsg = [message]
 320         for (shnum, peerid) in sorted([(s,p) for (p,s) in goal]):
 321             logmsg.append("sh%d to [%s]" % (shnum,
 322                                             idlib.shortnodeid_b2a(peerid)))
 323         self.log("current goal: %s" % (", ".join(logmsg)), level=log.NOISY)
 324         self.log("we are planning to push new seqnum=#%d" % self._new_seqnum,
 325                  level=log.NOISY)
 326
 327     def update_goal(self):
 328         # if log.recording_noisy
 329         if True:
 330             self.log_goal(self.goal, "before update: ")
 331
 332         # first, remove any bad peers from our goal
 333         self.goal = set([ (peerid, shnum)
 334                           for (peerid, shnum) in self.goal
 335                           if peerid not in self.bad_peers ])
 336
 337         # find the homeless shares:
 338         homefull_shares = set([shnum for (peerid, shnum) in self.goal])
 339         homeless_shares = set(range(self.total_shares)) - homefull_shares
 340         homeless_shares = sorted(list(homeless_shares))
 341         # place them somewhere. We prefer unused servers at the beginning of
 342         # the available peer list.
 343
 344         if not homeless_shares:
 345             return
 346
 347         # if an old share X is on a node, put the new share X there too.
 348         # TODO: 1: redistribute shares to achieve one-per-peer, by copying
 349         #       shares from existing peers to new (less-crowded) ones. The
 350         #       old shares must still be updated.
 351         # TODO: 2: move those shares instead of copying them, to reduce future
 352         #       update work
 353
 354         # this is a bit CPU intensive but easy to analyze. We create a sort
 355         # order for each peerid. If the peerid is marked as bad, we don't
 356         # even put them in the list. Then we care about the number of shares
 357         # which have already been assigned to them. After that we care about
 358         # their permutation order.
 359         old_assignments = DictOfSets()
 360         for (peerid, shnum) in self.goal:
 361             old_assignments.add(peerid, shnum)
 362
 363         peerlist = []
 364         for i, (peerid, ss) in enumerate(self.full_peerlist):
 365             if peerid in self.bad_peers:
 366                 continue
 367             entry = (len(old_assignments.get(peerid, [])), i, peerid, ss)
 368             peerlist.append(entry)
 369         peerlist.sort()
 370
 371         if not peerlist:
 372             raise NotEnoughServersError("Ran out of non-bad servers, "
 373                                         "first_error=%s" %
 374                                         str(self._first_write_error),
 375                                         self._first_write_error)
 376
 377         # we then index this peerlist with an integer, because we may have to
 378         # wrap. We update the goal as we go.
 379         i = 0
 380         for shnum in homeless_shares:
 381             (ignored1, ignored2, peerid, ss) = peerlist[i]
 382             # if we are forced to send a share to a server that already has
 383             # one, we may have two write requests in flight, and the
 384             # servermap (which was computed before either request was sent)
 385             # won't reflect the new shares, so the second response will be
 386             # surprising. There is code in _got_write_answer() to tolerate
 387             # this, otherwise it would cause the publish to fail with an
 388             # UncoordinatedWriteError. See #546 for details of the trouble
 389             # this used to cause.
 390             self.goal.add( (peerid, shnum) )
 391             self.connections[peerid] = ss
 392             i += 1
 393             if i >= len(peerlist):
 394                 i = 0
 395         if True:
 396             self.log_goal(self.goal, "after update: ")
 397
 398
 399
 400     def _encrypt_and_encode(self):
 401         # this returns a Deferred that fires with a list of (sharedata,
 402         # sharenum) tuples. TODO: cache the ciphertext, only produce the
 403         # shares that we care about.
 404         self.log("_encrypt_and_encode")
 405
 406         self._status.set_status("Encrypting")
 407         started = time.time()
 408
 409         key = hashutil.ssk_readkey_data_hash(self.salt, self.readkey)
 410         enc = AES(key)
 411         crypttext = enc.process(self.newdata)
 412         assert len(crypttext) == len(self.newdata)
 413
 414         now = time.time()
 415         self._status.timings["encrypt"] = now - started
 416         started = now
 417
 418         # now apply FEC
 419
 420         self._status.set_status("Encoding")
 421         fec = codec.CRSEncoder()
 422         fec.set_params(self.segment_size,
 423                        self.required_shares, self.total_shares)
 424         piece_size = fec.get_block_size()
 425         crypttext_pieces = [None] * self.required_shares
 426         for i in range(len(crypttext_pieces)):
 427             offset = i * piece_size
 428             piece = crypttext[offset:offset+piece_size]
 429             piece = piece + "\x00"*(piece_size - len(piece)) # padding
 430             crypttext_pieces[i] = piece
 431             assert len(piece) == piece_size
 432
 433         d = fec.encode(crypttext_pieces)
 434         def _done_encoding(res):
 435             elapsed = time.time() - started
 436             self._status.timings["encode"] = elapsed
 437             return res
 438         d.addCallback(_done_encoding)
 439         return d
 440
 441     def _generate_shares(self, shares_and_shareids):
 442         # this sets self.shares and self.root_hash
 443         self.log("_generate_shares")
 444         self._status.set_status("Generating Shares")
 445         started = time.time()
 446
 447         # we should know these by now
 448         privkey = self._privkey
 449         encprivkey = self._encprivkey
 450         pubkey = self._pubkey
 451
 452         (shares, share_ids) = shares_and_shareids
 453
 454         assert len(shares) == len(share_ids)
 455         assert len(shares) == self.total_shares
 456         all_shares = {}
 457         block_hash_trees = {}
 458         share_hash_leaves = [None] * len(shares)
 459         for i in range(len(shares)):
 460             share_data = shares[i]
 461             shnum = share_ids[i]
 462             all_shares[shnum] = share_data
 463
 464             # build the block hash tree. SDMF has only one leaf.
 465             leaves = [hashutil.block_hash(share_data)]
 466             t = hashtree.HashTree(leaves)
 467             block_hash_trees[shnum] = list(t)
 468             share_hash_leaves[shnum] = t[0]
 469         for leaf in share_hash_leaves:
 470             assert leaf is not None
 471         share_hash_tree = hashtree.HashTree(share_hash_leaves)
 472         share_hash_chain = {}
 473         for shnum in range(self.total_shares):
 474             needed_hashes = share_hash_tree.needed_hashes(shnum)
 475             share_hash_chain[shnum] = dict( [ (i, share_hash_tree[i])
 476                                               for i in needed_hashes ] )
 477         root_hash = share_hash_tree[0]
 478         assert len(root_hash) == 32
 479         self.log("my new root_hash is %s" % base32.b2a(root_hash))
 480         self._new_version_info = (self._new_seqnum, root_hash, self.salt)
 481
 482         prefix = pack_prefix(self._new_seqnum, root_hash, self.salt,
 483                              self.required_shares, self.total_shares,
 484                              self.segment_size, len(self.newdata))
 485
 486         # now pack the beginning of the share. All shares are the same up
 487         # to the signature, then they have divergent share hash chains,
 488         # then completely different block hash trees + salt + share data,
 489         # then they all share the same encprivkey at the end. The sizes
 490         # of everything are the same for all shares.
 491
 492         sign_started = time.time()
 493         signature = privkey.sign(prefix)
 494         self._status.timings["sign"] = time.time() - sign_started
 495
 496         verification_key = pubkey.serialize()
 497
 498         final_shares = {}
 499         for shnum in range(self.total_shares):
 500             final_share = pack_share(prefix,
 501                                      verification_key,
 502                                      signature,
 503                                      share_hash_chain[shnum],
 504                                      block_hash_trees[shnum],
 505                                      all_shares[shnum],
 506                                      encprivkey)
 507             final_shares[shnum] = final_share
 508         elapsed = time.time() - started
 509         self._status.timings["pack"] = elapsed
 510         self.shares = final_shares
 511         self.root_hash = root_hash
 512
 513         # we also need to build up the version identifier for what we're
 514         # pushing. Extract the offsets from one of our shares.
 515         assert final_shares
 516         offsets = unpack_header(final_shares.values()[0])[-1]
 517         offsets_tuple = tuple( [(key,value) for key,value in offsets.items()] )
 518         verinfo = (self._new_seqnum, root_hash, self.salt,
 519                    self.segment_size, len(self.newdata),
 520                    self.required_shares, self.total_shares,
 521                    prefix, offsets_tuple)
 522         self.versioninfo = verinfo
 523
 524
 525
 526     def _send_shares(self, needed):
 527         self.log("_send_shares")
 528
 529         # we're finally ready to send out our shares. If we encounter any
 530         # surprises here, it's because somebody else is writing at the same
 531         # time. (Note: in the future, when we remove the _query_peers() step
 532         # and instead speculate about [or remember] which shares are where,
 533         # surprises here are *not* indications of UncoordinatedWriteError,
 534         # and we'll need to respond to them more gracefully.)
 535
 536         # needed is a set of (peerid, shnum) tuples. The first thing we do is
 537         # organize it by peerid.
 538
 539         peermap = DictOfSets()
 540         for (peerid, shnum) in needed:
 541             peermap.add(peerid, shnum)
 542
 543         # the next thing is to build up a bunch of test vectors. The
 544         # semantics of Publish are that we perform the operation if the world
 545         # hasn't changed since the ServerMap was constructed (more or less).
 546         # For every share we're trying to place, we create a test vector that
 547         # tests to see if the server*share still corresponds to the
 548         # map.
 549
 550         all_tw_vectors = {} # maps peerid to tw_vectors
 551         sm = self._servermap.servermap
 552
 553         for key in needed:
 554             (peerid, shnum) = key
 555
 556             if key in sm:
 557                 # an old version of that share already exists on the
 558                 # server, according to our servermap. We will create a
 559                 # request that attempts to replace it.
 560                 old_versionid, old_timestamp = sm[key]
 561                 (old_seqnum, old_root_hash, old_salt, old_segsize,
 562                  old_datalength, old_k, old_N, old_prefix,
 563                  old_offsets_tuple) = old_versionid
 564                 old_checkstring = pack_checkstring(old_seqnum,
 565                                                    old_root_hash,
 566                                                    old_salt)
 567                 testv = (0, len(old_checkstring), "eq", old_checkstring)
 568
 569             elif key in self.bad_share_checkstrings:
 570                 old_checkstring = self.bad_share_checkstrings[key]
 571                 testv = (0, len(old_checkstring), "eq", old_checkstring)
 572
 573             else:
 574                 # add a testv that requires the share not exist
 575
 576                 # Unfortunately, foolscap-0.2.5 has a bug in the way inbound
 577                 # constraints are handled. If the same object is referenced
 578                 # multiple times inside the arguments, foolscap emits a
 579                 # 'reference' token instead of a distinct copy of the
 580                 # argument. The bug is that these 'reference' tokens are not
 581                 # accepted by the inbound constraint code. To work around
 582                 # this, we need to prevent python from interning the
 583                 # (constant) tuple, by creating a new copy of this vector
 584                 # each time.
 585
 586                 # This bug is fixed in foolscap-0.2.6, and even though this
 587                 # version of Tahoe requires foolscap-0.3.1 or newer, we are
 588                 # supposed to be able to interoperate with older versions of
 589                 # Tahoe which are allowed to use older versions of foolscap,
 590                 # including foolscap-0.2.5 . In addition, I've seen other
 591                 # foolscap problems triggered by 'reference' tokens (see #541
 592                 # for details). So we must keep this workaround in place.
 593
 594                 #testv = (0, 1, 'eq', "")
 595                 testv = tuple([0, 1, 'eq', ""])
 596
 597             testvs = [testv]
 598             # the write vector is simply the share
 599             writev = [(0, self.shares[shnum])]
 600
 601             if peerid not in all_tw_vectors:
 602                 all_tw_vectors[peerid] = {}
 603                 # maps shnum to (testvs, writevs, new_length)
 604             assert shnum not in all_tw_vectors[peerid]
 605
 606             all_tw_vectors[peerid][shnum] = (testvs, writev, None)
 607
 608         # we read the checkstring back from each share, however we only use
 609         # it to detect whether there was a new share that we didn't know
 610         # about. The success or failure of the write will tell us whether
 611         # there was a collision or not. If there is a collision, the first
 612         # thing we'll do is update the servermap, which will find out what
 613         # happened. We could conceivably reduce a roundtrip by using the
 614         # readv checkstring to populate the servermap, but really we'd have
 615         # to read enough data to validate the signatures too, so it wouldn't
 616         # be an overall win.
 617         read_vector = [(0, struct.calcsize(SIGNED_PREFIX))]
 618
 619         # ok, send the messages!
 620         self.log("sending %d shares" % len(all_tw_vectors), level=log.NOISY)
 621         started = time.time()
 622         for (peerid, tw_vectors) in all_tw_vectors.items():
 623
 624             write_enabler = self._node.get_write_enabler(peerid)
 625             renew_secret = self._node.get_renewal_secret(peerid)
 626             cancel_secret = self._node.get_cancel_secret(peerid)
 627             secrets = (write_enabler, renew_secret, cancel_secret)
 628             shnums = tw_vectors.keys()
 629
 630             for shnum in shnums:
 631                 self.outstanding.add( (peerid, shnum) )
 632
 633             d = self._do_testreadwrite(peerid, secrets,
 634                                        tw_vectors, read_vector)
 635             d.addCallbacks(self._got_write_answer, self._got_write_error,
 636                            callbackArgs=(peerid, shnums, started),
 637                            errbackArgs=(peerid, shnums, started))
 638             # tolerate immediate errback, like with DeadReferenceError
 639             d.addBoth(fireEventually)
 640             d.addCallback(self.loop)
 641             d.addErrback(self._fatal_error)
 642
 643         self._update_status()
 644         self.log("%d shares sent" % len(all_tw_vectors), level=log.NOISY)
 645
 646     def _do_testreadwrite(self, peerid, secrets,
 647                           tw_vectors, read_vector):
 648         storage_index = self._storage_index
 649         ss = self.connections[peerid]
 650
 651         #print "SS[%s] is %s" % (idlib.shortnodeid_b2a(peerid), ss), ss.tracker.interfaceName
 652         d = ss.callRemote("slot_testv_and_readv_and_writev",
 653                           storage_index,
 654                           secrets,
 655                           tw_vectors,
 656                           read_vector)
 657         return d
 658
 659     def _got_write_answer(self, answer, peerid, shnums, started):
 660         lp = self.log("_got_write_answer from %s" %
 661                       idlib.shortnodeid_b2a(peerid))
 662         for shnum in shnums:
 663             self.outstanding.discard( (peerid, shnum) )
 664
 665         now = time.time()
 666         elapsed = now - started
 667         self._status.add_per_server_time(peerid, elapsed)
 668
 669         wrote, read_data = answer
 670
 671         surprise_shares = set(read_data.keys()) - set(shnums)
 672
 673         surprised = False
 674         for shnum in surprise_shares:
 675             # read_data is a dict mapping shnum to checkstring (SIGNED_PREFIX)
 676             checkstring = read_data[shnum][0]
 677             their_version_info = unpack_checkstring(checkstring)
 678             if their_version_info == self._new_version_info:
 679                 # they have the right share, somehow
 680
 681                 if (peerid,shnum) in self.goal:
 682                     # and we want them to have it, so we probably sent them a
 683                     # copy in an earlier write. This is ok, and avoids the
 684                     # #546 problem.
 685                     continue
 686
 687                 # They aren't in our goal, but they are still for the right
 688                 # version. Somebody else wrote them, and it's a convergent
 689                 # uncoordinated write. Pretend this is ok (don't be
 690                 # surprised), since I suspect there's a decent chance that
 691                 # we'll hit this in normal operation.
 692                 continue
 693
 694             else:
 695                 # the new shares are of a different version
 696                 if peerid in self._servermap.reachable_peers:
 697                     # we asked them about their shares, so we had knowledge
 698                     # of what they used to have. Any surprising shares must
 699                     # have come from someone else, so UCW.
 700                     surprised = True
 701                 else:
 702                     # we didn't ask them, and now we've discovered that they
 703                     # have a share we didn't know about. This indicates that
 704                     # mapupdate should have wokred harder and asked more
 705                     # servers before concluding that it knew about them all.
 706
 707                     # signal UCW, but make sure to ask this peer next time,
 708                     # so we'll remember to update it if/when we retry.
 709                     surprised = True
 710                     # TODO: ask this peer next time. I don't yet have a good
 711                     # way to do this. Two insufficient possibilities are:
 712                     #
 713                     # self._servermap.add_new_share(peerid, shnum, verinfo, now)
 714                     #  but that requires fetching/validating/parsing the whole
 715                     #  version string, and all we have is the checkstring
 716                     # self._servermap.mark_bad_share(peerid, shnum, checkstring)
 717                     #  that will make publish overwrite the share next time,
 718                     #  but it won't re-query the server, and it won't make
 719                     #  mapupdate search further
 720
 721                     # TODO later: when publish starts, do
 722                     # servermap.get_best_version(), extract the seqnum,
 723                     # subtract one, and store as highest-replaceable-seqnum.
 724                     # Then, if this surprise-because-we-didn't-ask share is
 725                     # of highest-replaceable-seqnum or lower, we're allowed
 726                     # to replace it: send out a new writev (or rather add it
 727                     # to self.goal and loop).
 728                     pass
 729
 730                 surprised = True
 731
 732         if surprised:
 733             self.log("they had shares %s that we didn't know about" %
 734                      (list(surprise_shares),),
 735                      parent=lp, level=log.WEIRD, umid="un9CSQ")
 736             self.surprised = True
 737
 738         if not wrote:
 739             # TODO: there are two possibilities. The first is that the server
 740             # is full (or just doesn't want to give us any room), which means
 741             # we shouldn't ask them again, but is *not* an indication of an
 742             # uncoordinated write. The second is that our testv failed, which
 743             # *does* indicate an uncoordinated write. We currently don't have
 744             # a way to tell these two apart (in fact, the storage server code
 745             # doesn't have the option of refusing our share).
 746             #
 747             # If the server is full, mark the peer as bad (so we don't ask
 748             # them again), but don't set self.surprised. The loop() will find
 749             # a new server.
 750             #
 751             # If the testv failed, log it, set self.surprised, but don't
 752             # bother adding to self.bad_peers .
 753
 754             self.log("our testv failed, so the write did not happen",
 755                      parent=lp, level=log.WEIRD, umid="8sc26g")
 756             self.surprised = True
 757             self.bad_peers.add(peerid) # don't ask them again
 758             # use the checkstring to add information to the log message
 759             for (shnum,readv) in read_data.items():
 760                 checkstring = readv[0]
 761                 (other_seqnum,
 762                  other_roothash,
 763                  other_salt) = unpack_checkstring(checkstring)
 764                 expected_version = self._servermap.version_on_peer(peerid,
 765                                                                    shnum)
 766                 if expected_version:
 767                     (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
 768                      offsets_tuple) = expected_version
 769                     self.log("somebody modified the share on us:"
 770                              " shnum=%d: I thought they had #%d:R=%s,"
 771                              " but testv reported #%d:R=%s" %
 772                              (shnum,
 773                               seqnum, base32.b2a(root_hash)[:4],
 774                               other_seqnum, base32.b2a(other_roothash)[:4]),
 775                              parent=lp, level=log.NOISY)
 776                 # if expected_version==None, then we didn't expect to see a
 777                 # share on that peer, and the 'surprise_shares' clause above
 778                 # will have logged it.
 779             # self.loop() will take care of finding new homes
 780             return
 781
 782         for shnum in shnums:
 783             self.placed.add( (peerid, shnum) )
 784             # and update the servermap
 785             self._servermap.add_new_share(peerid, shnum,
 786                                           self.versioninfo, started)
 787
 788         # self.loop() will take care of checking to see if we're done
 789         return
 790
 791     def _got_write_error(self, f, peerid, shnums, started):
 792         for shnum in shnums:
 793             self.outstanding.discard( (peerid, shnum) )
 794         self.bad_peers.add(peerid)
 795         if self._first_write_error is None:
 796             self._first_write_error = f
 797         self.log(format="error while writing shares %(shnums)s to peerid %(peerid)s",
 798                  shnums=list(shnums), peerid=idlib.shortnodeid_b2a(peerid),
 799                  failure=f,
 800                  level=log.UNUSUAL)
 801         # self.loop() will take care of checking to see if we're done
 802         return
 803
 804
 805     def _done(self, res):
 806         if not self._running:
 807             return
 808         self._running = False
 809         now = time.time()
 810         self._status.timings["total"] = now - self._started
 811         self._status.set_active(False)
 812         if isinstance(res, failure.Failure):
 813             self.log("Publish done, with failure", failure=res,
 814                      level=log.WEIRD, umid="nRsR9Q")
 815             self._status.set_status("Failed")
 816         elif self.surprised:
 817             self.log("Publish done, UncoordinatedWriteError", level=log.UNUSUAL)
 818             self._status.set_status("UncoordinatedWriteError")
 819             # deliver a failure
 820             res = failure.Failure(UncoordinatedWriteError())
 821             # TODO: recovery
 822         else:
 823             self.log("Publish done, success")
 824             self._status.set_status("Finished")
 825             self._status.set_progress(1.0)
 826         eventually(self.done_deferred.callback, res)
 827
 828