2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
8 Hash = StringConstraint(maxLength=HASH_SIZE,
9 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
10 Nodeid = StringConstraint(maxLength=20,
11 minLength=20) # binary format 20-byte SHA1 hash
12 FURL = StringConstraint(1000)
13 StorageIndex = StringConstraint(16)
14 URI = StringConstraint(300) # kind of arbitrary
16 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
18 ShareData = StringConstraint(None)
19 URIExtensionData = StringConstraint(1000)
20 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
22 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
23 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
24 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
25 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
27 class RIStubClient(RemoteInterface):
28 """Each client publishes a service announcement for a dummy object called
29 the StubClient. This object doesn't actually offer any services, but the
30 announcement helps the Introducer keep track of which clients are
31 subscribed (so the grid admin can keep track of things like the size of
32 the grid and the client versions in use. This is the (empty)
33 RemoteInterface for the StubClient."""
35 class RIBucketWriter(RemoteInterface):
36 """ Objects of this kind live on the server side. """
37 def write(offset=Offset, data=ShareData):
42 If the data that has been written is incomplete or inconsistent then
43 the server will throw the data away, else it will store it for future
49 """Abandon all the data that has been written.
53 class RIBucketReader(RemoteInterface):
54 def read(offset=Offset, length=ReadSize):
57 def advise_corrupt_share(reason=str):
58 """Clients who discover hash failures in shares that they have
59 downloaded from me will use this method to inform me about the
60 failures. I will record their concern so that my operator can
61 manually inspect the shares in question. I return None.
63 This is a wrapper around RIStorageServer.advise_corrupt_share(),
64 which is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
85 class RIStorageServer(RemoteInterface):
86 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
90 Return a dictionary of version information.
92 return DictOf(str, Any())
94 def allocate_buckets(storage_index=StorageIndex,
95 renew_secret=LeaseRenewSecret,
96 cancel_secret=LeaseCancelSecret,
97 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
98 allocated_size=Offset, canary=Referenceable):
100 @param storage_index: the index of the bucket to be created or
102 @param sharenums: these are the share numbers (probably between 0 and
103 99) that the sender is proposing to store on this
105 @param renew_secret: This is the secret used to protect bucket refresh
106 This secret is generated by the client and
107 stored for later comparison by the server. Each
108 server is given a different secret.
109 @param cancel_secret: Like renew_secret, but protects bucket decref.
110 @param canary: If the canary is lost before close(), the bucket is
112 @return: tuple of (alreadygot, allocated), where alreadygot is what we
113 already have and allocated is what we hereby agree to accept.
114 New leases are added for shares in both lists.
116 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
117 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
119 def add_lease(storage_index=StorageIndex,
120 renew_secret=LeaseRenewSecret,
121 cancel_secret=LeaseCancelSecret):
123 Add a new lease on the given bucket. If the renew_secret matches an
124 existing lease, that lease will be renewed instead. If there is no
125 bucket for the given storage_index, return silently. (note that in
126 tahoe-1.3.0 and earlier, IndexError was raised if there was no
129 return Any() # returns None now, but future versions might change
131 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
133 Renew the lease on a given bucket, resetting the timer to 31 days.
134 Some networks will use this, some will not. If there is no bucket for
135 the given storage_index, IndexError will be raised.
137 For mutable shares, if the given renew_secret does not match an
138 existing lease, IndexError will be raised with a note listing the
139 server-nodeids on the existing leases, so leases on migrated shares
140 can be renewed or cancelled. For immutable shares, IndexError
141 (without the note) will be raised.
145 def cancel_lease(storage_index=StorageIndex,
146 cancel_secret=LeaseCancelSecret):
148 Cancel the lease on a given bucket. If this was the last lease on the
149 bucket, the bucket will be deleted. If there is no bucket for the
150 given storage_index, IndexError will be raised.
152 For mutable shares, if the given cancel_secret does not match an
153 existing lease, IndexError will be raised with a note listing the
154 server-nodeids on the existing leases, so leases on migrated shares
155 can be renewed or cancelled. For immutable shares, IndexError
156 (without the note) will be raised.
160 def get_buckets(storage_index=StorageIndex):
161 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
165 def slot_readv(storage_index=StorageIndex,
166 shares=ListOf(int), readv=ReadVector):
167 """Read a vector from the numbered shares associated with the given
168 storage index. An empty shares list means to return data from all
169 known shares. Returns a dictionary with one key per share."""
170 return DictOf(int, ReadData) # shnum -> results
172 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
173 secrets=TupleOf(WriteEnablerSecret,
176 tw_vectors=TestAndWriteVectorsForShares,
179 """General-purpose test-and-set operation for mutable slots. Perform
180 a bunch of comparisons against the existing shares. If they all pass,
181 then apply a bunch of write vectors to those shares. Then use the
182 read vectors to extract data from all the shares and return the data.
184 This method is, um, large. The goal is to allow clients to update all
185 the shares associated with a mutable file in a single round trip.
187 @param storage_index: the index of the bucket to be created or
189 @param write_enabler: a secret that is stored along with the slot.
190 Writes are accepted from any caller who can
191 present the matching secret. A different secret
192 should be used for each slot*server pair.
193 @param renew_secret: This is the secret used to protect bucket refresh
194 This secret is generated by the client and
195 stored for later comparison by the server. Each
196 server is given a different secret.
197 @param cancel_secret: Like renew_secret, but protects bucket decref.
199 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
200 cancel_secret). The first is required to perform any write. The
201 latter two are used when allocating new shares. To simply acquire a
202 new lease on existing shares, use an empty testv and an empty writev.
204 Each share can have a separate test vector (i.e. a list of
205 comparisons to perform). If all vectors for all shares pass, then all
206 writes for all shares are recorded. Each comparison is a 4-tuple of
207 (offset, length, operator, specimen), which effectively does a bool(
208 (read(offset, length)) OPERATOR specimen ) and only performs the
209 write if all these evaluate to True. Basic test-and-set uses 'eq'.
210 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
211 Write-if-same-or-newer uses 'le'.
213 Reads from the end of the container are truncated, and missing shares
214 behave like empty ones, so to assert that a share doesn't exist (for
215 use when creating a new share), use (0, 1, 'eq', '').
217 The write vector will be applied to the given share, expanding it if
218 necessary. A write vector applied to a share number that did not
219 exist previously will cause that share to be created.
221 Each write vector is accompanied by a 'new_length' argument. If
222 new_length is not None, use it to set the size of the container. This
223 can be used to pre-allocate space for a series of upcoming writes, or
224 truncate existing data. If the container is growing, new_length will
225 be applied before datav. If the container is shrinking, it will be
226 applied afterwards. If new_length==0, the share will be deleted.
228 The read vector is used to extract data from all known shares,
229 *before* any writes have been applied. The same vector is used for
230 all shares. This captures the state that was tested by the test
233 This method returns two values: a boolean and a dict. The boolean is
234 True if the write vectors were applied, False if not. The dict is
235 keyed by share number, and each value contains a list of strings, one
236 for each element of the read vector.
238 If the write_enabler is wrong, this will raise BadWriteEnablerError.
239 To enable share migration (using update_write_enabler), the exception
240 will have the nodeid used for the old write enabler embedded in it,
241 in the following string::
243 The write enabler was recorded by nodeid '%s'.
245 Note that the nodeid here is encoded using the same base32 encoding
246 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
249 return TupleOf(bool, DictOf(int, ReadData))
251 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
252 shnum=int, reason=str):
253 """Clients who discover hash failures in shares that they have
254 downloaded from me will use this method to inform me about the
255 failures. I will record their concern so that my operator can
256 manually inspect the shares in question. I return None.
258 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
259 (binary) storage index string, and 'shnum' is the integer share
260 number. 'reason' is a human-readable explanation of the problem,
261 probably including some expected hash values and the computed ones
262 which did not match. Corruption advisories for mutable shares should
263 include a hash of the public key (the same value that appears in the
264 mutable-file verify-cap), since the current share format does not
268 class IStorageBucketWriter(Interface):
270 Objects of this kind live on the client side.
272 def put_block(segmentnum=int, data=ShareData):
273 """@param data: For most segments, this data will be 'blocksize'
274 bytes in length. The last segment might be shorter.
275 @return: a Deferred that fires (with None) when the operation completes
278 def put_plaintext_hashes(hashes=ListOf(Hash)):
280 @return: a Deferred that fires (with None) when the operation completes
283 def put_crypttext_hashes(hashes=ListOf(Hash)):
285 @return: a Deferred that fires (with None) when the operation completes
288 def put_block_hashes(blockhashes=ListOf(Hash)):
290 @return: a Deferred that fires (with None) when the operation completes
293 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
295 @return: a Deferred that fires (with None) when the operation completes
298 def put_uri_extension(data=URIExtensionData):
299 """This block of data contains integrity-checking information (hashes
300 of plaintext, crypttext, and shares), as well as encoding parameters
301 that are necessary to recover the data. This is a serialized dict
302 mapping strings to other strings. The hash of this data is kept in
303 the URI and verified before any of the data is used. All buckets for
304 a given file contain identical copies of this data.
306 The serialization format is specified with the following pseudocode:
307 for k in sorted(dict.keys()):
308 assert re.match(r'^[a-zA-Z_\-]+$', k)
309 write(k + ':' + netstring(dict[k]))
311 @return: a Deferred that fires (with None) when the operation completes
315 """Finish writing and close the bucket. The share is not finalized
316 until this method is called: if the uploading client disconnects
317 before calling close(), the partially-written share will be
320 @return: a Deferred that fires (with None) when the operation completes
323 class IStorageBucketReader(Interface):
325 def get_block_data(blocknum=int, blocksize=int, size=int):
326 """Most blocks will be the same size. The last block might be shorter
332 def get_crypttext_hashes():
334 @return: ListOf(Hash)
337 def get_block_hashes(at_least_these=SetOf(int)):
339 @return: ListOf(Hash)
342 def get_share_hashes(at_least_these=SetOf(int)):
344 @return: ListOf(TupleOf(int, Hash))
347 def get_uri_extension():
349 @return: URIExtensionData
352 class IStorageBroker(Interface):
353 def get_servers_for_index(peer_selection_index):
355 @return: list of (peerid, versioned-rref) tuples
357 def get_all_servers():
359 @return: frozenset of (peerid, versioned-rref) tuples
361 def get_all_serverids():
363 @return: frozenset of serverid strings
365 def get_nickname_for_serverid(serverid):
367 @return: unicode nickname, or None
370 # methods moved from IntroducerClient, need review
371 def get_all_connections():
372 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
373 each active connection we've established to a remote service. This is
374 mostly useful for unit tests that need to wait until a certain number
375 of connections have been made."""
377 def get_all_connectors():
378 """Return a dict that maps from (nodeid, service_name) to a
379 RemoteServiceConnector instance for all services that we are actively
380 trying to connect to. Each RemoteServiceConnector has the following
383 service_name: the type of service provided, like 'storage'
384 announcement_time: when we first heard about this service
385 last_connect_time: when we last established a connection
386 last_loss_time: when we last lost a connection
388 version: the peer's version, from the most recent connection
389 oldest_supported: the peer's oldest supported version, same
391 rref: the RemoteReference, if connected, otherwise None
392 remote_host: the IAddress, if connected, otherwise None
394 This method is intended for monitoring interfaces, such as a web page
395 which describes connecting and connected peers.
398 def get_all_peerids():
399 """Return a frozenset of all peerids to whom we have a connection (to
400 one or more services) established. Mostly useful for unit tests."""
402 def get_all_connections_for(service_name):
403 """Return a frozenset of (nodeid, service_name, rref) tuples, one
404 for each active connection that provides the given SERVICE_NAME."""
406 def get_permuted_peers(service_name, key):
407 """Returns an ordered list of (peerid, rref) tuples, selecting from
408 the connections that provide SERVICE_NAME, using a hash-based
409 permutation keyed by KEY. This randomizes the service list in a
410 repeatable way, to distribute load over many peers.
414 class IURI(Interface):
415 def init_from_string(uri):
416 """Accept a string (as created by my to_string() method) and populate
417 this instance with its data. I am not normally called directly,
418 please use the module-level uri.from_string() function to convert
419 arbitrary URI strings into IURI-providing instances."""
422 """Return False if this URI be used to modify the data. Return True
423 if this URI cannot be used to modify the data."""
426 """Return True if the data can be modified by *somebody* (perhaps
427 someone who has a more powerful URI than this one)."""
430 """Return another IURI instance, which represents a read-only form of
431 this one. If is_readonly() is True, this returns self."""
433 def get_verify_cap():
434 """Return an instance that provides IVerifierURI, which can be used
435 to check on the availability of the file or directory, without
436 providing enough capabilities to actually read or modify the
437 contents. This may return None if the file does not need checking or
438 verification (e.g. LIT URIs).
442 """Return a string of printable ASCII characters, suitable for
443 passing into init_from_string."""
445 class IVerifierURI(Interface, IURI):
446 def init_from_string(uri):
447 """Accept a string (as created by my to_string() method) and populate
448 this instance with its data. I am not normally called directly,
449 please use the module-level uri.from_string() function to convert
450 arbitrary URI strings into IURI-providing instances."""
453 """Return a string of printable ASCII characters, suitable for
454 passing into init_from_string."""
456 class IDirnodeURI(Interface):
457 """I am a URI which represents a dirnode."""
460 class IFileURI(Interface):
461 """I am a URI which represents a filenode."""
463 """Return the length (in bytes) of the file that I represent."""
465 class IImmutableFileURI(IFileURI):
468 class IMutableFileURI(Interface):
469 """I am a URI which represents a mutable filenode."""
470 class IDirectoryURI(Interface):
472 class IReadonlyDirectoryURI(Interface):
475 class CannotPackUnknownNodeError(Exception):
476 """UnknownNodes (using filecaps from the future that we don't understand)
477 cannot yet be copied safely, so I refuse to copy them."""
479 class UnhandledCapTypeError(Exception):
480 """I recognize the cap/URI, but I cannot create an IFilesystemNode for
483 class IFilesystemNode(Interface):
486 Return the URI string that can be used by others to get access to
487 this node. If this node is read-only, the URI will only offer
488 read-only access. If this node is read-write, the URI will offer
491 If you have read-write access to a node and wish to share merely
492 read-only access with others, use get_readonly_uri().
495 def get_readonly_uri():
496 """Return the URI string that can be used by others to get read-only
497 access to this node. The result is a read-only URI, regardless of
498 whether this node is read-only or read-write.
500 If you have merely read-only access to this node, get_readonly_uri()
501 will return the same thing as get_uri().
504 def get_repair_cap():
505 """Return an IURI instance that can be used to repair the file, or
506 None if this node cannot be repaired (either because it is not
507 distributed, like a LIT file, or because the node does not represent
508 sufficient authority to create a repair-cap, like a read-only RSA
509 mutable file node [which cannot create the correct write-enablers]).
512 def get_verify_cap():
513 """Return an IVerifierURI instance that represents the
514 'verifiy/refresh capability' for this node. The holder of this
515 capability will be able to renew the lease for this node, protecting
516 it from garbage-collection. They will also be able to ask a server if
517 it holds a share for the file or directory.
520 def get_storage_index():
521 """Return a string with the (binary) storage index in use on this
522 download. This may be None if there is no storage index (i.e. LIT
526 """Return True if this reference provides mutable access to the given
527 file or directory (i.e. if you can modify it), or False if not. Note
528 that even if this reference is read-only, someone else may hold a
529 read-write reference to it."""
532 """Return True if this file or directory is mutable (by *somebody*,
533 not necessarily you), False if it is is immutable. Note that a file
534 might be mutable overall, but your reference to it might be
535 read-only. On the other hand, all references to an immutable file
536 will be read-only; there are no read-write references to an immutable
540 class IMutableFilesystemNode(IFilesystemNode):
543 class IFileNode(IFilesystemNode):
544 def download(target):
545 """Download the file's contents to a given IDownloadTarget"""
547 def download_to_data():
548 """Download the file's contents. Return a Deferred that fires
549 with those contents."""
552 """Return the length (in bytes) of the data this node represents."""
554 def read(consumer, offset=0, size=None):
555 """Download a portion (possibly all) of the file's contents, making
556 them available to the given IConsumer. Return a Deferred that fires
557 (with the consumer) when the consumer is unregistered (either because
558 the last byte has been given to it, or because the consumer threw an
559 exception during write(), possibly because it no longer wants to
560 receive data). The portion downloaded will start at 'offset' and
561 contain 'size' bytes (or the remainder of the file if size==None).
563 The consumer will be used in non-streaming mode: an IPullProducer
564 will be attached to it.
566 The consumer will not receive data right away: several network trips
567 must occur first. The order of events will be::
569 consumer.registerProducer(p, streaming)
570 (if streaming == False)::
571 consumer does p.resumeProducing()
573 consumer does p.resumeProducing()
574 consumer.write(data).. (repeat until all data is written)
575 consumer.unregisterProducer()
576 deferred.callback(consumer)
578 If a download error occurs, or an exception is raised by
579 consumer.registerProducer() or consumer.write(), I will call
580 consumer.unregisterProducer() and then deliver the exception via
581 deferred.errback(). To cancel the download, the consumer should call
582 p.stopProducing(), which will result in an exception being delivered
583 via deferred.errback().
585 A simple download-to-memory consumer example would look like this::
587 class MemoryConsumer:
588 implements(IConsumer)
592 def registerProducer(self, p, streaming):
593 assert streaming == False
596 def write(self, data):
597 self.chunks.append(data)
598 def unregisterProducer(self):
600 d = filenode.read(MemoryConsumer())
601 d.addCallback(lambda mc: "".join(mc.chunks))
606 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
607 """I provide access to a 'mutable file', which retains its identity
608 regardless of what contents are put in it.
610 The consistency-vs-availability problem means that there might be
611 multiple versions of a file present in the grid, some of which might be
612 unrecoverable (i.e. have fewer than 'k' shares). These versions are
613 loosely ordered: each has a sequence number and a hash, and any version
614 with seqnum=N was uploaded by a node which has seen at least one version
617 The 'servermap' (an instance of IMutableFileServerMap) is used to
618 describe the versions that are known to be present in the grid, and which
619 servers are hosting their shares. It is used to represent the 'state of
620 the world', and is used for this purpose by my test-and-set operations.
621 Downloading the contents of the mutable file will also return a
622 servermap. Uploading a new version into the mutable file requires a
623 servermap as input, and the semantics of the replace operation is
624 'replace the file with my new version if it looks like nobody else has
625 changed the file since my previous download'. Because the file is
626 distributed, this is not a perfect test-and-set operation, but it will do
627 its best. If the replace process sees evidence of a simultaneous write,
628 it will signal an UncoordinatedWriteError, so that the caller can take
632 Most readers will want to use the 'best' current version of the file, and
633 should use my 'download_best_version()' method.
635 To unconditionally replace the file, callers should use overwrite(). This
636 is the mode that user-visible mutable files will probably use.
638 To apply some delta to the file, call modify() with a callable modifier
639 function that can apply the modification that you want to make. This is
640 the mode that dirnodes will use, since most directory modification
641 operations can be expressed in terms of deltas to the directory state.
644 Three methods are available for users who need to perform more complex
645 operations. The first is get_servermap(), which returns an up-to-date
646 servermap using a specified mode. The second is download_version(), which
647 downloads a specific version (not necessarily the 'best' one). The third
648 is 'upload', which accepts new contents and a servermap (which must have
649 been updated with MODE_WRITE). The upload method will attempt to apply
650 the new contents as long as no other node has modified the file since the
651 servermap was updated. This might be useful to a caller who wants to
652 merge multiple versions into a single new one.
654 Note that each time the servermap is updated, a specific 'mode' is used,
655 which determines how many peers are queried. To use a servermap for my
656 replace() method, that servermap must have been updated in MODE_WRITE.
657 These modes are defined in allmydata.mutable.common, and consist of
658 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
659 allmydata/mutable/servermap.py for details about the differences.
661 Mutable files are currently limited in size (about 3.5MB max) and can
662 only be retrieved and updated all-at-once, as a single big string. Future
663 versions of our mutable files will remove this restriction.
666 def download_best_version():
667 """Download the 'best' available version of the file, meaning one of
668 the recoverable versions with the highest sequence number. If no
669 uncoordinated writes have occurred, and if enough shares are
670 available, then this will be the most recent version that has been
673 I update an internal servermap with MODE_READ, determine which
674 version of the file is indicated by
675 servermap.best_recoverable_version(), and return a Deferred that
676 fires with its contents. If no version is recoverable, the Deferred
677 will errback with UnrecoverableFileError.
680 def get_size_of_best_version():
681 """Find the size of the version that would be downloaded with
682 download_best_version(), without actually downloading the whole file.
684 I return a Deferred that fires with an integer.
687 def overwrite(new_contents):
688 """Unconditionally replace the contents of the mutable file with new
689 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
690 is only appropriate to use when the new contents of the file are
691 completely unrelated to the old ones, and you do not care about other
694 I return a Deferred that fires (with a PublishStatus object) when the
695 update has completed.
698 def modify(modifier_cb):
699 """Modify the contents of the file, by downloading the current
700 version, applying the modifier function (or bound method), then
701 uploading the new version. I return a Deferred that fires (with a
702 PublishStatus object) when the update is complete.
704 The modifier callable will be given three arguments: a string (with
705 the old contents), a 'first_time' boolean, and a servermap. As with
706 download_best_version(), the old contents will be from the best
707 recoverable version, but the modifier can use the servermap to make
708 other decisions (such as refusing to apply the delta if there are
709 multiple parallel versions, or if there is evidence of a newer
710 unrecoverable version). 'first_time' will be True the first time the
711 modifier is called, and False on any subsequent calls.
713 The callable should return a string with the new contents. The
714 callable must be prepared to be called multiple times, and must
715 examine the input string to see if the change that it wants to make
716 is already present in the old version. If it does not need to make
717 any changes, it can either return None, or return its input string.
719 If the modifier raises an exception, it will be returned in the
724 def get_servermap(mode):
725 """Return a Deferred that fires with an IMutableFileServerMap
726 instance, updated using the given mode.
729 def download_version(servermap, version):
730 """Download a specific version of the file, using the servermap
731 as a guide to where the shares are located.
733 I return a Deferred that fires with the requested contents, or
734 errbacks with UnrecoverableFileError. Note that a servermap which was
735 updated with MODE_ANYTHING or MODE_READ may not know about shares for
736 all versions (those modes stop querying servers as soon as they can
737 fulfil their goals), so you may want to use MODE_CHECK (which checks
738 everything) to get increased visibility.
741 def upload(new_contents, servermap):
742 """Replace the contents of the file with new ones. This requires a
743 servermap that was previously updated with MODE_WRITE.
745 I attempt to provide test-and-set semantics, in that I will avoid
746 modifying any share that is different than the version I saw in the
747 servermap. However, if another node is writing to the file at the
748 same time as me, I may manage to update some shares while they update
749 others. If I see any evidence of this, I will signal
750 UncoordinatedWriteError, and the file will be left in an inconsistent
751 state (possibly the version you provided, possibly the old version,
752 possibly somebody else's version, and possibly a mix of shares from
755 The recommended response to UncoordinatedWriteError is to either
756 return it to the caller (since they failed to coordinate their
757 writes), or to attempt some sort of recovery. It may be sufficient to
758 wait a random interval (with exponential backoff) and repeat your
759 operation. If I do not signal UncoordinatedWriteError, then I was
760 able to write the new version without incident.
762 I return a Deferred that fires (with a PublishStatus object) when the
763 publish has completed. I will update the servermap in-place with the
764 location of all new shares.
768 """Return this filenode's writekey, or None if the node does not have
769 write-capability. This may be used to assist with data structures
770 that need to make certain data available only to writers, such as the
771 read-write child caps in dirnodes. The recommended process is to have
772 reader-visible data be submitted to the filenode in the clear (where
773 it will be encrypted by the filenode using the readkey), but encrypt
774 writer-visible data using this writekey.
777 class NotEnoughSharesError(Exception):
778 """Download was unable to get enough shares, or upload was unable to
779 place 'shares_of_happiness' shares."""
781 class NoSharesError(Exception):
782 """Upload or Download was unable to get any shares at all."""
784 class UnableToFetchCriticalDownloadDataError(Exception):
785 """I was unable to fetch some piece of critical data which is supposed to
786 be identically present in all shares."""
788 class NoServersError(Exception):
789 """Upload wasn't given any servers to work with, usually indicating a
790 network or Introducer problem."""
792 class ExistingChildError(Exception):
793 """A directory node was asked to add or replace a child that already
794 exists, and overwrite= was set to False."""
796 class NoSuchChildError(Exception):
797 """A directory node was asked to fetch a child which does not exist."""
799 class IDirectoryNode(IMutableFilesystemNode):
800 """I represent a name-to-child mapping, holding the tahoe equivalent of a
801 directory. All child names are unicode strings, and all children are some
802 sort of IFilesystemNode (either files or subdirectories).
807 The dirnode ('1') URI returned by this method can be used in
808 set_uri() on a different directory ('2') to 'mount' a reference to
809 this directory ('1') under the other ('2'). This URI is just a
810 string, so it can be passed around through email or other out-of-band
814 def get_readonly_uri():
816 The dirnode ('1') URI returned by this method can be used in
817 set_uri() on a different directory ('2') to 'mount' a reference to
818 this directory ('1') under the other ('2'). This URI is just a
819 string, so it can be passed around through email or other out-of-band
824 """I return a Deferred that fires with a dictionary mapping child
825 name (a unicode string) to (node, metadata_dict) tuples, in which
826 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
827 is a dictionary of metadata."""
830 """I return a Deferred that fires with a boolean, True if there
831 exists a child of the given name, False if not. The child name must
832 be a unicode string."""
835 """I return a Deferred that fires with a specific named child node,
836 either an IFileNode or an IDirectoryNode. The child name must be a
837 unicode string. I raise NoSuchChildError if I do not have a child by
840 def get_metadata_for(name):
841 """I return a Deferred that fires with the metadata dictionary for a
842 specific named child node. This metadata is stored in the *edge*, not
843 in the child, so it is attached to the parent dirnode rather than the
844 child dir-or-file-node. The child name must be a unicode string. I
845 raise NoSuchChildError if I do not have a child by that name."""
847 def set_metadata_for(name, metadata):
848 """I replace any existing metadata for the named child with the new
849 metadata. The child name must be a unicode string. This metadata is
850 stored in the *edge*, not in the child, so it is attached to the
851 parent dirnode rather than the child dir-or-file-node. I return a
852 Deferred (that fires with this dirnode) when the operation is
853 complete. I raise NoSuchChildError if I do not have a child by that
856 def get_child_at_path(path):
857 """Transform a child path into an IDirectoryNode or IFileNode.
859 I perform a recursive series of 'get' operations to find the named
860 descendant node. I return a Deferred that fires with the node, or
861 errbacks with NoSuchChildError if the node could not be found.
863 The path can be either a single string (slash-separated) or a list of
864 path-name elements. All elements must be unicode strings.
867 def get_child_and_metadata_at_path(path):
868 """Transform a child path into an IDirectoryNode/IFileNode and
871 I am like get_child_at_path(), but my Deferred fires with a tuple of
872 (node, metadata). The metadata comes from the last edge. If the path
873 is empty, the metadata will be an empty dictionary.
876 def set_uri(name, child_uri, metadata=None, overwrite=True):
877 """I add a child (by URI) at the specific name. I return a Deferred
878 that fires when the operation finishes. If overwrite= is True, I will
879 replace any existing child of the same name, otherwise an existing
880 child will cause me to return ExistingChildError. The child name must
883 The child_uri could be for a file, or for a directory (either
884 read-write or read-only, using a URI that came from get_uri() ).
886 If metadata= is provided, I will use it as the metadata for the named
887 edge. This will replace any existing metadata. If metadata= is left
888 as the default value of None, I will set ['mtime'] to the current
889 time, and I will set ['ctime'] to the current time if there was not
890 already a child by this name present. This roughly matches the
891 ctime/mtime semantics of traditional filesystems.
893 If this directory node is read-only, the Deferred will errback with a
896 def set_children(entries, overwrite=True):
897 """Add multiple (name, child_uri) pairs (or (name, child_uri,
898 metadata) triples) to a directory node. Returns a Deferred that fires
899 (with None) when the operation finishes. This is equivalent to
900 calling set_uri() multiple times, but is much more efficient. All
901 child names must be unicode strings.
904 def set_node(name, child, metadata=None, overwrite=True):
905 """I add a child at the specific name. I return a Deferred that fires
906 when the operation finishes. This Deferred will fire with the child
907 node that was just added. I will replace any existing child of the
908 same name. The child name must be a unicode string. The 'child'
909 instance must be an instance providing IDirectoryNode or IFileNode.
911 If metadata= is provided, I will use it as the metadata for the named
912 edge. This will replace any existing metadata. If metadata= is left
913 as the default value of None, I will set ['mtime'] to the current
914 time, and I will set ['ctime'] to the current time if there was not
915 already a child by this name present. This roughly matches the
916 ctime/mtime semantics of traditional filesystems.
918 If this directory node is read-only, the Deferred will errback with a
921 def set_nodes(entries, overwrite=True):
922 """Add multiple (name, child_node) pairs (or (name, child_node,
923 metadata) triples) to a directory node. Returns a Deferred that fires
924 (with None) when the operation finishes. This is equivalent to
925 calling set_node() multiple times, but is much more efficient. All
926 child names must be unicode strings."""
929 def add_file(name, uploadable, metadata=None, overwrite=True):
930 """I upload a file (using the given IUploadable), then attach the
931 resulting FileNode to the directory at the given name. I set metadata
932 the same way as set_uri and set_node. The child name must be a
935 I return a Deferred that fires (with the IFileNode of the uploaded
936 file) when the operation completes."""
939 """I remove the child at the specific name. I return a Deferred that
940 fires when the operation finishes. The child name must be a unicode
941 string. I raise NoSuchChildError if I do not have a child by that
944 def create_empty_directory(name, overwrite=True):
945 """I create and attach an empty directory at the given name. The
946 child name must be a unicode string. I return a Deferred that fires
947 when the operation finishes."""
949 def move_child_to(current_child_name, new_parent, new_child_name=None,
951 """I take one of my children and move them to a new parent. The child
952 is referenced by name. On the new parent, the child will live under
953 'new_child_name', which defaults to 'current_child_name'. TODO: what
954 should we do about metadata? I return a Deferred that fires when the
955 operation finishes. The child name must be a unicode string. I raise
956 NoSuchChildError if I do not have a child by that name."""
958 def build_manifest():
959 """I generate a table of everything reachable from this directory.
960 I also compute deep-stats as described below.
962 I return a Monitor. The Monitor's results will be a dictionary with
965 res['manifest']: a list of (path, cap) tuples for all nodes
966 (directories and files) reachable from this one.
967 'path' will be a tuple of unicode strings. The
968 origin dirnode will be represented by an empty path
970 res['verifycaps']: a list of (printable) verifycap strings, one for
971 each reachable non-LIT node. This is a set:
972 it will contain no duplicates.
973 res['storage-index']: a list of (base32) storage index strings,
974 one for each reachable non-LIT node. This is
975 a set: it will contain no duplicates.
976 res['stats']: a dictionary, the same that is generated by
977 start_deep_stats() below.
979 The Monitor will also have an .origin_si attribute with the (binary)
980 storage index of the starting point.
983 def start_deep_stats():
984 """Return a Monitor, examining all nodes (directories and files)
985 reachable from this one. The Monitor's results will be a dictionary
986 with the following keys::
988 count-immutable-files: count of how many CHK files are in the set
989 count-mutable-files: same, for mutable files (does not include
991 count-literal-files: same, for LIT files
992 count-files: sum of the above three
994 count-directories: count of directories
996 size-immutable-files: total bytes for all CHK files in the set
997 size-mutable-files (TODO): same, for current version of all mutable
998 files, does not include directories
999 size-literal-files: same, for LIT files
1000 size-directories: size of mutable files used by directories
1002 largest-directory: number of bytes in the largest directory
1003 largest-directory-children: number of children in the largest
1005 largest-immutable-file: number of bytes in the largest CHK file
1007 size-mutable-files is not yet implemented, because it would involve
1008 even more queries than deep_stats does.
1010 The Monitor will also have an .origin_si attribute with the (binary)
1011 storage index of the starting point.
1013 This operation will visit every directory node underneath this one,
1014 and can take a long time to run. On a typical workstation with good
1015 bandwidth, this can examine roughly 15 directories per second (and
1016 takes several minutes of 100% CPU for ~1700 directories).
1019 class ICodecEncoder(Interface):
1020 def set_params(data_size, required_shares, max_shares):
1021 """Set up the parameters of this encoder.
1023 This prepares the encoder to perform an operation that converts a
1024 single block of data into a number of shares, such that a future
1025 ICodecDecoder can use a subset of these shares to recover the
1026 original data. This operation is invoked by calling encode(). Once
1027 the encoding parameters are set up, the encode operation can be
1028 invoked multiple times.
1030 set_params() prepares the encoder to accept blocks of input data that
1031 are exactly 'data_size' bytes in length. The encoder will be prepared
1032 to produce 'max_shares' shares for each encode() operation (although
1033 see the 'desired_share_ids' to use less CPU). The encoding math will
1034 be chosen such that the decoder can get by with as few as
1035 'required_shares' of these shares and still reproduce the original
1036 data. For example, set_params(1000, 5, 5) offers no redundancy at
1037 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1039 Numerical Restrictions: 'data_size' is required to be an integral
1040 multiple of 'required_shares'. In general, the caller should choose
1041 required_shares and max_shares based upon their reliability
1042 requirements and the number of peers available (the total storage
1043 space used is roughly equal to max_shares*data_size/required_shares),
1044 then choose data_size to achieve the memory footprint desired (larger
1045 data_size means more efficient operation, smaller data_size means
1046 smaller memory footprint).
1048 In addition, 'max_shares' must be equal to or greater than
1049 'required_shares'. Of course, setting them to be equal causes
1050 encode() to degenerate into a particularly slow form of the 'split'
1053 See encode() for more details about how these parameters are used.
1055 set_params() must be called before any other ICodecEncoder methods
1060 """Return the 3-tuple of data_size, required_shares, max_shares"""
1062 def get_encoder_type():
1063 """Return a short string that describes the type of this encoder.
1065 There is required to be a global table of encoder classes. This method
1066 returns an index into this table; the value at this index is an
1067 encoder class, and this encoder is an instance of that class.
1070 def get_block_size():
1071 """Return the length of the shares that encode() will produce.
1074 def encode_proposal(data, desired_share_ids=None):
1075 """Encode some data.
1077 'data' must be a string (or other buffer object), and len(data) must
1078 be equal to the 'data_size' value passed earlier to set_params().
1080 This will return a Deferred that will fire with two lists. The first
1081 is a list of shares, each of which is a string (or other buffer
1082 object) such that len(share) is the same as what get_share_size()
1083 returned earlier. The second is a list of shareids, in which each is
1084 an integer. The lengths of the two lists will always be equal to each
1085 other. The user should take care to keep each share closely
1086 associated with its shareid, as one is useless without the other.
1088 The length of this output list will normally be the same as the value
1089 provided to the 'max_shares' parameter of set_params(). This may be
1090 different if 'desired_share_ids' is provided.
1092 'desired_share_ids', if provided, is required to be a sequence of
1093 ints, each of which is required to be >= 0 and < max_shares. If not
1094 provided, encode() will produce 'max_shares' shares, as if
1095 'desired_share_ids' were set to range(max_shares). You might use this
1096 if you initially thought you were going to use 10 peers, started
1097 encoding, and then two of the peers dropped out: you could use
1098 desired_share_ids= to skip the work (both memory and CPU) of
1099 producing shares for the peers which are no longer available.
1103 def encode(inshares, desired_share_ids=None):
1104 """Encode some data. This may be called multiple times. Each call is
1107 inshares is a sequence of length required_shares, containing buffers
1108 (i.e. strings), where each buffer contains the next contiguous
1109 non-overlapping segment of the input data. Each buffer is required to
1110 be the same length, and the sum of the lengths of the buffers is
1111 required to be exactly the data_size promised by set_params(). (This
1112 implies that the data has to be padded before being passed to
1113 encode(), unless of course it already happens to be an even multiple
1114 of required_shares in length.)
1116 ALSO: the requirement to break up your data into 'required_shares'
1117 chunks before calling encode() feels a bit surprising, at least from
1118 the point of view of a user who doesn't know how FEC works. It feels
1119 like an implementation detail that has leaked outside the
1120 abstraction barrier. Can you imagine a use case in which the data to
1121 be encoded might already be available in pre-segmented chunks, such
1122 that it is faster or less work to make encode() take a list rather
1123 than splitting a single string?
1125 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1126 is supposed to *produce* shares, so what it *accepts* should be
1127 something other than shares. Other places in this interface use the
1128 word 'data' for that-which-is-not-shares.. maybe we should use that
1131 'desired_share_ids', if provided, is required to be a sequence of
1132 ints, each of which is required to be >= 0 and < max_shares. If not
1133 provided, encode() will produce 'max_shares' shares, as if
1134 'desired_share_ids' were set to range(max_shares). You might use this
1135 if you initially thought you were going to use 10 peers, started
1136 encoding, and then two of the peers dropped out: you could use
1137 desired_share_ids= to skip the work (both memory and CPU) of
1138 producing shares for the peers which are no longer available.
1140 For each call, encode() will return a Deferred that fires with two
1141 lists, one containing shares and the other containing the shareids.
1142 The get_share_size() method can be used to determine the length of
1143 the share strings returned by encode(). Each shareid is a small
1144 integer, exactly as passed into 'desired_share_ids' (or
1145 range(max_shares), if desired_share_ids was not provided).
1147 The shares and their corresponding shareids are required to be kept
1148 together during storage and retrieval. Specifically, the share data is
1149 useless by itself: the decoder needs to be told which share is which
1150 by providing it with both the shareid and the actual share data.
1152 This function will allocate an amount of memory roughly equal to::
1154 (max_shares - required_shares) * get_share_size()
1156 When combined with the memory that the caller must allocate to
1157 provide the input data, this leads to a memory footprint roughly
1158 equal to the size of the resulting encoded shares (i.e. the expansion
1159 factor times the size of the input segment).
1164 # returning a list of (shareidN,shareN) tuples instead of a pair of
1165 # lists (shareids..,shares..). Brian thought the tuples would
1166 # encourage users to keep the share and shareid together throughout
1167 # later processing, Zooko pointed out that the code to iterate
1168 # through two lists is not really more complicated than using a list
1169 # of tuples and there's also a performance improvement
1171 # having 'data_size' not required to be an integral multiple of
1172 # 'required_shares'. Doing this would require encode() to perform
1173 # padding internally, and we'd prefer to have any padding be done
1174 # explicitly by the caller. Yes, it is an abstraction leak, but
1175 # hopefully not an onerous one.
1178 class ICodecDecoder(Interface):
1179 def set_params(data_size, required_shares, max_shares):
1180 """Set the params. They have to be exactly the same ones that were
1181 used for encoding."""
1183 def get_needed_shares():
1184 """Return the number of shares needed to reconstruct the data.
1185 set_params() is required to be called before this."""
1187 def decode(some_shares, their_shareids):
1188 """Decode a partial list of shares into data.
1190 'some_shares' is required to be a sequence of buffers of sharedata, a
1191 subset of the shares returned by ICodecEncode.encode(). Each share is
1192 required to be of the same length. The i'th element of their_shareids
1193 is required to be the shareid of the i'th buffer in some_shares.
1195 This returns a Deferred which fires with a sequence of buffers. This
1196 sequence will contain all of the segments of the original data, in
1197 order. The sum of the lengths of all of the buffers will be the
1198 'data_size' value passed into the original ICodecEncode.set_params()
1199 call. To get back the single original input block of data, use
1200 ''.join(output_buffers), or you may wish to simply write them in
1201 order to an output file.
1203 Note that some of the elements in the result sequence may be
1204 references to the elements of the some_shares input sequence. In
1205 particular, this means that if those share objects are mutable (e.g.
1206 arrays) and if they are changed, then both the input (the
1207 'some_shares' parameter) and the output (the value given when the
1208 deferred is triggered) will change.
1210 The length of 'some_shares' is required to be exactly the value of
1211 'required_shares' passed into the original ICodecEncode.set_params()
1215 class IEncoder(Interface):
1216 """I take an object that provides IEncryptedUploadable, which provides
1217 encrypted data, and a list of shareholders. I then encode, hash, and
1218 deliver shares to those shareholders. I will compute all the necessary
1219 Merkle hash trees that are necessary to validate the crypttext that
1220 eventually comes back from the shareholders. I provide the URI Extension
1221 Block Hash, and the encoding parameters, both of which must be included
1224 I do not choose shareholders, that is left to the IUploader. I must be
1225 given a dict of RemoteReferences to storage buckets that are ready and
1226 willing to receive data.
1230 """Specify the number of bytes that will be encoded. This must be
1231 peformed before get_serialized_params() can be called.
1233 def set_params(params):
1234 """Override the default encoding parameters. 'params' is a tuple of
1235 (k,d,n), where 'k' is the number of required shares, 'd' is the
1236 shares_of_happiness, and 'n' is the total number of shares that will
1239 Encoding parameters can be set in three ways. 1: The Encoder class
1240 provides defaults (3/7/10). 2: the Encoder can be constructed with
1241 an 'options' dictionary, in which the
1242 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1243 set_params((k,d,n)) can be called.
1245 If you intend to use set_params(), you must call it before
1246 get_share_size or get_param are called.
1249 def set_encrypted_uploadable(u):
1250 """Provide a source of encrypted upload data. 'u' must implement
1251 IEncryptedUploadable.
1253 When this is called, the IEncryptedUploadable will be queried for its
1254 length and the storage_index that should be used.
1256 This returns a Deferred that fires with this Encoder instance.
1258 This must be performed before start() can be called.
1261 def get_param(name):
1262 """Return an encoding parameter, by name.
1264 'storage_index': return a string with the (16-byte truncated SHA-256
1265 hash) storage index to which these shares should be
1268 'share_counts': return a tuple describing how many shares are used:
1269 (needed_shares, shares_of_happiness, total_shares)
1271 'num_segments': return an int with the number of segments that
1274 'segment_size': return an int with the size of each segment.
1276 'block_size': return the size of the individual blocks that will
1277 be delivered to a shareholder's put_block() method. By
1278 knowing this, the shareholder will be able to keep all
1279 blocks in a single file and still provide random access
1280 when reading them. # TODO: can we avoid exposing this?
1282 'share_size': an int with the size of the data that will be stored
1283 on each shareholder. This is aggregate amount of data
1284 that will be sent to the shareholder, summed over all
1285 the put_block() calls I will ever make. It is useful to
1286 determine this size before asking potential
1287 shareholders whether they will grant a lease or not,
1288 since their answers will depend upon how much space we
1289 need. TODO: this might also include some amount of
1290 overhead, like the size of all the hashes. We need to
1291 decide whether this is useful or not.
1293 'serialized_params': a string with a concise description of the
1294 codec name and its parameters. This may be passed
1295 into the IUploadable to let it make sure that
1296 the same file encoded with different parameters
1297 will result in different storage indexes.
1299 Once this is called, set_size() and set_params() may not be called.
1302 def set_shareholders(shareholders):
1303 """Tell the encoder where to put the encoded shares. 'shareholders'
1304 must be a dictionary that maps share number (an integer ranging from
1305 0 to n-1) to an instance that provides IStorageBucketWriter. This
1306 must be performed before start() can be called."""
1309 """Begin the encode/upload process. This involves reading encrypted
1310 data from the IEncryptedUploadable, encoding it, uploading the shares
1311 to the shareholders, then sending the hash trees.
1313 set_encrypted_uploadable() and set_shareholders() must be called
1314 before this can be invoked.
1316 This returns a Deferred that fires with a verify cap when the upload
1317 process is complete. The verifycap, plus the encryption key, is
1318 sufficient to construct the read cap.
1321 class IDecoder(Interface):
1322 """I take a list of shareholders and some setup information, then
1323 download, validate, decode, and decrypt data from them, writing the
1324 results to an output file.
1326 I do not locate the shareholders, that is left to the IDownloader. I must
1327 be given a dict of RemoteReferences to storage buckets that are ready to
1332 """I take a file-like object (providing write and close) to which all
1333 the plaintext data will be written.
1335 TODO: producer/consumer . Maybe write() should return a Deferred that
1336 indicates when it will accept more data? But probably having the
1337 IDecoder be a producer is easier to glue to IConsumer pieces.
1340 def set_shareholders(shareholders):
1341 """I take a dictionary that maps share identifiers (small integers)
1342 to RemoteReferences that provide RIBucketReader. This must be called
1346 """I start the download. This process involves retrieving data and
1347 hash chains from the shareholders, using the hashes to validate the
1348 data, decoding the shares into segments, decrypting the segments,
1349 then writing the resulting plaintext to the output file.
1351 I return a Deferred that will fire (with self) when the download is
1355 class IDownloadTarget(Interface):
1356 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1357 # will register itself as a producer. This allows the target to invoke
1358 # downloader.pauseProducing, resumeProducing, and stopProducing.
1360 """Called before any calls to write() or close(). If an error
1361 occurs before any data is available, fail() may be called without
1362 a previous call to open().
1364 'size' is the length of the file being downloaded, in bytes."""
1367 """Output some data to the target."""
1369 """Inform the target that there is no more data to be written."""
1371 """fail() is called to indicate that the download has failed. 'why'
1372 is a Failure object indicating what went wrong. No further methods
1373 will be invoked on the IDownloadTarget after fail()."""
1374 def register_canceller(cb):
1375 """The CiphertextDownloader uses this to register a no-argument function
1376 that the target can call to cancel the download. Once this canceller
1377 is invoked, no further calls to write() or close() will be made."""
1379 """When the CiphertextDownloader is done, this finish() function will be
1380 called. Whatever it returns will be returned to the invoker of
1381 Downloader.download.
1383 # The following methods are just because that target might be a
1384 # repairer.DownUpConnector, and just because the current CHKUpload object
1385 # expects to find the storage index and encoding parameters in its
1387 def set_storageindex(storageindex):
1388 """ Set the storage index. """
1389 def set_encodingparams(encodingparams):
1390 """ Set the encoding parameters. """
1392 class IDownloader(Interface):
1393 def download(uri, target):
1394 """Perform a CHK download, sending the data to the given target.
1395 'target' must provide IDownloadTarget.
1397 Returns a Deferred that fires (with the results of target.finish)
1398 when the download is finished, or errbacks if something went wrong."""
1400 class IEncryptedUploadable(Interface):
1401 def set_upload_status(upload_status):
1402 """Provide an IUploadStatus object that should be filled with status
1403 information. The IEncryptedUploadable is responsible for setting
1404 key-determination progress ('chk'), size, storage_index, and
1405 ciphertext-fetch progress. It may delegate some of this
1406 responsibility to others, in particular to the IUploadable."""
1409 """This behaves just like IUploadable.get_size()."""
1411 def get_all_encoding_parameters():
1412 """Return a Deferred that fires with a tuple of
1413 (k,happy,n,segment_size). The segment_size will be used as-is, and
1414 must match the following constraints: it must be a multiple of k, and
1415 it shouldn't be unreasonably larger than the file size (if
1416 segment_size is larger than filesize, the difference must be stored
1419 This usually passes through to the IUploadable method of the same
1422 The encoder strictly obeys the values returned by this method. To
1423 make an upload use non-default encoding parameters, you must arrange
1424 to control the values that this method returns.
1427 def get_storage_index():
1428 """Return a Deferred that fires with a 16-byte storage index.
1431 def read_encrypted(length, hash_only):
1432 """This behaves just like IUploadable.read(), but returns crypttext
1433 instead of plaintext. If hash_only is True, then this discards the
1434 data (and returns an empty list); this improves efficiency when
1435 resuming an interrupted upload (where we need to compute the
1436 plaintext hashes, but don't need the redundant encrypted data)."""
1438 def get_plaintext_hashtree_leaves(first, last, num_segments):
1439 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1440 plaintext segments, i.e. get the tagged hashes of the given segments.
1441 The segment size is expected to be generated by the
1442 IEncryptedUploadable before any plaintext is read or ciphertext
1443 produced, so that the segment hashes can be generated with only a
1446 This returns a Deferred which fires with a sequence of hashes, using:
1448 tuple(segment_hashes[first:last])
1450 'num_segments' is used to assert that the number of segments that the
1451 IEncryptedUploadable handled matches the number of segments that the
1452 encoder was expecting.
1454 This method must not be called until the final byte has been read
1455 from read_encrypted(). Once this method is called, read_encrypted()
1456 can never be called again.
1459 def get_plaintext_hash():
1460 """OBSOLETE; Get the hash of the whole plaintext.
1462 This returns a Deferred which fires with a tagged SHA-256 hash of the
1463 whole plaintext, obtained from hashutil.plaintext_hash(data).
1467 """Just like IUploadable.close()."""
1469 class IUploadable(Interface):
1470 def set_upload_status(upload_status):
1471 """Provide an IUploadStatus object that should be filled with status
1472 information. The IUploadable is responsible for setting
1473 key-determination progress ('chk')."""
1475 def set_default_encoding_parameters(params):
1476 """Set the default encoding parameters, which must be a dict mapping
1477 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1478 'max_segment_size'. These might have an influence on the final
1479 encoding parameters returned by get_all_encoding_parameters(), if the
1480 Uploadable doesn't have more specific preferences.
1482 This call is optional: if it is not used, the Uploadable will use
1483 some built-in defaults. If used, this method must be called before
1484 any other IUploadable methods to have any effect.
1488 """Return a Deferred that will fire with the length of the data to be
1489 uploaded, in bytes. This will be called before the data is actually
1490 used, to compute encoding parameters.
1493 def get_all_encoding_parameters():
1494 """Return a Deferred that fires with a tuple of
1495 (k,happy,n,segment_size). The segment_size will be used as-is, and
1496 must match the following constraints: it must be a multiple of k, and
1497 it shouldn't be unreasonably larger than the file size (if
1498 segment_size is larger than filesize, the difference must be stored
1501 The relative values of k and n allow some IUploadables to request
1502 better redundancy than others (in exchange for consuming more space
1505 Larger values of segment_size reduce hash overhead, while smaller
1506 values reduce memory footprint and cause data to be delivered in
1507 smaller pieces (which may provide a smoother and more predictable
1508 download experience).
1510 The encoder strictly obeys the values returned by this method. To
1511 make an upload use non-default encoding parameters, you must arrange
1512 to control the values that this method returns. One way to influence
1513 them may be to call set_encoding_parameters() before calling
1514 get_all_encoding_parameters().
1517 def get_encryption_key():
1518 """Return a Deferred that fires with a 16-byte AES key. This key will
1519 be used to encrypt the data. The key will also be hashed to derive
1522 Uploadables which want to achieve convergence should hash their file
1523 contents and the serialized_encoding_parameters to form the key
1524 (which of course requires a full pass over the data). Uploadables can
1525 use the upload.ConvergentUploadMixin class to achieve this
1528 Uploadables which do not care about convergence (or do not wish to
1529 make multiple passes over the data) can simply return a
1530 strongly-random 16 byte string.
1532 get_encryption_key() may be called multiple times: the IUploadable is
1533 required to return the same value each time.
1537 """Return a Deferred that fires with a list of strings (perhaps with
1538 only a single element) which, when concatenated together, contain the
1539 next 'length' bytes of data. If EOF is near, this may provide fewer
1540 than 'length' bytes. The total number of bytes provided by read()
1541 before it signals EOF must equal the size provided by get_size().
1543 If the data must be acquired through multiple internal read
1544 operations, returning a list instead of a single string may help to
1545 reduce string copies.
1547 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1548 so a 10kB file means length=3kB, 100kB file means length=30kB,
1549 and >=1MB file means length=300kB.
1551 This method provides for a single full pass through the data. Later
1552 use cases may desire multiple passes or access to only parts of the
1553 data (such as a mutable file making small edits-in-place). This API
1554 will be expanded once those use cases are better understood.
1558 """The upload is finished, and whatever filehandle was in use may be
1561 class IUploadResults(Interface):
1562 """I am returned by upload() methods. I contain a number of public
1563 attributes which can be read to determine the results of the upload. Some
1564 of these are functional, some are timing information. All of these may be
1567 .file_size : the size of the file, in bytes
1568 .uri : the CHK read-cap for the file
1569 .ciphertext_fetched : how many bytes were fetched by the helper
1570 .sharemap: dict mapping share identifier to set of serverids
1571 (binary strings). This indicates which servers were given
1572 which shares. For immutable files, the shareid is an
1573 integer (the share number, from 0 to N-1). For mutable
1574 files, it is a string of the form 'seq%d-%s-sh%d',
1575 containing the sequence number, the roothash, and the
1577 .servermap : dict mapping server peerid to a set of share numbers
1578 .timings : dict of timing information, mapping name to seconds (float)
1579 total : total upload time, start to finish
1580 storage_index : time to compute the storage index
1581 peer_selection : time to decide which peers will be used
1582 contacting_helper : initial helper query to upload/no-upload decision
1583 existence_check : helper pre-upload existence check
1584 helper_total : initial helper query to helper finished pushing
1585 cumulative_fetch : helper waiting for ciphertext requests
1586 total_fetch : helper start to last ciphertext response
1587 cumulative_encoding : just time spent in zfec
1588 cumulative_sending : just time spent waiting for storage servers
1589 hashes_and_close : last segment push to shareholder close
1590 total_encode_and_push : first encode to shareholder close
1594 class IDownloadResults(Interface):
1595 """I am created internally by download() methods. I contain a number of
1596 public attributes which contain details about the download process.::
1598 .file_size : the size of the file, in bytes
1599 .servers_used : set of server peerids that were used during download
1600 .server_problems : dict mapping server peerid to a problem string. Only
1601 servers that had problems (bad hashes, disconnects)
1603 .servermap : dict mapping server peerid to a set of share numbers. Only
1604 servers that had any shares are listed here.
1605 .timings : dict of timing information, mapping name to seconds (float)
1606 peer_selection : time to ask servers about shares
1607 servers_peer_selection : dict of peerid to DYHB-query time
1608 uri_extension : time to fetch a copy of the URI extension block
1609 hashtrees : time to fetch the hash trees
1610 segments : time to fetch, decode, and deliver segments
1611 cumulative_fetch : time spent waiting for storage servers
1612 cumulative_decode : just time spent in zfec
1613 cumulative_decrypt : just time spent in decryption
1614 total : total download time, start to finish
1615 fetch_per_server : dict of peerid to list of per-segment fetch times
1619 class IUploader(Interface):
1620 def upload(uploadable):
1621 """Upload the file. 'uploadable' must impement IUploadable. This
1622 returns a Deferred which fires with an UploadResults instance, from
1623 which the URI of the file can be obtained as results.uri ."""
1625 def upload_ssk(write_capability, new_version, uploadable):
1626 """TODO: how should this work?"""
1628 class ICheckable(Interface):
1629 def check(monitor, verify=False, add_lease=False):
1630 """Check upon my health, optionally repairing any problems.
1632 This returns a Deferred that fires with an instance that provides
1633 ICheckResults, or None if the object is non-distributed (i.e. LIT
1636 The monitor will be checked periodically to see if the operation has
1637 been cancelled. If so, no new queries will be sent, and the Deferred
1638 will fire (with a OperationCancelledError) immediately.
1640 Filenodes and dirnodes (which provide IFilesystemNode) are also
1641 checkable. Instances that represent verifier-caps will be checkable
1642 but not downloadable. Some objects (like LIT files) do not actually
1643 live in the grid, and their checkers return None (non-distributed
1644 files are always healthy).
1646 If verify=False, a relatively lightweight check will be performed: I
1647 will ask all servers if they have a share for me, and I will believe
1648 whatever they say. If there are at least N distinct shares on the
1649 grid, my results will indicate r.is_healthy()==True. This requires a
1650 roundtrip to each server, but does not transfer very much data, so
1651 the network bandwidth is fairly low.
1653 If verify=True, a more resource-intensive check will be performed:
1654 every share will be downloaded, and the hashes will be validated on
1655 every bit. I will ignore any shares that failed their hash checks. If
1656 there are at least N distinct valid shares on the grid, my results
1657 will indicate r.is_healthy()==True. This requires N/k times as much
1658 download bandwidth (and server disk IO) as a regular download. If a
1659 storage server is holding a corrupt share, or is experiencing memory
1660 failures during retrieval, or is malicious or buggy, then
1661 verification will detect the problem, but checking will not.
1663 If add_lease=True, I will ensure that an up-to-date lease is present
1664 on each share. The lease secrets will be derived from by node secret
1665 (in BASEDIR/private/secret), so either I will add a new lease to the
1666 share, or I will merely renew the lease that I already had. In a
1667 future version of the storage-server protocol (once Accounting has
1668 been implemented), there may be additional options here to define the
1669 kind of lease that is obtained (which account number to claim, etc).
1671 TODO: any problems seen during checking will be reported to the
1672 health-manager.furl, a centralized object which is responsible for
1673 figuring out why files are unhealthy so corrective action can be
1677 def check_and_repair(monitor, verify=False, add_lease=False):
1678 """Like check(), but if the file/directory is not healthy, attempt to
1681 Any non-healthy result will cause an immediate repair operation, to
1682 generate and upload new shares. After repair, the file will be as
1683 healthy as we can make it. Details about what sort of repair is done
1684 will be put in the check-and-repair results. The Deferred will not
1685 fire until the repair is complete.
1687 This returns a Deferred which fires with an instance of
1688 ICheckAndRepairResults."""
1690 class IDeepCheckable(Interface):
1691 def start_deep_check(verify=False, add_lease=False):
1692 """Check upon the health of me and everything I can reach.
1694 This is a recursive form of check(), useable only on dirnodes.
1696 I return a Monitor, with results that are an IDeepCheckResults
1699 TODO: If any of the directories I traverse are unrecoverable, the
1700 Monitor will report failure. If any of the files I check upon are
1701 unrecoverable, those problems will be reported in the
1702 IDeepCheckResults as usual, and the Monitor will not report a
1706 def start_deep_check_and_repair(verify=False, add_lease=False):
1707 """Check upon the health of me and everything I can reach. Repair
1708 anything that isn't healthy.
1710 This is a recursive form of check_and_repair(), useable only on
1713 I return a Monitor, with results that are an
1714 IDeepCheckAndRepairResults object.
1716 TODO: If any of the directories I traverse are unrecoverable, the
1717 Monitor will report failure. If any of the files I check upon are
1718 unrecoverable, those problems will be reported in the
1719 IDeepCheckResults as usual, and the Monitor will not report a
1723 class ICheckResults(Interface):
1724 """I contain the detailed results of a check/verify operation.
1727 def get_storage_index():
1728 """Return a string with the (binary) storage index."""
1729 def get_storage_index_string():
1730 """Return a string with the (printable) abbreviated storage index."""
1732 """Return the (string) URI of the object that was checked."""
1735 """Return a boolean, True if the file/dir is fully healthy, False if
1736 it is damaged in any way. Non-distributed LIT files always return
1739 def is_recoverable():
1740 """Return a boolean, True if the file/dir can be recovered, False if
1741 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1742 files always return True."""
1744 def needs_rebalancing():
1745 """Return a boolean, True if the file/dir's reliability could be
1746 improved by moving shares to new servers. Non-distributed LIT files
1747 always return False."""
1751 """Return a dictionary that describes the state of the file/dir. LIT
1752 files always return an empty dictionary. Normal files and directories
1753 return a dictionary with the following keys (note that these use
1754 binary strings rather than base32-encoded ones) (also note that for
1755 mutable files, these counts are for the 'best' version):
1757 count-shares-good: the number of distinct good shares that were found
1758 count-shares-needed: 'k', the number of shares required for recovery
1759 count-shares-expected: 'N', the number of total shares generated
1760 count-good-share-hosts: the number of distinct storage servers with
1761 good shares. If this number is less than
1762 count-shares-good, then some shares are
1763 doubled up, increasing the correlation of
1764 failures. This indicates that one or more
1765 shares should be moved to an otherwise unused
1766 server, if one is available.
1767 count-corrupt-shares: the number of shares with integrity failures
1768 list-corrupt-shares: a list of 'share locators', one for each share
1769 that was found to be corrupt. Each share
1770 locator is a list of (serverid, storage_index,
1772 count-incompatible-shares: the number of shares which are of a share
1773 format unknown to this checker
1774 list-incompatible-shares: a list of 'share locators', one for each
1775 share that was found to be of an unknown
1776 format. Each share locator is a list of
1777 (serverid, storage_index, sharenum).
1778 servers-responding: list of (binary) storage server identifiers,
1779 one for each server which responded to the share
1780 query (even if they said they didn't have
1781 shares, and even if they said they did have
1782 shares but then didn't send them when asked, or
1783 dropped the connection, or returned a Failure,
1784 and even if they said they did have shares and
1785 sent incorrect ones when asked)
1786 sharemap: dict mapping share identifier to list of serverids
1787 (binary strings). This indicates which servers are holding
1788 which shares. For immutable files, the shareid is an
1789 integer (the share number, from 0 to N-1). For mutable
1790 files, it is a string of the form 'seq%d-%s-sh%d',
1791 containing the sequence number, the roothash, and the
1794 The following keys are most relevant for mutable files, but immutable
1795 files will provide sensible values too::
1797 count-wrong-shares: the number of shares for versions other than the
1798 'best' one (which is defined as being the
1799 recoverable version with the highest sequence
1800 number, then the highest roothash). These are
1801 either leftover shares from an older version
1802 (perhaps on a server that was offline when an
1803 update occurred), shares from an unrecoverable
1804 newer version, or shares from an alternate
1805 current version that results from an
1806 uncoordinated write collision. For a healthy
1807 file, this will equal 0.
1809 count-recoverable-versions: the number of recoverable versions of
1810 the file. For a healthy file, this will
1813 count-unrecoverable-versions: the number of unrecoverable versions
1814 of the file. For a healthy file, this
1820 """Return a string with a brief (one-line) summary of the results."""
1823 """Return a list of strings with more detailed results."""
1825 class ICheckAndRepairResults(Interface):
1826 """I contain the detailed results of a check/verify/repair operation.
1828 The IFilesystemNode.check()/verify()/repair() methods all return
1829 instances that provide ICheckAndRepairResults.
1832 def get_storage_index():
1833 """Return a string with the (binary) storage index."""
1834 def get_storage_index_string():
1835 """Return a string with the (printable) abbreviated storage index."""
1836 def get_repair_attempted():
1837 """Return a boolean, True if a repair was attempted. We might not
1838 attempt to repair the file because it was healthy, or healthy enough
1839 (i.e. some shares were missing but not enough to exceed some
1840 threshold), or because we don't know how to repair this object."""
1841 def get_repair_successful():
1842 """Return a boolean, True if repair was attempted and the file/dir
1843 was fully healthy afterwards. False if no repair was attempted or if
1844 a repair attempt failed."""
1845 def get_pre_repair_results():
1846 """Return an ICheckResults instance that describes the state of the
1847 file/dir before any repair was attempted."""
1848 def get_post_repair_results():
1849 """Return an ICheckResults instance that describes the state of the
1850 file/dir after any repair was attempted. If no repair was attempted,
1851 the pre-repair and post-repair results will be identical."""
1854 class IDeepCheckResults(Interface):
1855 """I contain the results of a deep-check operation.
1857 This is returned by a call to ICheckable.deep_check().
1860 def get_root_storage_index_string():
1861 """Return the storage index (abbreviated human-readable string) of
1862 the first object checked."""
1864 """Return a dictionary with the following keys::
1866 count-objects-checked: count of how many objects were checked
1867 count-objects-healthy: how many of those objects were completely
1869 count-objects-unhealthy: how many were damaged in some way
1870 count-objects-unrecoverable: how many were unrecoverable
1871 count-corrupt-shares: how many shares were found to have
1872 corruption, summed over all objects
1876 def get_corrupt_shares():
1877 """Return a set of (serverid, storage_index, sharenum) for all shares
1878 that were found to be corrupt. Both serverid and storage_index are
1881 def get_all_results():
1882 """Return a dictionary mapping pathname (a tuple of strings, ready to
1883 be slash-joined) to an ICheckResults instance, one for each object
1884 that was checked."""
1886 def get_results_for_storage_index(storage_index):
1887 """Retrive the ICheckResults instance for the given (binary)
1888 storage index. Raises KeyError if there are no results for that
1892 """Return a dictionary with the same keys as
1893 IDirectoryNode.deep_stats()."""
1895 class IDeepCheckAndRepairResults(Interface):
1896 """I contain the results of a deep-check-and-repair operation.
1898 This is returned by a call to ICheckable.deep_check_and_repair().
1901 def get_root_storage_index_string():
1902 """Return the storage index (abbreviated human-readable string) of
1903 the first object checked."""
1905 """Return a dictionary with the following keys::
1907 count-objects-checked: count of how many objects were checked
1908 count-objects-healthy-pre-repair: how many of those objects were
1909 completely healthy (before any
1911 count-objects-unhealthy-pre-repair: how many were damaged in
1913 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1914 count-objects-healthy-post-repair: how many of those objects were
1915 completely healthy (after any
1917 count-objects-unhealthy-post-repair: how many were damaged in
1919 count-objects-unrecoverable-post-repair: how many were
1921 count-repairs-attempted: repairs were attempted on this many
1922 objects. The count-repairs- keys will
1923 always be provided, however unless
1924 repair=true is present, they will all
1926 count-repairs-successful: how many repairs resulted in healthy
1928 count-repairs-unsuccessful: how many repairs resulted did not
1929 results in completely healthy objects
1930 count-corrupt-shares-pre-repair: how many shares were found to
1931 have corruption, summed over all
1932 objects examined (before any
1934 count-corrupt-shares-post-repair: how many shares were found to
1935 have corruption, summed over all
1936 objects examined (after any
1941 """Return a dictionary with the same keys as
1942 IDirectoryNode.deep_stats()."""
1944 def get_corrupt_shares():
1945 """Return a set of (serverid, storage_index, sharenum) for all shares
1946 that were found to be corrupt before any repair was attempted. Both
1947 serverid and storage_index are binary.
1949 def get_remaining_corrupt_shares():
1950 """Return a set of (serverid, storage_index, sharenum) for all shares
1951 that were found to be corrupt after any repair was completed. Both
1952 serverid and storage_index are binary. These are shares that need
1953 manual inspection and probably deletion.
1955 def get_all_results():
1956 """Return a dictionary mapping pathname (a tuple of strings, ready to
1957 be slash-joined) to an ICheckAndRepairResults instance, one for each
1958 object that was checked."""
1960 def get_results_for_storage_index(storage_index):
1961 """Retrive the ICheckAndRepairResults instance for the given (binary)
1962 storage index. Raises KeyError if there are no results for that
1966 class IRepairable(Interface):
1967 def repair(check_results):
1968 """Attempt to repair the given object. Returns a Deferred that fires
1969 with a IRepairResults object.
1971 I must be called with an object that implements ICheckResults, as
1972 proof that you have actually discovered a problem with this file. I
1973 will use the data in the checker results to guide the repair process,
1974 such as which servers provided bad data and should therefore be
1975 avoided. The ICheckResults object is inside the
1976 ICheckAndRepairResults object, which is returned by the
1977 ICheckable.check() method::
1979 d = filenode.check(repair=False)
1980 def _got_results(check_and_repair_results):
1981 check_results = check_and_repair_results.get_pre_repair_results()
1982 return filenode.repair(check_results)
1983 d.addCallback(_got_results)
1987 class IRepairResults(Interface):
1988 """I contain the results of a repair operation."""
1991 class IClient(Interface):
1992 def upload(uploadable):
1993 """Upload some data into a CHK, get back the UploadResults for it.
1994 @param uploadable: something that implements IUploadable
1995 @return: a Deferred that fires with the UploadResults instance.
1996 To get the URI for this file, use results.uri .
1999 def create_mutable_file(contents=""):
2000 """Create a new mutable file with contents, get back the URI string.
2001 @param contents: the initial contents to place in the file.
2002 @return: a Deferred that fires with tne (string) SSK URI for the new
2006 def create_empty_dirnode():
2007 """Create a new dirnode, empty and unattached.
2008 @return: a Deferred that fires with the new IDirectoryNode instance.
2011 def create_node_from_uri(uri, rouri):
2012 """Create a new IFilesystemNode instance from the uri, synchronously.
2013 @param uri: a string or IURI-providing instance, or None. This could
2014 be for a LiteralFileNode, a CHK file node, a mutable file
2015 node, or a directory node
2016 @param rouri: a string or IURI-providing instance, or None. If the
2017 main uri is None, I will use the rouri instead. If I
2018 recognize the format of the main uri, I will ignore the
2019 rouri (because it can be derived from the writecap).
2021 @return: an instance that provides IFilesystemNode (or more usefully
2022 one of its subclasses). File-specifying URIs will result in
2023 IFileNode or IMutableFileNode -providing instances, like
2024 FileNode, LiteralFileNode, or MutableFileNode.
2025 Directory-specifying URIs will result in
2026 IDirectoryNode-providing instances, like DirectoryNode.
2029 class IClientStatus(Interface):
2030 def list_all_uploads():
2031 """Return a list of uploader objects, one for each upload which
2032 currently has an object available (tracked with weakrefs). This is
2033 intended for debugging purposes."""
2034 def list_active_uploads():
2035 """Return a list of active IUploadStatus objects."""
2036 def list_recent_uploads():
2037 """Return a list of IUploadStatus objects for the most recently
2040 def list_all_downloads():
2041 """Return a list of downloader objects, one for each download which
2042 currently has an object available (tracked with weakrefs). This is
2043 intended for debugging purposes."""
2044 def list_active_downloads():
2045 """Return a list of active IDownloadStatus objects."""
2046 def list_recent_downloads():
2047 """Return a list of IDownloadStatus objects for the most recently
2048 started downloads."""
2050 class IUploadStatus(Interface):
2052 """Return a timestamp (float with seconds since epoch) indicating
2053 when the operation was started."""
2054 def get_storage_index():
2055 """Return a string with the (binary) storage index in use on this
2056 upload. Returns None if the storage index has not yet been
2059 """Return an integer with the number of bytes that will eventually
2060 be uploaded for this file. Returns None if the size is not yet known.
2063 """Return True if this upload is using a Helper, False if not."""
2065 """Return a string describing the current state of the upload
2068 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2069 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2070 made towards hashing the file to determine a CHK encryption key: if
2071 non-convergent encryption is in use, this will be trivial, otherwise
2072 the whole file must be hashed. 'ciphertext' describes how much of the
2073 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2074 uploads. 'encode_and_push' describes how much of the encode-and-push
2075 process has finished: for helper uploads this is dependent upon the
2076 helper providing progress reports. It might be reasonable to add all
2077 three numbers and report the sum to the user."""
2079 """Return True if the upload is currently active, False if not."""
2081 """Return an instance of UploadResults (which contains timing and
2082 sharemap information). Might return None if the upload is not yet
2085 """Each upload status gets a unique number: this method returns that
2086 number. This provides a handle to this particular upload, so a web
2087 page can generate a suitable hyperlink."""
2089 class IDownloadStatus(Interface):
2091 """Return a timestamp (float with seconds since epoch) indicating
2092 when the operation was started."""
2093 def get_storage_index():
2094 """Return a string with the (binary) storage index in use on this
2095 download. This may be None if there is no storage index (i.e. LIT
2098 """Return an integer with the number of bytes that will eventually be
2099 retrieved for this file. Returns None if the size is not yet known.
2102 """Return True if this download is using a Helper, False if not."""
2104 """Return a string describing the current state of the download
2107 """Returns a float (from 0.0 to 1.0) describing the amount of the
2108 download that has completed. This value will remain at 0.0 until the
2109 first byte of plaintext is pushed to the download target."""
2111 """Return True if the download is currently active, False if not."""
2113 """Each download status gets a unique number: this method returns
2114 that number. This provides a handle to this particular download, so a
2115 web page can generate a suitable hyperlink."""
2117 class IServermapUpdaterStatus(Interface):
2119 class IPublishStatus(Interface):
2121 class IRetrieveStatus(Interface):
2124 class NotCapableError(Exception):
2125 """You have tried to write to a read-only node."""
2127 class BadWriteEnablerError(Exception):
2130 class RIControlClient(RemoteInterface):
2132 def wait_for_client_connections(num_clients=int):
2133 """Do not return until we have connections to at least NUM_CLIENTS
2137 def upload_from_file_to_uri(filename=str,
2138 convergence=ChoiceOf(None,
2139 StringConstraint(2**20))):
2140 """Upload a file to the grid. This accepts a filename (which must be
2141 absolute) that points to a file on the node's local disk. The node will
2142 read the contents of this file, upload it to the grid, then return the
2143 URI at which it was uploaded. If convergence is None then a random
2144 encryption key will be used, else the plaintext will be hashed, then
2145 that hash will be mixed together with the "convergence" string to form
2150 def download_from_uri_to_file(uri=URI, filename=str):
2151 """Download a file from the grid, placing it on the node's local disk
2152 at the given filename (which must be absolute[?]). Returns the
2153 absolute filename where the file was written."""
2158 def get_memory_usage():
2159 """Return a dict describes the amount of memory currently in use. The
2160 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2161 measuring memory consupmtion in bytes."""
2162 return DictOf(str, int)
2164 def speed_test(count=int, size=int, mutable=Any()):
2165 """Write 'count' tempfiles to disk, all of the given size. Measure
2166 how long (in seconds) it takes to upload them all to the servers.
2167 Then measure how long it takes to download all of them. If 'mutable'
2168 is 'create', time creation of mutable files. If 'mutable' is
2169 'upload', then time access to the same mutable file instead of
2172 Returns a tuple of (upload_time, download_time).
2174 return (float, float)
2176 def measure_peer_response_time():
2177 """Send a short message to each connected peer, and measure the time
2178 it takes for them to respond to it. This is a rough measure of the
2179 application-level round trip time.
2181 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2184 return DictOf(Nodeid, float)
2186 UploadResults = Any() #DictOf(str, str)
2188 class RIEncryptedUploadable(RemoteInterface):
2189 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2194 def get_all_encoding_parameters():
2195 return (int, int, int, long)
2197 def read_encrypted(offset=Offset, length=ReadSize):
2204 class RICHKUploadHelper(RemoteInterface):
2205 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2209 Return a dictionary of version information.
2211 return DictOf(str, Any())
2213 def upload(reader=RIEncryptedUploadable):
2214 return UploadResults
2217 class RIHelper(RemoteInterface):
2218 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2222 Return a dictionary of version information.
2224 return DictOf(str, Any())
2226 def upload_chk(si=StorageIndex):
2227 """See if a file with a given storage index needs uploading. The
2228 helper will ask the appropriate storage servers to see if the file
2229 has already been uploaded. If so, the helper will return a set of
2230 'upload results' that includes whatever hashes are needed to build
2231 the read-cap, and perhaps a truncated sharemap.
2233 If the file has not yet been uploaded (or if it was only partially
2234 uploaded), the helper will return an empty upload-results dictionary
2235 and also an RICHKUploadHelper object that will take care of the
2236 upload process. The client should call upload() on this object and
2237 pass it a reference to an RIEncryptedUploadable object that will
2238 provide ciphertext. When the upload is finished, the upload() method
2239 will finish and return the upload results.
2241 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2244 class RIStatsProvider(RemoteInterface):
2245 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2247 Provides access to statistics and monitoring information.
2252 returns a dictionary containing 'counters' and 'stats', each a
2253 dictionary with string counter/stat name keys, and numeric values.
2254 counters are monotonically increasing measures of work done, and
2255 stats are instantaneous measures (potentially time averaged
2258 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2260 class RIStatsGatherer(RemoteInterface):
2261 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2263 Provides a monitoring service for centralised collection of stats
2266 def provide(provider=RIStatsProvider, nickname=str):
2268 @param provider: a stats collector instance which should be polled
2269 periodically by the gatherer to collect stats.
2270 @param nickname: a name useful to identify the provided client
2275 class IStatsProducer(Interface):
2278 returns a dictionary, with str keys representing the names of stats
2279 to be monitored, and numeric values.
2282 class RIKeyGenerator(RemoteInterface):
2283 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2285 Provides a service offering to make RSA key pairs.
2288 def get_rsa_key_pair(key_size=int):
2290 @param key_size: the size of the signature key.
2291 @return: tuple(verifying_key, signing_key)
2293 return TupleOf(str, str)
2296 class FileTooLargeError(Exception):
2299 class IValidatedThingProxy(Interface):
2301 """ Acquire a thing and validate it. Return a deferred which is
2302 eventually fired with self if the thing is valid or errbacked if it
2303 can't be acquired or validated."""
2305 class InsufficientVersionError(Exception):
2306 def __init__(self, needed, got):
2307 self.needed = needed
2310 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,