2 from zope.interface import Interface
3 from foolscap.schema import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any
5 from foolscap import RemoteInterface, Referenceable
9 Hash = StringConstraint(maxLength=HASH_SIZE,
10 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
11 Nodeid = StringConstraint(maxLength=20,
12 minLength=20) # binary format 20-byte SHA1 hash
13 FURL = StringConstraint(1000)
14 StorageIndex = StringConstraint(16)
15 URI = StringConstraint(300) # kind of arbitrary
17 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
19 ShareData = StringConstraint(None)
20 URIExtensionData = StringConstraint(1000)
21 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
23 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
24 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
25 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
26 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
28 class RIStubClient(RemoteInterface):
29 """Each client publishes a service announcement for a dummy object called
30 the StubClient. This object doesn't actually offer any services, but the
31 announcement helps the Introducer keep track of which clients are
32 subscribed (so the grid admin can keep track of things like the size of
33 the grid and the client versions in use. This is the (empty)
34 RemoteInterface for the StubClient."""
36 class RIBucketWriter(RemoteInterface):
37 """ Objects of this kind live on the server side. """
38 def write(offset=Offset, data=ShareData):
43 If the data that has been written is incomplete or inconsistent then
44 the server will throw the data away, else it will store it for future
50 """Abandon all the data that has been written.
54 class RIBucketReader(RemoteInterface):
55 def read(offset=Offset, length=ReadSize):
58 def advise_corrupt_share(reason=str):
59 """Clients who discover hash failures in shares that they have
60 downloaded from me will use this method to inform me about the
61 failures. I will record their concern so that my operator can
62 manually inspect the shares in question. I return None.
64 This is a wrapper around RIStorageServer.advise_corrupt_share(),
65 which is tied to a specific share, and therefore does not need the
66 extra share-identifying arguments. Please see that method for full
70 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
71 # elements are (offset, length, operator, specimen)
72 # operator is one of "lt, le, eq, ne, ge, gt"
73 # nop always passes and is used to fetch data while writing.
74 # you should use length==len(specimen) for everything except nop
75 DataVector = ListOf(TupleOf(Offset, ShareData))
76 # (offset, data). This limits us to 30 writes of 1MiB each per call
77 TestAndWriteVectorsForShares = DictOf(int,
80 ChoiceOf(None, Offset), # new_length
82 ReadVector = ListOf(TupleOf(Offset, ReadSize))
83 ReadData = ListOf(ShareData)
84 # returns data[offset:offset+length] for each element of TestVector
86 class RIStorageServer(RemoteInterface):
87 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
91 Return a dictionary of version information.
93 return DictOf(str, Any())
95 def allocate_buckets(storage_index=StorageIndex,
96 renew_secret=LeaseRenewSecret,
97 cancel_secret=LeaseCancelSecret,
98 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
99 allocated_size=Offset, canary=Referenceable):
101 @param storage_index: the index of the bucket to be created or
103 @param sharenums: these are the share numbers (probably between 0 and
104 99) that the sender is proposing to store on this
106 @param renew_secret: This is the secret used to protect bucket refresh
107 This secret is generated by the client and
108 stored for later comparison by the server. Each
109 server is given a different secret.
110 @param cancel_secret: Like renew_secret, but protects bucket decref.
111 @param canary: If the canary is lost before close(), the bucket is
113 @return: tuple of (alreadygot, allocated), where alreadygot is what we
114 already have and is what we hereby agree to accept. New
115 leases are added for shares in both lists.
117 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
118 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
120 def add_lease(storage_index=StorageIndex,
121 renew_secret=LeaseRenewSecret,
122 cancel_secret=LeaseCancelSecret):
124 Add a new lease on the given bucket. If the renew_secret matches an
125 existing lease, that lease will be renewed instead.
129 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
131 Renew the lease on a given bucket. Some networks will use this, some
136 def cancel_lease(storage_index=StorageIndex,
137 cancel_secret=LeaseCancelSecret):
139 Cancel the lease on a given bucket. If this was the last lease on the
140 bucket, the bucket will be deleted.
144 def get_buckets(storage_index=StorageIndex):
145 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
149 def slot_readv(storage_index=StorageIndex,
150 shares=ListOf(int), readv=ReadVector):
151 """Read a vector from the numbered shares associated with the given
152 storage index. An empty shares list means to return data from all
153 known shares. Returns a dictionary with one key per share."""
154 return DictOf(int, ReadData) # shnum -> results
156 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
157 secrets=TupleOf(WriteEnablerSecret,
160 tw_vectors=TestAndWriteVectorsForShares,
163 """General-purpose test-and-set operation for mutable slots. Perform
164 a bunch of comparisons against the existing shares. If they all pass,
165 then apply a bunch of write vectors to those shares. Then use the
166 read vectors to extract data from all the shares and return the data.
168 This method is, um, large. The goal is to allow clients to update all
169 the shares associated with a mutable file in a single round trip.
171 @param storage_index: the index of the bucket to be created or
173 @param write_enabler: a secret that is stored along with the slot.
174 Writes are accepted from any caller who can
175 present the matching secret. A different secret
176 should be used for each slot*server pair.
177 @param renew_secret: This is the secret used to protect bucket refresh
178 This secret is generated by the client and
179 stored for later comparison by the server. Each
180 server is given a different secret.
181 @param cancel_secret: Like renew_secret, but protects bucket decref.
183 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
184 cancel_secret). The first is required to perform any write. The
185 latter two are used when allocating new shares. To simply acquire a
186 new lease on existing shares, use an empty testv and an empty writev.
188 Each share can have a separate test vector (i.e. a list of
189 comparisons to perform). If all vectors for all shares pass, then all
190 writes for all shares are recorded. Each comparison is a 4-tuple of
191 (offset, length, operator, specimen), which effectively does a bool(
192 (read(offset, length)) OPERATOR specimen ) and only performs the
193 write if all these evaluate to True. Basic test-and-set uses 'eq'.
194 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
195 Write-if-same-or-newer uses 'le'.
197 Reads from the end of the container are truncated, and missing shares
198 behave like empty ones, so to assert that a share doesn't exist (for
199 use when creating a new share), use (0, 1, 'eq', '').
201 The write vector will be applied to the given share, expanding it if
202 necessary. A write vector applied to a share number that did not
203 exist previously will cause that share to be created.
205 Each write vector is accompanied by a 'new_length' argument. If
206 new_length is not None, use it to set the size of the container. This
207 can be used to pre-allocate space for a series of upcoming writes, or
208 truncate existing data. If the container is growing, new_length will
209 be applied before datav. If the container is shrinking, it will be
212 The read vector is used to extract data from all known shares,
213 *before* any writes have been applied. The same vector is used for
214 all shares. This captures the state that was tested by the test
217 This method returns two values: a boolean and a dict. The boolean is
218 True if the write vectors were applied, False if not. The dict is
219 keyed by share number, and each value contains a list of strings, one
220 for each element of the read vector.
222 If the write_enabler is wrong, this will raise BadWriteEnablerError.
223 To enable share migration (using update_write_enabler), the exception
224 will have the nodeid used for the old write enabler embedded in it,
225 in the following string::
227 The write enabler was recorded by nodeid '%s'.
229 Note that the nodeid here is encoded using the same base32 encoding
230 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
233 return TupleOf(bool, DictOf(int, ReadData))
235 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
236 shnum=int, reason=str):
237 """Clients who discover hash failures in shares that they have
238 downloaded from me will use this method to inform me about the
239 failures. I will record their concern so that my operator can
240 manually inspect the shares in question. I return None.
242 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
243 (binary) storage index string, and 'shnum' is the integer share
244 number. 'reason' is a human-readable explanation of the problem,
245 probably including some expected hash values and the computed ones
246 which did not match. Corruption advisories for mutable shares should
247 include a hash of the public key (the same value that appears in the
248 mutable-file verify-cap), since the current share format does not
252 class IStorageBucketWriter(Interface):
254 Objects of this kind live on the client side.
256 def put_block(segmentnum=int, data=ShareData):
257 """@param data: For most segments, this data will be 'blocksize'
258 bytes in length. The last segment might be shorter.
259 @return: a Deferred that fires (with None) when the operation completes
262 def put_plaintext_hashes(hashes=ListOf(Hash)):
264 @return: a Deferred that fires (with None) when the operation completes
267 def put_crypttext_hashes(hashes=ListOf(Hash)):
269 @return: a Deferred that fires (with None) when the operation completes
272 def put_block_hashes(blockhashes=ListOf(Hash)):
274 @return: a Deferred that fires (with None) when the operation completes
277 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
279 @return: a Deferred that fires (with None) when the operation completes
282 def put_uri_extension(data=URIExtensionData):
283 """This block of data contains integrity-checking information (hashes
284 of plaintext, crypttext, and shares), as well as encoding parameters
285 that are necessary to recover the data. This is a serialized dict
286 mapping strings to other strings. The hash of this data is kept in
287 the URI and verified before any of the data is used. All buckets for
288 a given file contain identical copies of this data.
290 The serialization format is specified with the following pseudocode:
291 for k in sorted(dict.keys()):
292 assert re.match(r'^[a-zA-Z_\-]+$', k)
293 write(k + ':' + netstring(dict[k]))
295 @return: a Deferred that fires (with None) when the operation completes
299 """Finish writing and close the bucket. The share is not finalized
300 until this method is called: if the uploading client disconnects
301 before calling close(), the partially-written share will be
304 @return: a Deferred that fires (with None) when the operation completes
307 class IStorageBucketReader(Interface):
309 def get_block(blocknum=int):
310 """Most blocks will be the same size. The last block might be shorter
316 def get_crypttext_hashes():
318 @return: ListOf(Hash)
321 def get_block_hashes():
323 @return: ListOf(Hash)
326 def get_share_hashes():
328 @return: ListOf(TupleOf(int, Hash))
331 def get_uri_extension():
333 @return: URIExtensionData
338 # hm, we need a solution for forward references in schemas
339 FileNode_ = Any() # TODO: foolscap needs constraints on copyables
340 DirectoryNode_ = Any() # TODO: same
341 AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
344 class IURI(Interface):
345 def init_from_string(uri):
346 """Accept a string (as created by my to_string() method) and populate
347 this instance with its data. I am not normally called directly,
348 please use the module-level uri.from_string() function to convert
349 arbitrary URI strings into IURI-providing instances."""
352 """Return False if this URI be used to modify the data. Return True
353 if this URI cannot be used to modify the data."""
356 """Return True if the data can be modified by *somebody* (perhaps
357 someone who has a more powerful URI than this one)."""
360 """Return another IURI instance, which represents a read-only form of
361 this one. If is_readonly() is True, this returns self."""
363 def get_verify_cap():
364 """Return an instance that provides IVerifierURI, which can be used
365 to check on the availability of the file or directory, without
366 providing enough capabilities to actually read or modify the
367 contents. This may return None if the file does not need checking or
368 verification (e.g. LIT URIs).
372 """Return a string of printable ASCII characters, suitable for
373 passing into init_from_string."""
375 class IVerifierURI(Interface, IURI):
376 def init_from_string(uri):
377 """Accept a string (as created by my to_string() method) and populate
378 this instance with its data. I am not normally called directly,
379 please use the module-level uri.from_string() function to convert
380 arbitrary URI strings into IURI-providing instances."""
383 """Return a string of printable ASCII characters, suitable for
384 passing into init_from_string."""
386 class IDirnodeURI(Interface):
387 """I am a URI which represents a dirnode."""
390 class IFileURI(Interface):
391 """I am a URI which represents a filenode."""
393 """Return the length (in bytes) of the file that I represent."""
395 class IMutableFileURI(Interface):
396 """I am a URI which represents a mutable filenode."""
397 class INewDirectoryURI(Interface):
399 class IReadonlyNewDirectoryURI(Interface):
403 class IFilesystemNode(Interface):
406 Return the URI that can be used by others to get access to this
407 node. If this node is read-only, the URI will only offer read-only
408 access. If this node is read-write, the URI will offer read-write
411 If you have read-write access to a node and wish to share merely
412 read-only access with others, use get_readonly_uri().
415 def get_readonly_uri():
416 """Return the directory URI that can be used by others to get
417 read-only access to this directory node. The result is a read-only
418 URI, regardless of whether this dirnode is read-only or read-write.
420 If you have merely read-only access to this dirnode,
421 get_readonly_uri() will return the same thing as get_uri().
424 def get_verify_cap():
425 """Return an IVerifierURI instance that represents the
426 'verifiy/refresh capability' for this node. The holder of this
427 capability will be able to renew the lease for this node, protecting
428 it from garbage-collection. They will also be able to ask a server if
429 it holds a share for the file or directory.
432 def get_storage_index():
433 """Return a string with the (binary) storage index in use on this
434 download. This may be None if there is no storage index (i.e. LIT
438 """Return True if this reference provides mutable access to the given
439 file or directory (i.e. if you can modify it), or False if not. Note
440 that even if this reference is read-only, someone else may hold a
441 read-write reference to it."""
444 """Return True if this file or directory is mutable (by *somebody*,
445 not necessarily you), False if it is is immutable. Note that a file
446 might be mutable overall, but your reference to it might be
447 read-only. On the other hand, all references to an immutable file
448 will be read-only; there are no read-write references to an immutable
452 class IMutableFilesystemNode(IFilesystemNode):
455 class IFileNode(IFilesystemNode):
456 def download(target):
457 """Download the file's contents to a given IDownloadTarget"""
459 def download_to_data():
460 """Download the file's contents. Return a Deferred that fires
461 with those contents."""
464 """Return the length (in bytes) of the data this node represents."""
466 def read(consumer, offset=0, size=None):
467 """Download a portion (possibly all) of the file's contents, making
468 them available to the given IConsumer. Return a Deferred that fires
469 (with the consumer) when the consumer is unregistered (either because
470 the last byte has been given to it, or because the consumer threw an
471 exception during write(), possibly because it no longer wants to
472 receive data). The portion downloaded will start at 'offset' and
473 contain 'size' bytes (or the remainder of the file if size==None).
475 The consumer will be used in non-streaming mode: an IPullProducer
476 will be attached to it.
478 The consumer will not receive data right away: several network trips
479 must occur first. The order of events will be::
481 consumer.registerProducer(p, streaming)
482 (if streaming == False)::
483 consumer does p.resumeProducing()
485 consumer does p.resumeProducing()
486 consumer.write(data).. (repeat until all data is written)
487 consumer.unregisterProducer()
488 deferred.callback(consumer)
490 If a download error occurs, or an exception is raised by
491 consumer.registerProducer() or consumer.write(), I will call
492 consumer.unregisterProducer() and then deliver the exception via
493 deferred.errback(). To cancel the download, the consumer should call
494 p.stopProducing(), which will result in an exception being delivered
495 via deferred.errback().
497 A simple download-to-memory consumer example would look like this::
499 class MemoryConsumer:
500 implements(IConsumer)
504 def registerProducer(self, p, streaming):
505 assert streaming == False
508 def write(self, data):
509 self.chunks.append(data)
510 def unregisterProducer(self):
512 d = filenode.read(MemoryConsumer())
513 d.addCallback(lambda mc: "".join(mc.chunks))
518 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
519 """I provide access to a 'mutable file', which retains its identity
520 regardless of what contents are put in it.
522 The consistency-vs-availability problem means that there might be
523 multiple versions of a file present in the grid, some of which might be
524 unrecoverable (i.e. have fewer than 'k' shares). These versions are
525 loosely ordered: each has a sequence number and a hash, and any version
526 with seqnum=N was uploaded by a node which has seen at least one version
529 The 'servermap' (an instance of IMutableFileServerMap) is used to
530 describe the versions that are known to be present in the grid, and which
531 servers are hosting their shares. It is used to represent the 'state of
532 the world', and is used for this purpose by my test-and-set operations.
533 Downloading the contents of the mutable file will also return a
534 servermap. Uploading a new version into the mutable file requires a
535 servermap as input, and the semantics of the replace operation is
536 'replace the file with my new version if it looks like nobody else has
537 changed the file since my previous download'. Because the file is
538 distributed, this is not a perfect test-and-set operation, but it will do
539 its best. If the replace process sees evidence of a simultaneous write,
540 it will signal an UncoordinatedWriteError, so that the caller can take
544 Most readers will want to use the 'best' current version of the file, and
545 should use my 'download_best_version()' method.
547 To unconditionally replace the file, callers should use overwrite(). This
548 is the mode that user-visible mutable files will probably use.
550 To apply some delta to the file, call modify() with a callable modifier
551 function that can apply the modification that you want to make. This is
552 the mode that dirnodes will use, since most directory modification
553 operations can be expressed in terms of deltas to the directory state.
556 Three methods are available for users who need to perform more complex
557 operations. The first is get_servermap(), which returns an up-to-date
558 servermap using a specified mode. The second is download_version(), which
559 downloads a specific version (not necessarily the 'best' one). The third
560 is 'upload', which accepts new contents and a servermap (which must have
561 been updated with MODE_WRITE). The upload method will attempt to apply
562 the new contents as long as no other node has modified the file since the
563 servermap was updated. This might be useful to a caller who wants to
564 merge multiple versions into a single new one.
566 Note that each time the servermap is updated, a specific 'mode' is used,
567 which determines how many peers are queried. To use a servermap for my
568 replace() method, that servermap must have been updated in MODE_WRITE.
569 These modes are defined in allmydata.mutable.common, and consist of
570 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
571 allmydata/mutable/servermap.py for details about the differences.
573 Mutable files are currently limited in size (about 3.5MB max) and can
574 only be retrieved and updated all-at-once, as a single big string. Future
575 versions of our mutable files will remove this restriction.
578 def download_best_version():
579 """Download the 'best' available version of the file, meaning one of
580 the recoverable versions with the highest sequence number. If no
581 uncoordinated writes have occurred, and if enough shares are
582 available, then this will be the most recent version that has been
585 I update an internal servermap with MODE_READ, determine which
586 version of the file is indicated by
587 servermap.best_recoverable_version(), and return a Deferred that
588 fires with its contents. If no version is recoverable, the Deferred
589 will errback with UnrecoverableFileError.
592 def get_size_of_best_version():
593 """Find the size of the version that would be downloaded with
594 download_best_version(), without actually downloading the whole file.
596 I return a Deferred that fires with an integer.
599 def overwrite(new_contents):
600 """Unconditionally replace the contents of the mutable file with new
601 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
602 is only appropriate to use when the new contents of the file are
603 completely unrelated to the old ones, and you do not care about other
606 I return a Deferred that fires (with a PublishStatus object) when the
607 update has completed.
610 def modify(modifier_cb):
611 """Modify the contents of the file, by downloading the current
612 version, applying the modifier function (or bound method), then
613 uploading the new version. I return a Deferred that fires (with a
614 PublishStatus object) when the update is complete.
616 The modifier callable will be given three arguments: a string (with
617 the old contents), a 'first_time' boolean, and a servermap. As with
618 download_best_version(), the old contents will be from the best
619 recoverable version, but the modifier can use the servermap to make
620 other decisions (such as refusing to apply the delta if there are
621 multiple parallel versions, or if there is evidence of a newer
622 unrecoverable version). 'first_time' will be True the first time the
623 modifier is called, and False on any subsequent calls.
625 The callable should return a string with the new contents. The
626 callable must be prepared to be called multiple times, and must
627 examine the input string to see if the change that it wants to make
628 is already present in the old version. If it does not need to make
629 any changes, it can either return None, or return its input string.
631 If the modifier raises an exception, it will be returned in the
636 def get_servermap(mode):
637 """Return a Deferred that fires with an IMutableFileServerMap
638 instance, updated using the given mode.
641 def download_version(servermap, version):
642 """Download a specific version of the file, using the servermap
643 as a guide to where the shares are located.
645 I return a Deferred that fires with the requested contents, or
646 errbacks with UnrecoverableFileError. Note that a servermap which was
647 updated with MODE_ANYTHING or MODE_READ may not know about shares for
648 all versions (those modes stop querying servers as soon as they can
649 fulfil their goals), so you may want to use MODE_CHECK (which checks
650 everything) to get increased visibility.
653 def upload(new_contents, servermap):
654 """Replace the contents of the file with new ones. This requires a
655 servermap that was previously updated with MODE_WRITE.
657 I attempt to provide test-and-set semantics, in that I will avoid
658 modifying any share that is different than the version I saw in the
659 servermap. However, if another node is writing to the file at the
660 same time as me, I may manage to update some shares while they update
661 others. If I see any evidence of this, I will signal
662 UncoordinatedWriteError, and the file will be left in an inconsistent
663 state (possibly the version you provided, possibly the old version,
664 possibly somebody else's version, and possibly a mix of shares from
667 The recommended response to UncoordinatedWriteError is to either
668 return it to the caller (since they failed to coordinate their
669 writes), or to attempt some sort of recovery. It may be sufficient to
670 wait a random interval (with exponential backoff) and repeat your
671 operation. If I do not signal UncoordinatedWriteError, then I was
672 able to write the new version without incident.
674 I return a Deferred that fires (with a PublishStatus object) when the
675 publish has completed. I will update the servermap in-place with the
676 location of all new shares.
680 """Return this filenode's writekey, or None if the node does not have
681 write-capability. This may be used to assist with data structures
682 that need to make certain data available only to writers, such as the
683 read-write child caps in dirnodes. The recommended process is to have
684 reader-visible data be submitted to the filenode in the clear (where
685 it will be encrypted by the filenode using the readkey), but encrypt
686 writer-visible data using this writekey.
689 class NotEnoughSharesError(Exception):
692 class ExistingChildError(Exception):
693 """A directory node was asked to add or replace a child that already
694 exists, and overwrite= was set to False."""
696 class NoSuchChildError(Exception):
697 """A directory node was asked to fetch a child which does not exist."""
699 class IDirectoryNode(IMutableFilesystemNode):
700 """I represent a name-to-child mapping, holding the tahoe equivalent of a
701 directory. All child names are unicode strings, and all children are some
702 sort of IFilesystemNode (either files or subdirectories).
707 The dirnode ('1') URI returned by this method can be used in
708 set_uri() on a different directory ('2') to 'mount' a reference to
709 this directory ('1') under the other ('2'). This URI is just a
710 string, so it can be passed around through email or other out-of-band
714 def get_readonly_uri():
716 The dirnode ('1') URI returned by this method can be used in
717 set_uri() on a different directory ('2') to 'mount' a reference to
718 this directory ('1') under the other ('2'). This URI is just a
719 string, so it can be passed around through email or other out-of-band
724 """I return a Deferred that fires with a dictionary mapping child
725 name (a unicode string) to (node, metadata_dict) tuples, in which
726 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
727 is a dictionary of metadata."""
730 """I return a Deferred that fires with a boolean, True if there
731 exists a child of the given name, False if not. The child name must
732 be a unicode string."""
735 """I return a Deferred that fires with a specific named child node,
736 either an IFileNode or an IDirectoryNode. The child name must be a
737 unicode string. I raise NoSuchChildError if I do not have a child by
740 def get_metadata_for(name):
741 """I return a Deferred that fires with the metadata dictionary for a
742 specific named child node. This metadata is stored in the *edge*, not
743 in the child, so it is attached to the parent dirnode rather than the
744 child dir-or-file-node. The child name must be a unicode string. I
745 raise NoSuchChildError if I do not have a child by that name."""
747 def set_metadata_for(name, metadata):
748 """I replace any existing metadata for the named child with the new
749 metadata. The child name must be a unicode string. This metadata is
750 stored in the *edge*, not in the child, so it is attached to the
751 parent dirnode rather than the child dir-or-file-node. I return a
752 Deferred (that fires with this dirnode) when the operation is
753 complete. I raise NoSuchChildError if I do not have a child by that
756 def get_child_at_path(path):
757 """Transform a child path into an IDirectoryNode or IFileNode.
759 I perform a recursive series of 'get' operations to find the named
760 descendant node. I return a Deferred that fires with the node, or
761 errbacks with NoSuchChildError if the node could not be found.
763 The path can be either a single string (slash-separated) or a list of
764 path-name elements. All elements must be unicode strings.
767 def get_child_and_metadata_at_path(path):
768 """Transform a child path into an IDirectoryNode/IFileNode and
771 I am like get_child_at_path(), but my Deferred fires with a tuple of
772 (node, metadata). The metadata comes from the last edge. If the path
773 is empty, the metadata will be an empty dictionary.
776 def set_uri(name, child_uri, metadata=None, overwrite=True):
777 """I add a child (by URI) at the specific name. I return a Deferred
778 that fires when the operation finishes. If overwrite= is True, I will
779 replace any existing child of the same name, otherwise an existing
780 child will cause me to return ExistingChildError. The child name must
783 The child_uri could be for a file, or for a directory (either
784 read-write or read-only, using a URI that came from get_uri() ).
786 If metadata= is provided, I will use it as the metadata for the named
787 edge. This will replace any existing metadata. If metadata= is left
788 as the default value of None, I will set ['mtime'] to the current
789 time, and I will set ['ctime'] to the current time if there was not
790 already a child by this name present. This roughly matches the
791 ctime/mtime semantics of traditional filesystems.
793 If this directory node is read-only, the Deferred will errback with a
796 def set_children(entries, overwrite=True):
797 """Add multiple (name, child_uri) pairs (or (name, child_uri,
798 metadata) triples) to a directory node. Returns a Deferred that fires
799 (with None) when the operation finishes. This is equivalent to
800 calling set_uri() multiple times, but is much more efficient. All
801 child names must be unicode strings.
804 def set_node(name, child, metadata=None, overwrite=True):
805 """I add a child at the specific name. I return a Deferred that fires
806 when the operation finishes. This Deferred will fire with the child
807 node that was just added. I will replace any existing child of the
808 same name. The child name must be a unicode string. The 'child'
809 instance must be an instance providing IDirectoryNode or IFileNode.
811 If metadata= is provided, I will use it as the metadata for the named
812 edge. This will replace any existing metadata. If metadata= is left
813 as the default value of None, I will set ['mtime'] to the current
814 time, and I will set ['ctime'] to the current time if there was not
815 already a child by this name present. This roughly matches the
816 ctime/mtime semantics of traditional filesystems.
818 If this directory node is read-only, the Deferred will errback with a
821 def set_nodes(entries, overwrite=True):
822 """Add multiple (name, child_node) pairs (or (name, child_node,
823 metadata) triples) to a directory node. Returns a Deferred that fires
824 (with None) when the operation finishes. This is equivalent to
825 calling set_node() multiple times, but is much more efficient. All
826 child names must be unicode strings."""
829 def add_file(name, uploadable, metadata=None, overwrite=True):
830 """I upload a file (using the given IUploadable), then attach the
831 resulting FileNode to the directory at the given name. I set metadata
832 the same way as set_uri and set_node. The child name must be a
835 I return a Deferred that fires (with the IFileNode of the uploaded
836 file) when the operation completes."""
839 """I remove the child at the specific name. I return a Deferred that
840 fires when the operation finishes. The child name must be a unicode
841 string. I raise NoSuchChildError if I do not have a child by that
844 def create_empty_directory(name, overwrite=True):
845 """I create and attach an empty directory at the given name. The
846 child name must be a unicode string. I return a Deferred that fires
847 when the operation finishes."""
849 def move_child_to(current_child_name, new_parent, new_child_name=None,
851 """I take one of my children and move them to a new parent. The child
852 is referenced by name. On the new parent, the child will live under
853 'new_child_name', which defaults to 'current_child_name'. TODO: what
854 should we do about metadata? I return a Deferred that fires when the
855 operation finishes. The child name must be a unicode string. I raise
856 NoSuchChildError if I do not have a child by that name."""
858 def build_manifest():
859 """I generate a table of everything reachable from this directory.
860 I also compute deep-stats as described below.
862 I return a Monitor. The Monitor's results will be a dictionary with
865 res['manifest']: a list of (path, cap) tuples for all nodes
866 (directories and files) reachable from this one.
867 'path' will be a tuple of unicode strings. The
868 origin dirnode will be represented by an empty path
870 res['verifycaps']: a list of (printable) verifycap strings, one for
871 each reachable non-LIT node. This is a set:
872 it will contain no duplicates.
873 res['storage-index']: a list of (base32) storage index strings,
874 one for each reachable non-LIT node. This is
875 a set: it will contain no duplicates.
876 res['stats']: a dictionary, the same that is generated by
877 start_deep_stats() below.
879 The Monitor will also have an .origin_si attribute with the (binary)
880 storage index of the starting point.
883 def start_deep_stats():
884 """Return a Monitor, examining all nodes (directories and files)
885 reachable from this one. The Monitor's results will be a dictionary
886 with the following keys::
888 count-immutable-files: count of how many CHK files are in the set
889 count-mutable-files: same, for mutable files (does not include
891 count-literal-files: same, for LIT files
892 count-files: sum of the above three
894 count-directories: count of directories
896 size-immutable-files: total bytes for all CHK files in the set
897 size-mutable-files (TODO): same, for current version of all mutable
898 files, does not include directories
899 size-literal-files: same, for LIT files
900 size-directories: size of mutable files used by directories
902 largest-directory: number of bytes in the largest directory
903 largest-directory-children: number of children in the largest
905 largest-immutable-file: number of bytes in the largest CHK file
907 size-mutable-files is not yet implemented, because it would involve
908 even more queries than deep_stats does.
910 The Monitor will also have an .origin_si attribute with the (binary)
911 storage index of the starting point.
913 This operation will visit every directory node underneath this one,
914 and can take a long time to run. On a typical workstation with good
915 bandwidth, this can examine roughly 15 directories per second (and
916 takes several minutes of 100% CPU for ~1700 directories).
919 class ICodecEncoder(Interface):
920 def set_params(data_size, required_shares, max_shares):
921 """Set up the parameters of this encoder.
923 This prepares the encoder to perform an operation that converts a
924 single block of data into a number of shares, such that a future
925 ICodecDecoder can use a subset of these shares to recover the
926 original data. This operation is invoked by calling encode(). Once
927 the encoding parameters are set up, the encode operation can be
928 invoked multiple times.
930 set_params() prepares the encoder to accept blocks of input data that
931 are exactly 'data_size' bytes in length. The encoder will be prepared
932 to produce 'max_shares' shares for each encode() operation (although
933 see the 'desired_share_ids' to use less CPU). The encoding math will
934 be chosen such that the decoder can get by with as few as
935 'required_shares' of these shares and still reproduce the original
936 data. For example, set_params(1000, 5, 5) offers no redundancy at
937 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
939 Numerical Restrictions: 'data_size' is required to be an integral
940 multiple of 'required_shares'. In general, the caller should choose
941 required_shares and max_shares based upon their reliability
942 requirements and the number of peers available (the total storage
943 space used is roughly equal to max_shares*data_size/required_shares),
944 then choose data_size to achieve the memory footprint desired (larger
945 data_size means more efficient operation, smaller data_size means
946 smaller memory footprint).
948 In addition, 'max_shares' must be equal to or greater than
949 'required_shares'. Of course, setting them to be equal causes
950 encode() to degenerate into a particularly slow form of the 'split'
953 See encode() for more details about how these parameters are used.
955 set_params() must be called before any other ICodecEncoder methods
960 """Return the 3-tuple of data_size, required_shares, max_shares"""
962 def get_encoder_type():
963 """Return a short string that describes the type of this encoder.
965 There is required to be a global table of encoder classes. This method
966 returns an index into this table; the value at this index is an
967 encoder class, and this encoder is an instance of that class.
970 def get_block_size():
971 """Return the length of the shares that encode() will produce.
974 def encode_proposal(data, desired_share_ids=None):
977 'data' must be a string (or other buffer object), and len(data) must
978 be equal to the 'data_size' value passed earlier to set_params().
980 This will return a Deferred that will fire with two lists. The first
981 is a list of shares, each of which is a string (or other buffer
982 object) such that len(share) is the same as what get_share_size()
983 returned earlier. The second is a list of shareids, in which each is
984 an integer. The lengths of the two lists will always be equal to each
985 other. The user should take care to keep each share closely
986 associated with its shareid, as one is useless without the other.
988 The length of this output list will normally be the same as the value
989 provided to the 'max_shares' parameter of set_params(). This may be
990 different if 'desired_share_ids' is provided.
992 'desired_share_ids', if provided, is required to be a sequence of
993 ints, each of which is required to be >= 0 and < max_shares. If not
994 provided, encode() will produce 'max_shares' shares, as if
995 'desired_share_ids' were set to range(max_shares). You might use this
996 if you initially thought you were going to use 10 peers, started
997 encoding, and then two of the peers dropped out: you could use
998 desired_share_ids= to skip the work (both memory and CPU) of
999 producing shares for the peers which are no longer available.
1003 def encode(inshares, desired_share_ids=None):
1004 """Encode some data. This may be called multiple times. Each call is
1007 inshares is a sequence of length required_shares, containing buffers
1008 (i.e. strings), where each buffer contains the next contiguous
1009 non-overlapping segment of the input data. Each buffer is required to
1010 be the same length, and the sum of the lengths of the buffers is
1011 required to be exactly the data_size promised by set_params(). (This
1012 implies that the data has to be padded before being passed to
1013 encode(), unless of course it already happens to be an even multiple
1014 of required_shares in length.)
1016 ALSO: the requirement to break up your data into 'required_shares'
1017 chunks before calling encode() feels a bit surprising, at least from
1018 the point of view of a user who doesn't know how FEC works. It feels
1019 like an implementation detail that has leaked outside the
1020 abstraction barrier. Can you imagine a use case in which the data to
1021 be encoded might already be available in pre-segmented chunks, such
1022 that it is faster or less work to make encode() take a list rather
1023 than splitting a single string?
1025 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1026 is supposed to *produce* shares, so what it *accepts* should be
1027 something other than shares. Other places in this interface use the
1028 word 'data' for that-which-is-not-shares.. maybe we should use that
1031 'desired_share_ids', if provided, is required to be a sequence of
1032 ints, each of which is required to be >= 0 and < max_shares. If not
1033 provided, encode() will produce 'max_shares' shares, as if
1034 'desired_share_ids' were set to range(max_shares). You might use this
1035 if you initially thought you were going to use 10 peers, started
1036 encoding, and then two of the peers dropped out: you could use
1037 desired_share_ids= to skip the work (both memory and CPU) of
1038 producing shares for the peers which are no longer available.
1040 For each call, encode() will return a Deferred that fires with two
1041 lists, one containing shares and the other containing the shareids.
1042 The get_share_size() method can be used to determine the length of
1043 the share strings returned by encode(). Each shareid is a small
1044 integer, exactly as passed into 'desired_share_ids' (or
1045 range(max_shares), if desired_share_ids was not provided).
1047 The shares and their corresponding shareids are required to be kept
1048 together during storage and retrieval. Specifically, the share data is
1049 useless by itself: the decoder needs to be told which share is which
1050 by providing it with both the shareid and the actual share data.
1052 This function will allocate an amount of memory roughly equal to::
1054 (max_shares - required_shares) * get_share_size()
1056 When combined with the memory that the caller must allocate to
1057 provide the input data, this leads to a memory footprint roughly
1058 equal to the size of the resulting encoded shares (i.e. the expansion
1059 factor times the size of the input segment).
1064 # returning a list of (shareidN,shareN) tuples instead of a pair of
1065 # lists (shareids..,shares..). Brian thought the tuples would
1066 # encourage users to keep the share and shareid together throughout
1067 # later processing, Zooko pointed out that the code to iterate
1068 # through two lists is not really more complicated than using a list
1069 # of tuples and there's also a performance improvement
1071 # having 'data_size' not required to be an integral multiple of
1072 # 'required_shares'. Doing this would require encode() to perform
1073 # padding internally, and we'd prefer to have any padding be done
1074 # explicitly by the caller. Yes, it is an abstraction leak, but
1075 # hopefully not an onerous one.
1078 class ICodecDecoder(Interface):
1079 def set_params(data_size, required_shares, max_shares):
1080 """Set the params. They have to be exactly the same ones that were used for encoding. """
1082 def get_needed_shares():
1083 """Return the number of shares needed to reconstruct the data.
1084 set_params() is required to be called before this."""
1086 def decode(some_shares, their_shareids):
1087 """Decode a partial list of shares into data.
1089 'some_shares' is required to be a sequence of buffers of sharedata, a
1090 subset of the shares returned by ICodecEncode.encode(). Each share is
1091 required to be of the same length. The i'th element of their_shareids
1092 is required to be the shareid of the i'th buffer in some_shares.
1094 This returns a Deferred which fires with a sequence of buffers. This
1095 sequence will contain all of the segments of the original data, in
1096 order. The sum of the lengths of all of the buffers will be the
1097 'data_size' value passed into the original ICodecEncode.set_params()
1098 call. To get back the single original input block of data, use
1099 ''.join(output_buffers), or you may wish to simply write them in
1100 order to an output file.
1102 Note that some of the elements in the result sequence may be
1103 references to the elements of the some_shares input sequence. In
1104 particular, this means that if those share objects are mutable (e.g.
1105 arrays) and if they are changed, then both the input (the
1106 'some_shares' parameter) and the output (the value given when the
1107 deferred is triggered) will change.
1109 The length of 'some_shares' is required to be exactly the value of
1110 'required_shares' passed into the original ICodecEncode.set_params()
1114 class IEncoder(Interface):
1115 """I take an object that provides IEncryptedUploadable, which provides
1116 encrypted data, and a list of shareholders. I then encode, hash, and
1117 deliver shares to those shareholders. I will compute all the necessary
1118 Merkle hash trees that are necessary to validate the crypttext that
1119 eventually comes back from the shareholders. I provide the URI Extension
1120 Block Hash, and the encoding parameters, both of which must be included
1123 I do not choose shareholders, that is left to the IUploader. I must be
1124 given a dict of RemoteReferences to storage buckets that are ready and
1125 willing to receive data.
1129 """Specify the number of bytes that will be encoded. This must be
1130 peformed before get_serialized_params() can be called.
1132 def set_params(params):
1133 """Override the default encoding parameters. 'params' is a tuple of
1134 (k,d,n), where 'k' is the number of required shares, 'd' is the
1135 shares_of_happiness, and 'n' is the total number of shares that will
1138 Encoding parameters can be set in three ways. 1: The Encoder class
1139 provides defaults (3/7/10). 2: the Encoder can be constructed with
1140 an 'options' dictionary, in which the
1141 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1142 set_params((k,d,n)) can be called.
1144 If you intend to use set_params(), you must call it before
1145 get_share_size or get_param are called.
1148 def set_encrypted_uploadable(u):
1149 """Provide a source of encrypted upload data. 'u' must implement
1150 IEncryptedUploadable.
1152 When this is called, the IEncryptedUploadable will be queried for its
1153 length and the storage_index that should be used.
1155 This returns a Deferred that fires with this Encoder instance.
1157 This must be performed before start() can be called.
1160 def get_param(name):
1161 """Return an encoding parameter, by name.
1163 'storage_index': return a string with the (16-byte truncated SHA-256
1164 hash) storage index to which these shares should be
1167 'share_counts': return a tuple describing how many shares are used:
1168 (needed_shares, shares_of_happiness, total_shares)
1170 'num_segments': return an int with the number of segments that
1173 'segment_size': return an int with the size of each segment.
1175 'block_size': return the size of the individual blocks that will
1176 be delivered to a shareholder's put_block() method. By
1177 knowing this, the shareholder will be able to keep all
1178 blocks in a single file and still provide random access
1179 when reading them. # TODO: can we avoid exposing this?
1181 'share_size': an int with the size of the data that will be stored
1182 on each shareholder. This is aggregate amount of data
1183 that will be sent to the shareholder, summed over all
1184 the put_block() calls I will ever make. It is useful to
1185 determine this size before asking potential
1186 shareholders whether they will grant a lease or not,
1187 since their answers will depend upon how much space we
1188 need. TODO: this might also include some amount of
1189 overhead, like the size of all the hashes. We need to
1190 decide whether this is useful or not.
1192 'serialized_params': a string with a concise description of the
1193 codec name and its parameters. This may be passed
1194 into the IUploadable to let it make sure that
1195 the same file encoded with different parameters
1196 will result in different storage indexes.
1198 Once this is called, set_size() and set_params() may not be called.
1201 def set_shareholders(shareholders):
1202 """Tell the encoder where to put the encoded shares. 'shareholders'
1203 must be a dictionary that maps share number (an integer ranging from
1204 0 to n-1) to an instance that provides IStorageBucketWriter. This
1205 must be performed before start() can be called."""
1208 """Begin the encode/upload process. This involves reading encrypted
1209 data from the IEncryptedUploadable, encoding it, uploading the shares
1210 to the shareholders, then sending the hash trees.
1212 set_encrypted_uploadable() and set_shareholders() must be called
1213 before this can be invoked.
1215 This returns a Deferred that fires with a tuple of
1216 (uri_extension_hash, needed_shares, total_shares, size) when the
1217 upload process is complete. This information, plus the encryption
1218 key, is sufficient to construct the URI.
1221 class IDecoder(Interface):
1222 """I take a list of shareholders and some setup information, then
1223 download, validate, decode, and decrypt data from them, writing the
1224 results to an output file.
1226 I do not locate the shareholders, that is left to the IDownloader. I must
1227 be given a dict of RemoteReferences to storage buckets that are ready to
1232 """I take a file-like object (providing write and close) to which all
1233 the plaintext data will be written.
1235 TODO: producer/consumer . Maybe write() should return a Deferred that
1236 indicates when it will accept more data? But probably having the
1237 IDecoder be a producer is easier to glue to IConsumer pieces.
1240 def set_shareholders(shareholders):
1241 """I take a dictionary that maps share identifiers (small integers)
1242 to RemoteReferences that provide RIBucketReader. This must be called
1246 """I start the download. This process involves retrieving data and
1247 hash chains from the shareholders, using the hashes to validate the
1248 data, decoding the shares into segments, decrypting the segments,
1249 then writing the resulting plaintext to the output file.
1251 I return a Deferred that will fire (with self) when the download is
1255 class IDownloadTarget(Interface):
1256 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1257 # will register itself as a producer. This allows the target to invoke
1258 # downloader.pauseProducing, resumeProducing, and stopProducing.
1260 """Called before any calls to write() or close(). If an error
1261 occurs before any data is available, fail() may be called without
1262 a previous call to open().
1264 'size' is the length of the file being downloaded, in bytes."""
1267 """Output some data to the target."""
1269 """Inform the target that there is no more data to be written."""
1271 """fail() is called to indicate that the download has failed. 'why'
1272 is a Failure object indicating what went wrong. No further methods
1273 will be invoked on the IDownloadTarget after fail()."""
1274 def register_canceller(cb):
1275 """The FileDownloader uses this to register a no-argument function
1276 that the target can call to cancel the download. Once this canceller
1277 is invoked, no further calls to write() or close() will be made."""
1279 """When the FileDownloader is done, this finish() function will be
1280 called. Whatever it returns will be returned to the invoker of
1281 Downloader.download.
1284 class IDownloader(Interface):
1285 def download(uri, target):
1286 """Perform a CHK download, sending the data to the given target.
1287 'target' must provide IDownloadTarget.
1289 Returns a Deferred that fires (with the results of target.finish)
1290 when the download is finished, or errbacks if something went wrong."""
1292 class IEncryptedUploadable(Interface):
1293 def set_upload_status(upload_status):
1294 """Provide an IUploadStatus object that should be filled with status
1295 information. The IEncryptedUploadable is responsible for setting
1296 key-determination progress ('chk'), size, storage_index, and
1297 ciphertext-fetch progress. It may delegate some of this
1298 responsibility to others, in particular to the IUploadable."""
1301 """This behaves just like IUploadable.get_size()."""
1303 def get_all_encoding_parameters():
1304 """Return a Deferred that fires with a tuple of
1305 (k,happy,n,segment_size). The segment_size will be used as-is, and
1306 must match the following constraints: it must be a multiple of k, and
1307 it shouldn't be unreasonably larger than the file size (if
1308 segment_size is larger than filesize, the difference must be stored
1311 This usually passes through to the IUploadable method of the same
1314 The encoder strictly obeys the values returned by this method. To
1315 make an upload use non-default encoding parameters, you must arrange
1316 to control the values that this method returns.
1319 def get_storage_index():
1320 """Return a Deferred that fires with a 16-byte storage index.
1323 def read_encrypted(length, hash_only):
1324 """This behaves just like IUploadable.read(), but returns crypttext
1325 instead of plaintext. If hash_only is True, then this discards the
1326 data (and returns an empty list); this improves efficiency when
1327 resuming an interrupted upload (where we need to compute the
1328 plaintext hashes, but don't need the redundant encrypted data)."""
1330 def get_plaintext_hashtree_leaves(first, last, num_segments):
1331 """Get the leaf nodes of a merkle hash tree over the plaintext
1332 segments, i.e. get the tagged hashes of the given segments. The
1333 segment size is expected to be generated by the IEncryptedUploadable
1334 before any plaintext is read or ciphertext produced, so that the
1335 segment hashes can be generated with only a single pass.
1337 This returns a Deferred which fires with a sequence of hashes, using:
1339 tuple(segment_hashes[first:last])
1341 'num_segments' is used to assert that the number of segments that the
1342 IEncryptedUploadable handled matches the number of segments that the
1343 encoder was expecting.
1345 This method must not be called until the final byte has been read
1346 from read_encrypted(). Once this method is called, read_encrypted()
1347 can never be called again.
1350 def get_plaintext_hash():
1351 """Get the hash of the whole plaintext.
1353 This returns a Deferred which fires with a tagged SHA-256 hash of the
1354 whole plaintext, obtained from hashutil.plaintext_hash(data).
1358 """Just like IUploadable.close()."""
1360 class IUploadable(Interface):
1361 def set_upload_status(upload_status):
1362 """Provide an IUploadStatus object that should be filled with status
1363 information. The IUploadable is responsible for setting
1364 key-determination progress ('chk')."""
1366 def set_default_encoding_parameters(params):
1367 """Set the default encoding parameters, which must be a dict mapping
1368 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1369 'max_segment_size'. These might have an influence on the final
1370 encoding parameters returned by get_all_encoding_parameters(), if the
1371 Uploadable doesn't have more specific preferences.
1373 This call is optional: if it is not used, the Uploadable will use
1374 some built-in defaults. If used, this method must be called before
1375 any other IUploadable methods to have any effect.
1379 """Return a Deferred that will fire with the length of the data to be
1380 uploaded, in bytes. This will be called before the data is actually
1381 used, to compute encoding parameters.
1384 def get_all_encoding_parameters():
1385 """Return a Deferred that fires with a tuple of
1386 (k,happy,n,segment_size). The segment_size will be used as-is, and
1387 must match the following constraints: it must be a multiple of k, and
1388 it shouldn't be unreasonably larger than the file size (if
1389 segment_size is larger than filesize, the difference must be stored
1392 The relative values of k and n allow some IUploadables to request
1393 better redundancy than others (in exchange for consuming more space
1396 Larger values of segment_size reduce hash overhead, while smaller
1397 values reduce memory footprint and cause data to be delivered in
1398 smaller pieces (which may provide a smoother and more predictable
1399 download experience).
1401 The encoder strictly obeys the values returned by this method. To
1402 make an upload use non-default encoding parameters, you must arrange
1403 to control the values that this method returns. One way to influence
1404 them may be to call set_encoding_parameters() before calling
1405 get_all_encoding_parameters().
1408 def get_encryption_key():
1409 """Return a Deferred that fires with a 16-byte AES key. This key will
1410 be used to encrypt the data. The key will also be hashed to derive
1413 Uploadables which want to achieve convergence should hash their file
1414 contents and the serialized_encoding_parameters to form the key
1415 (which of course requires a full pass over the data). Uploadables can
1416 use the upload.ConvergentUploadMixin class to achieve this
1419 Uploadables which do not care about convergence (or do not wish to
1420 make multiple passes over the data) can simply return a
1421 strongly-random 16 byte string.
1423 get_encryption_key() may be called multiple times: the IUploadable is
1424 required to return the same value each time.
1428 """Return a Deferred that fires with a list of strings (perhaps with
1429 only a single element) which, when concatenated together, contain the
1430 next 'length' bytes of data. If EOF is near, this may provide fewer
1431 than 'length' bytes. The total number of bytes provided by read()
1432 before it signals EOF must equal the size provided by get_size().
1434 If the data must be acquired through multiple internal read
1435 operations, returning a list instead of a single string may help to
1436 reduce string copies.
1438 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1439 so a 10kB file means length=3kB, 100kB file means length=30kB,
1440 and >=1MB file means length=300kB.
1442 This method provides for a single full pass through the data. Later
1443 use cases may desire multiple passes or access to only parts of the
1444 data (such as a mutable file making small edits-in-place). This API
1445 will be expanded once those use cases are better understood.
1449 """The upload is finished, and whatever filehandle was in use may be
1452 class IUploadResults(Interface):
1453 """I am returned by upload() methods. I contain a number of public
1454 attributes which can be read to determine the results of the upload. Some
1455 of these are functional, some are timing information. All of these may be
1458 .file_size : the size of the file, in bytes
1459 .uri : the CHK read-cap for the file
1460 .ciphertext_fetched : how many bytes were fetched by the helper
1461 .sharemap : dict mapping share number to placement string
1462 .servermap : dict mapping server peerid to a set of share numbers
1463 .timings : dict of timing information, mapping name to seconds (float)
1464 total : total upload time, start to finish
1465 storage_index : time to compute the storage index
1466 peer_selection : time to decide which peers will be used
1467 contacting_helper : initial helper query to upload/no-upload decision
1468 existence_check : helper pre-upload existence check
1469 helper_total : initial helper query to helper finished pushing
1470 cumulative_fetch : helper waiting for ciphertext requests
1471 total_fetch : helper start to last ciphertext response
1472 cumulative_encoding : just time spent in zfec
1473 cumulative_sending : just time spent waiting for storage servers
1474 hashes_and_close : last segment push to shareholder close
1475 total_encode_and_push : first encode to shareholder close
1479 class IDownloadResults(Interface):
1480 """I am created internally by download() methods. I contain a number of
1481 public attributes which contain details about the download process.::
1483 .file_size : the size of the file, in bytes
1484 .servers_used : set of server peerids that were used during download
1485 .server_problems : dict mapping server peerid to a problem string. Only
1486 servers that had problems (bad hashes, disconnects) are
1488 .servermap : dict mapping server peerid to a set of share numbers. Only
1489 servers that had any shares are listed here.
1490 .timings : dict of timing information, mapping name to seconds (float)
1491 peer_selection : time to ask servers about shares
1492 servers_peer_selection : dict of peerid to DYHB-query time
1493 uri_extension : time to fetch a copy of the URI extension block
1494 hashtrees : time to fetch the hash trees
1495 segments : time to fetch, decode, and deliver segments
1496 cumulative_fetch : time spent waiting for storage servers
1497 cumulative_decode : just time spent in zfec
1498 cumulative_decrypt : just time spent in decryption
1499 total : total download time, start to finish
1500 fetch_per_server : dict of peerid to list of per-segment fetch times
1504 class IUploader(Interface):
1505 def upload(uploadable):
1506 """Upload the file. 'uploadable' must impement IUploadable. This
1507 returns a Deferred which fires with an UploadResults instance, from
1508 which the URI of the file can be obtained as results.uri ."""
1510 def upload_ssk(write_capability, new_version, uploadable):
1511 """TODO: how should this work?"""
1513 class ICheckable(Interface):
1514 def check(monitor, verify=False):
1515 """Check upon my health, optionally repairing any problems.
1517 This returns a Deferred that fires with an instance that provides
1518 ICheckerResults, or None if the object is non-distributed (i.e. LIT
1521 The monitor will be checked periodically to see if the operation has
1522 been cancelled. If so, no new queries will be sent, and the Deferred
1523 will fire (with a OperationCancelledError) immediately.
1525 Filenodes and dirnodes (which provide IFilesystemNode) are also
1526 checkable. Instances that represent verifier-caps will be checkable
1527 but not downloadable. Some objects (like LIT files) do not actually
1528 live in the grid, and their checkers return None (non-distributed
1529 files are always healthy).
1531 If verify=False, a relatively lightweight check will be performed: I
1532 will ask all servers if they have a share for me, and I will believe
1533 whatever they say. If there are at least N distinct shares on the
1534 grid, my results will indicate r.is_healthy()==True. This requires a
1535 roundtrip to each server, but does not transfer very much data, so
1536 the network bandwidth is fairly low.
1538 If verify=True, a more resource-intensive check will be performed:
1539 every share will be downloaded, and the hashes will be validated on
1540 every bit. I will ignore any shares that failed their hash checks. If
1541 there are at least N distinct valid shares on the grid, my results
1542 will indicate r.is_healthy()==True. This requires N/k times as much
1543 download bandwidth (and server disk IO) as a regular download. If a
1544 storage server is holding a corrupt share, or is experiencing memory
1545 failures during retrieval, or is malicious or buggy, then
1546 verification will detect the problem, but checking will not.
1548 TODO: any problems seen during checking will be reported to the
1549 health-manager.furl, a centralized object which is responsible for
1550 figuring out why files are unhealthy so corrective action can be
1554 def check_and_repair(monitor, verify=False):
1555 """Like check(), but if the file/directory is not healthy, attempt to
1558 Any non-healthy result will cause an immediate repair operation, to
1559 generate and upload new shares. After repair, the file will be as
1560 healthy as we can make it. Details about what sort of repair is done
1561 will be put in the check-and-repair results. The Deferred will not
1562 fire until the repair is complete.
1564 This returns a Deferred which fires with an instance of
1565 ICheckAndRepairResults."""
1567 class IDeepCheckable(Interface):
1568 def start_deep_check(verify=False):
1569 """Check upon the health of me and everything I can reach.
1571 This is a recursive form of check(), useable only on dirnodes.
1573 I return a Monitor, with results that are an IDeepCheckResults
1577 def start_deep_check_and_repair(verify=False):
1578 """Check upon the health of me and everything I can reach. Repair
1579 anything that isn't healthy.
1581 This is a recursive form of check_and_repair(), useable only on
1584 I return a Monitor, with results that are an
1585 IDeepCheckAndRepairResults object.
1588 class ICheckerResults(Interface):
1589 """I contain the detailed results of a check/verify operation.
1592 def get_storage_index():
1593 """Return a string with the (binary) storage index."""
1594 def get_storage_index_string():
1595 """Return a string with the (printable) abbreviated storage index."""
1597 """Return the (string) URI of the object that was checked."""
1600 """Return a boolean, True if the file/dir is fully healthy, False if
1601 it is damaged in any way. Non-distributed LIT files always return
1604 def is_recoverable():
1605 """Return a boolean, True if the file/dir can be recovered, False if
1606 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1607 files always return True."""
1609 def needs_rebalancing():
1610 """Return a boolean, True if the file/dir's reliability could be
1611 improved by moving shares to new servers. Non-distributed LIT files
1612 always return False."""
1616 """Return a dictionary that describes the state of the file/dir. LIT
1617 files always return an empty dictionary. Normal files and directories return a
1618 dictionary with the following keys (note that these use binary strings rather than
1619 base32-encoded ones) (also note that for mutable files, these counts are for the 'best'
1622 count-shares-good: the number of distinct good shares that were found
1623 count-shares-needed: 'k', the number of shares required for recovery
1624 count-shares-expected: 'N', the number of total shares generated
1625 count-good-share-hosts: the number of distinct storage servers with
1626 good shares. If this number is less than
1627 count-shares-good, then some shares are
1628 doubled up, increasing the correlation of
1629 failures. This indicates that one or more
1630 shares should be moved to an otherwise unused
1631 server, if one is available.
1632 count-corrupt-shares: the number of shares with integrity failures
1633 list-corrupt-shares: a list of 'share locators', one for each share
1634 that was found to be corrupt. Each share
1635 locator is a list of (serverid, storage_index,
1637 servers-responding: list of (binary) storage server identifiers,
1638 one for each server which responded to the share
1639 query (even if they said they didn't have shares,
1640 and even if they said they did have shares but then
1641 refused to send them when asked, and even if they
1642 said they did have shares and sent incorrect ones
1644 sharemap: dict mapping share identifier to list of serverids
1645 (binary strings). This indicates which servers are holding
1646 which shares. For immutable files, the shareid is an
1647 integer (the share number, from 0 to N-1). For mutable
1648 files, it is a string of the form 'seq%d-%s-sh%d',
1649 containing the sequence number, the roothash, and the
1652 The following keys are most relevant for mutable files, but immutable
1653 files will provide sensible values too::
1655 count-wrong-shares: the number of shares for versions other than the
1656 'best' one (which is defined as being the
1657 recoverable version with the highest sequence
1658 number, then the highest roothash). These are
1659 either leftover shares from an older version
1660 (perhaps on a server that was offline when an
1661 update occurred), shares from an unrecoverable
1662 newer version, or shares from an alternate
1663 current version that results from an
1664 uncoordinated write collision. For a healthy
1665 file, this will equal 0.
1667 count-recoverable-versions: the number of recoverable versions of
1668 the file. For a healthy file, this will
1671 count-unrecoverable-versions: the number of unrecoverable versions
1672 of the file. For a healthy file, this
1678 """Return a string with a brief (one-line) summary of the results."""
1681 """Return a list of strings with more detailed results."""
1683 class ICheckAndRepairResults(Interface):
1684 """I contain the detailed results of a check/verify/repair operation.
1686 The IFilesystemNode.check()/verify()/repair() methods all return
1687 instances that provide ICheckAndRepairResults.
1690 def get_storage_index():
1691 """Return a string with the (binary) storage index."""
1692 def get_storage_index_string():
1693 """Return a string with the (printable) abbreviated storage index."""
1694 def get_repair_attempted():
1695 """Return a boolean, True if a repair was attempted."""
1696 def get_repair_successful():
1697 """Return a boolean, True if repair was attempted and the file/dir
1698 was fully healthy afterwards. False if no repair was attempted or if
1699 a repair attempt failed."""
1700 def get_pre_repair_results():
1701 """Return an ICheckerResults instance that describes the state of the
1702 file/dir before any repair was attempted."""
1703 def get_post_repair_results():
1704 """Return an ICheckerResults instance that describes the state of the
1705 file/dir after any repair was attempted. If no repair was attempted,
1706 the pre-repair and post-repair results will be identical."""
1709 class IDeepCheckResults(Interface):
1710 """I contain the results of a deep-check operation.
1712 This is returned by a call to ICheckable.deep_check().
1715 def get_root_storage_index_string():
1716 """Return the storage index (abbreviated human-readable string) of
1717 the first object checked."""
1719 """Return a dictionary with the following keys::
1721 count-objects-checked: count of how many objects were checked
1722 count-objects-healthy: how many of those objects were completely
1724 count-objects-unhealthy: how many were damaged in some way
1725 count-objects-unrecoverable: how many were unrecoverable
1726 count-corrupt-shares: how many shares were found to have
1727 corruption, summed over all objects
1731 def get_corrupt_shares():
1732 """Return a set of (serverid, storage_index, sharenum) for all shares
1733 that were found to be corrupt. Both serverid and storage_index are
1736 def get_all_results():
1737 """Return a dictionary mapping pathname (a tuple of strings, ready to
1738 be slash-joined) to an ICheckerResults instance, one for each object
1739 that was checked."""
1741 def get_results_for_storage_index(storage_index):
1742 """Retrive the ICheckerResults instance for the given (binary)
1743 storage index. Raises KeyError if there are no results for that
1747 """Return a dictionary with the same keys as
1748 IDirectoryNode.deep_stats()."""
1750 class IDeepCheckAndRepairResults(Interface):
1751 """I contain the results of a deep-check-and-repair operation.
1753 This is returned by a call to ICheckable.deep_check_and_repair().
1756 def get_root_storage_index_string():
1757 """Return the storage index (abbreviated human-readable string) of
1758 the first object checked."""
1760 """Return a dictionary with the following keys::
1762 count-objects-checked: count of how many objects were checked
1763 count-objects-healthy-pre-repair: how many of those objects were
1764 completely healthy (before any
1766 count-objects-unhealthy-pre-repair: how many were damaged in
1768 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1769 count-objects-healthy-post-repair: how many of those objects were
1770 completely healthy (after any
1772 count-objects-unhealthy-post-repair: how many were damaged in
1774 count-objects-unrecoverable-post-repair: how many were
1776 count-repairs-attempted: repairs were attempted on this many
1777 objects. The count-repairs- keys will
1778 always be provided, however unless
1779 repair=true is present, they will all
1781 count-repairs-successful: how many repairs resulted in healthy
1783 count-repairs-unsuccessful: how many repairs resulted did not
1784 results in completely healthy objects
1785 count-corrupt-shares-pre-repair: how many shares were found to
1786 have corruption, summed over all
1787 objects examined (before any
1789 count-corrupt-shares-post-repair: how many shares were found to
1790 have corruption, summed over all
1791 objects examined (after any
1796 """Return a dictionary with the same keys as
1797 IDirectoryNode.deep_stats()."""
1799 def get_corrupt_shares():
1800 """Return a set of (serverid, storage_index, sharenum) for all shares
1801 that were found to be corrupt before any repair was attempted. Both
1802 serverid and storage_index are binary.
1804 def get_remaining_corrupt_shares():
1805 """Return a set of (serverid, storage_index, sharenum) for all shares
1806 that were found to be corrupt after any repair was completed. Both
1807 serverid and storage_index are binary. These are shares that need
1808 manual inspection and probably deletion.
1810 def get_all_results():
1811 """Return a dictionary mapping pathname (a tuple of strings, ready to
1812 be slash-joined) to an ICheckAndRepairResults instance, one for each
1813 object that was checked."""
1816 class IRepairable(Interface):
1817 def repair(checker_results):
1818 """Attempt to repair the given object. Returns a Deferred that fires
1819 with a IRepairResults object.
1821 I must be called with an object that implements ICheckerResults, as
1822 proof that you have actually discovered a problem with this file. I
1823 will use the data in the checker results to guide the repair process,
1824 such as which servers provided bad data and should therefore be
1825 avoided. The ICheckerResults object is inside the
1826 ICheckAndRepairResults object, which is returned by the
1827 ICheckable.check() method::
1829 d = filenode.check(repair=False)
1830 def _got_results(check_and_repair_results):
1831 check_results = check_and_repair_results.get_pre_repair_results()
1832 return filenode.repair(check_results)
1833 d.addCallback(_got_results)
1837 class IRepairResults(Interface):
1838 """I contain the results of a repair operation."""
1841 class IClient(Interface):
1842 def upload(uploadable):
1843 """Upload some data into a CHK, get back the UploadResults for it.
1844 @param uploadable: something that implements IUploadable
1845 @return: a Deferred that fires with the UploadResults instance.
1846 To get the URI for this file, use results.uri .
1849 def create_mutable_file(contents=""):
1850 """Create a new mutable file with contents, get back the URI string.
1851 @param contents: the initial contents to place in the file.
1852 @return: a Deferred that fires with tne (string) SSK URI for the new
1856 def create_empty_dirnode():
1857 """Create a new dirnode, empty and unattached.
1858 @return: a Deferred that fires with the new IDirectoryNode instance.
1861 def create_node_from_uri(uri):
1862 """Create a new IFilesystemNode instance from the uri, synchronously.
1863 @param uri: a string or IURI-providing instance. This could be for a
1864 LiteralFileNode, a CHK file node, a mutable file node, or
1866 @return: an instance that provides IFilesystemNode (or more usefully one
1867 of its subclasses). File-specifying URIs will result in
1868 IFileNode or IMutableFileNode -providing instances, like
1869 FileNode, LiteralFileNode, or MutableFileNode.
1870 Directory-specifying URIs will result in
1871 IDirectoryNode-providing instances, like NewDirectoryNode.
1874 class IClientStatus(Interface):
1875 def list_all_uploads():
1876 """Return a list of uploader objects, one for each upload which
1877 currently has an object available (tracked with weakrefs). This is
1878 intended for debugging purposes."""
1879 def list_active_uploads():
1880 """Return a list of active IUploadStatus objects."""
1881 def list_recent_uploads():
1882 """Return a list of IUploadStatus objects for the most recently
1885 def list_all_downloads():
1886 """Return a list of downloader objects, one for each download which
1887 currently has an object available (tracked with weakrefs). This is
1888 intended for debugging purposes."""
1889 def list_active_downloads():
1890 """Return a list of active IDownloadStatus objects."""
1891 def list_recent_downloads():
1892 """Return a list of IDownloadStatus objects for the most recently
1893 started downloads."""
1895 class IUploadStatus(Interface):
1897 """Return a timestamp (float with seconds since epoch) indicating
1898 when the operation was started."""
1899 def get_storage_index():
1900 """Return a string with the (binary) storage index in use on this
1901 upload. Returns None if the storage index has not yet been
1904 """Return an integer with the number of bytes that will eventually
1905 be uploaded for this file. Returns None if the size is not yet known.
1908 """Return True if this upload is using a Helper, False if not."""
1910 """Return a string describing the current state of the upload
1913 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
1914 each from 0.0 to 1.0 . 'chk' describes how much progress has been
1915 made towards hashing the file to determine a CHK encryption key: if
1916 non-convergent encryption is in use, this will be trivial, otherwise
1917 the whole file must be hashed. 'ciphertext' describes how much of the
1918 ciphertext has been pushed to the helper, and is '1.0' for non-helper
1919 uploads. 'encode_and_push' describes how much of the encode-and-push
1920 process has finished: for helper uploads this is dependent upon the
1921 helper providing progress reports. It might be reasonable to add all
1922 three numbers and report the sum to the user."""
1924 """Return True if the upload is currently active, False if not."""
1926 """Return an instance of UploadResults (which contains timing and
1927 sharemap information). Might return None if the upload is not yet
1930 """Each upload status gets a unique number: this method returns that
1931 number. This provides a handle to this particular upload, so a web
1932 page can generate a suitable hyperlink."""
1934 class IDownloadStatus(Interface):
1936 """Return a timestamp (float with seconds since epoch) indicating
1937 when the operation was started."""
1938 def get_storage_index():
1939 """Return a string with the (binary) storage index in use on this
1940 download. This may be None if there is no storage index (i.e. LIT
1943 """Return an integer with the number of bytes that will eventually be
1944 retrieved for this file. Returns None if the size is not yet known.
1947 """Return True if this download is using a Helper, False if not."""
1949 """Return a string describing the current state of the download
1952 """Returns a float (from 0.0 to 1.0) describing the amount of the
1953 download that has completed. This value will remain at 0.0 until the
1954 first byte of plaintext is pushed to the download target."""
1956 """Return True if the download is currently active, False if not."""
1958 """Each download status gets a unique number: this method returns
1959 that number. This provides a handle to this particular download, so a
1960 web page can generate a suitable hyperlink."""
1962 class IServermapUpdaterStatus(Interface):
1964 class IPublishStatus(Interface):
1966 class IRetrieveStatus(Interface):
1969 class NotCapableError(Exception):
1970 """You have tried to write to a read-only node."""
1972 class BadWriteEnablerError(Exception):
1975 class RIControlClient(RemoteInterface):
1977 def wait_for_client_connections(num_clients=int):
1978 """Do not return until we have connections to at least NUM_CLIENTS
1982 def upload_from_file_to_uri(filename=str, convergence=ChoiceOf(None, StringConstraint(2**20))):
1983 """Upload a file to the grid. This accepts a filename (which must be
1984 absolute) that points to a file on the node's local disk. The node will
1985 read the contents of this file, upload it to the grid, then return the
1986 URI at which it was uploaded. If convergence is None then a random
1987 encryption key will be used, else the plaintext will be hashed, then
1988 that hash will be mixed together with the "convergence" string to form
1993 def download_from_uri_to_file(uri=URI, filename=str):
1994 """Download a file from the grid, placing it on the node's local disk
1995 at the given filename (which must be absolute[?]). Returns the
1996 absolute filename where the file was written."""
2001 def get_memory_usage():
2002 """Return a dict describes the amount of memory currently in use. The
2003 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2004 measuring memory consupmtion in bytes."""
2005 return DictOf(str, int)
2007 def speed_test(count=int, size=int, mutable=Any()):
2008 """Write 'count' tempfiles to disk, all of the given size. Measure
2009 how long (in seconds) it takes to upload them all to the servers.
2010 Then measure how long it takes to download all of them. If 'mutable'
2011 is 'create', time creation of mutable files. If 'mutable' is
2012 'upload', then time access to the same mutable file instead of
2015 Returns a tuple of (upload_time, download_time).
2017 return (float, float)
2019 def measure_peer_response_time():
2020 """Send a short message to each connected peer, and measure the time
2021 it takes for them to respond to it. This is a rough measure of the
2022 application-level round trip time.
2024 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2027 return DictOf(Nodeid, float)
2029 UploadResults = Any() #DictOf(str, str)
2031 class RIEncryptedUploadable(RemoteInterface):
2032 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2037 def get_all_encoding_parameters():
2038 return (int, int, int, long)
2040 def read_encrypted(offset=Offset, length=ReadSize):
2043 def get_plaintext_hashtree_leaves(first=int, last=int, num_segments=int):
2046 def get_plaintext_hash():
2053 class RICHKUploadHelper(RemoteInterface):
2054 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2058 Return a dictionary of version information.
2060 return DictOf(str, Any())
2062 def upload(reader=RIEncryptedUploadable):
2063 return UploadResults
2066 class RIHelper(RemoteInterface):
2067 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2071 Return a dictionary of version information.
2073 return DictOf(str, Any())
2075 def upload_chk(si=StorageIndex):
2076 """See if a file with a given storage index needs uploading. The
2077 helper will ask the appropriate storage servers to see if the file
2078 has already been uploaded. If so, the helper will return a set of
2079 'upload results' that includes whatever hashes are needed to build
2080 the read-cap, and perhaps a truncated sharemap.
2082 If the file has not yet been uploaded (or if it was only partially
2083 uploaded), the helper will return an empty upload-results dictionary
2084 and also an RICHKUploadHelper object that will take care of the
2085 upload process. The client should call upload() on this object and
2086 pass it a reference to an RIEncryptedUploadable object that will
2087 provide ciphertext. When the upload is finished, the upload() method
2088 will finish and return the upload results.
2090 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2093 class RIStatsProvider(RemoteInterface):
2094 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2096 Provides access to statistics and monitoring information.
2101 returns a dictionary containing 'counters' and 'stats', each a dictionary
2102 with string counter/stat name keys, and numeric values. counters are
2103 monotonically increasing measures of work done, and stats are instantaneous
2104 measures (potentially time averaged internally)
2106 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2108 class RIStatsGatherer(RemoteInterface):
2109 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2111 Provides a monitoring service for centralised collection of stats
2114 def provide(provider=RIStatsProvider, nickname=str):
2116 @param provider: a stats collector instance which should be polled
2117 periodically by the gatherer to collect stats.
2118 @param nickname: a name useful to identify the provided client
2123 class IStatsProducer(Interface):
2126 returns a dictionary, with str keys representing the names of stats
2127 to be monitored, and numeric values.
2130 class RIKeyGenerator(RemoteInterface):
2131 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2133 Provides a service offering to make RSA key pairs.
2136 def get_rsa_key_pair(key_size=int):
2138 @param key_size: the size of the signature key.
2139 @return: tuple(verifying_key, signing_key)
2141 return TupleOf(str, str)
2144 class FileTooLargeError(Exception):
2147 class IValidatedThingProxy(Interface):
2149 """ Acquire a thing and validate it. Return a deferred which is eventually fired with
2150 self if the thing is valid or errbacked if it can't be acquired or validated. """
2152 class InsufficientVersionError(Exception):
2153 def __init__(self, needed, got):
2154 self.needed = needed
2157 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,