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_plaintext_hashes():
318 @return: ListOf(Hash)
321 def get_crypttext_hashes():
323 @return: ListOf(Hash)
326 def get_block_hashes():
328 @return: ListOf(Hash)
331 def get_share_hashes():
333 @return: ListOf(TupleOf(int, Hash))
336 def get_uri_extension():
338 @return: URIExtensionData
343 # hm, we need a solution for forward references in schemas
344 FileNode_ = Any() # TODO: foolscap needs constraints on copyables
345 DirectoryNode_ = Any() # TODO: same
346 AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
349 class IURI(Interface):
350 def init_from_string(uri):
351 """Accept a string (as created by my to_string() method) and populate
352 this instance with its data. I am not normally called directly,
353 please use the module-level uri.from_string() function to convert
354 arbitrary URI strings into IURI-providing instances."""
357 """Return False if this URI be used to modify the data. Return True
358 if this URI cannot be used to modify the data."""
361 """Return True if the data can be modified by *somebody* (perhaps
362 someone who has a more powerful URI than this one)."""
365 """Return another IURI instance, which represents a read-only form of
366 this one. If is_readonly() is True, this returns self."""
368 def get_verify_cap():
369 """Return an instance that provides IVerifierURI, which can be used
370 to check on the availability of the file or directory, without
371 providing enough capabilities to actually read or modify the
372 contents. This may return None if the file does not need checking or
373 verification (e.g. LIT URIs).
377 """Return a string of printable ASCII characters, suitable for
378 passing into init_from_string."""
380 class IVerifierURI(Interface):
381 def init_from_string(uri):
382 """Accept a string (as created by my to_string() method) and populate
383 this instance with its data. I am not normally called directly,
384 please use the module-level uri.from_string() function to convert
385 arbitrary URI strings into IURI-providing instances."""
388 """Return a string of printable ASCII characters, suitable for
389 passing into init_from_string."""
391 class IDirnodeURI(Interface):
392 """I am a URI which represents a dirnode."""
395 class IFileURI(Interface):
396 """I am a URI which represents a filenode."""
398 """Return the length (in bytes) of the file that I represent."""
400 class IMutableFileURI(Interface):
401 """I am a URI which represents a mutable filenode."""
402 class INewDirectoryURI(Interface):
404 class IReadonlyNewDirectoryURI(Interface):
408 class IFilesystemNode(Interface):
411 Return the URI that can be used by others to get access to this
412 node. If this node is read-only, the URI will only offer read-only
413 access. If this node is read-write, the URI will offer read-write
416 If you have read-write access to a node and wish to share merely
417 read-only access with others, use get_readonly_uri().
420 def get_readonly_uri():
421 """Return the directory URI that can be used by others to get
422 read-only access to this directory node. The result is a read-only
423 URI, regardless of whether this dirnode is read-only or read-write.
425 If you have merely read-only access to this dirnode,
426 get_readonly_uri() will return the same thing as get_uri().
429 def get_verify_cap():
430 """Return an IVerifierURI instance that represents the
431 'verifiy/refresh capability' for this node. The holder of this
432 capability will be able to renew the lease for this node, protecting
433 it from garbage-collection. They will also be able to ask a server if
434 it holds a share for the file or directory.
437 def get_storage_index():
438 """Return a string with the (binary) storage index in use on this
439 download. This may be None if there is no storage index (i.e. LIT
443 """Return True if this reference provides mutable access to the given
444 file or directory (i.e. if you can modify it), or False if not. Note
445 that even if this reference is read-only, someone else may hold a
446 read-write reference to it."""
449 """Return True if this file or directory is mutable (by *somebody*,
450 not necessarily you), False if it is is immutable. Note that a file
451 might be mutable overall, but your reference to it might be
452 read-only. On the other hand, all references to an immutable file
453 will be read-only; there are no read-write references to an immutable
457 class IMutableFilesystemNode(IFilesystemNode):
460 class IFileNode(IFilesystemNode):
461 def download(target):
462 """Download the file's contents to a given IDownloadTarget"""
464 def download_to_data():
465 """Download the file's contents. Return a Deferred that fires
466 with those contents."""
469 """Return the length (in bytes) of the data this node represents."""
471 def read(consumer, offset=0, size=None):
472 """Download a portion (possibly all) of the file's contents, making
473 them available to the given IConsumer. Return a Deferred that fires
474 (with the consumer) when the consumer is unregistered (either because
475 the last byte has been given to it, or because the consumer threw an
476 exception during write(), possibly because it no longer wants to
477 receive data). The portion downloaded will start at 'offset' and
478 contain 'size' bytes (or the remainder of the file if size==None).
480 The consumer will be used in non-streaming mode: an IPullProducer
481 will be attached to it.
483 The consumer will not receive data right away: several network trips
484 must occur first. The order of events will be::
486 consumer.registerProducer(p, streaming)
487 (if streaming == False)::
488 consumer does p.resumeProducing()
490 consumer does p.resumeProducing()
491 consumer.write(data).. (repeat until all data is written)
492 consumer.unregisterProducer()
493 deferred.callback(consumer)
495 If a download error occurs, or an exception is raised by
496 consumer.registerProducer() or consumer.write(), I will call
497 consumer.unregisterProducer() and then deliver the exception via
498 deferred.errback(). To cancel the download, the consumer should call
499 p.stopProducing(), which will result in an exception being delivered
500 via deferred.errback().
502 A simple download-to-memory consumer example would look like this::
504 class MemoryConsumer:
505 implements(IConsumer)
509 def registerProducer(self, p, streaming):
510 assert streaming == False
513 def write(self, data):
514 self.chunks.append(data)
515 def unregisterProducer(self):
517 d = filenode.read(MemoryConsumer())
518 d.addCallback(lambda mc: "".join(mc.chunks))
523 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
524 """I provide access to a 'mutable file', which retains its identity
525 regardless of what contents are put in it.
527 The consistency-vs-availability problem means that there might be
528 multiple versions of a file present in the grid, some of which might be
529 unrecoverable (i.e. have fewer than 'k' shares). These versions are
530 loosely ordered: each has a sequence number and a hash, and any version
531 with seqnum=N was uploaded by a node which has seen at least one version
534 The 'servermap' (an instance of IMutableFileServerMap) is used to
535 describe the versions that are known to be present in the grid, and which
536 servers are hosting their shares. It is used to represent the 'state of
537 the world', and is used for this purpose by my test-and-set operations.
538 Downloading the contents of the mutable file will also return a
539 servermap. Uploading a new version into the mutable file requires a
540 servermap as input, and the semantics of the replace operation is
541 'replace the file with my new version if it looks like nobody else has
542 changed the file since my previous download'. Because the file is
543 distributed, this is not a perfect test-and-set operation, but it will do
544 its best. If the replace process sees evidence of a simultaneous write,
545 it will signal an UncoordinatedWriteError, so that the caller can take
549 Most readers will want to use the 'best' current version of the file, and
550 should use my 'download_best_version()' method.
552 To unconditionally replace the file, callers should use overwrite(). This
553 is the mode that user-visible mutable files will probably use.
555 To apply some delta to the file, call modify() with a callable modifier
556 function that can apply the modification that you want to make. This is
557 the mode that dirnodes will use, since most directory modification
558 operations can be expressed in terms of deltas to the directory state.
561 Three methods are available for users who need to perform more complex
562 operations. The first is get_servermap(), which returns an up-to-date
563 servermap using a specified mode. The second is download_version(), which
564 downloads a specific version (not necessarily the 'best' one). The third
565 is 'upload', which accepts new contents and a servermap (which must have
566 been updated with MODE_WRITE). The upload method will attempt to apply
567 the new contents as long as no other node has modified the file since the
568 servermap was updated. This might be useful to a caller who wants to
569 merge multiple versions into a single new one.
571 Note that each time the servermap is updated, a specific 'mode' is used,
572 which determines how many peers are queried. To use a servermap for my
573 replace() method, that servermap must have been updated in MODE_WRITE.
574 These modes are defined in allmydata.mutable.common, and consist of
575 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
576 allmydata/mutable/servermap.py for details about the differences.
578 Mutable files are currently limited in size (about 3.5MB max) and can
579 only be retrieved and updated all-at-once, as a single big string. Future
580 versions of our mutable files will remove this restriction.
583 def download_best_version():
584 """Download the 'best' available version of the file, meaning one of
585 the recoverable versions with the highest sequence number. If no
586 uncoordinated writes have occurred, and if enough shares are
587 available, then this will be the most recent version that has been
590 I update an internal servermap with MODE_READ, determine which
591 version of the file is indicated by
592 servermap.best_recoverable_version(), and return a Deferred that
593 fires with its contents. If no version is recoverable, the Deferred
594 will errback with UnrecoverableFileError.
597 def get_size_of_best_version():
598 """Find the size of the version that would be downloaded with
599 download_best_version(), without actually downloading the whole file.
601 I return a Deferred that fires with an integer.
604 def overwrite(new_contents):
605 """Unconditionally replace the contents of the mutable file with new
606 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
607 is only appropriate to use when the new contents of the file are
608 completely unrelated to the old ones, and you do not care about other
611 I return a Deferred that fires (with a PublishStatus object) when the
612 update has completed.
615 def modify(modifier_cb):
616 """Modify the contents of the file, by downloading the current
617 version, applying the modifier function (or bound method), then
618 uploading the new version. I return a Deferred that fires (with a
619 PublishStatus object) when the update is complete.
621 The modifier callable will be given three arguments: a string (with
622 the old contents), a 'first_time' boolean, and a servermap. As with
623 download_best_version(), the old contents will be from the best
624 recoverable version, but the modifier can use the servermap to make
625 other decisions (such as refusing to apply the delta if there are
626 multiple parallel versions, or if there is evidence of a newer
627 unrecoverable version). 'first_time' will be True the first time the
628 modifier is called, and False on any subsequent calls.
630 The callable should return a string with the new contents. The
631 callable must be prepared to be called multiple times, and must
632 examine the input string to see if the change that it wants to make
633 is already present in the old version. If it does not need to make
634 any changes, it can either return None, or return its input string.
636 If the modifier raises an exception, it will be returned in the
641 def get_servermap(mode):
642 """Return a Deferred that fires with an IMutableFileServerMap
643 instance, updated using the given mode.
646 def download_version(servermap, version):
647 """Download a specific version of the file, using the servermap
648 as a guide to where the shares are located.
650 I return a Deferred that fires with the requested contents, or
651 errbacks with UnrecoverableFileError. Note that a servermap which was
652 updated with MODE_ANYTHING or MODE_READ may not know about shares for
653 all versions (those modes stop querying servers as soon as they can
654 fulfil their goals), so you may want to use MODE_CHECK (which checks
655 everything) to get increased visibility.
658 def upload(new_contents, servermap):
659 """Replace the contents of the file with new ones. This requires a
660 servermap that was previously updated with MODE_WRITE.
662 I attempt to provide test-and-set semantics, in that I will avoid
663 modifying any share that is different than the version I saw in the
664 servermap. However, if another node is writing to the file at the
665 same time as me, I may manage to update some shares while they update
666 others. If I see any evidence of this, I will signal
667 UncoordinatedWriteError, and the file will be left in an inconsistent
668 state (possibly the version you provided, possibly the old version,
669 possibly somebody else's version, and possibly a mix of shares from
672 The recommended response to UncoordinatedWriteError is to either
673 return it to the caller (since they failed to coordinate their
674 writes), or to attempt some sort of recovery. It may be sufficient to
675 wait a random interval (with exponential backoff) and repeat your
676 operation. If I do not signal UncoordinatedWriteError, then I was
677 able to write the new version without incident.
679 I return a Deferred that fires (with a PublishStatus object) when the
680 publish has completed. I will update the servermap in-place with the
681 location of all new shares.
685 """Return this filenode's writekey, or None if the node does not have
686 write-capability. This may be used to assist with data structures
687 that need to make certain data available only to writers, such as the
688 read-write child caps in dirnodes. The recommended process is to have
689 reader-visible data be submitted to the filenode in the clear (where
690 it will be encrypted by the filenode using the readkey), but encrypt
691 writer-visible data using this writekey.
694 class NotEnoughSharesError(Exception):
697 class ExistingChildError(Exception):
698 """A directory node was asked to add or replace a child that already
699 exists, and overwrite= was set to False."""
701 class NoSuchChildError(Exception):
702 """A directory node was asked to fetch a child which does not exist."""
704 class IDirectoryNode(IMutableFilesystemNode):
705 """I represent a name-to-child mapping, holding the tahoe equivalent of a
706 directory. All child names are unicode strings, and all children are some
707 sort of IFilesystemNode (either files or subdirectories).
712 The dirnode ('1') URI returned by this method can be used in
713 set_uri() on a different directory ('2') to 'mount' a reference to
714 this directory ('1') under the other ('2'). This URI is just a
715 string, so it can be passed around through email or other out-of-band
719 def get_readonly_uri():
721 The dirnode ('1') URI returned by this method can be used in
722 set_uri() on a different directory ('2') to 'mount' a reference to
723 this directory ('1') under the other ('2'). This URI is just a
724 string, so it can be passed around through email or other out-of-band
729 """I return a Deferred that fires with a dictionary mapping child
730 name (a unicode string) to (node, metadata_dict) tuples, in which
731 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
732 is a dictionary of metadata."""
735 """I return a Deferred that fires with a boolean, True if there
736 exists a child of the given name, False if not. The child name must
737 be a unicode string."""
740 """I return a Deferred that fires with a specific named child node,
741 either an IFileNode or an IDirectoryNode. The child name must be a
742 unicode string. I raise NoSuchChildError if I do not have a child by
745 def get_metadata_for(name):
746 """I return a Deferred that fires with the metadata dictionary for a
747 specific named child node. This metadata is stored in the *edge*, not
748 in the child, so it is attached to the parent dirnode rather than the
749 child dir-or-file-node. The child name must be a unicode string. I
750 raise NoSuchChildError if I do not have a child by that name."""
752 def set_metadata_for(name, metadata):
753 """I replace any existing metadata for the named child with the new
754 metadata. The child name must be a unicode string. This metadata is
755 stored in the *edge*, not in the child, so it is attached to the
756 parent dirnode rather than the child dir-or-file-node. I return a
757 Deferred (that fires with this dirnode) when the operation is
758 complete. I raise NoSuchChildError if I do not have a child by that
761 def get_child_at_path(path):
762 """Transform a child path into an IDirectoryNode or IFileNode.
764 I perform a recursive series of 'get' operations to find the named
765 descendant node. I return a Deferred that fires with the node, or
766 errbacks with NoSuchChildError if the node could not be found.
768 The path can be either a single string (slash-separated) or a list of
769 path-name elements. All elements must be unicode strings.
772 def get_child_and_metadata_at_path(path):
773 """Transform a child path into an IDirectoryNode/IFileNode and
776 I am like get_child_at_path(), but my Deferred fires with a tuple of
777 (node, metadata). The metadata comes from the last edge. If the path
778 is empty, the metadata will be an empty dictionary.
781 def set_uri(name, child_uri, metadata=None, overwrite=True):
782 """I add a child (by URI) at the specific name. I return a Deferred
783 that fires when the operation finishes. If overwrite= is True, I will
784 replace any existing child of the same name, otherwise an existing
785 child will cause me to return ExistingChildError. The child name must
788 The child_uri could be for a file, or for a directory (either
789 read-write or read-only, using a URI that came from get_uri() ).
791 If metadata= is provided, I will use it as the metadata for the named
792 edge. This will replace any existing metadata. If metadata= is left
793 as the default value of None, I will set ['mtime'] to the current
794 time, and I will set ['ctime'] to the current time if there was not
795 already a child by this name present. This roughly matches the
796 ctime/mtime semantics of traditional filesystems.
798 If this directory node is read-only, the Deferred will errback with a
801 def set_children(entries, overwrite=True):
802 """Add multiple (name, child_uri) pairs (or (name, child_uri,
803 metadata) triples) to a directory node. Returns a Deferred that fires
804 (with None) when the operation finishes. This is equivalent to
805 calling set_uri() multiple times, but is much more efficient. All
806 child names must be unicode strings.
809 def set_node(name, child, metadata=None, overwrite=True):
810 """I add a child at the specific name. I return a Deferred that fires
811 when the operation finishes. This Deferred will fire with the child
812 node that was just added. I will replace any existing child of the
813 same name. The child name must be a unicode string. The 'child'
814 instance must be an instance providing IDirectoryNode or IFileNode.
816 If metadata= is provided, I will use it as the metadata for the named
817 edge. This will replace any existing metadata. If metadata= is left
818 as the default value of None, I will set ['mtime'] to the current
819 time, and I will set ['ctime'] to the current time if there was not
820 already a child by this name present. This roughly matches the
821 ctime/mtime semantics of traditional filesystems.
823 If this directory node is read-only, the Deferred will errback with a
826 def set_nodes(entries, overwrite=True):
827 """Add multiple (name, child_node) pairs (or (name, child_node,
828 metadata) triples) to a directory node. Returns a Deferred that fires
829 (with None) when the operation finishes. This is equivalent to
830 calling set_node() multiple times, but is much more efficient. All
831 child names must be unicode strings."""
834 def add_file(name, uploadable, metadata=None, overwrite=True):
835 """I upload a file (using the given IUploadable), then attach the
836 resulting FileNode to the directory at the given name. I set metadata
837 the same way as set_uri and set_node. The child name must be a
840 I return a Deferred that fires (with the IFileNode of the uploaded
841 file) when the operation completes."""
844 """I remove the child at the specific name. I return a Deferred that
845 fires when the operation finishes. The child name must be a unicode
846 string. I raise NoSuchChildError if I do not have a child by that
849 def create_empty_directory(name, overwrite=True):
850 """I create and attach an empty directory at the given name. The
851 child name must be a unicode string. I return a Deferred that fires
852 when the operation finishes."""
854 def move_child_to(current_child_name, new_parent, new_child_name=None,
856 """I take one of my children and move them to a new parent. The child
857 is referenced by name. On the new parent, the child will live under
858 'new_child_name', which defaults to 'current_child_name'. TODO: what
859 should we do about metadata? I return a Deferred that fires when the
860 operation finishes. The child name must be a unicode string. I raise
861 NoSuchChildError if I do not have a child by that name."""
863 def build_manifest():
864 """I generate a table of everything reachable from this directory.
865 I also compute deep-stats as described below.
867 I return a Monitor. The Monitor's results will be a dictionary with
870 res['manifest']: a list of (path, cap) tuples for all nodes
871 (directories and files) reachable from this one.
872 'path' will be a tuple of unicode strings. The
873 origin dirnode will be represented by an empty path
875 res['verifycaps']: a list of (printable) verifycap strings, one for
876 each reachable non-LIT node. This is a set:
877 it will contain no duplicates.
878 res['storage-index']: a list of (base32) storage index strings,
879 one for each reachable non-LIT node. This is
880 a set: it will contain no duplicates.
881 res['stats']: a dictionary, the same that is generated by
882 start_deep_stats() below.
884 The Monitor will also have an .origin_si attribute with the (binary)
885 storage index of the starting point.
888 def start_deep_stats():
889 """Return a Monitor, examining all nodes (directories and files)
890 reachable from this one. The Monitor's results will be a dictionary
891 with the following keys::
893 count-immutable-files: count of how many CHK files are in the set
894 count-mutable-files: same, for mutable files (does not include
896 count-literal-files: same, for LIT files
897 count-files: sum of the above three
899 count-directories: count of directories
901 size-immutable-files: total bytes for all CHK files in the set
902 size-mutable-files (TODO): same, for current version of all mutable
903 files, does not include directories
904 size-literal-files: same, for LIT files
905 size-directories: size of mutable files used by directories
907 largest-directory: number of bytes in the largest directory
908 largest-directory-children: number of children in the largest
910 largest-immutable-file: number of bytes in the largest CHK file
912 size-mutable-files is not yet implemented, because it would involve
913 even more queries than deep_stats does.
915 The Monitor will also have an .origin_si attribute with the (binary)
916 storage index of the starting point.
918 This operation will visit every directory node underneath this one,
919 and can take a long time to run. On a typical workstation with good
920 bandwidth, this can examine roughly 15 directories per second (and
921 takes several minutes of 100% CPU for ~1700 directories).
924 class ICodecEncoder(Interface):
925 def set_params(data_size, required_shares, max_shares):
926 """Set up the parameters of this encoder.
928 This prepares the encoder to perform an operation that converts a
929 single block of data into a number of shares, such that a future
930 ICodecDecoder can use a subset of these shares to recover the
931 original data. This operation is invoked by calling encode(). Once
932 the encoding parameters are set up, the encode operation can be
933 invoked multiple times.
935 set_params() prepares the encoder to accept blocks of input data that
936 are exactly 'data_size' bytes in length. The encoder will be prepared
937 to produce 'max_shares' shares for each encode() operation (although
938 see the 'desired_share_ids' to use less CPU). The encoding math will
939 be chosen such that the decoder can get by with as few as
940 'required_shares' of these shares and still reproduce the original
941 data. For example, set_params(1000, 5, 5) offers no redundancy at
942 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
944 Numerical Restrictions: 'data_size' is required to be an integral
945 multiple of 'required_shares'. In general, the caller should choose
946 required_shares and max_shares based upon their reliability
947 requirements and the number of peers available (the total storage
948 space used is roughly equal to max_shares*data_size/required_shares),
949 then choose data_size to achieve the memory footprint desired (larger
950 data_size means more efficient operation, smaller data_size means
951 smaller memory footprint).
953 In addition, 'max_shares' must be equal to or greater than
954 'required_shares'. Of course, setting them to be equal causes
955 encode() to degenerate into a particularly slow form of the 'split'
958 See encode() for more details about how these parameters are used.
960 set_params() must be called before any other ICodecEncoder methods
965 """Return the 3-tuple of data_size, required_shares, max_shares"""
967 def get_encoder_type():
968 """Return a short string that describes the type of this encoder.
970 There is required to be a global table of encoder classes. This method
971 returns an index into this table; the value at this index is an
972 encoder class, and this encoder is an instance of that class.
975 def get_block_size():
976 """Return the length of the shares that encode() will produce.
979 def encode_proposal(data, desired_share_ids=None):
982 'data' must be a string (or other buffer object), and len(data) must
983 be equal to the 'data_size' value passed earlier to set_params().
985 This will return a Deferred that will fire with two lists. The first
986 is a list of shares, each of which is a string (or other buffer
987 object) such that len(share) is the same as what get_share_size()
988 returned earlier. The second is a list of shareids, in which each is
989 an integer. The lengths of the two lists will always be equal to each
990 other. The user should take care to keep each share closely
991 associated with its shareid, as one is useless without the other.
993 The length of this output list will normally be the same as the value
994 provided to the 'max_shares' parameter of set_params(). This may be
995 different if 'desired_share_ids' is provided.
997 'desired_share_ids', if provided, is required to be a sequence of
998 ints, each of which is required to be >= 0 and < max_shares. If not
999 provided, encode() will produce 'max_shares' shares, as if
1000 'desired_share_ids' were set to range(max_shares). You might use this
1001 if you initially thought you were going to use 10 peers, started
1002 encoding, and then two of the peers dropped out: you could use
1003 desired_share_ids= to skip the work (both memory and CPU) of
1004 producing shares for the peers which are no longer available.
1008 def encode(inshares, desired_share_ids=None):
1009 """Encode some data. This may be called multiple times. Each call is
1012 inshares is a sequence of length required_shares, containing buffers
1013 (i.e. strings), where each buffer contains the next contiguous
1014 non-overlapping segment of the input data. Each buffer is required to
1015 be the same length, and the sum of the lengths of the buffers is
1016 required to be exactly the data_size promised by set_params(). (This
1017 implies that the data has to be padded before being passed to
1018 encode(), unless of course it already happens to be an even multiple
1019 of required_shares in length.)
1021 ALSO: the requirement to break up your data into 'required_shares'
1022 chunks before calling encode() feels a bit surprising, at least from
1023 the point of view of a user who doesn't know how FEC works. It feels
1024 like an implementation detail that has leaked outside the
1025 abstraction barrier. Can you imagine a use case in which the data to
1026 be encoded might already be available in pre-segmented chunks, such
1027 that it is faster or less work to make encode() take a list rather
1028 than splitting a single string?
1030 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1031 is supposed to *produce* shares, so what it *accepts* should be
1032 something other than shares. Other places in this interface use the
1033 word 'data' for that-which-is-not-shares.. maybe we should use that
1036 'desired_share_ids', if provided, is required to be a sequence of
1037 ints, each of which is required to be >= 0 and < max_shares. If not
1038 provided, encode() will produce 'max_shares' shares, as if
1039 'desired_share_ids' were set to range(max_shares). You might use this
1040 if you initially thought you were going to use 10 peers, started
1041 encoding, and then two of the peers dropped out: you could use
1042 desired_share_ids= to skip the work (both memory and CPU) of
1043 producing shares for the peers which are no longer available.
1045 For each call, encode() will return a Deferred that fires with two
1046 lists, one containing shares and the other containing the shareids.
1047 The get_share_size() method can be used to determine the length of
1048 the share strings returned by encode(). Each shareid is a small
1049 integer, exactly as passed into 'desired_share_ids' (or
1050 range(max_shares), if desired_share_ids was not provided).
1052 The shares and their corresponding shareids are required to be kept
1053 together during storage and retrieval. Specifically, the share data is
1054 useless by itself: the decoder needs to be told which share is which
1055 by providing it with both the shareid and the actual share data.
1057 This function will allocate an amount of memory roughly equal to::
1059 (max_shares - required_shares) * get_share_size()
1061 When combined with the memory that the caller must allocate to
1062 provide the input data, this leads to a memory footprint roughly
1063 equal to the size of the resulting encoded shares (i.e. the expansion
1064 factor times the size of the input segment).
1069 # returning a list of (shareidN,shareN) tuples instead of a pair of
1070 # lists (shareids..,shares..). Brian thought the tuples would
1071 # encourage users to keep the share and shareid together throughout
1072 # later processing, Zooko pointed out that the code to iterate
1073 # through two lists is not really more complicated than using a list
1074 # of tuples and there's also a performance improvement
1076 # having 'data_size' not required to be an integral multiple of
1077 # 'required_shares'. Doing this would require encode() to perform
1078 # padding internally, and we'd prefer to have any padding be done
1079 # explicitly by the caller. Yes, it is an abstraction leak, but
1080 # hopefully not an onerous one.
1083 class ICodecDecoder(Interface):
1084 def set_params(data_size, required_shares, max_shares):
1085 """Set the params. They have to be exactly the same ones that were used for encoding. """
1087 def get_needed_shares():
1088 """Return the number of shares needed to reconstruct the data.
1089 set_params() is required to be called before this."""
1091 def decode(some_shares, their_shareids):
1092 """Decode a partial list of shares into data.
1094 'some_shares' is required to be a sequence of buffers of sharedata, a
1095 subset of the shares returned by ICodecEncode.encode(). Each share is
1096 required to be of the same length. The i'th element of their_shareids
1097 is required to be the shareid of the i'th buffer in some_shares.
1099 This returns a Deferred which fires with a sequence of buffers. This
1100 sequence will contain all of the segments of the original data, in
1101 order. The sum of the lengths of all of the buffers will be the
1102 'data_size' value passed into the original ICodecEncode.set_params()
1103 call. To get back the single original input block of data, use
1104 ''.join(output_buffers), or you may wish to simply write them in
1105 order to an output file.
1107 Note that some of the elements in the result sequence may be
1108 references to the elements of the some_shares input sequence. In
1109 particular, this means that if those share objects are mutable (e.g.
1110 arrays) and if they are changed, then both the input (the
1111 'some_shares' parameter) and the output (the value given when the
1112 deferred is triggered) will change.
1114 The length of 'some_shares' is required to be exactly the value of
1115 'required_shares' passed into the original ICodecEncode.set_params()
1119 class IEncoder(Interface):
1120 """I take an object that provides IEncryptedUploadable, which provides
1121 encrypted data, and a list of shareholders. I then encode, hash, and
1122 deliver shares to those shareholders. I will compute all the necessary
1123 Merkle hash trees that are necessary to validate the crypttext that
1124 eventually comes back from the shareholders. I provide the URI Extension
1125 Block Hash, and the encoding parameters, both of which must be included
1128 I do not choose shareholders, that is left to the IUploader. I must be
1129 given a dict of RemoteReferences to storage buckets that are ready and
1130 willing to receive data.
1134 """Specify the number of bytes that will be encoded. This must be
1135 peformed before get_serialized_params() can be called.
1137 def set_params(params):
1138 """Override the default encoding parameters. 'params' is a tuple of
1139 (k,d,n), where 'k' is the number of required shares, 'd' is the
1140 shares_of_happiness, and 'n' is the total number of shares that will
1143 Encoding parameters can be set in three ways. 1: The Encoder class
1144 provides defaults (3/7/10). 2: the Encoder can be constructed with
1145 an 'options' dictionary, in which the
1146 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1147 set_params((k,d,n)) can be called.
1149 If you intend to use set_params(), you must call it before
1150 get_share_size or get_param are called.
1153 def set_encrypted_uploadable(u):
1154 """Provide a source of encrypted upload data. 'u' must implement
1155 IEncryptedUploadable.
1157 When this is called, the IEncryptedUploadable will be queried for its
1158 length and the storage_index that should be used.
1160 This returns a Deferred that fires with this Encoder instance.
1162 This must be performed before start() can be called.
1165 def get_param(name):
1166 """Return an encoding parameter, by name.
1168 'storage_index': return a string with the (16-byte truncated SHA-256
1169 hash) storage index to which these shares should be
1172 'share_counts': return a tuple describing how many shares are used:
1173 (needed_shares, shares_of_happiness, total_shares)
1175 'num_segments': return an int with the number of segments that
1178 'segment_size': return an int with the size of each segment.
1180 'block_size': return the size of the individual blocks that will
1181 be delivered to a shareholder's put_block() method. By
1182 knowing this, the shareholder will be able to keep all
1183 blocks in a single file and still provide random access
1184 when reading them. # TODO: can we avoid exposing this?
1186 'share_size': an int with the size of the data that will be stored
1187 on each shareholder. This is aggregate amount of data
1188 that will be sent to the shareholder, summed over all
1189 the put_block() calls I will ever make. It is useful to
1190 determine this size before asking potential
1191 shareholders whether they will grant a lease or not,
1192 since their answers will depend upon how much space we
1193 need. TODO: this might also include some amount of
1194 overhead, like the size of all the hashes. We need to
1195 decide whether this is useful or not.
1197 'serialized_params': a string with a concise description of the
1198 codec name and its parameters. This may be passed
1199 into the IUploadable to let it make sure that
1200 the same file encoded with different parameters
1201 will result in different storage indexes.
1203 Once this is called, set_size() and set_params() may not be called.
1206 def set_shareholders(shareholders):
1207 """Tell the encoder where to put the encoded shares. 'shareholders'
1208 must be a dictionary that maps share number (an integer ranging from
1209 0 to n-1) to an instance that provides IStorageBucketWriter. This
1210 must be performed before start() can be called."""
1213 """Begin the encode/upload process. This involves reading encrypted
1214 data from the IEncryptedUploadable, encoding it, uploading the shares
1215 to the shareholders, then sending the hash trees.
1217 set_encrypted_uploadable() and set_shareholders() must be called
1218 before this can be invoked.
1220 This returns a Deferred that fires with a tuple of
1221 (uri_extension_hash, needed_shares, total_shares, size) when the
1222 upload process is complete. This information, plus the encryption
1223 key, is sufficient to construct the URI.
1226 class IDecoder(Interface):
1227 """I take a list of shareholders and some setup information, then
1228 download, validate, decode, and decrypt data from them, writing the
1229 results to an output file.
1231 I do not locate the shareholders, that is left to the IDownloader. I must
1232 be given a dict of RemoteReferences to storage buckets that are ready to
1237 """I take a file-like object (providing write and close) to which all
1238 the plaintext data will be written.
1240 TODO: producer/consumer . Maybe write() should return a Deferred that
1241 indicates when it will accept more data? But probably having the
1242 IDecoder be a producer is easier to glue to IConsumer pieces.
1245 def set_shareholders(shareholders):
1246 """I take a dictionary that maps share identifiers (small integers)
1247 to RemoteReferences that provide RIBucketReader. This must be called
1251 """I start the download. This process involves retrieving data and
1252 hash chains from the shareholders, using the hashes to validate the
1253 data, decoding the shares into segments, decrypting the segments,
1254 then writing the resulting plaintext to the output file.
1256 I return a Deferred that will fire (with self) when the download is
1260 class IDownloadTarget(Interface):
1261 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1262 # will register itself as a producer. This allows the target to invoke
1263 # downloader.pauseProducing, resumeProducing, and stopProducing.
1265 """Called before any calls to write() or close(). If an error
1266 occurs before any data is available, fail() may be called without
1267 a previous call to open().
1269 'size' is the length of the file being downloaded, in bytes."""
1272 """Output some data to the target."""
1274 """Inform the target that there is no more data to be written."""
1276 """fail() is called to indicate that the download has failed. 'why'
1277 is a Failure object indicating what went wrong. No further methods
1278 will be invoked on the IDownloadTarget after fail()."""
1279 def register_canceller(cb):
1280 """The FileDownloader uses this to register a no-argument function
1281 that the target can call to cancel the download. Once this canceller
1282 is invoked, no further calls to write() or close() will be made."""
1284 """When the FileDownloader is done, this finish() function will be
1285 called. Whatever it returns will be returned to the invoker of
1286 Downloader.download.
1289 class IDownloader(Interface):
1290 def download(uri, target):
1291 """Perform a CHK download, sending the data to the given target.
1292 'target' must provide IDownloadTarget.
1294 Returns a Deferred that fires (with the results of target.finish)
1295 when the download is finished, or errbacks if something went wrong."""
1297 class IEncryptedUploadable(Interface):
1298 def set_upload_status(upload_status):
1299 """Provide an IUploadStatus object that should be filled with status
1300 information. The IEncryptedUploadable is responsible for setting
1301 key-determination progress ('chk'), size, storage_index, and
1302 ciphertext-fetch progress. It may delegate some of this
1303 responsibility to others, in particular to the IUploadable."""
1306 """This behaves just like IUploadable.get_size()."""
1308 def get_all_encoding_parameters():
1309 """Return a Deferred that fires with a tuple of
1310 (k,happy,n,segment_size). The segment_size will be used as-is, and
1311 must match the following constraints: it must be a multiple of k, and
1312 it shouldn't be unreasonably larger than the file size (if
1313 segment_size is larger than filesize, the difference must be stored
1316 This usually passes through to the IUploadable method of the same
1319 The encoder strictly obeys the values returned by this method. To
1320 make an upload use non-default encoding parameters, you must arrange
1321 to control the values that this method returns.
1324 def get_storage_index():
1325 """Return a Deferred that fires with a 16-byte storage index.
1328 def read_encrypted(length, hash_only):
1329 """This behaves just like IUploadable.read(), but returns crypttext
1330 instead of plaintext. If hash_only is True, then this discards the
1331 data (and returns an empty list); this improves efficiency when
1332 resuming an interrupted upload (where we need to compute the
1333 plaintext hashes, but don't need the redundant encrypted data)."""
1335 def get_plaintext_hashtree_leaves(first, last, num_segments):
1336 """Get the leaf nodes of a merkle hash tree over the plaintext
1337 segments, i.e. get the tagged hashes of the given segments. The
1338 segment size is expected to be generated by the IEncryptedUploadable
1339 before any plaintext is read or ciphertext produced, so that the
1340 segment hashes can be generated with only a single pass.
1342 This returns a Deferred which fires with a sequence of hashes, using:
1344 tuple(segment_hashes[first:last])
1346 'num_segments' is used to assert that the number of segments that the
1347 IEncryptedUploadable handled matches the number of segments that the
1348 encoder was expecting.
1350 This method must not be called until the final byte has been read
1351 from read_encrypted(). Once this method is called, read_encrypted()
1352 can never be called again.
1355 def get_plaintext_hash():
1356 """Get the hash of the whole plaintext.
1358 This returns a Deferred which fires with a tagged SHA-256 hash of the
1359 whole plaintext, obtained from hashutil.plaintext_hash(data).
1363 """Just like IUploadable.close()."""
1365 class IUploadable(Interface):
1366 def set_upload_status(upload_status):
1367 """Provide an IUploadStatus object that should be filled with status
1368 information. The IUploadable is responsible for setting
1369 key-determination progress ('chk')."""
1371 def set_default_encoding_parameters(params):
1372 """Set the default encoding parameters, which must be a dict mapping
1373 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1374 'max_segment_size'. These might have an influence on the final
1375 encoding parameters returned by get_all_encoding_parameters(), if the
1376 Uploadable doesn't have more specific preferences.
1378 This call is optional: if it is not used, the Uploadable will use
1379 some built-in defaults. If used, this method must be called before
1380 any other IUploadable methods to have any effect.
1384 """Return a Deferred that will fire with the length of the data to be
1385 uploaded, in bytes. This will be called before the data is actually
1386 used, to compute encoding parameters.
1389 def get_all_encoding_parameters():
1390 """Return a Deferred that fires with a tuple of
1391 (k,happy,n,segment_size). The segment_size will be used as-is, and
1392 must match the following constraints: it must be a multiple of k, and
1393 it shouldn't be unreasonably larger than the file size (if
1394 segment_size is larger than filesize, the difference must be stored
1397 The relative values of k and n allow some IUploadables to request
1398 better redundancy than others (in exchange for consuming more space
1401 Larger values of segment_size reduce hash overhead, while smaller
1402 values reduce memory footprint and cause data to be delivered in
1403 smaller pieces (which may provide a smoother and more predictable
1404 download experience).
1406 The encoder strictly obeys the values returned by this method. To
1407 make an upload use non-default encoding parameters, you must arrange
1408 to control the values that this method returns. One way to influence
1409 them may be to call set_encoding_parameters() before calling
1410 get_all_encoding_parameters().
1413 def get_encryption_key():
1414 """Return a Deferred that fires with a 16-byte AES key. This key will
1415 be used to encrypt the data. The key will also be hashed to derive
1418 Uploadables which want to achieve convergence should hash their file
1419 contents and the serialized_encoding_parameters to form the key
1420 (which of course requires a full pass over the data). Uploadables can
1421 use the upload.ConvergentUploadMixin class to achieve this
1424 Uploadables which do not care about convergence (or do not wish to
1425 make multiple passes over the data) can simply return a
1426 strongly-random 16 byte string.
1428 get_encryption_key() may be called multiple times: the IUploadable is
1429 required to return the same value each time.
1433 """Return a Deferred that fires with a list of strings (perhaps with
1434 only a single element) which, when concatenated together, contain the
1435 next 'length' bytes of data. If EOF is near, this may provide fewer
1436 than 'length' bytes. The total number of bytes provided by read()
1437 before it signals EOF must equal the size provided by get_size().
1439 If the data must be acquired through multiple internal read
1440 operations, returning a list instead of a single string may help to
1441 reduce string copies.
1443 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1444 so a 10kB file means length=3kB, 100kB file means length=30kB,
1445 and >=1MB file means length=300kB.
1447 This method provides for a single full pass through the data. Later
1448 use cases may desire multiple passes or access to only parts of the
1449 data (such as a mutable file making small edits-in-place). This API
1450 will be expanded once those use cases are better understood.
1454 """The upload is finished, and whatever filehandle was in use may be
1457 class IUploadResults(Interface):
1458 """I am returned by upload() methods. I contain a number of public
1459 attributes which can be read to determine the results of the upload. Some
1460 of these are functional, some are timing information. All of these may be
1463 .file_size : the size of the file, in bytes
1464 .uri : the CHK read-cap for the file
1465 .ciphertext_fetched : how many bytes were fetched by the helper
1466 .sharemap : dict mapping share number to placement string
1467 .servermap : dict mapping server peerid to a set of share numbers
1468 .timings : dict of timing information, mapping name to seconds (float)
1469 total : total upload time, start to finish
1470 storage_index : time to compute the storage index
1471 peer_selection : time to decide which peers will be used
1472 contacting_helper : initial helper query to upload/no-upload decision
1473 existence_check : helper pre-upload existence check
1474 helper_total : initial helper query to helper finished pushing
1475 cumulative_fetch : helper waiting for ciphertext requests
1476 total_fetch : helper start to last ciphertext response
1477 cumulative_encoding : just time spent in zfec
1478 cumulative_sending : just time spent waiting for storage servers
1479 hashes_and_close : last segment push to shareholder close
1480 total_encode_and_push : first encode to shareholder close
1484 class IDownloadResults(Interface):
1485 """I am created internally by download() methods. I contain a number of
1486 public attributes which contain details about the download process.::
1488 .file_size : the size of the file, in bytes
1489 .servers_used : set of server peerids that were used during download
1490 .server_problems : dict mapping server peerid to a problem string. Only
1491 servers that had problems (bad hashes, disconnects) are
1493 .servermap : dict mapping server peerid to a set of share numbers. Only
1494 servers that had any shares are listed here.
1495 .timings : dict of timing information, mapping name to seconds (float)
1496 peer_selection : time to ask servers about shares
1497 servers_peer_selection : dict of peerid to DYHB-query time
1498 uri_extension : time to fetch a copy of the URI extension block
1499 hashtrees : time to fetch the hash trees
1500 segments : time to fetch, decode, and deliver segments
1501 cumulative_fetch : time spent waiting for storage servers
1502 cumulative_decode : just time spent in zfec
1503 cumulative_decrypt : just time spent in decryption
1504 total : total download time, start to finish
1505 fetch_per_server : dict of peerid to list of per-segment fetch times
1509 class IUploader(Interface):
1510 def upload(uploadable):
1511 """Upload the file. 'uploadable' must impement IUploadable. This
1512 returns a Deferred which fires with an UploadResults instance, from
1513 which the URI of the file can be obtained as results.uri ."""
1515 def upload_ssk(write_capability, new_version, uploadable):
1516 """TODO: how should this work?"""
1518 class ICheckable(Interface):
1519 def check(monitor, verify=False):
1520 """Check upon my health, optionally repairing any problems.
1522 This returns a Deferred that fires with an instance that provides
1523 ICheckerResults, or None if the object is non-distributed (i.e. LIT
1526 The monitor will be checked periodically to see if the operation has
1527 been cancelled. If so, no new queries will be sent, and the Deferred
1528 will fire (with a OperationCancelledError) immediately.
1530 Filenodes and dirnodes (which provide IFilesystemNode) are also
1531 checkable. Instances that represent verifier-caps will be checkable
1532 but not downloadable. Some objects (like LIT files) do not actually
1533 live in the grid, and their checkers return None (non-distributed
1534 files are always healthy).
1536 If verify=False, a relatively lightweight check will be performed: I
1537 will ask all servers if they have a share for me, and I will believe
1538 whatever they say. If there are at least N distinct shares on the
1539 grid, my results will indicate r.is_healthy()==True. This requires a
1540 roundtrip to each server, but does not transfer very much data, so
1541 the network bandwidth is fairly low.
1543 If verify=True, a more resource-intensive check will be performed:
1544 every share will be downloaded, and the hashes will be validated on
1545 every bit. I will ignore any shares that failed their hash checks. If
1546 there are at least N distinct valid shares on the grid, my results
1547 will indicate r.is_healthy()==True. This requires N/k times as much
1548 download bandwidth (and server disk IO) as a regular download. If a
1549 storage server is holding a corrupt share, or is experiencing memory
1550 failures during retrieval, or is malicious or buggy, then
1551 verification will detect the problem, but checking will not.
1553 TODO: any problems seen during checking will be reported to the
1554 health-manager.furl, a centralized object which is responsible for
1555 figuring out why files are unhealthy so corrective action can be
1559 def check_and_repair(monitor, verify=False):
1560 """Like check(), but if the file/directory is not healthy, attempt to
1563 Any non-healthy result will cause an immediate repair operation, to
1564 generate and upload new shares. After repair, the file will be as
1565 healthy as we can make it. Details about what sort of repair is done
1566 will be put in the check-and-repair results. The Deferred will not
1567 fire until the repair is complete.
1569 This returns a Deferred which fires with an instance of
1570 ICheckAndRepairResults."""
1572 class IDeepCheckable(Interface):
1573 def start_deep_check(verify=False):
1574 """Check upon the health of me and everything I can reach.
1576 This is a recursive form of check(), useable only on dirnodes.
1578 I return a Monitor, with results that are an IDeepCheckResults
1582 def start_deep_check_and_repair(verify=False):
1583 """Check upon the health of me and everything I can reach. Repair
1584 anything that isn't healthy.
1586 This is a recursive form of check_and_repair(), useable only on
1589 I return a Monitor, with results that are an
1590 IDeepCheckAndRepairResults object.
1593 class ICheckerResults(Interface):
1594 """I contain the detailed results of a check/verify operation.
1597 def get_storage_index():
1598 """Return a string with the (binary) storage index."""
1599 def get_storage_index_string():
1600 """Return a string with the (printable) abbreviated storage index."""
1602 """Return the (string) URI of the object that was checked."""
1605 """Return a boolean, True if the file/dir is fully healthy, False if
1606 it is damaged in any way. Non-distributed LIT files always return
1609 def is_recoverable():
1610 """Return a boolean, True if the file/dir can be recovered, False if
1611 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1612 files always return True."""
1614 def needs_rebalancing():
1615 """Return a boolean, True if the file/dir's reliability could be
1616 improved by moving shares to new servers. Non-distributed LIT files
1617 always returne False."""
1621 """Return a dictionary that describes the state of the file/dir.
1622 Non-distributed LIT files always return an empty dictionary. Normal
1623 files and directories return a dictionary with the following keys
1624 (note that these use base32-encoded strings rather than binary ones)
1625 (also note that for mutable files, these counts are for the 'best'
1628 count-shares-good: the number of distinct good shares that were found
1629 count-shares-needed: 'k', the number of shares required for recovery
1630 count-shares-expected: 'N', the number of total shares generated
1631 count-good-share-hosts: the number of distinct storage servers with
1632 good shares. If this number is less than
1633 count-shares-good, then some shares are
1634 doubled up, increasing the correlation of
1635 failures. This indicates that one or more
1636 shares should be moved to an otherwise unused
1637 server, if one is available.
1638 count-corrupt-shares: the number of shares with integrity failures
1639 list-corrupt-shares: a list of 'share locators', one for each share
1640 that was found to be corrupt. Each share
1641 locator is a list of (serverid, storage_index,
1643 servers-responding: list of (binary) storage server identifiers,
1644 one for each server which responded to the share
1646 sharemap: dict mapping share identifier to list of serverids
1647 (binary strings). This indicates which servers are holding
1648 which shares. For immutable files, the shareid is an
1649 integer (the share number, from 0 to N-1). For mutable
1650 files, it is a string of the form 'seq%d-%s-sh%d',
1651 containing the sequence number, the roothash, and the
1654 The following keys are most relevant for mutable files, but immutable
1655 files will provide sensible values too::
1657 count-wrong-shares: the number of shares for versions other than the
1658 'best' one (which is defined as being the
1659 recoverable version with the highest sequence
1660 number, then the highest roothash). These are
1661 either leftover shares from an older version
1662 (perhaps on a server that was offline when an
1663 update occurred), shares from an unrecoverable
1664 newer version, or shares from an alternate
1665 current version that results from an
1666 uncoordinated write collision. For a healthy
1667 file, this will equal 0.
1669 count-recoverable-versions: the number of recoverable versions of
1670 the file. For a healthy file, this will
1673 count-unrecoverable-versions: the number of unrecoverable versions
1674 of the file. For a healthy file, this
1680 """Return a string with a brief (one-line) summary of the results."""
1683 """Return a list of strings with more detailed results."""
1685 class ICheckAndRepairResults(Interface):
1686 """I contain the detailed results of a check/verify/repair operation.
1688 The IFilesystemNode.check()/verify()/repair() methods all return
1689 instances that provide ICheckAndRepairResults.
1692 def get_storage_index():
1693 """Return a string with the (binary) storage index."""
1694 def get_storage_index_string():
1695 """Return a string with the (printable) abbreviated storage index."""
1696 def get_repair_attempted():
1697 """Return a boolean, True if a repair was attempted."""
1698 def get_repair_successful():
1699 """Return a boolean, True if repair was attempted and the file/dir
1700 was fully healthy afterwards. False if no repair was attempted or if
1701 a repair attempt failed."""
1702 def get_pre_repair_results():
1703 """Return an ICheckerResults instance that describes the state of the
1704 file/dir before any repair was attempted."""
1705 def get_post_repair_results():
1706 """Return an ICheckerResults instance that describes the state of the
1707 file/dir after any repair was attempted. If no repair was attempted,
1708 the pre-repair and post-repair results will be identical."""
1711 class IDeepCheckResults(Interface):
1712 """I contain the results of a deep-check operation.
1714 This is returned by a call to ICheckable.deep_check().
1717 def get_root_storage_index_string():
1718 """Return the storage index (abbreviated human-readable string) of
1719 the first object checked."""
1721 """Return a dictionary with the following keys::
1723 count-objects-checked: count of how many objects were checked
1724 count-objects-healthy: how many of those objects were completely
1726 count-objects-unhealthy: how many were damaged in some way
1727 count-objects-unrecoverable: how many were unrecoverable
1728 count-corrupt-shares: how many shares were found to have
1729 corruption, summed over all objects
1733 def get_corrupt_shares():
1734 """Return a set of (serverid, storage_index, sharenum) for all shares
1735 that were found to be corrupt. Both serverid and storage_index are
1738 def get_all_results():
1739 """Return a dictionary mapping pathname (a tuple of strings, ready to
1740 be slash-joined) to an ICheckerResults instance, one for each object
1741 that was checked."""
1743 def get_results_for_storage_index(storage_index):
1744 """Retrive the ICheckerResults instance for the given (binary)
1745 storage index. Raises KeyError if there are no results for that
1749 """Return a dictionary with the same keys as
1750 IDirectoryNode.deep_stats()."""
1752 class IDeepCheckAndRepairResults(Interface):
1753 """I contain the results of a deep-check-and-repair operation.
1755 This is returned by a call to ICheckable.deep_check_and_repair().
1758 def get_root_storage_index_string():
1759 """Return the storage index (abbreviated human-readable string) of
1760 the first object checked."""
1762 """Return a dictionary with the following keys::
1764 count-objects-checked: count of how many objects were checked
1765 count-objects-healthy-pre-repair: how many of those objects were
1766 completely healthy (before any
1768 count-objects-unhealthy-pre-repair: how many were damaged in
1770 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1771 count-objects-healthy-post-repair: how many of those objects were
1772 completely healthy (after any
1774 count-objects-unhealthy-post-repair: how many were damaged in
1776 count-objects-unrecoverable-post-repair: how many were
1778 count-repairs-attempted: repairs were attempted on this many
1779 objects. The count-repairs- keys will
1780 always be provided, however unless
1781 repair=true is present, they will all
1783 count-repairs-successful: how many repairs resulted in healthy
1785 count-repairs-unsuccessful: how many repairs resulted did not
1786 results in completely healthy objects
1787 count-corrupt-shares-pre-repair: how many shares were found to
1788 have corruption, summed over all
1789 objects examined (before any
1791 count-corrupt-shares-post-repair: how many shares were found to
1792 have corruption, summed over all
1793 objects examined (after any
1798 """Return a dictionary with the same keys as
1799 IDirectoryNode.deep_stats()."""
1801 def get_corrupt_shares():
1802 """Return a set of (serverid, storage_index, sharenum) for all shares
1803 that were found to be corrupt before any repair was attempted. Both
1804 serverid and storage_index are binary.
1806 def get_remaining_corrupt_shares():
1807 """Return a set of (serverid, storage_index, sharenum) for all shares
1808 that were found to be corrupt after any repair was completed. Both
1809 serverid and storage_index are binary. These are shares that need
1810 manual inspection and probably deletion.
1812 def get_all_results():
1813 """Return a dictionary mapping pathname (a tuple of strings, ready to
1814 be slash-joined) to an ICheckAndRepairResults instance, one for each
1815 object that was checked."""
1818 class IRepairable(Interface):
1819 def repair(checker_results):
1820 """Attempt to repair the given object. Returns a Deferred that fires
1821 with a IRepairResults object.
1823 I must be called with an object that implements ICheckerResults, as
1824 proof that you have actually discovered a problem with this file. I
1825 will use the data in the checker results to guide the repair process,
1826 such as which servers provided bad data and should therefore be
1827 avoided. The ICheckerResults object is inside the
1828 ICheckAndRepairResults object, which is returned by the
1829 ICheckable.check() method::
1831 d = filenode.check(repair=False)
1832 def _got_results(check_and_repair_results):
1833 check_results = check_and_repair_results.get_pre_repair_results()
1834 return filenode.repair(check_results)
1835 d.addCallback(_got_results)
1839 class IRepairResults(Interface):
1840 """I contain the results of a repair operation."""
1843 class IClient(Interface):
1844 def upload(uploadable):
1845 """Upload some data into a CHK, get back the UploadResults for it.
1846 @param uploadable: something that implements IUploadable
1847 @return: a Deferred that fires with the UploadResults instance.
1848 To get the URI for this file, use results.uri .
1851 def create_mutable_file(contents=""):
1852 """Create a new mutable file with contents, get back the URI string.
1853 @param contents: the initial contents to place in the file.
1854 @return: a Deferred that fires with tne (string) SSK URI for the new
1858 def create_empty_dirnode():
1859 """Create a new dirnode, empty and unattached.
1860 @return: a Deferred that fires with the new IDirectoryNode instance.
1863 def create_node_from_uri(uri):
1864 """Create a new IFilesystemNode instance from the uri, synchronously.
1865 @param uri: a string or IURI-providing instance. This could be for a
1866 LiteralFileNode, a CHK file node, a mutable file node, or
1868 @return: an instance that provides IFilesystemNode (or more usefully one
1869 of its subclasses). File-specifying URIs will result in
1870 IFileNode or IMutableFileNode -providing instances, like
1871 FileNode, LiteralFileNode, or MutableFileNode.
1872 Directory-specifying URIs will result in
1873 IDirectoryNode-providing instances, like NewDirectoryNode.
1876 class IClientStatus(Interface):
1877 def list_all_uploads():
1878 """Return a list of uploader objects, one for each upload which
1879 currently has an object available (tracked with weakrefs). This is
1880 intended for debugging purposes."""
1881 def list_active_uploads():
1882 """Return a list of active IUploadStatus objects."""
1883 def list_recent_uploads():
1884 """Return a list of IUploadStatus objects for the most recently
1887 def list_all_downloads():
1888 """Return a list of downloader objects, one for each download which
1889 currently has an object available (tracked with weakrefs). This is
1890 intended for debugging purposes."""
1891 def list_active_downloads():
1892 """Return a list of active IDownloadStatus objects."""
1893 def list_recent_downloads():
1894 """Return a list of IDownloadStatus objects for the most recently
1895 started downloads."""
1897 class IUploadStatus(Interface):
1899 """Return a timestamp (float with seconds since epoch) indicating
1900 when the operation was started."""
1901 def get_storage_index():
1902 """Return a string with the (binary) storage index in use on this
1903 upload. Returns None if the storage index has not yet been
1906 """Return an integer with the number of bytes that will eventually
1907 be uploaded for this file. Returns None if the size is not yet known.
1910 """Return True if this upload is using a Helper, False if not."""
1912 """Return a string describing the current state of the upload
1915 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
1916 each from 0.0 to 1.0 . 'chk' describes how much progress has been
1917 made towards hashing the file to determine a CHK encryption key: if
1918 non-convergent encryption is in use, this will be trivial, otherwise
1919 the whole file must be hashed. 'ciphertext' describes how much of the
1920 ciphertext has been pushed to the helper, and is '1.0' for non-helper
1921 uploads. 'encode_and_push' describes how much of the encode-and-push
1922 process has finished: for helper uploads this is dependent upon the
1923 helper providing progress reports. It might be reasonable to add all
1924 three numbers and report the sum to the user."""
1926 """Return True if the upload is currently active, False if not."""
1928 """Return an instance of UploadResults (which contains timing and
1929 sharemap information). Might return None if the upload is not yet
1932 """Each upload status gets a unique number: this method returns that
1933 number. This provides a handle to this particular upload, so a web
1934 page can generate a suitable hyperlink."""
1936 class IDownloadStatus(Interface):
1938 """Return a timestamp (float with seconds since epoch) indicating
1939 when the operation was started."""
1940 def get_storage_index():
1941 """Return a string with the (binary) storage index in use on this
1942 download. This may be None if there is no storage index (i.e. LIT
1945 """Return an integer with the number of bytes that will eventually be
1946 retrieved for this file. Returns None if the size is not yet known.
1949 """Return True if this download is using a Helper, False if not."""
1951 """Return a string describing the current state of the download
1954 """Returns a float (from 0.0 to 1.0) describing the amount of the
1955 download that has completed. This value will remain at 0.0 until the
1956 first byte of plaintext is pushed to the download target."""
1958 """Return True if the download is currently active, False if not."""
1960 """Each download status gets a unique number: this method returns
1961 that number. This provides a handle to this particular download, so a
1962 web page can generate a suitable hyperlink."""
1964 class IServermapUpdaterStatus(Interface):
1966 class IPublishStatus(Interface):
1968 class IRetrieveStatus(Interface):
1971 class NotCapableError(Exception):
1972 """You have tried to write to a read-only node."""
1974 class BadWriteEnablerError(Exception):
1977 class RIControlClient(RemoteInterface):
1979 def wait_for_client_connections(num_clients=int):
1980 """Do not return until we have connections to at least NUM_CLIENTS
1984 def upload_from_file_to_uri(filename=str, convergence=ChoiceOf(None, StringConstraint(2**20))):
1985 """Upload a file to the grid. This accepts a filename (which must be
1986 absolute) that points to a file on the node's local disk. The node will
1987 read the contents of this file, upload it to the grid, then return the
1988 URI at which it was uploaded. If convergence is None then a random
1989 encryption key will be used, else the plaintext will be hashed, then
1990 that hash will be mixed together with the "convergence" string to form
1995 def download_from_uri_to_file(uri=URI, filename=str):
1996 """Download a file from the grid, placing it on the node's local disk
1997 at the given filename (which must be absolute[?]). Returns the
1998 absolute filename where the file was written."""
2003 def get_memory_usage():
2004 """Return a dict describes the amount of memory currently in use. The
2005 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2006 measuring memory consupmtion in bytes."""
2007 return DictOf(str, int)
2009 def speed_test(count=int, size=int, mutable=Any()):
2010 """Write 'count' tempfiles to disk, all of the given size. Measure
2011 how long (in seconds) it takes to upload them all to the servers.
2012 Then measure how long it takes to download all of them. If 'mutable'
2013 is 'create', time creation of mutable files. If 'mutable' is
2014 'upload', then time access to the same mutable file instead of
2017 Returns a tuple of (upload_time, download_time).
2019 return (float, float)
2021 def measure_peer_response_time():
2022 """Send a short message to each connected peer, and measure the time
2023 it takes for them to respond to it. This is a rough measure of the
2024 application-level round trip time.
2026 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2029 return DictOf(Nodeid, float)
2031 UploadResults = Any() #DictOf(str, str)
2033 class RIEncryptedUploadable(RemoteInterface):
2034 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2039 def get_all_encoding_parameters():
2040 return (int, int, int, long)
2042 def read_encrypted(offset=Offset, length=ReadSize):
2045 def get_plaintext_hashtree_leaves(first=int, last=int, num_segments=int):
2048 def get_plaintext_hash():
2055 class RICHKUploadHelper(RemoteInterface):
2056 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2060 Return a dictionary of version information.
2062 return DictOf(str, Any())
2064 def upload(reader=RIEncryptedUploadable):
2065 return UploadResults
2068 class RIHelper(RemoteInterface):
2069 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2073 Return a dictionary of version information.
2075 return DictOf(str, Any())
2077 def upload_chk(si=StorageIndex):
2078 """See if a file with a given storage index needs uploading. The
2079 helper will ask the appropriate storage servers to see if the file
2080 has already been uploaded. If so, the helper will return a set of
2081 'upload results' that includes whatever hashes are needed to build
2082 the read-cap, and perhaps a truncated sharemap.
2084 If the file has not yet been uploaded (or if it was only partially
2085 uploaded), the helper will return an empty upload-results dictionary
2086 and also an RICHKUploadHelper object that will take care of the
2087 upload process. The client should call upload() on this object and
2088 pass it a reference to an RIEncryptedUploadable object that will
2089 provide ciphertext. When the upload is finished, the upload() method
2090 will finish and return the upload results.
2092 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2095 class RIStatsProvider(RemoteInterface):
2096 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2098 Provides access to statistics and monitoring information.
2103 returns a dictionary containing 'counters' and 'stats', each a dictionary
2104 with string counter/stat name keys, and numeric values. counters are
2105 monotonically increasing measures of work done, and stats are instantaneous
2106 measures (potentially time averaged internally)
2108 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2110 class RIStatsGatherer(RemoteInterface):
2111 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2113 Provides a monitoring service for centralised collection of stats
2116 def provide(provider=RIStatsProvider, nickname=str):
2118 @param provider: a stats collector instance which should be polled
2119 periodically by the gatherer to collect stats.
2120 @param nickname: a name useful to identify the provided client
2125 class IStatsProducer(Interface):
2128 returns a dictionary, with str keys representing the names of stats
2129 to be monitored, and numeric values.
2132 class RIKeyGenerator(RemoteInterface):
2133 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2135 Provides a service offering to make RSA key pairs.
2138 def get_rsa_key_pair(key_size=int):
2140 @param key_size: the size of the signature key.
2141 @return: tuple(verifying_key, signing_key)
2143 return TupleOf(str, str)
2146 class FileTooLargeError(Exception):
2149 class IValidatedThingProxy(Interface):
2151 """ Acquire a thing and validate it. Return a deferred which is eventually fired with
2152 self if the thing is valid or errbacked if it can't be acquired or validated. """
2154 class InsufficientVersionError(Exception):
2155 def __init__(self, needed, got):
2156 self.needed = needed
2159 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,