2 from zope.interface import Interface
3 from foolscap.schema import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint
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 tuple of (my_version, oldest_supported) strings. Each string can be parsed by
92 a pyutil.version_class.Version instance or a distutils.version.LooseVersion instance,
93 and then compared. The first goal is to make sure that nodes are not confused by
94 speaking to an incompatible peer. The second goal is to enable the development of
95 backwards-compatibility code.
97 The meaning of the oldest_supported element is that if you treat this storage server as
98 though it were of that version, then you will not be disappointed.
100 The precise meaning of this method might change in incompatible ways until we get the
101 whole compatibility scheme nailed down.
103 return TupleOf(str, str)
105 def allocate_buckets(storage_index=StorageIndex,
106 renew_secret=LeaseRenewSecret,
107 cancel_secret=LeaseCancelSecret,
108 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
109 allocated_size=Offset, canary=Referenceable):
111 @param storage_index: the index of the bucket to be created or
113 @param sharenums: these are the share numbers (probably between 0 and
114 99) that the sender is proposing to store on this
116 @param renew_secret: This is the secret used to protect bucket refresh
117 This secret is generated by the client and
118 stored for later comparison by the server. Each
119 server is given a different secret.
120 @param cancel_secret: Like renew_secret, but protects bucket decref.
121 @param canary: If the canary is lost before close(), the bucket is
123 @return: tuple of (alreadygot, allocated), where alreadygot is what we
124 already have and is what we hereby agree to accept. New
125 leases are added for shares in both lists.
127 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
128 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
130 def add_lease(storage_index=StorageIndex,
131 renew_secret=LeaseRenewSecret,
132 cancel_secret=LeaseCancelSecret):
134 Add a new lease on the given bucket. If the renew_secret matches an
135 existing lease, that lease will be renewed instead.
139 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
141 Renew the lease on a given bucket. Some networks will use this, some
146 def cancel_lease(storage_index=StorageIndex,
147 cancel_secret=LeaseCancelSecret):
149 Cancel the lease on a given bucket. If this was the last lease on the
150 bucket, the bucket will be deleted.
154 def get_buckets(storage_index=StorageIndex):
155 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
159 def slot_readv(storage_index=StorageIndex,
160 shares=ListOf(int), readv=ReadVector):
161 """Read a vector from the numbered shares associated with the given
162 storage index. An empty shares list means to return data from all
163 known shares. Returns a dictionary with one key per share."""
164 return DictOf(int, ReadData) # shnum -> results
166 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
167 secrets=TupleOf(WriteEnablerSecret,
170 tw_vectors=TestAndWriteVectorsForShares,
173 """General-purpose test-and-set operation for mutable slots. Perform
174 a bunch of comparisons against the existing shares. If they all pass,
175 then apply a bunch of write vectors to those shares. Then use the
176 read vectors to extract data from all the shares and return the data.
178 This method is, um, large. The goal is to allow clients to update all
179 the shares associated with a mutable file in a single round trip.
181 @param storage_index: the index of the bucket to be created or
183 @param write_enabler: a secret that is stored along with the slot.
184 Writes are accepted from any caller who can
185 present the matching secret. A different secret
186 should be used for each slot*server pair.
187 @param renew_secret: This is the secret used to protect bucket refresh
188 This secret is generated by the client and
189 stored for later comparison by the server. Each
190 server is given a different secret.
191 @param cancel_secret: Like renew_secret, but protects bucket decref.
193 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
194 cancel_secret). The first is required to perform any write. The
195 latter two are used when allocating new shares. To simply acquire a
196 new lease on existing shares, use an empty testv and an empty writev.
198 Each share can have a separate test vector (i.e. a list of
199 comparisons to perform). If all vectors for all shares pass, then all
200 writes for all shares are recorded. Each comparison is a 4-tuple of
201 (offset, length, operator, specimen), which effectively does a bool(
202 (read(offset, length)) OPERATOR specimen ) and only performs the
203 write if all these evaluate to True. Basic test-and-set uses 'eq'.
204 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
205 Write-if-same-or-newer uses 'le'.
207 Reads from the end of the container are truncated, and missing shares
208 behave like empty ones, so to assert that a share doesn't exist (for
209 use when creating a new share), use (0, 1, 'eq', '').
211 The write vector will be applied to the given share, expanding it if
212 necessary. A write vector applied to a share number that did not
213 exist previously will cause that share to be created.
215 Each write vector is accompanied by a 'new_length' argument. If
216 new_length is not None, use it to set the size of the container. This
217 can be used to pre-allocate space for a series of upcoming writes, or
218 truncate existing data. If the container is growing, new_length will
219 be applied before datav. If the container is shrinking, it will be
222 The read vector is used to extract data from all known shares,
223 *before* any writes have been applied. The same vector is used for
224 all shares. This captures the state that was tested by the test
227 This method returns two values: a boolean and a dict. The boolean is
228 True if the write vectors were applied, False if not. The dict is
229 keyed by share number, and each value contains a list of strings, one
230 for each element of the read vector.
232 If the write_enabler is wrong, this will raise BadWriteEnablerError.
233 To enable share migration (using update_write_enabler), the exception
234 will have the nodeid used for the old write enabler embedded in it,
235 in the following string::
237 The write enabler was recorded by nodeid '%s'.
239 Note that the nodeid here is encoded using the same base32 encoding
240 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
243 return TupleOf(bool, DictOf(int, ReadData))
245 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
246 shnum=int, reason=str):
247 """Clients who discover hash failures in shares that they have
248 downloaded from me will use this method to inform me about the
249 failures. I will record their concern so that my operator can
250 manually inspect the shares in question. I return None.
252 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
253 (binary) storage index string, and 'shnum' is the integer share
254 number. 'reason' is a human-readable explanation of the problem,
255 probably including some expected hash values and the computed ones
256 which did not match. Corruption advisories for mutable shares should
257 include a hash of the public key (the same value that appears in the
258 mutable-file verify-cap), since the current share format does not
262 class IStorageBucketWriter(Interface):
264 Objects of this kind live on the client side.
266 def put_block(segmentnum=int, data=ShareData):
267 """@param data: For most segments, this data will be 'blocksize'
268 bytes in length. The last segment might be shorter.
269 @return: a Deferred that fires (with None) when the operation completes
272 def put_plaintext_hashes(hashes=ListOf(Hash)):
274 @return: a Deferred that fires (with None) when the operation completes
277 def put_crypttext_hashes(hashes=ListOf(Hash)):
279 @return: a Deferred that fires (with None) when the operation completes
282 def put_block_hashes(blockhashes=ListOf(Hash)):
284 @return: a Deferred that fires (with None) when the operation completes
287 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
289 @return: a Deferred that fires (with None) when the operation completes
292 def put_uri_extension(data=URIExtensionData):
293 """This block of data contains integrity-checking information (hashes
294 of plaintext, crypttext, and shares), as well as encoding parameters
295 that are necessary to recover the data. This is a serialized dict
296 mapping strings to other strings. The hash of this data is kept in
297 the URI and verified before any of the data is used. All buckets for
298 a given file contain identical copies of this data.
300 The serialization format is specified with the following pseudocode:
301 for k in sorted(dict.keys()):
302 assert re.match(r'^[a-zA-Z_\-]+$', k)
303 write(k + ':' + netstring(dict[k]))
305 @return: a Deferred that fires (with None) when the operation completes
309 """Finish writing and close the bucket. The share is not finalized
310 until this method is called: if the uploading client disconnects
311 before calling close(), the partially-written share will be
314 @return: a Deferred that fires (with None) when the operation completes
317 class IStorageBucketReader(Interface):
319 def get_block(blocknum=int):
320 """Most blocks will be the same size. The last block might be shorter
326 def get_plaintext_hashes():
328 @return: ListOf(Hash)
331 def get_crypttext_hashes():
333 @return: ListOf(Hash)
336 def get_block_hashes():
338 @return: ListOf(Hash)
341 def get_share_hashes():
343 @return: ListOf(TupleOf(int, Hash))
346 def get_uri_extension():
348 @return: URIExtensionData
353 # hm, we need a solution for forward references in schemas
354 from foolscap.schema import Any
356 FileNode_ = Any() # TODO: foolscap needs constraints on copyables
357 DirectoryNode_ = Any() # TODO: same
358 AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
361 class IURI(Interface):
362 def init_from_string(uri):
363 """Accept a string (as created by my to_string() method) and populate
364 this instance with its data. I am not normally called directly,
365 please use the module-level uri.from_string() function to convert
366 arbitrary URI strings into IURI-providing instances."""
369 """Return False if this URI be used to modify the data. Return True
370 if this URI cannot be used to modify the data."""
373 """Return True if the data can be modified by *somebody* (perhaps
374 someone who has a more powerful URI than this one)."""
377 """Return another IURI instance, which represents a read-only form of
378 this one. If is_readonly() is True, this returns self."""
381 """Return an instance that provides IVerifierURI, which can be used
382 to check on the availability of the file or directory, without
383 providing enough capabilities to actually read or modify the
384 contents. This may return None if the file does not need checking or
385 verification (e.g. LIT URIs).
389 """Return a string of printable ASCII characters, suitable for
390 passing into init_from_string."""
392 class IVerifierURI(Interface):
393 def init_from_string(uri):
394 """Accept a string (as created by my to_string() method) and populate
395 this instance with its data. I am not normally called directly,
396 please use the module-level uri.from_string() function to convert
397 arbitrary URI strings into IURI-providing instances."""
400 """Return a string of printable ASCII characters, suitable for
401 passing into init_from_string."""
403 class IDirnodeURI(Interface):
404 """I am a URI which represents a dirnode."""
407 class IFileURI(Interface):
408 """I am a URI which represents a filenode."""
410 """Return the length (in bytes) of the file that I represent."""
412 class IMutableFileURI(Interface):
413 """I am a URI which represents a mutable filenode."""
414 class INewDirectoryURI(Interface):
416 class IReadonlyNewDirectoryURI(Interface):
420 class IFilesystemNode(Interface):
423 Return the URI that can be used by others to get access to this
424 node. If this node is read-only, the URI will only offer read-only
425 access. If this node is read-write, the URI will offer read-write
428 If you have read-write access to a node and wish to share merely
429 read-only access with others, use get_readonly_uri().
432 def get_readonly_uri():
433 """Return the directory URI that can be used by others to get
434 read-only access to this directory node. The result is a read-only
435 URI, regardless of whether this dirnode is read-only or read-write.
437 If you have merely read-only access to this dirnode,
438 get_readonly_uri() will return the same thing as get_uri().
442 """Return an IVerifierURI instance that represents the
443 'verifiy/refresh capability' for this node. The holder of this
444 capability will be able to renew the lease for this node, protecting
445 it from garbage-collection. They will also be able to ask a server if
446 it holds a share for the file or directory.
449 def get_storage_index():
450 """Return a string with the (binary) storage index in use on this
451 download. This may be None if there is no storage index (i.e. LIT
455 """Return True if this reference provides mutable access to the given
456 file or directory (i.e. if you can modify it), or False if not. Note
457 that even if this reference is read-only, someone else may hold a
458 read-write reference to it."""
461 """Return True if this file or directory is mutable (by *somebody*,
462 not necessarily you), False if it is is immutable. Note that a file
463 might be mutable overall, but your reference to it might be
464 read-only. On the other hand, all references to an immutable file
465 will be read-only; there are no read-write references to an immutable
469 class IMutableFilesystemNode(IFilesystemNode):
472 class IFileNode(IFilesystemNode):
473 def download(target):
474 """Download the file's contents to a given IDownloadTarget"""
476 def download_to_data():
477 """Download the file's contents. Return a Deferred that fires
478 with those contents."""
481 """Return the length (in bytes) of the data this node represents."""
483 def read(consumer, offset=0, size=None):
484 """Download a portion (possibly all) of the file's contents, making
485 them available to the given IConsumer. Return a Deferred that fires
486 (with the consumer) when the consumer is unregistered (either because
487 the last byte has been given to it, or because the consumer threw an
488 exception during write(), possibly because it no longer wants to
489 receive data). The portion downloaded will start at 'offset' and
490 contain 'size' bytes (or the remainder of the file if size==None).
492 The consumer will be used in non-streaming mode: an IPullProducer
493 will be attached to it.
495 The consumer will not receive data right away: several network trips
496 must occur first. The order of events will be::
498 consumer.registerProducer(p, streaming)
499 (if streaming == False)::
500 consumer does p.resumeProducing()
502 consumer does p.resumeProducing()
503 consumer.write(data).. (repeat until all data is written)
504 consumer.unregisterProducer()
505 deferred.callback(consumer)
507 If a download error occurs, or an exception is raised by
508 consumer.registerProducer() or consumer.write(), I will call
509 consumer.unregisterProducer() and then deliver the exception via
510 deferred.errback(). To cancel the download, the consumer should call
511 p.stopProducing(), which will result in an exception being delivered
512 via deferred.errback().
514 A simple download-to-memory consumer example would look like this::
516 class MemoryConsumer:
517 implements(IConsumer)
521 def registerProducer(self, p, streaming):
522 assert streaming == False
525 def write(self, data):
526 self.chunks.append(data)
527 def unregisterProducer(self):
529 d = filenode.read(MemoryConsumer())
530 d.addCallback(lambda mc: "".join(mc.chunks))
535 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
536 """I provide access to a 'mutable file', which retains its identity
537 regardless of what contents are put in it.
539 The consistency-vs-availability problem means that there might be
540 multiple versions of a file present in the grid, some of which might be
541 unrecoverable (i.e. have fewer than 'k' shares). These versions are
542 loosely ordered: each has a sequence number and a hash, and any version
543 with seqnum=N was uploaded by a node which has seen at least one version
546 The 'servermap' (an instance of IMutableFileServerMap) is used to
547 describe the versions that are known to be present in the grid, and which
548 servers are hosting their shares. It is used to represent the 'state of
549 the world', and is used for this purpose by my test-and-set operations.
550 Downloading the contents of the mutable file will also return a
551 servermap. Uploading a new version into the mutable file requires a
552 servermap as input, and the semantics of the replace operation is
553 'replace the file with my new version if it looks like nobody else has
554 changed the file since my previous download'. Because the file is
555 distributed, this is not a perfect test-and-set operation, but it will do
556 its best. If the replace process sees evidence of a simultaneous write,
557 it will signal an UncoordinatedWriteError, so that the caller can take
561 Most readers will want to use the 'best' current version of the file, and
562 should use my 'download_best_version()' method.
564 To unconditionally replace the file, callers should use overwrite(). This
565 is the mode that user-visible mutable files will probably use.
567 To apply some delta to the file, call modify() with a callable modifier
568 function that can apply the modification that you want to make. This is
569 the mode that dirnodes will use, since most directory modification
570 operations can be expressed in terms of deltas to the directory state.
573 Three methods are available for users who need to perform more complex
574 operations. The first is get_servermap(), which returns an up-to-date
575 servermap using a specified mode. The second is download_version(), which
576 downloads a specific version (not necessarily the 'best' one). The third
577 is 'upload', which accepts new contents and a servermap (which must have
578 been updated with MODE_WRITE). The upload method will attempt to apply
579 the new contents as long as no other node has modified the file since the
580 servermap was updated. This might be useful to a caller who wants to
581 merge multiple versions into a single new one.
583 Note that each time the servermap is updated, a specific 'mode' is used,
584 which determines how many peers are queried. To use a servermap for my
585 replace() method, that servermap must have been updated in MODE_WRITE.
586 These modes are defined in allmydata.mutable.common, and consist of
587 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
588 allmydata/mutable/servermap.py for details about the differences.
590 Mutable files are currently limited in size (about 3.5MB max) and can
591 only be retrieved and updated all-at-once, as a single big string. Future
592 versions of our mutable files will remove this restriction.
595 def download_best_version():
596 """Download the 'best' available version of the file, meaning one of
597 the recoverable versions with the highest sequence number. If no
598 uncoordinated writes have occurred, and if enough shares are
599 available, then this will be the most recent version that has been
602 I update an internal servermap with MODE_READ, determine which
603 version of the file is indicated by
604 servermap.best_recoverable_version(), and return a Deferred that
605 fires with its contents. If no version is recoverable, the Deferred
606 will errback with UnrecoverableFileError.
609 def get_size_of_best_version():
610 """Find the size of the version that would be downloaded with
611 download_best_version(), without actually downloading the whole file.
613 I return a Deferred that fires with an integer.
616 def overwrite(new_contents):
617 """Unconditionally replace the contents of the mutable file with new
618 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
619 is only appropriate to use when the new contents of the file are
620 completely unrelated to the old ones, and you do not care about other
623 I return a Deferred that fires (with a PublishStatus object) when the
624 update has completed.
627 def modify(modifier_cb):
628 """Modify the contents of the file, by downloading the current
629 version, applying the modifier function (or bound method), then
630 uploading the new version. I return a Deferred that fires (with a
631 PublishStatus object) when the update is complete.
633 The modifier callable will be given two arguments: a string (with the
634 old contents) and a servermap. As with download_best_version(), the
635 old contents will be from the best recoverable version, but the
636 modifier can use the servermap to make other decisions (such as
637 refusing to apply the delta if there are multiple parallel versions,
638 or if there is evidence of a newer unrecoverable version).
640 The callable should return a string with the new contents. The
641 callable must be prepared to be called multiple times, and must
642 examine the input string to see if the change that it wants to make
643 is already present in the old version. If it does not need to make
644 any changes, it can either return None, or return its input string.
646 If the modifier raises an exception, it will be returned in the
651 def get_servermap(mode):
652 """Return a Deferred that fires with an IMutableFileServerMap
653 instance, updated using the given mode.
656 def download_version(servermap, version):
657 """Download a specific version of the file, using the servermap
658 as a guide to where the shares are located.
660 I return a Deferred that fires with the requested contents, or
661 errbacks with UnrecoverableFileError. Note that a servermap which was
662 updated with MODE_ANYTHING or MODE_READ may not know about shares for
663 all versions (those modes stop querying servers as soon as they can
664 fulfil their goals), so you may want to use MODE_CHECK (which checks
665 everything) to get increased visibility.
668 def upload(new_contents, servermap):
669 """Replace the contents of the file with new ones. This requires a
670 servermap that was previously updated with MODE_WRITE.
672 I attempt to provide test-and-set semantics, in that I will avoid
673 modifying any share that is different than the version I saw in the
674 servermap. However, if another node is writing to the file at the
675 same time as me, I may manage to update some shares while they update
676 others. If I see any evidence of this, I will signal
677 UncoordinatedWriteError, and the file will be left in an inconsistent
678 state (possibly the version you provided, possibly the old version,
679 possibly somebody else's version, and possibly a mix of shares from
682 The recommended response to UncoordinatedWriteError is to either
683 return it to the caller (since they failed to coordinate their
684 writes), or to attempt some sort of recovery. It may be sufficient to
685 wait a random interval (with exponential backoff) and repeat your
686 operation. If I do not signal UncoordinatedWriteError, then I was
687 able to write the new version without incident.
689 I return a Deferred that fires (with a PublishStatus object) when the
690 publish has completed. I will update the servermap in-place with the
691 location of all new shares.
695 """Return this filenode's writekey, or None if the node does not have
696 write-capability. This may be used to assist with data structures
697 that need to make certain data available only to writers, such as the
698 read-write child caps in dirnodes. The recommended process is to have
699 reader-visible data be submitted to the filenode in the clear (where
700 it will be encrypted by the filenode using the readkey), but encrypt
701 writer-visible data using this writekey.
704 class NotEnoughSharesError(Exception):
707 class ExistingChildError(Exception):
708 """A directory node was asked to add or replace a child that already
709 exists, and overwrite= was set to False."""
711 class NoSuchChildError(Exception):
712 """A directory node was asked to fetch a child which does not exist."""
714 class IDirectoryNode(IMutableFilesystemNode):
715 """I represent a name-to-child mapping, holding the tahoe equivalent of a
716 directory. All child names are unicode strings, and all children are some
717 sort of IFilesystemNode (either files or subdirectories).
722 The dirnode ('1') URI returned by this method can be used in
723 set_uri() on a different directory ('2') to 'mount' a reference to
724 this directory ('1') under the other ('2'). This URI is just a
725 string, so it can be passed around through email or other out-of-band
729 def get_readonly_uri():
731 The dirnode ('1') URI returned by this method can be used in
732 set_uri() on a different directory ('2') to 'mount' a reference to
733 this directory ('1') under the other ('2'). This URI is just a
734 string, so it can be passed around through email or other out-of-band
739 """I return a Deferred that fires with a dictionary mapping child
740 name (a unicode string) to (node, metadata_dict) tuples, in which
741 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
742 is a dictionary of metadata."""
745 """I return a Deferred that fires with a boolean, True if there
746 exists a child of the given name, False if not. The child name must
747 be a unicode string."""
750 """I return a Deferred that fires with a specific named child node,
751 either an IFileNode or an IDirectoryNode. The child name must be a
752 unicode string. I raise NoSuchChildError if I do not have a child by
755 def get_metadata_for(name):
756 """I return a Deferred that fires with the metadata dictionary for a
757 specific named child node. This metadata is stored in the *edge*, not
758 in the child, so it is attached to the parent dirnode rather than the
759 child dir-or-file-node. The child name must be a unicode string. I
760 raise NoSuchChildError if I do not have a child by that name."""
762 def set_metadata_for(name, metadata):
763 """I replace any existing metadata for the named child with the new
764 metadata. The child name must be a unicode string. This metadata is
765 stored in the *edge*, not in the child, so it is attached to the
766 parent dirnode rather than the child dir-or-file-node. I return a
767 Deferred (that fires with this dirnode) when the operation is
768 complete. I raise NoSuchChildError if I do not have a child by that
771 def get_child_at_path(path):
772 """Transform a child path into an IDirectoryNode or IFileNode.
774 I perform a recursive series of 'get' operations to find the named
775 descendant node. I return a Deferred that fires with the node, or
776 errbacks with NoSuchChildError if the node could not be found.
778 The path can be either a single string (slash-separated) or a list of
779 path-name elements. All elements must be unicode strings.
782 def get_child_and_metadata_at_path(path):
783 """Transform a child path into an IDirectoryNode/IFileNode and
786 I am like get_child_at_path(), but my Deferred fires with a tuple of
787 (node, metadata). The metadata comes from the last edge. If the path
788 is empty, the metadata will be an empty dictionary.
791 def set_uri(name, child_uri, metadata=None, overwrite=True):
792 """I add a child (by URI) at the specific name. I return a Deferred
793 that fires when the operation finishes. If overwrite= is True, I will
794 replace any existing child of the same name, otherwise an existing
795 child will cause me to return ExistingChildError. The child name must
798 The child_uri could be for a file, or for a directory (either
799 read-write or read-only, using a URI that came from get_uri() ).
801 If metadata= is provided, I will use it as the metadata for the named
802 edge. This will replace any existing metadata. If metadata= is left
803 as the default value of None, I will set ['mtime'] to the current
804 time, and I will set ['ctime'] to the current time if there was not
805 already a child by this name present. This roughly matches the
806 ctime/mtime semantics of traditional filesystems.
808 If this directory node is read-only, the Deferred will errback with a
811 def set_children(entries, overwrite=True):
812 """Add multiple (name, child_uri) pairs (or (name, child_uri,
813 metadata) triples) to a directory node. Returns a Deferred that fires
814 (with None) when the operation finishes. This is equivalent to
815 calling set_uri() multiple times, but is much more efficient. All
816 child names must be unicode strings.
819 def set_node(name, child, metadata=None, overwrite=True):
820 """I add a child at the specific name. I return a Deferred that fires
821 when the operation finishes. This Deferred will fire with the child
822 node that was just added. I will replace any existing child of the
823 same name. The child name must be a unicode string. The 'child'
824 instance must be an instance providing IDirectoryNode or IFileNode.
826 If metadata= is provided, I will use it as the metadata for the named
827 edge. This will replace any existing metadata. If metadata= is left
828 as the default value of None, I will set ['mtime'] to the current
829 time, and I will set ['ctime'] to the current time if there was not
830 already a child by this name present. This roughly matches the
831 ctime/mtime semantics of traditional filesystems.
833 If this directory node is read-only, the Deferred will errback with a
836 def set_nodes(entries, overwrite=True):
837 """Add multiple (name, child_node) pairs (or (name, child_node,
838 metadata) triples) to a directory node. Returns a Deferred that fires
839 (with None) when the operation finishes. This is equivalent to
840 calling set_node() multiple times, but is much more efficient. All
841 child names must be unicode strings."""
844 def add_file(name, uploadable, metadata=None, overwrite=True):
845 """I upload a file (using the given IUploadable), then attach the
846 resulting FileNode to the directory at the given name. I set metadata
847 the same way as set_uri and set_node. The child name must be a
850 I return a Deferred that fires (with the IFileNode of the uploaded
851 file) when the operation completes."""
854 """I remove the child at the specific name. I return a Deferred that
855 fires when the operation finishes. The child name must be a unicode
856 string. I raise NoSuchChildError if I do not have a child by that
859 def create_empty_directory(name, overwrite=True):
860 """I create and attach an empty directory at the given name. The
861 child name must be a unicode string. I return a Deferred that fires
862 when the operation finishes."""
864 def move_child_to(current_child_name, new_parent, new_child_name=None,
866 """I take one of my children and move them to a new parent. The child
867 is referenced by name. On the new parent, the child will live under
868 'new_child_name', which defaults to 'current_child_name'. TODO: what
869 should we do about metadata? I return a Deferred that fires when the
870 operation finishes. The child name must be a unicode string. I raise
871 NoSuchChildError if I do not have a child by that name."""
873 def build_manifest():
874 """I generate a table of everything reachable from this directory.
875 I also compute deep-stats as described below.
877 I return a Monitor. The Monitor's results will be a dictionary with
878 two elements. The 'manifest' element is a list of (path, cap) tuples
879 for nodes (directories and files) reachable from this one. 'path'
880 will be a tuple of unicode strings. The origin dirnode will be
881 represented by an empty path tuple. The 'stats' element is a
882 dictionary, the same that is generated by 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
964 def get_encoder_type():
965 """Return a short string that describes the type of this encoder.
967 There is required to be a global table of encoder classes. This method
968 returns an index into this table; the value at this index is an
969 encoder class, and this encoder is an instance of that class.
972 def get_serialized_params(): # TODO: maybe, maybe not
973 """Return a string that describes the parameters of this encoder.
975 This string can be passed to the decoder to prepare it for handling
976 the encoded shares we create. It might contain more information than
977 was presented to set_params(), if there is some flexibility of
980 This string is intended to be embedded in the URI, so there are
981 several restrictions on its contents. At the moment I'm thinking that
982 this means it may contain hex digits and hyphens, and nothing else.
983 The idea is that the URI contains something like '%s:%s:%s' %
984 (encoder.get_encoder_name(), encoder.get_serialized_params(),
985 b2a(crypttext_hash)), and this is enough information to construct a
989 def get_block_size():
990 """Return the length of the shares that encode() will produce.
993 def encode_proposal(data, desired_share_ids=None):
996 'data' must be a string (or other buffer object), and len(data) must
997 be equal to the 'data_size' value passed earlier to set_params().
999 This will return a Deferred that will fire with two lists. The first
1000 is a list of shares, each of which is a string (or other buffer
1001 object) such that len(share) is the same as what get_share_size()
1002 returned earlier. The second is a list of shareids, in which each is
1003 an integer. The lengths of the two lists will always be equal to each
1004 other. The user should take care to keep each share closely
1005 associated with its shareid, as one is useless without the other.
1007 The length of this output list will normally be the same as the value
1008 provided to the 'max_shares' parameter of set_params(). This may be
1009 different if 'desired_share_ids' is provided.
1011 'desired_share_ids', if provided, is required to be a sequence of
1012 ints, each of which is required to be >= 0 and < max_shares. If not
1013 provided, encode() will produce 'max_shares' shares, as if
1014 'desired_share_ids' were set to range(max_shares). You might use this
1015 if you initially thought you were going to use 10 peers, started
1016 encoding, and then two of the peers dropped out: you could use
1017 desired_share_ids= to skip the work (both memory and CPU) of
1018 producing shares for the peers which are no longer available.
1022 def encode(inshares, desired_share_ids=None):
1023 """Encode some data. This may be called multiple times. Each call is
1026 inshares is a sequence of length required_shares, containing buffers
1027 (i.e. strings), where each buffer contains the next contiguous
1028 non-overlapping segment of the input data. Each buffer is required to
1029 be the same length, and the sum of the lengths of the buffers is
1030 required to be exactly the data_size promised by set_params(). (This
1031 implies that the data has to be padded before being passed to
1032 encode(), unless of course it already happens to be an even multiple
1033 of required_shares in length.)
1035 ALSO: the requirement to break up your data into 'required_shares'
1036 chunks before calling encode() feels a bit surprising, at least from
1037 the point of view of a user who doesn't know how FEC works. It feels
1038 like an implementation detail that has leaked outside the
1039 abstraction barrier. Can you imagine a use case in which the data to
1040 be encoded might already be available in pre-segmented chunks, such
1041 that it is faster or less work to make encode() take a list rather
1042 than splitting a single string?
1044 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1045 is supposed to *produce* shares, so what it *accepts* should be
1046 something other than shares. Other places in this interface use the
1047 word 'data' for that-which-is-not-shares.. maybe we should use that
1050 'desired_share_ids', if provided, is required to be a sequence of
1051 ints, each of which is required to be >= 0 and < max_shares. If not
1052 provided, encode() will produce 'max_shares' shares, as if
1053 'desired_share_ids' were set to range(max_shares). You might use this
1054 if you initially thought you were going to use 10 peers, started
1055 encoding, and then two of the peers dropped out: you could use
1056 desired_share_ids= to skip the work (both memory and CPU) of
1057 producing shares for the peers which are no longer available.
1059 For each call, encode() will return a Deferred that fires with two
1060 lists, one containing shares and the other containing the shareids.
1061 The get_share_size() method can be used to determine the length of
1062 the share strings returned by encode(). Each shareid is a small
1063 integer, exactly as passed into 'desired_share_ids' (or
1064 range(max_shares), if desired_share_ids was not provided).
1066 The shares and their corresponding shareids are required to be kept
1067 together during storage and retrieval. Specifically, the share data is
1068 useless by itself: the decoder needs to be told which share is which
1069 by providing it with both the shareid and the actual share data.
1071 This function will allocate an amount of memory roughly equal to::
1073 (max_shares - required_shares) * get_share_size()
1075 When combined with the memory that the caller must allocate to
1076 provide the input data, this leads to a memory footprint roughly
1077 equal to the size of the resulting encoded shares (i.e. the expansion
1078 factor times the size of the input segment).
1083 # returning a list of (shareidN,shareN) tuples instead of a pair of
1084 # lists (shareids..,shares..). Brian thought the tuples would
1085 # encourage users to keep the share and shareid together throughout
1086 # later processing, Zooko pointed out that the code to iterate
1087 # through two lists is not really more complicated than using a list
1088 # of tuples and there's also a performance improvement
1090 # having 'data_size' not required to be an integral multiple of
1091 # 'required_shares'. Doing this would require encode() to perform
1092 # padding internally, and we'd prefer to have any padding be done
1093 # explicitly by the caller. Yes, it is an abstraction leak, but
1094 # hopefully not an onerous one.
1097 class ICodecDecoder(Interface):
1098 def set_serialized_params(params):
1099 """Set up the parameters of this encoder, from a string returned by
1100 encoder.get_serialized_params()."""
1102 def get_needed_shares():
1103 """Return the number of shares needed to reconstruct the data.
1104 set_serialized_params() is required to be called before this."""
1106 def decode(some_shares, their_shareids):
1107 """Decode a partial list of shares into data.
1109 'some_shares' is required to be a sequence of buffers of sharedata, a
1110 subset of the shares returned by ICodecEncode.encode(). Each share is
1111 required to be of the same length. The i'th element of their_shareids
1112 is required to be the shareid of the i'th buffer in some_shares.
1114 This returns a Deferred which fires with a sequence of buffers. This
1115 sequence will contain all of the segments of the original data, in
1116 order. The sum of the lengths of all of the buffers will be the
1117 'data_size' value passed into the original ICodecEncode.set_params()
1118 call. To get back the single original input block of data, use
1119 ''.join(output_buffers), or you may wish to simply write them in
1120 order to an output file.
1122 Note that some of the elements in the result sequence may be
1123 references to the elements of the some_shares input sequence. In
1124 particular, this means that if those share objects are mutable (e.g.
1125 arrays) and if they are changed, then both the input (the
1126 'some_shares' parameter) and the output (the value given when the
1127 deferred is triggered) will change.
1129 The length of 'some_shares' is required to be exactly the value of
1130 'required_shares' passed into the original ICodecEncode.set_params()
1134 class IEncoder(Interface):
1135 """I take an object that provides IEncryptedUploadable, which provides
1136 encrypted data, and a list of shareholders. I then encode, hash, and
1137 deliver shares to those shareholders. I will compute all the necessary
1138 Merkle hash trees that are necessary to validate the crypttext that
1139 eventually comes back from the shareholders. I provide the URI Extension
1140 Block Hash, and the encoding parameters, both of which must be included
1143 I do not choose shareholders, that is left to the IUploader. I must be
1144 given a dict of RemoteReferences to storage buckets that are ready and
1145 willing to receive data.
1149 """Specify the number of bytes that will be encoded. This must be
1150 peformed before get_serialized_params() can be called.
1152 def set_params(params):
1153 """Override the default encoding parameters. 'params' is a tuple of
1154 (k,d,n), where 'k' is the number of required shares, 'd' is the
1155 shares_of_happiness, and 'n' is the total number of shares that will
1158 Encoding parameters can be set in three ways. 1: The Encoder class
1159 provides defaults (3/7/10). 2: the Encoder can be constructed with
1160 an 'options' dictionary, in which the
1161 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1162 set_params((k,d,n)) can be called.
1164 If you intend to use set_params(), you must call it before
1165 get_share_size or get_param are called.
1168 def set_encrypted_uploadable(u):
1169 """Provide a source of encrypted upload data. 'u' must implement
1170 IEncryptedUploadable.
1172 When this is called, the IEncryptedUploadable will be queried for its
1173 length and the storage_index that should be used.
1175 This returns a Deferred that fires with this Encoder instance.
1177 This must be performed before start() can be called.
1180 def get_param(name):
1181 """Return an encoding parameter, by name.
1183 'storage_index': return a string with the (16-byte truncated SHA-256
1184 hash) storage index to which these shares should be
1187 'share_counts': return a tuple describing how many shares are used:
1188 (needed_shares, shares_of_happiness, total_shares)
1190 'num_segments': return an int with the number of segments that
1193 'segment_size': return an int with the size of each segment.
1195 'block_size': return the size of the individual blocks that will
1196 be delivered to a shareholder's put_block() method. By
1197 knowing this, the shareholder will be able to keep all
1198 blocks in a single file and still provide random access
1199 when reading them. # TODO: can we avoid exposing this?
1201 'share_size': an int with the size of the data that will be stored
1202 on each shareholder. This is aggregate amount of data
1203 that will be sent to the shareholder, summed over all
1204 the put_block() calls I will ever make. It is useful to
1205 determine this size before asking potential
1206 shareholders whether they will grant a lease or not,
1207 since their answers will depend upon how much space we
1208 need. TODO: this might also include some amount of
1209 overhead, like the size of all the hashes. We need to
1210 decide whether this is useful or not.
1212 'serialized_params': a string with a concise description of the
1213 codec name and its parameters. This may be passed
1214 into the IUploadable to let it make sure that
1215 the same file encoded with different parameters
1216 will result in different storage indexes.
1218 Once this is called, set_size() and set_params() may not be called.
1221 def set_shareholders(shareholders):
1222 """Tell the encoder where to put the encoded shares. 'shareholders'
1223 must be a dictionary that maps share number (an integer ranging from
1224 0 to n-1) to an instance that provides IStorageBucketWriter. This
1225 must be performed before start() can be called."""
1228 """Begin the encode/upload process. This involves reading encrypted
1229 data from the IEncryptedUploadable, encoding it, uploading the shares
1230 to the shareholders, then sending the hash trees.
1232 set_encrypted_uploadable() and set_shareholders() must be called
1233 before this can be invoked.
1235 This returns a Deferred that fires with a tuple of
1236 (uri_extension_hash, needed_shares, total_shares, size) when the
1237 upload process is complete. This information, plus the encryption
1238 key, is sufficient to construct the URI.
1241 class IDecoder(Interface):
1242 """I take a list of shareholders and some setup information, then
1243 download, validate, decode, and decrypt data from them, writing the
1244 results to an output file.
1246 I do not locate the shareholders, that is left to the IDownloader. I must
1247 be given a dict of RemoteReferences to storage buckets that are ready to
1252 """I take a file-like object (providing write and close) to which all
1253 the plaintext data will be written.
1255 TODO: producer/consumer . Maybe write() should return a Deferred that
1256 indicates when it will accept more data? But probably having the
1257 IDecoder be a producer is easier to glue to IConsumer pieces.
1260 def set_shareholders(shareholders):
1261 """I take a dictionary that maps share identifiers (small integers)
1262 to RemoteReferences that provide RIBucketReader. This must be called
1266 """I start the download. This process involves retrieving data and
1267 hash chains from the shareholders, using the hashes to validate the
1268 data, decoding the shares into segments, decrypting the segments,
1269 then writing the resulting plaintext to the output file.
1271 I return a Deferred that will fire (with self) when the download is
1275 class IDownloadTarget(Interface):
1276 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1277 # will register itself as a producer. This allows the target to invoke
1278 # downloader.pauseProducing, resumeProducing, and stopProducing.
1280 """Called before any calls to write() or close(). If an error
1281 occurs before any data is available, fail() may be called without
1282 a previous call to open().
1284 'size' is the length of the file being downloaded, in bytes."""
1287 """Output some data to the target."""
1289 """Inform the target that there is no more data to be written."""
1291 """fail() is called to indicate that the download has failed. 'why'
1292 is a Failure object indicating what went wrong. No further methods
1293 will be invoked on the IDownloadTarget after fail()."""
1294 def register_canceller(cb):
1295 """The FileDownloader uses this to register a no-argument function
1296 that the target can call to cancel the download. Once this canceller
1297 is invoked, no further calls to write() or close() will be made."""
1299 """When the FileDownloader is done, this finish() function will be
1300 called. Whatever it returns will be returned to the invoker of
1301 Downloader.download.
1304 class IDownloader(Interface):
1305 def download(uri, target):
1306 """Perform a CHK download, sending the data to the given target.
1307 'target' must provide IDownloadTarget.
1309 Returns a Deferred that fires (with the results of target.finish)
1310 when the download is finished, or errbacks if something went wrong."""
1312 class IEncryptedUploadable(Interface):
1313 def set_upload_status(upload_status):
1314 """Provide an IUploadStatus object that should be filled with status
1315 information. The IEncryptedUploadable is responsible for setting
1316 key-determination progress ('chk'), size, storage_index, and
1317 ciphertext-fetch progress. It may delegate some of this
1318 responsibility to others, in particular to the IUploadable."""
1321 """This behaves just like IUploadable.get_size()."""
1323 def get_all_encoding_parameters():
1324 """Return a Deferred that fires with a tuple of
1325 (k,happy,n,segment_size). The segment_size will be used as-is, and
1326 must match the following constraints: it must be a multiple of k, and
1327 it shouldn't be unreasonably larger than the file size (if
1328 segment_size is larger than filesize, the difference must be stored
1331 This usually passes through to the IUploadable method of the same
1334 The encoder strictly obeys the values returned by this method. To
1335 make an upload use non-default encoding parameters, you must arrange
1336 to control the values that this method returns.
1339 def get_storage_index():
1340 """Return a Deferred that fires with a 16-byte storage index.
1343 def read_encrypted(length, hash_only):
1344 """This behaves just like IUploadable.read(), but returns crypttext
1345 instead of plaintext. If hash_only is True, then this discards the
1346 data (and returns an empty list); this improves efficiency when
1347 resuming an interrupted upload (where we need to compute the
1348 plaintext hashes, but don't need the redundant encrypted data)."""
1350 def get_plaintext_hashtree_leaves(first, last, num_segments):
1351 """Get the leaf nodes of a merkle hash tree over the plaintext
1352 segments, i.e. get the tagged hashes of the given segments. The
1353 segment size is expected to be generated by the IEncryptedUploadable
1354 before any plaintext is read or ciphertext produced, so that the
1355 segment hashes can be generated with only a single pass.
1357 This returns a Deferred which fires with a sequence of hashes, using:
1359 tuple(segment_hashes[first:last])
1361 'num_segments' is used to assert that the number of segments that the
1362 IEncryptedUploadable handled matches the number of segments that the
1363 encoder was expecting.
1365 This method must not be called until the final byte has been read
1366 from read_encrypted(). Once this method is called, read_encrypted()
1367 can never be called again.
1370 def get_plaintext_hash():
1371 """Get the hash of the whole plaintext.
1373 This returns a Deferred which fires with a tagged SHA-256 hash of the
1374 whole plaintext, obtained from hashutil.plaintext_hash(data).
1378 """Just like IUploadable.close()."""
1380 class IUploadable(Interface):
1381 def set_upload_status(upload_status):
1382 """Provide an IUploadStatus object that should be filled with status
1383 information. The IUploadable is responsible for setting
1384 key-determination progress ('chk')."""
1386 def set_default_encoding_parameters(params):
1387 """Set the default encoding parameters, which must be a dict mapping
1388 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1389 'max_segment_size'. These might have an influence on the final
1390 encoding parameters returned by get_all_encoding_parameters(), if the
1391 Uploadable doesn't have more specific preferences.
1393 This call is optional: if it is not used, the Uploadable will use
1394 some built-in defaults. If used, this method must be called before
1395 any other IUploadable methods to have any effect.
1399 """Return a Deferred that will fire with the length of the data to be
1400 uploaded, in bytes. This will be called before the data is actually
1401 used, to compute encoding parameters.
1404 def get_all_encoding_parameters():
1405 """Return a Deferred that fires with a tuple of
1406 (k,happy,n,segment_size). The segment_size will be used as-is, and
1407 must match the following constraints: it must be a multiple of k, and
1408 it shouldn't be unreasonably larger than the file size (if
1409 segment_size is larger than filesize, the difference must be stored
1412 The relative values of k and n allow some IUploadables to request
1413 better redundancy than others (in exchange for consuming more space
1416 Larger values of segment_size reduce hash overhead, while smaller
1417 values reduce memory footprint and cause data to be delivered in
1418 smaller pieces (which may provide a smoother and more predictable
1419 download experience).
1421 The encoder strictly obeys the values returned by this method. To
1422 make an upload use non-default encoding parameters, you must arrange
1423 to control the values that this method returns. One way to influence
1424 them may be to call set_encoding_parameters() before calling
1425 get_all_encoding_parameters().
1428 def get_encryption_key():
1429 """Return a Deferred that fires with a 16-byte AES key. This key will
1430 be used to encrypt the data. The key will also be hashed to derive
1433 Uploadables which want to achieve convergence should hash their file
1434 contents and the serialized_encoding_parameters to form the key
1435 (which of course requires a full pass over the data). Uploadables can
1436 use the upload.ConvergentUploadMixin class to achieve this
1439 Uploadables which do not care about convergence (or do not wish to
1440 make multiple passes over the data) can simply return a
1441 strongly-random 16 byte string.
1443 get_encryption_key() may be called multiple times: the IUploadable is
1444 required to return the same value each time.
1448 """Return a Deferred that fires with a list of strings (perhaps with
1449 only a single element) which, when concatenated together, contain the
1450 next 'length' bytes of data. If EOF is near, this may provide fewer
1451 than 'length' bytes. The total number of bytes provided by read()
1452 before it signals EOF must equal the size provided by get_size().
1454 If the data must be acquired through multiple internal read
1455 operations, returning a list instead of a single string may help to
1456 reduce string copies.
1458 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1459 so a 10kB file means length=3kB, 100kB file means length=30kB,
1460 and >=1MB file means length=300kB.
1462 This method provides for a single full pass through the data. Later
1463 use cases may desire multiple passes or access to only parts of the
1464 data (such as a mutable file making small edits-in-place). This API
1465 will be expanded once those use cases are better understood.
1469 """The upload is finished, and whatever filehandle was in use may be
1472 class IUploadResults(Interface):
1473 """I am returned by upload() methods. I contain a number of public
1474 attributes which can be read to determine the results of the upload. Some
1475 of these are functional, some are timing information. All of these may be
1478 .file_size : the size of the file, in bytes
1479 .uri : the CHK read-cap for the file
1480 .ciphertext_fetched : how many bytes were fetched by the helper
1481 .sharemap : dict mapping share number to placement string
1482 .servermap : dict mapping server peerid to a set of share numbers
1483 .timings : dict of timing information, mapping name to seconds (float)
1484 total : total upload time, start to finish
1485 storage_index : time to compute the storage index
1486 peer_selection : time to decide which peers will be used
1487 contacting_helper : initial helper query to upload/no-upload decision
1488 existence_check : helper pre-upload existence check
1489 helper_total : initial helper query to helper finished pushing
1490 cumulative_fetch : helper waiting for ciphertext requests
1491 total_fetch : helper start to last ciphertext response
1492 cumulative_encoding : just time spent in zfec
1493 cumulative_sending : just time spent waiting for storage servers
1494 hashes_and_close : last segment push to shareholder close
1495 total_encode_and_push : first encode to shareholder close
1499 class IDownloadResults(Interface):
1500 """I am created internally by download() methods. I contain a number of
1501 public attributes which contain details about the download process.::
1503 .file_size : the size of the file, in bytes
1504 .servers_used : set of server peerids that were used during download
1505 .server_problems : dict mapping server peerid to a problem string. Only
1506 servers that had problems (bad hashes, disconnects) are
1508 .servermap : dict mapping server peerid to a set of share numbers. Only
1509 servers that had any shares are listed here.
1510 .timings : dict of timing information, mapping name to seconds (float)
1511 peer_selection : time to ask servers about shares
1512 servers_peer_selection : dict of peerid to DYHB-query time
1513 uri_extension : time to fetch a copy of the URI extension block
1514 hashtrees : time to fetch the hash trees
1515 segments : time to fetch, decode, and deliver segments
1516 cumulative_fetch : time spent waiting for storage servers
1517 cumulative_decode : just time spent in zfec
1518 cumulative_decrypt : just time spent in decryption
1519 total : total download time, start to finish
1520 fetch_per_server : dict of peerid to list of per-segment fetch times
1524 class IUploader(Interface):
1525 def upload(uploadable):
1526 """Upload the file. 'uploadable' must impement IUploadable. This
1527 returns a Deferred which fires with an UploadResults instance, from
1528 which the URI of the file can be obtained as results.uri ."""
1530 def upload_ssk(write_capability, new_version, uploadable):
1531 """TODO: how should this work?"""
1533 class ICheckable(Interface):
1534 def check(monitor, verify=False):
1535 """Check upon my health, optionally repairing any problems.
1537 This returns a Deferred that fires with an instance that provides
1538 ICheckerResults, or None if the object is non-distributed (i.e. LIT
1541 The monitor will be checked periodically to see if the operation has
1542 been cancelled. If so, no new queries will be sent, and the Deferred
1543 will fire (with a OperationCancelledError) immediately.
1545 Filenodes and dirnodes (which provide IFilesystemNode) are also
1546 checkable. Instances that represent verifier-caps will be checkable
1547 but not downloadable. Some objects (like LIT files) do not actually
1548 live in the grid, and their checkers return None (non-distributed
1549 files are always healthy).
1551 If verify=False, a relatively lightweight check will be performed: I
1552 will ask all servers if they have a share for me, and I will believe
1553 whatever they say. If there are at least N distinct shares on the
1554 grid, my results will indicate r.is_healthy()==True. This requires a
1555 roundtrip to each server, but does not transfer very much data, so
1556 the network bandwidth is fairly low.
1558 If verify=True, a more resource-intensive check will be performed:
1559 every share will be downloaded, and the hashes will be validated on
1560 every bit. I will ignore any shares that failed their hash checks. If
1561 there are at least N distinct valid shares on the grid, my results
1562 will indicate r.is_healthy()==True. This requires N/k times as much
1563 download bandwidth (and server disk IO) as a regular download. If a
1564 storage server is holding a corrupt share, or is experiencing memory
1565 failures during retrieval, or is malicious or buggy, then
1566 verification will detect the problem, but checking will not.
1568 TODO: any problems seen during checking will be reported to the
1569 health-manager.furl, a centralized object which is responsible for
1570 figuring out why files are unhealthy so corrective action can be
1574 def check_and_repair(monitor, verify=False):
1575 """Like check(), but if the file/directory is not healthy, attempt to
1578 Any non-healthy result will cause an immediate repair operation, to
1579 generate and upload new shares. After repair, the file will be as
1580 healthy as we can make it. Details about what sort of repair is done
1581 will be put in the check-and-repair results. The Deferred will not
1582 fire until the repair is complete.
1584 This returns a Deferred which fires with an instance of
1585 ICheckAndRepairResults."""
1587 class IDeepCheckable(Interface):
1588 def start_deep_check(verify=False):
1589 """Check upon the health of me and everything I can reach.
1591 This is a recursive form of check(), useable only on dirnodes.
1593 I return a Monitor, with results that are an IDeepCheckResults
1597 def start_deep_check_and_repair(verify=False):
1598 """Check upon the health of me and everything I can reach. Repair
1599 anything that isn't healthy.
1601 This is a recursive form of check_and_repair(), useable only on
1604 I return a Monitor, with results that are an
1605 IDeepCheckAndRepairResults object.
1608 class ICheckerResults(Interface):
1609 """I contain the detailed results of a check/verify operation.
1612 def get_storage_index():
1613 """Return a string with the (binary) storage index."""
1614 def get_storage_index_string():
1615 """Return a string with the (printable) abbreviated storage index."""
1617 """Return the (string) URI of the object that was checked."""
1620 """Return a boolean, True if the file/dir is fully healthy, False if
1621 it is damaged in any way. Non-distributed LIT files always return
1624 def is_recoverable():
1625 """Return a boolean, True if the file/dir can be recovered, False if
1626 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1627 files always return True."""
1629 def needs_rebalancing():
1630 """Return a boolean, True if the file/dir's reliability could be
1631 improved by moving shares to new servers. Non-distributed LIT files
1632 always returne False."""
1636 """Return a dictionary that describes the state of the file/dir.
1637 Non-distributed LIT files always return an empty dictionary. Normal
1638 files and directories return a dictionary with the following keys
1639 (note that these use base32-encoded strings rather than binary ones)
1640 (also note that for mutable files, these counts are for the 'best'
1643 count-shares-good: the number of distinct good shares that were found
1644 count-shares-needed: 'k', the number of shares required for recovery
1645 count-shares-expected: 'N', the number of total shares generated
1646 count-good-share-hosts: the number of distinct storage servers with
1647 good shares. If this number is less than
1648 count-shares-good, then some shares are
1649 doubled up, increasing the correlation of
1650 failures. This indicates that one or more
1651 shares should be moved to an otherwise unused
1652 server, if one is available.
1653 count-corrupt-shares: the number of shares with integrity failures
1654 list-corrupt-shares: a list of 'share locators', one for each share
1655 that was found to be corrupt. Each share
1656 locator is a list of (serverid, storage_index,
1658 servers-responding: list of (binary) storage server identifiers,
1659 one for each server which responded to the share
1661 sharemap: dict mapping share identifier to list of serverids
1662 (binary strings). This indicates which servers are holding
1663 which shares. For immutable files, the shareid is an
1664 integer (the share number, from 0 to N-1). For mutable
1665 files, it is a string of the form 'seq%d-%s-sh%d',
1666 containing the sequence number, the roothash, and the
1669 The following keys are most relevant for mutable files, but immutable
1670 files will provide sensible values too::
1672 count-wrong-shares: the number of shares for versions other than the
1673 'best' one (which is defined as being the
1674 recoverable version with the highest sequence
1675 number, then the highest roothash). These are
1676 either leftover shares from an older version
1677 (perhaps on a server that was offline when an
1678 update occurred), shares from an unrecoverable
1679 newer version, or shares from an alternate
1680 current version that results from an
1681 uncoordinated write collision. For a healthy
1682 file, this will equal 0.
1684 count-recoverable-versions: the number of recoverable versions of
1685 the file. For a healthy file, this will
1688 count-unrecoverable-versions: the number of unrecoverable versions
1689 of the file. For a healthy file, this
1695 """Return a string with a brief (one-line) summary of the results."""
1698 """Return a list of strings with more detailed results."""
1700 class ICheckAndRepairResults(Interface):
1701 """I contain the detailed results of a check/verify/repair operation.
1703 The IFilesystemNode.check()/verify()/repair() methods all return
1704 instances that provide ICheckAndRepairResults.
1707 def get_storage_index():
1708 """Return a string with the (binary) storage index."""
1709 def get_storage_index_string():
1710 """Return a string with the (printable) abbreviated storage index."""
1711 def get_repair_attempted():
1712 """Return a boolean, True if a repair was attempted."""
1713 def get_repair_successful():
1714 """Return a boolean, True if repair was attempted and the file/dir
1715 was fully healthy afterwards. False if no repair was attempted or if
1716 a repair attempt failed."""
1717 def get_pre_repair_results():
1718 """Return an ICheckerResults instance that describes the state of the
1719 file/dir before any repair was attempted."""
1720 def get_post_repair_results():
1721 """Return an ICheckerResults instance that describes the state of the
1722 file/dir after any repair was attempted. If no repair was attempted,
1723 the pre-repair and post-repair results will be identical."""
1726 class IDeepCheckResults(Interface):
1727 """I contain the results of a deep-check operation.
1729 This is returned by a call to ICheckable.deep_check().
1732 def get_root_storage_index_string():
1733 """Return the storage index (abbreviated human-readable string) of
1734 the first object checked."""
1736 """Return a dictionary with the following keys::
1738 count-objects-checked: count of how many objects were checked
1739 count-objects-healthy: how many of those objects were completely
1741 count-objects-unhealthy: how many were damaged in some way
1742 count-objects-unrecoverable: how many were unrecoverable
1743 count-corrupt-shares: how many shares were found to have
1744 corruption, summed over all objects
1748 def get_corrupt_shares():
1749 """Return a set of (serverid, storage_index, sharenum) for all shares
1750 that were found to be corrupt. Both serverid and storage_index are
1753 def get_all_results():
1754 """Return a dictionary mapping pathname (a tuple of strings, ready to
1755 be slash-joined) to an ICheckerResults instance, one for each object
1756 that was checked."""
1758 def get_results_for_storage_index(storage_index):
1759 """Retrive the ICheckerResults instance for the given (binary)
1760 storage index. Raises KeyError if there are no results for that
1764 """Return a dictionary with the same keys as
1765 IDirectoryNode.deep_stats()."""
1767 class IDeepCheckAndRepairResults(Interface):
1768 """I contain the results of a deep-check-and-repair operation.
1770 This is returned by a call to ICheckable.deep_check_and_repair().
1773 def get_root_storage_index_string():
1774 """Return the storage index (abbreviated human-readable string) of
1775 the first object checked."""
1777 """Return a dictionary with the following keys::
1779 count-objects-checked: count of how many objects were checked
1780 count-objects-healthy-pre-repair: how many of those objects were
1781 completely healthy (before any
1783 count-objects-unhealthy-pre-repair: how many were damaged in
1785 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1786 count-objects-healthy-post-repair: how many of those objects were
1787 completely healthy (after any
1789 count-objects-unhealthy-post-repair: how many were damaged in
1791 count-objects-unrecoverable-post-repair: how many were
1793 count-repairs-attempted: repairs were attempted on this many
1794 objects. The count-repairs- keys will
1795 always be provided, however unless
1796 repair=true is present, they will all
1798 count-repairs-successful: how many repairs resulted in healthy
1800 count-repairs-unsuccessful: how many repairs resulted did not
1801 results in completely healthy objects
1802 count-corrupt-shares-pre-repair: how many shares were found to
1803 have corruption, summed over all
1804 objects examined (before any
1806 count-corrupt-shares-post-repair: how many shares were found to
1807 have corruption, summed over all
1808 objects examined (after any
1813 """Return a dictionary with the same keys as
1814 IDirectoryNode.deep_stats()."""
1816 def get_corrupt_shares():
1817 """Return a set of (serverid, storage_index, sharenum) for all shares
1818 that were found to be corrupt before any repair was attempted. Both
1819 serverid and storage_index are binary.
1821 def get_remaining_corrupt_shares():
1822 """Return a set of (serverid, storage_index, sharenum) for all shares
1823 that were found to be corrupt after any repair was completed. Both
1824 serverid and storage_index are binary. These are shares that need
1825 manual inspection and probably deletion.
1827 def get_all_results():
1828 """Return a dictionary mapping pathname (a tuple of strings, ready to
1829 be slash-joined) to an ICheckAndRepairResults instance, one for each
1830 object that was checked."""
1833 class IRepairable(Interface):
1834 def repair(checker_results):
1835 """Attempt to repair the given object. Returns a Deferred that fires
1836 with a IRepairResults object.
1838 I must be called with an object that implements ICheckerResults, as
1839 proof that you have actually discovered a problem with this file. I
1840 will use the data in the checker results to guide the repair process,
1841 such as which servers provided bad data and should therefore be
1842 avoided. The ICheckerResults object is inside the
1843 ICheckAndRepairResults object, which is returned by the
1844 ICheckable.check() method::
1846 d = filenode.check(repair=False)
1847 def _got_results(check_and_repair_results):
1848 check_results = check_and_repair_results.get_pre_repair_results()
1849 return filenode.repair(check_results)
1850 d.addCallback(_got_results)
1854 class IRepairResults(Interface):
1855 """I contain the results of a repair operation."""
1858 class IClient(Interface):
1859 def upload(uploadable):
1860 """Upload some data into a CHK, get back the UploadResults for it.
1861 @param uploadable: something that implements IUploadable
1862 @return: a Deferred that fires with the UploadResults instance.
1863 To get the URI for this file, use results.uri .
1866 def create_mutable_file(contents=""):
1867 """Create a new mutable file with contents, get back the URI string.
1868 @param contents: the initial contents to place in the file.
1869 @return: a Deferred that fires with tne (string) SSK URI for the new
1873 def create_empty_dirnode():
1874 """Create a new dirnode, empty and unattached.
1875 @return: a Deferred that fires with the new IDirectoryNode instance.
1878 def create_node_from_uri(uri):
1879 """Create a new IFilesystemNode instance from the uri, synchronously.
1880 @param uri: a string or IURI-providing instance. This could be for a
1881 LiteralFileNode, a CHK file node, a mutable file node, or
1883 @return: an instance that provides IFilesystemNode (or more usefully one
1884 of its subclasses). File-specifying URIs will result in
1885 IFileNode or IMutableFileNode -providing instances, like
1886 FileNode, LiteralFileNode, or MutableFileNode.
1887 Directory-specifying URIs will result in
1888 IDirectoryNode-providing instances, like NewDirectoryNode.
1891 class IClientStatus(Interface):
1892 def list_all_uploads():
1893 """Return a list of uploader objects, one for each upload which
1894 currently has an object available (tracked with weakrefs). This is
1895 intended for debugging purposes."""
1896 def list_active_uploads():
1897 """Return a list of active IUploadStatus objects."""
1898 def list_recent_uploads():
1899 """Return a list of IUploadStatus objects for the most recently
1902 def list_all_downloads():
1903 """Return a list of downloader objects, one for each download which
1904 currently has an object available (tracked with weakrefs). This is
1905 intended for debugging purposes."""
1906 def list_active_downloads():
1907 """Return a list of active IDownloadStatus objects."""
1908 def list_recent_downloads():
1909 """Return a list of IDownloadStatus objects for the most recently
1910 started downloads."""
1912 class IUploadStatus(Interface):
1914 """Return a timestamp (float with seconds since epoch) indicating
1915 when the operation was started."""
1916 def get_storage_index():
1917 """Return a string with the (binary) storage index in use on this
1918 upload. Returns None if the storage index has not yet been
1921 """Return an integer with the number of bytes that will eventually
1922 be uploaded for this file. Returns None if the size is not yet known.
1925 """Return True if this upload is using a Helper, False if not."""
1927 """Return a string describing the current state of the upload
1930 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
1931 each from 0.0 to 1.0 . 'chk' describes how much progress has been
1932 made towards hashing the file to determine a CHK encryption key: if
1933 non-convergent encryption is in use, this will be trivial, otherwise
1934 the whole file must be hashed. 'ciphertext' describes how much of the
1935 ciphertext has been pushed to the helper, and is '1.0' for non-helper
1936 uploads. 'encode_and_push' describes how much of the encode-and-push
1937 process has finished: for helper uploads this is dependent upon the
1938 helper providing progress reports. It might be reasonable to add all
1939 three numbers and report the sum to the user."""
1941 """Return True if the upload is currently active, False if not."""
1943 """Return an instance of UploadResults (which contains timing and
1944 sharemap information). Might return None if the upload is not yet
1947 """Each upload status gets a unique number: this method returns that
1948 number. This provides a handle to this particular upload, so a web
1949 page can generate a suitable hyperlink."""
1951 class IDownloadStatus(Interface):
1953 """Return a timestamp (float with seconds since epoch) indicating
1954 when the operation was started."""
1955 def get_storage_index():
1956 """Return a string with the (binary) storage index in use on this
1957 download. This may be None if there is no storage index (i.e. LIT
1960 """Return an integer with the number of bytes that will eventually be
1961 retrieved for this file. Returns None if the size is not yet known.
1964 """Return True if this download is using a Helper, False if not."""
1966 """Return a string describing the current state of the download
1969 """Returns a float (from 0.0 to 1.0) describing the amount of the
1970 download that has completed. This value will remain at 0.0 until the
1971 first byte of plaintext is pushed to the download target."""
1973 """Return True if the download is currently active, False if not."""
1975 """Each download status gets a unique number: this method returns
1976 that number. This provides a handle to this particular download, so a
1977 web page can generate a suitable hyperlink."""
1979 class IServermapUpdaterStatus(Interface):
1981 class IPublishStatus(Interface):
1983 class IRetrieveStatus(Interface):
1986 class NotCapableError(Exception):
1987 """You have tried to write to a read-only node."""
1989 class BadWriteEnablerError(Exception):
1992 class RIControlClient(RemoteInterface):
1994 def wait_for_client_connections(num_clients=int):
1995 """Do not return until we have connections to at least NUM_CLIENTS
1999 def upload_from_file_to_uri(filename=str, convergence=ChoiceOf(None, StringConstraint(2**20))):
2000 """Upload a file to the grid. This accepts a filename (which must be
2001 absolute) that points to a file on the node's local disk. The node will
2002 read the contents of this file, upload it to the grid, then return the
2003 URI at which it was uploaded. If convergence is None then a random
2004 encryption key will be used, else the plaintext will be hashed, then
2005 that hash will be mixed together with the "convergence" string to form
2010 def download_from_uri_to_file(uri=URI, filename=str):
2011 """Download a file from the grid, placing it on the node's local disk
2012 at the given filename (which must be absolute[?]). Returns the
2013 absolute filename where the file was written."""
2018 def get_memory_usage():
2019 """Return a dict describes the amount of memory currently in use. The
2020 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2021 measuring memory consupmtion in bytes."""
2022 return DictOf(str, int)
2024 def speed_test(count=int, size=int, mutable=Any()):
2025 """Write 'count' tempfiles to disk, all of the given size. Measure
2026 how long (in seconds) it takes to upload them all to the servers.
2027 Then measure how long it takes to download all of them. If 'mutable'
2028 is 'create', time creation of mutable files. If 'mutable' is
2029 'upload', then time access to the same mutable file instead of
2032 Returns a tuple of (upload_time, download_time).
2034 return (float, float)
2036 def measure_peer_response_time():
2037 """Send a short message to each connected peer, and measure the time
2038 it takes for them to respond to it. This is a rough measure of the
2039 application-level round trip time.
2041 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2044 return DictOf(Nodeid, float)
2046 UploadResults = Any() #DictOf(str, str)
2048 class RIEncryptedUploadable(RemoteInterface):
2049 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2054 def get_all_encoding_parameters():
2055 return (int, int, int, long)
2057 def read_encrypted(offset=Offset, length=ReadSize):
2060 def get_plaintext_hashtree_leaves(first=int, last=int, num_segments=int):
2063 def get_plaintext_hash():
2070 class RICHKUploadHelper(RemoteInterface):
2071 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2073 def upload(reader=RIEncryptedUploadable):
2074 return UploadResults
2077 class RIHelper(RemoteInterface):
2078 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2080 def upload_chk(si=StorageIndex):
2081 """See if a file with a given storage index needs uploading. The
2082 helper will ask the appropriate storage servers to see if the file
2083 has already been uploaded. If so, the helper will return a set of
2084 'upload results' that includes whatever hashes are needed to build
2085 the read-cap, and perhaps a truncated sharemap.
2087 If the file has not yet been uploaded (or if it was only partially
2088 uploaded), the helper will return an empty upload-results dictionary
2089 and also an RICHKUploadHelper object that will take care of the
2090 upload process. The client should call upload() on this object and
2091 pass it a reference to an RIEncryptedUploadable object that will
2092 provide ciphertext. When the upload is finished, the upload() method
2093 will finish and return the upload results.
2095 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2098 class RIStatsProvider(RemoteInterface):
2099 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2101 Provides access to statistics and monitoring information.
2106 returns a dictionary containing 'counters' and 'stats', each a dictionary
2107 with string counter/stat name keys, and numeric values. counters are
2108 monotonically increasing measures of work done, and stats are instantaneous
2109 measures (potentially time averaged internally)
2111 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2113 class RIStatsGatherer(RemoteInterface):
2114 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2116 Provides a monitoring service for centralised collection of stats
2119 def provide(provider=RIStatsProvider, nickname=str):
2121 @param provider: a stats collector instance which should be polled
2122 periodically by the gatherer to collect stats.
2123 @param nickname: a name useful to identify the provided client
2128 class IStatsProducer(Interface):
2131 returns a dictionary, with str keys representing the names of stats
2132 to be monitored, and numeric values.
2135 class RIKeyGenerator(RemoteInterface):
2136 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2138 Provides a service offering to make RSA key pairs.
2141 def get_rsa_key_pair(key_size=int):
2143 @param key_size: the size of the signature key.
2144 @return: tuple(verifying_key, signing_key)
2146 return TupleOf(str, str)
2149 class FileTooLargeError(Exception):