2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
8 Hash = StringConstraint(maxLength=HASH_SIZE,
9 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
10 Nodeid = StringConstraint(maxLength=20,
11 minLength=20) # binary format 20-byte SHA1 hash
12 FURL = StringConstraint(1000)
13 StorageIndex = StringConstraint(16)
14 URI = StringConstraint(300) # kind of arbitrary
16 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
18 ShareData = StringConstraint(None)
19 URIExtensionData = StringConstraint(1000)
20 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
22 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
23 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
24 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
25 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
27 class RIStubClient(RemoteInterface):
28 """Each client publishes a service announcement for a dummy object called
29 the StubClient. This object doesn't actually offer any services, but the
30 announcement helps the Introducer keep track of which clients are
31 subscribed (so the grid admin can keep track of things like the size of
32 the grid and the client versions in use. This is the (empty)
33 RemoteInterface for the StubClient."""
35 class RIBucketWriter(RemoteInterface):
36 """ Objects of this kind live on the server side. """
37 def write(offset=Offset, data=ShareData):
42 If the data that has been written is incomplete or inconsistent then
43 the server will throw the data away, else it will store it for future
49 """Abandon all the data that has been written.
53 class RIBucketReader(RemoteInterface):
54 def read(offset=Offset, length=ReadSize):
57 def advise_corrupt_share(reason=str):
58 """Clients who discover hash failures in shares that they have
59 downloaded from me will use this method to inform me about the
60 failures. I will record their concern so that my operator can
61 manually inspect the shares in question. I return None.
63 This is a wrapper around RIStorageServer.advise_corrupt_share(),
64 which is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
85 class RIStorageServer(RemoteInterface):
86 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
90 Return a dictionary of version information.
92 return DictOf(str, Any())
94 def allocate_buckets(storage_index=StorageIndex,
95 renew_secret=LeaseRenewSecret,
96 cancel_secret=LeaseCancelSecret,
97 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
98 allocated_size=Offset, canary=Referenceable):
100 @param storage_index: the index of the bucket to be created or
102 @param sharenums: these are the share numbers (probably between 0 and
103 99) that the sender is proposing to store on this
105 @param renew_secret: This is the secret used to protect bucket refresh
106 This secret is generated by the client and
107 stored for later comparison by the server. Each
108 server is given a different secret.
109 @param cancel_secret: Like renew_secret, but protects bucket decref.
110 @param canary: If the canary is lost before close(), the bucket is
112 @return: tuple of (alreadygot, allocated), where alreadygot is what we
113 already have and allocated is what we hereby agree to accept.
114 New leases are added for shares in both lists.
116 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
117 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
119 def add_lease(storage_index=StorageIndex,
120 renew_secret=LeaseRenewSecret,
121 cancel_secret=LeaseCancelSecret):
123 Add a new lease on the given bucket. If the renew_secret matches an
124 existing lease, that lease will be renewed instead. If there is no
125 bucket for the given storage_index, return silently. (note that in
126 tahoe-1.3.0 and earlier, IndexError was raised if there was no
129 return Any() # returns None now, but future versions might change
131 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
133 Renew the lease on a given bucket, resetting the timer to 31 days.
134 Some networks will use this, some will not. If there is no bucket for
135 the given storage_index, IndexError will be raised.
137 For mutable shares, if the given renew_secret does not match an
138 existing lease, IndexError will be raised with a note listing the
139 server-nodeids on the existing leases, so leases on migrated shares
140 can be renewed or cancelled. For immutable shares, IndexError
141 (without the note) will be raised.
145 def cancel_lease(storage_index=StorageIndex,
146 cancel_secret=LeaseCancelSecret):
148 Cancel the lease on a given bucket. If this was the last lease on the
149 bucket, the bucket will be deleted. If there is no bucket for the
150 given storage_index, IndexError will be raised.
152 For mutable shares, if the given cancel_secret does not match an
153 existing lease, IndexError will be raised with a note listing the
154 server-nodeids on the existing leases, so leases on migrated shares
155 can be renewed or cancelled. For immutable shares, IndexError
156 (without the note) will be raised.
160 def get_buckets(storage_index=StorageIndex):
161 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
165 def slot_readv(storage_index=StorageIndex,
166 shares=ListOf(int), readv=ReadVector):
167 """Read a vector from the numbered shares associated with the given
168 storage index. An empty shares list means to return data from all
169 known shares. Returns a dictionary with one key per share."""
170 return DictOf(int, ReadData) # shnum -> results
172 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
173 secrets=TupleOf(WriteEnablerSecret,
176 tw_vectors=TestAndWriteVectorsForShares,
179 """General-purpose test-and-set operation for mutable slots. Perform
180 a bunch of comparisons against the existing shares. If they all pass,
181 then apply a bunch of write vectors to those shares. Then use the
182 read vectors to extract data from all the shares and return the data.
184 This method is, um, large. The goal is to allow clients to update all
185 the shares associated with a mutable file in a single round trip.
187 @param storage_index: the index of the bucket to be created or
189 @param write_enabler: a secret that is stored along with the slot.
190 Writes are accepted from any caller who can
191 present the matching secret. A different secret
192 should be used for each slot*server pair.
193 @param renew_secret: This is the secret used to protect bucket refresh
194 This secret is generated by the client and
195 stored for later comparison by the server. Each
196 server is given a different secret.
197 @param cancel_secret: Like renew_secret, but protects bucket decref.
199 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
200 cancel_secret). The first is required to perform any write. The
201 latter two are used when allocating new shares. To simply acquire a
202 new lease on existing shares, use an empty testv and an empty writev.
204 Each share can have a separate test vector (i.e. a list of
205 comparisons to perform). If all vectors for all shares pass, then all
206 writes for all shares are recorded. Each comparison is a 4-tuple of
207 (offset, length, operator, specimen), which effectively does a bool(
208 (read(offset, length)) OPERATOR specimen ) and only performs the
209 write if all these evaluate to True. Basic test-and-set uses 'eq'.
210 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
211 Write-if-same-or-newer uses 'le'.
213 Reads from the end of the container are truncated, and missing shares
214 behave like empty ones, so to assert that a share doesn't exist (for
215 use when creating a new share), use (0, 1, 'eq', '').
217 The write vector will be applied to the given share, expanding it if
218 necessary. A write vector applied to a share number that did not
219 exist previously will cause that share to be created.
221 Each write vector is accompanied by a 'new_length' argument. If
222 new_length is not None, use it to set the size of the container. This
223 can be used to pre-allocate space for a series of upcoming writes, or
224 truncate existing data. If the container is growing, new_length will
225 be applied before datav. If the container is shrinking, it will be
226 applied afterwards. If new_length==0, the share will be deleted.
228 The read vector is used to extract data from all known shares,
229 *before* any writes have been applied. The same vector is used for
230 all shares. This captures the state that was tested by the test
233 This method returns two values: a boolean and a dict. The boolean is
234 True if the write vectors were applied, False if not. The dict is
235 keyed by share number, and each value contains a list of strings, one
236 for each element of the read vector.
238 If the write_enabler is wrong, this will raise BadWriteEnablerError.
239 To enable share migration (using update_write_enabler), the exception
240 will have the nodeid used for the old write enabler embedded in it,
241 in the following string::
243 The write enabler was recorded by nodeid '%s'.
245 Note that the nodeid here is encoded using the same base32 encoding
246 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
249 return TupleOf(bool, DictOf(int, ReadData))
251 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
252 shnum=int, reason=str):
253 """Clients who discover hash failures in shares that they have
254 downloaded from me will use this method to inform me about the
255 failures. I will record their concern so that my operator can
256 manually inspect the shares in question. I return None.
258 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
259 (binary) storage index string, and 'shnum' is the integer share
260 number. 'reason' is a human-readable explanation of the problem,
261 probably including some expected hash values and the computed ones
262 which did not match. Corruption advisories for mutable shares should
263 include a hash of the public key (the same value that appears in the
264 mutable-file verify-cap), since the current share format does not
268 class IStorageBucketWriter(Interface):
270 Objects of this kind live on the client side.
272 def put_block(segmentnum=int, data=ShareData):
273 """@param data: For most segments, this data will be 'blocksize'
274 bytes in length. The last segment might be shorter.
275 @return: a Deferred that fires (with None) when the operation completes
278 def put_plaintext_hashes(hashes=ListOf(Hash)):
280 @return: a Deferred that fires (with None) when the operation completes
283 def put_crypttext_hashes(hashes=ListOf(Hash)):
285 @return: a Deferred that fires (with None) when the operation completes
288 def put_block_hashes(blockhashes=ListOf(Hash)):
290 @return: a Deferred that fires (with None) when the operation completes
293 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
295 @return: a Deferred that fires (with None) when the operation completes
298 def put_uri_extension(data=URIExtensionData):
299 """This block of data contains integrity-checking information (hashes
300 of plaintext, crypttext, and shares), as well as encoding parameters
301 that are necessary to recover the data. This is a serialized dict
302 mapping strings to other strings. The hash of this data is kept in
303 the URI and verified before any of the data is used. All buckets for
304 a given file contain identical copies of this data.
306 The serialization format is specified with the following pseudocode:
307 for k in sorted(dict.keys()):
308 assert re.match(r'^[a-zA-Z_\-]+$', k)
309 write(k + ':' + netstring(dict[k]))
311 @return: a Deferred that fires (with None) when the operation completes
315 """Finish writing and close the bucket. The share is not finalized
316 until this method is called: if the uploading client disconnects
317 before calling close(), the partially-written share will be
320 @return: a Deferred that fires (with None) when the operation completes
323 class IStorageBucketReader(Interface):
325 def get_block_data(blocknum=int, blocksize=int, size=int):
326 """Most blocks will be the same size. The last block might be shorter
332 def get_crypttext_hashes():
334 @return: ListOf(Hash)
337 def get_block_hashes(at_least_these=SetOf(int)):
339 @return: ListOf(Hash)
342 def get_share_hashes(at_least_these=SetOf(int)):
344 @return: ListOf(TupleOf(int, Hash))
347 def get_uri_extension():
349 @return: URIExtensionData
352 class IStorageBroker(Interface):
353 def get_servers_for_index(peer_selection_index):
355 @return: list of (peerid, versioned-rref) tuples
357 def get_all_servers():
359 @return: frozenset of (peerid, versioned-rref) tuples
361 def get_all_serverids():
363 @return: frozenset of serverid strings
365 def get_nickname_for_serverid(serverid):
367 @return: unicode nickname, or None
370 # methods moved from IntroducerClient, need review
371 def get_all_connections():
372 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
373 each active connection we've established to a remote service. This is
374 mostly useful for unit tests that need to wait until a certain number
375 of connections have been made."""
377 def get_all_connectors():
378 """Return a dict that maps from (nodeid, service_name) to a
379 RemoteServiceConnector instance for all services that we are actively
380 trying to connect to. Each RemoteServiceConnector has the following
383 service_name: the type of service provided, like 'storage'
384 announcement_time: when we first heard about this service
385 last_connect_time: when we last established a connection
386 last_loss_time: when we last lost a connection
388 version: the peer's version, from the most recent connection
389 oldest_supported: the peer's oldest supported version, same
391 rref: the RemoteReference, if connected, otherwise None
392 remote_host: the IAddress, if connected, otherwise None
394 This method is intended for monitoring interfaces, such as a web page
395 which describes connecting and connected peers.
398 def get_all_peerids():
399 """Return a frozenset of all peerids to whom we have a connection (to
400 one or more services) established. Mostly useful for unit tests."""
402 def get_all_connections_for(service_name):
403 """Return a frozenset of (nodeid, service_name, rref) tuples, one
404 for each active connection that provides the given SERVICE_NAME."""
406 def get_permuted_peers(service_name, key):
407 """Returns an ordered list of (peerid, rref) tuples, selecting from
408 the connections that provide SERVICE_NAME, using a hash-based
409 permutation keyed by KEY. This randomizes the service list in a
410 repeatable way, to distribute load over many peers.
414 # hm, we need a solution for forward references in schemas
415 FileNode_ = Any() # TODO: foolscap needs constraints on copyables
416 DirectoryNode_ = Any() # TODO: same
417 AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
420 class IURI(Interface):
421 def init_from_string(uri):
422 """Accept a string (as created by my to_string() method) and populate
423 this instance with its data. I am not normally called directly,
424 please use the module-level uri.from_string() function to convert
425 arbitrary URI strings into IURI-providing instances."""
428 """Return False if this URI be used to modify the data. Return True
429 if this URI cannot be used to modify the data."""
432 """Return True if the data can be modified by *somebody* (perhaps
433 someone who has a more powerful URI than this one)."""
436 """Return another IURI instance, which represents a read-only form of
437 this one. If is_readonly() is True, this returns self."""
439 def get_verify_cap():
440 """Return an instance that provides IVerifierURI, which can be used
441 to check on the availability of the file or directory, without
442 providing enough capabilities to actually read or modify the
443 contents. This may return None if the file does not need checking or
444 verification (e.g. LIT URIs).
448 """Return a string of printable ASCII characters, suitable for
449 passing into init_from_string."""
451 class IVerifierURI(Interface, IURI):
452 def init_from_string(uri):
453 """Accept a string (as created by my to_string() method) and populate
454 this instance with its data. I am not normally called directly,
455 please use the module-level uri.from_string() function to convert
456 arbitrary URI strings into IURI-providing instances."""
459 """Return a string of printable ASCII characters, suitable for
460 passing into init_from_string."""
462 class IDirnodeURI(Interface):
463 """I am a URI which represents a dirnode."""
466 class IFileURI(Interface):
467 """I am a URI which represents a filenode."""
469 """Return the length (in bytes) of the file that I represent."""
471 class IImmutableFileURI(IFileURI):
474 class IMutableFileURI(Interface):
475 """I am a URI which represents a mutable filenode."""
476 class INewDirectoryURI(Interface):
478 class IReadonlyNewDirectoryURI(Interface):
481 class CannotPackUnknownNodeError(Exception):
482 """UnknownNodes (using filecaps from the future that we don't understand)
483 cannot yet be copied safely, so I refuse to copy them."""
485 class UnhandledCapTypeError(Exception):
486 """I recognize the cap/URI, but I cannot create an IFilesystemNode for
489 class IFilesystemNode(Interface):
492 Return the URI that can be used by others to get access to this
493 node. If this node is read-only, the URI will only offer read-only
494 access. If this node is read-write, the URI will offer read-write
497 If you have read-write access to a node and wish to share merely
498 read-only access with others, use get_readonly_uri().
501 def get_readonly_uri():
502 """Return the directory URI that can be used by others to get
503 read-only access to this directory node. The result is a read-only
504 URI, regardless of whether this dirnode is read-only or read-write.
506 If you have merely read-only access to this dirnode,
507 get_readonly_uri() will return the same thing as get_uri().
510 def get_repair_cap():
511 """Return an IURI instance that can be used to repair the file, or
512 None if this node cannot be repaired (either because it is not
513 distributed, like a LIT file, or because the node does not represent
514 sufficient authority to create a repair-cap, like a read-only RSA
515 mutable file node [which cannot create the correct write-enablers]).
518 def get_verify_cap():
519 """Return an IVerifierURI instance that represents the
520 'verifiy/refresh capability' for this node. The holder of this
521 capability will be able to renew the lease for this node, protecting
522 it from garbage-collection. They will also be able to ask a server if
523 it holds a share for the file or directory.
526 def get_storage_index():
527 """Return a string with the (binary) storage index in use on this
528 download. This may be None if there is no storage index (i.e. LIT
532 """Return True if this reference provides mutable access to the given
533 file or directory (i.e. if you can modify it), or False if not. Note
534 that even if this reference is read-only, someone else may hold a
535 read-write reference to it."""
538 """Return True if this file or directory is mutable (by *somebody*,
539 not necessarily you), False if it is is immutable. Note that a file
540 might be mutable overall, but your reference to it might be
541 read-only. On the other hand, all references to an immutable file
542 will be read-only; there are no read-write references to an immutable
546 class IMutableFilesystemNode(IFilesystemNode):
549 class IFileNode(IFilesystemNode):
550 def download(target):
551 """Download the file's contents to a given IDownloadTarget"""
553 def download_to_data():
554 """Download the file's contents. Return a Deferred that fires
555 with those contents."""
558 """Return the length (in bytes) of the data this node represents."""
560 def read(consumer, offset=0, size=None):
561 """Download a portion (possibly all) of the file's contents, making
562 them available to the given IConsumer. Return a Deferred that fires
563 (with the consumer) when the consumer is unregistered (either because
564 the last byte has been given to it, or because the consumer threw an
565 exception during write(), possibly because it no longer wants to
566 receive data). The portion downloaded will start at 'offset' and
567 contain 'size' bytes (or the remainder of the file if size==None).
569 The consumer will be used in non-streaming mode: an IPullProducer
570 will be attached to it.
572 The consumer will not receive data right away: several network trips
573 must occur first. The order of events will be::
575 consumer.registerProducer(p, streaming)
576 (if streaming == False)::
577 consumer does p.resumeProducing()
579 consumer does p.resumeProducing()
580 consumer.write(data).. (repeat until all data is written)
581 consumer.unregisterProducer()
582 deferred.callback(consumer)
584 If a download error occurs, or an exception is raised by
585 consumer.registerProducer() or consumer.write(), I will call
586 consumer.unregisterProducer() and then deliver the exception via
587 deferred.errback(). To cancel the download, the consumer should call
588 p.stopProducing(), which will result in an exception being delivered
589 via deferred.errback().
591 A simple download-to-memory consumer example would look like this::
593 class MemoryConsumer:
594 implements(IConsumer)
598 def registerProducer(self, p, streaming):
599 assert streaming == False
602 def write(self, data):
603 self.chunks.append(data)
604 def unregisterProducer(self):
606 d = filenode.read(MemoryConsumer())
607 d.addCallback(lambda mc: "".join(mc.chunks))
612 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
613 """I provide access to a 'mutable file', which retains its identity
614 regardless of what contents are put in it.
616 The consistency-vs-availability problem means that there might be
617 multiple versions of a file present in the grid, some of which might be
618 unrecoverable (i.e. have fewer than 'k' shares). These versions are
619 loosely ordered: each has a sequence number and a hash, and any version
620 with seqnum=N was uploaded by a node which has seen at least one version
623 The 'servermap' (an instance of IMutableFileServerMap) is used to
624 describe the versions that are known to be present in the grid, and which
625 servers are hosting their shares. It is used to represent the 'state of
626 the world', and is used for this purpose by my test-and-set operations.
627 Downloading the contents of the mutable file will also return a
628 servermap. Uploading a new version into the mutable file requires a
629 servermap as input, and the semantics of the replace operation is
630 'replace the file with my new version if it looks like nobody else has
631 changed the file since my previous download'. Because the file is
632 distributed, this is not a perfect test-and-set operation, but it will do
633 its best. If the replace process sees evidence of a simultaneous write,
634 it will signal an UncoordinatedWriteError, so that the caller can take
638 Most readers will want to use the 'best' current version of the file, and
639 should use my 'download_best_version()' method.
641 To unconditionally replace the file, callers should use overwrite(). This
642 is the mode that user-visible mutable files will probably use.
644 To apply some delta to the file, call modify() with a callable modifier
645 function that can apply the modification that you want to make. This is
646 the mode that dirnodes will use, since most directory modification
647 operations can be expressed in terms of deltas to the directory state.
650 Three methods are available for users who need to perform more complex
651 operations. The first is get_servermap(), which returns an up-to-date
652 servermap using a specified mode. The second is download_version(), which
653 downloads a specific version (not necessarily the 'best' one). The third
654 is 'upload', which accepts new contents and a servermap (which must have
655 been updated with MODE_WRITE). The upload method will attempt to apply
656 the new contents as long as no other node has modified the file since the
657 servermap was updated. This might be useful to a caller who wants to
658 merge multiple versions into a single new one.
660 Note that each time the servermap is updated, a specific 'mode' is used,
661 which determines how many peers are queried. To use a servermap for my
662 replace() method, that servermap must have been updated in MODE_WRITE.
663 These modes are defined in allmydata.mutable.common, and consist of
664 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
665 allmydata/mutable/servermap.py for details about the differences.
667 Mutable files are currently limited in size (about 3.5MB max) and can
668 only be retrieved and updated all-at-once, as a single big string. Future
669 versions of our mutable files will remove this restriction.
672 def download_best_version():
673 """Download the 'best' available version of the file, meaning one of
674 the recoverable versions with the highest sequence number. If no
675 uncoordinated writes have occurred, and if enough shares are
676 available, then this will be the most recent version that has been
679 I update an internal servermap with MODE_READ, determine which
680 version of the file is indicated by
681 servermap.best_recoverable_version(), and return a Deferred that
682 fires with its contents. If no version is recoverable, the Deferred
683 will errback with UnrecoverableFileError.
686 def get_size_of_best_version():
687 """Find the size of the version that would be downloaded with
688 download_best_version(), without actually downloading the whole file.
690 I return a Deferred that fires with an integer.
693 def overwrite(new_contents):
694 """Unconditionally replace the contents of the mutable file with new
695 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
696 is only appropriate to use when the new contents of the file are
697 completely unrelated to the old ones, and you do not care about other
700 I return a Deferred that fires (with a PublishStatus object) when the
701 update has completed.
704 def modify(modifier_cb):
705 """Modify the contents of the file, by downloading the current
706 version, applying the modifier function (or bound method), then
707 uploading the new version. I return a Deferred that fires (with a
708 PublishStatus object) when the update is complete.
710 The modifier callable will be given three arguments: a string (with
711 the old contents), a 'first_time' boolean, and a servermap. As with
712 download_best_version(), the old contents will be from the best
713 recoverable version, but the modifier can use the servermap to make
714 other decisions (such as refusing to apply the delta if there are
715 multiple parallel versions, or if there is evidence of a newer
716 unrecoverable version). 'first_time' will be True the first time the
717 modifier is called, and False on any subsequent calls.
719 The callable should return a string with the new contents. The
720 callable must be prepared to be called multiple times, and must
721 examine the input string to see if the change that it wants to make
722 is already present in the old version. If it does not need to make
723 any changes, it can either return None, or return its input string.
725 If the modifier raises an exception, it will be returned in the
730 def get_servermap(mode):
731 """Return a Deferred that fires with an IMutableFileServerMap
732 instance, updated using the given mode.
735 def download_version(servermap, version):
736 """Download a specific version of the file, using the servermap
737 as a guide to where the shares are located.
739 I return a Deferred that fires with the requested contents, or
740 errbacks with UnrecoverableFileError. Note that a servermap which was
741 updated with MODE_ANYTHING or MODE_READ may not know about shares for
742 all versions (those modes stop querying servers as soon as they can
743 fulfil their goals), so you may want to use MODE_CHECK (which checks
744 everything) to get increased visibility.
747 def upload(new_contents, servermap):
748 """Replace the contents of the file with new ones. This requires a
749 servermap that was previously updated with MODE_WRITE.
751 I attempt to provide test-and-set semantics, in that I will avoid
752 modifying any share that is different than the version I saw in the
753 servermap. However, if another node is writing to the file at the
754 same time as me, I may manage to update some shares while they update
755 others. If I see any evidence of this, I will signal
756 UncoordinatedWriteError, and the file will be left in an inconsistent
757 state (possibly the version you provided, possibly the old version,
758 possibly somebody else's version, and possibly a mix of shares from
761 The recommended response to UncoordinatedWriteError is to either
762 return it to the caller (since they failed to coordinate their
763 writes), or to attempt some sort of recovery. It may be sufficient to
764 wait a random interval (with exponential backoff) and repeat your
765 operation. If I do not signal UncoordinatedWriteError, then I was
766 able to write the new version without incident.
768 I return a Deferred that fires (with a PublishStatus object) when the
769 publish has completed. I will update the servermap in-place with the
770 location of all new shares.
774 """Return this filenode's writekey, or None if the node does not have
775 write-capability. This may be used to assist with data structures
776 that need to make certain data available only to writers, such as the
777 read-write child caps in dirnodes. The recommended process is to have
778 reader-visible data be submitted to the filenode in the clear (where
779 it will be encrypted by the filenode using the readkey), but encrypt
780 writer-visible data using this writekey.
783 class NotEnoughSharesError(Exception):
784 """Download was unable to get enough shares, or upload was unable to
785 place 'shares_of_happiness' shares."""
787 class NoSharesError(Exception):
788 """Upload or Download was unable to get any shares at all."""
790 class UnableToFetchCriticalDownloadDataError(Exception):
791 """I was unable to fetch some piece of critical data which is supposed to
792 be identically present in all shares."""
794 class NoServersError(Exception):
795 """Upload wasn't given any servers to work with, usually indicating a
796 network or Introducer problem."""
798 class ExistingChildError(Exception):
799 """A directory node was asked to add or replace a child that already
800 exists, and overwrite= was set to False."""
802 class NoSuchChildError(Exception):
803 """A directory node was asked to fetch a child which does not exist."""
805 class IDirectoryNode(IMutableFilesystemNode):
806 """I represent a name-to-child mapping, holding the tahoe equivalent of a
807 directory. All child names are unicode strings, and all children are some
808 sort of IFilesystemNode (either files or subdirectories).
813 The dirnode ('1') URI returned by this method can be used in
814 set_uri() on a different directory ('2') to 'mount' a reference to
815 this directory ('1') under the other ('2'). This URI is just a
816 string, so it can be passed around through email or other out-of-band
820 def get_readonly_uri():
822 The dirnode ('1') URI returned by this method can be used in
823 set_uri() on a different directory ('2') to 'mount' a reference to
824 this directory ('1') under the other ('2'). This URI is just a
825 string, so it can be passed around through email or other out-of-band
830 """I return a Deferred that fires with a dictionary mapping child
831 name (a unicode string) to (node, metadata_dict) tuples, in which
832 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
833 is a dictionary of metadata."""
836 """I return a Deferred that fires with a boolean, True if there
837 exists a child of the given name, False if not. The child name must
838 be a unicode string."""
841 """I return a Deferred that fires with a specific named child node,
842 either an IFileNode or an IDirectoryNode. The child name must be a
843 unicode string. I raise NoSuchChildError if I do not have a child by
846 def get_metadata_for(name):
847 """I return a Deferred that fires with the metadata dictionary for a
848 specific named child node. This metadata is stored in the *edge*, not
849 in the child, so it is attached to the parent dirnode rather than the
850 child dir-or-file-node. The child name must be a unicode string. I
851 raise NoSuchChildError if I do not have a child by that name."""
853 def set_metadata_for(name, metadata):
854 """I replace any existing metadata for the named child with the new
855 metadata. The child name must be a unicode string. This metadata is
856 stored in the *edge*, not in the child, so it is attached to the
857 parent dirnode rather than the child dir-or-file-node. I return a
858 Deferred (that fires with this dirnode) when the operation is
859 complete. I raise NoSuchChildError if I do not have a child by that
862 def get_child_at_path(path):
863 """Transform a child path into an IDirectoryNode or IFileNode.
865 I perform a recursive series of 'get' operations to find the named
866 descendant node. I return a Deferred that fires with the node, or
867 errbacks with NoSuchChildError if the node could not be found.
869 The path can be either a single string (slash-separated) or a list of
870 path-name elements. All elements must be unicode strings.
873 def get_child_and_metadata_at_path(path):
874 """Transform a child path into an IDirectoryNode/IFileNode and
877 I am like get_child_at_path(), but my Deferred fires with a tuple of
878 (node, metadata). The metadata comes from the last edge. If the path
879 is empty, the metadata will be an empty dictionary.
882 def set_uri(name, child_uri, metadata=None, overwrite=True):
883 """I add a child (by URI) at the specific name. I return a Deferred
884 that fires when the operation finishes. If overwrite= is True, I will
885 replace any existing child of the same name, otherwise an existing
886 child will cause me to return ExistingChildError. The child name must
889 The child_uri could be for a file, or for a directory (either
890 read-write or read-only, using a URI that came from get_uri() ).
892 If metadata= is provided, I will use it as the metadata for the named
893 edge. This will replace any existing metadata. If metadata= is left
894 as the default value of None, I will set ['mtime'] to the current
895 time, and I will set ['ctime'] to the current time if there was not
896 already a child by this name present. This roughly matches the
897 ctime/mtime semantics of traditional filesystems.
899 If this directory node is read-only, the Deferred will errback with a
902 def set_children(entries, overwrite=True):
903 """Add multiple (name, child_uri) pairs (or (name, child_uri,
904 metadata) triples) to a directory node. Returns a Deferred that fires
905 (with None) when the operation finishes. This is equivalent to
906 calling set_uri() multiple times, but is much more efficient. All
907 child names must be unicode strings.
910 def set_node(name, child, metadata=None, overwrite=True):
911 """I add a child at the specific name. I return a Deferred that fires
912 when the operation finishes. This Deferred will fire with the child
913 node that was just added. I will replace any existing child of the
914 same name. The child name must be a unicode string. The 'child'
915 instance must be an instance providing IDirectoryNode or IFileNode.
917 If metadata= is provided, I will use it as the metadata for the named
918 edge. This will replace any existing metadata. If metadata= is left
919 as the default value of None, I will set ['mtime'] to the current
920 time, and I will set ['ctime'] to the current time if there was not
921 already a child by this name present. This roughly matches the
922 ctime/mtime semantics of traditional filesystems.
924 If this directory node is read-only, the Deferred will errback with a
927 def set_nodes(entries, overwrite=True):
928 """Add multiple (name, child_node) pairs (or (name, child_node,
929 metadata) triples) to a directory node. Returns a Deferred that fires
930 (with None) when the operation finishes. This is equivalent to
931 calling set_node() multiple times, but is much more efficient. All
932 child names must be unicode strings."""
935 def add_file(name, uploadable, metadata=None, overwrite=True):
936 """I upload a file (using the given IUploadable), then attach the
937 resulting FileNode to the directory at the given name. I set metadata
938 the same way as set_uri and set_node. The child name must be a
941 I return a Deferred that fires (with the IFileNode of the uploaded
942 file) when the operation completes."""
945 """I remove the child at the specific name. I return a Deferred that
946 fires when the operation finishes. The child name must be a unicode
947 string. I raise NoSuchChildError if I do not have a child by that
950 def create_empty_directory(name, overwrite=True):
951 """I create and attach an empty directory at the given name. The
952 child name must be a unicode string. I return a Deferred that fires
953 when the operation finishes."""
955 def move_child_to(current_child_name, new_parent, new_child_name=None,
957 """I take one of my children and move them to a new parent. The child
958 is referenced by name. On the new parent, the child will live under
959 'new_child_name', which defaults to 'current_child_name'. TODO: what
960 should we do about metadata? I return a Deferred that fires when the
961 operation finishes. The child name must be a unicode string. I raise
962 NoSuchChildError if I do not have a child by that name."""
964 def build_manifest():
965 """I generate a table of everything reachable from this directory.
966 I also compute deep-stats as described below.
968 I return a Monitor. The Monitor's results will be a dictionary with
971 res['manifest']: a list of (path, cap) tuples for all nodes
972 (directories and files) reachable from this one.
973 'path' will be a tuple of unicode strings. The
974 origin dirnode will be represented by an empty path
976 res['verifycaps']: a list of (printable) verifycap strings, one for
977 each reachable non-LIT node. This is a set:
978 it will contain no duplicates.
979 res['storage-index']: a list of (base32) storage index strings,
980 one for each reachable non-LIT node. This is
981 a set: it will contain no duplicates.
982 res['stats']: a dictionary, the same that is generated by
983 start_deep_stats() below.
985 The Monitor will also have an .origin_si attribute with the (binary)
986 storage index of the starting point.
989 def start_deep_stats():
990 """Return a Monitor, examining all nodes (directories and files)
991 reachable from this one. The Monitor's results will be a dictionary
992 with the following keys::
994 count-immutable-files: count of how many CHK files are in the set
995 count-mutable-files: same, for mutable files (does not include
997 count-literal-files: same, for LIT files
998 count-files: sum of the above three
1000 count-directories: count of directories
1002 size-immutable-files: total bytes for all CHK files in the set
1003 size-mutable-files (TODO): same, for current version of all mutable
1004 files, does not include directories
1005 size-literal-files: same, for LIT files
1006 size-directories: size of mutable files used by directories
1008 largest-directory: number of bytes in the largest directory
1009 largest-directory-children: number of children in the largest
1011 largest-immutable-file: number of bytes in the largest CHK file
1013 size-mutable-files is not yet implemented, because it would involve
1014 even more queries than deep_stats does.
1016 The Monitor will also have an .origin_si attribute with the (binary)
1017 storage index of the starting point.
1019 This operation will visit every directory node underneath this one,
1020 and can take a long time to run. On a typical workstation with good
1021 bandwidth, this can examine roughly 15 directories per second (and
1022 takes several minutes of 100% CPU for ~1700 directories).
1025 class ICodecEncoder(Interface):
1026 def set_params(data_size, required_shares, max_shares):
1027 """Set up the parameters of this encoder.
1029 This prepares the encoder to perform an operation that converts a
1030 single block of data into a number of shares, such that a future
1031 ICodecDecoder can use a subset of these shares to recover the
1032 original data. This operation is invoked by calling encode(). Once
1033 the encoding parameters are set up, the encode operation can be
1034 invoked multiple times.
1036 set_params() prepares the encoder to accept blocks of input data that
1037 are exactly 'data_size' bytes in length. The encoder will be prepared
1038 to produce 'max_shares' shares for each encode() operation (although
1039 see the 'desired_share_ids' to use less CPU). The encoding math will
1040 be chosen such that the decoder can get by with as few as
1041 'required_shares' of these shares and still reproduce the original
1042 data. For example, set_params(1000, 5, 5) offers no redundancy at
1043 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1045 Numerical Restrictions: 'data_size' is required to be an integral
1046 multiple of 'required_shares'. In general, the caller should choose
1047 required_shares and max_shares based upon their reliability
1048 requirements and the number of peers available (the total storage
1049 space used is roughly equal to max_shares*data_size/required_shares),
1050 then choose data_size to achieve the memory footprint desired (larger
1051 data_size means more efficient operation, smaller data_size means
1052 smaller memory footprint).
1054 In addition, 'max_shares' must be equal to or greater than
1055 'required_shares'. Of course, setting them to be equal causes
1056 encode() to degenerate into a particularly slow form of the 'split'
1059 See encode() for more details about how these parameters are used.
1061 set_params() must be called before any other ICodecEncoder methods
1066 """Return the 3-tuple of data_size, required_shares, max_shares"""
1068 def get_encoder_type():
1069 """Return a short string that describes the type of this encoder.
1071 There is required to be a global table of encoder classes. This method
1072 returns an index into this table; the value at this index is an
1073 encoder class, and this encoder is an instance of that class.
1076 def get_block_size():
1077 """Return the length of the shares that encode() will produce.
1080 def encode_proposal(data, desired_share_ids=None):
1081 """Encode some data.
1083 'data' must be a string (or other buffer object), and len(data) must
1084 be equal to the 'data_size' value passed earlier to set_params().
1086 This will return a Deferred that will fire with two lists. The first
1087 is a list of shares, each of which is a string (or other buffer
1088 object) such that len(share) is the same as what get_share_size()
1089 returned earlier. The second is a list of shareids, in which each is
1090 an integer. The lengths of the two lists will always be equal to each
1091 other. The user should take care to keep each share closely
1092 associated with its shareid, as one is useless without the other.
1094 The length of this output list will normally be the same as the value
1095 provided to the 'max_shares' parameter of set_params(). This may be
1096 different if 'desired_share_ids' is provided.
1098 'desired_share_ids', if provided, is required to be a sequence of
1099 ints, each of which is required to be >= 0 and < max_shares. If not
1100 provided, encode() will produce 'max_shares' shares, as if
1101 'desired_share_ids' were set to range(max_shares). You might use this
1102 if you initially thought you were going to use 10 peers, started
1103 encoding, and then two of the peers dropped out: you could use
1104 desired_share_ids= to skip the work (both memory and CPU) of
1105 producing shares for the peers which are no longer available.
1109 def encode(inshares, desired_share_ids=None):
1110 """Encode some data. This may be called multiple times. Each call is
1113 inshares is a sequence of length required_shares, containing buffers
1114 (i.e. strings), where each buffer contains the next contiguous
1115 non-overlapping segment of the input data. Each buffer is required to
1116 be the same length, and the sum of the lengths of the buffers is
1117 required to be exactly the data_size promised by set_params(). (This
1118 implies that the data has to be padded before being passed to
1119 encode(), unless of course it already happens to be an even multiple
1120 of required_shares in length.)
1122 ALSO: the requirement to break up your data into 'required_shares'
1123 chunks before calling encode() feels a bit surprising, at least from
1124 the point of view of a user who doesn't know how FEC works. It feels
1125 like an implementation detail that has leaked outside the
1126 abstraction barrier. Can you imagine a use case in which the data to
1127 be encoded might already be available in pre-segmented chunks, such
1128 that it is faster or less work to make encode() take a list rather
1129 than splitting a single string?
1131 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1132 is supposed to *produce* shares, so what it *accepts* should be
1133 something other than shares. Other places in this interface use the
1134 word 'data' for that-which-is-not-shares.. maybe we should use that
1137 'desired_share_ids', if provided, is required to be a sequence of
1138 ints, each of which is required to be >= 0 and < max_shares. If not
1139 provided, encode() will produce 'max_shares' shares, as if
1140 'desired_share_ids' were set to range(max_shares). You might use this
1141 if you initially thought you were going to use 10 peers, started
1142 encoding, and then two of the peers dropped out: you could use
1143 desired_share_ids= to skip the work (both memory and CPU) of
1144 producing shares for the peers which are no longer available.
1146 For each call, encode() will return a Deferred that fires with two
1147 lists, one containing shares and the other containing the shareids.
1148 The get_share_size() method can be used to determine the length of
1149 the share strings returned by encode(). Each shareid is a small
1150 integer, exactly as passed into 'desired_share_ids' (or
1151 range(max_shares), if desired_share_ids was not provided).
1153 The shares and their corresponding shareids are required to be kept
1154 together during storage and retrieval. Specifically, the share data is
1155 useless by itself: the decoder needs to be told which share is which
1156 by providing it with both the shareid and the actual share data.
1158 This function will allocate an amount of memory roughly equal to::
1160 (max_shares - required_shares) * get_share_size()
1162 When combined with the memory that the caller must allocate to
1163 provide the input data, this leads to a memory footprint roughly
1164 equal to the size of the resulting encoded shares (i.e. the expansion
1165 factor times the size of the input segment).
1170 # returning a list of (shareidN,shareN) tuples instead of a pair of
1171 # lists (shareids..,shares..). Brian thought the tuples would
1172 # encourage users to keep the share and shareid together throughout
1173 # later processing, Zooko pointed out that the code to iterate
1174 # through two lists is not really more complicated than using a list
1175 # of tuples and there's also a performance improvement
1177 # having 'data_size' not required to be an integral multiple of
1178 # 'required_shares'. Doing this would require encode() to perform
1179 # padding internally, and we'd prefer to have any padding be done
1180 # explicitly by the caller. Yes, it is an abstraction leak, but
1181 # hopefully not an onerous one.
1184 class ICodecDecoder(Interface):
1185 def set_params(data_size, required_shares, max_shares):
1186 """Set the params. They have to be exactly the same ones that were
1187 used for encoding."""
1189 def get_needed_shares():
1190 """Return the number of shares needed to reconstruct the data.
1191 set_params() is required to be called before this."""
1193 def decode(some_shares, their_shareids):
1194 """Decode a partial list of shares into data.
1196 'some_shares' is required to be a sequence of buffers of sharedata, a
1197 subset of the shares returned by ICodecEncode.encode(). Each share is
1198 required to be of the same length. The i'th element of their_shareids
1199 is required to be the shareid of the i'th buffer in some_shares.
1201 This returns a Deferred which fires with a sequence of buffers. This
1202 sequence will contain all of the segments of the original data, in
1203 order. The sum of the lengths of all of the buffers will be the
1204 'data_size' value passed into the original ICodecEncode.set_params()
1205 call. To get back the single original input block of data, use
1206 ''.join(output_buffers), or you may wish to simply write them in
1207 order to an output file.
1209 Note that some of the elements in the result sequence may be
1210 references to the elements of the some_shares input sequence. In
1211 particular, this means that if those share objects are mutable (e.g.
1212 arrays) and if they are changed, then both the input (the
1213 'some_shares' parameter) and the output (the value given when the
1214 deferred is triggered) will change.
1216 The length of 'some_shares' is required to be exactly the value of
1217 'required_shares' passed into the original ICodecEncode.set_params()
1221 class IEncoder(Interface):
1222 """I take an object that provides IEncryptedUploadable, which provides
1223 encrypted data, and a list of shareholders. I then encode, hash, and
1224 deliver shares to those shareholders. I will compute all the necessary
1225 Merkle hash trees that are necessary to validate the crypttext that
1226 eventually comes back from the shareholders. I provide the URI Extension
1227 Block Hash, and the encoding parameters, both of which must be included
1230 I do not choose shareholders, that is left to the IUploader. I must be
1231 given a dict of RemoteReferences to storage buckets that are ready and
1232 willing to receive data.
1236 """Specify the number of bytes that will be encoded. This must be
1237 peformed before get_serialized_params() can be called.
1239 def set_params(params):
1240 """Override the default encoding parameters. 'params' is a tuple of
1241 (k,d,n), where 'k' is the number of required shares, 'd' is the
1242 shares_of_happiness, and 'n' is the total number of shares that will
1245 Encoding parameters can be set in three ways. 1: The Encoder class
1246 provides defaults (3/7/10). 2: the Encoder can be constructed with
1247 an 'options' dictionary, in which the
1248 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1249 set_params((k,d,n)) can be called.
1251 If you intend to use set_params(), you must call it before
1252 get_share_size or get_param are called.
1255 def set_encrypted_uploadable(u):
1256 """Provide a source of encrypted upload data. 'u' must implement
1257 IEncryptedUploadable.
1259 When this is called, the IEncryptedUploadable will be queried for its
1260 length and the storage_index that should be used.
1262 This returns a Deferred that fires with this Encoder instance.
1264 This must be performed before start() can be called.
1267 def get_param(name):
1268 """Return an encoding parameter, by name.
1270 'storage_index': return a string with the (16-byte truncated SHA-256
1271 hash) storage index to which these shares should be
1274 'share_counts': return a tuple describing how many shares are used:
1275 (needed_shares, shares_of_happiness, total_shares)
1277 'num_segments': return an int with the number of segments that
1280 'segment_size': return an int with the size of each segment.
1282 'block_size': return the size of the individual blocks that will
1283 be delivered to a shareholder's put_block() method. By
1284 knowing this, the shareholder will be able to keep all
1285 blocks in a single file and still provide random access
1286 when reading them. # TODO: can we avoid exposing this?
1288 'share_size': an int with the size of the data that will be stored
1289 on each shareholder. This is aggregate amount of data
1290 that will be sent to the shareholder, summed over all
1291 the put_block() calls I will ever make. It is useful to
1292 determine this size before asking potential
1293 shareholders whether they will grant a lease or not,
1294 since their answers will depend upon how much space we
1295 need. TODO: this might also include some amount of
1296 overhead, like the size of all the hashes. We need to
1297 decide whether this is useful or not.
1299 'serialized_params': a string with a concise description of the
1300 codec name and its parameters. This may be passed
1301 into the IUploadable to let it make sure that
1302 the same file encoded with different parameters
1303 will result in different storage indexes.
1305 Once this is called, set_size() and set_params() may not be called.
1308 def set_shareholders(shareholders):
1309 """Tell the encoder where to put the encoded shares. 'shareholders'
1310 must be a dictionary that maps share number (an integer ranging from
1311 0 to n-1) to an instance that provides IStorageBucketWriter. This
1312 must be performed before start() can be called."""
1315 """Begin the encode/upload process. This involves reading encrypted
1316 data from the IEncryptedUploadable, encoding it, uploading the shares
1317 to the shareholders, then sending the hash trees.
1319 set_encrypted_uploadable() and set_shareholders() must be called
1320 before this can be invoked.
1322 This returns a Deferred that fires with a verify cap when the upload
1323 process is complete. The verifycap, plus the encryption key, is
1324 sufficient to construct the read cap.
1327 class IDecoder(Interface):
1328 """I take a list of shareholders and some setup information, then
1329 download, validate, decode, and decrypt data from them, writing the
1330 results to an output file.
1332 I do not locate the shareholders, that is left to the IDownloader. I must
1333 be given a dict of RemoteReferences to storage buckets that are ready to
1338 """I take a file-like object (providing write and close) to which all
1339 the plaintext data will be written.
1341 TODO: producer/consumer . Maybe write() should return a Deferred that
1342 indicates when it will accept more data? But probably having the
1343 IDecoder be a producer is easier to glue to IConsumer pieces.
1346 def set_shareholders(shareholders):
1347 """I take a dictionary that maps share identifiers (small integers)
1348 to RemoteReferences that provide RIBucketReader. This must be called
1352 """I start the download. This process involves retrieving data and
1353 hash chains from the shareholders, using the hashes to validate the
1354 data, decoding the shares into segments, decrypting the segments,
1355 then writing the resulting plaintext to the output file.
1357 I return a Deferred that will fire (with self) when the download is
1361 class IDownloadTarget(Interface):
1362 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1363 # will register itself as a producer. This allows the target to invoke
1364 # downloader.pauseProducing, resumeProducing, and stopProducing.
1366 """Called before any calls to write() or close(). If an error
1367 occurs before any data is available, fail() may be called without
1368 a previous call to open().
1370 'size' is the length of the file being downloaded, in bytes."""
1373 """Output some data to the target."""
1375 """Inform the target that there is no more data to be written."""
1377 """fail() is called to indicate that the download has failed. 'why'
1378 is a Failure object indicating what went wrong. No further methods
1379 will be invoked on the IDownloadTarget after fail()."""
1380 def register_canceller(cb):
1381 """The CiphertextDownloader uses this to register a no-argument function
1382 that the target can call to cancel the download. Once this canceller
1383 is invoked, no further calls to write() or close() will be made."""
1385 """When the CiphertextDownloader is done, this finish() function will be
1386 called. Whatever it returns will be returned to the invoker of
1387 Downloader.download.
1389 # The following methods are just because that target might be a
1390 # repairer.DownUpConnector, and just because the current CHKUpload object
1391 # expects to find the storage index and encoding parameters in its
1393 def set_storageindex(storageindex):
1394 """ Set the storage index. """
1395 def set_encodingparams(encodingparams):
1396 """ Set the encoding parameters. """
1398 class IDownloader(Interface):
1399 def download(uri, target):
1400 """Perform a CHK download, sending the data to the given target.
1401 'target' must provide IDownloadTarget.
1403 Returns a Deferred that fires (with the results of target.finish)
1404 when the download is finished, or errbacks if something went wrong."""
1406 class IEncryptedUploadable(Interface):
1407 def set_upload_status(upload_status):
1408 """Provide an IUploadStatus object that should be filled with status
1409 information. The IEncryptedUploadable is responsible for setting
1410 key-determination progress ('chk'), size, storage_index, and
1411 ciphertext-fetch progress. It may delegate some of this
1412 responsibility to others, in particular to the IUploadable."""
1415 """This behaves just like IUploadable.get_size()."""
1417 def get_all_encoding_parameters():
1418 """Return a Deferred that fires with a tuple of
1419 (k,happy,n,segment_size). The segment_size will be used as-is, and
1420 must match the following constraints: it must be a multiple of k, and
1421 it shouldn't be unreasonably larger than the file size (if
1422 segment_size is larger than filesize, the difference must be stored
1425 This usually passes through to the IUploadable method of the same
1428 The encoder strictly obeys the values returned by this method. To
1429 make an upload use non-default encoding parameters, you must arrange
1430 to control the values that this method returns.
1433 def get_storage_index():
1434 """Return a Deferred that fires with a 16-byte storage index.
1437 def read_encrypted(length, hash_only):
1438 """This behaves just like IUploadable.read(), but returns crypttext
1439 instead of plaintext. If hash_only is True, then this discards the
1440 data (and returns an empty list); this improves efficiency when
1441 resuming an interrupted upload (where we need to compute the
1442 plaintext hashes, but don't need the redundant encrypted data)."""
1444 def get_plaintext_hashtree_leaves(first, last, num_segments):
1445 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1446 plaintext segments, i.e. get the tagged hashes of the given segments.
1447 The segment size is expected to be generated by the
1448 IEncryptedUploadable before any plaintext is read or ciphertext
1449 produced, so that the segment hashes can be generated with only a
1452 This returns a Deferred which fires with a sequence of hashes, using:
1454 tuple(segment_hashes[first:last])
1456 'num_segments' is used to assert that the number of segments that the
1457 IEncryptedUploadable handled matches the number of segments that the
1458 encoder was expecting.
1460 This method must not be called until the final byte has been read
1461 from read_encrypted(). Once this method is called, read_encrypted()
1462 can never be called again.
1465 def get_plaintext_hash():
1466 """OBSOLETE; Get the hash of the whole plaintext.
1468 This returns a Deferred which fires with a tagged SHA-256 hash of the
1469 whole plaintext, obtained from hashutil.plaintext_hash(data).
1473 """Just like IUploadable.close()."""
1475 class IUploadable(Interface):
1476 def set_upload_status(upload_status):
1477 """Provide an IUploadStatus object that should be filled with status
1478 information. The IUploadable is responsible for setting
1479 key-determination progress ('chk')."""
1481 def set_default_encoding_parameters(params):
1482 """Set the default encoding parameters, which must be a dict mapping
1483 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1484 'max_segment_size'. These might have an influence on the final
1485 encoding parameters returned by get_all_encoding_parameters(), if the
1486 Uploadable doesn't have more specific preferences.
1488 This call is optional: if it is not used, the Uploadable will use
1489 some built-in defaults. If used, this method must be called before
1490 any other IUploadable methods to have any effect.
1494 """Return a Deferred that will fire with the length of the data to be
1495 uploaded, in bytes. This will be called before the data is actually
1496 used, to compute encoding parameters.
1499 def get_all_encoding_parameters():
1500 """Return a Deferred that fires with a tuple of
1501 (k,happy,n,segment_size). The segment_size will be used as-is, and
1502 must match the following constraints: it must be a multiple of k, and
1503 it shouldn't be unreasonably larger than the file size (if
1504 segment_size is larger than filesize, the difference must be stored
1507 The relative values of k and n allow some IUploadables to request
1508 better redundancy than others (in exchange for consuming more space
1511 Larger values of segment_size reduce hash overhead, while smaller
1512 values reduce memory footprint and cause data to be delivered in
1513 smaller pieces (which may provide a smoother and more predictable
1514 download experience).
1516 The encoder strictly obeys the values returned by this method. To
1517 make an upload use non-default encoding parameters, you must arrange
1518 to control the values that this method returns. One way to influence
1519 them may be to call set_encoding_parameters() before calling
1520 get_all_encoding_parameters().
1523 def get_encryption_key():
1524 """Return a Deferred that fires with a 16-byte AES key. This key will
1525 be used to encrypt the data. The key will also be hashed to derive
1528 Uploadables which want to achieve convergence should hash their file
1529 contents and the serialized_encoding_parameters to form the key
1530 (which of course requires a full pass over the data). Uploadables can
1531 use the upload.ConvergentUploadMixin class to achieve this
1534 Uploadables which do not care about convergence (or do not wish to
1535 make multiple passes over the data) can simply return a
1536 strongly-random 16 byte string.
1538 get_encryption_key() may be called multiple times: the IUploadable is
1539 required to return the same value each time.
1543 """Return a Deferred that fires with a list of strings (perhaps with
1544 only a single element) which, when concatenated together, contain the
1545 next 'length' bytes of data. If EOF is near, this may provide fewer
1546 than 'length' bytes. The total number of bytes provided by read()
1547 before it signals EOF must equal the size provided by get_size().
1549 If the data must be acquired through multiple internal read
1550 operations, returning a list instead of a single string may help to
1551 reduce string copies.
1553 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1554 so a 10kB file means length=3kB, 100kB file means length=30kB,
1555 and >=1MB file means length=300kB.
1557 This method provides for a single full pass through the data. Later
1558 use cases may desire multiple passes or access to only parts of the
1559 data (such as a mutable file making small edits-in-place). This API
1560 will be expanded once those use cases are better understood.
1564 """The upload is finished, and whatever filehandle was in use may be
1567 class IUploadResults(Interface):
1568 """I am returned by upload() methods. I contain a number of public
1569 attributes which can be read to determine the results of the upload. Some
1570 of these are functional, some are timing information. All of these may be
1573 .file_size : the size of the file, in bytes
1574 .uri : the CHK read-cap for the file
1575 .ciphertext_fetched : how many bytes were fetched by the helper
1576 .sharemap: dict mapping share identifier to set of serverids
1577 (binary strings). This indicates which servers were given
1578 which shares. For immutable files, the shareid is an
1579 integer (the share number, from 0 to N-1). For mutable
1580 files, it is a string of the form 'seq%d-%s-sh%d',
1581 containing the sequence number, the roothash, and the
1583 .servermap : dict mapping server peerid to a set of share numbers
1584 .timings : dict of timing information, mapping name to seconds (float)
1585 total : total upload time, start to finish
1586 storage_index : time to compute the storage index
1587 peer_selection : time to decide which peers will be used
1588 contacting_helper : initial helper query to upload/no-upload decision
1589 existence_check : helper pre-upload existence check
1590 helper_total : initial helper query to helper finished pushing
1591 cumulative_fetch : helper waiting for ciphertext requests
1592 total_fetch : helper start to last ciphertext response
1593 cumulative_encoding : just time spent in zfec
1594 cumulative_sending : just time spent waiting for storage servers
1595 hashes_and_close : last segment push to shareholder close
1596 total_encode_and_push : first encode to shareholder close
1600 class IDownloadResults(Interface):
1601 """I am created internally by download() methods. I contain a number of
1602 public attributes which contain details about the download process.::
1604 .file_size : the size of the file, in bytes
1605 .servers_used : set of server peerids that were used during download
1606 .server_problems : dict mapping server peerid to a problem string. Only
1607 servers that had problems (bad hashes, disconnects)
1609 .servermap : dict mapping server peerid to a set of share numbers. Only
1610 servers that had any shares are listed here.
1611 .timings : dict of timing information, mapping name to seconds (float)
1612 peer_selection : time to ask servers about shares
1613 servers_peer_selection : dict of peerid to DYHB-query time
1614 uri_extension : time to fetch a copy of the URI extension block
1615 hashtrees : time to fetch the hash trees
1616 segments : time to fetch, decode, and deliver segments
1617 cumulative_fetch : time spent waiting for storage servers
1618 cumulative_decode : just time spent in zfec
1619 cumulative_decrypt : just time spent in decryption
1620 total : total download time, start to finish
1621 fetch_per_server : dict of peerid to list of per-segment fetch times
1625 class IUploader(Interface):
1626 def upload(uploadable):
1627 """Upload the file. 'uploadable' must impement IUploadable. This
1628 returns a Deferred which fires with an UploadResults instance, from
1629 which the URI of the file can be obtained as results.uri ."""
1631 def upload_ssk(write_capability, new_version, uploadable):
1632 """TODO: how should this work?"""
1634 class ICheckable(Interface):
1635 def check(monitor, verify=False, add_lease=False):
1636 """Check upon my health, optionally repairing any problems.
1638 This returns a Deferred that fires with an instance that provides
1639 ICheckResults, or None if the object is non-distributed (i.e. LIT
1642 The monitor will be checked periodically to see if the operation has
1643 been cancelled. If so, no new queries will be sent, and the Deferred
1644 will fire (with a OperationCancelledError) immediately.
1646 Filenodes and dirnodes (which provide IFilesystemNode) are also
1647 checkable. Instances that represent verifier-caps will be checkable
1648 but not downloadable. Some objects (like LIT files) do not actually
1649 live in the grid, and their checkers return None (non-distributed
1650 files are always healthy).
1652 If verify=False, a relatively lightweight check will be performed: I
1653 will ask all servers if they have a share for me, and I will believe
1654 whatever they say. If there are at least N distinct shares on the
1655 grid, my results will indicate r.is_healthy()==True. This requires a
1656 roundtrip to each server, but does not transfer very much data, so
1657 the network bandwidth is fairly low.
1659 If verify=True, a more resource-intensive check will be performed:
1660 every share will be downloaded, and the hashes will be validated on
1661 every bit. I will ignore any shares that failed their hash checks. If
1662 there are at least N distinct valid shares on the grid, my results
1663 will indicate r.is_healthy()==True. This requires N/k times as much
1664 download bandwidth (and server disk IO) as a regular download. If a
1665 storage server is holding a corrupt share, or is experiencing memory
1666 failures during retrieval, or is malicious or buggy, then
1667 verification will detect the problem, but checking will not.
1669 If add_lease=True, I will ensure that an up-to-date lease is present
1670 on each share. The lease secrets will be derived from by node secret
1671 (in BASEDIR/private/secret), so either I will add a new lease to the
1672 share, or I will merely renew the lease that I already had. In a
1673 future version of the storage-server protocol (once Accounting has
1674 been implemented), there may be additional options here to define the
1675 kind of lease that is obtained (which account number to claim, etc).
1677 TODO: any problems seen during checking will be reported to the
1678 health-manager.furl, a centralized object which is responsible for
1679 figuring out why files are unhealthy so corrective action can be
1683 def check_and_repair(monitor, verify=False, add_lease=False):
1684 """Like check(), but if the file/directory is not healthy, attempt to
1687 Any non-healthy result will cause an immediate repair operation, to
1688 generate and upload new shares. After repair, the file will be as
1689 healthy as we can make it. Details about what sort of repair is done
1690 will be put in the check-and-repair results. The Deferred will not
1691 fire until the repair is complete.
1693 This returns a Deferred which fires with an instance of
1694 ICheckAndRepairResults."""
1696 class IDeepCheckable(Interface):
1697 def start_deep_check(verify=False, add_lease=False):
1698 """Check upon the health of me and everything I can reach.
1700 This is a recursive form of check(), useable only on dirnodes.
1702 I return a Monitor, with results that are an IDeepCheckResults
1705 TODO: If any of the directories I traverse are unrecoverable, the
1706 Monitor will report failure. If any of the files I check upon are
1707 unrecoverable, those problems will be reported in the
1708 IDeepCheckResults as usual, and the Monitor will not report a
1712 def start_deep_check_and_repair(verify=False, add_lease=False):
1713 """Check upon the health of me and everything I can reach. Repair
1714 anything that isn't healthy.
1716 This is a recursive form of check_and_repair(), useable only on
1719 I return a Monitor, with results that are an
1720 IDeepCheckAndRepairResults object.
1722 TODO: If any of the directories I traverse are unrecoverable, the
1723 Monitor will report failure. If any of the files I check upon are
1724 unrecoverable, those problems will be reported in the
1725 IDeepCheckResults as usual, and the Monitor will not report a
1729 class ICheckResults(Interface):
1730 """I contain the detailed results of a check/verify operation.
1733 def get_storage_index():
1734 """Return a string with the (binary) storage index."""
1735 def get_storage_index_string():
1736 """Return a string with the (printable) abbreviated storage index."""
1738 """Return the (string) URI of the object that was checked."""
1741 """Return a boolean, True if the file/dir is fully healthy, False if
1742 it is damaged in any way. Non-distributed LIT files always return
1745 def is_recoverable():
1746 """Return a boolean, True if the file/dir can be recovered, False if
1747 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1748 files always return True."""
1750 def needs_rebalancing():
1751 """Return a boolean, True if the file/dir's reliability could be
1752 improved by moving shares to new servers. Non-distributed LIT files
1753 always return False."""
1757 """Return a dictionary that describes the state of the file/dir. LIT
1758 files always return an empty dictionary. Normal files and directories
1759 return a dictionary with the following keys (note that these use
1760 binary strings rather than base32-encoded ones) (also note that for
1761 mutable files, these counts are for the 'best' version):
1763 count-shares-good: the number of distinct good shares that were found
1764 count-shares-needed: 'k', the number of shares required for recovery
1765 count-shares-expected: 'N', the number of total shares generated
1766 count-good-share-hosts: the number of distinct storage servers with
1767 good shares. If this number is less than
1768 count-shares-good, then some shares are
1769 doubled up, increasing the correlation of
1770 failures. This indicates that one or more
1771 shares should be moved to an otherwise unused
1772 server, if one is available.
1773 count-corrupt-shares: the number of shares with integrity failures
1774 list-corrupt-shares: a list of 'share locators', one for each share
1775 that was found to be corrupt. Each share
1776 locator is a list of (serverid, storage_index,
1778 count-incompatible-shares: the number of shares which are of a share
1779 format unknown to this checker
1780 list-incompatible-shares: a list of 'share locators', one for each
1781 share that was found to be of an unknown
1782 format. Each share locator is a list of
1783 (serverid, storage_index, sharenum).
1784 servers-responding: list of (binary) storage server identifiers,
1785 one for each server which responded to the share
1786 query (even if they said they didn't have
1787 shares, and even if they said they did have
1788 shares but then didn't send them when asked, or
1789 dropped the connection, or returned a Failure,
1790 and even if they said they did have shares and
1791 sent incorrect ones when asked)
1792 sharemap: dict mapping share identifier to list of serverids
1793 (binary strings). This indicates which servers are holding
1794 which shares. For immutable files, the shareid is an
1795 integer (the share number, from 0 to N-1). For mutable
1796 files, it is a string of the form 'seq%d-%s-sh%d',
1797 containing the sequence number, the roothash, and the
1800 The following keys are most relevant for mutable files, but immutable
1801 files will provide sensible values too::
1803 count-wrong-shares: the number of shares for versions other than the
1804 'best' one (which is defined as being the
1805 recoverable version with the highest sequence
1806 number, then the highest roothash). These are
1807 either leftover shares from an older version
1808 (perhaps on a server that was offline when an
1809 update occurred), shares from an unrecoverable
1810 newer version, or shares from an alternate
1811 current version that results from an
1812 uncoordinated write collision. For a healthy
1813 file, this will equal 0.
1815 count-recoverable-versions: the number of recoverable versions of
1816 the file. For a healthy file, this will
1819 count-unrecoverable-versions: the number of unrecoverable versions
1820 of the file. For a healthy file, this
1826 """Return a string with a brief (one-line) summary of the results."""
1829 """Return a list of strings with more detailed results."""
1831 class ICheckAndRepairResults(Interface):
1832 """I contain the detailed results of a check/verify/repair operation.
1834 The IFilesystemNode.check()/verify()/repair() methods all return
1835 instances that provide ICheckAndRepairResults.
1838 def get_storage_index():
1839 """Return a string with the (binary) storage index."""
1840 def get_storage_index_string():
1841 """Return a string with the (printable) abbreviated storage index."""
1842 def get_repair_attempted():
1843 """Return a boolean, True if a repair was attempted. We might not
1844 attempt to repair the file because it was healthy, or healthy enough
1845 (i.e. some shares were missing but not enough to exceed some
1846 threshold), or because we don't know how to repair this object."""
1847 def get_repair_successful():
1848 """Return a boolean, True if repair was attempted and the file/dir
1849 was fully healthy afterwards. False if no repair was attempted or if
1850 a repair attempt failed."""
1851 def get_pre_repair_results():
1852 """Return an ICheckResults instance that describes the state of the
1853 file/dir before any repair was attempted."""
1854 def get_post_repair_results():
1855 """Return an ICheckResults instance that describes the state of the
1856 file/dir after any repair was attempted. If no repair was attempted,
1857 the pre-repair and post-repair results will be identical."""
1860 class IDeepCheckResults(Interface):
1861 """I contain the results of a deep-check operation.
1863 This is returned by a call to ICheckable.deep_check().
1866 def get_root_storage_index_string():
1867 """Return the storage index (abbreviated human-readable string) of
1868 the first object checked."""
1870 """Return a dictionary with the following keys::
1872 count-objects-checked: count of how many objects were checked
1873 count-objects-healthy: how many of those objects were completely
1875 count-objects-unhealthy: how many were damaged in some way
1876 count-objects-unrecoverable: how many were unrecoverable
1877 count-corrupt-shares: how many shares were found to have
1878 corruption, summed over all objects
1882 def get_corrupt_shares():
1883 """Return a set of (serverid, storage_index, sharenum) for all shares
1884 that were found to be corrupt. Both serverid and storage_index are
1887 def get_all_results():
1888 """Return a dictionary mapping pathname (a tuple of strings, ready to
1889 be slash-joined) to an ICheckResults instance, one for each object
1890 that was checked."""
1892 def get_results_for_storage_index(storage_index):
1893 """Retrive the ICheckResults instance for the given (binary)
1894 storage index. Raises KeyError if there are no results for that
1898 """Return a dictionary with the same keys as
1899 IDirectoryNode.deep_stats()."""
1901 class IDeepCheckAndRepairResults(Interface):
1902 """I contain the results of a deep-check-and-repair operation.
1904 This is returned by a call to ICheckable.deep_check_and_repair().
1907 def get_root_storage_index_string():
1908 """Return the storage index (abbreviated human-readable string) of
1909 the first object checked."""
1911 """Return a dictionary with the following keys::
1913 count-objects-checked: count of how many objects were checked
1914 count-objects-healthy-pre-repair: how many of those objects were
1915 completely healthy (before any
1917 count-objects-unhealthy-pre-repair: how many were damaged in
1919 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1920 count-objects-healthy-post-repair: how many of those objects were
1921 completely healthy (after any
1923 count-objects-unhealthy-post-repair: how many were damaged in
1925 count-objects-unrecoverable-post-repair: how many were
1927 count-repairs-attempted: repairs were attempted on this many
1928 objects. The count-repairs- keys will
1929 always be provided, however unless
1930 repair=true is present, they will all
1932 count-repairs-successful: how many repairs resulted in healthy
1934 count-repairs-unsuccessful: how many repairs resulted did not
1935 results in completely healthy objects
1936 count-corrupt-shares-pre-repair: how many shares were found to
1937 have corruption, summed over all
1938 objects examined (before any
1940 count-corrupt-shares-post-repair: how many shares were found to
1941 have corruption, summed over all
1942 objects examined (after any
1947 """Return a dictionary with the same keys as
1948 IDirectoryNode.deep_stats()."""
1950 def get_corrupt_shares():
1951 """Return a set of (serverid, storage_index, sharenum) for all shares
1952 that were found to be corrupt before any repair was attempted. Both
1953 serverid and storage_index are binary.
1955 def get_remaining_corrupt_shares():
1956 """Return a set of (serverid, storage_index, sharenum) for all shares
1957 that were found to be corrupt after any repair was completed. Both
1958 serverid and storage_index are binary. These are shares that need
1959 manual inspection and probably deletion.
1961 def get_all_results():
1962 """Return a dictionary mapping pathname (a tuple of strings, ready to
1963 be slash-joined) to an ICheckAndRepairResults instance, one for each
1964 object that was checked."""
1966 def get_results_for_storage_index(storage_index):
1967 """Retrive the ICheckAndRepairResults instance for the given (binary)
1968 storage index. Raises KeyError if there are no results for that
1972 class IRepairable(Interface):
1973 def repair(check_results):
1974 """Attempt to repair the given object. Returns a Deferred that fires
1975 with a IRepairResults object.
1977 I must be called with an object that implements ICheckResults, as
1978 proof that you have actually discovered a problem with this file. I
1979 will use the data in the checker results to guide the repair process,
1980 such as which servers provided bad data and should therefore be
1981 avoided. The ICheckResults object is inside the
1982 ICheckAndRepairResults object, which is returned by the
1983 ICheckable.check() method::
1985 d = filenode.check(repair=False)
1986 def _got_results(check_and_repair_results):
1987 check_results = check_and_repair_results.get_pre_repair_results()
1988 return filenode.repair(check_results)
1989 d.addCallback(_got_results)
1993 class IRepairResults(Interface):
1994 """I contain the results of a repair operation."""
1997 class IClient(Interface):
1998 def upload(uploadable):
1999 """Upload some data into a CHK, get back the UploadResults for it.
2000 @param uploadable: something that implements IUploadable
2001 @return: a Deferred that fires with the UploadResults instance.
2002 To get the URI for this file, use results.uri .
2005 def create_mutable_file(contents=""):
2006 """Create a new mutable file with contents, get back the URI string.
2007 @param contents: the initial contents to place in the file.
2008 @return: a Deferred that fires with tne (string) SSK URI for the new
2012 def create_empty_dirnode():
2013 """Create a new dirnode, empty and unattached.
2014 @return: a Deferred that fires with the new IDirectoryNode instance.
2017 def create_node_from_uri(uri, rouri):
2018 """Create a new IFilesystemNode instance from the uri, synchronously.
2019 @param uri: a string or IURI-providing instance, or None. This could
2020 be for a LiteralFileNode, a CHK file node, a mutable file
2021 node, or a directory node
2022 @param rouri: a string or IURI-providing instance, or None. If the
2023 main uri is None, I will use the rouri instead. If I
2024 recognize the format of the main uri, I will ignore the
2025 rouri (because it can be derived from the writecap).
2027 @return: an instance that provides IFilesystemNode (or more usefully
2028 one of its subclasses). File-specifying URIs will result in
2029 IFileNode or IMutableFileNode -providing instances, like
2030 FileNode, LiteralFileNode, or MutableFileNode.
2031 Directory-specifying URIs will result in
2032 IDirectoryNode-providing instances, like NewDirectoryNode.
2035 class IClientStatus(Interface):
2036 def list_all_uploads():
2037 """Return a list of uploader objects, one for each upload which
2038 currently has an object available (tracked with weakrefs). This is
2039 intended for debugging purposes."""
2040 def list_active_uploads():
2041 """Return a list of active IUploadStatus objects."""
2042 def list_recent_uploads():
2043 """Return a list of IUploadStatus objects for the most recently
2046 def list_all_downloads():
2047 """Return a list of downloader objects, one for each download which
2048 currently has an object available (tracked with weakrefs). This is
2049 intended for debugging purposes."""
2050 def list_active_downloads():
2051 """Return a list of active IDownloadStatus objects."""
2052 def list_recent_downloads():
2053 """Return a list of IDownloadStatus objects for the most recently
2054 started downloads."""
2056 class IUploadStatus(Interface):
2058 """Return a timestamp (float with seconds since epoch) indicating
2059 when the operation was started."""
2060 def get_storage_index():
2061 """Return a string with the (binary) storage index in use on this
2062 upload. Returns None if the storage index has not yet been
2065 """Return an integer with the number of bytes that will eventually
2066 be uploaded for this file. Returns None if the size is not yet known.
2069 """Return True if this upload is using a Helper, False if not."""
2071 """Return a string describing the current state of the upload
2074 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2075 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2076 made towards hashing the file to determine a CHK encryption key: if
2077 non-convergent encryption is in use, this will be trivial, otherwise
2078 the whole file must be hashed. 'ciphertext' describes how much of the
2079 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2080 uploads. 'encode_and_push' describes how much of the encode-and-push
2081 process has finished: for helper uploads this is dependent upon the
2082 helper providing progress reports. It might be reasonable to add all
2083 three numbers and report the sum to the user."""
2085 """Return True if the upload is currently active, False if not."""
2087 """Return an instance of UploadResults (which contains timing and
2088 sharemap information). Might return None if the upload is not yet
2091 """Each upload status gets a unique number: this method returns that
2092 number. This provides a handle to this particular upload, so a web
2093 page can generate a suitable hyperlink."""
2095 class IDownloadStatus(Interface):
2097 """Return a timestamp (float with seconds since epoch) indicating
2098 when the operation was started."""
2099 def get_storage_index():
2100 """Return a string with the (binary) storage index in use on this
2101 download. This may be None if there is no storage index (i.e. LIT
2104 """Return an integer with the number of bytes that will eventually be
2105 retrieved for this file. Returns None if the size is not yet known.
2108 """Return True if this download is using a Helper, False if not."""
2110 """Return a string describing the current state of the download
2113 """Returns a float (from 0.0 to 1.0) describing the amount of the
2114 download that has completed. This value will remain at 0.0 until the
2115 first byte of plaintext is pushed to the download target."""
2117 """Return True if the download is currently active, False if not."""
2119 """Each download status gets a unique number: this method returns
2120 that number. This provides a handle to this particular download, so a
2121 web page can generate a suitable hyperlink."""
2123 class IServermapUpdaterStatus(Interface):
2125 class IPublishStatus(Interface):
2127 class IRetrieveStatus(Interface):
2130 class NotCapableError(Exception):
2131 """You have tried to write to a read-only node."""
2133 class BadWriteEnablerError(Exception):
2136 class RIControlClient(RemoteInterface):
2138 def wait_for_client_connections(num_clients=int):
2139 """Do not return until we have connections to at least NUM_CLIENTS
2143 def upload_from_file_to_uri(filename=str,
2144 convergence=ChoiceOf(None,
2145 StringConstraint(2**20))):
2146 """Upload a file to the grid. This accepts a filename (which must be
2147 absolute) that points to a file on the node's local disk. The node will
2148 read the contents of this file, upload it to the grid, then return the
2149 URI at which it was uploaded. If convergence is None then a random
2150 encryption key will be used, else the plaintext will be hashed, then
2151 that hash will be mixed together with the "convergence" string to form
2156 def download_from_uri_to_file(uri=URI, filename=str):
2157 """Download a file from the grid, placing it on the node's local disk
2158 at the given filename (which must be absolute[?]). Returns the
2159 absolute filename where the file was written."""
2164 def get_memory_usage():
2165 """Return a dict describes the amount of memory currently in use. The
2166 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2167 measuring memory consupmtion in bytes."""
2168 return DictOf(str, int)
2170 def speed_test(count=int, size=int, mutable=Any()):
2171 """Write 'count' tempfiles to disk, all of the given size. Measure
2172 how long (in seconds) it takes to upload them all to the servers.
2173 Then measure how long it takes to download all of them. If 'mutable'
2174 is 'create', time creation of mutable files. If 'mutable' is
2175 'upload', then time access to the same mutable file instead of
2178 Returns a tuple of (upload_time, download_time).
2180 return (float, float)
2182 def measure_peer_response_time():
2183 """Send a short message to each connected peer, and measure the time
2184 it takes for them to respond to it. This is a rough measure of the
2185 application-level round trip time.
2187 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2190 return DictOf(Nodeid, float)
2192 UploadResults = Any() #DictOf(str, str)
2194 class RIEncryptedUploadable(RemoteInterface):
2195 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2200 def get_all_encoding_parameters():
2201 return (int, int, int, long)
2203 def read_encrypted(offset=Offset, length=ReadSize):
2210 class RICHKUploadHelper(RemoteInterface):
2211 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2215 Return a dictionary of version information.
2217 return DictOf(str, Any())
2219 def upload(reader=RIEncryptedUploadable):
2220 return UploadResults
2223 class RIHelper(RemoteInterface):
2224 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2228 Return a dictionary of version information.
2230 return DictOf(str, Any())
2232 def upload_chk(si=StorageIndex):
2233 """See if a file with a given storage index needs uploading. The
2234 helper will ask the appropriate storage servers to see if the file
2235 has already been uploaded. If so, the helper will return a set of
2236 'upload results' that includes whatever hashes are needed to build
2237 the read-cap, and perhaps a truncated sharemap.
2239 If the file has not yet been uploaded (or if it was only partially
2240 uploaded), the helper will return an empty upload-results dictionary
2241 and also an RICHKUploadHelper object that will take care of the
2242 upload process. The client should call upload() on this object and
2243 pass it a reference to an RIEncryptedUploadable object that will
2244 provide ciphertext. When the upload is finished, the upload() method
2245 will finish and return the upload results.
2247 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2250 class RIStatsProvider(RemoteInterface):
2251 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2253 Provides access to statistics and monitoring information.
2258 returns a dictionary containing 'counters' and 'stats', each a
2259 dictionary with string counter/stat name keys, and numeric values.
2260 counters are monotonically increasing measures of work done, and
2261 stats are instantaneous measures (potentially time averaged
2264 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2266 class RIStatsGatherer(RemoteInterface):
2267 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2269 Provides a monitoring service for centralised collection of stats
2272 def provide(provider=RIStatsProvider, nickname=str):
2274 @param provider: a stats collector instance which should be polled
2275 periodically by the gatherer to collect stats.
2276 @param nickname: a name useful to identify the provided client
2281 class IStatsProducer(Interface):
2284 returns a dictionary, with str keys representing the names of stats
2285 to be monitored, and numeric values.
2288 class RIKeyGenerator(RemoteInterface):
2289 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2291 Provides a service offering to make RSA key pairs.
2294 def get_rsa_key_pair(key_size=int):
2296 @param key_size: the size of the signature key.
2297 @return: tuple(verifying_key, signing_key)
2299 return TupleOf(str, str)
2302 class FileTooLargeError(Exception):
2305 class IValidatedThingProxy(Interface):
2307 """ Acquire a thing and validate it. Return a deferred which is
2308 eventually fired with self if the thing is valid or errbacked if it
2309 can't be acquired or validated."""
2311 class InsufficientVersionError(Exception):
2312 def __init__(self, needed, got):
2313 self.needed = needed
2316 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,