2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable share modifications
30 LeaseRenewSecret = Hash # used to protect lease renewal requests
31 LeaseCancelSecret = Hash # was used to protect lease cancellation requests
34 class RIBucketWriter(RemoteInterface):
35 """ Objects of this kind live on the server side. """
36 def write(offset=Offset, data=ShareData):
41 If the data that has been written is incomplete or inconsistent then
42 the server will throw the data away, else it will store it for future
48 """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 that is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
70 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
71 # elements are (offset, length, operator, specimen)
72 # operator is one of "lt, le, eq, ne, ge, gt"
73 # nop always passes and is used to fetch data while writing.
74 # you should use length==len(specimen) for everything except nop
75 DataVector = ListOf(TupleOf(Offset, ShareData))
76 # (offset, data). This limits us to 30 writes of 1MiB each per call
77 TestAndWriteVectorsForShares = DictOf(int,
80 ChoiceOf(None, Offset), # new_length
82 ReadVector = ListOf(TupleOf(Offset, ReadSize))
83 ReadData = ListOf(ShareData)
84 # returns data[offset:offset+length] for each element of TestVector
87 class RIStorageServer(RemoteInterface):
88 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
92 Return a dictionary of version information.
94 return DictOf(str, Any())
96 def allocate_buckets(storage_index=StorageIndex,
97 renew_secret=LeaseRenewSecret,
98 cancel_secret=LeaseCancelSecret,
99 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
100 allocated_size=Offset, canary=Referenceable):
102 @param storage_index: the index of the bucket to be created or
104 @param sharenums: these are the share numbers (probably between 0 and
105 99) that the sender is proposing to store on this
107 @param renew_secret: This is the secret used to protect bucket refresh
108 This secret is generated by the client and
109 stored for later comparison by the server. Each
110 server is given a different secret.
111 @param cancel_secret: This no longer allows lease cancellation, but
112 must still be a unique value identifying the
113 lease. XXX stop relying on it to be unique.
114 @param canary: If the canary is lost before close(), the bucket is
116 @return: tuple of (alreadygot, allocated), where alreadygot is what we
117 already have and allocated is what we hereby agree to accept.
118 New leases are added for shares in both lists.
120 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
121 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
123 def add_lease(storage_index=StorageIndex,
124 renew_secret=LeaseRenewSecret,
125 cancel_secret=LeaseCancelSecret):
127 Add a new lease on the given bucket. If the renew_secret matches an
128 existing lease, that lease will be renewed instead. If there is no
129 bucket for the given storage_index, return silently. (note that in
130 tahoe-1.3.0 and earlier, IndexError was raised if there was no
133 return Any() # returns None now, but future versions might change
135 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
137 Renew the lease on a given bucket, resetting the timer to 31 days.
138 Some networks will use this, some will not. If there is no bucket for
139 the given storage_index, IndexError will be raised.
141 For mutable shares, if the given renew_secret does not match an
142 existing lease, IndexError will be raised with a note listing the
143 server-nodeids on the existing leases, so leases on migrated shares
144 can be renewed. For immutable shares, IndexError (without the note)
149 def get_buckets(storage_index=StorageIndex):
150 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
154 def slot_readv(storage_index=StorageIndex,
155 shares=ListOf(int), readv=ReadVector):
156 """Read a vector from the numbered shares associated with the given
157 storage index. An empty shares list means to return data from all
158 known shares. Returns a dictionary with one key per share."""
159 return DictOf(int, ReadData) # shnum -> results
161 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
162 secrets=TupleOf(WriteEnablerSecret,
165 tw_vectors=TestAndWriteVectorsForShares,
169 General-purpose test-read-and-set operation for mutable slots:
170 (1) For submitted shnums, compare the test vectors against extant
171 shares, or against an empty share for shnums that do not exist.
172 (2) Use the read vectors to extract "old data" from extant shares.
173 (3) If all tests in (1) passed, then apply the write vectors
174 (possibly creating new shares).
175 (4) Return whether the tests passed, and the "old data", which does
176 not include any modifications made by the writes.
178 The operation does not interleave with other operations on the same
181 This method is, um, large. The goal is to allow clients to update all
182 the shares associated with a mutable file in a single round trip.
184 @param storage_index: the index of the bucket to be created or
186 @param write_enabler: a secret that is stored along with the slot.
187 Writes are accepted from any caller who can
188 present the matching secret. A different secret
189 should be used for each slot*server pair.
190 @param renew_secret: This is the secret used to protect bucket refresh
191 This secret is generated by the client and
192 stored for later comparison by the server. Each
193 server is given a different secret.
194 @param cancel_secret: This no longer allows lease cancellation, but
195 must still be a unique value identifying the
196 lease. XXX stop relying on it to be unique.
198 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
199 cancel_secret). The first is required to perform any write. The
200 latter two are used when allocating new shares. To simply acquire a
201 new lease on existing shares, use an empty testv and an empty writev.
203 Each share can have a separate test vector (i.e. a list of
204 comparisons to perform). If all vectors for all shares pass, then all
205 writes for all shares are recorded. Each comparison is a 4-tuple of
206 (offset, length, operator, specimen), which effectively does a
207 bool( (read(offset, length)) OPERATOR specimen ) and only performs
208 the write if all these evaluate to True. Basic test-and-set uses 'eq'.
209 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
210 Write-if-same-or-newer uses 'le'.
212 Reads from the end of the container are truncated, and missing shares
213 behave like empty ones, so to assert that a share doesn't exist (for
214 use when creating a new share), use (0, 1, 'eq', '').
216 The write vector will be applied to the given share, expanding it if
217 necessary. A write vector applied to a share number that did not
218 exist previously will cause that share to be created. Write vectors
219 must not overlap (if they do, this will either cause an error or
220 apply them in an unspecified order). Duplicate write vectors, with
221 the same offset and data, are currently tolerated but are not
224 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
225 vector whose offset is beyond the end of the current data, the space
226 between the end of the current data and the beginning of the write
227 vector will be filled with zero bytes. In earlier versions the
228 contents of this space was unspecified (and might end up containing
229 secrets). Storage servers with the new zero-filling behavior will
230 advertise a true value for the 'fills-holes-with-zero-bytes' key
231 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
234 Each write vector is accompanied by a 'new_length' argument, which
235 can be used to truncate the data. If new_length is not None and it is
236 less than the current size of the data (after applying all write
237 vectors), then the data will be truncated to new_length. If
238 new_length==0, the share will be deleted.
240 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
241 enlarge the file by sending a number larger than the size of the data
242 after applying all write vectors. That behavior was not used, and as
243 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
246 If a storage client knows that the server supports zero-filling, for
247 example from the 'fills-holes-with-zero-bytes' key in its version
248 information, it can extend the file efficiently by writing a single
249 zero byte just before the new end-of-file. Otherwise it must
250 explicitly write zeroes to all bytes between the old and new
251 end-of-file. In any case it should avoid sending new_length larger
252 than the size of the data after applying all write vectors.
254 The read vector is used to extract data from all known shares,
255 *before* any writes have been applied. The same read vector is used
256 for all shares. This captures the state that was tested by the test
257 vector, for extant shares.
259 This method returns two values: a boolean and a dict. The boolean is
260 True if the write vectors were applied, False if not. The dict is
261 keyed by share number, and each value contains a list of strings, one
262 for each element of the read vector.
264 If the write_enabler is wrong, this will raise BadWriteEnablerError.
265 To enable share migration (using update_write_enabler), the exception
266 will have the nodeid used for the old write enabler embedded in it,
267 in the following string::
269 The write enabler was recorded by nodeid '%s'.
271 Note that the nodeid here is encoded using the same base32 encoding
272 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
274 return TupleOf(bool, DictOf(int, ReadData))
276 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
277 shnum=int, reason=str):
278 """Clients who discover hash failures in shares that they have
279 downloaded from me will use this method to inform me about the
280 failures. I will record their concern so that my operator can
281 manually inspect the shares in question. I return None.
283 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
284 (binary) storage index string, and 'shnum' is the integer share
285 number. 'reason' is a human-readable explanation of the problem,
286 probably including some expected hash values and the computed ones
287 that did not match. Corruption advisories for mutable shares should
288 include a hash of the public key (the same value that appears in the
289 mutable-file verify-cap), since the current share format does not
294 class IStorageBucketWriter(Interface):
296 Objects of this kind live on the client side.
298 def put_block(segmentnum, data):
300 @param segmentnum=int
301 @param data=ShareData: For most segments, this data will be 'blocksize'
302 bytes in length. The last segment might be shorter.
303 @return: a Deferred that fires (with None) when the operation completes
306 def put_crypttext_hashes(hashes):
308 @param hashes=ListOf(Hash)
309 @return: a Deferred that fires (with None) when the operation completes
312 def put_block_hashes(blockhashes):
314 @param blockhashes=ListOf(Hash)
315 @return: a Deferred that fires (with None) when the operation completes
318 def put_share_hashes(sharehashes):
320 @param sharehashes=ListOf(TupleOf(int, Hash))
321 @return: a Deferred that fires (with None) when the operation completes
324 def put_uri_extension(data):
325 """This block of data contains integrity-checking information (hashes
326 of plaintext, crypttext, and shares), as well as encoding parameters
327 that are necessary to recover the data. This is a serialized dict
328 mapping strings to other strings. The hash of this data is kept in
329 the URI and verified before any of the data is used. All buckets for
330 a given file contain identical copies of this data.
332 The serialization format is specified with the following pseudocode:
333 for k in sorted(dict.keys()):
334 assert re.match(r'^[a-zA-Z_\-]+$', k)
335 write(k + ':' + netstring(dict[k]))
337 @param data=URIExtensionData
338 @return: a Deferred that fires (with None) when the operation completes
342 """Finish writing and close the bucket. The share is not finalized
343 until this method is called: if the uploading client disconnects
344 before calling close(), the partially-written share will be
347 @return: a Deferred that fires (with None) when the operation completes
351 class IStorageBucketReader(Interface):
352 def get_block_data(blocknum, blocksize, size):
353 """Most blocks will be the same size. The last block might be shorter
362 def get_crypttext_hashes():
364 @return: ListOf(Hash)
367 def get_block_hashes(at_least_these=()):
369 @param at_least_these=SetOf(int)
370 @return: ListOf(Hash)
373 def get_share_hashes():
375 @return: ListOf(TupleOf(int, Hash))
378 def get_uri_extension():
380 @return: URIExtensionData
384 class IStorageBroker(Interface):
385 def get_servers_for_psi(peer_selection_index):
387 @return: list of IServer instances
389 def get_connected_servers():
391 @return: frozenset of connected IServer instances
393 def get_known_servers():
395 @return: frozenset of IServer instances
397 def get_all_serverids():
399 @return: frozenset of serverid strings
401 def get_nickname_for_serverid(serverid):
403 @return: unicode nickname, or None
406 # methods moved from IntroducerClient, need review
407 def get_all_connections():
408 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
409 each active connection we've established to a remote service. This is
410 mostly useful for unit tests that need to wait until a certain number
411 of connections have been made."""
413 def get_all_connectors():
414 """Return a dict that maps from (nodeid, service_name) to a
415 RemoteServiceConnector instance for all services that we are actively
416 trying to connect to. Each RemoteServiceConnector has the following
419 service_name: the type of service provided, like 'storage'
420 announcement_time: when we first heard about this service
421 last_connect_time: when we last established a connection
422 last_loss_time: when we last lost a connection
424 version: the peer's version, from the most recent connection
425 oldest_supported: the peer's oldest supported version, same
427 rref: the RemoteReference, if connected, otherwise None
428 remote_host: the IAddress, if connected, otherwise None
430 This method is intended for monitoring interfaces, such as a web page
431 that describes connecting and connected peers.
434 def get_all_peerids():
435 """Return a frozenset of all peerids to whom we have a connection (to
436 one or more services) established. Mostly useful for unit tests."""
438 def get_all_connections_for(service_name):
439 """Return a frozenset of (nodeid, service_name, rref) tuples, one
440 for each active connection that provides the given SERVICE_NAME."""
442 def get_permuted_peers(service_name, key):
443 """Returns an ordered list of (peerid, rref) tuples, selecting from
444 the connections that provide SERVICE_NAME, using a hash-based
445 permutation keyed by KEY. This randomizes the service list in a
446 repeatable way, to distribute load over many peers.
450 class IDisplayableServer(Interface):
461 class IServer(IDisplayableServer):
462 """I live in the client, and represent a single server."""
463 def start_connecting(tub, trigger_cb):
467 """Once a server is connected, I return a RemoteReference.
468 Before a server is connected for the first time, I return None.
470 Note that the rref I return will start producing DeadReferenceErrors
471 once the connection is lost.
475 class IMutableSlotWriter(Interface):
477 The interface for a writer around a mutable slot on a remote server.
479 def set_checkstring(seqnum_or_checkstring, root_hash=None, salt=None):
481 Set the checkstring that I will pass to the remote server when
484 @param checkstring A packed checkstring to use.
486 Note that implementations can differ in which semantics they
487 wish to support for set_checkstring -- they can, for example,
488 build the checkstring themselves from its constituents, or
492 def get_checkstring():
494 Get the checkstring that I think currently exists on the remote
498 def put_block(data, segnum, salt):
500 Add a block and salt to the share.
503 def put_encprivkey(encprivkey):
505 Add the encrypted private key to the share.
508 def put_blockhashes(blockhashes):
510 @param blockhashes=list
511 Add the block hash tree to the share.
514 def put_sharehashes(sharehashes):
516 @param sharehashes=dict
517 Add the share hash chain to the share.
522 Return the part of the share that needs to be signed.
525 def put_signature(signature):
527 Add the signature to the share.
530 def put_verification_key(verification_key):
532 Add the verification key to the share.
535 def finish_publishing():
537 Do anything necessary to finish writing the share to a remote
538 server. I require that no further publishing needs to take place
539 after this method has been called.
543 class IURI(Interface):
544 def init_from_string(uri):
545 """Accept a string (as created by my to_string() method) and populate
546 this instance with its data. I am not normally called directly,
547 please use the module-level uri.from_string() function to convert
548 arbitrary URI strings into IURI-providing instances."""
551 """Return False if this URI be used to modify the data. Return True
552 if this URI cannot be used to modify the data."""
555 """Return True if the data can be modified by *somebody* (perhaps
556 someone who has a more powerful URI than this one)."""
558 # TODO: rename to get_read_cap()
560 """Return another IURI instance that represents a read-only form of
561 this one. If is_readonly() is True, this returns self."""
563 def get_verify_cap():
564 """Return an instance that provides IVerifierURI, which can be used
565 to check on the availability of the file or directory, without
566 providing enough capabilities to actually read or modify the
567 contents. This may return None if the file does not need checking or
568 verification (e.g. LIT URIs).
572 """Return a string of printable ASCII characters, suitable for
573 passing into init_from_string."""
576 class IVerifierURI(Interface, IURI):
577 def init_from_string(uri):
578 """Accept a string (as created by my to_string() method) and populate
579 this instance with its data. I am not normally called directly,
580 please use the module-level uri.from_string() function to convert
581 arbitrary URI strings into IURI-providing instances."""
584 """Return a string of printable ASCII characters, suitable for
585 passing into init_from_string."""
588 class IDirnodeURI(Interface):
589 """I am a URI that represents a dirnode."""
592 class IFileURI(Interface):
593 """I am a URI that represents a filenode."""
595 """Return the length (in bytes) of the file that I represent."""
598 class IImmutableFileURI(IFileURI):
601 class IMutableFileURI(Interface):
604 class IDirectoryURI(Interface):
607 class IReadonlyDirectoryURI(Interface):
611 class CapConstraintError(Exception):
612 """A constraint on a cap was violated."""
614 class MustBeDeepImmutableError(CapConstraintError):
615 """Mutable children cannot be added to an immutable directory.
616 Also, caps obtained from an immutable directory can trigger this error
617 if they are later found to refer to a mutable object and then used."""
619 class MustBeReadonlyError(CapConstraintError):
620 """Known write caps cannot be specified in a ro_uri field. Also,
621 caps obtained from a ro_uri field can trigger this error if they
622 are later found to be write caps and then used."""
624 class MustNotBeUnknownRWError(CapConstraintError):
625 """Cannot add an unknown child cap specified in a rw_uri field."""
628 class IReadable(Interface):
629 """I represent a readable object -- either an immutable file, or a
630 specific version of a mutable file.
634 """Return True if this reference provides mutable access to the given
635 file or directory (i.e. if you can modify it), or False if not. Note
636 that even if this reference is read-only, someone else may hold a
637 read-write reference to it.
639 For an IReadable returned by get_best_readable_version(), this will
640 always return True, but for instances of subinterfaces such as
641 IMutableFileVersion, it may return False."""
644 """Return True if this file or directory is mutable (by *somebody*,
645 not necessarily you), False if it is is immutable. Note that a file
646 might be mutable overall, but your reference to it might be
647 read-only. On the other hand, all references to an immutable file
648 will be read-only; there are no read-write references to an immutable
651 def get_storage_index():
652 """Return the storage index of the file."""
655 """Return the length (in bytes) of this readable object."""
657 def download_to_data():
658 """Download all of the file contents. I return a Deferred that fires
659 with the contents as a byte string."""
661 def read(consumer, offset=0, size=None):
662 """Download a portion (possibly all) of the file's contents, making
663 them available to the given IConsumer. Return a Deferred that fires
664 (with the consumer) when the consumer is unregistered (either because
665 the last byte has been given to it, or because the consumer threw an
666 exception during write(), possibly because it no longer wants to
667 receive data). The portion downloaded will start at 'offset' and
668 contain 'size' bytes (or the remainder of the file if size==None). It
669 is an error to read beyond the end of the file: callers must use
670 get_size() and clip any non-default offset= and size= parameters. It
671 is permissible to read zero bytes.
673 The consumer will be used in non-streaming mode: an IPullProducer
674 will be attached to it.
676 The consumer will not receive data right away: several network trips
677 must occur first. The order of events will be::
679 consumer.registerProducer(p, streaming)
680 (if streaming == False)::
681 consumer does p.resumeProducing()
683 consumer does p.resumeProducing()
684 consumer.write(data).. (repeat until all data is written)
685 consumer.unregisterProducer()
686 deferred.callback(consumer)
688 If a download error occurs, or an exception is raised by
689 consumer.registerProducer() or consumer.write(), I will call
690 consumer.unregisterProducer() and then deliver the exception via
691 deferred.errback(). To cancel the download, the consumer should call
692 p.stopProducing(), which will result in an exception being delivered
693 via deferred.errback().
695 See src/allmydata/util/consumer.py for an example of a simple
696 download-to-memory consumer.
700 class IWriteable(Interface):
702 I define methods that callers can use to update SDMF and MDMF
703 mutable files on a Tahoe-LAFS grid.
705 # XXX: For the moment, we have only this. It is possible that we
706 # want to move overwrite() and modify() in here too.
707 def update(data, offset):
709 I write the data from my data argument to the MDMF file,
710 starting at offset. I continue writing data until my data
711 argument is exhausted, appending data to the file as necessary.
713 # assert IMutableUploadable.providedBy(data)
714 # to append data: offset=node.get_size_of_best_version()
715 # do we want to support compacting MDMF?
716 # for an MDMF file, this can be done with O(data.get_size())
717 # memory. For an SDMF file, any modification takes
718 # O(node.get_size_of_best_version()).
721 class IMutableFileVersion(IReadable):
722 """I provide access to a particular version of a mutable file. The
723 access is read/write if I was obtained from a filenode derived from
724 a write cap, or read-only if the filenode was derived from a read cap.
727 def get_sequence_number():
728 """Return the sequence number of this version."""
731 """Return the IMutableFileServerMap instance that was used to create
736 """Return this filenode's writekey, or None if the node does not have
737 write-capability. This may be used to assist with data structures
738 that need to make certain data available only to writers, such as the
739 read-write child caps in dirnodes. The recommended process is to have
740 reader-visible data be submitted to the filenode in the clear (where
741 it will be encrypted by the filenode using the readkey), but encrypt
742 writer-visible data using this writekey.
745 def overwrite(new_contents):
746 """Replace the contents of the mutable file, provided that no other
747 node has published (or is attempting to publish, concurrently) a
748 newer version of the file than this one.
750 I will avoid modifying any share that is different than the version
751 given by get_sequence_number(). However, if another node is writing
752 to the file at the same time as me, I may manage to update some shares
753 while they update others. If I see any evidence of this, I will signal
754 UncoordinatedWriteError, and the file will be left in an inconsistent
755 state (possibly the version you provided, possibly the old version,
756 possibly somebody else's version, and possibly a mix of shares from
759 The recommended response to UncoordinatedWriteError is to either
760 return it to the caller (since they failed to coordinate their
761 writes), or to attempt some sort of recovery. It may be sufficient to
762 wait a random interval (with exponential backoff) and repeat your
763 operation. If I do not signal UncoordinatedWriteError, then I was
764 able to write the new version without incident.
766 I return a Deferred that fires (with a PublishStatus object) when the
767 update has completed.
770 def modify(modifier_cb):
771 """Modify the contents of the file, by downloading this version,
772 applying the modifier function (or bound method), then uploading
773 the new version. This will succeed as long as no other node
774 publishes a version between the download and the upload.
775 I return a Deferred that fires (with a PublishStatus object) when
776 the update is complete.
778 The modifier callable will be given three arguments: a string (with
779 the old contents), a 'first_time' boolean, and a servermap. As with
780 download_to_data(), the old contents will be from this version,
781 but the modifier can use the servermap to make other decisions
782 (such as refusing to apply the delta if there are multiple parallel
783 versions, or if there is evidence of a newer unrecoverable version).
784 'first_time' will be True the first time the modifier is called,
785 and False on any subsequent calls.
787 The callable should return a string with the new contents. The
788 callable must be prepared to be called multiple times, and must
789 examine the input string to see if the change that it wants to make
790 is already present in the old version. If it does not need to make
791 any changes, it can either return None, or return its input string.
793 If the modifier raises an exception, it will be returned in the
798 # The hierarchy looks like this:
805 class IFilesystemNode(Interface):
807 """Return the strongest 'cap instance' associated with this node.
808 (writecap for writeable-mutable files/directories, readcap for
809 immutable or readonly-mutable files/directories). To convert this
810 into a string, call .to_string() on the result."""
813 """Return a readonly cap instance for this node. For immutable or
814 readonly nodes, get_cap() and get_readcap() return the same thing."""
816 def get_repair_cap():
817 """Return an IURI instance that can be used to repair the file, or
818 None if this node cannot be repaired (either because it is not
819 distributed, like a LIT file, or because the node does not represent
820 sufficient authority to create a repair-cap, like a read-only RSA
821 mutable file node [which cannot create the correct write-enablers]).
824 def get_verify_cap():
825 """Return an IVerifierURI instance that represents the
826 'verifiy/refresh capability' for this node. The holder of this
827 capability will be able to renew the lease for this node, protecting
828 it from garbage-collection. They will also be able to ask a server if
829 it holds a share for the file or directory.
833 """Return the URI string corresponding to the strongest cap associated
834 with this node. If this node is read-only, the URI will only offer
835 read-only access. If this node is read-write, the URI will offer
838 If you have read-write access to a node and wish to share merely
839 read-only access with others, use get_readonly_uri().
843 """Return the URI string that can be used by others to get write
844 access to this node, if it is writeable. If this is a read-only node,
847 def get_readonly_uri():
848 """Return the URI string that can be used by others to get read-only
849 access to this node. The result is a read-only URI, regardless of
850 whether this node is read-only or read-write.
852 If you have merely read-only access to this node, get_readonly_uri()
853 will return the same thing as get_uri().
856 def get_storage_index():
857 """Return a string with the (binary) storage index in use on this
858 download. This may be None if there is no storage index (i.e. LIT
859 files and directories)."""
862 """Return True if this reference provides mutable access to the given
863 file or directory (i.e. if you can modify it), or False if not. Note
864 that even if this reference is read-only, someone else may hold a
865 read-write reference to it."""
868 """Return True if this file or directory is mutable (by *somebody*,
869 not necessarily you), False if it is is immutable. Note that a file
870 might be mutable overall, but your reference to it might be
871 read-only. On the other hand, all references to an immutable file
872 will be read-only; there are no read-write references to an immutable
877 """Return True if this is an unknown node."""
879 def is_allowed_in_immutable_directory():
880 """Return True if this node is allowed as a child of a deep-immutable
881 directory. This is true if either the node is of a known-immutable type,
882 or it is unknown and read-only.
886 """Raise any error associated with this node."""
888 # XXX: These may not be appropriate outside the context of an IReadable.
890 """Return the length (in bytes) of the data this node represents. For
891 directory nodes, I return the size of the backing store. I return
892 synchronously and do not consult the network, so for mutable objects,
893 I will return the most recently observed size for the object, or None
894 if I don't remember a size. Use get_current_size, which returns a
895 Deferred, if you want more up-to-date information."""
897 def get_current_size():
898 """I return a Deferred that fires with the length (in bytes) of the
899 data this node represents.
903 class IFileNode(IFilesystemNode):
904 """I am a node that represents a file: a sequence of bytes. I am not a
905 container, like IDirectoryNode."""
906 def get_best_readable_version():
907 """Return a Deferred that fires with an IReadable for the 'best'
908 available version of the file. The IReadable provides only read
909 access, even if this filenode was derived from a write cap.
911 For an immutable file, there is only one version. For a mutable
912 file, the 'best' version is the recoverable version with the
913 highest sequence number. If no uncoordinated writes have occurred,
914 and if enough shares are available, then this will be the most
915 recent version that has been uploaded. If no version is recoverable,
916 the Deferred will errback with an UnrecoverableFileError.
919 def download_best_version():
920 """Download the contents of the version that would be returned
921 by get_best_readable_version(). This is equivalent to calling
922 download_to_data() on the IReadable given by that method.
924 I return a Deferred that fires with a byte string when the file
925 has been fully downloaded. To support streaming download, use
926 the 'read' method of IReadable. If no version is recoverable,
927 the Deferred will errback with an UnrecoverableFileError.
930 def get_size_of_best_version():
931 """Find the size of the version that would be returned by
932 get_best_readable_version().
934 I return a Deferred that fires with an integer. If no version
935 is recoverable, the Deferred will errback with an
936 UnrecoverableFileError.
940 class IImmutableFileNode(IFileNode, IReadable):
941 """I am a node representing an immutable file. Immutable files have
945 class IMutableFileNode(IFileNode):
946 """I provide access to a 'mutable file', which retains its identity
947 regardless of what contents are put in it.
949 The consistency-vs-availability problem means that there might be
950 multiple versions of a file present in the grid, some of which might be
951 unrecoverable (i.e. have fewer than 'k' shares). These versions are
952 loosely ordered: each has a sequence number and a hash, and any version
953 with seqnum=N was uploaded by a node that has seen at least one version
956 The 'servermap' (an instance of IMutableFileServerMap) is used to
957 describe the versions that are known to be present in the grid, and which
958 servers are hosting their shares. It is used to represent the 'state of
959 the world', and is used for this purpose by my test-and-set operations.
960 Downloading the contents of the mutable file will also return a
961 servermap. Uploading a new version into the mutable file requires a
962 servermap as input, and the semantics of the replace operation is
963 'replace the file with my new version if it looks like nobody else has
964 changed the file since my previous download'. Because the file is
965 distributed, this is not a perfect test-and-set operation, but it will do
966 its best. If the replace process sees evidence of a simultaneous write,
967 it will signal an UncoordinatedWriteError, so that the caller can take
971 Most readers will want to use the 'best' current version of the file, and
972 should use my 'download_best_version()' method.
974 To unconditionally replace the file, callers should use overwrite(). This
975 is the mode that user-visible mutable files will probably use.
977 To apply some delta to the file, call modify() with a callable modifier
978 function that can apply the modification that you want to make. This is
979 the mode that dirnodes will use, since most directory modification
980 operations can be expressed in terms of deltas to the directory state.
983 Three methods are available for users who need to perform more complex
984 operations. The first is get_servermap(), which returns an up-to-date
985 servermap using a specified mode. The second is download_version(), which
986 downloads a specific version (not necessarily the 'best' one). The third
987 is 'upload', which accepts new contents and a servermap (which must have
988 been updated with MODE_WRITE). The upload method will attempt to apply
989 the new contents as long as no other node has modified the file since the
990 servermap was updated. This might be useful to a caller who wants to
991 merge multiple versions into a single new one.
993 Note that each time the servermap is updated, a specific 'mode' is used,
994 which determines how many peers are queried. To use a servermap for my
995 replace() method, that servermap must have been updated in MODE_WRITE.
996 These modes are defined in allmydata.mutable.common, and consist of
997 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
998 allmydata/mutable/servermap.py for details about the differences.
1000 Mutable files are currently limited in size (about 3.5MB max) and can
1001 only be retrieved and updated all-at-once, as a single big string. Future
1002 versions of our mutable files will remove this restriction.
1004 def get_best_mutable_version():
1005 """Return a Deferred that fires with an IMutableFileVersion for
1006 the 'best' available version of the file. The best version is
1007 the recoverable version with the highest sequence number. If no
1008 uncoordinated writes have occurred, and if enough shares are
1009 available, then this will be the most recent version that has
1012 If no version is recoverable, the Deferred will errback with an
1013 UnrecoverableFileError.
1016 def overwrite(new_contents):
1017 """Unconditionally replace the contents of the mutable file with new
1018 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1019 is only appropriate to use when the new contents of the file are
1020 completely unrelated to the old ones, and you do not care about other
1023 I return a Deferred that fires (with a PublishStatus object) when the
1024 update has completed.
1027 def modify(modifier_cb):
1028 """Modify the contents of the file, by downloading the current
1029 version, applying the modifier function (or bound method), then
1030 uploading the new version. I return a Deferred that fires (with a
1031 PublishStatus object) when the update is complete.
1033 The modifier callable will be given three arguments: a string (with
1034 the old contents), a 'first_time' boolean, and a servermap. As with
1035 download_best_version(), the old contents will be from the best
1036 recoverable version, but the modifier can use the servermap to make
1037 other decisions (such as refusing to apply the delta if there are
1038 multiple parallel versions, or if there is evidence of a newer
1039 unrecoverable version). 'first_time' will be True the first time the
1040 modifier is called, and False on any subsequent calls.
1042 The callable should return a string with the new contents. The
1043 callable must be prepared to be called multiple times, and must
1044 examine the input string to see if the change that it wants to make
1045 is already present in the old version. If it does not need to make
1046 any changes, it can either return None, or return its input string.
1048 If the modifier raises an exception, it will be returned in the
1052 def get_servermap(mode):
1053 """Return a Deferred that fires with an IMutableFileServerMap
1054 instance, updated using the given mode.
1057 def download_version(servermap, version):
1058 """Download a specific version of the file, using the servermap
1059 as a guide to where the shares are located.
1061 I return a Deferred that fires with the requested contents, or
1062 errbacks with UnrecoverableFileError. Note that a servermap that was
1063 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1064 all versions (those modes stop querying servers as soon as they can
1065 fulfil their goals), so you may want to use MODE_CHECK (which checks
1066 everything) to get increased visibility.
1069 def upload(new_contents, servermap):
1070 """Replace the contents of the file with new ones. This requires a
1071 servermap that was previously updated with MODE_WRITE.
1073 I attempt to provide test-and-set semantics, in that I will avoid
1074 modifying any share that is different than the version I saw in the
1075 servermap. However, if another node is writing to the file at the
1076 same time as me, I may manage to update some shares while they update
1077 others. If I see any evidence of this, I will signal
1078 UncoordinatedWriteError, and the file will be left in an inconsistent
1079 state (possibly the version you provided, possibly the old version,
1080 possibly somebody else's version, and possibly a mix of shares from
1083 The recommended response to UncoordinatedWriteError is to either
1084 return it to the caller (since they failed to coordinate their
1085 writes), or to attempt some sort of recovery. It may be sufficient to
1086 wait a random interval (with exponential backoff) and repeat your
1087 operation. If I do not signal UncoordinatedWriteError, then I was
1088 able to write the new version without incident.
1090 I return a Deferred that fires (with a PublishStatus object) when the
1091 publish has completed. I will update the servermap in-place with the
1092 location of all new shares.
1096 """Return this filenode's writekey, or None if the node does not have
1097 write-capability. This may be used to assist with data structures
1098 that need to make certain data available only to writers, such as the
1099 read-write child caps in dirnodes. The recommended process is to have
1100 reader-visible data be submitted to the filenode in the clear (where
1101 it will be encrypted by the filenode using the readkey), but encrypt
1102 writer-visible data using this writekey.
1106 """Returns the mutable file protocol version."""
1109 class NotEnoughSharesError(Exception):
1110 """Download was unable to get enough shares"""
1112 class NoSharesError(Exception):
1113 """Download was unable to get any shares at all."""
1115 class DownloadStopped(Exception):
1118 class UploadUnhappinessError(Exception):
1119 """Upload was unable to satisfy 'servers_of_happiness'"""
1121 class UnableToFetchCriticalDownloadDataError(Exception):
1122 """I was unable to fetch some piece of critical data that is supposed to
1123 be identically present in all shares."""
1125 class NoServersError(Exception):
1126 """Upload wasn't given any servers to work with, usually indicating a
1127 network or Introducer problem."""
1129 class ExistingChildError(Exception):
1130 """A directory node was asked to add or replace a child that already
1131 exists, and overwrite= was set to False."""
1133 class NoSuchChildError(Exception):
1134 """A directory node was asked to fetch a child that does not exist."""
1136 # avoid UnicodeEncodeErrors when converting to str
1137 return self.__repr__()
1139 class ChildOfWrongTypeError(Exception):
1140 """An operation was attempted on a child of the wrong type (file or directory)."""
1143 class IDirectoryNode(IFilesystemNode):
1144 """I represent a filesystem node that is a container, with a
1145 name-to-child mapping, holding the tahoe equivalent of a directory. All
1146 child names are unicode strings, and all children are some sort of
1147 IFilesystemNode (a file, subdirectory, or unknown node).
1152 The dirnode ('1') URI returned by this method can be used in
1153 set_uri() on a different directory ('2') to 'mount' a reference to
1154 this directory ('1') under the other ('2'). This URI is just a
1155 string, so it can be passed around through email or other out-of-band
1159 def get_readonly_uri():
1161 The dirnode ('1') URI returned by this method can be used in
1162 set_uri() on a different directory ('2') to 'mount' a reference to
1163 this directory ('1') under the other ('2'). This URI is just a
1164 string, so it can be passed around through email or other out-of-band
1169 """I return a Deferred that fires with a dictionary mapping child
1170 name (a unicode string) to (node, metadata_dict) tuples, in which
1171 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1174 def has_child(name):
1175 """I return a Deferred that fires with a boolean, True if there
1176 exists a child of the given name, False if not. The child name must
1177 be a unicode string."""
1180 """I return a Deferred that fires with a specific named child node,
1181 which is an IFilesystemNode. The child name must be a unicode string.
1182 I raise NoSuchChildError if I do not have a child by that name."""
1184 def get_metadata_for(name):
1185 """I return a Deferred that fires with the metadata dictionary for
1186 a specific named child node. The child name must be a unicode string.
1187 This metadata is stored in the *edge*, not in the child, so it is
1188 attached to the parent dirnode rather than the child node.
1189 I raise NoSuchChildError if I do not have a child by that name."""
1191 def set_metadata_for(name, metadata):
1192 """I replace any existing metadata for the named child with the new
1193 metadata. The child name must be a unicode string. This metadata is
1194 stored in the *edge*, not in the child, so it is attached to the
1195 parent dirnode rather than the child node. I return a Deferred
1196 (that fires with this dirnode) when the operation is complete.
1197 I raise NoSuchChildError if I do not have a child by that name."""
1199 def get_child_at_path(path):
1200 """Transform a child path into an IFilesystemNode.
1202 I perform a recursive series of 'get' operations to find the named
1203 descendant node. I return a Deferred that fires with the node, or
1204 errbacks with NoSuchChildError if the node could not be found.
1206 The path can be either a single string (slash-separated) or a list of
1207 path-name elements. All elements must be unicode strings.
1210 def get_child_and_metadata_at_path(path):
1211 """Transform a child path into an IFilesystemNode and metadata.
1213 I am like get_child_at_path(), but my Deferred fires with a tuple of
1214 (node, metadata). The metadata comes from the last edge. If the path
1215 is empty, the metadata will be an empty dictionary.
1218 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1219 """I add a child (by writecap+readcap) at the specific name. I return
1220 a Deferred that fires when the operation finishes. If overwrite= is
1221 True, I will replace any existing child of the same name, otherwise
1222 an existing child will cause me to return ExistingChildError. The
1223 child name must be a unicode string.
1225 The child caps could be for a file, or for a directory. If you have
1226 both the writecap and readcap, you should provide both arguments.
1227 If you have only one cap and don't know whether it is read-only,
1228 provide it as the writecap argument and leave the readcap as None.
1229 If you have only one cap that is known to be read-only, provide it
1230 as the readcap argument and leave the writecap as None.
1231 The filecaps are typically obtained from an IFilesystemNode with
1232 get_uri() and get_readonly_uri().
1234 If metadata= is provided, I will use it as the metadata for the named
1235 edge. This will replace any existing metadata. If metadata= is left
1236 as the default value of None, I will set ['mtime'] to the current
1237 time, and I will set ['ctime'] to the current time if there was not
1238 already a child by this name present. This roughly matches the
1239 ctime/mtime semantics of traditional filesystems. See the
1240 "About the metadata" section of webapi.txt for futher information.
1242 If this directory node is read-only, the Deferred will errback with a
1243 NotWriteableError."""
1245 def set_children(entries, overwrite=True):
1246 """Add multiple children (by writecap+readcap) to a directory node.
1247 Takes a dictionary, with childname as keys and (writecap, readcap)
1248 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1249 a Deferred that fires (with this dirnode) when the operation
1250 finishes. This is equivalent to calling set_uri() multiple times, but
1251 is much more efficient. All child names must be unicode strings.
1254 def set_node(name, child, metadata=None, overwrite=True):
1255 """I add a child at the specific name. I return a Deferred that fires
1256 when the operation finishes. This Deferred will fire with the child
1257 node that was just added. I will replace any existing child of the
1258 same name. The child name must be a unicode string. The 'child'
1259 instance must be an instance providing IFilesystemNode.
1261 If metadata= is provided, I will use it as the metadata for the named
1262 edge. This will replace any existing metadata. If metadata= is left
1263 as the default value of None, I will set ['mtime'] to the current
1264 time, and I will set ['ctime'] to the current time if there was not
1265 already a child by this name present. This roughly matches the
1266 ctime/mtime semantics of traditional filesystems. See the
1267 "About the metadata" section of webapi.txt for futher information.
1269 If this directory node is read-only, the Deferred will errback with a
1270 NotWriteableError."""
1272 def set_nodes(entries, overwrite=True):
1273 """Add multiple children to a directory node. Takes a dict mapping
1274 unicode childname to (child_node, metdata) tuples. If metdata=None,
1275 the original metadata is left unmodified. Returns a Deferred that
1276 fires (with this dirnode) when the operation finishes. This is
1277 equivalent to calling set_node() multiple times, but is much more
1280 def add_file(name, uploadable, metadata=None, overwrite=True):
1281 """I upload a file (using the given IUploadable), then attach the
1282 resulting ImmutableFileNode to the directory at the given name. I set
1283 metadata the same way as set_uri and set_node. The child name must be
1286 I return a Deferred that fires (with the IFileNode of the uploaded
1287 file) when the operation completes."""
1289 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1290 """I remove the child at the specific name. I return a Deferred that
1291 fires when the operation finishes. The child name must be a unicode
1292 string. If must_exist is True and I do not have a child by that name,
1293 I raise NoSuchChildError. If must_be_directory is True and the child
1294 is a file, or if must_be_file is True and the child is a directory,
1295 I raise ChildOfWrongTypeError."""
1297 def create_subdirectory(name, initial_children={}, overwrite=True,
1298 mutable=True, mutable_version=None, metadata=None):
1299 """I create and attach a directory at the given name. The new
1300 directory can be empty, or it can be populated with children
1301 according to 'initial_children', which takes a dictionary in the same
1302 format as set_nodes (i.e. mapping unicode child name to (childnode,
1303 metadata) tuples). The child name must be a unicode string. I return
1304 a Deferred that fires (with the new directory node) when the
1305 operation finishes."""
1307 def move_child_to(current_child_name, new_parent, new_child_name=None,
1309 """I take one of my children and move them to a new parent. The child
1310 is referenced by name. On the new parent, the child will live under
1311 'new_child_name', which defaults to 'current_child_name'. TODO: what
1312 should we do about metadata? I return a Deferred that fires when the
1313 operation finishes. The child name must be a unicode string. I raise
1314 NoSuchChildError if I do not have a child by that name."""
1316 def build_manifest():
1317 """I generate a table of everything reachable from this directory.
1318 I also compute deep-stats as described below.
1320 I return a Monitor. The Monitor's results will be a dictionary with
1323 res['manifest']: a list of (path, cap) tuples for all nodes
1324 (directories and files) reachable from this one.
1325 'path' will be a tuple of unicode strings. The
1326 origin dirnode will be represented by an empty path
1328 res['verifycaps']: a list of (printable) verifycap strings, one for
1329 each reachable non-LIT node. This is a set:
1330 it will contain no duplicates.
1331 res['storage-index']: a list of (base32) storage index strings,
1332 one for each reachable non-LIT node. This is
1333 a set: it will contain no duplicates.
1334 res['stats']: a dictionary, the same that is generated by
1335 start_deep_stats() below.
1337 The Monitor will also have an .origin_si attribute with the (binary)
1338 storage index of the starting point.
1341 def start_deep_stats():
1342 """Return a Monitor, examining all nodes (directories and files)
1343 reachable from this one. The Monitor's results will be a dictionary
1344 with the following keys::
1346 count-immutable-files: count of how many CHK files are in the set
1347 count-mutable-files: same, for mutable files (does not include
1349 count-literal-files: same, for LIT files
1350 count-files: sum of the above three
1352 count-directories: count of directories
1354 size-immutable-files: total bytes for all CHK files in the set
1355 size-mutable-files (TODO): same, for current version of all mutable
1356 files, does not include directories
1357 size-literal-files: same, for LIT files
1358 size-directories: size of mutable files used by directories
1360 largest-directory: number of bytes in the largest directory
1361 largest-directory-children: number of children in the largest
1363 largest-immutable-file: number of bytes in the largest CHK file
1365 size-mutable-files is not yet implemented, because it would involve
1366 even more queries than deep_stats does.
1368 The Monitor will also have an .origin_si attribute with the (binary)
1369 storage index of the starting point.
1371 This operation will visit every directory node underneath this one,
1372 and can take a long time to run. On a typical workstation with good
1373 bandwidth, this can examine roughly 15 directories per second (and
1374 takes several minutes of 100% CPU for ~1700 directories).
1378 class ICodecEncoder(Interface):
1379 def set_params(data_size, required_shares, max_shares):
1380 """Set up the parameters of this encoder.
1382 This prepares the encoder to perform an operation that converts a
1383 single block of data into a number of shares, such that a future
1384 ICodecDecoder can use a subset of these shares to recover the
1385 original data. This operation is invoked by calling encode(). Once
1386 the encoding parameters are set up, the encode operation can be
1387 invoked multiple times.
1389 set_params() prepares the encoder to accept blocks of input data that
1390 are exactly 'data_size' bytes in length. The encoder will be prepared
1391 to produce 'max_shares' shares for each encode() operation (although
1392 see the 'desired_share_ids' to use less CPU). The encoding math will
1393 be chosen such that the decoder can get by with as few as
1394 'required_shares' of these shares and still reproduce the original
1395 data. For example, set_params(1000, 5, 5) offers no redundancy at
1396 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1398 Numerical Restrictions: 'data_size' is required to be an integral
1399 multiple of 'required_shares'. In general, the caller should choose
1400 required_shares and max_shares based upon their reliability
1401 requirements and the number of peers available (the total storage
1402 space used is roughly equal to max_shares*data_size/required_shares),
1403 then choose data_size to achieve the memory footprint desired (larger
1404 data_size means more efficient operation, smaller data_size means
1405 smaller memory footprint).
1407 In addition, 'max_shares' must be equal to or greater than
1408 'required_shares'. Of course, setting them to be equal causes
1409 encode() to degenerate into a particularly slow form of the 'split'
1412 See encode() for more details about how these parameters are used.
1414 set_params() must be called before any other ICodecEncoder methods
1419 """Return the 3-tuple of data_size, required_shares, max_shares"""
1421 def get_encoder_type():
1422 """Return a short string that describes the type of this encoder.
1424 There is required to be a global table of encoder classes. This method
1425 returns an index into this table; the value at this index is an
1426 encoder class, and this encoder is an instance of that class.
1429 def get_block_size():
1430 """Return the length of the shares that encode() will produce.
1433 def encode_proposal(data, desired_share_ids=None):
1434 """Encode some data.
1436 'data' must be a string (or other buffer object), and len(data) must
1437 be equal to the 'data_size' value passed earlier to set_params().
1439 This will return a Deferred that will fire with two lists. The first
1440 is a list of shares, each of which is a string (or other buffer
1441 object) such that len(share) is the same as what get_share_size()
1442 returned earlier. The second is a list of shareids, in which each is
1443 an integer. The lengths of the two lists will always be equal to each
1444 other. The user should take care to keep each share closely
1445 associated with its shareid, as one is useless without the other.
1447 The length of this output list will normally be the same as the value
1448 provided to the 'max_shares' parameter of set_params(). This may be
1449 different if 'desired_share_ids' is provided.
1451 'desired_share_ids', if provided, is required to be a sequence of
1452 ints, each of which is required to be >= 0 and < max_shares. If not
1453 provided, encode() will produce 'max_shares' shares, as if
1454 'desired_share_ids' were set to range(max_shares). You might use this
1455 if you initially thought you were going to use 10 peers, started
1456 encoding, and then two of the peers dropped out: you could use
1457 desired_share_ids= to skip the work (both memory and CPU) of
1458 producing shares for the peers that are no longer available.
1462 def encode(inshares, desired_share_ids=None):
1463 """Encode some data. This may be called multiple times. Each call is
1466 inshares is a sequence of length required_shares, containing buffers
1467 (i.e. strings), where each buffer contains the next contiguous
1468 non-overlapping segment of the input data. Each buffer is required to
1469 be the same length, and the sum of the lengths of the buffers is
1470 required to be exactly the data_size promised by set_params(). (This
1471 implies that the data has to be padded before being passed to
1472 encode(), unless of course it already happens to be an even multiple
1473 of required_shares in length.)
1475 Note: the requirement to break up your data into
1476 'required_shares' chunks of exactly the right length before
1477 calling encode() is surprising from point of view of a user
1478 who doesn't know how FEC works. It feels like an
1479 implementation detail that has leaked outside the abstraction
1480 barrier. Is there a use case in which the data to be encoded
1481 might already be available in pre-segmented chunks, such that
1482 it is faster or less work to make encode() take a list rather
1483 than splitting a single string?
1485 Yes, there is: suppose you are uploading a file with K=64,
1486 N=128, segsize=262,144. Then each in-share will be of size
1487 4096. If you use this .encode() API then your code could first
1488 read each successive 4096-byte chunk from the file and store
1489 each one in a Python string and store each such Python string
1490 in a Python list. Then you could call .encode(), passing that
1491 list as "inshares". The encoder would generate the other 64
1492 "secondary shares" and return to you a new list containing
1493 references to the same 64 Python strings that you passed in
1494 (as the primary shares) plus references to the new 64 Python
1497 (You could even imagine that your code could use readv() so
1498 that the operating system can arrange to get all of those
1499 bytes copied from the file into the Python list of Python
1500 strings as efficiently as possible instead of having a loop
1501 written in C or in Python to copy the next part of the file
1502 into the next string.)
1504 On the other hand if you instead use the .encode_proposal()
1505 API (above), then your code can first read in all of the
1506 262,144 bytes of the segment from the file into a Python
1507 string, then call .encode_proposal() passing the segment data
1508 as the "data" argument. The encoder would basically first
1509 split the "data" argument into a list of 64 in-shares of 4096
1510 byte each, and then do the same thing that .encode() does. So
1511 this would result in a little bit more copying of data and a
1512 little bit higher of a "maximum memory usage" during the
1513 process, although it might or might not make a practical
1514 difference for our current use cases.
1516 Note that "inshares" is a strange name for the parameter if
1517 you think of the parameter as being just for feeding in data
1518 to the codec. It makes more sense if you think of the result
1519 of this encoding as being the set of shares from inshares plus
1520 an extra set of "secondary shares" (or "check shares"). It is
1521 a surprising name! If the API is going to be surprising then
1522 the name should be surprising. If we switch to
1523 encode_proposal() above then we should also switch to an
1526 'desired_share_ids', if provided, is required to be a sequence of
1527 ints, each of which is required to be >= 0 and < max_shares. If not
1528 provided, encode() will produce 'max_shares' shares, as if
1529 'desired_share_ids' were set to range(max_shares). You might use this
1530 if you initially thought you were going to use 10 peers, started
1531 encoding, and then two of the peers dropped out: you could use
1532 desired_share_ids= to skip the work (both memory and CPU) of
1533 producing shares for the peers that are no longer available.
1535 For each call, encode() will return a Deferred that fires with two
1536 lists, one containing shares and the other containing the shareids.
1537 The get_share_size() method can be used to determine the length of
1538 the share strings returned by encode(). Each shareid is a small
1539 integer, exactly as passed into 'desired_share_ids' (or
1540 range(max_shares), if desired_share_ids was not provided).
1542 The shares and their corresponding shareids are required to be kept
1543 together during storage and retrieval. Specifically, the share data is
1544 useless by itself: the decoder needs to be told which share is which
1545 by providing it with both the shareid and the actual share data.
1547 This function will allocate an amount of memory roughly equal to::
1549 (max_shares - required_shares) * get_share_size()
1551 When combined with the memory that the caller must allocate to
1552 provide the input data, this leads to a memory footprint roughly
1553 equal to the size of the resulting encoded shares (i.e. the expansion
1554 factor times the size of the input segment).
1559 # returning a list of (shareidN,shareN) tuples instead of a pair of
1560 # lists (shareids..,shares..). Brian thought the tuples would
1561 # encourage users to keep the share and shareid together throughout
1562 # later processing, Zooko pointed out that the code to iterate
1563 # through two lists is not really more complicated than using a list
1564 # of tuples and there's also a performance improvement
1566 # having 'data_size' not required to be an integral multiple of
1567 # 'required_shares'. Doing this would require encode() to perform
1568 # padding internally, and we'd prefer to have any padding be done
1569 # explicitly by the caller. Yes, it is an abstraction leak, but
1570 # hopefully not an onerous one.
1573 class ICodecDecoder(Interface):
1574 def set_params(data_size, required_shares, max_shares):
1575 """Set the params. They have to be exactly the same ones that were
1576 used for encoding."""
1578 def get_needed_shares():
1579 """Return the number of shares needed to reconstruct the data.
1580 set_params() is required to be called before this."""
1582 def decode(some_shares, their_shareids):
1583 """Decode a partial list of shares into data.
1585 'some_shares' is required to be a sequence of buffers of sharedata, a
1586 subset of the shares returned by ICodecEncode.encode(). Each share is
1587 required to be of the same length. The i'th element of their_shareids
1588 is required to be the shareid of the i'th buffer in some_shares.
1590 This returns a Deferred that fires with a sequence of buffers. This
1591 sequence will contain all of the segments of the original data, in
1592 order. The sum of the lengths of all of the buffers will be the
1593 'data_size' value passed into the original ICodecEncode.set_params()
1594 call. To get back the single original input block of data, use
1595 ''.join(output_buffers), or you may wish to simply write them in
1596 order to an output file.
1598 Note that some of the elements in the result sequence may be
1599 references to the elements of the some_shares input sequence. In
1600 particular, this means that if those share objects are mutable (e.g.
1601 arrays) and if they are changed, then both the input (the
1602 'some_shares' parameter) and the output (the value given when the
1603 deferred is triggered) will change.
1605 The length of 'some_shares' is required to be exactly the value of
1606 'required_shares' passed into the original ICodecEncode.set_params()
1611 class IEncoder(Interface):
1612 """I take an object that provides IEncryptedUploadable, which provides
1613 encrypted data, and a list of shareholders. I then encode, hash, and
1614 deliver shares to those shareholders. I will compute all the necessary
1615 Merkle hash trees that are necessary to validate the crypttext that
1616 eventually comes back from the shareholders. I provide the URI Extension
1617 Block Hash, and the encoding parameters, both of which must be included
1620 I do not choose shareholders, that is left to the IUploader. I must be
1621 given a dict of RemoteReferences to storage buckets that are ready and
1622 willing to receive data.
1626 """Specify the number of bytes that will be encoded. This must be
1627 peformed before get_serialized_params() can be called.
1630 def set_encrypted_uploadable(u):
1631 """Provide a source of encrypted upload data. 'u' must implement
1632 IEncryptedUploadable.
1634 When this is called, the IEncryptedUploadable will be queried for its
1635 length and the storage_index that should be used.
1637 This returns a Deferred that fires with this Encoder instance.
1639 This must be performed before start() can be called.
1642 def get_param(name):
1643 """Return an encoding parameter, by name.
1645 'storage_index': return a string with the (16-byte truncated SHA-256
1646 hash) storage index to which these shares should be
1649 'share_counts': return a tuple describing how many shares are used:
1650 (needed_shares, servers_of_happiness, total_shares)
1652 'num_segments': return an int with the number of segments that
1655 'segment_size': return an int with the size of each segment.
1657 'block_size': return the size of the individual blocks that will
1658 be delivered to a shareholder's put_block() method. By
1659 knowing this, the shareholder will be able to keep all
1660 blocks in a single file and still provide random access
1661 when reading them. # TODO: can we avoid exposing this?
1663 'share_size': an int with the size of the data that will be stored
1664 on each shareholder. This is aggregate amount of data
1665 that will be sent to the shareholder, summed over all
1666 the put_block() calls I will ever make. It is useful to
1667 determine this size before asking potential
1668 shareholders whether they will grant a lease or not,
1669 since their answers will depend upon how much space we
1670 need. TODO: this might also include some amount of
1671 overhead, like the size of all the hashes. We need to
1672 decide whether this is useful or not.
1674 'serialized_params': a string with a concise description of the
1675 codec name and its parameters. This may be passed
1676 into the IUploadable to let it make sure that
1677 the same file encoded with different parameters
1678 will result in different storage indexes.
1680 Once this is called, set_size() and set_params() may not be called.
1683 def set_shareholders(shareholders, servermap):
1684 """Tell the encoder where to put the encoded shares. 'shareholders'
1685 must be a dictionary that maps share number (an integer ranging from
1686 0 to n-1) to an instance that provides IStorageBucketWriter.
1687 'servermap' is a dictionary that maps share number (as defined above)
1688 to a set of peerids. This must be performed before start() can be
1692 """Begin the encode/upload process. This involves reading encrypted
1693 data from the IEncryptedUploadable, encoding it, uploading the shares
1694 to the shareholders, then sending the hash trees.
1696 set_encrypted_uploadable() and set_shareholders() must be called
1697 before this can be invoked.
1699 This returns a Deferred that fires with a verify cap when the upload
1700 process is complete. The verifycap, plus the encryption key, is
1701 sufficient to construct the read cap.
1705 class IDecoder(Interface):
1706 """I take a list of shareholders and some setup information, then
1707 download, validate, decode, and decrypt data from them, writing the
1708 results to an output file.
1710 I do not locate the shareholders, that is left to the IDownloader. I must
1711 be given a dict of RemoteReferences to storage buckets that are ready to
1716 """I take a file-like object (providing write and close) to which all
1717 the plaintext data will be written.
1719 TODO: producer/consumer . Maybe write() should return a Deferred that
1720 indicates when it will accept more data? But probably having the
1721 IDecoder be a producer is easier to glue to IConsumer pieces.
1724 def set_shareholders(shareholders):
1725 """I take a dictionary that maps share identifiers (small integers)
1726 to RemoteReferences that provide RIBucketReader. This must be called
1730 """I start the download. This process involves retrieving data and
1731 hash chains from the shareholders, using the hashes to validate the
1732 data, decoding the shares into segments, decrypting the segments,
1733 then writing the resulting plaintext to the output file.
1735 I return a Deferred that will fire (with self) when the download is
1740 class IDownloadTarget(Interface):
1741 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1742 # will register itself as a producer. This allows the target to invoke
1743 # downloader.pauseProducing, resumeProducing, and stopProducing.
1745 """Called before any calls to write() or close(). If an error
1746 occurs before any data is available, fail() may be called without
1747 a previous call to open().
1749 'size' is the length of the file being downloaded, in bytes."""
1752 """Output some data to the target."""
1755 """Inform the target that there is no more data to be written."""
1758 """fail() is called to indicate that the download has failed. 'why'
1759 is a Failure object indicating what went wrong. No further methods
1760 will be invoked on the IDownloadTarget after fail()."""
1762 def register_canceller(cb):
1763 """The CiphertextDownloader uses this to register a no-argument function
1764 that the target can call to cancel the download. Once this canceller
1765 is invoked, no further calls to write() or close() will be made."""
1768 """When the CiphertextDownloader is done, this finish() function will be
1769 called. Whatever it returns will be returned to the invoker of
1770 Downloader.download.
1774 class IDownloader(Interface):
1775 def download(uri, target):
1776 """Perform a CHK download, sending the data to the given target.
1777 'target' must provide IDownloadTarget.
1779 Returns a Deferred that fires (with the results of target.finish)
1780 when the download is finished, or errbacks if something went wrong."""
1783 class IEncryptedUploadable(Interface):
1784 def set_upload_status(upload_status):
1785 """Provide an IUploadStatus object that should be filled with status
1786 information. The IEncryptedUploadable is responsible for setting
1787 key-determination progress ('chk'), size, storage_index, and
1788 ciphertext-fetch progress. It may delegate some of this
1789 responsibility to others, in particular to the IUploadable."""
1792 """This behaves just like IUploadable.get_size()."""
1794 def get_all_encoding_parameters():
1795 """Return a Deferred that fires with a tuple of
1796 (k,happy,n,segment_size). The segment_size will be used as-is, and
1797 must match the following constraints: it must be a multiple of k, and
1798 it shouldn't be unreasonably larger than the file size (if
1799 segment_size is larger than filesize, the difference must be stored
1802 This usually passes through to the IUploadable method of the same
1805 The encoder strictly obeys the values returned by this method. To
1806 make an upload use non-default encoding parameters, you must arrange
1807 to control the values that this method returns.
1810 def get_storage_index():
1811 """Return a Deferred that fires with a 16-byte storage index.
1814 def read_encrypted(length, hash_only):
1815 """This behaves just like IUploadable.read(), but returns crypttext
1816 instead of plaintext. If hash_only is True, then this discards the
1817 data (and returns an empty list); this improves efficiency when
1818 resuming an interrupted upload (where we need to compute the
1819 plaintext hashes, but don't need the redundant encrypted data)."""
1822 """Just like IUploadable.close()."""
1825 class IUploadable(Interface):
1826 def set_upload_status(upload_status):
1827 """Provide an IUploadStatus object that should be filled with status
1828 information. The IUploadable is responsible for setting
1829 key-determination progress ('chk')."""
1831 def set_default_encoding_parameters(params):
1832 """Set the default encoding parameters, which must be a dict mapping
1833 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1834 'max_segment_size'. These might have an influence on the final
1835 encoding parameters returned by get_all_encoding_parameters(), if the
1836 Uploadable doesn't have more specific preferences.
1838 This call is optional: if it is not used, the Uploadable will use
1839 some built-in defaults. If used, this method must be called before
1840 any other IUploadable methods to have any effect.
1844 """Return a Deferred that will fire with the length of the data to be
1845 uploaded, in bytes. This will be called before the data is actually
1846 used, to compute encoding parameters.
1849 def get_all_encoding_parameters():
1850 """Return a Deferred that fires with a tuple of
1851 (k,happy,n,segment_size). The segment_size will be used as-is, and
1852 must match the following constraints: it must be a multiple of k, and
1853 it shouldn't be unreasonably larger than the file size (if
1854 segment_size is larger than filesize, the difference must be stored
1857 The relative values of k and n allow some IUploadables to request
1858 better redundancy than others (in exchange for consuming more space
1861 Larger values of segment_size reduce hash overhead, while smaller
1862 values reduce memory footprint and cause data to be delivered in
1863 smaller pieces (which may provide a smoother and more predictable
1864 download experience).
1866 The encoder strictly obeys the values returned by this method. To
1867 make an upload use non-default encoding parameters, you must arrange
1868 to control the values that this method returns. One way to influence
1869 them may be to call set_encoding_parameters() before calling
1870 get_all_encoding_parameters().
1873 def get_encryption_key():
1874 """Return a Deferred that fires with a 16-byte AES key. This key will
1875 be used to encrypt the data. The key will also be hashed to derive
1878 Uploadables that want to achieve convergence should hash their file
1879 contents and the serialized_encoding_parameters to form the key
1880 (which of course requires a full pass over the data). Uploadables can
1881 use the upload.ConvergentUploadMixin class to achieve this
1884 Uploadables that do not care about convergence (or do not wish to
1885 make multiple passes over the data) can simply return a
1886 strongly-random 16 byte string.
1888 get_encryption_key() may be called multiple times: the IUploadable is
1889 required to return the same value each time.
1893 """Return a Deferred that fires with a list of strings (perhaps with
1894 only a single element) that, when concatenated together, contain the
1895 next 'length' bytes of data. If EOF is near, this may provide fewer
1896 than 'length' bytes. The total number of bytes provided by read()
1897 before it signals EOF must equal the size provided by get_size().
1899 If the data must be acquired through multiple internal read
1900 operations, returning a list instead of a single string may help to
1901 reduce string copies. However, the length of the concatenated strings
1902 must equal the amount of data requested, unless EOF is encountered.
1903 Long reads, or short reads without EOF, are not allowed. read()
1904 should return the same amount of data as a local disk file read, just
1905 in a different shape and asynchronously.
1907 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1908 so a 10kB file means length=3kB, 100kB file means length=30kB,
1909 and >=1MB file means length=300kB.
1911 This method provides for a single full pass through the data. Later
1912 use cases may desire multiple passes or access to only parts of the
1913 data (such as a mutable file making small edits-in-place). This API
1914 will be expanded once those use cases are better understood.
1918 """The upload is finished, and whatever filehandle was in use may be
1922 class IMutableUploadable(Interface):
1924 I represent content that is due to be uploaded to a mutable filecap.
1926 # This is somewhat simpler than the IUploadable interface above
1927 # because mutable files do not need to be concerned with possibly
1928 # generating a CHK, nor with per-file keys. It is a subset of the
1929 # methods in IUploadable, though, so we could just as well implement
1930 # the mutable uploadables as IUploadables that don't happen to use
1931 # those methods (with the understanding that the unused methods will
1932 # never be called on such objects)
1935 Returns a Deferred that fires with the size of the content held
1941 Returns a list of strings that, when concatenated, are the next
1942 length bytes of the file, or fewer if there are fewer bytes
1943 between the current location and the end of the file.
1948 The process that used the Uploadable is finished using it, so
1949 the uploadable may be closed.
1953 class IUploadResults(Interface):
1954 """I am returned by immutable upload() methods and contain the results of
1957 Note that some of my methods return empty values (0 or an empty dict)
1958 when called for non-distributed LIT files."""
1960 def get_file_size():
1961 """Return the file size, in bytes."""
1964 """Return the (string) URI of the object uploaded, a CHK readcap."""
1966 def get_ciphertext_fetched():
1967 """Return the number of bytes fetched by the helpe for this upload,
1968 or 0 if the helper did not need to fetch any bytes (or if there was
1971 def get_preexisting_shares():
1972 """Return the number of shares that were already present in the grid."""
1974 def get_pushed_shares():
1975 """Return the number of shares that were uploaded."""
1978 """Return a dict mapping share identifier to set of IServer
1979 instances. This indicates which servers were given which shares. For
1980 immutable files, the shareid is an integer (the share number, from 0
1981 to N-1). For mutable files, it is a string of the form
1982 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1983 the share number."""
1985 def get_servermap():
1986 """Return dict mapping IServer instance to a set of share numbers."""
1989 """Return dict of timing information, mapping name to seconds. All
1991 total : total upload time, start to finish
1992 storage_index : time to compute the storage index
1993 peer_selection : time to decide which peers will be used
1994 contacting_helper : initial helper query to upload/no-upload decision
1995 helper_total : initial helper query to helper finished pushing
1996 cumulative_fetch : helper waiting for ciphertext requests
1997 total_fetch : helper start to last ciphertext response
1998 cumulative_encoding : just time spent in zfec
1999 cumulative_sending : just time spent waiting for storage servers
2000 hashes_and_close : last segment push to shareholder close
2001 total_encode_and_push : first encode to shareholder close
2004 def get_uri_extension_data():
2005 """Return the dict of UEB data created for this file."""
2007 def get_verifycapstr():
2008 """Return the (string) verify-cap URI for the uploaded object."""
2011 class IDownloadResults(Interface):
2012 """I am created internally by download() methods. I contain a number of
2013 public attributes that contain details about the download process.::
2015 .file_size : the size of the file, in bytes
2016 .servers_used : set of server peerids that were used during download
2017 .server_problems : dict mapping server peerid to a problem string. Only
2018 servers that had problems (bad hashes, disconnects)
2020 .servermap : dict mapping server peerid to a set of share numbers. Only
2021 servers that had any shares are listed here.
2022 .timings : dict of timing information, mapping name to seconds (float)
2023 peer_selection : time to ask servers about shares
2024 servers_peer_selection : dict of peerid to DYHB-query time
2025 uri_extension : time to fetch a copy of the URI extension block
2026 hashtrees : time to fetch the hash trees
2027 segments : time to fetch, decode, and deliver segments
2028 cumulative_fetch : time spent waiting for storage servers
2029 cumulative_decode : just time spent in zfec
2030 cumulative_decrypt : just time spent in decryption
2031 total : total download time, start to finish
2032 fetch_per_server : dict of server to list of per-segment fetch times
2036 class IUploader(Interface):
2037 def upload(uploadable):
2038 """Upload the file. 'uploadable' must impement IUploadable. This
2039 returns a Deferred that fires with an IUploadResults instance, from
2040 which the URI of the file can be obtained as results.uri ."""
2043 class ICheckable(Interface):
2044 def check(monitor, verify=False, add_lease=False):
2045 """Check up on my health, optionally repairing any problems.
2047 This returns a Deferred that fires with an instance that provides
2048 ICheckResults, or None if the object is non-distributed (i.e. LIT
2051 The monitor will be checked periodically to see if the operation has
2052 been cancelled. If so, no new queries will be sent, and the Deferred
2053 will fire (with a OperationCancelledError) immediately.
2055 Filenodes and dirnodes (which provide IFilesystemNode) are also
2056 checkable. Instances that represent verifier-caps will be checkable
2057 but not downloadable. Some objects (like LIT files) do not actually
2058 live in the grid, and their checkers return None (non-distributed
2059 files are always healthy).
2061 If verify=False, a relatively lightweight check will be performed: I
2062 will ask all servers if they have a share for me, and I will believe
2063 whatever they say. If there are at least N distinct shares on the
2064 grid, my results will indicate r.is_healthy()==True. This requires a
2065 roundtrip to each server, but does not transfer very much data, so
2066 the network bandwidth is fairly low.
2068 If verify=True, a more resource-intensive check will be performed:
2069 every share will be downloaded, and the hashes will be validated on
2070 every bit. I will ignore any shares that failed their hash checks. If
2071 there are at least N distinct valid shares on the grid, my results
2072 will indicate r.is_healthy()==True. This requires N/k times as much
2073 download bandwidth (and server disk IO) as a regular download. If a
2074 storage server is holding a corrupt share, or is experiencing memory
2075 failures during retrieval, or is malicious or buggy, then
2076 verification will detect the problem, but checking will not.
2078 If add_lease=True, I will ensure that an up-to-date lease is present
2079 on each share. The lease secrets will be derived from by node secret
2080 (in BASEDIR/private/secret), so either I will add a new lease to the
2081 share, or I will merely renew the lease that I already had. In a
2082 future version of the storage-server protocol (once Accounting has
2083 been implemented), there may be additional options here to define the
2084 kind of lease that is obtained (which account number to claim, etc).
2086 TODO: any problems seen during checking will be reported to the
2087 health-manager.furl, a centralized object that is responsible for
2088 figuring out why files are unhealthy so corrective action can be
2092 def check_and_repair(monitor, verify=False, add_lease=False):
2093 """Like check(), but if the file/directory is not healthy, attempt to
2096 Any non-healthy result will cause an immediate repair operation, to
2097 generate and upload new shares. After repair, the file will be as
2098 healthy as we can make it. Details about what sort of repair is done
2099 will be put in the check-and-repair results. The Deferred will not
2100 fire until the repair is complete.
2102 This returns a Deferred that fires with an instance of
2103 ICheckAndRepairResults."""
2106 class IDeepCheckable(Interface):
2107 def start_deep_check(verify=False, add_lease=False):
2108 """Check upon the health of me and everything I can reach.
2110 This is a recursive form of check(), useable only on dirnodes.
2112 I return a Monitor, with results that are an IDeepCheckResults
2115 TODO: If any of the directories I traverse are unrecoverable, the
2116 Monitor will report failure. If any of the files I check upon are
2117 unrecoverable, those problems will be reported in the
2118 IDeepCheckResults as usual, and the Monitor will not report a
2122 def start_deep_check_and_repair(verify=False, add_lease=False):
2123 """Check upon the health of me and everything I can reach. Repair
2124 anything that isn't healthy.
2126 This is a recursive form of check_and_repair(), useable only on
2129 I return a Monitor, with results that are an
2130 IDeepCheckAndRepairResults object.
2132 TODO: If any of the directories I traverse are unrecoverable, the
2133 Monitor will report failure. If any of the files I check upon are
2134 unrecoverable, those problems will be reported in the
2135 IDeepCheckResults as usual, and the Monitor will not report a
2140 class ICheckResults(Interface):
2141 """I contain the detailed results of a check/verify operation.
2144 def get_storage_index():
2145 """Return a string with the (binary) storage index."""
2147 def get_storage_index_string():
2148 """Return a string with the (printable) abbreviated storage index."""
2151 """Return the (string) URI of the object that was checked."""
2154 """Return a boolean, True if the file/dir is fully healthy, False if
2155 it is damaged in any way. Non-distributed LIT files always return
2158 def is_recoverable():
2159 """Return a boolean, True if the file/dir can be recovered, False if
2160 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2161 files always return True."""
2163 # the following methods all return None for non-distributed LIT files
2165 def get_happiness():
2166 """Return the happiness count of the file."""
2168 def get_encoding_needed():
2169 """Return 'k', the number of shares required for recovery."""
2171 def get_encoding_expected():
2172 """Return 'N', the number of total shares generated."""
2174 def get_share_counter_good():
2175 """Return the number of distinct good shares that were found. For
2176 mutable files, this counts shares for the 'best' version."""
2178 def get_share_counter_wrong():
2179 """For mutable files, return the number of shares for versions other
2180 than the 'best' one (which is defined as being the recoverable
2181 version with the highest sequence number, then the highest roothash).
2182 These are either leftover shares from an older version (perhaps on a
2183 server that was offline when an update occurred), shares from an
2184 unrecoverable newer version, or shares from an alternate current
2185 version that results from an uncoordinated write collision. For a
2186 healthy file, this will equal 0. For immutable files, this will
2189 def get_corrupt_shares():
2190 """Return a list of 'share locators', one for each share that was
2191 found to be corrupt (integrity failure). Each share locator is a list
2192 of (IServer, storage_index, sharenum)."""
2194 def get_incompatible_shares():
2195 """Return a list of 'share locators', one for each share that was
2196 found to be of an unknown format. Each share locator is a list of
2197 (IServer, storage_index, sharenum)."""
2199 def get_servers_responding():
2200 """Return a list of IServer objects, one for each server that
2201 responded to the share query (even if they said they didn't have
2202 shares, and even if they said they did have shares but then didn't
2203 send them when asked, or dropped the connection, or returned a
2204 Failure, and even if they said they did have shares and sent
2205 incorrect ones when asked)"""
2207 def get_host_counter_good_shares():
2208 """Return the number of distinct storage servers with good shares. If
2209 this number is less than get_share_counters()[good], then some shares
2210 are doubled up, increasing the correlation of failures. This
2211 indicates that one or more shares should be moved to an otherwise
2212 unused server, if one is available.
2215 def get_version_counter_recoverable():
2216 """Return the number of recoverable versions of the file. For a
2217 healthy file, this will equal 1."""
2219 def get_version_counter_unrecoverable():
2220 """Return the number of unrecoverable versions of the file. For a
2221 healthy file, this will be 0."""
2224 """Return a dict mapping share identifier to list of IServer objects.
2225 This indicates which servers are holding which shares. For immutable
2226 files, the shareid is an integer (the share number, from 0 to N-1).
2227 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2228 containing the sequence number, the roothash, and the share number."""
2231 """Return a string with a brief (one-line) summary of the results."""
2234 """Return a list of strings with more detailed results."""
2237 class ICheckAndRepairResults(Interface):
2238 """I contain the detailed results of a check/verify/repair operation.
2240 The IFilesystemNode.check()/verify()/repair() methods all return
2241 instances that provide ICheckAndRepairResults.
2244 def get_storage_index():
2245 """Return a string with the (binary) storage index."""
2247 def get_storage_index_string():
2248 """Return a string with the (printable) abbreviated storage index."""
2250 def get_repair_attempted():
2251 """Return a boolean, True if a repair was attempted. We might not
2252 attempt to repair the file because it was healthy, or healthy enough
2253 (i.e. some shares were missing but not enough to exceed some
2254 threshold), or because we don't know how to repair this object."""
2256 def get_repair_successful():
2257 """Return a boolean, True if repair was attempted and the file/dir
2258 was fully healthy afterwards. False if no repair was attempted or if
2259 a repair attempt failed."""
2261 def get_pre_repair_results():
2262 """Return an ICheckResults instance that describes the state of the
2263 file/dir before any repair was attempted."""
2265 def get_post_repair_results():
2266 """Return an ICheckResults instance that describes the state of the
2267 file/dir after any repair was attempted. If no repair was attempted,
2268 the pre-repair and post-repair results will be identical."""
2271 class IDeepCheckResults(Interface):
2272 """I contain the results of a deep-check operation.
2274 This is returned by a call to ICheckable.deep_check().
2277 def get_root_storage_index_string():
2278 """Return the storage index (abbreviated human-readable string) of
2279 the first object checked."""
2282 """Return a dictionary with the following keys::
2284 count-objects-checked: count of how many objects were checked
2285 count-objects-healthy: how many of those objects were completely
2287 count-objects-unhealthy: how many were damaged in some way
2288 count-objects-unrecoverable: how many were unrecoverable
2289 count-corrupt-shares: how many shares were found to have
2290 corruption, summed over all objects
2294 def get_corrupt_shares():
2295 """Return a set of (IServer, storage_index, sharenum) for all shares
2296 that were found to be corrupt. storage_index is binary."""
2298 def get_all_results():
2299 """Return a dictionary mapping pathname (a tuple of strings, ready to
2300 be slash-joined) to an ICheckResults instance, one for each object
2301 that was checked."""
2303 def get_results_for_storage_index(storage_index):
2304 """Retrive the ICheckResults instance for the given (binary)
2305 storage index. Raises KeyError if there are no results for that
2309 """Return a dictionary with the same keys as
2310 IDirectoryNode.deep_stats()."""
2313 class IDeepCheckAndRepairResults(Interface):
2314 """I contain the results of a deep-check-and-repair operation.
2316 This is returned by a call to ICheckable.deep_check_and_repair().
2319 def get_root_storage_index_string():
2320 """Return the storage index (abbreviated human-readable string) of
2321 the first object checked."""
2324 """Return a dictionary with the following keys::
2326 count-objects-checked: count of how many objects were checked
2327 count-objects-healthy-pre-repair: how many of those objects were
2328 completely healthy (before any
2330 count-objects-unhealthy-pre-repair: how many were damaged in
2332 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2333 count-objects-healthy-post-repair: how many of those objects were
2334 completely healthy (after any
2336 count-objects-unhealthy-post-repair: how many were damaged in
2338 count-objects-unrecoverable-post-repair: how many were
2340 count-repairs-attempted: repairs were attempted on this many
2341 objects. The count-repairs- keys will
2342 always be provided, however unless
2343 repair=true is present, they will all
2345 count-repairs-successful: how many repairs resulted in healthy
2347 count-repairs-unsuccessful: how many repairs resulted did not
2348 results in completely healthy objects
2349 count-corrupt-shares-pre-repair: how many shares were found to
2350 have corruption, summed over all
2351 objects examined (before any
2353 count-corrupt-shares-post-repair: how many shares were found to
2354 have corruption, summed over all
2355 objects examined (after any
2360 """Return a dictionary with the same keys as
2361 IDirectoryNode.deep_stats()."""
2363 def get_corrupt_shares():
2364 """Return a set of (IServer, storage_index, sharenum) for all shares
2365 that were found to be corrupt before any repair was attempted.
2366 storage_index is binary.
2368 def get_remaining_corrupt_shares():
2369 """Return a set of (IServer, storage_index, sharenum) for all shares
2370 that were found to be corrupt after any repair was completed.
2371 storage_index is binary. These are shares that need manual inspection
2372 and probably deletion.
2374 def get_all_results():
2375 """Return a dictionary mapping pathname (a tuple of strings, ready to
2376 be slash-joined) to an ICheckAndRepairResults instance, one for each
2377 object that was checked."""
2379 def get_results_for_storage_index(storage_index):
2380 """Retrive the ICheckAndRepairResults instance for the given (binary)
2381 storage index. Raises KeyError if there are no results for that
2385 class IRepairable(Interface):
2386 def repair(check_results):
2387 """Attempt to repair the given object. Returns a Deferred that fires
2388 with a IRepairResults object.
2390 I must be called with an object that implements ICheckResults, as
2391 proof that you have actually discovered a problem with this file. I
2392 will use the data in the checker results to guide the repair process,
2393 such as which servers provided bad data and should therefore be
2394 avoided. The ICheckResults object is inside the
2395 ICheckAndRepairResults object, which is returned by the
2396 ICheckable.check() method::
2398 d = filenode.check(repair=False)
2399 def _got_results(check_and_repair_results):
2400 check_results = check_and_repair_results.get_pre_repair_results()
2401 return filenode.repair(check_results)
2402 d.addCallback(_got_results)
2407 class IRepairResults(Interface):
2408 """I contain the results of a repair operation."""
2409 def get_successful():
2410 """Returns a boolean: True if the repair made the file healthy, False
2411 if not. Repair failure generally indicates a file that has been
2412 damaged beyond repair."""
2415 class IClient(Interface):
2416 def upload(uploadable):
2417 """Upload some data into a CHK, get back the UploadResults for it.
2418 @param uploadable: something that implements IUploadable
2419 @return: a Deferred that fires with the UploadResults instance.
2420 To get the URI for this file, use results.uri .
2423 def create_mutable_file(contents=""):
2424 """Create a new mutable file (with initial) contents, get back the
2427 @param contents: (bytestring, callable, or None): this provides the
2428 initial contents of the mutable file. If 'contents' is a bytestring,
2429 it will be used as-is. If 'contents' is a callable, it will be
2430 invoked with the new MutableFileNode instance and is expected to
2431 return a bytestring with the initial contents of the file (the
2432 callable can use node.get_writekey() to decide how to encrypt the
2433 initial contents, e.g. for a brand new dirnode with initial
2434 children). contents=None is equivalent to an empty string. Using
2435 content_maker= is more efficient than creating a mutable file and
2436 setting its contents in two separate operations.
2438 @return: a Deferred that fires with an IMutableFileNode instance.
2441 def create_dirnode(initial_children={}):
2442 """Create a new unattached dirnode, possibly with initial children.
2444 @param initial_children: dict with keys that are unicode child names,
2445 and values that are (childnode, metadata) tuples.
2447 @return: a Deferred that fires with the new IDirectoryNode instance.
2450 def create_node_from_uri(uri, rouri):
2451 """Create a new IFilesystemNode instance from the uri, synchronously.
2452 @param uri: a string or IURI-providing instance, or None. This could
2453 be for a LiteralFileNode, a CHK file node, a mutable file
2454 node, or a directory node
2455 @param rouri: a string or IURI-providing instance, or None. If the
2456 main uri is None, I will use the rouri instead. If I
2457 recognize the format of the main uri, I will ignore the
2458 rouri (because it can be derived from the writecap).
2460 @return: an instance that provides IFilesystemNode (or more usefully
2461 one of its subclasses). File-specifying URIs will result in
2462 IFileNode-providing instances, like ImmutableFileNode,
2463 LiteralFileNode, or MutableFileNode. Directory-specifying
2464 URIs will result in IDirectoryNode-providing instances, like
2469 class INodeMaker(Interface):
2470 """The NodeMaker is used to create IFilesystemNode instances. It can
2471 accept a filecap/dircap string and return the node right away. It can
2472 also create new nodes (i.e. upload a file, or create a mutable file)
2473 asynchronously. Once you have one of these nodes, you can use other
2474 methods to determine whether it is a file or directory, and to download
2475 or modify its contents.
2477 The NodeMaker encapsulates all the authorities that these
2478 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2479 Tahoe process will typically have a single NodeMaker, but unit tests may
2480 create simplified/mocked forms for testing purposes.
2483 def create_from_cap(writecap, readcap=None, deep_immutable=False, name=u"<unknown name>"):
2484 """I create an IFilesystemNode from the given writecap/readcap. I can
2485 only provide nodes for existing file/directory objects: use my other
2486 methods to create new objects. I return synchronously."""
2488 def create_mutable_file(contents=None, keysize=None):
2489 """I create a new mutable file, and return a Deferred that will fire
2490 with the IMutableFileNode instance when it is ready. If contents= is
2491 provided (a bytestring), it will be used as the initial contents of
2492 the new file, otherwise the file will contain zero bytes. keysize= is
2493 for use by unit tests, to create mutable files that are smaller than
2496 def create_new_mutable_directory(initial_children={}):
2497 """I create a new mutable directory, and return a Deferred that will
2498 fire with the IDirectoryNode instance when it is ready. If
2499 initial_children= is provided (a dict mapping unicode child name to
2500 (childnode, metadata_dict) tuples), the directory will be populated
2501 with those children, otherwise it will be empty."""
2504 class IClientStatus(Interface):
2505 def list_all_uploads():
2506 """Return a list of uploader objects, one for each upload that
2507 currently has an object available (tracked with weakrefs). This is
2508 intended for debugging purposes."""
2510 def list_active_uploads():
2511 """Return a list of active IUploadStatus objects."""
2513 def list_recent_uploads():
2514 """Return a list of IUploadStatus objects for the most recently
2517 def list_all_downloads():
2518 """Return a list of downloader objects, one for each download that
2519 currently has an object available (tracked with weakrefs). This is
2520 intended for debugging purposes."""
2522 def list_active_downloads():
2523 """Return a list of active IDownloadStatus objects."""
2525 def list_recent_downloads():
2526 """Return a list of IDownloadStatus objects for the most recently
2527 started downloads."""
2530 class IUploadStatus(Interface):
2532 """Return a timestamp (float with seconds since epoch) indicating
2533 when the operation was started."""
2535 def get_storage_index():
2536 """Return a string with the (binary) storage index in use on this
2537 upload. Returns None if the storage index has not yet been
2541 """Return an integer with the number of bytes that will eventually
2542 be uploaded for this file. Returns None if the size is not yet known.
2545 """Return True if this upload is using a Helper, False if not."""
2548 """Return a string describing the current state of the upload
2552 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2553 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2554 made towards hashing the file to determine a CHK encryption key: if
2555 non-convergent encryption is in use, this will be trivial, otherwise
2556 the whole file must be hashed. 'ciphertext' describes how much of the
2557 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2558 uploads. 'encode_and_push' describes how much of the encode-and-push
2559 process has finished: for helper uploads this is dependent upon the
2560 helper providing progress reports. It might be reasonable to add all
2561 three numbers and report the sum to the user."""
2564 """Return True if the upload is currently active, False if not."""
2567 """Return an instance of UploadResults (which contains timing and
2568 sharemap information). Might return None if the upload is not yet
2572 """Each upload status gets a unique number: this method returns that
2573 number. This provides a handle to this particular upload, so a web
2574 page can generate a suitable hyperlink."""
2577 class IDownloadStatus(Interface):
2579 """Return a timestamp (float with seconds since epoch) indicating
2580 when the operation was started."""
2582 def get_storage_index():
2583 """Return a string with the (binary) storage index in use on this
2584 download. This may be None if there is no storage index (i.e. LIT
2588 """Return an integer with the number of bytes that will eventually be
2589 retrieved for this file. Returns None if the size is not yet known.
2593 """Return True if this download is using a Helper, False if not."""
2596 """Return a string describing the current state of the download
2600 """Returns a float (from 0.0 to 1.0) describing the amount of the
2601 download that has completed. This value will remain at 0.0 until the
2602 first byte of plaintext is pushed to the download target."""
2605 """Return True if the download is currently active, False if not."""
2608 """Each download status gets a unique number: this method returns
2609 that number. This provides a handle to this particular download, so a
2610 web page can generate a suitable hyperlink."""
2613 class IServermapUpdaterStatus(Interface):
2616 class IPublishStatus(Interface):
2619 class IRetrieveStatus(Interface):
2623 class NotCapableError(Exception):
2624 """You have tried to write to a read-only node."""
2626 class BadWriteEnablerError(Exception):
2630 class RIControlClient(RemoteInterface):
2631 def wait_for_client_connections(num_clients=int):
2632 """Do not return until we have connections to at least NUM_CLIENTS
2638 def upload_random_data_from_file(size=int, convergence=str):
2641 def download_to_tempfile_and_delete(uri=str):
2644 def get_memory_usage():
2645 """Return a dict describes the amount of memory currently in use. The
2646 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2647 measuring memory consupmtion in bytes."""
2648 return DictOf(str, int)
2650 def speed_test(count=int, size=int, mutable=Any()):
2651 """Write 'count' tempfiles to disk, all of the given size. Measure
2652 how long (in seconds) it takes to upload them all to the servers.
2653 Then measure how long it takes to download all of them. If 'mutable'
2654 is 'create', time creation of mutable files. If 'mutable' is
2655 'upload', then time access to the same mutable file instead of
2658 Returns a tuple of (upload_time, download_time).
2660 return (float, float)
2662 def measure_peer_response_time():
2663 """Send a short message to each connected peer, and measure the time
2664 it takes for them to respond to it. This is a rough measure of the
2665 application-level round trip time.
2667 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2670 return DictOf(str, float)
2673 UploadResults = Any() #DictOf(str, str)
2676 class RIEncryptedUploadable(RemoteInterface):
2677 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2682 def get_all_encoding_parameters():
2683 return (int, int, int, long)
2685 def read_encrypted(offset=Offset, length=ReadSize):
2692 class RICHKUploadHelper(RemoteInterface):
2693 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2697 Return a dictionary of version information.
2699 return DictOf(str, Any())
2701 def upload(reader=RIEncryptedUploadable):
2702 return UploadResults
2705 class RIHelper(RemoteInterface):
2706 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2710 Return a dictionary of version information.
2712 return DictOf(str, Any())
2714 def upload_chk(si=StorageIndex):
2715 """See if a file with a given storage index needs uploading. The
2716 helper will ask the appropriate storage servers to see if the file
2717 has already been uploaded. If so, the helper will return a set of
2718 'upload results' that includes whatever hashes are needed to build
2719 the read-cap, and perhaps a truncated sharemap.
2721 If the file has not yet been uploaded (or if it was only partially
2722 uploaded), the helper will return an empty upload-results dictionary
2723 and also an RICHKUploadHelper object that will take care of the
2724 upload process. The client should call upload() on this object and
2725 pass it a reference to an RIEncryptedUploadable object that will
2726 provide ciphertext. When the upload is finished, the upload() method
2727 will finish and return the upload results.
2729 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2732 class RIStatsProvider(RemoteInterface):
2733 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2735 Provides access to statistics and monitoring information.
2740 returns a dictionary containing 'counters' and 'stats', each a
2741 dictionary with string counter/stat name keys, and numeric or None values.
2742 counters are monotonically increasing measures of work done, and
2743 stats are instantaneous measures (potentially time averaged
2746 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2749 class RIStatsGatherer(RemoteInterface):
2750 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2752 Provides a monitoring service for centralised collection of stats
2755 def provide(provider=RIStatsProvider, nickname=str):
2757 @param provider: a stats collector instance that should be polled
2758 periodically by the gatherer to collect stats.
2759 @param nickname: a name useful to identify the provided client
2764 class IStatsProducer(Interface):
2767 returns a dictionary, with str keys representing the names of stats
2768 to be monitored, and numeric values.
2771 class RIKeyGenerator(RemoteInterface):
2772 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2774 Provides a service offering to make RSA key pairs.
2777 def get_rsa_key_pair(key_size=int):
2779 @param key_size: the size of the signature key.
2780 @return: tuple(verifying_key, signing_key)
2782 return TupleOf(str, str)
2785 class FileTooLargeError(Exception):
2789 class IValidatedThingProxy(Interface):
2791 """ Acquire a thing and validate it. Return a deferred that is
2792 eventually fired with self if the thing is valid or errbacked if it
2793 can't be acquired or validated."""
2796 class InsufficientVersionError(Exception):
2797 def __init__(self, needed, got):
2798 self.needed = needed
2802 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2805 class EmptyPathnameComponentError(Exception):
2806 """The webapi disallows empty pathname components."""