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 last_connect_time: when we last established a connection
421 last_loss_time: when we last lost a connection
423 version: the peer's version, from the most recent connection
424 oldest_supported: the peer's oldest supported version, same
426 rref: the RemoteReference, if connected, otherwise None
427 remote_host: the IAddress, if connected, otherwise None
429 This method is intended for monitoring interfaces, such as a web page
430 that describes connecting and connected peers.
433 def get_all_peerids():
434 """Return a frozenset of all peerids to whom we have a connection (to
435 one or more services) established. Mostly useful for unit tests."""
437 def get_all_connections_for(service_name):
438 """Return a frozenset of (nodeid, service_name, rref) tuples, one
439 for each active connection that provides the given SERVICE_NAME."""
441 def get_permuted_peers(service_name, key):
442 """Returns an ordered list of (peerid, rref) tuples, selecting from
443 the connections that provide SERVICE_NAME, using a hash-based
444 permutation keyed by KEY. This randomizes the service list in a
445 repeatable way, to distribute load over many peers.
449 class IDisplayableServer(Interface):
460 class IServer(IDisplayableServer):
461 """I live in the client, and represent a single server."""
462 def start_connecting(tub, trigger_cb):
466 """Once a server is connected, I return a RemoteReference.
467 Before a server is connected for the first time, I return None.
469 Note that the rref I return will start producing DeadReferenceErrors
470 once the connection is lost.
474 class IMutableSlotWriter(Interface):
476 The interface for a writer around a mutable slot on a remote server.
478 def set_checkstring(seqnum_or_checkstring, root_hash=None, salt=None):
480 Set the checkstring that I will pass to the remote server when
483 @param checkstring A packed checkstring to use.
485 Note that implementations can differ in which semantics they
486 wish to support for set_checkstring -- they can, for example,
487 build the checkstring themselves from its constituents, or
491 def get_checkstring():
493 Get the checkstring that I think currently exists on the remote
497 def put_block(data, segnum, salt):
499 Add a block and salt to the share.
502 def put_encprivkey(encprivkey):
504 Add the encrypted private key to the share.
507 def put_blockhashes(blockhashes):
509 @param blockhashes=list
510 Add the block hash tree to the share.
513 def put_sharehashes(sharehashes):
515 @param sharehashes=dict
516 Add the share hash chain to the share.
521 Return the part of the share that needs to be signed.
524 def put_signature(signature):
526 Add the signature to the share.
529 def put_verification_key(verification_key):
531 Add the verification key to the share.
534 def finish_publishing():
536 Do anything necessary to finish writing the share to a remote
537 server. I require that no further publishing needs to take place
538 after this method has been called.
542 class IURI(Interface):
543 def init_from_string(uri):
544 """Accept a string (as created by my to_string() method) and populate
545 this instance with its data. I am not normally called directly,
546 please use the module-level uri.from_string() function to convert
547 arbitrary URI strings into IURI-providing instances."""
550 """Return False if this URI be used to modify the data. Return True
551 if this URI cannot be used to modify the data."""
554 """Return True if the data can be modified by *somebody* (perhaps
555 someone who has a more powerful URI than this one)."""
557 # TODO: rename to get_read_cap()
559 """Return another IURI instance that represents a read-only form of
560 this one. If is_readonly() is True, this returns self."""
562 def get_verify_cap():
563 """Return an instance that provides IVerifierURI, which can be used
564 to check on the availability of the file or directory, without
565 providing enough capabilities to actually read or modify the
566 contents. This may return None if the file does not need checking or
567 verification (e.g. LIT URIs).
571 """Return a string of printable ASCII characters, suitable for
572 passing into init_from_string."""
575 class IVerifierURI(Interface, IURI):
576 def init_from_string(uri):
577 """Accept a string (as created by my to_string() method) and populate
578 this instance with its data. I am not normally called directly,
579 please use the module-level uri.from_string() function to convert
580 arbitrary URI strings into IURI-providing instances."""
583 """Return a string of printable ASCII characters, suitable for
584 passing into init_from_string."""
587 class IDirnodeURI(Interface):
588 """I am a URI that represents a dirnode."""
591 class IFileURI(Interface):
592 """I am a URI that represents a filenode."""
594 """Return the length (in bytes) of the file that I represent."""
597 class IImmutableFileURI(IFileURI):
600 class IMutableFileURI(Interface):
603 class IDirectoryURI(Interface):
606 class IReadonlyDirectoryURI(Interface):
610 class CapConstraintError(Exception):
611 """A constraint on a cap was violated."""
613 class MustBeDeepImmutableError(CapConstraintError):
614 """Mutable children cannot be added to an immutable directory.
615 Also, caps obtained from an immutable directory can trigger this error
616 if they are later found to refer to a mutable object and then used."""
618 class MustBeReadonlyError(CapConstraintError):
619 """Known write caps cannot be specified in a ro_uri field. Also,
620 caps obtained from a ro_uri field can trigger this error if they
621 are later found to be write caps and then used."""
623 class MustNotBeUnknownRWError(CapConstraintError):
624 """Cannot add an unknown child cap specified in a rw_uri field."""
627 class IReadable(Interface):
628 """I represent a readable object -- either an immutable file, or a
629 specific version of a mutable file.
633 """Return True if this reference provides mutable access to the given
634 file or directory (i.e. if you can modify it), or False if not. Note
635 that even if this reference is read-only, someone else may hold a
636 read-write reference to it.
638 For an IReadable returned by get_best_readable_version(), this will
639 always return True, but for instances of subinterfaces such as
640 IMutableFileVersion, it may return False."""
643 """Return True if this file or directory is mutable (by *somebody*,
644 not necessarily you), False if it is is immutable. Note that a file
645 might be mutable overall, but your reference to it might be
646 read-only. On the other hand, all references to an immutable file
647 will be read-only; there are no read-write references to an immutable
650 def get_storage_index():
651 """Return the storage index of the file."""
654 """Return the length (in bytes) of this readable object."""
656 def download_to_data():
657 """Download all of the file contents. I return a Deferred that fires
658 with the contents as a byte string."""
660 def read(consumer, offset=0, size=None):
661 """Download a portion (possibly all) of the file's contents, making
662 them available to the given IConsumer. Return a Deferred that fires
663 (with the consumer) when the consumer is unregistered (either because
664 the last byte has been given to it, or because the consumer threw an
665 exception during write(), possibly because it no longer wants to
666 receive data). The portion downloaded will start at 'offset' and
667 contain 'size' bytes (or the remainder of the file if size==None).
669 The consumer will be used in non-streaming mode: an IPullProducer
670 will be attached to it.
672 The consumer will not receive data right away: several network trips
673 must occur first. The order of events will be::
675 consumer.registerProducer(p, streaming)
676 (if streaming == False)::
677 consumer does p.resumeProducing()
679 consumer does p.resumeProducing()
680 consumer.write(data).. (repeat until all data is written)
681 consumer.unregisterProducer()
682 deferred.callback(consumer)
684 If a download error occurs, or an exception is raised by
685 consumer.registerProducer() or consumer.write(), I will call
686 consumer.unregisterProducer() and then deliver the exception via
687 deferred.errback(). To cancel the download, the consumer should call
688 p.stopProducing(), which will result in an exception being delivered
689 via deferred.errback().
691 See src/allmydata/util/consumer.py for an example of a simple
692 download-to-memory consumer.
696 class IWriteable(Interface):
698 I define methods that callers can use to update SDMF and MDMF
699 mutable files on a Tahoe-LAFS grid.
701 # XXX: For the moment, we have only this. It is possible that we
702 # want to move overwrite() and modify() in here too.
703 def update(data, offset):
705 I write the data from my data argument to the MDMF file,
706 starting at offset. I continue writing data until my data
707 argument is exhausted, appending data to the file as necessary.
709 # assert IMutableUploadable.providedBy(data)
710 # to append data: offset=node.get_size_of_best_version()
711 # do we want to support compacting MDMF?
712 # for an MDMF file, this can be done with O(data.get_size())
713 # memory. For an SDMF file, any modification takes
714 # O(node.get_size_of_best_version()).
717 class IMutableFileVersion(IReadable):
718 """I provide access to a particular version of a mutable file. The
719 access is read/write if I was obtained from a filenode derived from
720 a write cap, or read-only if the filenode was derived from a read cap.
723 def get_sequence_number():
724 """Return the sequence number of this version."""
727 """Return the IMutableFileServerMap instance that was used to create
732 """Return this filenode's writekey, or None if the node does not have
733 write-capability. This may be used to assist with data structures
734 that need to make certain data available only to writers, such as the
735 read-write child caps in dirnodes. The recommended process is to have
736 reader-visible data be submitted to the filenode in the clear (where
737 it will be encrypted by the filenode using the readkey), but encrypt
738 writer-visible data using this writekey.
741 def overwrite(new_contents):
742 """Replace the contents of the mutable file, provided that no other
743 node has published (or is attempting to publish, concurrently) a
744 newer version of the file than this one.
746 I will avoid modifying any share that is different than the version
747 given by get_sequence_number(). However, if another node is writing
748 to the file at the same time as me, I may manage to update some shares
749 while they update others. If I see any evidence of this, I will signal
750 UncoordinatedWriteError, and the file will be left in an inconsistent
751 state (possibly the version you provided, possibly the old version,
752 possibly somebody else's version, and possibly a mix of shares from
755 The recommended response to UncoordinatedWriteError is to either
756 return it to the caller (since they failed to coordinate their
757 writes), or to attempt some sort of recovery. It may be sufficient to
758 wait a random interval (with exponential backoff) and repeat your
759 operation. If I do not signal UncoordinatedWriteError, then I was
760 able to write the new version without incident.
762 I return a Deferred that fires (with a PublishStatus object) when the
763 update has completed.
766 def modify(modifier_cb):
767 """Modify the contents of the file, by downloading this version,
768 applying the modifier function (or bound method), then uploading
769 the new version. This will succeed as long as no other node
770 publishes a version between the download and the upload.
771 I return a Deferred that fires (with a PublishStatus object) when
772 the update is complete.
774 The modifier callable will be given three arguments: a string (with
775 the old contents), a 'first_time' boolean, and a servermap. As with
776 download_to_data(), the old contents will be from this version,
777 but the modifier can use the servermap to make other decisions
778 (such as refusing to apply the delta if there are multiple parallel
779 versions, or if there is evidence of a newer unrecoverable version).
780 'first_time' will be True the first time the modifier is called,
781 and False on any subsequent calls.
783 The callable should return a string with the new contents. The
784 callable must be prepared to be called multiple times, and must
785 examine the input string to see if the change that it wants to make
786 is already present in the old version. If it does not need to make
787 any changes, it can either return None, or return its input string.
789 If the modifier raises an exception, it will be returned in the
794 # The hierarchy looks like this:
801 class IFilesystemNode(Interface):
803 """Return the strongest 'cap instance' associated with this node.
804 (writecap for writeable-mutable files/directories, readcap for
805 immutable or readonly-mutable files/directories). To convert this
806 into a string, call .to_string() on the result."""
809 """Return a readonly cap instance for this node. For immutable or
810 readonly nodes, get_cap() and get_readcap() return the same thing."""
812 def get_repair_cap():
813 """Return an IURI instance that can be used to repair the file, or
814 None if this node cannot be repaired (either because it is not
815 distributed, like a LIT file, or because the node does not represent
816 sufficient authority to create a repair-cap, like a read-only RSA
817 mutable file node [which cannot create the correct write-enablers]).
820 def get_verify_cap():
821 """Return an IVerifierURI instance that represents the
822 'verifiy/refresh capability' for this node. The holder of this
823 capability will be able to renew the lease for this node, protecting
824 it from garbage-collection. They will also be able to ask a server if
825 it holds a share for the file or directory.
829 """Return the URI string corresponding to the strongest cap associated
830 with this node. If this node is read-only, the URI will only offer
831 read-only access. If this node is read-write, the URI will offer
834 If you have read-write access to a node and wish to share merely
835 read-only access with others, use get_readonly_uri().
839 """Return the URI string that can be used by others to get write
840 access to this node, if it is writeable. If this is a read-only node,
843 def get_readonly_uri():
844 """Return the URI string that can be used by others to get read-only
845 access to this node. The result is a read-only URI, regardless of
846 whether this node is read-only or read-write.
848 If you have merely read-only access to this node, get_readonly_uri()
849 will return the same thing as get_uri().
852 def get_storage_index():
853 """Return a string with the (binary) storage index in use on this
854 download. This may be None if there is no storage index (i.e. LIT
855 files and directories)."""
858 """Return True if this reference provides mutable access to the given
859 file or directory (i.e. if you can modify it), or False if not. Note
860 that even if this reference is read-only, someone else may hold a
861 read-write reference to it."""
864 """Return True if this file or directory is mutable (by *somebody*,
865 not necessarily you), False if it is is immutable. Note that a file
866 might be mutable overall, but your reference to it might be
867 read-only. On the other hand, all references to an immutable file
868 will be read-only; there are no read-write references to an immutable
873 """Return True if this is an unknown node."""
875 def is_allowed_in_immutable_directory():
876 """Return True if this node is allowed as a child of a deep-immutable
877 directory. This is true if either the node is of a known-immutable type,
878 or it is unknown and read-only.
882 """Raise any error associated with this node."""
884 # XXX: These may not be appropriate outside the context of an IReadable.
886 """Return the length (in bytes) of the data this node represents. For
887 directory nodes, I return the size of the backing store. I return
888 synchronously and do not consult the network, so for mutable objects,
889 I will return the most recently observed size for the object, or None
890 if I don't remember a size. Use get_current_size, which returns a
891 Deferred, if you want more up-to-date information."""
893 def get_current_size():
894 """I return a Deferred that fires with the length (in bytes) of the
895 data this node represents.
899 class IFileNode(IFilesystemNode):
900 """I am a node that represents a file: a sequence of bytes. I am not a
901 container, like IDirectoryNode."""
902 def get_best_readable_version():
903 """Return a Deferred that fires with an IReadable for the 'best'
904 available version of the file. The IReadable provides only read
905 access, even if this filenode was derived from a write cap.
907 For an immutable file, there is only one version. For a mutable
908 file, the 'best' version is the recoverable version with the
909 highest sequence number. If no uncoordinated writes have occurred,
910 and if enough shares are available, then this will be the most
911 recent version that has been uploaded. If no version is recoverable,
912 the Deferred will errback with an UnrecoverableFileError.
915 def download_best_version():
916 """Download the contents of the version that would be returned
917 by get_best_readable_version(). This is equivalent to calling
918 download_to_data() on the IReadable given by that method.
920 I return a Deferred that fires with a byte string when the file
921 has been fully downloaded. To support streaming download, use
922 the 'read' method of IReadable. If no version is recoverable,
923 the Deferred will errback with an UnrecoverableFileError.
926 def get_size_of_best_version():
927 """Find the size of the version that would be returned by
928 get_best_readable_version().
930 I return a Deferred that fires with an integer. If no version
931 is recoverable, the Deferred will errback with an
932 UnrecoverableFileError.
936 class IImmutableFileNode(IFileNode, IReadable):
937 """I am a node representing an immutable file. Immutable files have
941 class IMutableFileNode(IFileNode):
942 """I provide access to a 'mutable file', which retains its identity
943 regardless of what contents are put in it.
945 The consistency-vs-availability problem means that there might be
946 multiple versions of a file present in the grid, some of which might be
947 unrecoverable (i.e. have fewer than 'k' shares). These versions are
948 loosely ordered: each has a sequence number and a hash, and any version
949 with seqnum=N was uploaded by a node that has seen at least one version
952 The 'servermap' (an instance of IMutableFileServerMap) is used to
953 describe the versions that are known to be present in the grid, and which
954 servers are hosting their shares. It is used to represent the 'state of
955 the world', and is used for this purpose by my test-and-set operations.
956 Downloading the contents of the mutable file will also return a
957 servermap. Uploading a new version into the mutable file requires a
958 servermap as input, and the semantics of the replace operation is
959 'replace the file with my new version if it looks like nobody else has
960 changed the file since my previous download'. Because the file is
961 distributed, this is not a perfect test-and-set operation, but it will do
962 its best. If the replace process sees evidence of a simultaneous write,
963 it will signal an UncoordinatedWriteError, so that the caller can take
967 Most readers will want to use the 'best' current version of the file, and
968 should use my 'download_best_version()' method.
970 To unconditionally replace the file, callers should use overwrite(). This
971 is the mode that user-visible mutable files will probably use.
973 To apply some delta to the file, call modify() with a callable modifier
974 function that can apply the modification that you want to make. This is
975 the mode that dirnodes will use, since most directory modification
976 operations can be expressed in terms of deltas to the directory state.
979 Three methods are available for users who need to perform more complex
980 operations. The first is get_servermap(), which returns an up-to-date
981 servermap using a specified mode. The second is download_version(), which
982 downloads a specific version (not necessarily the 'best' one). The third
983 is 'upload', which accepts new contents and a servermap (which must have
984 been updated with MODE_WRITE). The upload method will attempt to apply
985 the new contents as long as no other node has modified the file since the
986 servermap was updated. This might be useful to a caller who wants to
987 merge multiple versions into a single new one.
989 Note that each time the servermap is updated, a specific 'mode' is used,
990 which determines how many peers are queried. To use a servermap for my
991 replace() method, that servermap must have been updated in MODE_WRITE.
992 These modes are defined in allmydata.mutable.common, and consist of
993 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
994 allmydata/mutable/servermap.py for details about the differences.
996 Mutable files are currently limited in size (about 3.5MB max) and can
997 only be retrieved and updated all-at-once, as a single big string. Future
998 versions of our mutable files will remove this restriction.
1000 def get_best_mutable_version():
1001 """Return a Deferred that fires with an IMutableFileVersion for
1002 the 'best' available version of the file. The best version is
1003 the recoverable version with the highest sequence number. If no
1004 uncoordinated writes have occurred, and if enough shares are
1005 available, then this will be the most recent version that has
1008 If no version is recoverable, the Deferred will errback with an
1009 UnrecoverableFileError.
1012 def overwrite(new_contents):
1013 """Unconditionally replace the contents of the mutable file with new
1014 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1015 is only appropriate to use when the new contents of the file are
1016 completely unrelated to the old ones, and you do not care about other
1019 I return a Deferred that fires (with a PublishStatus object) when the
1020 update has completed.
1023 def modify(modifier_cb):
1024 """Modify the contents of the file, by downloading the current
1025 version, applying the modifier function (or bound method), then
1026 uploading the new version. I return a Deferred that fires (with a
1027 PublishStatus object) when the update is complete.
1029 The modifier callable will be given three arguments: a string (with
1030 the old contents), a 'first_time' boolean, and a servermap. As with
1031 download_best_version(), the old contents will be from the best
1032 recoverable version, but the modifier can use the servermap to make
1033 other decisions (such as refusing to apply the delta if there are
1034 multiple parallel versions, or if there is evidence of a newer
1035 unrecoverable version). 'first_time' will be True the first time the
1036 modifier is called, and False on any subsequent calls.
1038 The callable should return a string with the new contents. The
1039 callable must be prepared to be called multiple times, and must
1040 examine the input string to see if the change that it wants to make
1041 is already present in the old version. If it does not need to make
1042 any changes, it can either return None, or return its input string.
1044 If the modifier raises an exception, it will be returned in the
1048 def get_servermap(mode):
1049 """Return a Deferred that fires with an IMutableFileServerMap
1050 instance, updated using the given mode.
1053 def download_version(servermap, version):
1054 """Download a specific version of the file, using the servermap
1055 as a guide to where the shares are located.
1057 I return a Deferred that fires with the requested contents, or
1058 errbacks with UnrecoverableFileError. Note that a servermap that was
1059 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1060 all versions (those modes stop querying servers as soon as they can
1061 fulfil their goals), so you may want to use MODE_CHECK (which checks
1062 everything) to get increased visibility.
1065 def upload(new_contents, servermap):
1066 """Replace the contents of the file with new ones. This requires a
1067 servermap that was previously updated with MODE_WRITE.
1069 I attempt to provide test-and-set semantics, in that I will avoid
1070 modifying any share that is different than the version I saw in the
1071 servermap. However, if another node is writing to the file at the
1072 same time as me, I may manage to update some shares while they update
1073 others. If I see any evidence of this, I will signal
1074 UncoordinatedWriteError, and the file will be left in an inconsistent
1075 state (possibly the version you provided, possibly the old version,
1076 possibly somebody else's version, and possibly a mix of shares from
1079 The recommended response to UncoordinatedWriteError is to either
1080 return it to the caller (since they failed to coordinate their
1081 writes), or to attempt some sort of recovery. It may be sufficient to
1082 wait a random interval (with exponential backoff) and repeat your
1083 operation. If I do not signal UncoordinatedWriteError, then I was
1084 able to write the new version without incident.
1086 I return a Deferred that fires (with a PublishStatus object) when the
1087 publish has completed. I will update the servermap in-place with the
1088 location of all new shares.
1092 """Return this filenode's writekey, or None if the node does not have
1093 write-capability. This may be used to assist with data structures
1094 that need to make certain data available only to writers, such as the
1095 read-write child caps in dirnodes. The recommended process is to have
1096 reader-visible data be submitted to the filenode in the clear (where
1097 it will be encrypted by the filenode using the readkey), but encrypt
1098 writer-visible data using this writekey.
1102 """Returns the mutable file protocol version."""
1105 class NotEnoughSharesError(Exception):
1106 """Download was unable to get enough shares"""
1108 class NoSharesError(Exception):
1109 """Download was unable to get any shares at all."""
1111 class DownloadStopped(Exception):
1114 class UploadUnhappinessError(Exception):
1115 """Upload was unable to satisfy 'servers_of_happiness'"""
1117 class UnableToFetchCriticalDownloadDataError(Exception):
1118 """I was unable to fetch some piece of critical data that is supposed to
1119 be identically present in all shares."""
1121 class NoServersError(Exception):
1122 """Upload wasn't given any servers to work with, usually indicating a
1123 network or Introducer problem."""
1125 class ExistingChildError(Exception):
1126 """A directory node was asked to add or replace a child that already
1127 exists, and overwrite= was set to False."""
1129 class NoSuchChildError(Exception):
1130 """A directory node was asked to fetch a child that does not exist."""
1132 # avoid UnicodeEncodeErrors when converting to str
1133 return self.__repr__()
1135 class ChildOfWrongTypeError(Exception):
1136 """An operation was attempted on a child of the wrong type (file or directory)."""
1139 class IDirectoryNode(IFilesystemNode):
1140 """I represent a filesystem node that is a container, with a
1141 name-to-child mapping, holding the tahoe equivalent of a directory. All
1142 child names are unicode strings, and all children are some sort of
1143 IFilesystemNode (a file, subdirectory, or unknown node).
1148 The dirnode ('1') URI returned by this method can be used in
1149 set_uri() on a different directory ('2') to 'mount' a reference to
1150 this directory ('1') under the other ('2'). This URI is just a
1151 string, so it can be passed around through email or other out-of-band
1155 def get_readonly_uri():
1157 The dirnode ('1') URI returned by this method can be used in
1158 set_uri() on a different directory ('2') to 'mount' a reference to
1159 this directory ('1') under the other ('2'). This URI is just a
1160 string, so it can be passed around through email or other out-of-band
1165 """I return a Deferred that fires with a dictionary mapping child
1166 name (a unicode string) to (node, metadata_dict) tuples, in which
1167 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1170 def has_child(name):
1171 """I return a Deferred that fires with a boolean, True if there
1172 exists a child of the given name, False if not. The child name must
1173 be a unicode string."""
1176 """I return a Deferred that fires with a specific named child node,
1177 which is an IFilesystemNode. The child name must be a unicode string.
1178 I raise NoSuchChildError if I do not have a child by that name."""
1180 def get_metadata_for(name):
1181 """I return a Deferred that fires with the metadata dictionary for
1182 a specific named child node. The child name must be a unicode string.
1183 This metadata is stored in the *edge*, not in the child, so it is
1184 attached to the parent dirnode rather than the child node.
1185 I raise NoSuchChildError if I do not have a child by that name."""
1187 def set_metadata_for(name, metadata):
1188 """I replace any existing metadata for the named child with the new
1189 metadata. The child name must be a unicode string. This metadata is
1190 stored in the *edge*, not in the child, so it is attached to the
1191 parent dirnode rather than the child node. I return a Deferred
1192 (that fires with this dirnode) when the operation is complete.
1193 I raise NoSuchChildError if I do not have a child by that name."""
1195 def get_child_at_path(path):
1196 """Transform a child path into an IFilesystemNode.
1198 I perform a recursive series of 'get' operations to find the named
1199 descendant node. I return a Deferred that fires with the node, or
1200 errbacks with NoSuchChildError if the node could not be found.
1202 The path can be either a single string (slash-separated) or a list of
1203 path-name elements. All elements must be unicode strings.
1206 def get_child_and_metadata_at_path(path):
1207 """Transform a child path into an IFilesystemNode and metadata.
1209 I am like get_child_at_path(), but my Deferred fires with a tuple of
1210 (node, metadata). The metadata comes from the last edge. If the path
1211 is empty, the metadata will be an empty dictionary.
1214 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1215 """I add a child (by writecap+readcap) at the specific name. I return
1216 a Deferred that fires when the operation finishes. If overwrite= is
1217 True, I will replace any existing child of the same name, otherwise
1218 an existing child will cause me to return ExistingChildError. The
1219 child name must be a unicode string.
1221 The child caps could be for a file, or for a directory. If you have
1222 both the writecap and readcap, you should provide both arguments.
1223 If you have only one cap and don't know whether it is read-only,
1224 provide it as the writecap argument and leave the readcap as None.
1225 If you have only one cap that is known to be read-only, provide it
1226 as the readcap argument and leave the writecap as None.
1227 The filecaps are typically obtained from an IFilesystemNode with
1228 get_uri() and get_readonly_uri().
1230 If metadata= is provided, I will use it as the metadata for the named
1231 edge. This will replace any existing metadata. If metadata= is left
1232 as the default value of None, I will set ['mtime'] to the current
1233 time, and I will set ['ctime'] to the current time if there was not
1234 already a child by this name present. This roughly matches the
1235 ctime/mtime semantics of traditional filesystems. See the
1236 "About the metadata" section of webapi.txt for futher information.
1238 If this directory node is read-only, the Deferred will errback with a
1239 NotWriteableError."""
1241 def set_children(entries, overwrite=True):
1242 """Add multiple children (by writecap+readcap) to a directory node.
1243 Takes a dictionary, with childname as keys and (writecap, readcap)
1244 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1245 a Deferred that fires (with this dirnode) when the operation
1246 finishes. This is equivalent to calling set_uri() multiple times, but
1247 is much more efficient. All child names must be unicode strings.
1250 def set_node(name, child, metadata=None, overwrite=True):
1251 """I add a child at the specific name. I return a Deferred that fires
1252 when the operation finishes. This Deferred will fire with the child
1253 node that was just added. I will replace any existing child of the
1254 same name. The child name must be a unicode string. The 'child'
1255 instance must be an instance providing IFilesystemNode.
1257 If metadata= is provided, I will use it as the metadata for the named
1258 edge. This will replace any existing metadata. If metadata= is left
1259 as the default value of None, I will set ['mtime'] to the current
1260 time, and I will set ['ctime'] to the current time if there was not
1261 already a child by this name present. This roughly matches the
1262 ctime/mtime semantics of traditional filesystems. See the
1263 "About the metadata" section of webapi.txt for futher information.
1265 If this directory node is read-only, the Deferred will errback with a
1266 NotWriteableError."""
1268 def set_nodes(entries, overwrite=True):
1269 """Add multiple children to a directory node. Takes a dict mapping
1270 unicode childname to (child_node, metdata) tuples. If metdata=None,
1271 the original metadata is left unmodified. Returns a Deferred that
1272 fires (with this dirnode) when the operation finishes. This is
1273 equivalent to calling set_node() multiple times, but is much more
1276 def add_file(name, uploadable, metadata=None, overwrite=True):
1277 """I upload a file (using the given IUploadable), then attach the
1278 resulting ImmutableFileNode to the directory at the given name. I set
1279 metadata the same way as set_uri and set_node. The child name must be
1282 I return a Deferred that fires (with the IFileNode of the uploaded
1283 file) when the operation completes."""
1285 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1286 """I remove the child at the specific name. I return a Deferred that
1287 fires when the operation finishes. The child name must be a unicode
1288 string. If must_exist is True and I do not have a child by that name,
1289 I raise NoSuchChildError. If must_be_directory is True and the child
1290 is a file, or if must_be_file is True and the child is a directory,
1291 I raise ChildOfWrongTypeError."""
1293 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1294 """I create and attach a directory at the given name. The new
1295 directory can be empty, or it can be populated with children
1296 according to 'initial_children', which takes a dictionary in the same
1297 format as set_nodes (i.e. mapping unicode child name to (childnode,
1298 metadata) tuples). The child name must be a unicode string. I return
1299 a Deferred that fires (with the new directory node) when the
1300 operation finishes."""
1302 def move_child_to(current_child_name, new_parent, new_child_name=None,
1304 """I take one of my children and move them to a new parent. The child
1305 is referenced by name. On the new parent, the child will live under
1306 'new_child_name', which defaults to 'current_child_name'. TODO: what
1307 should we do about metadata? I return a Deferred that fires when the
1308 operation finishes. The child name must be a unicode string. I raise
1309 NoSuchChildError if I do not have a child by that name."""
1311 def build_manifest():
1312 """I generate a table of everything reachable from this directory.
1313 I also compute deep-stats as described below.
1315 I return a Monitor. The Monitor's results will be a dictionary with
1318 res['manifest']: a list of (path, cap) tuples for all nodes
1319 (directories and files) reachable from this one.
1320 'path' will be a tuple of unicode strings. The
1321 origin dirnode will be represented by an empty path
1323 res['verifycaps']: a list of (printable) verifycap strings, one for
1324 each reachable non-LIT node. This is a set:
1325 it will contain no duplicates.
1326 res['storage-index']: a list of (base32) storage index strings,
1327 one for each reachable non-LIT node. This is
1328 a set: it will contain no duplicates.
1329 res['stats']: a dictionary, the same that is generated by
1330 start_deep_stats() below.
1332 The Monitor will also have an .origin_si attribute with the (binary)
1333 storage index of the starting point.
1336 def start_deep_stats():
1337 """Return a Monitor, examining all nodes (directories and files)
1338 reachable from this one. The Monitor's results will be a dictionary
1339 with the following keys::
1341 count-immutable-files: count of how many CHK files are in the set
1342 count-mutable-files: same, for mutable files (does not include
1344 count-literal-files: same, for LIT files
1345 count-files: sum of the above three
1347 count-directories: count of directories
1349 size-immutable-files: total bytes for all CHK files in the set
1350 size-mutable-files (TODO): same, for current version of all mutable
1351 files, does not include directories
1352 size-literal-files: same, for LIT files
1353 size-directories: size of mutable files used by directories
1355 largest-directory: number of bytes in the largest directory
1356 largest-directory-children: number of children in the largest
1358 largest-immutable-file: number of bytes in the largest CHK file
1360 size-mutable-files is not yet implemented, because it would involve
1361 even more queries than deep_stats does.
1363 The Monitor will also have an .origin_si attribute with the (binary)
1364 storage index of the starting point.
1366 This operation will visit every directory node underneath this one,
1367 and can take a long time to run. On a typical workstation with good
1368 bandwidth, this can examine roughly 15 directories per second (and
1369 takes several minutes of 100% CPU for ~1700 directories).
1373 class ICodecEncoder(Interface):
1374 def set_params(data_size, required_shares, max_shares):
1375 """Set up the parameters of this encoder.
1377 This prepares the encoder to perform an operation that converts a
1378 single block of data into a number of shares, such that a future
1379 ICodecDecoder can use a subset of these shares to recover the
1380 original data. This operation is invoked by calling encode(). Once
1381 the encoding parameters are set up, the encode operation can be
1382 invoked multiple times.
1384 set_params() prepares the encoder to accept blocks of input data that
1385 are exactly 'data_size' bytes in length. The encoder will be prepared
1386 to produce 'max_shares' shares for each encode() operation (although
1387 see the 'desired_share_ids' to use less CPU). The encoding math will
1388 be chosen such that the decoder can get by with as few as
1389 'required_shares' of these shares and still reproduce the original
1390 data. For example, set_params(1000, 5, 5) offers no redundancy at
1391 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1393 Numerical Restrictions: 'data_size' is required to be an integral
1394 multiple of 'required_shares'. In general, the caller should choose
1395 required_shares and max_shares based upon their reliability
1396 requirements and the number of peers available (the total storage
1397 space used is roughly equal to max_shares*data_size/required_shares),
1398 then choose data_size to achieve the memory footprint desired (larger
1399 data_size means more efficient operation, smaller data_size means
1400 smaller memory footprint).
1402 In addition, 'max_shares' must be equal to or greater than
1403 'required_shares'. Of course, setting them to be equal causes
1404 encode() to degenerate into a particularly slow form of the 'split'
1407 See encode() for more details about how these parameters are used.
1409 set_params() must be called before any other ICodecEncoder methods
1414 """Return the 3-tuple of data_size, required_shares, max_shares"""
1416 def get_encoder_type():
1417 """Return a short string that describes the type of this encoder.
1419 There is required to be a global table of encoder classes. This method
1420 returns an index into this table; the value at this index is an
1421 encoder class, and this encoder is an instance of that class.
1424 def get_block_size():
1425 """Return the length of the shares that encode() will produce.
1428 def encode_proposal(data, desired_share_ids=None):
1429 """Encode some data.
1431 'data' must be a string (or other buffer object), and len(data) must
1432 be equal to the 'data_size' value passed earlier to set_params().
1434 This will return a Deferred that will fire with two lists. The first
1435 is a list of shares, each of which is a string (or other buffer
1436 object) such that len(share) is the same as what get_share_size()
1437 returned earlier. The second is a list of shareids, in which each is
1438 an integer. The lengths of the two lists will always be equal to each
1439 other. The user should take care to keep each share closely
1440 associated with its shareid, as one is useless without the other.
1442 The length of this output list will normally be the same as the value
1443 provided to the 'max_shares' parameter of set_params(). This may be
1444 different if 'desired_share_ids' is provided.
1446 'desired_share_ids', if provided, is required to be a sequence of
1447 ints, each of which is required to be >= 0 and < max_shares. If not
1448 provided, encode() will produce 'max_shares' shares, as if
1449 'desired_share_ids' were set to range(max_shares). You might use this
1450 if you initially thought you were going to use 10 peers, started
1451 encoding, and then two of the peers dropped out: you could use
1452 desired_share_ids= to skip the work (both memory and CPU) of
1453 producing shares for the peers that are no longer available.
1457 def encode(inshares, desired_share_ids=None):
1458 """Encode some data. This may be called multiple times. Each call is
1461 inshares is a sequence of length required_shares, containing buffers
1462 (i.e. strings), where each buffer contains the next contiguous
1463 non-overlapping segment of the input data. Each buffer is required to
1464 be the same length, and the sum of the lengths of the buffers is
1465 required to be exactly the data_size promised by set_params(). (This
1466 implies that the data has to be padded before being passed to
1467 encode(), unless of course it already happens to be an even multiple
1468 of required_shares in length.)
1470 Note: the requirement to break up your data into
1471 'required_shares' chunks of exactly the right length before
1472 calling encode() is surprising from point of view of a user
1473 who doesn't know how FEC works. It feels like an
1474 implementation detail that has leaked outside the abstraction
1475 barrier. Is there a use case in which the data to be encoded
1476 might already be available in pre-segmented chunks, such that
1477 it is faster or less work to make encode() take a list rather
1478 than splitting a single string?
1480 Yes, there is: suppose you are uploading a file with K=64,
1481 N=128, segsize=262,144. Then each in-share will be of size
1482 4096. If you use this .encode() API then your code could first
1483 read each successive 4096-byte chunk from the file and store
1484 each one in a Python string and store each such Python string
1485 in a Python list. Then you could call .encode(), passing that
1486 list as "inshares". The encoder would generate the other 64
1487 "secondary shares" and return to you a new list containing
1488 references to the same 64 Python strings that you passed in
1489 (as the primary shares) plus references to the new 64 Python
1492 (You could even imagine that your code could use readv() so
1493 that the operating system can arrange to get all of those
1494 bytes copied from the file into the Python list of Python
1495 strings as efficiently as possible instead of having a loop
1496 written in C or in Python to copy the next part of the file
1497 into the next string.)
1499 On the other hand if you instead use the .encode_proposal()
1500 API (above), then your code can first read in all of the
1501 262,144 bytes of the segment from the file into a Python
1502 string, then call .encode_proposal() passing the segment data
1503 as the "data" argument. The encoder would basically first
1504 split the "data" argument into a list of 64 in-shares of 4096
1505 byte each, and then do the same thing that .encode() does. So
1506 this would result in a little bit more copying of data and a
1507 little bit higher of a "maximum memory usage" during the
1508 process, although it might or might not make a practical
1509 difference for our current use cases.
1511 Note that "inshares" is a strange name for the parameter if
1512 you think of the parameter as being just for feeding in data
1513 to the codec. It makes more sense if you think of the result
1514 of this encoding as being the set of shares from inshares plus
1515 an extra set of "secondary shares" (or "check shares"). It is
1516 a surprising name! If the API is going to be surprising then
1517 the name should be surprising. If we switch to
1518 encode_proposal() above then we should also switch to an
1521 'desired_share_ids', if provided, is required to be a sequence of
1522 ints, each of which is required to be >= 0 and < max_shares. If not
1523 provided, encode() will produce 'max_shares' shares, as if
1524 'desired_share_ids' were set to range(max_shares). You might use this
1525 if you initially thought you were going to use 10 peers, started
1526 encoding, and then two of the peers dropped out: you could use
1527 desired_share_ids= to skip the work (both memory and CPU) of
1528 producing shares for the peers that are no longer available.
1530 For each call, encode() will return a Deferred that fires with two
1531 lists, one containing shares and the other containing the shareids.
1532 The get_share_size() method can be used to determine the length of
1533 the share strings returned by encode(). Each shareid is a small
1534 integer, exactly as passed into 'desired_share_ids' (or
1535 range(max_shares), if desired_share_ids was not provided).
1537 The shares and their corresponding shareids are required to be kept
1538 together during storage and retrieval. Specifically, the share data is
1539 useless by itself: the decoder needs to be told which share is which
1540 by providing it with both the shareid and the actual share data.
1542 This function will allocate an amount of memory roughly equal to::
1544 (max_shares - required_shares) * get_share_size()
1546 When combined with the memory that the caller must allocate to
1547 provide the input data, this leads to a memory footprint roughly
1548 equal to the size of the resulting encoded shares (i.e. the expansion
1549 factor times the size of the input segment).
1554 # returning a list of (shareidN,shareN) tuples instead of a pair of
1555 # lists (shareids..,shares..). Brian thought the tuples would
1556 # encourage users to keep the share and shareid together throughout
1557 # later processing, Zooko pointed out that the code to iterate
1558 # through two lists is not really more complicated than using a list
1559 # of tuples and there's also a performance improvement
1561 # having 'data_size' not required to be an integral multiple of
1562 # 'required_shares'. Doing this would require encode() to perform
1563 # padding internally, and we'd prefer to have any padding be done
1564 # explicitly by the caller. Yes, it is an abstraction leak, but
1565 # hopefully not an onerous one.
1568 class ICodecDecoder(Interface):
1569 def set_params(data_size, required_shares, max_shares):
1570 """Set the params. They have to be exactly the same ones that were
1571 used for encoding."""
1573 def get_needed_shares():
1574 """Return the number of shares needed to reconstruct the data.
1575 set_params() is required to be called before this."""
1577 def decode(some_shares, their_shareids):
1578 """Decode a partial list of shares into data.
1580 'some_shares' is required to be a sequence of buffers of sharedata, a
1581 subset of the shares returned by ICodecEncode.encode(). Each share is
1582 required to be of the same length. The i'th element of their_shareids
1583 is required to be the shareid of the i'th buffer in some_shares.
1585 This returns a Deferred that fires with a sequence of buffers. This
1586 sequence will contain all of the segments of the original data, in
1587 order. The sum of the lengths of all of the buffers will be the
1588 'data_size' value passed into the original ICodecEncode.set_params()
1589 call. To get back the single original input block of data, use
1590 ''.join(output_buffers), or you may wish to simply write them in
1591 order to an output file.
1593 Note that some of the elements in the result sequence may be
1594 references to the elements of the some_shares input sequence. In
1595 particular, this means that if those share objects are mutable (e.g.
1596 arrays) and if they are changed, then both the input (the
1597 'some_shares' parameter) and the output (the value given when the
1598 deferred is triggered) will change.
1600 The length of 'some_shares' is required to be exactly the value of
1601 'required_shares' passed into the original ICodecEncode.set_params()
1606 class IEncoder(Interface):
1607 """I take an object that provides IEncryptedUploadable, which provides
1608 encrypted data, and a list of shareholders. I then encode, hash, and
1609 deliver shares to those shareholders. I will compute all the necessary
1610 Merkle hash trees that are necessary to validate the crypttext that
1611 eventually comes back from the shareholders. I provide the URI Extension
1612 Block Hash, and the encoding parameters, both of which must be included
1615 I do not choose shareholders, that is left to the IUploader. I must be
1616 given a dict of RemoteReferences to storage buckets that are ready and
1617 willing to receive data.
1621 """Specify the number of bytes that will be encoded. This must be
1622 peformed before get_serialized_params() can be called.
1625 def set_encrypted_uploadable(u):
1626 """Provide a source of encrypted upload data. 'u' must implement
1627 IEncryptedUploadable.
1629 When this is called, the IEncryptedUploadable will be queried for its
1630 length and the storage_index that should be used.
1632 This returns a Deferred that fires with this Encoder instance.
1634 This must be performed before start() can be called.
1637 def get_param(name):
1638 """Return an encoding parameter, by name.
1640 'storage_index': return a string with the (16-byte truncated SHA-256
1641 hash) storage index to which these shares should be
1644 'share_counts': return a tuple describing how many shares are used:
1645 (needed_shares, servers_of_happiness, total_shares)
1647 'num_segments': return an int with the number of segments that
1650 'segment_size': return an int with the size of each segment.
1652 'block_size': return the size of the individual blocks that will
1653 be delivered to a shareholder's put_block() method. By
1654 knowing this, the shareholder will be able to keep all
1655 blocks in a single file and still provide random access
1656 when reading them. # TODO: can we avoid exposing this?
1658 'share_size': an int with the size of the data that will be stored
1659 on each shareholder. This is aggregate amount of data
1660 that will be sent to the shareholder, summed over all
1661 the put_block() calls I will ever make. It is useful to
1662 determine this size before asking potential
1663 shareholders whether they will grant a lease or not,
1664 since their answers will depend upon how much space we
1665 need. TODO: this might also include some amount of
1666 overhead, like the size of all the hashes. We need to
1667 decide whether this is useful or not.
1669 'serialized_params': a string with a concise description of the
1670 codec name and its parameters. This may be passed
1671 into the IUploadable to let it make sure that
1672 the same file encoded with different parameters
1673 will result in different storage indexes.
1675 Once this is called, set_size() and set_params() may not be called.
1678 def set_shareholders(shareholders, servermap):
1679 """Tell the encoder where to put the encoded shares. 'shareholders'
1680 must be a dictionary that maps share number (an integer ranging from
1681 0 to n-1) to an instance that provides IStorageBucketWriter.
1682 'servermap' is a dictionary that maps share number (as defined above)
1683 to a set of peerids. This must be performed before start() can be
1687 """Begin the encode/upload process. This involves reading encrypted
1688 data from the IEncryptedUploadable, encoding it, uploading the shares
1689 to the shareholders, then sending the hash trees.
1691 set_encrypted_uploadable() and set_shareholders() must be called
1692 before this can be invoked.
1694 This returns a Deferred that fires with a verify cap when the upload
1695 process is complete. The verifycap, plus the encryption key, is
1696 sufficient to construct the read cap.
1700 class IDecoder(Interface):
1701 """I take a list of shareholders and some setup information, then
1702 download, validate, decode, and decrypt data from them, writing the
1703 results to an output file.
1705 I do not locate the shareholders, that is left to the IDownloader. I must
1706 be given a dict of RemoteReferences to storage buckets that are ready to
1711 """I take a file-like object (providing write and close) to which all
1712 the plaintext data will be written.
1714 TODO: producer/consumer . Maybe write() should return a Deferred that
1715 indicates when it will accept more data? But probably having the
1716 IDecoder be a producer is easier to glue to IConsumer pieces.
1719 def set_shareholders(shareholders):
1720 """I take a dictionary that maps share identifiers (small integers)
1721 to RemoteReferences that provide RIBucketReader. This must be called
1725 """I start the download. This process involves retrieving data and
1726 hash chains from the shareholders, using the hashes to validate the
1727 data, decoding the shares into segments, decrypting the segments,
1728 then writing the resulting plaintext to the output file.
1730 I return a Deferred that will fire (with self) when the download is
1735 class IDownloadTarget(Interface):
1736 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1737 # will register itself as a producer. This allows the target to invoke
1738 # downloader.pauseProducing, resumeProducing, and stopProducing.
1740 """Called before any calls to write() or close(). If an error
1741 occurs before any data is available, fail() may be called without
1742 a previous call to open().
1744 'size' is the length of the file being downloaded, in bytes."""
1747 """Output some data to the target."""
1750 """Inform the target that there is no more data to be written."""
1753 """fail() is called to indicate that the download has failed. 'why'
1754 is a Failure object indicating what went wrong. No further methods
1755 will be invoked on the IDownloadTarget after fail()."""
1757 def register_canceller(cb):
1758 """The CiphertextDownloader uses this to register a no-argument function
1759 that the target can call to cancel the download. Once this canceller
1760 is invoked, no further calls to write() or close() will be made."""
1763 """When the CiphertextDownloader is done, this finish() function will be
1764 called. Whatever it returns will be returned to the invoker of
1765 Downloader.download.
1769 class IDownloader(Interface):
1770 def download(uri, target):
1771 """Perform a CHK download, sending the data to the given target.
1772 'target' must provide IDownloadTarget.
1774 Returns a Deferred that fires (with the results of target.finish)
1775 when the download is finished, or errbacks if something went wrong."""
1778 class IEncryptedUploadable(Interface):
1779 def set_upload_status(upload_status):
1780 """Provide an IUploadStatus object that should be filled with status
1781 information. The IEncryptedUploadable is responsible for setting
1782 key-determination progress ('chk'), size, storage_index, and
1783 ciphertext-fetch progress. It may delegate some of this
1784 responsibility to others, in particular to the IUploadable."""
1787 """This behaves just like IUploadable.get_size()."""
1789 def get_all_encoding_parameters():
1790 """Return a Deferred that fires with a tuple of
1791 (k,happy,n,segment_size). The segment_size will be used as-is, and
1792 must match the following constraints: it must be a multiple of k, and
1793 it shouldn't be unreasonably larger than the file size (if
1794 segment_size is larger than filesize, the difference must be stored
1797 This usually passes through to the IUploadable method of the same
1800 The encoder strictly obeys the values returned by this method. To
1801 make an upload use non-default encoding parameters, you must arrange
1802 to control the values that this method returns.
1805 def get_storage_index():
1806 """Return a Deferred that fires with a 16-byte storage index.
1809 def read_encrypted(length, hash_only):
1810 """This behaves just like IUploadable.read(), but returns crypttext
1811 instead of plaintext. If hash_only is True, then this discards the
1812 data (and returns an empty list); this improves efficiency when
1813 resuming an interrupted upload (where we need to compute the
1814 plaintext hashes, but don't need the redundant encrypted data)."""
1817 """Just like IUploadable.close()."""
1820 class IUploadable(Interface):
1821 def set_upload_status(upload_status):
1822 """Provide an IUploadStatus object that should be filled with status
1823 information. The IUploadable is responsible for setting
1824 key-determination progress ('chk')."""
1826 def set_default_encoding_parameters(params):
1827 """Set the default encoding parameters, which must be a dict mapping
1828 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1829 'max_segment_size'. These might have an influence on the final
1830 encoding parameters returned by get_all_encoding_parameters(), if the
1831 Uploadable doesn't have more specific preferences.
1833 This call is optional: if it is not used, the Uploadable will use
1834 some built-in defaults. If used, this method must be called before
1835 any other IUploadable methods to have any effect.
1839 """Return a Deferred that will fire with the length of the data to be
1840 uploaded, in bytes. This will be called before the data is actually
1841 used, to compute encoding parameters.
1844 def get_all_encoding_parameters():
1845 """Return a Deferred that fires with a tuple of
1846 (k,happy,n,segment_size). The segment_size will be used as-is, and
1847 must match the following constraints: it must be a multiple of k, and
1848 it shouldn't be unreasonably larger than the file size (if
1849 segment_size is larger than filesize, the difference must be stored
1852 The relative values of k and n allow some IUploadables to request
1853 better redundancy than others (in exchange for consuming more space
1856 Larger values of segment_size reduce hash overhead, while smaller
1857 values reduce memory footprint and cause data to be delivered in
1858 smaller pieces (which may provide a smoother and more predictable
1859 download experience).
1861 The encoder strictly obeys the values returned by this method. To
1862 make an upload use non-default encoding parameters, you must arrange
1863 to control the values that this method returns. One way to influence
1864 them may be to call set_encoding_parameters() before calling
1865 get_all_encoding_parameters().
1868 def get_encryption_key():
1869 """Return a Deferred that fires with a 16-byte AES key. This key will
1870 be used to encrypt the data. The key will also be hashed to derive
1873 Uploadables that want to achieve convergence should hash their file
1874 contents and the serialized_encoding_parameters to form the key
1875 (which of course requires a full pass over the data). Uploadables can
1876 use the upload.ConvergentUploadMixin class to achieve this
1879 Uploadables that do not care about convergence (or do not wish to
1880 make multiple passes over the data) can simply return a
1881 strongly-random 16 byte string.
1883 get_encryption_key() may be called multiple times: the IUploadable is
1884 required to return the same value each time.
1888 """Return a Deferred that fires with a list of strings (perhaps with
1889 only a single element) that, when concatenated together, contain the
1890 next 'length' bytes of data. If EOF is near, this may provide fewer
1891 than 'length' bytes. The total number of bytes provided by read()
1892 before it signals EOF must equal the size provided by get_size().
1894 If the data must be acquired through multiple internal read
1895 operations, returning a list instead of a single string may help to
1896 reduce string copies. However, the length of the concatenated strings
1897 must equal the amount of data requested, unless EOF is encountered.
1898 Long reads, or short reads without EOF, are not allowed. read()
1899 should return the same amount of data as a local disk file read, just
1900 in a different shape and asynchronously.
1902 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1903 so a 10kB file means length=3kB, 100kB file means length=30kB,
1904 and >=1MB file means length=300kB.
1906 This method provides for a single full pass through the data. Later
1907 use cases may desire multiple passes or access to only parts of the
1908 data (such as a mutable file making small edits-in-place). This API
1909 will be expanded once those use cases are better understood.
1913 """The upload is finished, and whatever filehandle was in use may be
1917 class IMutableUploadable(Interface):
1919 I represent content that is due to be uploaded to a mutable filecap.
1921 # This is somewhat simpler than the IUploadable interface above
1922 # because mutable files do not need to be concerned with possibly
1923 # generating a CHK, nor with per-file keys. It is a subset of the
1924 # methods in IUploadable, though, so we could just as well implement
1925 # the mutable uploadables as IUploadables that don't happen to use
1926 # those methods (with the understanding that the unused methods will
1927 # never be called on such objects)
1930 Returns a Deferred that fires with the size of the content held
1936 Returns a list of strings that, when concatenated, are the next
1937 length bytes of the file, or fewer if there are fewer bytes
1938 between the current location and the end of the file.
1943 The process that used the Uploadable is finished using it, so
1944 the uploadable may be closed.
1948 class IUploadResults(Interface):
1949 """I am returned by immutable upload() methods and contain the results of
1952 Note that some of my methods return empty values (0 or an empty dict)
1953 when called for non-distributed LIT files."""
1955 def get_file_size():
1956 """Return the file size, in bytes."""
1959 """Return the (string) URI of the object uploaded, a CHK readcap."""
1961 def get_ciphertext_fetched():
1962 """Return the number of bytes fetched by the helpe for this upload,
1963 or 0 if the helper did not need to fetch any bytes (or if there was
1966 def get_preexisting_shares():
1967 """Return the number of shares that were already present in the grid."""
1969 def get_pushed_shares():
1970 """Return the number of shares that were uploaded."""
1973 """Return a dict mapping share identifier to set of IServer
1974 instances. This indicates which servers were given which shares. For
1975 immutable files, the shareid is an integer (the share number, from 0
1976 to N-1). For mutable files, it is a string of the form
1977 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1978 the share number."""
1980 def get_servermap():
1981 """Return dict mapping IServer instance to a set of share numbers."""
1984 """Return dict of timing information, mapping name to seconds. All
1986 total : total upload time, start to finish
1987 storage_index : time to compute the storage index
1988 peer_selection : time to decide which peers will be used
1989 contacting_helper : initial helper query to upload/no-upload decision
1990 helper_total : initial helper query to helper finished pushing
1991 cumulative_fetch : helper waiting for ciphertext requests
1992 total_fetch : helper start to last ciphertext response
1993 cumulative_encoding : just time spent in zfec
1994 cumulative_sending : just time spent waiting for storage servers
1995 hashes_and_close : last segment push to shareholder close
1996 total_encode_and_push : first encode to shareholder close
1999 def get_uri_extension_data():
2000 """Return the dict of UEB data created for this file."""
2002 def get_verifycapstr():
2003 """Return the (string) verify-cap URI for the uploaded object."""
2006 class IDownloadResults(Interface):
2007 """I am created internally by download() methods. I contain a number of
2008 public attributes that contain details about the download process.::
2010 .file_size : the size of the file, in bytes
2011 .servers_used : set of server peerids that were used during download
2012 .server_problems : dict mapping server peerid to a problem string. Only
2013 servers that had problems (bad hashes, disconnects)
2015 .servermap : dict mapping server peerid to a set of share numbers. Only
2016 servers that had any shares are listed here.
2017 .timings : dict of timing information, mapping name to seconds (float)
2018 peer_selection : time to ask servers about shares
2019 servers_peer_selection : dict of peerid to DYHB-query time
2020 uri_extension : time to fetch a copy of the URI extension block
2021 hashtrees : time to fetch the hash trees
2022 segments : time to fetch, decode, and deliver segments
2023 cumulative_fetch : time spent waiting for storage servers
2024 cumulative_decode : just time spent in zfec
2025 cumulative_decrypt : just time spent in decryption
2026 total : total download time, start to finish
2027 fetch_per_server : dict of server to list of per-segment fetch times
2031 class IUploader(Interface):
2032 def upload(uploadable):
2033 """Upload the file. 'uploadable' must impement IUploadable. This
2034 returns a Deferred that fires with an IUploadResults instance, from
2035 which the URI of the file can be obtained as results.uri ."""
2038 class ICheckable(Interface):
2039 def check(monitor, verify=False, add_lease=False):
2040 """Check up on my health, optionally repairing any problems.
2042 This returns a Deferred that fires with an instance that provides
2043 ICheckResults, or None if the object is non-distributed (i.e. LIT
2046 The monitor will be checked periodically to see if the operation has
2047 been cancelled. If so, no new queries will be sent, and the Deferred
2048 will fire (with a OperationCancelledError) immediately.
2050 Filenodes and dirnodes (which provide IFilesystemNode) are also
2051 checkable. Instances that represent verifier-caps will be checkable
2052 but not downloadable. Some objects (like LIT files) do not actually
2053 live in the grid, and their checkers return None (non-distributed
2054 files are always healthy).
2056 If verify=False, a relatively lightweight check will be performed: I
2057 will ask all servers if they have a share for me, and I will believe
2058 whatever they say. If there are at least N distinct shares on the
2059 grid, my results will indicate r.is_healthy()==True. This requires a
2060 roundtrip to each server, but does not transfer very much data, so
2061 the network bandwidth is fairly low.
2063 If verify=True, a more resource-intensive check will be performed:
2064 every share will be downloaded, and the hashes will be validated on
2065 every bit. I will ignore any shares that failed their hash checks. If
2066 there are at least N distinct valid shares on the grid, my results
2067 will indicate r.is_healthy()==True. This requires N/k times as much
2068 download bandwidth (and server disk IO) as a regular download. If a
2069 storage server is holding a corrupt share, or is experiencing memory
2070 failures during retrieval, or is malicious or buggy, then
2071 verification will detect the problem, but checking will not.
2073 If add_lease=True, I will ensure that an up-to-date lease is present
2074 on each share. The lease secrets will be derived from by node secret
2075 (in BASEDIR/private/secret), so either I will add a new lease to the
2076 share, or I will merely renew the lease that I already had. In a
2077 future version of the storage-server protocol (once Accounting has
2078 been implemented), there may be additional options here to define the
2079 kind of lease that is obtained (which account number to claim, etc).
2081 TODO: any problems seen during checking will be reported to the
2082 health-manager.furl, a centralized object that is responsible for
2083 figuring out why files are unhealthy so corrective action can be
2087 def check_and_repair(monitor, verify=False, add_lease=False):
2088 """Like check(), but if the file/directory is not healthy, attempt to
2091 Any non-healthy result will cause an immediate repair operation, to
2092 generate and upload new shares. After repair, the file will be as
2093 healthy as we can make it. Details about what sort of repair is done
2094 will be put in the check-and-repair results. The Deferred will not
2095 fire until the repair is complete.
2097 This returns a Deferred that fires with an instance of
2098 ICheckAndRepairResults."""
2101 class IDeepCheckable(Interface):
2102 def start_deep_check(verify=False, add_lease=False):
2103 """Check upon the health of me and everything I can reach.
2105 This is a recursive form of check(), useable only on dirnodes.
2107 I return a Monitor, with results that are an IDeepCheckResults
2110 TODO: If any of the directories I traverse are unrecoverable, the
2111 Monitor will report failure. If any of the files I check upon are
2112 unrecoverable, those problems will be reported in the
2113 IDeepCheckResults as usual, and the Monitor will not report a
2117 def start_deep_check_and_repair(verify=False, add_lease=False):
2118 """Check upon the health of me and everything I can reach. Repair
2119 anything that isn't healthy.
2121 This is a recursive form of check_and_repair(), useable only on
2124 I return a Monitor, with results that are an
2125 IDeepCheckAndRepairResults object.
2127 TODO: If any of the directories I traverse are unrecoverable, the
2128 Monitor will report failure. If any of the files I check upon are
2129 unrecoverable, those problems will be reported in the
2130 IDeepCheckResults as usual, and the Monitor will not report a
2135 class ICheckResults(Interface):
2136 """I contain the detailed results of a check/verify operation.
2139 def get_storage_index():
2140 """Return a string with the (binary) storage index."""
2142 def get_storage_index_string():
2143 """Return a string with the (printable) abbreviated storage index."""
2146 """Return the (string) URI of the object that was checked."""
2149 """Return a boolean, True if the file/dir is fully healthy, False if
2150 it is damaged in any way. Non-distributed LIT files always return
2153 def is_recoverable():
2154 """Return a boolean, True if the file/dir can be recovered, False if
2155 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2156 files always return True."""
2158 # the following methods all return None for non-distributed LIT files
2160 def get_happiness():
2161 """Return the happiness count of the file."""
2163 def get_encoding_needed():
2164 """Return 'k', the number of shares required for recovery."""
2166 def get_encoding_expected():
2167 """Return 'N', the number of total shares generated."""
2169 def get_share_counter_good():
2170 """Return the number of distinct good shares that were found. For
2171 mutable files, this counts shares for the 'best' version."""
2173 def get_share_counter_wrong():
2174 """For mutable files, return the number of shares for versions other
2175 than the 'best' one (which is defined as being the recoverable
2176 version with the highest sequence number, then the highest roothash).
2177 These are either leftover shares from an older version (perhaps on a
2178 server that was offline when an update occurred), shares from an
2179 unrecoverable newer version, or shares from an alternate current
2180 version that results from an uncoordinated write collision. For a
2181 healthy file, this will equal 0. For immutable files, this will
2184 def get_corrupt_shares():
2185 """Return a list of 'share locators', one for each share that was
2186 found to be corrupt (integrity failure). Each share locator is a list
2187 of (IServer, storage_index, sharenum)."""
2189 def get_incompatible_shares():
2190 """Return a list of 'share locators', one for each share that was
2191 found to be of an unknown format. Each share locator is a list of
2192 (IServer, storage_index, sharenum)."""
2194 def get_servers_responding():
2195 """Return a list of IServer objects, one for each server that
2196 responded to the share query (even if they said they didn't have
2197 shares, and even if they said they did have shares but then didn't
2198 send them when asked, or dropped the connection, or returned a
2199 Failure, and even if they said they did have shares and sent
2200 incorrect ones when asked)"""
2202 def get_host_counter_good_shares():
2203 """Return the number of distinct storage servers with good shares. If
2204 this number is less than get_share_counters()[good], then some shares
2205 are doubled up, increasing the correlation of failures. This
2206 indicates that one or more shares should be moved to an otherwise
2207 unused server, if one is available.
2210 def get_version_counter_recoverable():
2211 """Return the number of recoverable versions of the file. For a
2212 healthy file, this will equal 1."""
2214 def get_version_counter_unrecoverable():
2215 """Return the number of unrecoverable versions of the file. For a
2216 healthy file, this will be 0."""
2219 """Return a dict mapping share identifier to list of IServer objects.
2220 This indicates which servers are holding which shares. For immutable
2221 files, the shareid is an integer (the share number, from 0 to N-1).
2222 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2223 containing the sequence number, the roothash, and the share number."""
2226 """Return a string with a brief (one-line) summary of the results."""
2229 """Return a list of strings with more detailed results."""
2232 class ICheckAndRepairResults(Interface):
2233 """I contain the detailed results of a check/verify/repair operation.
2235 The IFilesystemNode.check()/verify()/repair() methods all return
2236 instances that provide ICheckAndRepairResults.
2239 def get_storage_index():
2240 """Return a string with the (binary) storage index."""
2242 def get_storage_index_string():
2243 """Return a string with the (printable) abbreviated storage index."""
2245 def get_repair_attempted():
2246 """Return a boolean, True if a repair was attempted. We might not
2247 attempt to repair the file because it was healthy, or healthy enough
2248 (i.e. some shares were missing but not enough to exceed some
2249 threshold), or because we don't know how to repair this object."""
2251 def get_repair_successful():
2252 """Return a boolean, True if repair was attempted and the file/dir
2253 was fully healthy afterwards. False if no repair was attempted or if
2254 a repair attempt failed."""
2256 def get_pre_repair_results():
2257 """Return an ICheckResults instance that describes the state of the
2258 file/dir before any repair was attempted."""
2260 def get_post_repair_results():
2261 """Return an ICheckResults instance that describes the state of the
2262 file/dir after any repair was attempted. If no repair was attempted,
2263 the pre-repair and post-repair results will be identical."""
2266 class IDeepCheckResults(Interface):
2267 """I contain the results of a deep-check operation.
2269 This is returned by a call to ICheckable.deep_check().
2272 def get_root_storage_index_string():
2273 """Return the storage index (abbreviated human-readable string) of
2274 the first object checked."""
2277 """Return a dictionary with the following keys::
2279 count-objects-checked: count of how many objects were checked
2280 count-objects-healthy: how many of those objects were completely
2282 count-objects-unhealthy: how many were damaged in some way
2283 count-objects-unrecoverable: how many were unrecoverable
2284 count-corrupt-shares: how many shares were found to have
2285 corruption, summed over all objects
2289 def get_corrupt_shares():
2290 """Return a set of (IServer, storage_index, sharenum) for all shares
2291 that were found to be corrupt. storage_index is binary."""
2293 def get_all_results():
2294 """Return a dictionary mapping pathname (a tuple of strings, ready to
2295 be slash-joined) to an ICheckResults instance, one for each object
2296 that was checked."""
2298 def get_results_for_storage_index(storage_index):
2299 """Retrive the ICheckResults instance for the given (binary)
2300 storage index. Raises KeyError if there are no results for that
2304 """Return a dictionary with the same keys as
2305 IDirectoryNode.deep_stats()."""
2308 class IDeepCheckAndRepairResults(Interface):
2309 """I contain the results of a deep-check-and-repair operation.
2311 This is returned by a call to ICheckable.deep_check_and_repair().
2314 def get_root_storage_index_string():
2315 """Return the storage index (abbreviated human-readable string) of
2316 the first object checked."""
2319 """Return a dictionary with the following keys::
2321 count-objects-checked: count of how many objects were checked
2322 count-objects-healthy-pre-repair: how many of those objects were
2323 completely healthy (before any
2325 count-objects-unhealthy-pre-repair: how many were damaged in
2327 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2328 count-objects-healthy-post-repair: how many of those objects were
2329 completely healthy (after any
2331 count-objects-unhealthy-post-repair: how many were damaged in
2333 count-objects-unrecoverable-post-repair: how many were
2335 count-repairs-attempted: repairs were attempted on this many
2336 objects. The count-repairs- keys will
2337 always be provided, however unless
2338 repair=true is present, they will all
2340 count-repairs-successful: how many repairs resulted in healthy
2342 count-repairs-unsuccessful: how many repairs resulted did not
2343 results in completely healthy objects
2344 count-corrupt-shares-pre-repair: how many shares were found to
2345 have corruption, summed over all
2346 objects examined (before any
2348 count-corrupt-shares-post-repair: how many shares were found to
2349 have corruption, summed over all
2350 objects examined (after any
2355 """Return a dictionary with the same keys as
2356 IDirectoryNode.deep_stats()."""
2358 def get_corrupt_shares():
2359 """Return a set of (IServer, storage_index, sharenum) for all shares
2360 that were found to be corrupt before any repair was attempted.
2361 storage_index is binary.
2363 def get_remaining_corrupt_shares():
2364 """Return a set of (IServer, storage_index, sharenum) for all shares
2365 that were found to be corrupt after any repair was completed.
2366 storage_index is binary. These are shares that need manual inspection
2367 and probably deletion.
2369 def get_all_results():
2370 """Return a dictionary mapping pathname (a tuple of strings, ready to
2371 be slash-joined) to an ICheckAndRepairResults instance, one for each
2372 object that was checked."""
2374 def get_results_for_storage_index(storage_index):
2375 """Retrive the ICheckAndRepairResults instance for the given (binary)
2376 storage index. Raises KeyError if there are no results for that
2380 class IRepairable(Interface):
2381 def repair(check_results):
2382 """Attempt to repair the given object. Returns a Deferred that fires
2383 with a IRepairResults object.
2385 I must be called with an object that implements ICheckResults, as
2386 proof that you have actually discovered a problem with this file. I
2387 will use the data in the checker results to guide the repair process,
2388 such as which servers provided bad data and should therefore be
2389 avoided. The ICheckResults object is inside the
2390 ICheckAndRepairResults object, which is returned by the
2391 ICheckable.check() method::
2393 d = filenode.check(repair=False)
2394 def _got_results(check_and_repair_results):
2395 check_results = check_and_repair_results.get_pre_repair_results()
2396 return filenode.repair(check_results)
2397 d.addCallback(_got_results)
2402 class IRepairResults(Interface):
2403 """I contain the results of a repair operation."""
2404 def get_successful():
2405 """Returns a boolean: True if the repair made the file healthy, False
2406 if not. Repair failure generally indicates a file that has been
2407 damaged beyond repair."""
2410 class IClient(Interface):
2411 def upload(uploadable):
2412 """Upload some data into a CHK, get back the UploadResults for it.
2413 @param uploadable: something that implements IUploadable
2414 @return: a Deferred that fires with the UploadResults instance.
2415 To get the URI for this file, use results.uri .
2418 def create_mutable_file(contents=""):
2419 """Create a new mutable file (with initial) contents, get back the
2422 @param contents: (bytestring, callable, or None): this provides the
2423 initial contents of the mutable file. If 'contents' is a bytestring,
2424 it will be used as-is. If 'contents' is a callable, it will be
2425 invoked with the new MutableFileNode instance and is expected to
2426 return a bytestring with the initial contents of the file (the
2427 callable can use node.get_writekey() to decide how to encrypt the
2428 initial contents, e.g. for a brand new dirnode with initial
2429 children). contents=None is equivalent to an empty string. Using
2430 content_maker= is more efficient than creating a mutable file and
2431 setting its contents in two separate operations.
2433 @return: a Deferred that fires with an IMutableFileNode instance.
2436 def create_dirnode(initial_children={}):
2437 """Create a new unattached dirnode, possibly with initial children.
2439 @param initial_children: dict with keys that are unicode child names,
2440 and values that are (childnode, metadata) tuples.
2442 @return: a Deferred that fires with the new IDirectoryNode instance.
2445 def create_node_from_uri(uri, rouri):
2446 """Create a new IFilesystemNode instance from the uri, synchronously.
2447 @param uri: a string or IURI-providing instance, or None. This could
2448 be for a LiteralFileNode, a CHK file node, a mutable file
2449 node, or a directory node
2450 @param rouri: a string or IURI-providing instance, or None. If the
2451 main uri is None, I will use the rouri instead. If I
2452 recognize the format of the main uri, I will ignore the
2453 rouri (because it can be derived from the writecap).
2455 @return: an instance that provides IFilesystemNode (or more usefully
2456 one of its subclasses). File-specifying URIs will result in
2457 IFileNode-providing instances, like ImmutableFileNode,
2458 LiteralFileNode, or MutableFileNode. Directory-specifying
2459 URIs will result in IDirectoryNode-providing instances, like
2464 class INodeMaker(Interface):
2465 """The NodeMaker is used to create IFilesystemNode instances. It can
2466 accept a filecap/dircap string and return the node right away. It can
2467 also create new nodes (i.e. upload a file, or create a mutable file)
2468 asynchronously. Once you have one of these nodes, you can use other
2469 methods to determine whether it is a file or directory, and to download
2470 or modify its contents.
2472 The NodeMaker encapsulates all the authorities that these
2473 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2474 Tahoe process will typically have a single NodeMaker, but unit tests may
2475 create simplified/mocked forms for testing purposes.
2478 def create_from_cap(writecap, readcap=None, deep_immutable=False, name=u"<unknown name>"):
2479 """I create an IFilesystemNode from the given writecap/readcap. I can
2480 only provide nodes for existing file/directory objects: use my other
2481 methods to create new objects. I return synchronously."""
2483 def create_mutable_file(contents=None, keysize=None):
2484 """I create a new mutable file, and return a Deferred that will fire
2485 with the IMutableFileNode instance when it is ready. If contents= is
2486 provided (a bytestring), it will be used as the initial contents of
2487 the new file, otherwise the file will contain zero bytes. keysize= is
2488 for use by unit tests, to create mutable files that are smaller than
2491 def create_new_mutable_directory(initial_children={}):
2492 """I create a new mutable directory, and return a Deferred that will
2493 fire with the IDirectoryNode instance when it is ready. If
2494 initial_children= is provided (a dict mapping unicode child name to
2495 (childnode, metadata_dict) tuples), the directory will be populated
2496 with those children, otherwise it will be empty."""
2499 class IClientStatus(Interface):
2500 def list_all_uploads():
2501 """Return a list of uploader objects, one for each upload that
2502 currently has an object available (tracked with weakrefs). This is
2503 intended for debugging purposes."""
2505 def list_active_uploads():
2506 """Return a list of active IUploadStatus objects."""
2508 def list_recent_uploads():
2509 """Return a list of IUploadStatus objects for the most recently
2512 def list_all_downloads():
2513 """Return a list of downloader objects, one for each download that
2514 currently has an object available (tracked with weakrefs). This is
2515 intended for debugging purposes."""
2517 def list_active_downloads():
2518 """Return a list of active IDownloadStatus objects."""
2520 def list_recent_downloads():
2521 """Return a list of IDownloadStatus objects for the most recently
2522 started downloads."""
2525 class IUploadStatus(Interface):
2527 """Return a timestamp (float with seconds since epoch) indicating
2528 when the operation was started."""
2530 def get_storage_index():
2531 """Return a string with the (binary) storage index in use on this
2532 upload. Returns None if the storage index has not yet been
2536 """Return an integer with the number of bytes that will eventually
2537 be uploaded for this file. Returns None if the size is not yet known.
2540 """Return True if this upload is using a Helper, False if not."""
2543 """Return a string describing the current state of the upload
2547 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2548 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2549 made towards hashing the file to determine a CHK encryption key: if
2550 non-convergent encryption is in use, this will be trivial, otherwise
2551 the whole file must be hashed. 'ciphertext' describes how much of the
2552 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2553 uploads. 'encode_and_push' describes how much of the encode-and-push
2554 process has finished: for helper uploads this is dependent upon the
2555 helper providing progress reports. It might be reasonable to add all
2556 three numbers and report the sum to the user."""
2559 """Return True if the upload is currently active, False if not."""
2562 """Return an instance of UploadResults (which contains timing and
2563 sharemap information). Might return None if the upload is not yet
2567 """Each upload status gets a unique number: this method returns that
2568 number. This provides a handle to this particular upload, so a web
2569 page can generate a suitable hyperlink."""
2572 class IDownloadStatus(Interface):
2574 """Return a timestamp (float with seconds since epoch) indicating
2575 when the operation was started."""
2577 def get_storage_index():
2578 """Return a string with the (binary) storage index in use on this
2579 download. This may be None if there is no storage index (i.e. LIT
2583 """Return an integer with the number of bytes that will eventually be
2584 retrieved for this file. Returns None if the size is not yet known.
2588 """Return True if this download is using a Helper, False if not."""
2591 """Return a string describing the current state of the download
2595 """Returns a float (from 0.0 to 1.0) describing the amount of the
2596 download that has completed. This value will remain at 0.0 until the
2597 first byte of plaintext is pushed to the download target."""
2600 """Return True if the download is currently active, False if not."""
2603 """Each download status gets a unique number: this method returns
2604 that number. This provides a handle to this particular download, so a
2605 web page can generate a suitable hyperlink."""
2608 class IServermapUpdaterStatus(Interface):
2611 class IPublishStatus(Interface):
2614 class IRetrieveStatus(Interface):
2618 class NotCapableError(Exception):
2619 """You have tried to write to a read-only node."""
2621 class BadWriteEnablerError(Exception):
2625 class RIControlClient(RemoteInterface):
2626 def wait_for_client_connections(num_clients=int):
2627 """Do not return until we have connections to at least NUM_CLIENTS
2631 def upload_from_file_to_uri(filename=str,
2632 convergence=ChoiceOf(None,
2633 StringConstraint(2**20))):
2634 """Upload a file to the grid. This accepts a filename (which must be
2635 absolute) that points to a file on the node's local disk. The node will
2636 read the contents of this file, upload it to the grid, then return the
2637 URI at which it was uploaded. If convergence is None then a random
2638 encryption key will be used, else the plaintext will be hashed, then
2639 that hash will be mixed together with the "convergence" string to form
2644 def download_from_uri_to_file(uri=URI, filename=str):
2645 """Download a file from the grid, placing it on the node's local disk
2646 at the given filename (which must be absolute[?]). Returns the
2647 absolute filename where the file was written."""
2652 def get_memory_usage():
2653 """Return a dict describes the amount of memory currently in use. The
2654 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2655 measuring memory consupmtion in bytes."""
2656 return DictOf(str, int)
2658 def speed_test(count=int, size=int, mutable=Any()):
2659 """Write 'count' tempfiles to disk, all of the given size. Measure
2660 how long (in seconds) it takes to upload them all to the servers.
2661 Then measure how long it takes to download all of them. If 'mutable'
2662 is 'create', time creation of mutable files. If 'mutable' is
2663 'upload', then time access to the same mutable file instead of
2666 Returns a tuple of (upload_time, download_time).
2668 return (float, float)
2670 def measure_peer_response_time():
2671 """Send a short message to each connected peer, and measure the time
2672 it takes for them to respond to it. This is a rough measure of the
2673 application-level round trip time.
2675 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2678 return DictOf(str, float)
2681 UploadResults = Any() #DictOf(str, str)
2684 class RIEncryptedUploadable(RemoteInterface):
2685 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2690 def get_all_encoding_parameters():
2691 return (int, int, int, long)
2693 def read_encrypted(offset=Offset, length=ReadSize):
2700 class RICHKUploadHelper(RemoteInterface):
2701 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2705 Return a dictionary of version information.
2707 return DictOf(str, Any())
2709 def upload(reader=RIEncryptedUploadable):
2710 return UploadResults
2713 class RIHelper(RemoteInterface):
2714 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2718 Return a dictionary of version information.
2720 return DictOf(str, Any())
2722 def upload_chk(si=StorageIndex):
2723 """See if a file with a given storage index needs uploading. The
2724 helper will ask the appropriate storage servers to see if the file
2725 has already been uploaded. If so, the helper will return a set of
2726 'upload results' that includes whatever hashes are needed to build
2727 the read-cap, and perhaps a truncated sharemap.
2729 If the file has not yet been uploaded (or if it was only partially
2730 uploaded), the helper will return an empty upload-results dictionary
2731 and also an RICHKUploadHelper object that will take care of the
2732 upload process. The client should call upload() on this object and
2733 pass it a reference to an RIEncryptedUploadable object that will
2734 provide ciphertext. When the upload is finished, the upload() method
2735 will finish and return the upload results.
2737 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2740 class RIStatsProvider(RemoteInterface):
2741 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2743 Provides access to statistics and monitoring information.
2748 returns a dictionary containing 'counters' and 'stats', each a
2749 dictionary with string counter/stat name keys, and numeric or None values.
2750 counters are monotonically increasing measures of work done, and
2751 stats are instantaneous measures (potentially time averaged
2754 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2757 class RIStatsGatherer(RemoteInterface):
2758 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2760 Provides a monitoring service for centralised collection of stats
2763 def provide(provider=RIStatsProvider, nickname=str):
2765 @param provider: a stats collector instance that should be polled
2766 periodically by the gatherer to collect stats.
2767 @param nickname: a name useful to identify the provided client
2772 class IStatsProducer(Interface):
2775 returns a dictionary, with str keys representing the names of stats
2776 to be monitored, and numeric values.
2779 class RIKeyGenerator(RemoteInterface):
2780 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2782 Provides a service offering to make RSA key pairs.
2785 def get_rsa_key_pair(key_size=int):
2787 @param key_size: the size of the signature key.
2788 @return: tuple(verifying_key, signing_key)
2790 return TupleOf(str, str)
2793 class FileTooLargeError(Exception):
2797 class IValidatedThingProxy(Interface):
2799 """ Acquire a thing and validate it. Return a deferred that is
2800 eventually fired with self if the thing is valid or errbacked if it
2801 can't be acquired or validated."""
2804 class InsufficientVersionError(Exception):
2805 def __init__(self, needed, got):
2806 self.needed = needed
2810 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2813 class EmptyPathnameComponentError(Exception):
2814 """The webapi disallows empty pathname components."""