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 bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIBucketWriter(RemoteInterface):
34 """ Objects of this kind live on the server side. """
35 def write(offset=Offset, data=ShareData):
40 If the data that has been written is incomplete or inconsistent then
41 the server will throw the data away, else it will store it for future
47 """Abandon all the data that has been written.
52 class RIBucketReader(RemoteInterface):
53 def read(offset=Offset, length=ReadSize):
56 def advise_corrupt_share(reason=str):
57 """Clients who discover hash failures in shares that they have
58 downloaded from me will use this method to inform me about the
59 failures. I will record their concern so that my operator can
60 manually inspect the shares in question. I return None.
62 This is a wrapper around RIStorageServer.advise_corrupt_share(),
63 which is tied to a specific share, and therefore does not need the
64 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
86 class RIStorageServer(RemoteInterface):
87 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
91 Return a dictionary of version information.
93 return DictOf(str, Any())
95 def allocate_buckets(storage_index=StorageIndex,
96 renew_secret=LeaseRenewSecret,
97 cancel_secret=LeaseCancelSecret,
98 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
99 allocated_size=Offset, canary=Referenceable):
101 @param storage_index: the index of the bucket to be created or
103 @param sharenums: these are the share numbers (probably between 0 and
104 99) that the sender is proposing to store on this
106 @param renew_secret: This is the secret used to protect bucket refresh
107 This secret is generated by the client and
108 stored for later comparison by the server. Each
109 server is given a different secret.
110 @param cancel_secret: Like renew_secret, but protects bucket decref.
111 @param canary: If the canary is lost before close(), the bucket is
113 @return: tuple of (alreadygot, allocated), where alreadygot is what we
114 already have and allocated is what we hereby agree to accept.
115 New leases are added for shares in both lists.
117 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
118 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
120 def add_lease(storage_index=StorageIndex,
121 renew_secret=LeaseRenewSecret,
122 cancel_secret=LeaseCancelSecret):
124 Add a new lease on the given bucket. If the renew_secret matches an
125 existing lease, that lease will be renewed instead. If there is no
126 bucket for the given storage_index, return silently. (note that in
127 tahoe-1.3.0 and earlier, IndexError was raised if there was no
130 return Any() # returns None now, but future versions might change
132 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
134 Renew the lease on a given bucket, resetting the timer to 31 days.
135 Some networks will use this, some will not. If there is no bucket for
136 the given storage_index, IndexError will be raised.
138 For mutable shares, if the given renew_secret does not match an
139 existing lease, IndexError will be raised with a note listing the
140 server-nodeids on the existing leases, so leases on migrated shares
141 can be renewed or cancelled. For immutable shares, IndexError
142 (without the note) will be raised.
146 def get_buckets(storage_index=StorageIndex):
147 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
151 def slot_readv(storage_index=StorageIndex,
152 shares=ListOf(int), readv=ReadVector):
153 """Read a vector from the numbered shares associated with the given
154 storage index. An empty shares list means to return data from all
155 known shares. Returns a dictionary with one key per share."""
156 return DictOf(int, ReadData) # shnum -> results
158 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
159 secrets=TupleOf(WriteEnablerSecret,
162 tw_vectors=TestAndWriteVectorsForShares,
166 General-purpose test-read-and-set operation for mutable slots:
167 (1) For submitted shnums, compare the test vectors against extant
168 shares, or against an empty share for shnums that do not exist.
169 (2) Use the read vectors to extract "old data" from extant shares.
170 (3) If all tests in (1) passed, then apply the write vectors
171 (possibly creating new shares).
172 (4) Return whether the tests passed, and the "old data", which does
173 not include any modifications made by the writes.
175 The operation does not interleave with other operations on the same
178 This method is, um, large. The goal is to allow clients to update all
179 the shares associated with a mutable file in a single round trip.
181 @param storage_index: the index of the bucket to be created or
183 @param write_enabler: a secret that is stored along with the slot.
184 Writes are accepted from any caller who can
185 present the matching secret. A different secret
186 should be used for each slot*server pair.
187 @param renew_secret: This is the secret used to protect bucket refresh
188 This secret is generated by the client and
189 stored for later comparison by the server. Each
190 server is given a different secret.
191 @param cancel_secret: Like renew_secret, but protects bucket decref.
193 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
194 cancel_secret). The first is required to perform any write. The
195 latter two are used when allocating new shares. To simply acquire a
196 new lease on existing shares, use an empty testv and an empty writev.
198 Each share can have a separate test vector (i.e. a list of
199 comparisons to perform). If all vectors for all shares pass, then all
200 writes for all shares are recorded. Each comparison is a 4-tuple of
201 (offset, length, operator, specimen), which effectively does a
202 bool( (read(offset, length)) OPERATOR specimen ) and only performs
203 the write if all these evaluate to True. Basic test-and-set uses 'eq'.
204 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
205 Write-if-same-or-newer uses 'le'.
207 Reads from the end of the container are truncated, and missing shares
208 behave like empty ones, so to assert that a share doesn't exist (for
209 use when creating a new share), use (0, 1, 'eq', '').
211 The write vector will be applied to the given share, expanding it if
212 necessary. A write vector applied to a share number that did not
213 exist previously will cause that share to be created. Write vectors
214 must not overlap (if they do, this will either cause an error or
215 apply them in an unspecified order). Duplicate write vectors, with
216 the same offset and data, are currently tolerated but are not
219 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
220 vector whose offset is beyond the end of the current data, the space
221 between the end of the current data and the beginning of the write
222 vector will be filled with zero bytes. In earlier versions the
223 contents of this space was unspecified (and might end up containing
224 secrets). Storage servers with the new zero-filling behavior will
225 advertise a true value for the 'fills-holes-with-zero-bytes' key
226 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
229 Each write vector is accompanied by a 'new_length' argument, which
230 can be used to truncate the data. If new_length is not None and it is
231 less than the current size of the data (after applying all write
232 vectors), then the data will be truncated to new_length. If
233 new_length==0, the share will be deleted.
235 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
236 enlarge the file by sending a number larger than the size of the data
237 after applying all write vectors. That behavior was not used, and as
238 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
241 If a storage client knows that the server supports zero-filling, for
242 example from the 'fills-holes-with-zero-bytes' key in its version
243 information, it can extend the file efficiently by writing a single
244 zero byte just before the new end-of-file. Otherwise it must
245 explicitly write zeroes to all bytes between the old and new
246 end-of-file. In any case it should avoid sending new_length larger
247 than the size of the data after applying all write vectors.
249 The read vector is used to extract data from all known shares,
250 *before* any writes have been applied. The same read vector is used
251 for all shares. This captures the state that was tested by the test
252 vector, for extant shares.
254 This method returns two values: a boolean and a dict. The boolean is
255 True if the write vectors were applied, False if not. The dict is
256 keyed by share number, and each value contains a list of strings, one
257 for each element of the read vector.
259 If the write_enabler is wrong, this will raise BadWriteEnablerError.
260 To enable share migration (using update_write_enabler), the exception
261 will have the nodeid used for the old write enabler embedded in it,
262 in the following string::
264 The write enabler was recorded by nodeid '%s'.
266 Note that the nodeid here is encoded using the same base32 encoding
267 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
269 return TupleOf(bool, DictOf(int, ReadData))
271 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
272 shnum=int, reason=str):
273 """Clients who discover hash failures in shares that they have
274 downloaded from me will use this method to inform me about the
275 failures. I will record their concern so that my operator can
276 manually inspect the shares in question. I return None.
278 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
279 (binary) storage index string, and 'shnum' is the integer share
280 number. 'reason' is a human-readable explanation of the problem,
281 probably including some expected hash values and the computed ones
282 which did not match. Corruption advisories for mutable shares should
283 include a hash of the public key (the same value that appears in the
284 mutable-file verify-cap), since the current share format does not
289 class IStorageBucketWriter(Interface):
291 Objects of this kind live on the client side.
293 def put_block(segmentnum=int, data=ShareData):
294 """@param data: For most segments, this data will be 'blocksize'
295 bytes in length. The last segment might be shorter.
296 @return: a Deferred that fires (with None) when the operation completes
299 def put_plaintext_hashes(hashes=ListOf(Hash)):
301 @return: a Deferred that fires (with None) when the operation completes
304 def put_crypttext_hashes(hashes=ListOf(Hash)):
306 @return: a Deferred that fires (with None) when the operation completes
309 def put_block_hashes(blockhashes=ListOf(Hash)):
311 @return: a Deferred that fires (with None) when the operation completes
314 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
316 @return: a Deferred that fires (with None) when the operation completes
319 def put_uri_extension(data=URIExtensionData):
320 """This block of data contains integrity-checking information (hashes
321 of plaintext, crypttext, and shares), as well as encoding parameters
322 that are necessary to recover the data. This is a serialized dict
323 mapping strings to other strings. The hash of this data is kept in
324 the URI and verified before any of the data is used. All buckets for
325 a given file contain identical copies of this data.
327 The serialization format is specified with the following pseudocode:
328 for k in sorted(dict.keys()):
329 assert re.match(r'^[a-zA-Z_\-]+$', k)
330 write(k + ':' + netstring(dict[k]))
332 @return: a Deferred that fires (with None) when the operation completes
336 """Finish writing and close the bucket. The share is not finalized
337 until this method is called: if the uploading client disconnects
338 before calling close(), the partially-written share will be
341 @return: a Deferred that fires (with None) when the operation completes
344 class IStorageBucketReader(Interface):
346 def get_block_data(blocknum=int, blocksize=int, size=int):
347 """Most blocks will be the same size. The last block might be shorter
353 def get_crypttext_hashes():
355 @return: ListOf(Hash)
358 def get_block_hashes(at_least_these=SetOf(int)):
360 @return: ListOf(Hash)
363 def get_share_hashes(at_least_these=SetOf(int)):
365 @return: ListOf(TupleOf(int, Hash))
368 def get_uri_extension():
370 @return: URIExtensionData
374 class IStorageBroker(Interface):
375 def get_servers_for_psi(peer_selection_index):
377 @return: list of IServer instances
379 def get_connected_servers():
381 @return: frozenset of connected IServer instances
383 def get_known_servers():
385 @return: frozenset of IServer instances
387 def get_all_serverids():
389 @return: frozenset of serverid strings
391 def get_nickname_for_serverid(serverid):
393 @return: unicode nickname, or None
396 # methods moved from IntroducerClient, need review
397 def get_all_connections():
398 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
399 each active connection we've established to a remote service. This is
400 mostly useful for unit tests that need to wait until a certain number
401 of connections have been made."""
403 def get_all_connectors():
404 """Return a dict that maps from (nodeid, service_name) to a
405 RemoteServiceConnector instance for all services that we are actively
406 trying to connect to. Each RemoteServiceConnector has the following
409 service_name: the type of service provided, like 'storage'
410 announcement_time: when we first heard about this service
411 last_connect_time: when we last established a connection
412 last_loss_time: when we last lost a connection
414 version: the peer's version, from the most recent connection
415 oldest_supported: the peer's oldest supported version, same
417 rref: the RemoteReference, if connected, otherwise None
418 remote_host: the IAddress, if connected, otherwise None
420 This method is intended for monitoring interfaces, such as a web page
421 which describes connecting and connected peers.
424 def get_all_peerids():
425 """Return a frozenset of all peerids to whom we have a connection (to
426 one or more services) established. Mostly useful for unit tests."""
428 def get_all_connections_for(service_name):
429 """Return a frozenset of (nodeid, service_name, rref) tuples, one
430 for each active connection that provides the given SERVICE_NAME."""
432 def get_permuted_peers(service_name, key):
433 """Returns an ordered list of (peerid, rref) tuples, selecting from
434 the connections that provide SERVICE_NAME, using a hash-based
435 permutation keyed by KEY. This randomizes the service list in a
436 repeatable way, to distribute load over many peers.
440 class IDisplayableServer(Interface):
451 class IServer(IDisplayableServer):
452 """I live in the client, and represent a single server."""
453 def start_connecting(tub, trigger_cb):
457 """Once a server is connected, I return a RemoteReference.
458 Before a server is connected for the first time, I return None.
460 Note that the rref I return will start producing DeadReferenceErrors
461 once the connection is lost.
465 class IMutableSlotWriter(Interface):
467 The interface for a writer around a mutable slot on a remote server.
469 def set_checkstring(checkstring, *args):
471 Set the checkstring that I will pass to the remote server when
474 @param checkstring A packed checkstring to use.
476 Note that implementations can differ in which semantics they
477 wish to support for set_checkstring -- they can, for example,
478 build the checkstring themselves from its constituents, or
482 def get_checkstring():
484 Get the checkstring that I think currently exists on the remote
488 def put_block(data, segnum, salt):
490 Add a block and salt to the share.
493 def put_encprivkey(encprivkey):
495 Add the encrypted private key to the share.
498 def put_blockhashes(blockhashes=list):
500 Add the block hash tree to the share.
503 def put_sharehashes(sharehashes=dict):
505 Add the share hash chain to the share.
510 Return the part of the share that needs to be signed.
513 def put_signature(signature):
515 Add the signature to the share.
518 def put_verification_key(verification_key):
520 Add the verification key to the share.
523 def finish_publishing():
525 Do anything necessary to finish writing the share to a remote
526 server. I require that no further publishing needs to take place
527 after this method has been called.
531 class IURI(Interface):
532 def init_from_string(uri):
533 """Accept a string (as created by my to_string() method) and populate
534 this instance with its data. I am not normally called directly,
535 please use the module-level uri.from_string() function to convert
536 arbitrary URI strings into IURI-providing instances."""
539 """Return False if this URI be used to modify the data. Return True
540 if this URI cannot be used to modify the data."""
543 """Return True if the data can be modified by *somebody* (perhaps
544 someone who has a more powerful URI than this one)."""
546 # TODO: rename to get_read_cap()
548 """Return another IURI instance, which represents a read-only form of
549 this one. If is_readonly() is True, this returns self."""
551 def get_verify_cap():
552 """Return an instance that provides IVerifierURI, which can be used
553 to check on the availability of the file or directory, without
554 providing enough capabilities to actually read or modify the
555 contents. This may return None if the file does not need checking or
556 verification (e.g. LIT URIs).
560 """Return a string of printable ASCII characters, suitable for
561 passing into init_from_string."""
564 class IVerifierURI(Interface, IURI):
565 def init_from_string(uri):
566 """Accept a string (as created by my to_string() method) and populate
567 this instance with its data. I am not normally called directly,
568 please use the module-level uri.from_string() function to convert
569 arbitrary URI strings into IURI-providing instances."""
572 """Return a string of printable ASCII characters, suitable for
573 passing into init_from_string."""
576 class IDirnodeURI(Interface):
577 """I am a URI which represents a dirnode."""
579 class IFileURI(Interface):
580 """I am a URI which represents a filenode."""
582 """Return the length (in bytes) of the file that I represent."""
585 class IImmutableFileURI(IFileURI):
588 class IMutableFileURI(Interface):
591 class IDirectoryURI(Interface):
594 class IReadonlyDirectoryURI(Interface):
598 class CapConstraintError(Exception):
599 """A constraint on a cap was violated."""
601 class MustBeDeepImmutableError(CapConstraintError):
602 """Mutable children cannot be added to an immutable directory.
603 Also, caps obtained from an immutable directory can trigger this error
604 if they are later found to refer to a mutable object and then used."""
606 class MustBeReadonlyError(CapConstraintError):
607 """Known write caps cannot be specified in a ro_uri field. Also,
608 caps obtained from a ro_uri field can trigger this error if they
609 are later found to be write caps and then used."""
611 class MustNotBeUnknownRWError(CapConstraintError):
612 """Cannot add an unknown child cap specified in a rw_uri field."""
615 class IReadable(Interface):
616 """I represent a readable object -- either an immutable file, or a
617 specific version of a mutable file.
621 """Return True if this reference provides mutable access to the given
622 file or directory (i.e. if you can modify it), or False if not. Note
623 that even if this reference is read-only, someone else may hold a
624 read-write reference to it.
626 For an IReadable returned by get_best_readable_version(), this will
627 always return True, but for instances of subinterfaces such as
628 IMutableFileVersion, it may return False."""
631 """Return True if this file or directory is mutable (by *somebody*,
632 not necessarily you), False if it is is immutable. Note that a file
633 might be mutable overall, but your reference to it might be
634 read-only. On the other hand, all references to an immutable file
635 will be read-only; there are no read-write references to an immutable
638 def get_storage_index():
639 """Return the storage index of the file."""
642 """Return the length (in bytes) of this readable object."""
644 def download_to_data():
645 """Download all of the file contents. I return a Deferred that fires
646 with the contents as a byte string."""
648 def read(consumer, offset=0, size=None):
649 """Download a portion (possibly all) of the file's contents, making
650 them available to the given IConsumer. Return a Deferred that fires
651 (with the consumer) when the consumer is unregistered (either because
652 the last byte has been given to it, or because the consumer threw an
653 exception during write(), possibly because it no longer wants to
654 receive data). The portion downloaded will start at 'offset' and
655 contain 'size' bytes (or the remainder of the file if size==None).
657 The consumer will be used in non-streaming mode: an IPullProducer
658 will be attached to it.
660 The consumer will not receive data right away: several network trips
661 must occur first. The order of events will be::
663 consumer.registerProducer(p, streaming)
664 (if streaming == False)::
665 consumer does p.resumeProducing()
667 consumer does p.resumeProducing()
668 consumer.write(data).. (repeat until all data is written)
669 consumer.unregisterProducer()
670 deferred.callback(consumer)
672 If a download error occurs, or an exception is raised by
673 consumer.registerProducer() or consumer.write(), I will call
674 consumer.unregisterProducer() and then deliver the exception via
675 deferred.errback(). To cancel the download, the consumer should call
676 p.stopProducing(), which will result in an exception being delivered
677 via deferred.errback().
679 See src/allmydata/util/consumer.py for an example of a simple
680 download-to-memory consumer.
684 class IWriteable(Interface):
686 I define methods that callers can use to update SDMF and MDMF
687 mutable files on a Tahoe-LAFS grid.
689 # XXX: For the moment, we have only this. It is possible that we
690 # want to move overwrite() and modify() in here too.
691 def update(data, offset):
693 I write the data from my data argument to the MDMF file,
694 starting at offset. I continue writing data until my data
695 argument is exhausted, appending data to the file as necessary.
697 # assert IMutableUploadable.providedBy(data)
698 # to append data: offset=node.get_size_of_best_version()
699 # do we want to support compacting MDMF?
700 # for an MDMF file, this can be done with O(data.get_size())
701 # memory. For an SDMF file, any modification takes
702 # O(node.get_size_of_best_version()).
705 class IMutableFileVersion(IReadable):
706 """I provide access to a particular version of a mutable file. The
707 access is read/write if I was obtained from a filenode derived from
708 a write cap, or read-only if the filenode was derived from a read cap.
711 def get_sequence_number():
712 """Return the sequence number of this version."""
715 """Return the IMutableFileServerMap instance that was used to create
720 """Return this filenode's writekey, or None if the node does not have
721 write-capability. This may be used to assist with data structures
722 that need to make certain data available only to writers, such as the
723 read-write child caps in dirnodes. The recommended process is to have
724 reader-visible data be submitted to the filenode in the clear (where
725 it will be encrypted by the filenode using the readkey), but encrypt
726 writer-visible data using this writekey.
729 # TODO: Can this be overwrite instead of replace?
730 def replace(new_contents):
731 """Replace the contents of the mutable file, provided that no other
732 node has published (or is attempting to publish, concurrently) a
733 newer version of the file than this one.
735 I will avoid modifying any share that is different than the version
736 given by get_sequence_number(). However, if another node is writing
737 to the file at the same time as me, I may manage to update some shares
738 while they update others. If I see any evidence of this, I will signal
739 UncoordinatedWriteError, and the file will be left in an inconsistent
740 state (possibly the version you provided, possibly the old version,
741 possibly somebody else's version, and possibly a mix of shares from
744 The recommended response to UncoordinatedWriteError is to either
745 return it to the caller (since they failed to coordinate their
746 writes), or to attempt some sort of recovery. It may be sufficient to
747 wait a random interval (with exponential backoff) and repeat your
748 operation. If I do not signal UncoordinatedWriteError, then I was
749 able to write the new version without incident.
751 I return a Deferred that fires (with a PublishStatus object) when the
752 update has completed.
755 def modify(modifier_cb):
756 """Modify the contents of the file, by downloading this version,
757 applying the modifier function (or bound method), then uploading
758 the new version. This will succeed as long as no other node
759 publishes a version between the download and the upload.
760 I return a Deferred that fires (with a PublishStatus object) when
761 the update is complete.
763 The modifier callable will be given three arguments: a string (with
764 the old contents), a 'first_time' boolean, and a servermap. As with
765 download_to_data(), the old contents will be from this version,
766 but the modifier can use the servermap to make other decisions
767 (such as refusing to apply the delta if there are multiple parallel
768 versions, or if there is evidence of a newer unrecoverable version).
769 'first_time' will be True the first time the modifier is called,
770 and False on any subsequent calls.
772 The callable should return a string with the new contents. The
773 callable must be prepared to be called multiple times, and must
774 examine the input string to see if the change that it wants to make
775 is already present in the old version. If it does not need to make
776 any changes, it can either return None, or return its input string.
778 If the modifier raises an exception, it will be returned in the
783 # The hierarchy looks like this:
790 class IFilesystemNode(Interface):
792 """Return the strongest 'cap instance' associated with this node.
793 (writecap for writeable-mutable files/directories, readcap for
794 immutable or readonly-mutable files/directories). To convert this
795 into a string, call .to_string() on the result."""
798 """Return a readonly cap instance for this node. For immutable or
799 readonly nodes, get_cap() and get_readcap() return the same thing."""
801 def get_repair_cap():
802 """Return an IURI instance that can be used to repair the file, or
803 None if this node cannot be repaired (either because it is not
804 distributed, like a LIT file, or because the node does not represent
805 sufficient authority to create a repair-cap, like a read-only RSA
806 mutable file node [which cannot create the correct write-enablers]).
809 def get_verify_cap():
810 """Return an IVerifierURI instance that represents the
811 'verifiy/refresh capability' for this node. The holder of this
812 capability will be able to renew the lease for this node, protecting
813 it from garbage-collection. They will also be able to ask a server if
814 it holds a share for the file or directory.
818 """Return the URI string corresponding to the strongest cap associated
819 with this node. If this node is read-only, the URI will only offer
820 read-only access. If this node is read-write, the URI will offer
823 If you have read-write access to a node and wish to share merely
824 read-only access with others, use get_readonly_uri().
828 """Return the URI string that can be used by others to get write
829 access to this node, if it is writeable. If this is a read-only node,
832 def get_readonly_uri():
833 """Return the URI string that can be used by others to get read-only
834 access to this node. The result is a read-only URI, regardless of
835 whether this node is read-only or read-write.
837 If you have merely read-only access to this node, get_readonly_uri()
838 will return the same thing as get_uri().
841 def get_storage_index():
842 """Return a string with the (binary) storage index in use on this
843 download. This may be None if there is no storage index (i.e. LIT
844 files and directories)."""
847 """Return True if this reference provides mutable access to the given
848 file or directory (i.e. if you can modify it), or False if not. Note
849 that even if this reference is read-only, someone else may hold a
850 read-write reference to it."""
853 """Return True if this file or directory is mutable (by *somebody*,
854 not necessarily you), False if it is is immutable. Note that a file
855 might be mutable overall, but your reference to it might be
856 read-only. On the other hand, all references to an immutable file
857 will be read-only; there are no read-write references to an immutable
862 """Return True if this is an unknown node."""
864 def is_allowed_in_immutable_directory():
865 """Return True if this node is allowed as a child of a deep-immutable
866 directory. This is true if either the node is of a known-immutable type,
867 or it is unknown and read-only.
871 """Raise any error associated with this node."""
873 # XXX: These may not be appropriate outside the context of an IReadable.
875 """Return the length (in bytes) of the data this node represents. For
876 directory nodes, I return the size of the backing store. I return
877 synchronously and do not consult the network, so for mutable objects,
878 I will return the most recently observed size for the object, or None
879 if I don't remember a size. Use get_current_size, which returns a
880 Deferred, if you want more up-to-date information."""
882 def get_current_size():
883 """I return a Deferred that fires with the length (in bytes) of the
884 data this node represents.
888 class IFileNode(IFilesystemNode):
889 """I am a node which represents a file: a sequence of bytes. I am not a
890 container, like IDirectoryNode."""
891 def get_best_readable_version():
892 """Return a Deferred that fires with an IReadable for the 'best'
893 available version of the file. The IReadable provides only read
894 access, even if this filenode was derived from a write cap.
896 For an immutable file, there is only one version. For a mutable
897 file, the 'best' version is the recoverable version with the
898 highest sequence number. If no uncoordinated writes have occurred,
899 and if enough shares are available, then this will be the most
900 recent version that has been uploaded. If no version is recoverable,
901 the Deferred will errback with an UnrecoverableFileError.
904 def download_best_version():
905 """Download the contents of the version that would be returned
906 by get_best_readable_version(). This is equivalent to calling
907 download_to_data() on the IReadable given by that method.
909 I return a Deferred that fires with a byte string when the file
910 has been fully downloaded. To support streaming download, use
911 the 'read' method of IReadable. If no version is recoverable,
912 the Deferred will errback with an UnrecoverableFileError.
915 def get_size_of_best_version():
916 """Find the size of the version that would be returned by
917 get_best_readable_version().
919 I return a Deferred that fires with an integer. If no version
920 is recoverable, the Deferred will errback with an
921 UnrecoverableFileError.
925 class IImmutableFileNode(IFileNode, IReadable):
926 """I am a node representing an immutable file. Immutable files have
930 class IMutableFileNode(IFileNode):
931 """I provide access to a 'mutable file', which retains its identity
932 regardless of what contents are put in it.
934 The consistency-vs-availability problem means that there might be
935 multiple versions of a file present in the grid, some of which might be
936 unrecoverable (i.e. have fewer than 'k' shares). These versions are
937 loosely ordered: each has a sequence number and a hash, and any version
938 with seqnum=N was uploaded by a node which has seen at least one version
941 The 'servermap' (an instance of IMutableFileServerMap) is used to
942 describe the versions that are known to be present in the grid, and which
943 servers are hosting their shares. It is used to represent the 'state of
944 the world', and is used for this purpose by my test-and-set operations.
945 Downloading the contents of the mutable file will also return a
946 servermap. Uploading a new version into the mutable file requires a
947 servermap as input, and the semantics of the replace operation is
948 'replace the file with my new version if it looks like nobody else has
949 changed the file since my previous download'. Because the file is
950 distributed, this is not a perfect test-and-set operation, but it will do
951 its best. If the replace process sees evidence of a simultaneous write,
952 it will signal an UncoordinatedWriteError, so that the caller can take
956 Most readers will want to use the 'best' current version of the file, and
957 should use my 'download_best_version()' method.
959 To unconditionally replace the file, callers should use overwrite(). This
960 is the mode that user-visible mutable files will probably use.
962 To apply some delta to the file, call modify() with a callable modifier
963 function that can apply the modification that you want to make. This is
964 the mode that dirnodes will use, since most directory modification
965 operations can be expressed in terms of deltas to the directory state.
968 Three methods are available for users who need to perform more complex
969 operations. The first is get_servermap(), which returns an up-to-date
970 servermap using a specified mode. The second is download_version(), which
971 downloads a specific version (not necessarily the 'best' one). The third
972 is 'upload', which accepts new contents and a servermap (which must have
973 been updated with MODE_WRITE). The upload method will attempt to apply
974 the new contents as long as no other node has modified the file since the
975 servermap was updated. This might be useful to a caller who wants to
976 merge multiple versions into a single new one.
978 Note that each time the servermap is updated, a specific 'mode' is used,
979 which determines how many peers are queried. To use a servermap for my
980 replace() method, that servermap must have been updated in MODE_WRITE.
981 These modes are defined in allmydata.mutable.common, and consist of
982 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
983 allmydata/mutable/servermap.py for details about the differences.
985 Mutable files are currently limited in size (about 3.5MB max) and can
986 only be retrieved and updated all-at-once, as a single big string. Future
987 versions of our mutable files will remove this restriction.
989 def get_best_mutable_version():
990 """Return a Deferred that fires with an IMutableFileVersion for
991 the 'best' available version of the file. The best version is
992 the recoverable version with the highest sequence number. If no
993 uncoordinated writes have occurred, and if enough shares are
994 available, then this will be the most recent version that has
997 If no version is recoverable, the Deferred will errback with an
998 UnrecoverableFileError.
1001 def overwrite(new_contents):
1002 """Unconditionally replace the contents of the mutable file with new
1003 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1004 is only appropriate to use when the new contents of the file are
1005 completely unrelated to the old ones, and you do not care about other
1008 I return a Deferred that fires (with a PublishStatus object) when the
1009 update has completed.
1012 def modify(modifier_cb):
1013 """Modify the contents of the file, by downloading the current
1014 version, applying the modifier function (or bound method), then
1015 uploading the new version. I return a Deferred that fires (with a
1016 PublishStatus object) when the update is complete.
1018 The modifier callable will be given three arguments: a string (with
1019 the old contents), a 'first_time' boolean, and a servermap. As with
1020 download_best_version(), the old contents will be from the best
1021 recoverable version, but the modifier can use the servermap to make
1022 other decisions (such as refusing to apply the delta if there are
1023 multiple parallel versions, or if there is evidence of a newer
1024 unrecoverable version). 'first_time' will be True the first time the
1025 modifier is called, and False on any subsequent calls.
1027 The callable should return a string with the new contents. The
1028 callable must be prepared to be called multiple times, and must
1029 examine the input string to see if the change that it wants to make
1030 is already present in the old version. If it does not need to make
1031 any changes, it can either return None, or return its input string.
1033 If the modifier raises an exception, it will be returned in the
1037 def get_servermap(mode):
1038 """Return a Deferred that fires with an IMutableFileServerMap
1039 instance, updated using the given mode.
1042 def download_version(servermap, version):
1043 """Download a specific version of the file, using the servermap
1044 as a guide to where the shares are located.
1046 I return a Deferred that fires with the requested contents, or
1047 errbacks with UnrecoverableFileError. Note that a servermap which was
1048 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1049 all versions (those modes stop querying servers as soon as they can
1050 fulfil their goals), so you may want to use MODE_CHECK (which checks
1051 everything) to get increased visibility.
1054 def upload(new_contents, servermap):
1055 """Replace the contents of the file with new ones. This requires a
1056 servermap that was previously updated with MODE_WRITE.
1058 I attempt to provide test-and-set semantics, in that I will avoid
1059 modifying any share that is different than the version I saw in the
1060 servermap. However, if another node is writing to the file at the
1061 same time as me, I may manage to update some shares while they update
1062 others. If I see any evidence of this, I will signal
1063 UncoordinatedWriteError, and the file will be left in an inconsistent
1064 state (possibly the version you provided, possibly the old version,
1065 possibly somebody else's version, and possibly a mix of shares from
1068 The recommended response to UncoordinatedWriteError is to either
1069 return it to the caller (since they failed to coordinate their
1070 writes), or to attempt some sort of recovery. It may be sufficient to
1071 wait a random interval (with exponential backoff) and repeat your
1072 operation. If I do not signal UncoordinatedWriteError, then I was
1073 able to write the new version without incident.
1075 I return a Deferred that fires (with a PublishStatus object) when the
1076 publish has completed. I will update the servermap in-place with the
1077 location of all new shares.
1081 """Return this filenode's writekey, or None if the node does not have
1082 write-capability. This may be used to assist with data structures
1083 that need to make certain data available only to writers, such as the
1084 read-write child caps in dirnodes. The recommended process is to have
1085 reader-visible data be submitted to the filenode in the clear (where
1086 it will be encrypted by the filenode using the readkey), but encrypt
1087 writer-visible data using this writekey.
1091 """Returns the mutable file protocol version."""
1094 class NotEnoughSharesError(Exception):
1095 """Download was unable to get enough shares"""
1097 class NoSharesError(Exception):
1098 """Download was unable to get any shares at all."""
1100 class DownloadStopped(Exception):
1103 class UploadUnhappinessError(Exception):
1104 """Upload was unable to satisfy 'servers_of_happiness'"""
1106 class UnableToFetchCriticalDownloadDataError(Exception):
1107 """I was unable to fetch some piece of critical data which is supposed to
1108 be identically present in all shares."""
1110 class NoServersError(Exception):
1111 """Upload wasn't given any servers to work with, usually indicating a
1112 network or Introducer problem."""
1114 class ExistingChildError(Exception):
1115 """A directory node was asked to add or replace a child that already
1116 exists, and overwrite= was set to False."""
1118 class NoSuchChildError(Exception):
1119 """A directory node was asked to fetch a child which does not exist."""
1121 # avoid UnicodeEncodeErrors when converting to str
1122 return self.__repr__()
1124 class ChildOfWrongTypeError(Exception):
1125 """An operation was attempted on a child of the wrong type (file or directory)."""
1128 class IDirectoryNode(IFilesystemNode):
1129 """I represent a filesystem node that is a container, with a
1130 name-to-child mapping, holding the tahoe equivalent of a directory. All
1131 child names are unicode strings, and all children are some sort of
1132 IFilesystemNode (a file, subdirectory, or unknown node).
1137 The dirnode ('1') URI returned by this method can be used in
1138 set_uri() on a different directory ('2') to 'mount' a reference to
1139 this directory ('1') under the other ('2'). This URI is just a
1140 string, so it can be passed around through email or other out-of-band
1144 def get_readonly_uri():
1146 The dirnode ('1') URI returned by this method can be used in
1147 set_uri() on a different directory ('2') to 'mount' a reference to
1148 this directory ('1') under the other ('2'). This URI is just a
1149 string, so it can be passed around through email or other out-of-band
1154 """I return a Deferred that fires with a dictionary mapping child
1155 name (a unicode string) to (node, metadata_dict) tuples, in which
1156 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1159 def has_child(name):
1160 """I return a Deferred that fires with a boolean, True if there
1161 exists a child of the given name, False if not. The child name must
1162 be a unicode string."""
1165 """I return a Deferred that fires with a specific named child node,
1166 which is an IFilesystemNode. The child name must be a unicode string.
1167 I raise NoSuchChildError if I do not have a child by that name."""
1169 def get_metadata_for(name):
1170 """I return a Deferred that fires with the metadata dictionary for
1171 a specific named child node. The child name must be a unicode string.
1172 This metadata is stored in the *edge*, not in the child, so it is
1173 attached to the parent dirnode rather than the child node.
1174 I raise NoSuchChildError if I do not have a child by that name."""
1176 def set_metadata_for(name, metadata):
1177 """I replace any existing metadata for the named child with the new
1178 metadata. The child name must be a unicode string. This metadata is
1179 stored in the *edge*, not in the child, so it is attached to the
1180 parent dirnode rather than the child node. I return a Deferred
1181 (that fires with this dirnode) when the operation is complete.
1182 I raise NoSuchChildError if I do not have a child by that name."""
1184 def get_child_at_path(path):
1185 """Transform a child path into an IFilesystemNode.
1187 I perform a recursive series of 'get' operations to find the named
1188 descendant node. I return a Deferred that fires with the node, or
1189 errbacks with NoSuchChildError if the node could not be found.
1191 The path can be either a single string (slash-separated) or a list of
1192 path-name elements. All elements must be unicode strings.
1195 def get_child_and_metadata_at_path(path):
1196 """Transform a child path into an IFilesystemNode and metadata.
1198 I am like get_child_at_path(), but my Deferred fires with a tuple of
1199 (node, metadata). The metadata comes from the last edge. If the path
1200 is empty, the metadata will be an empty dictionary.
1203 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1204 """I add a child (by writecap+readcap) at the specific name. I return
1205 a Deferred that fires when the operation finishes. If overwrite= is
1206 True, I will replace any existing child of the same name, otherwise
1207 an existing child will cause me to return ExistingChildError. The
1208 child name must be a unicode string.
1210 The child caps could be for a file, or for a directory. If you have
1211 both the writecap and readcap, you should provide both arguments.
1212 If you have only one cap and don't know whether it is read-only,
1213 provide it as the writecap argument and leave the readcap as None.
1214 If you have only one cap that is known to be read-only, provide it
1215 as the readcap argument and leave the writecap as None.
1216 The filecaps are typically obtained from an IFilesystemNode with
1217 get_uri() and get_readonly_uri().
1219 If metadata= is provided, I will use it as the metadata for the named
1220 edge. This will replace any existing metadata. If metadata= is left
1221 as the default value of None, I will set ['mtime'] to the current
1222 time, and I will set ['ctime'] to the current time if there was not
1223 already a child by this name present. This roughly matches the
1224 ctime/mtime semantics of traditional filesystems. See the
1225 "About the metadata" section of webapi.txt for futher information.
1227 If this directory node is read-only, the Deferred will errback with a
1228 NotWriteableError."""
1230 def set_children(entries, overwrite=True):
1231 """Add multiple children (by writecap+readcap) to a directory node.
1232 Takes a dictionary, with childname as keys and (writecap, readcap)
1233 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1234 a Deferred that fires (with this dirnode) when the operation
1235 finishes. This is equivalent to calling set_uri() multiple times, but
1236 is much more efficient. All child names must be unicode strings.
1239 def set_node(name, child, metadata=None, overwrite=True):
1240 """I add a child at the specific name. I return a Deferred that fires
1241 when the operation finishes. This Deferred will fire with the child
1242 node that was just added. I will replace any existing child of the
1243 same name. The child name must be a unicode string. The 'child'
1244 instance must be an instance providing IFilesystemNode.
1246 If metadata= is provided, I will use it as the metadata for the named
1247 edge. This will replace any existing metadata. If metadata= is left
1248 as the default value of None, I will set ['mtime'] to the current
1249 time, and I will set ['ctime'] to the current time if there was not
1250 already a child by this name present. This roughly matches the
1251 ctime/mtime semantics of traditional filesystems. See the
1252 "About the metadata" section of webapi.txt for futher information.
1254 If this directory node is read-only, the Deferred will errback with a
1255 NotWriteableError."""
1257 def set_nodes(entries, overwrite=True):
1258 """Add multiple children to a directory node. Takes a dict mapping
1259 unicode childname to (child_node, metdata) tuples. If metdata=None,
1260 the original metadata is left unmodified. Returns a Deferred that
1261 fires (with this dirnode) when the operation finishes. This is
1262 equivalent to calling set_node() multiple times, but is much more
1265 def add_file(name, uploadable, metadata=None, overwrite=True):
1266 """I upload a file (using the given IUploadable), then attach the
1267 resulting ImmutableFileNode to the directory at the given name. I set
1268 metadata the same way as set_uri and set_node. The child name must be
1271 I return a Deferred that fires (with the IFileNode of the uploaded
1272 file) when the operation completes."""
1274 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1275 """I remove the child at the specific name. I return a Deferred that
1276 fires when the operation finishes. The child name must be a unicode
1277 string. If must_exist is True and I do not have a child by that name,
1278 I raise NoSuchChildError. If must_be_directory is True and the child
1279 is a file, or if must_be_file is True and the child is a directory,
1280 I raise ChildOfWrongTypeError."""
1282 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1283 """I create and attach a directory at the given name. The new
1284 directory can be empty, or it can be populated with children
1285 according to 'initial_children', which takes a dictionary in the same
1286 format as set_nodes (i.e. mapping unicode child name to (childnode,
1287 metadata) tuples). The child name must be a unicode string. I return
1288 a Deferred that fires (with the new directory node) when the
1289 operation finishes."""
1291 def move_child_to(current_child_name, new_parent, new_child_name=None,
1293 """I take one of my children and move them to a new parent. The child
1294 is referenced by name. On the new parent, the child will live under
1295 'new_child_name', which defaults to 'current_child_name'. TODO: what
1296 should we do about metadata? I return a Deferred that fires when the
1297 operation finishes. The child name must be a unicode string. I raise
1298 NoSuchChildError if I do not have a child by that name."""
1300 def build_manifest():
1301 """I generate a table of everything reachable from this directory.
1302 I also compute deep-stats as described below.
1304 I return a Monitor. The Monitor's results will be a dictionary with
1307 res['manifest']: a list of (path, cap) tuples for all nodes
1308 (directories and files) reachable from this one.
1309 'path' will be a tuple of unicode strings. The
1310 origin dirnode will be represented by an empty path
1312 res['verifycaps']: a list of (printable) verifycap strings, one for
1313 each reachable non-LIT node. This is a set:
1314 it will contain no duplicates.
1315 res['storage-index']: a list of (base32) storage index strings,
1316 one for each reachable non-LIT node. This is
1317 a set: it will contain no duplicates.
1318 res['stats']: a dictionary, the same that is generated by
1319 start_deep_stats() below.
1321 The Monitor will also have an .origin_si attribute with the (binary)
1322 storage index of the starting point.
1325 def start_deep_stats():
1326 """Return a Monitor, examining all nodes (directories and files)
1327 reachable from this one. The Monitor's results will be a dictionary
1328 with the following keys::
1330 count-immutable-files: count of how many CHK files are in the set
1331 count-mutable-files: same, for mutable files (does not include
1333 count-literal-files: same, for LIT files
1334 count-files: sum of the above three
1336 count-directories: count of directories
1338 size-immutable-files: total bytes for all CHK files in the set
1339 size-mutable-files (TODO): same, for current version of all mutable
1340 files, does not include directories
1341 size-literal-files: same, for LIT files
1342 size-directories: size of mutable files used by directories
1344 largest-directory: number of bytes in the largest directory
1345 largest-directory-children: number of children in the largest
1347 largest-immutable-file: number of bytes in the largest CHK file
1349 size-mutable-files is not yet implemented, because it would involve
1350 even more queries than deep_stats does.
1352 The Monitor will also have an .origin_si attribute with the (binary)
1353 storage index of the starting point.
1355 This operation will visit every directory node underneath this one,
1356 and can take a long time to run. On a typical workstation with good
1357 bandwidth, this can examine roughly 15 directories per second (and
1358 takes several minutes of 100% CPU for ~1700 directories).
1362 class ICodecEncoder(Interface):
1363 def set_params(data_size, required_shares, max_shares):
1364 """Set up the parameters of this encoder.
1366 This prepares the encoder to perform an operation that converts a
1367 single block of data into a number of shares, such that a future
1368 ICodecDecoder can use a subset of these shares to recover the
1369 original data. This operation is invoked by calling encode(). Once
1370 the encoding parameters are set up, the encode operation can be
1371 invoked multiple times.
1373 set_params() prepares the encoder to accept blocks of input data that
1374 are exactly 'data_size' bytes in length. The encoder will be prepared
1375 to produce 'max_shares' shares for each encode() operation (although
1376 see the 'desired_share_ids' to use less CPU). The encoding math will
1377 be chosen such that the decoder can get by with as few as
1378 'required_shares' of these shares and still reproduce the original
1379 data. For example, set_params(1000, 5, 5) offers no redundancy at
1380 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1382 Numerical Restrictions: 'data_size' is required to be an integral
1383 multiple of 'required_shares'. In general, the caller should choose
1384 required_shares and max_shares based upon their reliability
1385 requirements and the number of peers available (the total storage
1386 space used is roughly equal to max_shares*data_size/required_shares),
1387 then choose data_size to achieve the memory footprint desired (larger
1388 data_size means more efficient operation, smaller data_size means
1389 smaller memory footprint).
1391 In addition, 'max_shares' must be equal to or greater than
1392 'required_shares'. Of course, setting them to be equal causes
1393 encode() to degenerate into a particularly slow form of the 'split'
1396 See encode() for more details about how these parameters are used.
1398 set_params() must be called before any other ICodecEncoder methods
1403 """Return the 3-tuple of data_size, required_shares, max_shares"""
1405 def get_encoder_type():
1406 """Return a short string that describes the type of this encoder.
1408 There is required to be a global table of encoder classes. This method
1409 returns an index into this table; the value at this index is an
1410 encoder class, and this encoder is an instance of that class.
1413 def get_block_size():
1414 """Return the length of the shares that encode() will produce.
1417 def encode_proposal(data, desired_share_ids=None):
1418 """Encode some data.
1420 'data' must be a string (or other buffer object), and len(data) must
1421 be equal to the 'data_size' value passed earlier to set_params().
1423 This will return a Deferred that will fire with two lists. The first
1424 is a list of shares, each of which is a string (or other buffer
1425 object) such that len(share) is the same as what get_share_size()
1426 returned earlier. The second is a list of shareids, in which each is
1427 an integer. The lengths of the two lists will always be equal to each
1428 other. The user should take care to keep each share closely
1429 associated with its shareid, as one is useless without the other.
1431 The length of this output list will normally be the same as the value
1432 provided to the 'max_shares' parameter of set_params(). This may be
1433 different if 'desired_share_ids' is provided.
1435 'desired_share_ids', if provided, is required to be a sequence of
1436 ints, each of which is required to be >= 0 and < max_shares. If not
1437 provided, encode() will produce 'max_shares' shares, as if
1438 'desired_share_ids' were set to range(max_shares). You might use this
1439 if you initially thought you were going to use 10 peers, started
1440 encoding, and then two of the peers dropped out: you could use
1441 desired_share_ids= to skip the work (both memory and CPU) of
1442 producing shares for the peers which are no longer available.
1446 def encode(inshares, desired_share_ids=None):
1447 """Encode some data. This may be called multiple times. Each call is
1450 inshares is a sequence of length required_shares, containing buffers
1451 (i.e. strings), where each buffer contains the next contiguous
1452 non-overlapping segment of the input data. Each buffer is required to
1453 be the same length, and the sum of the lengths of the buffers is
1454 required to be exactly the data_size promised by set_params(). (This
1455 implies that the data has to be padded before being passed to
1456 encode(), unless of course it already happens to be an even multiple
1457 of required_shares in length.)
1459 Note: the requirement to break up your data into
1460 'required_shares' chunks of exactly the right length before
1461 calling encode() is surprising from point of view of a user
1462 who doesn't know how FEC works. It feels like an
1463 implementation detail that has leaked outside the abstraction
1464 barrier. Is there a use case in which the data to be encoded
1465 might already be available in pre-segmented chunks, such that
1466 it is faster or less work to make encode() take a list rather
1467 than splitting a single string?
1469 Yes, there is: suppose you are uploading a file with K=64,
1470 N=128, segsize=262,144. Then each in-share will be of size
1471 4096. If you use this .encode() API then your code could first
1472 read each successive 4096-byte chunk from the file and store
1473 each one in a Python string and store each such Python string
1474 in a Python list. Then you could call .encode(), passing that
1475 list as "inshares". The encoder would generate the other 64
1476 "secondary shares" and return to you a new list containing
1477 references to the same 64 Python strings that you passed in
1478 (as the primary shares) plus references to the new 64 Python
1481 (You could even imagine that your code could use readv() so
1482 that the operating system can arrange to get all of those
1483 bytes copied from the file into the Python list of Python
1484 strings as efficiently as possible instead of having a loop
1485 written in C or in Python to copy the next part of the file
1486 into the next string.)
1488 On the other hand if you instead use the .encode_proposal()
1489 API (above), then your code can first read in all of the
1490 262,144 bytes of the segment from the file into a Python
1491 string, then call .encode_proposal() passing the segment data
1492 as the "data" argument. The encoder would basically first
1493 split the "data" argument into a list of 64 in-shares of 4096
1494 byte each, and then do the same thing that .encode() does. So
1495 this would result in a little bit more copying of data and a
1496 little bit higher of a "maximum memory usage" during the
1497 process, although it might or might not make a practical
1498 difference for our current use cases.
1500 Note that "inshares" is a strange name for the parameter if
1501 you think of the parameter as being just for feeding in data
1502 to the codec. It makes more sense if you think of the result
1503 of this encoding as being the set of shares from inshares plus
1504 an extra set of "secondary shares" (or "check shares"). It is
1505 a surprising name! If the API is going to be surprising then
1506 the name should be surprising. If we switch to
1507 encode_proposal() above then we should also switch to an
1510 'desired_share_ids', if provided, is required to be a sequence of
1511 ints, each of which is required to be >= 0 and < max_shares. If not
1512 provided, encode() will produce 'max_shares' shares, as if
1513 'desired_share_ids' were set to range(max_shares). You might use this
1514 if you initially thought you were going to use 10 peers, started
1515 encoding, and then two of the peers dropped out: you could use
1516 desired_share_ids= to skip the work (both memory and CPU) of
1517 producing shares for the peers which are no longer available.
1519 For each call, encode() will return a Deferred that fires with two
1520 lists, one containing shares and the other containing the shareids.
1521 The get_share_size() method can be used to determine the length of
1522 the share strings returned by encode(). Each shareid is a small
1523 integer, exactly as passed into 'desired_share_ids' (or
1524 range(max_shares), if desired_share_ids was not provided).
1526 The shares and their corresponding shareids are required to be kept
1527 together during storage and retrieval. Specifically, the share data is
1528 useless by itself: the decoder needs to be told which share is which
1529 by providing it with both the shareid and the actual share data.
1531 This function will allocate an amount of memory roughly equal to::
1533 (max_shares - required_shares) * get_share_size()
1535 When combined with the memory that the caller must allocate to
1536 provide the input data, this leads to a memory footprint roughly
1537 equal to the size of the resulting encoded shares (i.e. the expansion
1538 factor times the size of the input segment).
1543 # returning a list of (shareidN,shareN) tuples instead of a pair of
1544 # lists (shareids..,shares..). Brian thought the tuples would
1545 # encourage users to keep the share and shareid together throughout
1546 # later processing, Zooko pointed out that the code to iterate
1547 # through two lists is not really more complicated than using a list
1548 # of tuples and there's also a performance improvement
1550 # having 'data_size' not required to be an integral multiple of
1551 # 'required_shares'. Doing this would require encode() to perform
1552 # padding internally, and we'd prefer to have any padding be done
1553 # explicitly by the caller. Yes, it is an abstraction leak, but
1554 # hopefully not an onerous one.
1557 class ICodecDecoder(Interface):
1558 def set_params(data_size, required_shares, max_shares):
1559 """Set the params. They have to be exactly the same ones that were
1560 used for encoding."""
1562 def get_needed_shares():
1563 """Return the number of shares needed to reconstruct the data.
1564 set_params() is required to be called before this."""
1566 def decode(some_shares, their_shareids):
1567 """Decode a partial list of shares into data.
1569 'some_shares' is required to be a sequence of buffers of sharedata, a
1570 subset of the shares returned by ICodecEncode.encode(). Each share is
1571 required to be of the same length. The i'th element of their_shareids
1572 is required to be the shareid of the i'th buffer in some_shares.
1574 This returns a Deferred which fires with a sequence of buffers. This
1575 sequence will contain all of the segments of the original data, in
1576 order. The sum of the lengths of all of the buffers will be the
1577 'data_size' value passed into the original ICodecEncode.set_params()
1578 call. To get back the single original input block of data, use
1579 ''.join(output_buffers), or you may wish to simply write them in
1580 order to an output file.
1582 Note that some of the elements in the result sequence may be
1583 references to the elements of the some_shares input sequence. In
1584 particular, this means that if those share objects are mutable (e.g.
1585 arrays) and if they are changed, then both the input (the
1586 'some_shares' parameter) and the output (the value given when the
1587 deferred is triggered) will change.
1589 The length of 'some_shares' is required to be exactly the value of
1590 'required_shares' passed into the original ICodecEncode.set_params()
1595 class IEncoder(Interface):
1596 """I take an object that provides IEncryptedUploadable, which provides
1597 encrypted data, and a list of shareholders. I then encode, hash, and
1598 deliver shares to those shareholders. I will compute all the necessary
1599 Merkle hash trees that are necessary to validate the crypttext that
1600 eventually comes back from the shareholders. I provide the URI Extension
1601 Block Hash, and the encoding parameters, both of which must be included
1604 I do not choose shareholders, that is left to the IUploader. I must be
1605 given a dict of RemoteReferences to storage buckets that are ready and
1606 willing to receive data.
1610 """Specify the number of bytes that will be encoded. This must be
1611 peformed before get_serialized_params() can be called.
1613 def set_params(params):
1614 """Override the default encoding parameters. 'params' is a tuple of
1615 (k,d,n), where 'k' is the number of required shares, 'd' is the
1616 servers_of_happiness, and 'n' is the total number of shares that will
1619 Encoding parameters can be set in three ways. 1: The Encoder class
1620 provides defaults (3/7/10). 2: the Encoder can be constructed with
1621 an 'options' dictionary, in which the
1622 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1623 set_params((k,d,n)) can be called.
1625 If you intend to use set_params(), you must call it before
1626 get_share_size or get_param are called.
1629 def set_encrypted_uploadable(u):
1630 """Provide a source of encrypted upload data. 'u' must implement
1631 IEncryptedUploadable.
1633 When this is called, the IEncryptedUploadable will be queried for its
1634 length and the storage_index that should be used.
1636 This returns a Deferred that fires with this Encoder instance.
1638 This must be performed before start() can be called.
1641 def get_param(name):
1642 """Return an encoding parameter, by name.
1644 'storage_index': return a string with the (16-byte truncated SHA-256
1645 hash) storage index to which these shares should be
1648 'share_counts': return a tuple describing how many shares are used:
1649 (needed_shares, servers_of_happiness, total_shares)
1651 'num_segments': return an int with the number of segments that
1654 'segment_size': return an int with the size of each segment.
1656 'block_size': return the size of the individual blocks that will
1657 be delivered to a shareholder's put_block() method. By
1658 knowing this, the shareholder will be able to keep all
1659 blocks in a single file and still provide random access
1660 when reading them. # TODO: can we avoid exposing this?
1662 'share_size': an int with the size of the data that will be stored
1663 on each shareholder. This is aggregate amount of data
1664 that will be sent to the shareholder, summed over all
1665 the put_block() calls I will ever make. It is useful to
1666 determine this size before asking potential
1667 shareholders whether they will grant a lease or not,
1668 since their answers will depend upon how much space we
1669 need. TODO: this might also include some amount of
1670 overhead, like the size of all the hashes. We need to
1671 decide whether this is useful or not.
1673 'serialized_params': a string with a concise description of the
1674 codec name and its parameters. This may be passed
1675 into the IUploadable to let it make sure that
1676 the same file encoded with different parameters
1677 will result in different storage indexes.
1679 Once this is called, set_size() and set_params() may not be called.
1682 def set_shareholders(shareholders, servermap):
1683 """Tell the encoder where to put the encoded shares. 'shareholders'
1684 must be a dictionary that maps share number (an integer ranging from
1685 0 to n-1) to an instance that provides IStorageBucketWriter.
1686 'servermap' is a dictionary that maps share number (as defined above)
1687 to a set of peerids. This must be performed before start() can be
1691 """Begin the encode/upload process. This involves reading encrypted
1692 data from the IEncryptedUploadable, encoding it, uploading the shares
1693 to the shareholders, then sending the hash trees.
1695 set_encrypted_uploadable() and set_shareholders() must be called
1696 before this can be invoked.
1698 This returns a Deferred that fires with a verify cap when the upload
1699 process is complete. The verifycap, plus the encryption key, is
1700 sufficient to construct the read cap.
1704 class IDecoder(Interface):
1705 """I take a list of shareholders and some setup information, then
1706 download, validate, decode, and decrypt data from them, writing the
1707 results to an output file.
1709 I do not locate the shareholders, that is left to the IDownloader. I must
1710 be given a dict of RemoteReferences to storage buckets that are ready to
1715 """I take a file-like object (providing write and close) to which all
1716 the plaintext data will be written.
1718 TODO: producer/consumer . Maybe write() should return a Deferred that
1719 indicates when it will accept more data? But probably having the
1720 IDecoder be a producer is easier to glue to IConsumer pieces.
1723 def set_shareholders(shareholders):
1724 """I take a dictionary that maps share identifiers (small integers)
1725 to RemoteReferences that provide RIBucketReader. This must be called
1729 """I start the download. This process involves retrieving data and
1730 hash chains from the shareholders, using the hashes to validate the
1731 data, decoding the shares into segments, decrypting the segments,
1732 then writing the resulting plaintext to the output file.
1734 I return a Deferred that will fire (with self) when the download is
1739 class IDownloadTarget(Interface):
1740 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1741 # will register itself as a producer. This allows the target to invoke
1742 # downloader.pauseProducing, resumeProducing, and stopProducing.
1744 """Called before any calls to write() or close(). If an error
1745 occurs before any data is available, fail() may be called without
1746 a previous call to open().
1748 'size' is the length of the file being downloaded, in bytes."""
1751 """Output some data to the target."""
1754 """Inform the target that there is no more data to be written."""
1757 """fail() is called to indicate that the download has failed. 'why'
1758 is a Failure object indicating what went wrong. No further methods
1759 will be invoked on the IDownloadTarget after fail()."""
1761 def register_canceller(cb):
1762 """The CiphertextDownloader uses this to register a no-argument function
1763 that the target can call to cancel the download. Once this canceller
1764 is invoked, no further calls to write() or close() will be made."""
1767 """When the CiphertextDownloader is done, this finish() function will be
1768 called. Whatever it returns will be returned to the invoker of
1769 Downloader.download.
1773 class IDownloader(Interface):
1774 def download(uri, target):
1775 """Perform a CHK download, sending the data to the given target.
1776 'target' must provide IDownloadTarget.
1778 Returns a Deferred that fires (with the results of target.finish)
1779 when the download is finished, or errbacks if something went wrong."""
1782 class IEncryptedUploadable(Interface):
1783 def set_upload_status(upload_status):
1784 """Provide an IUploadStatus object that should be filled with status
1785 information. The IEncryptedUploadable is responsible for setting
1786 key-determination progress ('chk'), size, storage_index, and
1787 ciphertext-fetch progress. It may delegate some of this
1788 responsibility to others, in particular to the IUploadable."""
1791 """This behaves just like IUploadable.get_size()."""
1793 def get_all_encoding_parameters():
1794 """Return a Deferred that fires with a tuple of
1795 (k,happy,n,segment_size). The segment_size will be used as-is, and
1796 must match the following constraints: it must be a multiple of k, and
1797 it shouldn't be unreasonably larger than the file size (if
1798 segment_size is larger than filesize, the difference must be stored
1801 This usually passes through to the IUploadable method of the same
1804 The encoder strictly obeys the values returned by this method. To
1805 make an upload use non-default encoding parameters, you must arrange
1806 to control the values that this method returns.
1809 def get_storage_index():
1810 """Return a Deferred that fires with a 16-byte storage index.
1813 def read_encrypted(length, hash_only):
1814 """This behaves just like IUploadable.read(), but returns crypttext
1815 instead of plaintext. If hash_only is True, then this discards the
1816 data (and returns an empty list); this improves efficiency when
1817 resuming an interrupted upload (where we need to compute the
1818 plaintext hashes, but don't need the redundant encrypted data)."""
1820 def get_plaintext_hashtree_leaves(first, last, num_segments):
1821 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1822 plaintext segments, i.e. get the tagged hashes of the given segments.
1823 The segment size is expected to be generated by the
1824 IEncryptedUploadable before any plaintext is read or ciphertext
1825 produced, so that the segment hashes can be generated with only a
1828 This returns a Deferred which fires with a sequence of hashes, using:
1830 tuple(segment_hashes[first:last])
1832 'num_segments' is used to assert that the number of segments that the
1833 IEncryptedUploadable handled matches the number of segments that the
1834 encoder was expecting.
1836 This method must not be called until the final byte has been read
1837 from read_encrypted(). Once this method is called, read_encrypted()
1838 can never be called again.
1841 def get_plaintext_hash():
1842 """OBSOLETE; Get the hash of the whole plaintext.
1844 This returns a Deferred which fires with a tagged SHA-256 hash of the
1845 whole plaintext, obtained from hashutil.plaintext_hash(data).
1849 """Just like IUploadable.close()."""
1852 class IUploadable(Interface):
1853 def set_upload_status(upload_status):
1854 """Provide an IUploadStatus object that should be filled with status
1855 information. The IUploadable is responsible for setting
1856 key-determination progress ('chk')."""
1858 def set_default_encoding_parameters(params):
1859 """Set the default encoding parameters, which must be a dict mapping
1860 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1861 'max_segment_size'. These might have an influence on the final
1862 encoding parameters returned by get_all_encoding_parameters(), if the
1863 Uploadable doesn't have more specific preferences.
1865 This call is optional: if it is not used, the Uploadable will use
1866 some built-in defaults. If used, this method must be called before
1867 any other IUploadable methods to have any effect.
1871 """Return a Deferred that will fire with the length of the data to be
1872 uploaded, in bytes. This will be called before the data is actually
1873 used, to compute encoding parameters.
1876 def get_all_encoding_parameters():
1877 """Return a Deferred that fires with a tuple of
1878 (k,happy,n,segment_size). The segment_size will be used as-is, and
1879 must match the following constraints: it must be a multiple of k, and
1880 it shouldn't be unreasonably larger than the file size (if
1881 segment_size is larger than filesize, the difference must be stored
1884 The relative values of k and n allow some IUploadables to request
1885 better redundancy than others (in exchange for consuming more space
1888 Larger values of segment_size reduce hash overhead, while smaller
1889 values reduce memory footprint and cause data to be delivered in
1890 smaller pieces (which may provide a smoother and more predictable
1891 download experience).
1893 The encoder strictly obeys the values returned by this method. To
1894 make an upload use non-default encoding parameters, you must arrange
1895 to control the values that this method returns. One way to influence
1896 them may be to call set_encoding_parameters() before calling
1897 get_all_encoding_parameters().
1900 def get_encryption_key():
1901 """Return a Deferred that fires with a 16-byte AES key. This key will
1902 be used to encrypt the data. The key will also be hashed to derive
1905 Uploadables which want to achieve convergence should hash their file
1906 contents and the serialized_encoding_parameters to form the key
1907 (which of course requires a full pass over the data). Uploadables can
1908 use the upload.ConvergentUploadMixin class to achieve this
1911 Uploadables which do not care about convergence (or do not wish to
1912 make multiple passes over the data) can simply return a
1913 strongly-random 16 byte string.
1915 get_encryption_key() may be called multiple times: the IUploadable is
1916 required to return the same value each time.
1920 """Return a Deferred that fires with a list of strings (perhaps with
1921 only a single element) which, when concatenated together, contain the
1922 next 'length' bytes of data. If EOF is near, this may provide fewer
1923 than 'length' bytes. The total number of bytes provided by read()
1924 before it signals EOF must equal the size provided by get_size().
1926 If the data must be acquired through multiple internal read
1927 operations, returning a list instead of a single string may help to
1928 reduce string copies. However, the length of the concatenated strings
1929 must equal the amount of data requested, unless EOF is encountered.
1930 Long reads, or short reads without EOF, are not allowed. read()
1931 should return the same amount of data as a local disk file read, just
1932 in a different shape and asynchronously.
1934 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1935 so a 10kB file means length=3kB, 100kB file means length=30kB,
1936 and >=1MB file means length=300kB.
1938 This method provides for a single full pass through the data. Later
1939 use cases may desire multiple passes or access to only parts of the
1940 data (such as a mutable file making small edits-in-place). This API
1941 will be expanded once those use cases are better understood.
1945 """The upload is finished, and whatever filehandle was in use may be
1949 class IMutableUploadable(Interface):
1951 I represent content that is due to be uploaded to a mutable filecap.
1953 # This is somewhat simpler than the IUploadable interface above
1954 # because mutable files do not need to be concerned with possibly
1955 # generating a CHK, nor with per-file keys. It is a subset of the
1956 # methods in IUploadable, though, so we could just as well implement
1957 # the mutable uploadables as IUploadables that don't happen to use
1958 # those methods (with the understanding that the unused methods will
1959 # never be called on such objects)
1962 Returns a Deferred that fires with the size of the content held
1968 Returns a list of strings which, when concatenated, are the next
1969 length bytes of the file, or fewer if there are fewer bytes
1970 between the current location and the end of the file.
1975 The process that used the Uploadable is finished using it, so
1976 the uploadable may be closed.
1980 class IUploadResults(Interface):
1981 """I am returned by immutable upload() methods and contain the results of
1984 Note that some of my methods return empty values (0 or an empty dict)
1985 when called for non-distributed LIT files."""
1987 def get_file_size():
1988 """Return the file size, in bytes."""
1991 """Return the (string) URI of the object uploaded, a CHK readcap."""
1993 def get_ciphertext_fetched():
1994 """Return the number of bytes fetched by the helpe for this upload,
1995 or 0 if the helper did not need to fetch any bytes (or if there was
1998 def get_preexisting_shares():
1999 """Return the number of shares that were already present in the grid."""
2001 def get_pushed_shares():
2002 """Return the number of shares that were uploaded."""
2005 """Return a dict mapping share identifier to set of IServer
2006 instances. This indicates which servers were given which shares. For
2007 immutable files, the shareid is an integer (the share number, from 0
2008 to N-1). For mutable files, it is a string of the form
2009 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
2010 the share number."""
2012 def get_servermap():
2013 """Return dict mapping IServer instance to a set of share numbers."""
2016 """Return dict of timing information, mapping name to seconds. All
2018 total : total upload time, start to finish
2019 storage_index : time to compute the storage index
2020 peer_selection : time to decide which peers will be used
2021 contacting_helper : initial helper query to upload/no-upload decision
2022 helper_total : initial helper query to helper finished pushing
2023 cumulative_fetch : helper waiting for ciphertext requests
2024 total_fetch : helper start to last ciphertext response
2025 cumulative_encoding : just time spent in zfec
2026 cumulative_sending : just time spent waiting for storage servers
2027 hashes_and_close : last segment push to shareholder close
2028 total_encode_and_push : first encode to shareholder close
2031 def get_uri_extension_data():
2032 """Return the dict of UEB data created for this file."""
2034 def get_verifycapstr():
2035 """Return the (string) verify-cap URI for the uploaded object."""
2038 class IDownloadResults(Interface):
2039 """I am created internally by download() methods. I contain a number of
2040 public attributes which contain details about the download process.::
2042 .file_size : the size of the file, in bytes
2043 .servers_used : set of server peerids that were used during download
2044 .server_problems : dict mapping server peerid to a problem string. Only
2045 servers that had problems (bad hashes, disconnects)
2047 .servermap : dict mapping server peerid to a set of share numbers. Only
2048 servers that had any shares are listed here.
2049 .timings : dict of timing information, mapping name to seconds (float)
2050 peer_selection : time to ask servers about shares
2051 servers_peer_selection : dict of peerid to DYHB-query time
2052 uri_extension : time to fetch a copy of the URI extension block
2053 hashtrees : time to fetch the hash trees
2054 segments : time to fetch, decode, and deliver segments
2055 cumulative_fetch : time spent waiting for storage servers
2056 cumulative_decode : just time spent in zfec
2057 cumulative_decrypt : just time spent in decryption
2058 total : total download time, start to finish
2059 fetch_per_server : dict of server to list of per-segment fetch times
2063 class IUploader(Interface):
2064 def upload(uploadable):
2065 """Upload the file. 'uploadable' must impement IUploadable. This
2066 returns a Deferred which fires with an IUploadResults instance, from
2067 which the URI of the file can be obtained as results.uri ."""
2069 def upload_ssk(write_capability, new_version, uploadable):
2070 """TODO: how should this work?"""
2072 class ICheckable(Interface):
2073 def check(monitor, verify=False, add_lease=False):
2074 """Check up on my health, optionally repairing any problems.
2076 This returns a Deferred that fires with an instance that provides
2077 ICheckResults, or None if the object is non-distributed (i.e. LIT
2080 The monitor will be checked periodically to see if the operation has
2081 been cancelled. If so, no new queries will be sent, and the Deferred
2082 will fire (with a OperationCancelledError) immediately.
2084 Filenodes and dirnodes (which provide IFilesystemNode) are also
2085 checkable. Instances that represent verifier-caps will be checkable
2086 but not downloadable. Some objects (like LIT files) do not actually
2087 live in the grid, and their checkers return None (non-distributed
2088 files are always healthy).
2090 If verify=False, a relatively lightweight check will be performed: I
2091 will ask all servers if they have a share for me, and I will believe
2092 whatever they say. If there are at least N distinct shares on the
2093 grid, my results will indicate r.is_healthy()==True. This requires a
2094 roundtrip to each server, but does not transfer very much data, so
2095 the network bandwidth is fairly low.
2097 If verify=True, a more resource-intensive check will be performed:
2098 every share will be downloaded, and the hashes will be validated on
2099 every bit. I will ignore any shares that failed their hash checks. If
2100 there are at least N distinct valid shares on the grid, my results
2101 will indicate r.is_healthy()==True. This requires N/k times as much
2102 download bandwidth (and server disk IO) as a regular download. If a
2103 storage server is holding a corrupt share, or is experiencing memory
2104 failures during retrieval, or is malicious or buggy, then
2105 verification will detect the problem, but checking will not.
2107 If add_lease=True, I will ensure that an up-to-date lease is present
2108 on each share. The lease secrets will be derived from by node secret
2109 (in BASEDIR/private/secret), so either I will add a new lease to the
2110 share, or I will merely renew the lease that I already had. In a
2111 future version of the storage-server protocol (once Accounting has
2112 been implemented), there may be additional options here to define the
2113 kind of lease that is obtained (which account number to claim, etc).
2115 TODO: any problems seen during checking will be reported to the
2116 health-manager.furl, a centralized object which is responsible for
2117 figuring out why files are unhealthy so corrective action can be
2121 def check_and_repair(monitor, verify=False, add_lease=False):
2122 """Like check(), but if the file/directory is not healthy, attempt to
2125 Any non-healthy result will cause an immediate repair operation, to
2126 generate and upload new shares. After repair, the file will be as
2127 healthy as we can make it. Details about what sort of repair is done
2128 will be put in the check-and-repair results. The Deferred will not
2129 fire until the repair is complete.
2131 This returns a Deferred which fires with an instance of
2132 ICheckAndRepairResults."""
2135 class IDeepCheckable(Interface):
2136 def start_deep_check(verify=False, add_lease=False):
2137 """Check upon the health of me and everything I can reach.
2139 This is a recursive form of check(), useable only on dirnodes.
2141 I return a Monitor, with results that are an IDeepCheckResults
2144 TODO: If any of the directories I traverse are unrecoverable, the
2145 Monitor will report failure. If any of the files I check upon are
2146 unrecoverable, those problems will be reported in the
2147 IDeepCheckResults as usual, and the Monitor will not report a
2151 def start_deep_check_and_repair(verify=False, add_lease=False):
2152 """Check upon the health of me and everything I can reach. Repair
2153 anything that isn't healthy.
2155 This is a recursive form of check_and_repair(), useable only on
2158 I return a Monitor, with results that are an
2159 IDeepCheckAndRepairResults object.
2161 TODO: If any of the directories I traverse are unrecoverable, the
2162 Monitor will report failure. If any of the files I check upon are
2163 unrecoverable, those problems will be reported in the
2164 IDeepCheckResults as usual, and the Monitor will not report a
2169 class ICheckResults(Interface):
2170 """I contain the detailed results of a check/verify operation.
2173 def get_storage_index():
2174 """Return a string with the (binary) storage index."""
2176 def get_storage_index_string():
2177 """Return a string with the (printable) abbreviated storage index."""
2180 """Return the (string) URI of the object that was checked."""
2183 """Return a boolean, True if the file/dir is fully healthy, False if
2184 it is damaged in any way. Non-distributed LIT files always return
2187 def is_recoverable():
2188 """Return a boolean, True if the file/dir can be recovered, False if
2189 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2190 files always return True."""
2192 def needs_rebalancing():
2193 """Return a boolean, True if the file/dirs reliability could be
2194 improved by moving shares to new servers. Non-distributed LIT files
2195 always return False."""
2197 # the following methods all return None for non-distributed LIT files
2199 def get_encoding_needed():
2200 """Return 'k', the number of shares required for recovery"""
2202 def get_encoding_expected():
2203 """Return 'N', the number of total shares generated"""
2205 def get_share_counter_good():
2206 """Return the number of distinct good shares that were found. For
2207 mutable files, this counts shares for the 'best' version."""
2209 def get_share_counter_wrong():
2210 """For mutable files, return the number of shares for versions other
2211 than the 'best' one (which is defined as being the recoverable
2212 version with the highest sequence number, then the highest roothash).
2213 These are either leftover shares from an older version (perhaps on a
2214 server that was offline when an update occurred), shares from an
2215 unrecoverable newer version, or shares from an alternate current
2216 version that results from an uncoordinated write collision. For a
2217 healthy file, this will equal 0. For immutable files, this will
2220 def get_corrupt_shares():
2221 """Return a list of 'share locators', one for each share that was
2222 found to be corrupt (integrity failure). Each share locator is a list
2223 of (IServer, storage_index, sharenum)."""
2225 def get_incompatible_shares():
2226 """Return a list of 'share locators', one for each share that was
2227 found to be of an unknown format. Each share locator is a list of
2228 (IServer, storage_index, sharenum)."""
2230 def get_servers_responding():
2231 """Return a list of IServer objects, one for each server which
2232 responded to the share query (even if they said they didn't have
2233 shares, and even if they said they did have shares but then didn't
2234 send them when asked, or dropped the connection, or returned a
2235 Failure, and even if they said they did have shares and sent
2236 incorrect ones when asked)"""
2238 def get_host_counter_good_shares():
2239 """Return the number of distinct storage servers with good shares. If
2240 this number is less than get_share_counters()[good], then some shares
2241 are doubled up, increasing the correlation of failures. This
2242 indicates that one or more shares should be moved to an otherwise
2243 unused server, if one is available.
2246 def get_version_counter_recoverable():
2247 """Return the number of recoverable versions of the file. For a
2248 healthy file, this will equal 1."""
2250 def get_version_counter_unrecoverable():
2251 """Return the number of unrecoverable versions of the file. For a
2252 healthy file, this will be 0."""
2255 """Return a dict mapping share identifier to list of IServer objects.
2256 This indicates which servers are holding which shares. For immutable
2257 files, the shareid is an integer (the share number, from 0 to N-1).
2258 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2259 containing the sequence number, the roothash, and the share number."""
2262 """Return a string with a brief (one-line) summary of the results."""
2265 """Return a list of strings with more detailed results."""
2268 class ICheckAndRepairResults(Interface):
2269 """I contain the detailed results of a check/verify/repair operation.
2271 The IFilesystemNode.check()/verify()/repair() methods all return
2272 instances that provide ICheckAndRepairResults.
2275 def get_storage_index():
2276 """Return a string with the (binary) storage index."""
2278 def get_storage_index_string():
2279 """Return a string with the (printable) abbreviated storage index."""
2281 def get_repair_attempted():
2282 """Return a boolean, True if a repair was attempted. We might not
2283 attempt to repair the file because it was healthy, or healthy enough
2284 (i.e. some shares were missing but not enough to exceed some
2285 threshold), or because we don't know how to repair this object."""
2287 def get_repair_successful():
2288 """Return a boolean, True if repair was attempted and the file/dir
2289 was fully healthy afterwards. False if no repair was attempted or if
2290 a repair attempt failed."""
2292 def get_pre_repair_results():
2293 """Return an ICheckResults instance that describes the state of the
2294 file/dir before any repair was attempted."""
2296 def get_post_repair_results():
2297 """Return an ICheckResults instance that describes the state of the
2298 file/dir after any repair was attempted. If no repair was attempted,
2299 the pre-repair and post-repair results will be identical."""
2302 class IDeepCheckResults(Interface):
2303 """I contain the results of a deep-check operation.
2305 This is returned by a call to ICheckable.deep_check().
2308 def get_root_storage_index_string():
2309 """Return the storage index (abbreviated human-readable string) of
2310 the first object checked."""
2313 """Return a dictionary with the following keys::
2315 count-objects-checked: count of how many objects were checked
2316 count-objects-healthy: how many of those objects were completely
2318 count-objects-unhealthy: how many were damaged in some way
2319 count-objects-unrecoverable: how many were unrecoverable
2320 count-corrupt-shares: how many shares were found to have
2321 corruption, summed over all objects
2325 def get_corrupt_shares():
2326 """Return a set of (IServer, storage_index, sharenum) for all shares
2327 that were found to be corrupt. storage_index is binary."""
2329 def get_all_results():
2330 """Return a dictionary mapping pathname (a tuple of strings, ready to
2331 be slash-joined) to an ICheckResults instance, one for each object
2332 that was checked."""
2334 def get_results_for_storage_index(storage_index):
2335 """Retrive the ICheckResults instance for the given (binary)
2336 storage index. Raises KeyError if there are no results for that
2340 """Return a dictionary with the same keys as
2341 IDirectoryNode.deep_stats()."""
2344 class IDeepCheckAndRepairResults(Interface):
2345 """I contain the results of a deep-check-and-repair operation.
2347 This is returned by a call to ICheckable.deep_check_and_repair().
2350 def get_root_storage_index_string():
2351 """Return the storage index (abbreviated human-readable string) of
2352 the first object checked."""
2355 """Return a dictionary with the following keys::
2357 count-objects-checked: count of how many objects were checked
2358 count-objects-healthy-pre-repair: how many of those objects were
2359 completely healthy (before any
2361 count-objects-unhealthy-pre-repair: how many were damaged in
2363 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2364 count-objects-healthy-post-repair: how many of those objects were
2365 completely healthy (after any
2367 count-objects-unhealthy-post-repair: how many were damaged in
2369 count-objects-unrecoverable-post-repair: how many were
2371 count-repairs-attempted: repairs were attempted on this many
2372 objects. The count-repairs- keys will
2373 always be provided, however unless
2374 repair=true is present, they will all
2376 count-repairs-successful: how many repairs resulted in healthy
2378 count-repairs-unsuccessful: how many repairs resulted did not
2379 results in completely healthy objects
2380 count-corrupt-shares-pre-repair: how many shares were found to
2381 have corruption, summed over all
2382 objects examined (before any
2384 count-corrupt-shares-post-repair: how many shares were found to
2385 have corruption, summed over all
2386 objects examined (after any
2391 """Return a dictionary with the same keys as
2392 IDirectoryNode.deep_stats()."""
2394 def get_corrupt_shares():
2395 """Return a set of (IServer, storage_index, sharenum) for all shares
2396 that were found to be corrupt before any repair was attempted.
2397 storage_index is binary.
2399 def get_remaining_corrupt_shares():
2400 """Return a set of (IServer, storage_index, sharenum) for all shares
2401 that were found to be corrupt after any repair was completed.
2402 storage_index is binary. These are shares that need manual inspection
2403 and probably deletion.
2405 def get_all_results():
2406 """Return a dictionary mapping pathname (a tuple of strings, ready to
2407 be slash-joined) to an ICheckAndRepairResults instance, one for each
2408 object that was checked."""
2410 def get_results_for_storage_index(storage_index):
2411 """Retrive the ICheckAndRepairResults instance for the given (binary)
2412 storage index. Raises KeyError if there are no results for that
2416 class IRepairable(Interface):
2417 def repair(check_results):
2418 """Attempt to repair the given object. Returns a Deferred that fires
2419 with a IRepairResults object.
2421 I must be called with an object that implements ICheckResults, as
2422 proof that you have actually discovered a problem with this file. I
2423 will use the data in the checker results to guide the repair process,
2424 such as which servers provided bad data and should therefore be
2425 avoided. The ICheckResults object is inside the
2426 ICheckAndRepairResults object, which is returned by the
2427 ICheckable.check() method::
2429 d = filenode.check(repair=False)
2430 def _got_results(check_and_repair_results):
2431 check_results = check_and_repair_results.get_pre_repair_results()
2432 return filenode.repair(check_results)
2433 d.addCallback(_got_results)
2438 class IRepairResults(Interface):
2439 """I contain the results of a repair operation."""
2440 def get_successful():
2441 """Returns a boolean: True if the repair made the file healthy, False
2442 if not. Repair failure generally indicates a file that has been
2443 damaged beyond repair."""
2446 class IClient(Interface):
2447 def upload(uploadable):
2448 """Upload some data into a CHK, get back the UploadResults for it.
2449 @param uploadable: something that implements IUploadable
2450 @return: a Deferred that fires with the UploadResults instance.
2451 To get the URI for this file, use results.uri .
2454 def create_mutable_file(contents=""):
2455 """Create a new mutable file (with initial) contents, get back the
2458 @param contents: (bytestring, callable, or None): this provides the
2459 initial contents of the mutable file. If 'contents' is a bytestring,
2460 it will be used as-is. If 'contents' is a callable, it will be
2461 invoked with the new MutableFileNode instance and is expected to
2462 return a bytestring with the initial contents of the file (the
2463 callable can use node.get_writekey() to decide how to encrypt the
2464 initial contents, e.g. for a brand new dirnode with initial
2465 children). contents=None is equivalent to an empty string. Using
2466 content_maker= is more efficient than creating a mutable file and
2467 setting its contents in two separate operations.
2469 @return: a Deferred that fires with an IMutableFileNode instance.
2472 def create_dirnode(initial_children={}):
2473 """Create a new unattached dirnode, possibly with initial children.
2475 @param initial_children: dict with keys that are unicode child names,
2476 and values that are (childnode, metadata) tuples.
2478 @return: a Deferred that fires with the new IDirectoryNode instance.
2481 def create_node_from_uri(uri, rouri):
2482 """Create a new IFilesystemNode instance from the uri, synchronously.
2483 @param uri: a string or IURI-providing instance, or None. This could
2484 be for a LiteralFileNode, a CHK file node, a mutable file
2485 node, or a directory node
2486 @param rouri: a string or IURI-providing instance, or None. If the
2487 main uri is None, I will use the rouri instead. If I
2488 recognize the format of the main uri, I will ignore the
2489 rouri (because it can be derived from the writecap).
2491 @return: an instance that provides IFilesystemNode (or more usefully
2492 one of its subclasses). File-specifying URIs will result in
2493 IFileNode-providing instances, like ImmutableFileNode,
2494 LiteralFileNode, or MutableFileNode. Directory-specifying
2495 URIs will result in IDirectoryNode-providing instances, like
2500 class INodeMaker(Interface):
2501 """The NodeMaker is used to create IFilesystemNode instances. It can
2502 accept a filecap/dircap string and return the node right away. It can
2503 also create new nodes (i.e. upload a file, or create a mutable file)
2504 asynchronously. Once you have one of these nodes, you can use other
2505 methods to determine whether it is a file or directory, and to download
2506 or modify its contents.
2508 The NodeMaker encapsulates all the authorities that these
2509 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2510 Tahoe process will typically have a single NodeMaker, but unit tests may
2511 create simplified/mocked forms for testing purposes.
2513 def create_from_cap(writecap, readcap=None, **kwargs):
2514 """I create an IFilesystemNode from the given writecap/readcap. I can
2515 only provide nodes for existing file/directory objects: use my other
2516 methods to create new objects. I return synchronously."""
2518 def create_mutable_file(contents=None, keysize=None):
2519 """I create a new mutable file, and return a Deferred which will fire
2520 with the IMutableFileNode instance when it is ready. If contents= is
2521 provided (a bytestring), it will be used as the initial contents of
2522 the new file, otherwise the file will contain zero bytes. keysize= is
2523 for use by unit tests, to create mutable files that are smaller than
2526 def create_new_mutable_directory(initial_children={}):
2527 """I create a new mutable directory, and return a Deferred which will
2528 fire with the IDirectoryNode instance when it is ready. If
2529 initial_children= is provided (a dict mapping unicode child name to
2530 (childnode, metadata_dict) tuples), the directory will be populated
2531 with those children, otherwise it will be empty."""
2534 class IClientStatus(Interface):
2535 def list_all_uploads():
2536 """Return a list of uploader objects, one for each upload which
2537 currently has an object available (tracked with weakrefs). This is
2538 intended for debugging purposes."""
2540 def list_active_uploads():
2541 """Return a list of active IUploadStatus objects."""
2543 def list_recent_uploads():
2544 """Return a list of IUploadStatus objects for the most recently
2547 def list_all_downloads():
2548 """Return a list of downloader objects, one for each download which
2549 currently has an object available (tracked with weakrefs). This is
2550 intended for debugging purposes."""
2552 def list_active_downloads():
2553 """Return a list of active IDownloadStatus objects."""
2555 def list_recent_downloads():
2556 """Return a list of IDownloadStatus objects for the most recently
2557 started downloads."""
2560 class IUploadStatus(Interface):
2562 """Return a timestamp (float with seconds since epoch) indicating
2563 when the operation was started."""
2565 def get_storage_index():
2566 """Return a string with the (binary) storage index in use on this
2567 upload. Returns None if the storage index has not yet been
2571 """Return an integer with the number of bytes that will eventually
2572 be uploaded for this file. Returns None if the size is not yet known.
2575 """Return True if this upload is using a Helper, False if not."""
2578 """Return a string describing the current state of the upload
2582 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2583 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2584 made towards hashing the file to determine a CHK encryption key: if
2585 non-convergent encryption is in use, this will be trivial, otherwise
2586 the whole file must be hashed. 'ciphertext' describes how much of the
2587 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2588 uploads. 'encode_and_push' describes how much of the encode-and-push
2589 process has finished: for helper uploads this is dependent upon the
2590 helper providing progress reports. It might be reasonable to add all
2591 three numbers and report the sum to the user."""
2594 """Return True if the upload is currently active, False if not."""
2597 """Return an instance of UploadResults (which contains timing and
2598 sharemap information). Might return None if the upload is not yet
2602 """Each upload status gets a unique number: this method returns that
2603 number. This provides a handle to this particular upload, so a web
2604 page can generate a suitable hyperlink."""
2607 class IDownloadStatus(Interface):
2609 """Return a timestamp (float with seconds since epoch) indicating
2610 when the operation was started."""
2612 def get_storage_index():
2613 """Return a string with the (binary) storage index in use on this
2614 download. This may be None if there is no storage index (i.e. LIT
2618 """Return an integer with the number of bytes that will eventually be
2619 retrieved for this file. Returns None if the size is not yet known.
2623 """Return True if this download is using a Helper, False if not."""
2626 """Return a string describing the current state of the download
2630 """Returns a float (from 0.0 to 1.0) describing the amount of the
2631 download that has completed. This value will remain at 0.0 until the
2632 first byte of plaintext is pushed to the download target."""
2635 """Return True if the download is currently active, False if not."""
2638 """Each download status gets a unique number: this method returns
2639 that number. This provides a handle to this particular download, so a
2640 web page can generate a suitable hyperlink."""
2643 class IServermapUpdaterStatus(Interface):
2646 class IPublishStatus(Interface):
2649 class IRetrieveStatus(Interface):
2653 class NotCapableError(Exception):
2654 """You have tried to write to a read-only node."""
2656 class BadWriteEnablerError(Exception):
2659 class RIControlClient(RemoteInterface):
2661 def wait_for_client_connections(num_clients=int):
2662 """Do not return until we have connections to at least NUM_CLIENTS
2666 def upload_from_file_to_uri(filename=str,
2667 convergence=ChoiceOf(None,
2668 StringConstraint(2**20))):
2669 """Upload a file to the grid. This accepts a filename (which must be
2670 absolute) that points to a file on the node's local disk. The node will
2671 read the contents of this file, upload it to the grid, then return the
2672 URI at which it was uploaded. If convergence is None then a random
2673 encryption key will be used, else the plaintext will be hashed, then
2674 that hash will be mixed together with the "convergence" string to form
2679 def download_from_uri_to_file(uri=URI, filename=str):
2680 """Download a file from the grid, placing it on the node's local disk
2681 at the given filename (which must be absolute[?]). Returns the
2682 absolute filename where the file was written."""
2687 def get_memory_usage():
2688 """Return a dict describes the amount of memory currently in use. The
2689 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2690 measuring memory consupmtion in bytes."""
2691 return DictOf(str, int)
2693 def speed_test(count=int, size=int, mutable=Any()):
2694 """Write 'count' tempfiles to disk, all of the given size. Measure
2695 how long (in seconds) it takes to upload them all to the servers.
2696 Then measure how long it takes to download all of them. If 'mutable'
2697 is 'create', time creation of mutable files. If 'mutable' is
2698 'upload', then time access to the same mutable file instead of
2701 Returns a tuple of (upload_time, download_time).
2703 return (float, float)
2705 def measure_peer_response_time():
2706 """Send a short message to each connected peer, and measure the time
2707 it takes for them to respond to it. This is a rough measure of the
2708 application-level round trip time.
2710 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2713 return DictOf(str, float)
2716 UploadResults = Any() #DictOf(str, str)
2719 class RIEncryptedUploadable(RemoteInterface):
2720 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2725 def get_all_encoding_parameters():
2726 return (int, int, int, long)
2728 def read_encrypted(offset=Offset, length=ReadSize):
2735 class RICHKUploadHelper(RemoteInterface):
2736 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2740 Return a dictionary of version information.
2742 return DictOf(str, Any())
2744 def upload(reader=RIEncryptedUploadable):
2745 return UploadResults
2748 class RIHelper(RemoteInterface):
2749 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2753 Return a dictionary of version information.
2755 return DictOf(str, Any())
2757 def upload_chk(si=StorageIndex):
2758 """See if a file with a given storage index needs uploading. The
2759 helper will ask the appropriate storage servers to see if the file
2760 has already been uploaded. If so, the helper will return a set of
2761 'upload results' that includes whatever hashes are needed to build
2762 the read-cap, and perhaps a truncated sharemap.
2764 If the file has not yet been uploaded (or if it was only partially
2765 uploaded), the helper will return an empty upload-results dictionary
2766 and also an RICHKUploadHelper object that will take care of the
2767 upload process. The client should call upload() on this object and
2768 pass it a reference to an RIEncryptedUploadable object that will
2769 provide ciphertext. When the upload is finished, the upload() method
2770 will finish and return the upload results.
2772 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2775 class RIStatsProvider(RemoteInterface):
2776 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2778 Provides access to statistics and monitoring information.
2783 returns a dictionary containing 'counters' and 'stats', each a
2784 dictionary with string counter/stat name keys, and numeric or None values.
2785 counters are monotonically increasing measures of work done, and
2786 stats are instantaneous measures (potentially time averaged
2789 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2792 class RIStatsGatherer(RemoteInterface):
2793 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2795 Provides a monitoring service for centralised collection of stats
2798 def provide(provider=RIStatsProvider, nickname=str):
2800 @param provider: a stats collector instance which should be polled
2801 periodically by the gatherer to collect stats.
2802 @param nickname: a name useful to identify the provided client
2807 class IStatsProducer(Interface):
2810 returns a dictionary, with str keys representing the names of stats
2811 to be monitored, and numeric values.
2814 class RIKeyGenerator(RemoteInterface):
2815 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2817 Provides a service offering to make RSA key pairs.
2820 def get_rsa_key_pair(key_size=int):
2822 @param key_size: the size of the signature key.
2823 @return: tuple(verifying_key, signing_key)
2825 return TupleOf(str, str)
2828 class FileTooLargeError(Exception):
2832 class IValidatedThingProxy(Interface):
2834 """ Acquire a thing and validate it. Return a deferred which is
2835 eventually fired with self if the thing is valid or errbacked if it
2836 can't be acquired or validated."""
2839 class InsufficientVersionError(Exception):
2840 def __init__(self, needed, got):
2841 self.needed = needed
2845 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2848 class EmptyPathnameComponentError(Exception):
2849 """The webapi disallows empty pathname components."""