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 RIStubClient(RemoteInterface):
34 """Each client publishes a service announcement for a dummy object called
35 the StubClient. This object doesn't actually offer any services, but the
36 announcement helps the Introducer keep track of which clients are
37 subscribed (so the grid admin can keep track of things like the size of
38 the grid and the client versions in use. This is the (empty)
39 RemoteInterface for the StubClient."""
41 class RIBucketWriter(RemoteInterface):
42 """ Objects of this kind live on the server side. """
43 def write(offset=Offset, data=ShareData):
48 If the data that has been written is incomplete or inconsistent then
49 the server will throw the data away, else it will store it for future
55 """Abandon all the data that has been written.
59 class RIBucketReader(RemoteInterface):
60 def read(offset=Offset, length=ReadSize):
63 def advise_corrupt_share(reason=str):
64 """Clients who discover hash failures in shares that they have
65 downloaded from me will use this method to inform me about the
66 failures. I will record their concern so that my operator can
67 manually inspect the shares in question. I return None.
69 This is a wrapper around RIStorageServer.advise_corrupt_share(),
70 which is tied to a specific share, and therefore does not need the
71 extra share-identifying arguments. Please see that method for full
75 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
76 # elements are (offset, length, operator, specimen)
77 # operator is one of "lt, le, eq, ne, ge, gt"
78 # nop always passes and is used to fetch data while writing.
79 # you should use length==len(specimen) for everything except nop
80 DataVector = ListOf(TupleOf(Offset, ShareData))
81 # (offset, data). This limits us to 30 writes of 1MiB each per call
82 TestAndWriteVectorsForShares = DictOf(int,
85 ChoiceOf(None, Offset), # new_length
87 ReadVector = ListOf(TupleOf(Offset, ReadSize))
88 ReadData = ListOf(ShareData)
89 # returns data[offset:offset+length] for each element of TestVector
91 class RIStorageServer(RemoteInterface):
92 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
96 Return a dictionary of version information.
98 return DictOf(str, Any())
100 def allocate_buckets(storage_index=StorageIndex,
101 renew_secret=LeaseRenewSecret,
102 cancel_secret=LeaseCancelSecret,
103 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
104 allocated_size=Offset, canary=Referenceable):
106 @param storage_index: the index of the bucket to be created or
108 @param sharenums: these are the share numbers (probably between 0 and
109 99) that the sender is proposing to store on this
111 @param renew_secret: This is the secret used to protect bucket refresh
112 This secret is generated by the client and
113 stored for later comparison by the server. Each
114 server is given a different secret.
115 @param cancel_secret: Like renew_secret, but protects bucket decref.
116 @param canary: If the canary is lost before close(), the bucket is
118 @return: tuple of (alreadygot, allocated), where alreadygot is what we
119 already have and allocated is what we hereby agree to accept.
120 New leases are added for shares in both lists.
122 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
123 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
125 def add_lease(storage_index=StorageIndex,
126 renew_secret=LeaseRenewSecret,
127 cancel_secret=LeaseCancelSecret):
129 Add a new lease on the given bucket. If the renew_secret matches an
130 existing lease, that lease will be renewed instead. If there is no
131 bucket for the given storage_index, return silently. (note that in
132 tahoe-1.3.0 and earlier, IndexError was raised if there was no
135 return Any() # returns None now, but future versions might change
137 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
139 Renew the lease on a given bucket, resetting the timer to 31 days.
140 Some networks will use this, some will not. If there is no bucket for
141 the given storage_index, IndexError will be raised.
143 For mutable shares, if the given renew_secret does not match an
144 existing lease, IndexError will be raised with a note listing the
145 server-nodeids on the existing leases, so leases on migrated shares
146 can be renewed or cancelled. For immutable shares, IndexError
147 (without the note) will be raised.
151 def get_buckets(storage_index=StorageIndex):
152 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
156 def slot_readv(storage_index=StorageIndex,
157 shares=ListOf(int), readv=ReadVector):
158 """Read a vector from the numbered shares associated with the given
159 storage index. An empty shares list means to return data from all
160 known shares. Returns a dictionary with one key per share."""
161 return DictOf(int, ReadData) # shnum -> results
163 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
164 secrets=TupleOf(WriteEnablerSecret,
167 tw_vectors=TestAndWriteVectorsForShares,
170 """General-purpose test-and-set operation for mutable slots. Perform
171 a bunch of comparisons against the existing shares. If they all pass,
172 then apply a bunch of write vectors to those shares. Then use the
173 read vectors to extract data from all the shares and return the data.
175 This method is, um, large. The goal is to allow clients to update all
176 the shares associated with a mutable file in a single round trip.
178 @param storage_index: the index of the bucket to be created or
180 @param write_enabler: a secret that is stored along with the slot.
181 Writes are accepted from any caller who can
182 present the matching secret. A different secret
183 should be used for each slot*server pair.
184 @param renew_secret: This is the secret used to protect bucket refresh
185 This secret is generated by the client and
186 stored for later comparison by the server. Each
187 server is given a different secret.
188 @param cancel_secret: Like renew_secret, but protects bucket decref.
190 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
191 cancel_secret). The first is required to perform any write. The
192 latter two are used when allocating new shares. To simply acquire a
193 new lease on existing shares, use an empty testv and an empty writev.
195 Each share can have a separate test vector (i.e. a list of
196 comparisons to perform). If all vectors for all shares pass, then all
197 writes for all shares are recorded. Each comparison is a 4-tuple of
198 (offset, length, operator, specimen), which effectively does a bool(
199 (read(offset, length)) OPERATOR specimen ) and only performs the
200 write if all these evaluate to True. Basic test-and-set uses 'eq'.
201 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
202 Write-if-same-or-newer uses 'le'.
204 Reads from the end of the container are truncated, and missing shares
205 behave like empty ones, so to assert that a share doesn't exist (for
206 use when creating a new share), use (0, 1, 'eq', '').
208 The write vector will be applied to the given share, expanding it if
209 necessary. A write vector applied to a share number that did not
210 exist previously will cause that share to be created.
212 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
213 vector whose offset is beyond the end of the current data, the space
214 between the end of the current data and the beginning of the write
215 vector will be filled with zero bytes. In earlier versions the
216 contents of this space was unspecified (and might end up containing
219 Each write vector is accompanied by a 'new_length' argument, which
220 can be used to truncate the data. If new_length is not None and it is
221 less than the current size of the data (after applying all write
222 vectors), then the data will be truncated to new_length. If
223 new_length==0, the share will be deleted.
225 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
226 enlarge the file by sending a number larger than the size of the data
227 after applying all write vectors. That behavior was not used, and as
228 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
231 If a storage client can rely on a server being of version v1.8.3 or
232 later, it can extend the file efficiently by writing a single zero
233 byte just before the new end-of-file. Otherwise it must explicitly
234 write zeroes to all bytes between the old and new end-of-file. In any
235 case it should avoid sending new_length larger than the size of the
236 data after applying all write vectors.
238 The read vector is used to extract data from all known shares,
239 *before* any writes have been applied. The same vector is used for
240 all shares. This captures the state that was tested by the test
243 This method returns two values: a boolean and a dict. The boolean is
244 True if the write vectors were applied, False if not. The dict is
245 keyed by share number, and each value contains a list of strings, one
246 for each element of the read vector.
248 If the write_enabler is wrong, this will raise BadWriteEnablerError.
249 To enable share migration (using update_write_enabler), the exception
250 will have the nodeid used for the old write enabler embedded in it,
251 in the following string::
253 The write enabler was recorded by nodeid '%s'.
255 Note that the nodeid here is encoded using the same base32 encoding
256 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
259 return TupleOf(bool, DictOf(int, ReadData))
261 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
262 shnum=int, reason=str):
263 """Clients who discover hash failures in shares that they have
264 downloaded from me will use this method to inform me about the
265 failures. I will record their concern so that my operator can
266 manually inspect the shares in question. I return None.
268 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
269 (binary) storage index string, and 'shnum' is the integer share
270 number. 'reason' is a human-readable explanation of the problem,
271 probably including some expected hash values and the computed ones
272 which did not match. Corruption advisories for mutable shares should
273 include a hash of the public key (the same value that appears in the
274 mutable-file verify-cap), since the current share format does not
278 class IStorageBucketWriter(Interface):
280 Objects of this kind live on the client side.
282 def put_block(segmentnum=int, data=ShareData):
283 """@param data: For most segments, this data will be 'blocksize'
284 bytes in length. The last segment might be shorter.
285 @return: a Deferred that fires (with None) when the operation completes
288 def put_plaintext_hashes(hashes=ListOf(Hash)):
290 @return: a Deferred that fires (with None) when the operation completes
293 def put_crypttext_hashes(hashes=ListOf(Hash)):
295 @return: a Deferred that fires (with None) when the operation completes
298 def put_block_hashes(blockhashes=ListOf(Hash)):
300 @return: a Deferred that fires (with None) when the operation completes
303 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
305 @return: a Deferred that fires (with None) when the operation completes
308 def put_uri_extension(data=URIExtensionData):
309 """This block of data contains integrity-checking information (hashes
310 of plaintext, crypttext, and shares), as well as encoding parameters
311 that are necessary to recover the data. This is a serialized dict
312 mapping strings to other strings. The hash of this data is kept in
313 the URI and verified before any of the data is used. All buckets for
314 a given file contain identical copies of this data.
316 The serialization format is specified with the following pseudocode:
317 for k in sorted(dict.keys()):
318 assert re.match(r'^[a-zA-Z_\-]+$', k)
319 write(k + ':' + netstring(dict[k]))
321 @return: a Deferred that fires (with None) when the operation completes
325 """Finish writing and close the bucket. The share is not finalized
326 until this method is called: if the uploading client disconnects
327 before calling close(), the partially-written share will be
330 @return: a Deferred that fires (with None) when the operation completes
333 class IStorageBucketReader(Interface):
335 def get_block_data(blocknum=int, blocksize=int, size=int):
336 """Most blocks will be the same size. The last block might be shorter
342 def get_crypttext_hashes():
344 @return: ListOf(Hash)
347 def get_block_hashes(at_least_these=SetOf(int)):
349 @return: ListOf(Hash)
352 def get_share_hashes(at_least_these=SetOf(int)):
354 @return: ListOf(TupleOf(int, Hash))
357 def get_uri_extension():
359 @return: URIExtensionData
362 class IStorageBroker(Interface):
363 def get_servers_for_psi(peer_selection_index):
365 @return: list of IServer instances
367 def get_connected_servers():
369 @return: frozenset of connected IServer instances
371 def get_known_servers():
373 @return: frozenset of IServer instances
375 def get_all_serverids():
377 @return: frozenset of serverid strings
379 def get_nickname_for_serverid(serverid):
381 @return: unicode nickname, or None
384 # methods moved from IntroducerClient, need review
385 def get_all_connections():
386 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
387 each active connection we've established to a remote service. This is
388 mostly useful for unit tests that need to wait until a certain number
389 of connections have been made."""
391 def get_all_connectors():
392 """Return a dict that maps from (nodeid, service_name) to a
393 RemoteServiceConnector instance for all services that we are actively
394 trying to connect to. Each RemoteServiceConnector has the following
397 service_name: the type of service provided, like 'storage'
398 announcement_time: when we first heard about this service
399 last_connect_time: when we last established a connection
400 last_loss_time: when we last lost a connection
402 version: the peer's version, from the most recent connection
403 oldest_supported: the peer's oldest supported version, same
405 rref: the RemoteReference, if connected, otherwise None
406 remote_host: the IAddress, if connected, otherwise None
408 This method is intended for monitoring interfaces, such as a web page
409 which describes connecting and connected peers.
412 def get_all_peerids():
413 """Return a frozenset of all peerids to whom we have a connection (to
414 one or more services) established. Mostly useful for unit tests."""
416 def get_all_connections_for(service_name):
417 """Return a frozenset of (nodeid, service_name, rref) tuples, one
418 for each active connection that provides the given SERVICE_NAME."""
420 def get_permuted_peers(service_name, key):
421 """Returns an ordered list of (peerid, rref) tuples, selecting from
422 the connections that provide SERVICE_NAME, using a hash-based
423 permutation keyed by KEY. This randomizes the service list in a
424 repeatable way, to distribute load over many peers.
428 class IMutableSlotWriter(Interface):
430 The interface for a writer around a mutable slot on a remote server.
432 def set_checkstring(checkstring, *args):
434 Set the checkstring that I will pass to the remote server when
437 @param checkstring A packed checkstring to use.
439 Note that implementations can differ in which semantics they
440 wish to support for set_checkstring -- they can, for example,
441 build the checkstring themselves from its constituents, or
445 def get_checkstring():
447 Get the checkstring that I think currently exists on the remote
451 def put_block(data, segnum, salt):
453 Add a block and salt to the share.
456 def put_encprivey(encprivkey):
458 Add the encrypted private key to the share.
461 def put_blockhashes(blockhashes=list):
463 Add the block hash tree to the share.
466 def put_sharehashes(sharehashes=dict):
468 Add the share hash chain to the share.
473 Return the part of the share that needs to be signed.
476 def put_signature(signature):
478 Add the signature to the share.
481 def put_verification_key(verification_key):
483 Add the verification key to the share.
486 def finish_publishing():
488 Do anything necessary to finish writing the share to a remote
489 server. I require that no further publishing needs to take place
490 after this method has been called.
494 class IURI(Interface):
495 def init_from_string(uri):
496 """Accept a string (as created by my to_string() method) and populate
497 this instance with its data. I am not normally called directly,
498 please use the module-level uri.from_string() function to convert
499 arbitrary URI strings into IURI-providing instances."""
502 """Return False if this URI be used to modify the data. Return True
503 if this URI cannot be used to modify the data."""
506 """Return True if the data can be modified by *somebody* (perhaps
507 someone who has a more powerful URI than this one)."""
509 # TODO: rename to get_read_cap()
511 """Return another IURI instance, which represents a read-only form of
512 this one. If is_readonly() is True, this returns self."""
514 def get_verify_cap():
515 """Return an instance that provides IVerifierURI, which can be used
516 to check on the availability of the file or directory, without
517 providing enough capabilities to actually read or modify the
518 contents. This may return None if the file does not need checking or
519 verification (e.g. LIT URIs).
523 """Return a string of printable ASCII characters, suitable for
524 passing into init_from_string."""
526 class IVerifierURI(Interface, IURI):
527 def init_from_string(uri):
528 """Accept a string (as created by my to_string() method) and populate
529 this instance with its data. I am not normally called directly,
530 please use the module-level uri.from_string() function to convert
531 arbitrary URI strings into IURI-providing instances."""
534 """Return a string of printable ASCII characters, suitable for
535 passing into init_from_string."""
537 class IDirnodeURI(Interface):
538 """I am a URI which represents a dirnode."""
540 class IFileURI(Interface):
541 """I am a URI which represents a filenode."""
543 """Return the length (in bytes) of the file that I represent."""
545 class IImmutableFileURI(IFileURI):
548 class IMutableFileURI(Interface):
549 """I am a URI which represents a mutable filenode."""
550 def get_extension_params():
551 """Return the extension parameters in the URI"""
553 def set_extension_params():
554 """Set the extension parameters that should be in the URI"""
556 class IDirectoryURI(Interface):
559 class IReadonlyDirectoryURI(Interface):
562 class CapConstraintError(Exception):
563 """A constraint on a cap was violated."""
565 class MustBeDeepImmutableError(CapConstraintError):
566 """Mutable children cannot be added to an immutable directory.
567 Also, caps obtained from an immutable directory can trigger this error
568 if they are later found to refer to a mutable object and then used."""
570 class MustBeReadonlyError(CapConstraintError):
571 """Known write caps cannot be specified in a ro_uri field. Also,
572 caps obtained from a ro_uri field can trigger this error if they
573 are later found to be write caps and then used."""
575 class MustNotBeUnknownRWError(CapConstraintError):
576 """Cannot add an unknown child cap specified in a rw_uri field."""
579 class IReadable(Interface):
580 """I represent a readable object -- either an immutable file, or a
581 specific version of a mutable file.
585 """Return True if this reference provides mutable access to the given
586 file or directory (i.e. if you can modify it), or False if not. Note
587 that even if this reference is read-only, someone else may hold a
588 read-write reference to it.
590 For an IReadable returned by get_best_readable_version(), this will
591 always return True, but for instances of subinterfaces such as
592 IMutableFileVersion, it may return False."""
595 """Return True if this file or directory is mutable (by *somebody*,
596 not necessarily you), False if it is is immutable. Note that a file
597 might be mutable overall, but your reference to it might be
598 read-only. On the other hand, all references to an immutable file
599 will be read-only; there are no read-write references to an immutable
602 def get_storage_index():
603 """Return the storage index of the file."""
606 """Return the length (in bytes) of this readable object."""
608 def download_to_data():
609 """Download all of the file contents. I return a Deferred that fires
610 with the contents as a byte string."""
612 def read(consumer, offset=0, size=None):
613 """Download a portion (possibly all) of the file's contents, making
614 them available to the given IConsumer. Return a Deferred that fires
615 (with the consumer) when the consumer is unregistered (either because
616 the last byte has been given to it, or because the consumer threw an
617 exception during write(), possibly because it no longer wants to
618 receive data). The portion downloaded will start at 'offset' and
619 contain 'size' bytes (or the remainder of the file if size==None).
621 The consumer will be used in non-streaming mode: an IPullProducer
622 will be attached to it.
624 The consumer will not receive data right away: several network trips
625 must occur first. The order of events will be::
627 consumer.registerProducer(p, streaming)
628 (if streaming == False)::
629 consumer does p.resumeProducing()
631 consumer does p.resumeProducing()
632 consumer.write(data).. (repeat until all data is written)
633 consumer.unregisterProducer()
634 deferred.callback(consumer)
636 If a download error occurs, or an exception is raised by
637 consumer.registerProducer() or consumer.write(), I will call
638 consumer.unregisterProducer() and then deliver the exception via
639 deferred.errback(). To cancel the download, the consumer should call
640 p.stopProducing(), which will result in an exception being delivered
641 via deferred.errback().
643 See src/allmydata/util/consumer.py for an example of a simple
644 download-to-memory consumer.
648 class IWriteable(Interface):
650 I define methods that callers can use to update SDMF and MDMF
651 mutable files on a Tahoe-LAFS grid.
653 # XXX: For the moment, we have only this. It is possible that we
654 # want to move overwrite() and modify() in here too.
655 def update(data, offset):
657 I write the data from my data argument to the MDMF file,
658 starting at offset. I continue writing data until my data
659 argument is exhausted, appending data to the file as necessary.
661 # assert IMutableUploadable.providedBy(data)
662 # to append data: offset=node.get_size_of_best_version()
663 # do we want to support compacting MDMF?
664 # for an MDMF file, this can be done with O(data.get_size())
665 # memory. For an SDMF file, any modification takes
666 # O(node.get_size_of_best_version()).
669 class IMutableFileVersion(IReadable):
670 """I provide access to a particular version of a mutable file. The
671 access is read/write if I was obtained from a filenode derived from
672 a write cap, or read-only if the filenode was derived from a read cap.
675 def get_sequence_number():
676 """Return the sequence number of this version."""
679 """Return the IMutableFileServerMap instance that was used to create
684 """Return this filenode's writekey, or None if the node does not have
685 write-capability. This may be used to assist with data structures
686 that need to make certain data available only to writers, such as the
687 read-write child caps in dirnodes. The recommended process is to have
688 reader-visible data be submitted to the filenode in the clear (where
689 it will be encrypted by the filenode using the readkey), but encrypt
690 writer-visible data using this writekey.
693 # TODO: Can this be overwrite instead of replace?
694 def replace(new_contents):
695 """Replace the contents of the mutable file, provided that no other
696 node has published (or is attempting to publish, concurrently) a
697 newer version of the file than this one.
699 I will avoid modifying any share that is different than the version
700 given by get_sequence_number(). However, if another node is writing
701 to the file at the same time as me, I may manage to update some shares
702 while they update others. If I see any evidence of this, I will signal
703 UncoordinatedWriteError, and the file will be left in an inconsistent
704 state (possibly the version you provided, possibly the old version,
705 possibly somebody else's version, and possibly a mix of shares from
708 The recommended response to UncoordinatedWriteError is to either
709 return it to the caller (since they failed to coordinate their
710 writes), or to attempt some sort of recovery. It may be sufficient to
711 wait a random interval (with exponential backoff) and repeat your
712 operation. If I do not signal UncoordinatedWriteError, then I was
713 able to write the new version without incident.
715 I return a Deferred that fires (with a PublishStatus object) when the
716 update has completed.
719 def modify(modifier_cb):
720 """Modify the contents of the file, by downloading this version,
721 applying the modifier function (or bound method), then uploading
722 the new version. This will succeed as long as no other node
723 publishes a version between the download and the upload.
724 I return a Deferred that fires (with a PublishStatus object) when
725 the update is complete.
727 The modifier callable will be given three arguments: a string (with
728 the old contents), a 'first_time' boolean, and a servermap. As with
729 download_to_data(), the old contents will be from this version,
730 but the modifier can use the servermap to make other decisions
731 (such as refusing to apply the delta if there are multiple parallel
732 versions, or if there is evidence of a newer unrecoverable version).
733 'first_time' will be True the first time the modifier is called,
734 and False on any subsequent calls.
736 The callable should return a string with the new contents. The
737 callable must be prepared to be called multiple times, and must
738 examine the input string to see if the change that it wants to make
739 is already present in the old version. If it does not need to make
740 any changes, it can either return None, or return its input string.
742 If the modifier raises an exception, it will be returned in the
747 # The hierarchy looks like this:
754 class IFilesystemNode(Interface):
756 """Return the strongest 'cap instance' associated with this node.
757 (writecap for writeable-mutable files/directories, readcap for
758 immutable or readonly-mutable files/directories). To convert this
759 into a string, call .to_string() on the result."""
762 """Return a readonly cap instance for this node. For immutable or
763 readonly nodes, get_cap() and get_readcap() return the same thing."""
765 def get_repair_cap():
766 """Return an IURI instance that can be used to repair the file, or
767 None if this node cannot be repaired (either because it is not
768 distributed, like a LIT file, or because the node does not represent
769 sufficient authority to create a repair-cap, like a read-only RSA
770 mutable file node [which cannot create the correct write-enablers]).
773 def get_verify_cap():
774 """Return an IVerifierURI instance that represents the
775 'verifiy/refresh capability' for this node. The holder of this
776 capability will be able to renew the lease for this node, protecting
777 it from garbage-collection. They will also be able to ask a server if
778 it holds a share for the file or directory.
782 """Return the URI string corresponding to the strongest cap associated
783 with this node. If this node is read-only, the URI will only offer
784 read-only access. If this node is read-write, the URI will offer
787 If you have read-write access to a node and wish to share merely
788 read-only access with others, use get_readonly_uri().
792 """Return the URI string that can be used by others to get write
793 access to this node, if it is writeable. If this is a read-only node,
796 def get_readonly_uri():
797 """Return the URI string that can be used by others to get read-only
798 access to this node. The result is a read-only URI, regardless of
799 whether this node is read-only or read-write.
801 If you have merely read-only access to this node, get_readonly_uri()
802 will return the same thing as get_uri().
805 def get_storage_index():
806 """Return a string with the (binary) storage index in use on this
807 download. This may be None if there is no storage index (i.e. LIT
808 files and directories)."""
811 """Return True if this reference provides mutable access to the given
812 file or directory (i.e. if you can modify it), or False if not. Note
813 that even if this reference is read-only, someone else may hold a
814 read-write reference to it."""
817 """Return True if this file or directory is mutable (by *somebody*,
818 not necessarily you), False if it is is immutable. Note that a file
819 might be mutable overall, but your reference to it might be
820 read-only. On the other hand, all references to an immutable file
821 will be read-only; there are no read-write references to an immutable
826 """Return True if this is an unknown node."""
828 def is_allowed_in_immutable_directory():
829 """Return True if this node is allowed as a child of a deep-immutable
830 directory. This is true if either the node is of a known-immutable type,
831 or it is unknown and read-only.
835 """Raise any error associated with this node."""
837 # XXX: These may not be appropriate outside the context of an IReadable.
839 """Return the length (in bytes) of the data this node represents. For
840 directory nodes, I return the size of the backing store. I return
841 synchronously and do not consult the network, so for mutable objects,
842 I will return the most recently observed size for the object, or None
843 if I don't remember a size. Use get_current_size, which returns a
844 Deferred, if you want more up-to-date information."""
846 def get_current_size():
847 """I return a Deferred that fires with the length (in bytes) of the
848 data this node represents.
851 class IFileNode(IFilesystemNode):
852 """I am a node which represents a file: a sequence of bytes. I am not a
853 container, like IDirectoryNode."""
854 def get_best_readable_version():
855 """Return a Deferred that fires with an IReadable for the 'best'
856 available version of the file. The IReadable provides only read
857 access, even if this filenode was derived from a write cap.
859 For an immutable file, there is only one version. For a mutable
860 file, the 'best' version is the recoverable version with the
861 highest sequence number. If no uncoordinated writes have occurred,
862 and if enough shares are available, then this will be the most
863 recent version that has been uploaded. If no version is recoverable,
864 the Deferred will errback with an UnrecoverableFileError.
867 def download_best_version():
868 """Download the contents of the version that would be returned
869 by get_best_readable_version(). This is equivalent to calling
870 download_to_data() on the IReadable given by that method.
872 I return a Deferred that fires with a byte string when the file
873 has been fully downloaded. To support streaming download, use
874 the 'read' method of IReadable. If no version is recoverable,
875 the Deferred will errback with an UnrecoverableFileError.
878 def get_size_of_best_version():
879 """Find the size of the version that would be returned by
880 get_best_readable_version().
882 I return a Deferred that fires with an integer. If no version
883 is recoverable, the Deferred will errback with an
884 UnrecoverableFileError.
888 class IImmutableFileNode(IFileNode, IReadable):
889 """I am a node representing an immutable file. Immutable files have
893 class IMutableFileNode(IFileNode):
894 """I provide access to a 'mutable file', which retains its identity
895 regardless of what contents are put in it.
897 The consistency-vs-availability problem means that there might be
898 multiple versions of a file present in the grid, some of which might be
899 unrecoverable (i.e. have fewer than 'k' shares). These versions are
900 loosely ordered: each has a sequence number and a hash, and any version
901 with seqnum=N was uploaded by a node which has seen at least one version
904 The 'servermap' (an instance of IMutableFileServerMap) is used to
905 describe the versions that are known to be present in the grid, and which
906 servers are hosting their shares. It is used to represent the 'state of
907 the world', and is used for this purpose by my test-and-set operations.
908 Downloading the contents of the mutable file will also return a
909 servermap. Uploading a new version into the mutable file requires a
910 servermap as input, and the semantics of the replace operation is
911 'replace the file with my new version if it looks like nobody else has
912 changed the file since my previous download'. Because the file is
913 distributed, this is not a perfect test-and-set operation, but it will do
914 its best. If the replace process sees evidence of a simultaneous write,
915 it will signal an UncoordinatedWriteError, so that the caller can take
919 Most readers will want to use the 'best' current version of the file, and
920 should use my 'download_best_version()' method.
922 To unconditionally replace the file, callers should use overwrite(). This
923 is the mode that user-visible mutable files will probably use.
925 To apply some delta to the file, call modify() with a callable modifier
926 function that can apply the modification that you want to make. This is
927 the mode that dirnodes will use, since most directory modification
928 operations can be expressed in terms of deltas to the directory state.
931 Three methods are available for users who need to perform more complex
932 operations. The first is get_servermap(), which returns an up-to-date
933 servermap using a specified mode. The second is download_version(), which
934 downloads a specific version (not necessarily the 'best' one). The third
935 is 'upload', which accepts new contents and a servermap (which must have
936 been updated with MODE_WRITE). The upload method will attempt to apply
937 the new contents as long as no other node has modified the file since the
938 servermap was updated. This might be useful to a caller who wants to
939 merge multiple versions into a single new one.
941 Note that each time the servermap is updated, a specific 'mode' is used,
942 which determines how many peers are queried. To use a servermap for my
943 replace() method, that servermap must have been updated in MODE_WRITE.
944 These modes are defined in allmydata.mutable.common, and consist of
945 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
946 allmydata/mutable/servermap.py for details about the differences.
948 Mutable files are currently limited in size (about 3.5MB max) and can
949 only be retrieved and updated all-at-once, as a single big string. Future
950 versions of our mutable files will remove this restriction.
952 def get_best_mutable_version():
953 """Return a Deferred that fires with an IMutableFileVersion for
954 the 'best' available version of the file. The best version is
955 the recoverable version with the highest sequence number. If no
956 uncoordinated writes have occurred, and if enough shares are
957 available, then this will be the most recent version that has
960 If no version is recoverable, the Deferred will errback with an
961 UnrecoverableFileError.
964 def overwrite(new_contents):
965 """Unconditionally replace the contents of the mutable file with new
966 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
967 is only appropriate to use when the new contents of the file are
968 completely unrelated to the old ones, and you do not care about other
971 I return a Deferred that fires (with a PublishStatus object) when the
972 update has completed.
975 def modify(modifier_cb):
976 """Modify the contents of the file, by downloading the current
977 version, applying the modifier function (or bound method), then
978 uploading the new version. I return a Deferred that fires (with a
979 PublishStatus object) when the update is complete.
981 The modifier callable will be given three arguments: a string (with
982 the old contents), a 'first_time' boolean, and a servermap. As with
983 download_best_version(), the old contents will be from the best
984 recoverable version, but the modifier can use the servermap to make
985 other decisions (such as refusing to apply the delta if there are
986 multiple parallel versions, or if there is evidence of a newer
987 unrecoverable version). 'first_time' will be True the first time the
988 modifier is called, and False on any subsequent calls.
990 The callable should return a string with the new contents. The
991 callable must be prepared to be called multiple times, and must
992 examine the input string to see if the change that it wants to make
993 is already present in the old version. If it does not need to make
994 any changes, it can either return None, or return its input string.
996 If the modifier raises an exception, it will be returned in the
1000 def get_servermap(mode):
1001 """Return a Deferred that fires with an IMutableFileServerMap
1002 instance, updated using the given mode.
1005 def download_version(servermap, version):
1006 """Download a specific version of the file, using the servermap
1007 as a guide to where the shares are located.
1009 I return a Deferred that fires with the requested contents, or
1010 errbacks with UnrecoverableFileError. Note that a servermap which was
1011 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1012 all versions (those modes stop querying servers as soon as they can
1013 fulfil their goals), so you may want to use MODE_CHECK (which checks
1014 everything) to get increased visibility.
1017 def upload(new_contents, servermap):
1018 """Replace the contents of the file with new ones. This requires a
1019 servermap that was previously updated with MODE_WRITE.
1021 I attempt to provide test-and-set semantics, in that I will avoid
1022 modifying any share that is different than the version I saw in the
1023 servermap. However, if another node is writing to the file at the
1024 same time as me, I may manage to update some shares while they update
1025 others. If I see any evidence of this, I will signal
1026 UncoordinatedWriteError, and the file will be left in an inconsistent
1027 state (possibly the version you provided, possibly the old version,
1028 possibly somebody else's version, and possibly a mix of shares from
1031 The recommended response to UncoordinatedWriteError is to either
1032 return it to the caller (since they failed to coordinate their
1033 writes), or to attempt some sort of recovery. It may be sufficient to
1034 wait a random interval (with exponential backoff) and repeat your
1035 operation. If I do not signal UncoordinatedWriteError, then I was
1036 able to write the new version without incident.
1038 I return a Deferred that fires (with a PublishStatus object) when the
1039 publish has completed. I will update the servermap in-place with the
1040 location of all new shares.
1044 """Return this filenode's writekey, or None if the node does not have
1045 write-capability. This may be used to assist with data structures
1046 that need to make certain data available only to writers, such as the
1047 read-write child caps in dirnodes. The recommended process is to have
1048 reader-visible data be submitted to the filenode in the clear (where
1049 it will be encrypted by the filenode using the readkey), but encrypt
1050 writer-visible data using this writekey.
1054 """Returns the mutable file protocol version."""
1056 class NotEnoughSharesError(Exception):
1057 """Download was unable to get enough shares"""
1059 class NoSharesError(Exception):
1060 """Download was unable to get any shares at all."""
1062 class DownloadStopped(Exception):
1065 class UploadUnhappinessError(Exception):
1066 """Upload was unable to satisfy 'servers_of_happiness'"""
1068 class UnableToFetchCriticalDownloadDataError(Exception):
1069 """I was unable to fetch some piece of critical data which is supposed to
1070 be identically present in all shares."""
1072 class NoServersError(Exception):
1073 """Upload wasn't given any servers to work with, usually indicating a
1074 network or Introducer problem."""
1076 class ExistingChildError(Exception):
1077 """A directory node was asked to add or replace a child that already
1078 exists, and overwrite= was set to False."""
1080 class NoSuchChildError(Exception):
1081 """A directory node was asked to fetch a child which does not exist."""
1083 class ChildOfWrongTypeError(Exception):
1084 """An operation was attempted on a child of the wrong type (file or directory)."""
1086 class IDirectoryNode(IFilesystemNode):
1087 """I represent a filesystem node that is a container, with a
1088 name-to-child mapping, holding the tahoe equivalent of a directory. All
1089 child names are unicode strings, and all children are some sort of
1090 IFilesystemNode (a file, subdirectory, or unknown node).
1095 The dirnode ('1') URI returned by this method can be used in
1096 set_uri() on a different directory ('2') to 'mount' a reference to
1097 this directory ('1') under the other ('2'). This URI is just a
1098 string, so it can be passed around through email or other out-of-band
1102 def get_readonly_uri():
1104 The dirnode ('1') URI returned by this method can be used in
1105 set_uri() on a different directory ('2') to 'mount' a reference to
1106 this directory ('1') under the other ('2'). This URI is just a
1107 string, so it can be passed around through email or other out-of-band
1112 """I return a Deferred that fires with a dictionary mapping child
1113 name (a unicode string) to (node, metadata_dict) tuples, in which
1114 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1117 def has_child(name):
1118 """I return a Deferred that fires with a boolean, True if there
1119 exists a child of the given name, False if not. The child name must
1120 be a unicode string."""
1123 """I return a Deferred that fires with a specific named child node,
1124 which is an IFilesystemNode. The child name must be a unicode string.
1125 I raise NoSuchChildError if I do not have a child by that name."""
1127 def get_metadata_for(name):
1128 """I return a Deferred that fires with the metadata dictionary for
1129 a specific named child node. The child name must be a unicode string.
1130 This metadata is stored in the *edge*, not in the child, so it is
1131 attached to the parent dirnode rather than the child node.
1132 I raise NoSuchChildError if I do not have a child by that name."""
1134 def set_metadata_for(name, metadata):
1135 """I replace any existing metadata for the named child with the new
1136 metadata. The child name must be a unicode string. This metadata is
1137 stored in the *edge*, not in the child, so it is attached to the
1138 parent dirnode rather than the child node. I return a Deferred
1139 (that fires with this dirnode) when the operation is complete.
1140 I raise NoSuchChildError if I do not have a child by that name."""
1142 def get_child_at_path(path):
1143 """Transform a child path into an IFilesystemNode.
1145 I perform a recursive series of 'get' operations to find the named
1146 descendant node. I return a Deferred that fires with the node, or
1147 errbacks with NoSuchChildError if the node could not be found.
1149 The path can be either a single string (slash-separated) or a list of
1150 path-name elements. All elements must be unicode strings.
1153 def get_child_and_metadata_at_path(path):
1154 """Transform a child path into an IFilesystemNode and metadata.
1156 I am like get_child_at_path(), but my Deferred fires with a tuple of
1157 (node, metadata). The metadata comes from the last edge. If the path
1158 is empty, the metadata will be an empty dictionary.
1161 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1162 """I add a child (by writecap+readcap) at the specific name. I return
1163 a Deferred that fires when the operation finishes. If overwrite= is
1164 True, I will replace any existing child of the same name, otherwise
1165 an existing child will cause me to return ExistingChildError. The
1166 child name must be a unicode string.
1168 The child caps could be for a file, or for a directory. If you have
1169 both the writecap and readcap, you should provide both arguments.
1170 If you have only one cap and don't know whether it is read-only,
1171 provide it as the writecap argument and leave the readcap as None.
1172 If you have only one cap that is known to be read-only, provide it
1173 as the readcap argument and leave the writecap as None.
1174 The filecaps are typically obtained from an IFilesystemNode with
1175 get_uri() and get_readonly_uri().
1177 If metadata= is provided, I will use it as the metadata for the named
1178 edge. This will replace any existing metadata. If metadata= is left
1179 as the default value of None, I will set ['mtime'] to the current
1180 time, and I will set ['ctime'] to the current time if there was not
1181 already a child by this name present. This roughly matches the
1182 ctime/mtime semantics of traditional filesystems. See the
1183 "About the metadata" section of webapi.txt for futher information.
1185 If this directory node is read-only, the Deferred will errback with a
1186 NotWriteableError."""
1188 def set_children(entries, overwrite=True):
1189 """Add multiple children (by writecap+readcap) to a directory node.
1190 Takes a dictionary, with childname as keys and (writecap, readcap)
1191 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1192 a Deferred that fires (with this dirnode) when the operation
1193 finishes. This is equivalent to calling set_uri() multiple times, but
1194 is much more efficient. All child names must be unicode strings.
1197 def set_node(name, child, metadata=None, overwrite=True):
1198 """I add a child at the specific name. I return a Deferred that fires
1199 when the operation finishes. This Deferred will fire with the child
1200 node that was just added. I will replace any existing child of the
1201 same name. The child name must be a unicode string. The 'child'
1202 instance must be an instance providing IFilesystemNode.
1204 If metadata= is provided, I will use it as the metadata for the named
1205 edge. This will replace any existing metadata. If metadata= is left
1206 as the default value of None, I will set ['mtime'] to the current
1207 time, and I will set ['ctime'] to the current time if there was not
1208 already a child by this name present. This roughly matches the
1209 ctime/mtime semantics of traditional filesystems. See the
1210 "About the metadata" section of webapi.txt for futher information.
1212 If this directory node is read-only, the Deferred will errback with a
1213 NotWriteableError."""
1215 def set_nodes(entries, overwrite=True):
1216 """Add multiple children to a directory node. Takes a dict mapping
1217 unicode childname to (child_node, metdata) tuples. If metdata=None,
1218 the original metadata is left unmodified. Returns a Deferred that
1219 fires (with this dirnode) when the operation finishes. This is
1220 equivalent to calling set_node() multiple times, but is much more
1223 def add_file(name, uploadable, metadata=None, overwrite=True):
1224 """I upload a file (using the given IUploadable), then attach the
1225 resulting ImmutableFileNode to the directory at the given name. I set
1226 metadata the same way as set_uri and set_node. The child name must be
1229 I return a Deferred that fires (with the IFileNode of the uploaded
1230 file) when the operation completes."""
1232 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1233 """I remove the child at the specific name. I return a Deferred that
1234 fires when the operation finishes. The child name must be a unicode
1235 string. If must_exist is True and I do not have a child by that name,
1236 I raise NoSuchChildError. If must_be_directory is True and the child
1237 is a file, or if must_be_file is True and the child is a directory,
1238 I raise ChildOfWrongTypeError."""
1240 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1241 """I create and attach a directory at the given name. The new
1242 directory can be empty, or it can be populated with children
1243 according to 'initial_children', which takes a dictionary in the same
1244 format as set_nodes (i.e. mapping unicode child name to (childnode,
1245 metadata) tuples). The child name must be a unicode string. I return
1246 a Deferred that fires (with the new directory node) when the
1247 operation finishes."""
1249 def move_child_to(current_child_name, new_parent, new_child_name=None,
1251 """I take one of my children and move them to a new parent. The child
1252 is referenced by name. On the new parent, the child will live under
1253 'new_child_name', which defaults to 'current_child_name'. TODO: what
1254 should we do about metadata? I return a Deferred that fires when the
1255 operation finishes. The child name must be a unicode string. I raise
1256 NoSuchChildError if I do not have a child by that name."""
1258 def build_manifest():
1259 """I generate a table of everything reachable from this directory.
1260 I also compute deep-stats as described below.
1262 I return a Monitor. The Monitor's results will be a dictionary with
1265 res['manifest']: a list of (path, cap) tuples for all nodes
1266 (directories and files) reachable from this one.
1267 'path' will be a tuple of unicode strings. The
1268 origin dirnode will be represented by an empty path
1270 res['verifycaps']: a list of (printable) verifycap strings, one for
1271 each reachable non-LIT node. This is a set:
1272 it will contain no duplicates.
1273 res['storage-index']: a list of (base32) storage index strings,
1274 one for each reachable non-LIT node. This is
1275 a set: it will contain no duplicates.
1276 res['stats']: a dictionary, the same that is generated by
1277 start_deep_stats() below.
1279 The Monitor will also have an .origin_si attribute with the (binary)
1280 storage index of the starting point.
1283 def start_deep_stats():
1284 """Return a Monitor, examining all nodes (directories and files)
1285 reachable from this one. The Monitor's results will be a dictionary
1286 with the following keys::
1288 count-immutable-files: count of how many CHK files are in the set
1289 count-mutable-files: same, for mutable files (does not include
1291 count-literal-files: same, for LIT files
1292 count-files: sum of the above three
1294 count-directories: count of directories
1296 size-immutable-files: total bytes for all CHK files in the set
1297 size-mutable-files (TODO): same, for current version of all mutable
1298 files, does not include directories
1299 size-literal-files: same, for LIT files
1300 size-directories: size of mutable files used by directories
1302 largest-directory: number of bytes in the largest directory
1303 largest-directory-children: number of children in the largest
1305 largest-immutable-file: number of bytes in the largest CHK file
1307 size-mutable-files is not yet implemented, because it would involve
1308 even more queries than deep_stats does.
1310 The Monitor will also have an .origin_si attribute with the (binary)
1311 storage index of the starting point.
1313 This operation will visit every directory node underneath this one,
1314 and can take a long time to run. On a typical workstation with good
1315 bandwidth, this can examine roughly 15 directories per second (and
1316 takes several minutes of 100% CPU for ~1700 directories).
1319 class ICodecEncoder(Interface):
1320 def set_params(data_size, required_shares, max_shares):
1321 """Set up the parameters of this encoder.
1323 This prepares the encoder to perform an operation that converts a
1324 single block of data into a number of shares, such that a future
1325 ICodecDecoder can use a subset of these shares to recover the
1326 original data. This operation is invoked by calling encode(). Once
1327 the encoding parameters are set up, the encode operation can be
1328 invoked multiple times.
1330 set_params() prepares the encoder to accept blocks of input data that
1331 are exactly 'data_size' bytes in length. The encoder will be prepared
1332 to produce 'max_shares' shares for each encode() operation (although
1333 see the 'desired_share_ids' to use less CPU). The encoding math will
1334 be chosen such that the decoder can get by with as few as
1335 'required_shares' of these shares and still reproduce the original
1336 data. For example, set_params(1000, 5, 5) offers no redundancy at
1337 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1339 Numerical Restrictions: 'data_size' is required to be an integral
1340 multiple of 'required_shares'. In general, the caller should choose
1341 required_shares and max_shares based upon their reliability
1342 requirements and the number of peers available (the total storage
1343 space used is roughly equal to max_shares*data_size/required_shares),
1344 then choose data_size to achieve the memory footprint desired (larger
1345 data_size means more efficient operation, smaller data_size means
1346 smaller memory footprint).
1348 In addition, 'max_shares' must be equal to or greater than
1349 'required_shares'. Of course, setting them to be equal causes
1350 encode() to degenerate into a particularly slow form of the 'split'
1353 See encode() for more details about how these parameters are used.
1355 set_params() must be called before any other ICodecEncoder methods
1360 """Return the 3-tuple of data_size, required_shares, max_shares"""
1362 def get_encoder_type():
1363 """Return a short string that describes the type of this encoder.
1365 There is required to be a global table of encoder classes. This method
1366 returns an index into this table; the value at this index is an
1367 encoder class, and this encoder is an instance of that class.
1370 def get_block_size():
1371 """Return the length of the shares that encode() will produce.
1374 def encode_proposal(data, desired_share_ids=None):
1375 """Encode some data.
1377 'data' must be a string (or other buffer object), and len(data) must
1378 be equal to the 'data_size' value passed earlier to set_params().
1380 This will return a Deferred that will fire with two lists. The first
1381 is a list of shares, each of which is a string (or other buffer
1382 object) such that len(share) is the same as what get_share_size()
1383 returned earlier. The second is a list of shareids, in which each is
1384 an integer. The lengths of the two lists will always be equal to each
1385 other. The user should take care to keep each share closely
1386 associated with its shareid, as one is useless without the other.
1388 The length of this output list will normally be the same as the value
1389 provided to the 'max_shares' parameter of set_params(). This may be
1390 different if 'desired_share_ids' is provided.
1392 'desired_share_ids', if provided, is required to be a sequence of
1393 ints, each of which is required to be >= 0 and < max_shares. If not
1394 provided, encode() will produce 'max_shares' shares, as if
1395 'desired_share_ids' were set to range(max_shares). You might use this
1396 if you initially thought you were going to use 10 peers, started
1397 encoding, and then two of the peers dropped out: you could use
1398 desired_share_ids= to skip the work (both memory and CPU) of
1399 producing shares for the peers which are no longer available.
1403 def encode(inshares, desired_share_ids=None):
1404 """Encode some data. This may be called multiple times. Each call is
1407 inshares is a sequence of length required_shares, containing buffers
1408 (i.e. strings), where each buffer contains the next contiguous
1409 non-overlapping segment of the input data. Each buffer is required to
1410 be the same length, and the sum of the lengths of the buffers is
1411 required to be exactly the data_size promised by set_params(). (This
1412 implies that the data has to be padded before being passed to
1413 encode(), unless of course it already happens to be an even multiple
1414 of required_shares in length.)
1416 Note: the requirement to break up your data into
1417 'required_shares' chunks of exactly the right length before
1418 calling encode() is surprising from point of view of a user
1419 who doesn't know how FEC works. It feels like an
1420 implementation detail that has leaked outside the abstraction
1421 barrier. Is there a use case in which the data to be encoded
1422 might already be available in pre-segmented chunks, such that
1423 it is faster or less work to make encode() take a list rather
1424 than splitting a single string?
1426 Yes, there is: suppose you are uploading a file with K=64,
1427 N=128, segsize=262,144. Then each in-share will be of size
1428 4096. If you use this .encode() API then your code could first
1429 read each successive 4096-byte chunk from the file and store
1430 each one in a Python string and store each such Python string
1431 in a Python list. Then you could call .encode(), passing that
1432 list as "inshares". The encoder would generate the other 64
1433 "secondary shares" and return to you a new list containing
1434 references to the same 64 Python strings that you passed in
1435 (as the primary shares) plus references to the new 64 Python
1438 (You could even imagine that your code could use readv() so
1439 that the operating system can arrange to get all of those
1440 bytes copied from the file into the Python list of Python
1441 strings as efficiently as possible instead of having a loop
1442 written in C or in Python to copy the next part of the file
1443 into the next string.)
1445 On the other hand if you instead use the .encode_proposal()
1446 API (above), then your code can first read in all of the
1447 262,144 bytes of the segment from the file into a Python
1448 string, then call .encode_proposal() passing the segment data
1449 as the "data" argument. The encoder would basically first
1450 split the "data" argument into a list of 64 in-shares of 4096
1451 byte each, and then do the same thing that .encode() does. So
1452 this would result in a little bit more copying of data and a
1453 little bit higher of a "maximum memory usage" during the
1454 process, although it might or might not make a practical
1455 difference for our current use cases.
1457 Note that "inshares" is a strange name for the parameter if
1458 you think of the parameter as being just for feeding in data
1459 to the codec. It makes more sense if you think of the result
1460 of this encoding as being the set of shares from inshares plus
1461 an extra set of "secondary shares" (or "check shares"). It is
1462 a surprising name! If the API is going to be surprising then
1463 the name should be surprising. If we switch to
1464 encode_proposal() above then we should also switch to an
1467 'desired_share_ids', if provided, is required to be a sequence of
1468 ints, each of which is required to be >= 0 and < max_shares. If not
1469 provided, encode() will produce 'max_shares' shares, as if
1470 'desired_share_ids' were set to range(max_shares). You might use this
1471 if you initially thought you were going to use 10 peers, started
1472 encoding, and then two of the peers dropped out: you could use
1473 desired_share_ids= to skip the work (both memory and CPU) of
1474 producing shares for the peers which are no longer available.
1476 For each call, encode() will return a Deferred that fires with two
1477 lists, one containing shares and the other containing the shareids.
1478 The get_share_size() method can be used to determine the length of
1479 the share strings returned by encode(). Each shareid is a small
1480 integer, exactly as passed into 'desired_share_ids' (or
1481 range(max_shares), if desired_share_ids was not provided).
1483 The shares and their corresponding shareids are required to be kept
1484 together during storage and retrieval. Specifically, the share data is
1485 useless by itself: the decoder needs to be told which share is which
1486 by providing it with both the shareid and the actual share data.
1488 This function will allocate an amount of memory roughly equal to::
1490 (max_shares - required_shares) * get_share_size()
1492 When combined with the memory that the caller must allocate to
1493 provide the input data, this leads to a memory footprint roughly
1494 equal to the size of the resulting encoded shares (i.e. the expansion
1495 factor times the size of the input segment).
1500 # returning a list of (shareidN,shareN) tuples instead of a pair of
1501 # lists (shareids..,shares..). Brian thought the tuples would
1502 # encourage users to keep the share and shareid together throughout
1503 # later processing, Zooko pointed out that the code to iterate
1504 # through two lists is not really more complicated than using a list
1505 # of tuples and there's also a performance improvement
1507 # having 'data_size' not required to be an integral multiple of
1508 # 'required_shares'. Doing this would require encode() to perform
1509 # padding internally, and we'd prefer to have any padding be done
1510 # explicitly by the caller. Yes, it is an abstraction leak, but
1511 # hopefully not an onerous one.
1514 class ICodecDecoder(Interface):
1515 def set_params(data_size, required_shares, max_shares):
1516 """Set the params. They have to be exactly the same ones that were
1517 used for encoding."""
1519 def get_needed_shares():
1520 """Return the number of shares needed to reconstruct the data.
1521 set_params() is required to be called before this."""
1523 def decode(some_shares, their_shareids):
1524 """Decode a partial list of shares into data.
1526 'some_shares' is required to be a sequence of buffers of sharedata, a
1527 subset of the shares returned by ICodecEncode.encode(). Each share is
1528 required to be of the same length. The i'th element of their_shareids
1529 is required to be the shareid of the i'th buffer in some_shares.
1531 This returns a Deferred which fires with a sequence of buffers. This
1532 sequence will contain all of the segments of the original data, in
1533 order. The sum of the lengths of all of the buffers will be the
1534 'data_size' value passed into the original ICodecEncode.set_params()
1535 call. To get back the single original input block of data, use
1536 ''.join(output_buffers), or you may wish to simply write them in
1537 order to an output file.
1539 Note that some of the elements in the result sequence may be
1540 references to the elements of the some_shares input sequence. In
1541 particular, this means that if those share objects are mutable (e.g.
1542 arrays) and if they are changed, then both the input (the
1543 'some_shares' parameter) and the output (the value given when the
1544 deferred is triggered) will change.
1546 The length of 'some_shares' is required to be exactly the value of
1547 'required_shares' passed into the original ICodecEncode.set_params()
1551 class IEncoder(Interface):
1552 """I take an object that provides IEncryptedUploadable, which provides
1553 encrypted data, and a list of shareholders. I then encode, hash, and
1554 deliver shares to those shareholders. I will compute all the necessary
1555 Merkle hash trees that are necessary to validate the crypttext that
1556 eventually comes back from the shareholders. I provide the URI Extension
1557 Block Hash, and the encoding parameters, both of which must be included
1560 I do not choose shareholders, that is left to the IUploader. I must be
1561 given a dict of RemoteReferences to storage buckets that are ready and
1562 willing to receive data.
1566 """Specify the number of bytes that will be encoded. This must be
1567 peformed before get_serialized_params() can be called.
1569 def set_params(params):
1570 """Override the default encoding parameters. 'params' is a tuple of
1571 (k,d,n), where 'k' is the number of required shares, 'd' is the
1572 servers_of_happiness, and 'n' is the total number of shares that will
1575 Encoding parameters can be set in three ways. 1: The Encoder class
1576 provides defaults (3/7/10). 2: the Encoder can be constructed with
1577 an 'options' dictionary, in which the
1578 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1579 set_params((k,d,n)) can be called.
1581 If you intend to use set_params(), you must call it before
1582 get_share_size or get_param are called.
1585 def set_encrypted_uploadable(u):
1586 """Provide a source of encrypted upload data. 'u' must implement
1587 IEncryptedUploadable.
1589 When this is called, the IEncryptedUploadable will be queried for its
1590 length and the storage_index that should be used.
1592 This returns a Deferred that fires with this Encoder instance.
1594 This must be performed before start() can be called.
1597 def get_param(name):
1598 """Return an encoding parameter, by name.
1600 'storage_index': return a string with the (16-byte truncated SHA-256
1601 hash) storage index to which these shares should be
1604 'share_counts': return a tuple describing how many shares are used:
1605 (needed_shares, servers_of_happiness, total_shares)
1607 'num_segments': return an int with the number of segments that
1610 'segment_size': return an int with the size of each segment.
1612 'block_size': return the size of the individual blocks that will
1613 be delivered to a shareholder's put_block() method. By
1614 knowing this, the shareholder will be able to keep all
1615 blocks in a single file and still provide random access
1616 when reading them. # TODO: can we avoid exposing this?
1618 'share_size': an int with the size of the data that will be stored
1619 on each shareholder. This is aggregate amount of data
1620 that will be sent to the shareholder, summed over all
1621 the put_block() calls I will ever make. It is useful to
1622 determine this size before asking potential
1623 shareholders whether they will grant a lease or not,
1624 since their answers will depend upon how much space we
1625 need. TODO: this might also include some amount of
1626 overhead, like the size of all the hashes. We need to
1627 decide whether this is useful or not.
1629 'serialized_params': a string with a concise description of the
1630 codec name and its parameters. This may be passed
1631 into the IUploadable to let it make sure that
1632 the same file encoded with different parameters
1633 will result in different storage indexes.
1635 Once this is called, set_size() and set_params() may not be called.
1638 def set_shareholders(shareholders, servermap):
1639 """Tell the encoder where to put the encoded shares. 'shareholders'
1640 must be a dictionary that maps share number (an integer ranging from
1641 0 to n-1) to an instance that provides IStorageBucketWriter.
1642 'servermap' is a dictionary that maps share number (as defined above)
1643 to a set of peerids. This must be performed before start() can be
1647 """Begin the encode/upload process. This involves reading encrypted
1648 data from the IEncryptedUploadable, encoding it, uploading the shares
1649 to the shareholders, then sending the hash trees.
1651 set_encrypted_uploadable() and set_shareholders() must be called
1652 before this can be invoked.
1654 This returns a Deferred that fires with a verify cap when the upload
1655 process is complete. The verifycap, plus the encryption key, is
1656 sufficient to construct the read cap.
1659 class IDecoder(Interface):
1660 """I take a list of shareholders and some setup information, then
1661 download, validate, decode, and decrypt data from them, writing the
1662 results to an output file.
1664 I do not locate the shareholders, that is left to the IDownloader. I must
1665 be given a dict of RemoteReferences to storage buckets that are ready to
1670 """I take a file-like object (providing write and close) to which all
1671 the plaintext data will be written.
1673 TODO: producer/consumer . Maybe write() should return a Deferred that
1674 indicates when it will accept more data? But probably having the
1675 IDecoder be a producer is easier to glue to IConsumer pieces.
1678 def set_shareholders(shareholders):
1679 """I take a dictionary that maps share identifiers (small integers)
1680 to RemoteReferences that provide RIBucketReader. This must be called
1684 """I start the download. This process involves retrieving data and
1685 hash chains from the shareholders, using the hashes to validate the
1686 data, decoding the shares into segments, decrypting the segments,
1687 then writing the resulting plaintext to the output file.
1689 I return a Deferred that will fire (with self) when the download is
1693 class IDownloadTarget(Interface):
1694 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1695 # will register itself as a producer. This allows the target to invoke
1696 # downloader.pauseProducing, resumeProducing, and stopProducing.
1698 """Called before any calls to write() or close(). If an error
1699 occurs before any data is available, fail() may be called without
1700 a previous call to open().
1702 'size' is the length of the file being downloaded, in bytes."""
1705 """Output some data to the target."""
1707 """Inform the target that there is no more data to be written."""
1709 """fail() is called to indicate that the download has failed. 'why'
1710 is a Failure object indicating what went wrong. No further methods
1711 will be invoked on the IDownloadTarget after fail()."""
1712 def register_canceller(cb):
1713 """The CiphertextDownloader uses this to register a no-argument function
1714 that the target can call to cancel the download. Once this canceller
1715 is invoked, no further calls to write() or close() will be made."""
1717 """When the CiphertextDownloader is done, this finish() function will be
1718 called. Whatever it returns will be returned to the invoker of
1719 Downloader.download.
1722 class IDownloader(Interface):
1723 def download(uri, target):
1724 """Perform a CHK download, sending the data to the given target.
1725 'target' must provide IDownloadTarget.
1727 Returns a Deferred that fires (with the results of target.finish)
1728 when the download is finished, or errbacks if something went wrong."""
1730 class IEncryptedUploadable(Interface):
1731 def set_upload_status(upload_status):
1732 """Provide an IUploadStatus object that should be filled with status
1733 information. The IEncryptedUploadable is responsible for setting
1734 key-determination progress ('chk'), size, storage_index, and
1735 ciphertext-fetch progress. It may delegate some of this
1736 responsibility to others, in particular to the IUploadable."""
1739 """This behaves just like IUploadable.get_size()."""
1741 def get_all_encoding_parameters():
1742 """Return a Deferred that fires with a tuple of
1743 (k,happy,n,segment_size). The segment_size will be used as-is, and
1744 must match the following constraints: it must be a multiple of k, and
1745 it shouldn't be unreasonably larger than the file size (if
1746 segment_size is larger than filesize, the difference must be stored
1749 This usually passes through to the IUploadable method of the same
1752 The encoder strictly obeys the values returned by this method. To
1753 make an upload use non-default encoding parameters, you must arrange
1754 to control the values that this method returns.
1757 def get_storage_index():
1758 """Return a Deferred that fires with a 16-byte storage index.
1761 def read_encrypted(length, hash_only):
1762 """This behaves just like IUploadable.read(), but returns crypttext
1763 instead of plaintext. If hash_only is True, then this discards the
1764 data (and returns an empty list); this improves efficiency when
1765 resuming an interrupted upload (where we need to compute the
1766 plaintext hashes, but don't need the redundant encrypted data)."""
1768 def get_plaintext_hashtree_leaves(first, last, num_segments):
1769 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1770 plaintext segments, i.e. get the tagged hashes of the given segments.
1771 The segment size is expected to be generated by the
1772 IEncryptedUploadable before any plaintext is read or ciphertext
1773 produced, so that the segment hashes can be generated with only a
1776 This returns a Deferred which fires with a sequence of hashes, using:
1778 tuple(segment_hashes[first:last])
1780 'num_segments' is used to assert that the number of segments that the
1781 IEncryptedUploadable handled matches the number of segments that the
1782 encoder was expecting.
1784 This method must not be called until the final byte has been read
1785 from read_encrypted(). Once this method is called, read_encrypted()
1786 can never be called again.
1789 def get_plaintext_hash():
1790 """OBSOLETE; Get the hash of the whole plaintext.
1792 This returns a Deferred which fires with a tagged SHA-256 hash of the
1793 whole plaintext, obtained from hashutil.plaintext_hash(data).
1797 """Just like IUploadable.close()."""
1799 class IUploadable(Interface):
1800 def set_upload_status(upload_status):
1801 """Provide an IUploadStatus object that should be filled with status
1802 information. The IUploadable is responsible for setting
1803 key-determination progress ('chk')."""
1805 def set_default_encoding_parameters(params):
1806 """Set the default encoding parameters, which must be a dict mapping
1807 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1808 'max_segment_size'. These might have an influence on the final
1809 encoding parameters returned by get_all_encoding_parameters(), if the
1810 Uploadable doesn't have more specific preferences.
1812 This call is optional: if it is not used, the Uploadable will use
1813 some built-in defaults. If used, this method must be called before
1814 any other IUploadable methods to have any effect.
1818 """Return a Deferred that will fire with the length of the data to be
1819 uploaded, in bytes. This will be called before the data is actually
1820 used, to compute encoding parameters.
1823 def get_all_encoding_parameters():
1824 """Return a Deferred that fires with a tuple of
1825 (k,happy,n,segment_size). The segment_size will be used as-is, and
1826 must match the following constraints: it must be a multiple of k, and
1827 it shouldn't be unreasonably larger than the file size (if
1828 segment_size is larger than filesize, the difference must be stored
1831 The relative values of k and n allow some IUploadables to request
1832 better redundancy than others (in exchange for consuming more space
1835 Larger values of segment_size reduce hash overhead, while smaller
1836 values reduce memory footprint and cause data to be delivered in
1837 smaller pieces (which may provide a smoother and more predictable
1838 download experience).
1840 The encoder strictly obeys the values returned by this method. To
1841 make an upload use non-default encoding parameters, you must arrange
1842 to control the values that this method returns. One way to influence
1843 them may be to call set_encoding_parameters() before calling
1844 get_all_encoding_parameters().
1847 def get_encryption_key():
1848 """Return a Deferred that fires with a 16-byte AES key. This key will
1849 be used to encrypt the data. The key will also be hashed to derive
1852 Uploadables which want to achieve convergence should hash their file
1853 contents and the serialized_encoding_parameters to form the key
1854 (which of course requires a full pass over the data). Uploadables can
1855 use the upload.ConvergentUploadMixin class to achieve this
1858 Uploadables which do not care about convergence (or do not wish to
1859 make multiple passes over the data) can simply return a
1860 strongly-random 16 byte string.
1862 get_encryption_key() may be called multiple times: the IUploadable is
1863 required to return the same value each time.
1867 """Return a Deferred that fires with a list of strings (perhaps with
1868 only a single element) which, when concatenated together, contain the
1869 next 'length' bytes of data. If EOF is near, this may provide fewer
1870 than 'length' bytes. The total number of bytes provided by read()
1871 before it signals EOF must equal the size provided by get_size().
1873 If the data must be acquired through multiple internal read
1874 operations, returning a list instead of a single string may help to
1875 reduce string copies. However, the length of the concatenated strings
1876 must equal the amount of data requested, unless EOF is encountered.
1877 Long reads, or short reads without EOF, are not allowed. read()
1878 should return the same amount of data as a local disk file read, just
1879 in a different shape and asynchronously.
1881 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1882 so a 10kB file means length=3kB, 100kB file means length=30kB,
1883 and >=1MB file means length=300kB.
1885 This method provides for a single full pass through the data. Later
1886 use cases may desire multiple passes or access to only parts of the
1887 data (such as a mutable file making small edits-in-place). This API
1888 will be expanded once those use cases are better understood.
1892 """The upload is finished, and whatever filehandle was in use may be
1896 class IMutableUploadable(Interface):
1898 I represent content that is due to be uploaded to a mutable filecap.
1900 # This is somewhat simpler than the IUploadable interface above
1901 # because mutable files do not need to be concerned with possibly
1902 # generating a CHK, nor with per-file keys. It is a subset of the
1903 # methods in IUploadable, though, so we could just as well implement
1904 # the mutable uploadables as IUploadables that don't happen to use
1905 # those methods (with the understanding that the unused methods will
1906 # never be called on such objects)
1909 Returns a Deferred that fires with the size of the content held
1915 Returns a list of strings which, when concatenated, are the next
1916 length bytes of the file, or fewer if there are fewer bytes
1917 between the current location and the end of the file.
1922 The process that used the Uploadable is finished using it, so
1923 the uploadable may be closed.
1926 class IUploadResults(Interface):
1927 """I am returned by upload() methods. I contain a number of public
1928 attributes which can be read to determine the results of the upload. Some
1929 of these are functional, some are timing information. All of these may be
1932 .file_size : the size of the file, in bytes
1933 .uri : the CHK read-cap for the file
1934 .ciphertext_fetched : how many bytes were fetched by the helper
1935 .sharemap: dict mapping share identifier to set of serverids
1936 (binary strings). This indicates which servers were given
1937 which shares. For immutable files, the shareid is an
1938 integer (the share number, from 0 to N-1). For mutable
1939 files, it is a string of the form 'seq%d-%s-sh%d',
1940 containing the sequence number, the roothash, and the
1942 .servermap : dict mapping server peerid to a set of share numbers
1943 .timings : dict of timing information, mapping name to seconds (float)
1944 total : total upload time, start to finish
1945 storage_index : time to compute the storage index
1946 peer_selection : time to decide which peers will be used
1947 contacting_helper : initial helper query to upload/no-upload decision
1948 existence_check : helper pre-upload existence check
1949 helper_total : initial helper query to helper finished pushing
1950 cumulative_fetch : helper waiting for ciphertext requests
1951 total_fetch : helper start to last ciphertext response
1952 cumulative_encoding : just time spent in zfec
1953 cumulative_sending : just time spent waiting for storage servers
1954 hashes_and_close : last segment push to shareholder close
1955 total_encode_and_push : first encode to shareholder close
1959 class IDownloadResults(Interface):
1960 """I am created internally by download() methods. I contain a number of
1961 public attributes which contain details about the download process.::
1963 .file_size : the size of the file, in bytes
1964 .servers_used : set of server peerids that were used during download
1965 .server_problems : dict mapping server peerid to a problem string. Only
1966 servers that had problems (bad hashes, disconnects)
1968 .servermap : dict mapping server peerid to a set of share numbers. Only
1969 servers that had any shares are listed here.
1970 .timings : dict of timing information, mapping name to seconds (float)
1971 peer_selection : time to ask servers about shares
1972 servers_peer_selection : dict of peerid to DYHB-query time
1973 uri_extension : time to fetch a copy of the URI extension block
1974 hashtrees : time to fetch the hash trees
1975 segments : time to fetch, decode, and deliver segments
1976 cumulative_fetch : time spent waiting for storage servers
1977 cumulative_decode : just time spent in zfec
1978 cumulative_decrypt : just time spent in decryption
1979 total : total download time, start to finish
1980 fetch_per_server : dict of peerid to list of per-segment fetch times
1984 class IUploader(Interface):
1985 def upload(uploadable):
1986 """Upload the file. 'uploadable' must impement IUploadable. This
1987 returns a Deferred which fires with an IUploadResults instance, from
1988 which the URI of the file can be obtained as results.uri ."""
1990 def upload_ssk(write_capability, new_version, uploadable):
1991 """TODO: how should this work?"""
1993 class ICheckable(Interface):
1994 def check(monitor, verify=False, add_lease=False):
1995 """Check up on my health, optionally repairing any problems.
1997 This returns a Deferred that fires with an instance that provides
1998 ICheckResults, or None if the object is non-distributed (i.e. LIT
2001 The monitor will be checked periodically to see if the operation has
2002 been cancelled. If so, no new queries will be sent, and the Deferred
2003 will fire (with a OperationCancelledError) immediately.
2005 Filenodes and dirnodes (which provide IFilesystemNode) are also
2006 checkable. Instances that represent verifier-caps will be checkable
2007 but not downloadable. Some objects (like LIT files) do not actually
2008 live in the grid, and their checkers return None (non-distributed
2009 files are always healthy).
2011 If verify=False, a relatively lightweight check will be performed: I
2012 will ask all servers if they have a share for me, and I will believe
2013 whatever they say. If there are at least N distinct shares on the
2014 grid, my results will indicate r.is_healthy()==True. This requires a
2015 roundtrip to each server, but does not transfer very much data, so
2016 the network bandwidth is fairly low.
2018 If verify=True, a more resource-intensive check will be performed:
2019 every share will be downloaded, and the hashes will be validated on
2020 every bit. I will ignore any shares that failed their hash checks. If
2021 there are at least N distinct valid shares on the grid, my results
2022 will indicate r.is_healthy()==True. This requires N/k times as much
2023 download bandwidth (and server disk IO) as a regular download. If a
2024 storage server is holding a corrupt share, or is experiencing memory
2025 failures during retrieval, or is malicious or buggy, then
2026 verification will detect the problem, but checking will not.
2028 If add_lease=True, I will ensure that an up-to-date lease is present
2029 on each share. The lease secrets will be derived from by node secret
2030 (in BASEDIR/private/secret), so either I will add a new lease to the
2031 share, or I will merely renew the lease that I already had. In a
2032 future version of the storage-server protocol (once Accounting has
2033 been implemented), there may be additional options here to define the
2034 kind of lease that is obtained (which account number to claim, etc).
2036 TODO: any problems seen during checking will be reported to the
2037 health-manager.furl, a centralized object which is responsible for
2038 figuring out why files are unhealthy so corrective action can be
2042 def check_and_repair(monitor, verify=False, add_lease=False):
2043 """Like check(), but if the file/directory is not healthy, attempt to
2046 Any non-healthy result will cause an immediate repair operation, to
2047 generate and upload new shares. After repair, the file will be as
2048 healthy as we can make it. Details about what sort of repair is done
2049 will be put in the check-and-repair results. The Deferred will not
2050 fire until the repair is complete.
2052 This returns a Deferred which fires with an instance of
2053 ICheckAndRepairResults."""
2055 class IDeepCheckable(Interface):
2056 def start_deep_check(verify=False, add_lease=False):
2057 """Check upon the health of me and everything I can reach.
2059 This is a recursive form of check(), useable only on dirnodes.
2061 I return a Monitor, with results that are an IDeepCheckResults
2064 TODO: If any of the directories I traverse are unrecoverable, the
2065 Monitor will report failure. If any of the files I check upon are
2066 unrecoverable, those problems will be reported in the
2067 IDeepCheckResults as usual, and the Monitor will not report a
2071 def start_deep_check_and_repair(verify=False, add_lease=False):
2072 """Check upon the health of me and everything I can reach. Repair
2073 anything that isn't healthy.
2075 This is a recursive form of check_and_repair(), useable only on
2078 I return a Monitor, with results that are an
2079 IDeepCheckAndRepairResults object.
2081 TODO: If any of the directories I traverse are unrecoverable, the
2082 Monitor will report failure. If any of the files I check upon are
2083 unrecoverable, those problems will be reported in the
2084 IDeepCheckResults as usual, and the Monitor will not report a
2088 class ICheckResults(Interface):
2089 """I contain the detailed results of a check/verify operation.
2092 def get_storage_index():
2093 """Return a string with the (binary) storage index."""
2094 def get_storage_index_string():
2095 """Return a string with the (printable) abbreviated storage index."""
2097 """Return the (string) URI of the object that was checked."""
2100 """Return a boolean, True if the file/dir is fully healthy, False if
2101 it is damaged in any way. Non-distributed LIT files always return
2104 def is_recoverable():
2105 """Return a boolean, True if the file/dir can be recovered, False if
2106 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2107 files always return True."""
2109 def needs_rebalancing():
2110 """Return a boolean, True if the file/dir's reliability could be
2111 improved by moving shares to new servers. Non-distributed LIT files
2112 always return False."""
2116 """Return a dictionary that describes the state of the file/dir. LIT
2117 files always return an empty dictionary. Normal files and directories
2118 return a dictionary with the following keys (note that these use
2119 binary strings rather than base32-encoded ones) (also note that for
2120 mutable files, these counts are for the 'best' version):
2122 count-shares-good: the number of distinct good shares that were found
2123 count-shares-needed: 'k', the number of shares required for recovery
2124 count-shares-expected: 'N', the number of total shares generated
2125 count-good-share-hosts: the number of distinct storage servers with
2126 good shares. If this number is less than
2127 count-shares-good, then some shares are
2128 doubled up, increasing the correlation of
2129 failures. This indicates that one or more
2130 shares should be moved to an otherwise unused
2131 server, if one is available.
2132 count-corrupt-shares: the number of shares with integrity failures
2133 list-corrupt-shares: a list of 'share locators', one for each share
2134 that was found to be corrupt. Each share
2135 locator is a list of (serverid, storage_index,
2137 count-incompatible-shares: the number of shares which are of a share
2138 format unknown to this checker
2139 list-incompatible-shares: a list of 'share locators', one for each
2140 share that was found to be of an unknown
2141 format. Each share locator is a list of
2142 (serverid, storage_index, sharenum).
2143 servers-responding: list of (binary) storage server identifiers,
2144 one for each server which responded to the share
2145 query (even if they said they didn't have
2146 shares, and even if they said they did have
2147 shares but then didn't send them when asked, or
2148 dropped the connection, or returned a Failure,
2149 and even if they said they did have shares and
2150 sent incorrect ones when asked)
2151 sharemap: dict mapping share identifier to list of serverids
2152 (binary strings). This indicates which servers are holding
2153 which shares. For immutable files, the shareid is an
2154 integer (the share number, from 0 to N-1). For mutable
2155 files, it is a string of the form 'seq%d-%s-sh%d',
2156 containing the sequence number, the roothash, and the
2159 The following keys are most relevant for mutable files, but immutable
2160 files will provide sensible values too::
2162 count-wrong-shares: the number of shares for versions other than the
2163 'best' one (which is defined as being the
2164 recoverable version with the highest sequence
2165 number, then the highest roothash). These are
2166 either leftover shares from an older version
2167 (perhaps on a server that was offline when an
2168 update occurred), shares from an unrecoverable
2169 newer version, or shares from an alternate
2170 current version that results from an
2171 uncoordinated write collision. For a healthy
2172 file, this will equal 0.
2174 count-recoverable-versions: the number of recoverable versions of
2175 the file. For a healthy file, this will
2178 count-unrecoverable-versions: the number of unrecoverable versions
2179 of the file. For a healthy file, this
2185 """Return a string with a brief (one-line) summary of the results."""
2188 """Return a list of strings with more detailed results."""
2190 class ICheckAndRepairResults(Interface):
2191 """I contain the detailed results of a check/verify/repair operation.
2193 The IFilesystemNode.check()/verify()/repair() methods all return
2194 instances that provide ICheckAndRepairResults.
2197 def get_storage_index():
2198 """Return a string with the (binary) storage index."""
2199 def get_storage_index_string():
2200 """Return a string with the (printable) abbreviated storage index."""
2201 def get_repair_attempted():
2202 """Return a boolean, True if a repair was attempted. We might not
2203 attempt to repair the file because it was healthy, or healthy enough
2204 (i.e. some shares were missing but not enough to exceed some
2205 threshold), or because we don't know how to repair this object."""
2206 def get_repair_successful():
2207 """Return a boolean, True if repair was attempted and the file/dir
2208 was fully healthy afterwards. False if no repair was attempted or if
2209 a repair attempt failed."""
2210 def get_pre_repair_results():
2211 """Return an ICheckResults instance that describes the state of the
2212 file/dir before any repair was attempted."""
2213 def get_post_repair_results():
2214 """Return an ICheckResults instance that describes the state of the
2215 file/dir after any repair was attempted. If no repair was attempted,
2216 the pre-repair and post-repair results will be identical."""
2219 class IDeepCheckResults(Interface):
2220 """I contain the results of a deep-check operation.
2222 This is returned by a call to ICheckable.deep_check().
2225 def get_root_storage_index_string():
2226 """Return the storage index (abbreviated human-readable string) of
2227 the first object checked."""
2229 """Return a dictionary with the following keys::
2231 count-objects-checked: count of how many objects were checked
2232 count-objects-healthy: how many of those objects were completely
2234 count-objects-unhealthy: how many were damaged in some way
2235 count-objects-unrecoverable: how many were unrecoverable
2236 count-corrupt-shares: how many shares were found to have
2237 corruption, summed over all objects
2241 def get_corrupt_shares():
2242 """Return a set of (serverid, storage_index, sharenum) for all shares
2243 that were found to be corrupt. Both serverid and storage_index are
2246 def get_all_results():
2247 """Return a dictionary mapping pathname (a tuple of strings, ready to
2248 be slash-joined) to an ICheckResults instance, one for each object
2249 that was checked."""
2251 def get_results_for_storage_index(storage_index):
2252 """Retrive the ICheckResults instance for the given (binary)
2253 storage index. Raises KeyError if there are no results for that
2257 """Return a dictionary with the same keys as
2258 IDirectoryNode.deep_stats()."""
2260 class IDeepCheckAndRepairResults(Interface):
2261 """I contain the results of a deep-check-and-repair operation.
2263 This is returned by a call to ICheckable.deep_check_and_repair().
2266 def get_root_storage_index_string():
2267 """Return the storage index (abbreviated human-readable string) of
2268 the first object checked."""
2270 """Return a dictionary with the following keys::
2272 count-objects-checked: count of how many objects were checked
2273 count-objects-healthy-pre-repair: how many of those objects were
2274 completely healthy (before any
2276 count-objects-unhealthy-pre-repair: how many were damaged in
2278 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2279 count-objects-healthy-post-repair: how many of those objects were
2280 completely healthy (after any
2282 count-objects-unhealthy-post-repair: how many were damaged in
2284 count-objects-unrecoverable-post-repair: how many were
2286 count-repairs-attempted: repairs were attempted on this many
2287 objects. The count-repairs- keys will
2288 always be provided, however unless
2289 repair=true is present, they will all
2291 count-repairs-successful: how many repairs resulted in healthy
2293 count-repairs-unsuccessful: how many repairs resulted did not
2294 results in completely healthy objects
2295 count-corrupt-shares-pre-repair: how many shares were found to
2296 have corruption, summed over all
2297 objects examined (before any
2299 count-corrupt-shares-post-repair: how many shares were found to
2300 have corruption, summed over all
2301 objects examined (after any
2306 """Return a dictionary with the same keys as
2307 IDirectoryNode.deep_stats()."""
2309 def get_corrupt_shares():
2310 """Return a set of (serverid, storage_index, sharenum) for all shares
2311 that were found to be corrupt before any repair was attempted. Both
2312 serverid and storage_index are binary.
2314 def get_remaining_corrupt_shares():
2315 """Return a set of (serverid, storage_index, sharenum) for all shares
2316 that were found to be corrupt after any repair was completed. Both
2317 serverid and storage_index are binary. These are shares that need
2318 manual inspection and probably deletion.
2320 def get_all_results():
2321 """Return a dictionary mapping pathname (a tuple of strings, ready to
2322 be slash-joined) to an ICheckAndRepairResults instance, one for each
2323 object that was checked."""
2325 def get_results_for_storage_index(storage_index):
2326 """Retrive the ICheckAndRepairResults instance for the given (binary)
2327 storage index. Raises KeyError if there are no results for that
2331 class IRepairable(Interface):
2332 def repair(check_results):
2333 """Attempt to repair the given object. Returns a Deferred that fires
2334 with a IRepairResults object.
2336 I must be called with an object that implements ICheckResults, as
2337 proof that you have actually discovered a problem with this file. I
2338 will use the data in the checker results to guide the repair process,
2339 such as which servers provided bad data and should therefore be
2340 avoided. The ICheckResults object is inside the
2341 ICheckAndRepairResults object, which is returned by the
2342 ICheckable.check() method::
2344 d = filenode.check(repair=False)
2345 def _got_results(check_and_repair_results):
2346 check_results = check_and_repair_results.get_pre_repair_results()
2347 return filenode.repair(check_results)
2348 d.addCallback(_got_results)
2352 class IRepairResults(Interface):
2353 """I contain the results of a repair operation."""
2354 def get_successful(self):
2355 """Returns a boolean: True if the repair made the file healthy, False
2356 if not. Repair failure generally indicates a file that has been
2357 damaged beyond repair."""
2360 class IClient(Interface):
2361 def upload(uploadable):
2362 """Upload some data into a CHK, get back the UploadResults for it.
2363 @param uploadable: something that implements IUploadable
2364 @return: a Deferred that fires with the UploadResults instance.
2365 To get the URI for this file, use results.uri .
2368 def create_mutable_file(contents=""):
2369 """Create a new mutable file (with initial) contents, get back the
2372 @param contents: (bytestring, callable, or None): this provides the
2373 initial contents of the mutable file. If 'contents' is a bytestring,
2374 it will be used as-is. If 'contents' is a callable, it will be
2375 invoked with the new MutableFileNode instance and is expected to
2376 return a bytestring with the initial contents of the file (the
2377 callable can use node.get_writekey() to decide how to encrypt the
2378 initial contents, e.g. for a brand new dirnode with initial
2379 children). contents=None is equivalent to an empty string. Using
2380 content_maker= is more efficient than creating a mutable file and
2381 setting its contents in two separate operations.
2383 @return: a Deferred that fires with an IMutableFileNode instance.
2386 def create_dirnode(initial_children={}):
2387 """Create a new unattached dirnode, possibly with initial children.
2389 @param initial_children: dict with keys that are unicode child names,
2390 and values that are (childnode, metadata) tuples.
2392 @return: a Deferred that fires with the new IDirectoryNode instance.
2395 def create_node_from_uri(uri, rouri):
2396 """Create a new IFilesystemNode instance from the uri, synchronously.
2397 @param uri: a string or IURI-providing instance, or None. This could
2398 be for a LiteralFileNode, a CHK file node, a mutable file
2399 node, or a directory node
2400 @param rouri: a string or IURI-providing instance, or None. If the
2401 main uri is None, I will use the rouri instead. If I
2402 recognize the format of the main uri, I will ignore the
2403 rouri (because it can be derived from the writecap).
2405 @return: an instance that provides IFilesystemNode (or more usefully
2406 one of its subclasses). File-specifying URIs will result in
2407 IFileNode-providing instances, like ImmutableFileNode,
2408 LiteralFileNode, or MutableFileNode. Directory-specifying
2409 URIs will result in IDirectoryNode-providing instances, like
2413 class INodeMaker(Interface):
2414 """The NodeMaker is used to create IFilesystemNode instances. It can
2415 accept a filecap/dircap string and return the node right away. It can
2416 also create new nodes (i.e. upload a file, or create a mutable file)
2417 asynchronously. Once you have one of these nodes, you can use other
2418 methods to determine whether it is a file or directory, and to download
2419 or modify its contents.
2421 The NodeMaker encapsulates all the authorities that these
2422 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2423 Tahoe process will typically have a single NodeMaker, but unit tests may
2424 create simplified/mocked forms for testing purposes.
2426 def create_from_cap(writecap, readcap=None, **kwargs):
2427 """I create an IFilesystemNode from the given writecap/readcap. I can
2428 only provide nodes for existing file/directory objects: use my other
2429 methods to create new objects. I return synchronously."""
2431 def create_mutable_file(contents=None, keysize=None):
2432 """I create a new mutable file, and return a Deferred which will fire
2433 with the IMutableFileNode instance when it is ready. If contents= is
2434 provided (a bytestring), it will be used as the initial contents of
2435 the new file, otherwise the file will contain zero bytes. keysize= is
2436 for use by unit tests, to create mutable files that are smaller than
2439 def create_new_mutable_directory(initial_children={}):
2440 """I create a new mutable directory, and return a Deferred which will
2441 fire with the IDirectoryNode instance when it is ready. If
2442 initial_children= is provided (a dict mapping unicode child name to
2443 (childnode, metadata_dict) tuples), the directory will be populated
2444 with those children, otherwise it will be empty."""
2446 class IClientStatus(Interface):
2447 def list_all_uploads():
2448 """Return a list of uploader objects, one for each upload which
2449 currently has an object available (tracked with weakrefs). This is
2450 intended for debugging purposes."""
2451 def list_active_uploads():
2452 """Return a list of active IUploadStatus objects."""
2453 def list_recent_uploads():
2454 """Return a list of IUploadStatus objects for the most recently
2457 def list_all_downloads():
2458 """Return a list of downloader objects, one for each download which
2459 currently has an object available (tracked with weakrefs). This is
2460 intended for debugging purposes."""
2461 def list_active_downloads():
2462 """Return a list of active IDownloadStatus objects."""
2463 def list_recent_downloads():
2464 """Return a list of IDownloadStatus objects for the most recently
2465 started downloads."""
2467 class IUploadStatus(Interface):
2469 """Return a timestamp (float with seconds since epoch) indicating
2470 when the operation was started."""
2471 def get_storage_index():
2472 """Return a string with the (binary) storage index in use on this
2473 upload. Returns None if the storage index has not yet been
2476 """Return an integer with the number of bytes that will eventually
2477 be uploaded for this file. Returns None if the size is not yet known.
2480 """Return True if this upload is using a Helper, False if not."""
2482 """Return a string describing the current state of the upload
2485 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2486 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2487 made towards hashing the file to determine a CHK encryption key: if
2488 non-convergent encryption is in use, this will be trivial, otherwise
2489 the whole file must be hashed. 'ciphertext' describes how much of the
2490 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2491 uploads. 'encode_and_push' describes how much of the encode-and-push
2492 process has finished: for helper uploads this is dependent upon the
2493 helper providing progress reports. It might be reasonable to add all
2494 three numbers and report the sum to the user."""
2496 """Return True if the upload is currently active, False if not."""
2498 """Return an instance of UploadResults (which contains timing and
2499 sharemap information). Might return None if the upload is not yet
2502 """Each upload status gets a unique number: this method returns that
2503 number. This provides a handle to this particular upload, so a web
2504 page can generate a suitable hyperlink."""
2506 class IDownloadStatus(Interface):
2508 """Return a timestamp (float with seconds since epoch) indicating
2509 when the operation was started."""
2510 def get_storage_index():
2511 """Return a string with the (binary) storage index in use on this
2512 download. This may be None if there is no storage index (i.e. LIT
2515 """Return an integer with the number of bytes that will eventually be
2516 retrieved for this file. Returns None if the size is not yet known.
2519 """Return True if this download is using a Helper, False if not."""
2521 """Return a string describing the current state of the download
2524 """Returns a float (from 0.0 to 1.0) describing the amount of the
2525 download that has completed. This value will remain at 0.0 until the
2526 first byte of plaintext is pushed to the download target."""
2528 """Return True if the download is currently active, False if not."""
2530 """Each download status gets a unique number: this method returns
2531 that number. This provides a handle to this particular download, so a
2532 web page can generate a suitable hyperlink."""
2534 class IServermapUpdaterStatus(Interface):
2536 class IPublishStatus(Interface):
2538 class IRetrieveStatus(Interface):
2541 class NotCapableError(Exception):
2542 """You have tried to write to a read-only node."""
2544 class BadWriteEnablerError(Exception):
2547 class RIControlClient(RemoteInterface):
2549 def wait_for_client_connections(num_clients=int):
2550 """Do not return until we have connections to at least NUM_CLIENTS
2554 def upload_from_file_to_uri(filename=str,
2555 convergence=ChoiceOf(None,
2556 StringConstraint(2**20))):
2557 """Upload a file to the grid. This accepts a filename (which must be
2558 absolute) that points to a file on the node's local disk. The node will
2559 read the contents of this file, upload it to the grid, then return the
2560 URI at which it was uploaded. If convergence is None then a random
2561 encryption key will be used, else the plaintext will be hashed, then
2562 that hash will be mixed together with the "convergence" string to form
2567 def download_from_uri_to_file(uri=URI, filename=str):
2568 """Download a file from the grid, placing it on the node's local disk
2569 at the given filename (which must be absolute[?]). Returns the
2570 absolute filename where the file was written."""
2575 def get_memory_usage():
2576 """Return a dict describes the amount of memory currently in use. The
2577 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2578 measuring memory consupmtion in bytes."""
2579 return DictOf(str, int)
2581 def speed_test(count=int, size=int, mutable=Any()):
2582 """Write 'count' tempfiles to disk, all of the given size. Measure
2583 how long (in seconds) it takes to upload them all to the servers.
2584 Then measure how long it takes to download all of them. If 'mutable'
2585 is 'create', time creation of mutable files. If 'mutable' is
2586 'upload', then time access to the same mutable file instead of
2589 Returns a tuple of (upload_time, download_time).
2591 return (float, float)
2593 def measure_peer_response_time():
2594 """Send a short message to each connected peer, and measure the time
2595 it takes for them to respond to it. This is a rough measure of the
2596 application-level round trip time.
2598 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2601 return DictOf(str, float)
2603 UploadResults = Any() #DictOf(str, str)
2605 class RIEncryptedUploadable(RemoteInterface):
2606 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2611 def get_all_encoding_parameters():
2612 return (int, int, int, long)
2614 def read_encrypted(offset=Offset, length=ReadSize):
2621 class RICHKUploadHelper(RemoteInterface):
2622 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2626 Return a dictionary of version information.
2628 return DictOf(str, Any())
2630 def upload(reader=RIEncryptedUploadable):
2631 return UploadResults
2634 class RIHelper(RemoteInterface):
2635 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2639 Return a dictionary of version information.
2641 return DictOf(str, Any())
2643 def upload_chk(si=StorageIndex):
2644 """See if a file with a given storage index needs uploading. The
2645 helper will ask the appropriate storage servers to see if the file
2646 has already been uploaded. If so, the helper will return a set of
2647 'upload results' that includes whatever hashes are needed to build
2648 the read-cap, and perhaps a truncated sharemap.
2650 If the file has not yet been uploaded (or if it was only partially
2651 uploaded), the helper will return an empty upload-results dictionary
2652 and also an RICHKUploadHelper object that will take care of the
2653 upload process. The client should call upload() on this object and
2654 pass it a reference to an RIEncryptedUploadable object that will
2655 provide ciphertext. When the upload is finished, the upload() method
2656 will finish and return the upload results.
2658 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2661 class RIStatsProvider(RemoteInterface):
2662 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2664 Provides access to statistics and monitoring information.
2669 returns a dictionary containing 'counters' and 'stats', each a
2670 dictionary with string counter/stat name keys, and numeric or None values.
2671 counters are monotonically increasing measures of work done, and
2672 stats are instantaneous measures (potentially time averaged
2675 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2677 class RIStatsGatherer(RemoteInterface):
2678 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2680 Provides a monitoring service for centralised collection of stats
2683 def provide(provider=RIStatsProvider, nickname=str):
2685 @param provider: a stats collector instance which should be polled
2686 periodically by the gatherer to collect stats.
2687 @param nickname: a name useful to identify the provided client
2692 class IStatsProducer(Interface):
2695 returns a dictionary, with str keys representing the names of stats
2696 to be monitored, and numeric values.
2699 class RIKeyGenerator(RemoteInterface):
2700 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2702 Provides a service offering to make RSA key pairs.
2705 def get_rsa_key_pair(key_size=int):
2707 @param key_size: the size of the signature key.
2708 @return: tuple(verifying_key, signing_key)
2710 return TupleOf(str, str)
2713 class FileTooLargeError(Exception):
2716 class IValidatedThingProxy(Interface):
2718 """ Acquire a thing and validate it. Return a deferred which is
2719 eventually fired with self if the thing is valid or errbacked if it
2720 can't be acquired or validated."""
2722 class InsufficientVersionError(Exception):
2723 def __init__(self, needed, got):
2724 self.needed = needed
2727 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2730 class EmptyPathnameComponentError(Exception):
2731 """The webapi disallows empty pathname components."""