2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIBucketWriter(RemoteInterface):
34 """ Objects of this kind live on the server side. """
35 def write(offset=Offset, data=ShareData):
40 If the data that has been written is incomplete or inconsistent then
41 the server will throw the data away, else it will store it for future
47 """Abandon all the data that has been written.
52 class RIBucketReader(RemoteInterface):
53 def read(offset=Offset, length=ReadSize):
56 def advise_corrupt_share(reason=str):
57 """Clients who discover hash failures in shares that they have
58 downloaded from me will use this method to inform me about the
59 failures. I will record their concern so that my operator can
60 manually inspect the shares in question. I return None.
62 This is a wrapper around RIStorageServer.advise_corrupt_share()
63 that is tied to a specific share, and therefore does not need the
64 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
86 class RIStorageServer(RemoteInterface):
87 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
91 Return a dictionary of version information.
93 return DictOf(str, Any())
95 def allocate_buckets(storage_index=StorageIndex,
96 renew_secret=LeaseRenewSecret,
97 cancel_secret=LeaseCancelSecret,
98 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
99 allocated_size=Offset, canary=Referenceable):
101 @param storage_index: the index of the bucket to be created or
103 @param sharenums: these are the share numbers (probably between 0 and
104 99) that the sender is proposing to store on this
106 @param renew_secret: This is the secret used to protect bucket refresh
107 This secret is generated by the client and
108 stored for later comparison by the server. Each
109 server is given a different secret.
110 @param cancel_secret: Like renew_secret, but protects bucket decref.
111 @param canary: If the canary is lost before close(), the bucket is
113 @return: tuple of (alreadygot, allocated), where alreadygot is what we
114 already have and allocated is what we hereby agree to accept.
115 New leases are added for shares in both lists.
117 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
118 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
120 def add_lease(storage_index=StorageIndex,
121 renew_secret=LeaseRenewSecret,
122 cancel_secret=LeaseCancelSecret):
124 Add a new lease on the given bucket. If the renew_secret matches an
125 existing lease, that lease will be renewed instead. If there is no
126 bucket for the given storage_index, return silently. (note that in
127 tahoe-1.3.0 and earlier, IndexError was raised if there was no
130 return Any() # returns None now, but future versions might change
132 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
134 Renew the lease on a given bucket, resetting the timer to 31 days.
135 Some networks will use this, some will not. If there is no bucket for
136 the given storage_index, IndexError will be raised.
138 For mutable shares, if the given renew_secret does not match an
139 existing lease, IndexError will be raised with a note listing the
140 server-nodeids on the existing leases, so leases on migrated shares
141 can be renewed or cancelled. For immutable shares, IndexError
142 (without the note) will be raised.
146 def get_buckets(storage_index=StorageIndex):
147 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
151 def slot_readv(storage_index=StorageIndex,
152 shares=ListOf(int), readv=ReadVector):
153 """Read a vector from the numbered shares associated with the given
154 storage index. An empty shares list means to return data from all
155 known shares. Returns a dictionary with one key per share."""
156 return DictOf(int, ReadData) # shnum -> results
158 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
159 secrets=TupleOf(WriteEnablerSecret,
162 tw_vectors=TestAndWriteVectorsForShares,
166 General-purpose test-read-and-set operation for mutable slots:
167 (1) For submitted shnums, compare the test vectors against extant
168 shares, or against an empty share for shnums that do not exist.
169 (2) Use the read vectors to extract "old data" from extant shares.
170 (3) If all tests in (1) passed, then apply the write vectors
171 (possibly creating new shares).
172 (4) Return whether the tests passed, and the "old data", which does
173 not include any modifications made by the writes.
175 The operation does not interleave with other operations on the same
178 This method is, um, large. The goal is to allow clients to update all
179 the shares associated with a mutable file in a single round trip.
181 @param storage_index: the index of the bucket to be created or
183 @param write_enabler: a secret that is stored along with the slot.
184 Writes are accepted from any caller who can
185 present the matching secret. A different secret
186 should be used for each slot*server pair.
187 @param renew_secret: This is the secret used to protect bucket refresh
188 This secret is generated by the client and
189 stored for later comparison by the server. Each
190 server is given a different secret.
191 @param cancel_secret: Like renew_secret, but protects bucket decref.
193 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
194 cancel_secret). The first is required to perform any write. The
195 latter two are used when allocating new shares. To simply acquire a
196 new lease on existing shares, use an empty testv and an empty writev.
198 Each share can have a separate test vector (i.e. a list of
199 comparisons to perform). If all vectors for all shares pass, then all
200 writes for all shares are recorded. Each comparison is a 4-tuple of
201 (offset, length, operator, specimen), which effectively does a
202 bool( (read(offset, length)) OPERATOR specimen ) and only performs
203 the write if all these evaluate to True. Basic test-and-set uses 'eq'.
204 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
205 Write-if-same-or-newer uses 'le'.
207 Reads from the end of the container are truncated, and missing shares
208 behave like empty ones, so to assert that a share doesn't exist (for
209 use when creating a new share), use (0, 1, 'eq', '').
211 The write vector will be applied to the given share, expanding it if
212 necessary. A write vector applied to a share number that did not
213 exist previously will cause that share to be created. Write vectors
214 must not overlap (if they do, this will either cause an error or
215 apply them in an unspecified order). Duplicate write vectors, with
216 the same offset and data, are currently tolerated but are not
219 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
220 vector whose offset is beyond the end of the current data, the space
221 between the end of the current data and the beginning of the write
222 vector will be filled with zero bytes. In earlier versions the
223 contents of this space was unspecified (and might end up containing
224 secrets). Storage servers with the new zero-filling behavior will
225 advertise a true value for the 'fills-holes-with-zero-bytes' key
226 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
229 Each write vector is accompanied by a 'new_length' argument, which
230 can be used to truncate the data. If new_length is not None and it is
231 less than the current size of the data (after applying all write
232 vectors), then the data will be truncated to new_length. If
233 new_length==0, the share will be deleted.
235 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
236 enlarge the file by sending a number larger than the size of the data
237 after applying all write vectors. That behavior was not used, and as
238 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
241 If a storage client knows that the server supports zero-filling, for
242 example from the 'fills-holes-with-zero-bytes' key in its version
243 information, it can extend the file efficiently by writing a single
244 zero byte just before the new end-of-file. Otherwise it must
245 explicitly write zeroes to all bytes between the old and new
246 end-of-file. In any case it should avoid sending new_length larger
247 than the size of the data after applying all write vectors.
249 The read vector is used to extract data from all known shares,
250 *before* any writes have been applied. The same read vector is used
251 for all shares. This captures the state that was tested by the test
252 vector, for extant shares.
254 This method returns two values: a boolean and a dict. The boolean is
255 True if the write vectors were applied, False if not. The dict is
256 keyed by share number, and each value contains a list of strings, one
257 for each element of the read vector.
259 If the write_enabler is wrong, this will raise BadWriteEnablerError.
260 To enable share migration (using update_write_enabler), the exception
261 will have the nodeid used for the old write enabler embedded in it,
262 in the following string::
264 The write enabler was recorded by nodeid '%s'.
266 Note that the nodeid here is encoded using the same base32 encoding
267 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
269 return TupleOf(bool, DictOf(int, ReadData))
271 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
272 shnum=int, reason=str):
273 """Clients who discover hash failures in shares that they have
274 downloaded from me will use this method to inform me about the
275 failures. I will record their concern so that my operator can
276 manually inspect the shares in question. I return None.
278 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
279 (binary) storage index string, and 'shnum' is the integer share
280 number. 'reason' is a human-readable explanation of the problem,
281 probably including some expected hash values and the computed ones
282 that did not match. Corruption advisories for mutable shares should
283 include a hash of the public key (the same value that appears in the
284 mutable-file verify-cap), since the current share format does not
289 class IStorageBucketWriter(Interface):
291 Objects of this kind live on the client side.
293 def put_block(segmentnum, data):
295 @param segmentnum=int
296 @param data=ShareData: For most segments, this data will be 'blocksize'
297 bytes in length. The last segment might be shorter.
298 @return: a Deferred that fires (with None) when the operation completes
301 def put_crypttext_hashes(hashes):
303 @param hashes=ListOf(Hash)
304 @return: a Deferred that fires (with None) when the operation completes
307 def put_block_hashes(blockhashes):
309 @param blockhashes=ListOf(Hash)
310 @return: a Deferred that fires (with None) when the operation completes
313 def put_share_hashes(sharehashes):
315 @param sharehashes=ListOf(TupleOf(int, Hash))
316 @return: a Deferred that fires (with None) when the operation completes
319 def put_uri_extension(data):
320 """This block of data contains integrity-checking information (hashes
321 of plaintext, crypttext, and shares), as well as encoding parameters
322 that are necessary to recover the data. This is a serialized dict
323 mapping strings to other strings. The hash of this data is kept in
324 the URI and verified before any of the data is used. All buckets for
325 a given file contain identical copies of this data.
327 The serialization format is specified with the following pseudocode:
328 for k in sorted(dict.keys()):
329 assert re.match(r'^[a-zA-Z_\-]+$', k)
330 write(k + ':' + netstring(dict[k]))
332 @param data=URIExtensionData
333 @return: a Deferred that fires (with None) when the operation completes
337 """Finish writing and close the bucket. The share is not finalized
338 until this method is called: if the uploading client disconnects
339 before calling close(), the partially-written share will be
342 @return: a Deferred that fires (with None) when the operation completes
346 class IStorageBucketReader(Interface):
347 def get_block_data(blocknum, blocksize, size):
348 """Most blocks will be the same size. The last block might be shorter
357 def get_crypttext_hashes():
359 @return: ListOf(Hash)
362 def get_block_hashes(at_least_these=()):
364 @param at_least_these=SetOf(int)
365 @return: ListOf(Hash)
368 def get_share_hashes():
370 @return: ListOf(TupleOf(int, Hash))
373 def get_uri_extension():
375 @return: URIExtensionData
379 class IStorageBroker(Interface):
380 def get_servers_for_psi(peer_selection_index):
382 @return: list of IServer instances
384 def get_connected_servers():
386 @return: frozenset of connected IServer instances
388 def get_known_servers():
390 @return: frozenset of IServer instances
392 def get_all_serverids():
394 @return: frozenset of serverid strings
396 def get_nickname_for_serverid(serverid):
398 @return: unicode nickname, or None
401 # methods moved from IntroducerClient, need review
402 def get_all_connections():
403 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
404 each active connection we've established to a remote service. This is
405 mostly useful for unit tests that need to wait until a certain number
406 of connections have been made."""
408 def get_all_connectors():
409 """Return a dict that maps from (nodeid, service_name) to a
410 RemoteServiceConnector instance for all services that we are actively
411 trying to connect to. Each RemoteServiceConnector has the following
414 service_name: the type of service provided, like 'storage'
415 announcement_time: when we first heard about this service
416 last_connect_time: when we last established a connection
417 last_loss_time: when we last lost a connection
419 version: the peer's version, from the most recent connection
420 oldest_supported: the peer's oldest supported version, same
422 rref: the RemoteReference, if connected, otherwise None
423 remote_host: the IAddress, if connected, otherwise None
425 This method is intended for monitoring interfaces, such as a web page
426 that describes connecting and connected peers.
429 def get_all_peerids():
430 """Return a frozenset of all peerids to whom we have a connection (to
431 one or more services) established. Mostly useful for unit tests."""
433 def get_all_connections_for(service_name):
434 """Return a frozenset of (nodeid, service_name, rref) tuples, one
435 for each active connection that provides the given SERVICE_NAME."""
437 def get_permuted_peers(service_name, key):
438 """Returns an ordered list of (peerid, rref) tuples, selecting from
439 the connections that provide SERVICE_NAME, using a hash-based
440 permutation keyed by KEY. This randomizes the service list in a
441 repeatable way, to distribute load over many peers.
445 class IDisplayableServer(Interface):
456 class IServer(IDisplayableServer):
457 """I live in the client, and represent a single server."""
458 def start_connecting(tub, trigger_cb):
462 """Once a server is connected, I return a RemoteReference.
463 Before a server is connected for the first time, I return None.
465 Note that the rref I return will start producing DeadReferenceErrors
466 once the connection is lost.
470 class IMutableSlotWriter(Interface):
472 The interface for a writer around a mutable slot on a remote server.
474 def set_checkstring(seqnum_or_checkstring, root_hash=None, salt=None):
476 Set the checkstring that I will pass to the remote server when
479 @param checkstring A packed checkstring to use.
481 Note that implementations can differ in which semantics they
482 wish to support for set_checkstring -- they can, for example,
483 build the checkstring themselves from its constituents, or
487 def get_checkstring():
489 Get the checkstring that I think currently exists on the remote
493 def put_block(data, segnum, salt):
495 Add a block and salt to the share.
498 def put_encprivkey(encprivkey):
500 Add the encrypted private key to the share.
503 def put_blockhashes(blockhashes):
505 @param blockhashes=list
506 Add the block hash tree to the share.
509 def put_sharehashes(sharehashes):
511 @param sharehashes=dict
512 Add the share hash chain to the share.
517 Return the part of the share that needs to be signed.
520 def put_signature(signature):
522 Add the signature to the share.
525 def put_verification_key(verification_key):
527 Add the verification key to the share.
530 def finish_publishing():
532 Do anything necessary to finish writing the share to a remote
533 server. I require that no further publishing needs to take place
534 after this method has been called.
538 class IURI(Interface):
539 def init_from_string(uri):
540 """Accept a string (as created by my to_string() method) and populate
541 this instance with its data. I am not normally called directly,
542 please use the module-level uri.from_string() function to convert
543 arbitrary URI strings into IURI-providing instances."""
546 """Return False if this URI be used to modify the data. Return True
547 if this URI cannot be used to modify the data."""
550 """Return True if the data can be modified by *somebody* (perhaps
551 someone who has a more powerful URI than this one)."""
553 # TODO: rename to get_read_cap()
555 """Return another IURI instance that represents a read-only form of
556 this one. If is_readonly() is True, this returns self."""
558 def get_verify_cap():
559 """Return an instance that provides IVerifierURI, which can be used
560 to check on the availability of the file or directory, without
561 providing enough capabilities to actually read or modify the
562 contents. This may return None if the file does not need checking or
563 verification (e.g. LIT URIs).
567 """Return a string of printable ASCII characters, suitable for
568 passing into init_from_string."""
571 class IVerifierURI(Interface, IURI):
572 def init_from_string(uri):
573 """Accept a string (as created by my to_string() method) and populate
574 this instance with its data. I am not normally called directly,
575 please use the module-level uri.from_string() function to convert
576 arbitrary URI strings into IURI-providing instances."""
579 """Return a string of printable ASCII characters, suitable for
580 passing into init_from_string."""
583 class IDirnodeURI(Interface):
584 """I am a URI that represents a dirnode."""
587 class IFileURI(Interface):
588 """I am a URI that represents a filenode."""
590 """Return the length (in bytes) of the file that I represent."""
593 class IImmutableFileURI(IFileURI):
596 class IMutableFileURI(Interface):
599 class IDirectoryURI(Interface):
602 class IReadonlyDirectoryURI(Interface):
606 class CapConstraintError(Exception):
607 """A constraint on a cap was violated."""
609 class MustBeDeepImmutableError(CapConstraintError):
610 """Mutable children cannot be added to an immutable directory.
611 Also, caps obtained from an immutable directory can trigger this error
612 if they are later found to refer to a mutable object and then used."""
614 class MustBeReadonlyError(CapConstraintError):
615 """Known write caps cannot be specified in a ro_uri field. Also,
616 caps obtained from a ro_uri field can trigger this error if they
617 are later found to be write caps and then used."""
619 class MustNotBeUnknownRWError(CapConstraintError):
620 """Cannot add an unknown child cap specified in a rw_uri field."""
623 class IReadable(Interface):
624 """I represent a readable object -- either an immutable file, or a
625 specific version of a mutable file.
629 """Return True if this reference provides mutable access to the given
630 file or directory (i.e. if you can modify it), or False if not. Note
631 that even if this reference is read-only, someone else may hold a
632 read-write reference to it.
634 For an IReadable returned by get_best_readable_version(), this will
635 always return True, but for instances of subinterfaces such as
636 IMutableFileVersion, it may return False."""
639 """Return True if this file or directory is mutable (by *somebody*,
640 not necessarily you), False if it is is immutable. Note that a file
641 might be mutable overall, but your reference to it might be
642 read-only. On the other hand, all references to an immutable file
643 will be read-only; there are no read-write references to an immutable
646 def get_storage_index():
647 """Return the storage index of the file."""
650 """Return the length (in bytes) of this readable object."""
652 def download_to_data():
653 """Download all of the file contents. I return a Deferred that fires
654 with the contents as a byte string."""
656 def read(consumer, offset=0, size=None):
657 """Download a portion (possibly all) of the file's contents, making
658 them available to the given IConsumer. Return a Deferred that fires
659 (with the consumer) when the consumer is unregistered (either because
660 the last byte has been given to it, or because the consumer threw an
661 exception during write(), possibly because it no longer wants to
662 receive data). The portion downloaded will start at 'offset' and
663 contain 'size' bytes (or the remainder of the file if size==None).
665 The consumer will be used in non-streaming mode: an IPullProducer
666 will be attached to it.
668 The consumer will not receive data right away: several network trips
669 must occur first. The order of events will be::
671 consumer.registerProducer(p, streaming)
672 (if streaming == False)::
673 consumer does p.resumeProducing()
675 consumer does p.resumeProducing()
676 consumer.write(data).. (repeat until all data is written)
677 consumer.unregisterProducer()
678 deferred.callback(consumer)
680 If a download error occurs, or an exception is raised by
681 consumer.registerProducer() or consumer.write(), I will call
682 consumer.unregisterProducer() and then deliver the exception via
683 deferred.errback(). To cancel the download, the consumer should call
684 p.stopProducing(), which will result in an exception being delivered
685 via deferred.errback().
687 See src/allmydata/util/consumer.py for an example of a simple
688 download-to-memory consumer.
692 class IWriteable(Interface):
694 I define methods that callers can use to update SDMF and MDMF
695 mutable files on a Tahoe-LAFS grid.
697 # XXX: For the moment, we have only this. It is possible that we
698 # want to move overwrite() and modify() in here too.
699 def update(data, offset):
701 I write the data from my data argument to the MDMF file,
702 starting at offset. I continue writing data until my data
703 argument is exhausted, appending data to the file as necessary.
705 # assert IMutableUploadable.providedBy(data)
706 # to append data: offset=node.get_size_of_best_version()
707 # do we want to support compacting MDMF?
708 # for an MDMF file, this can be done with O(data.get_size())
709 # memory. For an SDMF file, any modification takes
710 # O(node.get_size_of_best_version()).
713 class IMutableFileVersion(IReadable):
714 """I provide access to a particular version of a mutable file. The
715 access is read/write if I was obtained from a filenode derived from
716 a write cap, or read-only if the filenode was derived from a read cap.
719 def get_sequence_number():
720 """Return the sequence number of this version."""
723 """Return the IMutableFileServerMap instance that was used to create
728 """Return this filenode's writekey, or None if the node does not have
729 write-capability. This may be used to assist with data structures
730 that need to make certain data available only to writers, such as the
731 read-write child caps in dirnodes. The recommended process is to have
732 reader-visible data be submitted to the filenode in the clear (where
733 it will be encrypted by the filenode using the readkey), but encrypt
734 writer-visible data using this writekey.
737 def overwrite(new_contents):
738 """Replace the contents of the mutable file, provided that no other
739 node has published (or is attempting to publish, concurrently) a
740 newer version of the file than this one.
742 I will avoid modifying any share that is different than the version
743 given by get_sequence_number(). However, if another node is writing
744 to the file at the same time as me, I may manage to update some shares
745 while they update others. If I see any evidence of this, I will signal
746 UncoordinatedWriteError, and the file will be left in an inconsistent
747 state (possibly the version you provided, possibly the old version,
748 possibly somebody else's version, and possibly a mix of shares from
751 The recommended response to UncoordinatedWriteError is to either
752 return it to the caller (since they failed to coordinate their
753 writes), or to attempt some sort of recovery. It may be sufficient to
754 wait a random interval (with exponential backoff) and repeat your
755 operation. If I do not signal UncoordinatedWriteError, then I was
756 able to write the new version without incident.
758 I return a Deferred that fires (with a PublishStatus object) when the
759 update has completed.
762 def modify(modifier_cb):
763 """Modify the contents of the file, by downloading this version,
764 applying the modifier function (or bound method), then uploading
765 the new version. This will succeed as long as no other node
766 publishes a version between the download and the upload.
767 I return a Deferred that fires (with a PublishStatus object) when
768 the update is complete.
770 The modifier callable will be given three arguments: a string (with
771 the old contents), a 'first_time' boolean, and a servermap. As with
772 download_to_data(), the old contents will be from this version,
773 but the modifier can use the servermap to make other decisions
774 (such as refusing to apply the delta if there are multiple parallel
775 versions, or if there is evidence of a newer unrecoverable version).
776 'first_time' will be True the first time the modifier is called,
777 and False on any subsequent calls.
779 The callable should return a string with the new contents. The
780 callable must be prepared to be called multiple times, and must
781 examine the input string to see if the change that it wants to make
782 is already present in the old version. If it does not need to make
783 any changes, it can either return None, or return its input string.
785 If the modifier raises an exception, it will be returned in the
790 # The hierarchy looks like this:
797 class IFilesystemNode(Interface):
799 """Return the strongest 'cap instance' associated with this node.
800 (writecap for writeable-mutable files/directories, readcap for
801 immutable or readonly-mutable files/directories). To convert this
802 into a string, call .to_string() on the result."""
805 """Return a readonly cap instance for this node. For immutable or
806 readonly nodes, get_cap() and get_readcap() return the same thing."""
808 def get_repair_cap():
809 """Return an IURI instance that can be used to repair the file, or
810 None if this node cannot be repaired (either because it is not
811 distributed, like a LIT file, or because the node does not represent
812 sufficient authority to create a repair-cap, like a read-only RSA
813 mutable file node [which cannot create the correct write-enablers]).
816 def get_verify_cap():
817 """Return an IVerifierURI instance that represents the
818 'verifiy/refresh capability' for this node. The holder of this
819 capability will be able to renew the lease for this node, protecting
820 it from garbage-collection. They will also be able to ask a server if
821 it holds a share for the file or directory.
825 """Return the URI string corresponding to the strongest cap associated
826 with this node. If this node is read-only, the URI will only offer
827 read-only access. If this node is read-write, the URI will offer
830 If you have read-write access to a node and wish to share merely
831 read-only access with others, use get_readonly_uri().
835 """Return the URI string that can be used by others to get write
836 access to this node, if it is writeable. If this is a read-only node,
839 def get_readonly_uri():
840 """Return the URI string that can be used by others to get read-only
841 access to this node. The result is a read-only URI, regardless of
842 whether this node is read-only or read-write.
844 If you have merely read-only access to this node, get_readonly_uri()
845 will return the same thing as get_uri().
848 def get_storage_index():
849 """Return a string with the (binary) storage index in use on this
850 download. This may be None if there is no storage index (i.e. LIT
851 files and directories)."""
854 """Return True if this reference provides mutable access to the given
855 file or directory (i.e. if you can modify it), or False if not. Note
856 that even if this reference is read-only, someone else may hold a
857 read-write reference to it."""
860 """Return True if this file or directory is mutable (by *somebody*,
861 not necessarily you), False if it is is immutable. Note that a file
862 might be mutable overall, but your reference to it might be
863 read-only. On the other hand, all references to an immutable file
864 will be read-only; there are no read-write references to an immutable
869 """Return True if this is an unknown node."""
871 def is_allowed_in_immutable_directory():
872 """Return True if this node is allowed as a child of a deep-immutable
873 directory. This is true if either the node is of a known-immutable type,
874 or it is unknown and read-only.
878 """Raise any error associated with this node."""
880 # XXX: These may not be appropriate outside the context of an IReadable.
882 """Return the length (in bytes) of the data this node represents. For
883 directory nodes, I return the size of the backing store. I return
884 synchronously and do not consult the network, so for mutable objects,
885 I will return the most recently observed size for the object, or None
886 if I don't remember a size. Use get_current_size, which returns a
887 Deferred, if you want more up-to-date information."""
889 def get_current_size():
890 """I return a Deferred that fires with the length (in bytes) of the
891 data this node represents.
895 class IFileNode(IFilesystemNode):
896 """I am a node that represents a file: a sequence of bytes. I am not a
897 container, like IDirectoryNode."""
898 def get_best_readable_version():
899 """Return a Deferred that fires with an IReadable for the 'best'
900 available version of the file. The IReadable provides only read
901 access, even if this filenode was derived from a write cap.
903 For an immutable file, there is only one version. For a mutable
904 file, the 'best' version is the recoverable version with the
905 highest sequence number. If no uncoordinated writes have occurred,
906 and if enough shares are available, then this will be the most
907 recent version that has been uploaded. If no version is recoverable,
908 the Deferred will errback with an UnrecoverableFileError.
911 def download_best_version():
912 """Download the contents of the version that would be returned
913 by get_best_readable_version(). This is equivalent to calling
914 download_to_data() on the IReadable given by that method.
916 I return a Deferred that fires with a byte string when the file
917 has been fully downloaded. To support streaming download, use
918 the 'read' method of IReadable. If no version is recoverable,
919 the Deferred will errback with an UnrecoverableFileError.
922 def get_size_of_best_version():
923 """Find the size of the version that would be returned by
924 get_best_readable_version().
926 I return a Deferred that fires with an integer. If no version
927 is recoverable, the Deferred will errback with an
928 UnrecoverableFileError.
932 class IImmutableFileNode(IFileNode, IReadable):
933 """I am a node representing an immutable file. Immutable files have
937 class IMutableFileNode(IFileNode):
938 """I provide access to a 'mutable file', which retains its identity
939 regardless of what contents are put in it.
941 The consistency-vs-availability problem means that there might be
942 multiple versions of a file present in the grid, some of which might be
943 unrecoverable (i.e. have fewer than 'k' shares). These versions are
944 loosely ordered: each has a sequence number and a hash, and any version
945 with seqnum=N was uploaded by a node that has seen at least one version
948 The 'servermap' (an instance of IMutableFileServerMap) is used to
949 describe the versions that are known to be present in the grid, and which
950 servers are hosting their shares. It is used to represent the 'state of
951 the world', and is used for this purpose by my test-and-set operations.
952 Downloading the contents of the mutable file will also return a
953 servermap. Uploading a new version into the mutable file requires a
954 servermap as input, and the semantics of the replace operation is
955 'replace the file with my new version if it looks like nobody else has
956 changed the file since my previous download'. Because the file is
957 distributed, this is not a perfect test-and-set operation, but it will do
958 its best. If the replace process sees evidence of a simultaneous write,
959 it will signal an UncoordinatedWriteError, so that the caller can take
963 Most readers will want to use the 'best' current version of the file, and
964 should use my 'download_best_version()' method.
966 To unconditionally replace the file, callers should use overwrite(). This
967 is the mode that user-visible mutable files will probably use.
969 To apply some delta to the file, call modify() with a callable modifier
970 function that can apply the modification that you want to make. This is
971 the mode that dirnodes will use, since most directory modification
972 operations can be expressed in terms of deltas to the directory state.
975 Three methods are available for users who need to perform more complex
976 operations. The first is get_servermap(), which returns an up-to-date
977 servermap using a specified mode. The second is download_version(), which
978 downloads a specific version (not necessarily the 'best' one). The third
979 is 'upload', which accepts new contents and a servermap (which must have
980 been updated with MODE_WRITE). The upload method will attempt to apply
981 the new contents as long as no other node has modified the file since the
982 servermap was updated. This might be useful to a caller who wants to
983 merge multiple versions into a single new one.
985 Note that each time the servermap is updated, a specific 'mode' is used,
986 which determines how many peers are queried. To use a servermap for my
987 replace() method, that servermap must have been updated in MODE_WRITE.
988 These modes are defined in allmydata.mutable.common, and consist of
989 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
990 allmydata/mutable/servermap.py for details about the differences.
992 Mutable files are currently limited in size (about 3.5MB max) and can
993 only be retrieved and updated all-at-once, as a single big string. Future
994 versions of our mutable files will remove this restriction.
996 def get_best_mutable_version():
997 """Return a Deferred that fires with an IMutableFileVersion for
998 the 'best' available version of the file. The best version is
999 the recoverable version with the highest sequence number. If no
1000 uncoordinated writes have occurred, and if enough shares are
1001 available, then this will be the most recent version that has
1004 If no version is recoverable, the Deferred will errback with an
1005 UnrecoverableFileError.
1008 def overwrite(new_contents):
1009 """Unconditionally replace the contents of the mutable file with new
1010 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1011 is only appropriate to use when the new contents of the file are
1012 completely unrelated to the old ones, and you do not care about other
1015 I return a Deferred that fires (with a PublishStatus object) when the
1016 update has completed.
1019 def modify(modifier_cb):
1020 """Modify the contents of the file, by downloading the current
1021 version, applying the modifier function (or bound method), then
1022 uploading the new version. I return a Deferred that fires (with a
1023 PublishStatus object) when the update is complete.
1025 The modifier callable will be given three arguments: a string (with
1026 the old contents), a 'first_time' boolean, and a servermap. As with
1027 download_best_version(), the old contents will be from the best
1028 recoverable version, but the modifier can use the servermap to make
1029 other decisions (such as refusing to apply the delta if there are
1030 multiple parallel versions, or if there is evidence of a newer
1031 unrecoverable version). 'first_time' will be True the first time the
1032 modifier is called, and False on any subsequent calls.
1034 The callable should return a string with the new contents. The
1035 callable must be prepared to be called multiple times, and must
1036 examine the input string to see if the change that it wants to make
1037 is already present in the old version. If it does not need to make
1038 any changes, it can either return None, or return its input string.
1040 If the modifier raises an exception, it will be returned in the
1044 def get_servermap(mode):
1045 """Return a Deferred that fires with an IMutableFileServerMap
1046 instance, updated using the given mode.
1049 def download_version(servermap, version):
1050 """Download a specific version of the file, using the servermap
1051 as a guide to where the shares are located.
1053 I return a Deferred that fires with the requested contents, or
1054 errbacks with UnrecoverableFileError. Note that a servermap that was
1055 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1056 all versions (those modes stop querying servers as soon as they can
1057 fulfil their goals), so you may want to use MODE_CHECK (which checks
1058 everything) to get increased visibility.
1061 def upload(new_contents, servermap):
1062 """Replace the contents of the file with new ones. This requires a
1063 servermap that was previously updated with MODE_WRITE.
1065 I attempt to provide test-and-set semantics, in that I will avoid
1066 modifying any share that is different than the version I saw in the
1067 servermap. However, if another node is writing to the file at the
1068 same time as me, I may manage to update some shares while they update
1069 others. If I see any evidence of this, I will signal
1070 UncoordinatedWriteError, and the file will be left in an inconsistent
1071 state (possibly the version you provided, possibly the old version,
1072 possibly somebody else's version, and possibly a mix of shares from
1075 The recommended response to UncoordinatedWriteError is to either
1076 return it to the caller (since they failed to coordinate their
1077 writes), or to attempt some sort of recovery. It may be sufficient to
1078 wait a random interval (with exponential backoff) and repeat your
1079 operation. If I do not signal UncoordinatedWriteError, then I was
1080 able to write the new version without incident.
1082 I return a Deferred that fires (with a PublishStatus object) when the
1083 publish has completed. I will update the servermap in-place with the
1084 location of all new shares.
1088 """Return this filenode's writekey, or None if the node does not have
1089 write-capability. This may be used to assist with data structures
1090 that need to make certain data available only to writers, such as the
1091 read-write child caps in dirnodes. The recommended process is to have
1092 reader-visible data be submitted to the filenode in the clear (where
1093 it will be encrypted by the filenode using the readkey), but encrypt
1094 writer-visible data using this writekey.
1098 """Returns the mutable file protocol version."""
1101 class NotEnoughSharesError(Exception):
1102 """Download was unable to get enough shares"""
1104 class NoSharesError(Exception):
1105 """Download was unable to get any shares at all."""
1107 class DownloadStopped(Exception):
1110 class UploadUnhappinessError(Exception):
1111 """Upload was unable to satisfy 'servers_of_happiness'"""
1113 class UnableToFetchCriticalDownloadDataError(Exception):
1114 """I was unable to fetch some piece of critical data that is supposed to
1115 be identically present in all shares."""
1117 class NoServersError(Exception):
1118 """Upload wasn't given any servers to work with, usually indicating a
1119 network or Introducer problem."""
1121 class ExistingChildError(Exception):
1122 """A directory node was asked to add or replace a child that already
1123 exists, and overwrite= was set to False."""
1125 class NoSuchChildError(Exception):
1126 """A directory node was asked to fetch a child that does not exist."""
1128 # avoid UnicodeEncodeErrors when converting to str
1129 return self.__repr__()
1131 class ChildOfWrongTypeError(Exception):
1132 """An operation was attempted on a child of the wrong type (file or directory)."""
1135 class IDirectoryNode(IFilesystemNode):
1136 """I represent a filesystem node that is a container, with a
1137 name-to-child mapping, holding the tahoe equivalent of a directory. All
1138 child names are unicode strings, and all children are some sort of
1139 IFilesystemNode (a file, subdirectory, or unknown node).
1144 The dirnode ('1') URI returned by this method can be used in
1145 set_uri() on a different directory ('2') to 'mount' a reference to
1146 this directory ('1') under the other ('2'). This URI is just a
1147 string, so it can be passed around through email or other out-of-band
1151 def get_readonly_uri():
1153 The dirnode ('1') URI returned by this method can be used in
1154 set_uri() on a different directory ('2') to 'mount' a reference to
1155 this directory ('1') under the other ('2'). This URI is just a
1156 string, so it can be passed around through email or other out-of-band
1161 """I return a Deferred that fires with a dictionary mapping child
1162 name (a unicode string) to (node, metadata_dict) tuples, in which
1163 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1166 def has_child(name):
1167 """I return a Deferred that fires with a boolean, True if there
1168 exists a child of the given name, False if not. The child name must
1169 be a unicode string."""
1172 """I return a Deferred that fires with a specific named child node,
1173 which is an IFilesystemNode. The child name must be a unicode string.
1174 I raise NoSuchChildError if I do not have a child by that name."""
1176 def get_metadata_for(name):
1177 """I return a Deferred that fires with the metadata dictionary for
1178 a specific named child node. The child name must be a unicode string.
1179 This metadata is stored in the *edge*, not in the child, so it is
1180 attached to the parent dirnode rather than the child node.
1181 I raise NoSuchChildError if I do not have a child by that name."""
1183 def set_metadata_for(name, metadata):
1184 """I replace any existing metadata for the named child with the new
1185 metadata. The child name must be a unicode string. This metadata is
1186 stored in the *edge*, not in the child, so it is attached to the
1187 parent dirnode rather than the child node. I return a Deferred
1188 (that fires with this dirnode) when the operation is complete.
1189 I raise NoSuchChildError if I do not have a child by that name."""
1191 def get_child_at_path(path):
1192 """Transform a child path into an IFilesystemNode.
1194 I perform a recursive series of 'get' operations to find the named
1195 descendant node. I return a Deferred that fires with the node, or
1196 errbacks with NoSuchChildError if the node could not be found.
1198 The path can be either a single string (slash-separated) or a list of
1199 path-name elements. All elements must be unicode strings.
1202 def get_child_and_metadata_at_path(path):
1203 """Transform a child path into an IFilesystemNode and metadata.
1205 I am like get_child_at_path(), but my Deferred fires with a tuple of
1206 (node, metadata). The metadata comes from the last edge. If the path
1207 is empty, the metadata will be an empty dictionary.
1210 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1211 """I add a child (by writecap+readcap) at the specific name. I return
1212 a Deferred that fires when the operation finishes. If overwrite= is
1213 True, I will replace any existing child of the same name, otherwise
1214 an existing child will cause me to return ExistingChildError. The
1215 child name must be a unicode string.
1217 The child caps could be for a file, or for a directory. If you have
1218 both the writecap and readcap, you should provide both arguments.
1219 If you have only one cap and don't know whether it is read-only,
1220 provide it as the writecap argument and leave the readcap as None.
1221 If you have only one cap that is known to be read-only, provide it
1222 as the readcap argument and leave the writecap as None.
1223 The filecaps are typically obtained from an IFilesystemNode with
1224 get_uri() and get_readonly_uri().
1226 If metadata= is provided, I will use it as the metadata for the named
1227 edge. This will replace any existing metadata. If metadata= is left
1228 as the default value of None, I will set ['mtime'] to the current
1229 time, and I will set ['ctime'] to the current time if there was not
1230 already a child by this name present. This roughly matches the
1231 ctime/mtime semantics of traditional filesystems. See the
1232 "About the metadata" section of webapi.txt for futher information.
1234 If this directory node is read-only, the Deferred will errback with a
1235 NotWriteableError."""
1237 def set_children(entries, overwrite=True):
1238 """Add multiple children (by writecap+readcap) to a directory node.
1239 Takes a dictionary, with childname as keys and (writecap, readcap)
1240 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1241 a Deferred that fires (with this dirnode) when the operation
1242 finishes. This is equivalent to calling set_uri() multiple times, but
1243 is much more efficient. All child names must be unicode strings.
1246 def set_node(name, child, metadata=None, overwrite=True):
1247 """I add a child at the specific name. I return a Deferred that fires
1248 when the operation finishes. This Deferred will fire with the child
1249 node that was just added. I will replace any existing child of the
1250 same name. The child name must be a unicode string. The 'child'
1251 instance must be an instance providing IFilesystemNode.
1253 If metadata= is provided, I will use it as the metadata for the named
1254 edge. This will replace any existing metadata. If metadata= is left
1255 as the default value of None, I will set ['mtime'] to the current
1256 time, and I will set ['ctime'] to the current time if there was not
1257 already a child by this name present. This roughly matches the
1258 ctime/mtime semantics of traditional filesystems. See the
1259 "About the metadata" section of webapi.txt for futher information.
1261 If this directory node is read-only, the Deferred will errback with a
1262 NotWriteableError."""
1264 def set_nodes(entries, overwrite=True):
1265 """Add multiple children to a directory node. Takes a dict mapping
1266 unicode childname to (child_node, metdata) tuples. If metdata=None,
1267 the original metadata is left unmodified. Returns a Deferred that
1268 fires (with this dirnode) when the operation finishes. This is
1269 equivalent to calling set_node() multiple times, but is much more
1272 def add_file(name, uploadable, metadata=None, overwrite=True):
1273 """I upload a file (using the given IUploadable), then attach the
1274 resulting ImmutableFileNode to the directory at the given name. I set
1275 metadata the same way as set_uri and set_node. The child name must be
1278 I return a Deferred that fires (with the IFileNode of the uploaded
1279 file) when the operation completes."""
1281 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1282 """I remove the child at the specific name. I return a Deferred that
1283 fires when the operation finishes. The child name must be a unicode
1284 string. If must_exist is True and I do not have a child by that name,
1285 I raise NoSuchChildError. If must_be_directory is True and the child
1286 is a file, or if must_be_file is True and the child is a directory,
1287 I raise ChildOfWrongTypeError."""
1289 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1290 """I create and attach a directory at the given name. The new
1291 directory can be empty, or it can be populated with children
1292 according to 'initial_children', which takes a dictionary in the same
1293 format as set_nodes (i.e. mapping unicode child name to (childnode,
1294 metadata) tuples). The child name must be a unicode string. I return
1295 a Deferred that fires (with the new directory node) when the
1296 operation finishes."""
1298 def move_child_to(current_child_name, new_parent, new_child_name=None,
1300 """I take one of my children and move them to a new parent. The child
1301 is referenced by name. On the new parent, the child will live under
1302 'new_child_name', which defaults to 'current_child_name'. TODO: what
1303 should we do about metadata? I return a Deferred that fires when the
1304 operation finishes. The child name must be a unicode string. I raise
1305 NoSuchChildError if I do not have a child by that name."""
1307 def build_manifest():
1308 """I generate a table of everything reachable from this directory.
1309 I also compute deep-stats as described below.
1311 I return a Monitor. The Monitor's results will be a dictionary with
1314 res['manifest']: a list of (path, cap) tuples for all nodes
1315 (directories and files) reachable from this one.
1316 'path' will be a tuple of unicode strings. The
1317 origin dirnode will be represented by an empty path
1319 res['verifycaps']: a list of (printable) verifycap strings, one for
1320 each reachable non-LIT node. This is a set:
1321 it will contain no duplicates.
1322 res['storage-index']: a list of (base32) storage index strings,
1323 one for each reachable non-LIT node. This is
1324 a set: it will contain no duplicates.
1325 res['stats']: a dictionary, the same that is generated by
1326 start_deep_stats() below.
1328 The Monitor will also have an .origin_si attribute with the (binary)
1329 storage index of the starting point.
1332 def start_deep_stats():
1333 """Return a Monitor, examining all nodes (directories and files)
1334 reachable from this one. The Monitor's results will be a dictionary
1335 with the following keys::
1337 count-immutable-files: count of how many CHK files are in the set
1338 count-mutable-files: same, for mutable files (does not include
1340 count-literal-files: same, for LIT files
1341 count-files: sum of the above three
1343 count-directories: count of directories
1345 size-immutable-files: total bytes for all CHK files in the set
1346 size-mutable-files (TODO): same, for current version of all mutable
1347 files, does not include directories
1348 size-literal-files: same, for LIT files
1349 size-directories: size of mutable files used by directories
1351 largest-directory: number of bytes in the largest directory
1352 largest-directory-children: number of children in the largest
1354 largest-immutable-file: number of bytes in the largest CHK file
1356 size-mutable-files is not yet implemented, because it would involve
1357 even more queries than deep_stats does.
1359 The Monitor will also have an .origin_si attribute with the (binary)
1360 storage index of the starting point.
1362 This operation will visit every directory node underneath this one,
1363 and can take a long time to run. On a typical workstation with good
1364 bandwidth, this can examine roughly 15 directories per second (and
1365 takes several minutes of 100% CPU for ~1700 directories).
1369 class ICodecEncoder(Interface):
1370 def set_params(data_size, required_shares, max_shares):
1371 """Set up the parameters of this encoder.
1373 This prepares the encoder to perform an operation that converts a
1374 single block of data into a number of shares, such that a future
1375 ICodecDecoder can use a subset of these shares to recover the
1376 original data. This operation is invoked by calling encode(). Once
1377 the encoding parameters are set up, the encode operation can be
1378 invoked multiple times.
1380 set_params() prepares the encoder to accept blocks of input data that
1381 are exactly 'data_size' bytes in length. The encoder will be prepared
1382 to produce 'max_shares' shares for each encode() operation (although
1383 see the 'desired_share_ids' to use less CPU). The encoding math will
1384 be chosen such that the decoder can get by with as few as
1385 'required_shares' of these shares and still reproduce the original
1386 data. For example, set_params(1000, 5, 5) offers no redundancy at
1387 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1389 Numerical Restrictions: 'data_size' is required to be an integral
1390 multiple of 'required_shares'. In general, the caller should choose
1391 required_shares and max_shares based upon their reliability
1392 requirements and the number of peers available (the total storage
1393 space used is roughly equal to max_shares*data_size/required_shares),
1394 then choose data_size to achieve the memory footprint desired (larger
1395 data_size means more efficient operation, smaller data_size means
1396 smaller memory footprint).
1398 In addition, 'max_shares' must be equal to or greater than
1399 'required_shares'. Of course, setting them to be equal causes
1400 encode() to degenerate into a particularly slow form of the 'split'
1403 See encode() for more details about how these parameters are used.
1405 set_params() must be called before any other ICodecEncoder methods
1410 """Return the 3-tuple of data_size, required_shares, max_shares"""
1412 def get_encoder_type():
1413 """Return a short string that describes the type of this encoder.
1415 There is required to be a global table of encoder classes. This method
1416 returns an index into this table; the value at this index is an
1417 encoder class, and this encoder is an instance of that class.
1420 def get_block_size():
1421 """Return the length of the shares that encode() will produce.
1424 def encode_proposal(data, desired_share_ids=None):
1425 """Encode some data.
1427 'data' must be a string (or other buffer object), and len(data) must
1428 be equal to the 'data_size' value passed earlier to set_params().
1430 This will return a Deferred that will fire with two lists. The first
1431 is a list of shares, each of which is a string (or other buffer
1432 object) such that len(share) is the same as what get_share_size()
1433 returned earlier. The second is a list of shareids, in which each is
1434 an integer. The lengths of the two lists will always be equal to each
1435 other. The user should take care to keep each share closely
1436 associated with its shareid, as one is useless without the other.
1438 The length of this output list will normally be the same as the value
1439 provided to the 'max_shares' parameter of set_params(). This may be
1440 different if 'desired_share_ids' is provided.
1442 'desired_share_ids', if provided, is required to be a sequence of
1443 ints, each of which is required to be >= 0 and < max_shares. If not
1444 provided, encode() will produce 'max_shares' shares, as if
1445 'desired_share_ids' were set to range(max_shares). You might use this
1446 if you initially thought you were going to use 10 peers, started
1447 encoding, and then two of the peers dropped out: you could use
1448 desired_share_ids= to skip the work (both memory and CPU) of
1449 producing shares for the peers that are no longer available.
1453 def encode(inshares, desired_share_ids=None):
1454 """Encode some data. This may be called multiple times. Each call is
1457 inshares is a sequence of length required_shares, containing buffers
1458 (i.e. strings), where each buffer contains the next contiguous
1459 non-overlapping segment of the input data. Each buffer is required to
1460 be the same length, and the sum of the lengths of the buffers is
1461 required to be exactly the data_size promised by set_params(). (This
1462 implies that the data has to be padded before being passed to
1463 encode(), unless of course it already happens to be an even multiple
1464 of required_shares in length.)
1466 Note: the requirement to break up your data into
1467 'required_shares' chunks of exactly the right length before
1468 calling encode() is surprising from point of view of a user
1469 who doesn't know how FEC works. It feels like an
1470 implementation detail that has leaked outside the abstraction
1471 barrier. Is there a use case in which the data to be encoded
1472 might already be available in pre-segmented chunks, such that
1473 it is faster or less work to make encode() take a list rather
1474 than splitting a single string?
1476 Yes, there is: suppose you are uploading a file with K=64,
1477 N=128, segsize=262,144. Then each in-share will be of size
1478 4096. If you use this .encode() API then your code could first
1479 read each successive 4096-byte chunk from the file and store
1480 each one in a Python string and store each such Python string
1481 in a Python list. Then you could call .encode(), passing that
1482 list as "inshares". The encoder would generate the other 64
1483 "secondary shares" and return to you a new list containing
1484 references to the same 64 Python strings that you passed in
1485 (as the primary shares) plus references to the new 64 Python
1488 (You could even imagine that your code could use readv() so
1489 that the operating system can arrange to get all of those
1490 bytes copied from the file into the Python list of Python
1491 strings as efficiently as possible instead of having a loop
1492 written in C or in Python to copy the next part of the file
1493 into the next string.)
1495 On the other hand if you instead use the .encode_proposal()
1496 API (above), then your code can first read in all of the
1497 262,144 bytes of the segment from the file into a Python
1498 string, then call .encode_proposal() passing the segment data
1499 as the "data" argument. The encoder would basically first
1500 split the "data" argument into a list of 64 in-shares of 4096
1501 byte each, and then do the same thing that .encode() does. So
1502 this would result in a little bit more copying of data and a
1503 little bit higher of a "maximum memory usage" during the
1504 process, although it might or might not make a practical
1505 difference for our current use cases.
1507 Note that "inshares" is a strange name for the parameter if
1508 you think of the parameter as being just for feeding in data
1509 to the codec. It makes more sense if you think of the result
1510 of this encoding as being the set of shares from inshares plus
1511 an extra set of "secondary shares" (or "check shares"). It is
1512 a surprising name! If the API is going to be surprising then
1513 the name should be surprising. If we switch to
1514 encode_proposal() above then we should also switch to an
1517 'desired_share_ids', if provided, is required to be a sequence of
1518 ints, each of which is required to be >= 0 and < max_shares. If not
1519 provided, encode() will produce 'max_shares' shares, as if
1520 'desired_share_ids' were set to range(max_shares). You might use this
1521 if you initially thought you were going to use 10 peers, started
1522 encoding, and then two of the peers dropped out: you could use
1523 desired_share_ids= to skip the work (both memory and CPU) of
1524 producing shares for the peers that are no longer available.
1526 For each call, encode() will return a Deferred that fires with two
1527 lists, one containing shares and the other containing the shareids.
1528 The get_share_size() method can be used to determine the length of
1529 the share strings returned by encode(). Each shareid is a small
1530 integer, exactly as passed into 'desired_share_ids' (or
1531 range(max_shares), if desired_share_ids was not provided).
1533 The shares and their corresponding shareids are required to be kept
1534 together during storage and retrieval. Specifically, the share data is
1535 useless by itself: the decoder needs to be told which share is which
1536 by providing it with both the shareid and the actual share data.
1538 This function will allocate an amount of memory roughly equal to::
1540 (max_shares - required_shares) * get_share_size()
1542 When combined with the memory that the caller must allocate to
1543 provide the input data, this leads to a memory footprint roughly
1544 equal to the size of the resulting encoded shares (i.e. the expansion
1545 factor times the size of the input segment).
1550 # returning a list of (shareidN,shareN) tuples instead of a pair of
1551 # lists (shareids..,shares..). Brian thought the tuples would
1552 # encourage users to keep the share and shareid together throughout
1553 # later processing, Zooko pointed out that the code to iterate
1554 # through two lists is not really more complicated than using a list
1555 # of tuples and there's also a performance improvement
1557 # having 'data_size' not required to be an integral multiple of
1558 # 'required_shares'. Doing this would require encode() to perform
1559 # padding internally, and we'd prefer to have any padding be done
1560 # explicitly by the caller. Yes, it is an abstraction leak, but
1561 # hopefully not an onerous one.
1564 class ICodecDecoder(Interface):
1565 def set_params(data_size, required_shares, max_shares):
1566 """Set the params. They have to be exactly the same ones that were
1567 used for encoding."""
1569 def get_needed_shares():
1570 """Return the number of shares needed to reconstruct the data.
1571 set_params() is required to be called before this."""
1573 def decode(some_shares, their_shareids):
1574 """Decode a partial list of shares into data.
1576 'some_shares' is required to be a sequence of buffers of sharedata, a
1577 subset of the shares returned by ICodecEncode.encode(). Each share is
1578 required to be of the same length. The i'th element of their_shareids
1579 is required to be the shareid of the i'th buffer in some_shares.
1581 This returns a Deferred that fires with a sequence of buffers. This
1582 sequence will contain all of the segments of the original data, in
1583 order. The sum of the lengths of all of the buffers will be the
1584 'data_size' value passed into the original ICodecEncode.set_params()
1585 call. To get back the single original input block of data, use
1586 ''.join(output_buffers), or you may wish to simply write them in
1587 order to an output file.
1589 Note that some of the elements in the result sequence may be
1590 references to the elements of the some_shares input sequence. In
1591 particular, this means that if those share objects are mutable (e.g.
1592 arrays) and if they are changed, then both the input (the
1593 'some_shares' parameter) and the output (the value given when the
1594 deferred is triggered) will change.
1596 The length of 'some_shares' is required to be exactly the value of
1597 'required_shares' passed into the original ICodecEncode.set_params()
1602 class IEncoder(Interface):
1603 """I take an object that provides IEncryptedUploadable, which provides
1604 encrypted data, and a list of shareholders. I then encode, hash, and
1605 deliver shares to those shareholders. I will compute all the necessary
1606 Merkle hash trees that are necessary to validate the crypttext that
1607 eventually comes back from the shareholders. I provide the URI Extension
1608 Block Hash, and the encoding parameters, both of which must be included
1611 I do not choose shareholders, that is left to the IUploader. I must be
1612 given a dict of RemoteReferences to storage buckets that are ready and
1613 willing to receive data.
1617 """Specify the number of bytes that will be encoded. This must be
1618 peformed before get_serialized_params() can be called.
1621 def set_encrypted_uploadable(u):
1622 """Provide a source of encrypted upload data. 'u' must implement
1623 IEncryptedUploadable.
1625 When this is called, the IEncryptedUploadable will be queried for its
1626 length and the storage_index that should be used.
1628 This returns a Deferred that fires with this Encoder instance.
1630 This must be performed before start() can be called.
1633 def get_param(name):
1634 """Return an encoding parameter, by name.
1636 'storage_index': return a string with the (16-byte truncated SHA-256
1637 hash) storage index to which these shares should be
1640 'share_counts': return a tuple describing how many shares are used:
1641 (needed_shares, servers_of_happiness, total_shares)
1643 'num_segments': return an int with the number of segments that
1646 'segment_size': return an int with the size of each segment.
1648 'block_size': return the size of the individual blocks that will
1649 be delivered to a shareholder's put_block() method. By
1650 knowing this, the shareholder will be able to keep all
1651 blocks in a single file and still provide random access
1652 when reading them. # TODO: can we avoid exposing this?
1654 'share_size': an int with the size of the data that will be stored
1655 on each shareholder. This is aggregate amount of data
1656 that will be sent to the shareholder, summed over all
1657 the put_block() calls I will ever make. It is useful to
1658 determine this size before asking potential
1659 shareholders whether they will grant a lease or not,
1660 since their answers will depend upon how much space we
1661 need. TODO: this might also include some amount of
1662 overhead, like the size of all the hashes. We need to
1663 decide whether this is useful or not.
1665 'serialized_params': a string with a concise description of the
1666 codec name and its parameters. This may be passed
1667 into the IUploadable to let it make sure that
1668 the same file encoded with different parameters
1669 will result in different storage indexes.
1671 Once this is called, set_size() and set_params() may not be called.
1674 def set_shareholders(shareholders, servermap):
1675 """Tell the encoder where to put the encoded shares. 'shareholders'
1676 must be a dictionary that maps share number (an integer ranging from
1677 0 to n-1) to an instance that provides IStorageBucketWriter.
1678 'servermap' is a dictionary that maps share number (as defined above)
1679 to a set of peerids. This must be performed before start() can be
1683 """Begin the encode/upload process. This involves reading encrypted
1684 data from the IEncryptedUploadable, encoding it, uploading the shares
1685 to the shareholders, then sending the hash trees.
1687 set_encrypted_uploadable() and set_shareholders() must be called
1688 before this can be invoked.
1690 This returns a Deferred that fires with a verify cap when the upload
1691 process is complete. The verifycap, plus the encryption key, is
1692 sufficient to construct the read cap.
1696 class IDecoder(Interface):
1697 """I take a list of shareholders and some setup information, then
1698 download, validate, decode, and decrypt data from them, writing the
1699 results to an output file.
1701 I do not locate the shareholders, that is left to the IDownloader. I must
1702 be given a dict of RemoteReferences to storage buckets that are ready to
1707 """I take a file-like object (providing write and close) to which all
1708 the plaintext data will be written.
1710 TODO: producer/consumer . Maybe write() should return a Deferred that
1711 indicates when it will accept more data? But probably having the
1712 IDecoder be a producer is easier to glue to IConsumer pieces.
1715 def set_shareholders(shareholders):
1716 """I take a dictionary that maps share identifiers (small integers)
1717 to RemoteReferences that provide RIBucketReader. This must be called
1721 """I start the download. This process involves retrieving data and
1722 hash chains from the shareholders, using the hashes to validate the
1723 data, decoding the shares into segments, decrypting the segments,
1724 then writing the resulting plaintext to the output file.
1726 I return a Deferred that will fire (with self) when the download is
1731 class IDownloadTarget(Interface):
1732 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1733 # will register itself as a producer. This allows the target to invoke
1734 # downloader.pauseProducing, resumeProducing, and stopProducing.
1736 """Called before any calls to write() or close(). If an error
1737 occurs before any data is available, fail() may be called without
1738 a previous call to open().
1740 'size' is the length of the file being downloaded, in bytes."""
1743 """Output some data to the target."""
1746 """Inform the target that there is no more data to be written."""
1749 """fail() is called to indicate that the download has failed. 'why'
1750 is a Failure object indicating what went wrong. No further methods
1751 will be invoked on the IDownloadTarget after fail()."""
1753 def register_canceller(cb):
1754 """The CiphertextDownloader uses this to register a no-argument function
1755 that the target can call to cancel the download. Once this canceller
1756 is invoked, no further calls to write() or close() will be made."""
1759 """When the CiphertextDownloader is done, this finish() function will be
1760 called. Whatever it returns will be returned to the invoker of
1761 Downloader.download.
1765 class IDownloader(Interface):
1766 def download(uri, target):
1767 """Perform a CHK download, sending the data to the given target.
1768 'target' must provide IDownloadTarget.
1770 Returns a Deferred that fires (with the results of target.finish)
1771 when the download is finished, or errbacks if something went wrong."""
1774 class IEncryptedUploadable(Interface):
1775 def set_upload_status(upload_status):
1776 """Provide an IUploadStatus object that should be filled with status
1777 information. The IEncryptedUploadable is responsible for setting
1778 key-determination progress ('chk'), size, storage_index, and
1779 ciphertext-fetch progress. It may delegate some of this
1780 responsibility to others, in particular to the IUploadable."""
1783 """This behaves just like IUploadable.get_size()."""
1785 def get_all_encoding_parameters():
1786 """Return a Deferred that fires with a tuple of
1787 (k,happy,n,segment_size). The segment_size will be used as-is, and
1788 must match the following constraints: it must be a multiple of k, and
1789 it shouldn't be unreasonably larger than the file size (if
1790 segment_size is larger than filesize, the difference must be stored
1793 This usually passes through to the IUploadable method of the same
1796 The encoder strictly obeys the values returned by this method. To
1797 make an upload use non-default encoding parameters, you must arrange
1798 to control the values that this method returns.
1801 def get_storage_index():
1802 """Return a Deferred that fires with a 16-byte storage index.
1805 def read_encrypted(length, hash_only):
1806 """This behaves just like IUploadable.read(), but returns crypttext
1807 instead of plaintext. If hash_only is True, then this discards the
1808 data (and returns an empty list); this improves efficiency when
1809 resuming an interrupted upload (where we need to compute the
1810 plaintext hashes, but don't need the redundant encrypted data)."""
1813 """Just like IUploadable.close()."""
1816 class IUploadable(Interface):
1817 def set_upload_status(upload_status):
1818 """Provide an IUploadStatus object that should be filled with status
1819 information. The IUploadable is responsible for setting
1820 key-determination progress ('chk')."""
1822 def set_default_encoding_parameters(params):
1823 """Set the default encoding parameters, which must be a dict mapping
1824 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1825 'max_segment_size'. These might have an influence on the final
1826 encoding parameters returned by get_all_encoding_parameters(), if the
1827 Uploadable doesn't have more specific preferences.
1829 This call is optional: if it is not used, the Uploadable will use
1830 some built-in defaults. If used, this method must be called before
1831 any other IUploadable methods to have any effect.
1835 """Return a Deferred that will fire with the length of the data to be
1836 uploaded, in bytes. This will be called before the data is actually
1837 used, to compute encoding parameters.
1840 def get_all_encoding_parameters():
1841 """Return a Deferred that fires with a tuple of
1842 (k,happy,n,segment_size). The segment_size will be used as-is, and
1843 must match the following constraints: it must be a multiple of k, and
1844 it shouldn't be unreasonably larger than the file size (if
1845 segment_size is larger than filesize, the difference must be stored
1848 The relative values of k and n allow some IUploadables to request
1849 better redundancy than others (in exchange for consuming more space
1852 Larger values of segment_size reduce hash overhead, while smaller
1853 values reduce memory footprint and cause data to be delivered in
1854 smaller pieces (which may provide a smoother and more predictable
1855 download experience).
1857 The encoder strictly obeys the values returned by this method. To
1858 make an upload use non-default encoding parameters, you must arrange
1859 to control the values that this method returns. One way to influence
1860 them may be to call set_encoding_parameters() before calling
1861 get_all_encoding_parameters().
1864 def get_encryption_key():
1865 """Return a Deferred that fires with a 16-byte AES key. This key will
1866 be used to encrypt the data. The key will also be hashed to derive
1869 Uploadables that want to achieve convergence should hash their file
1870 contents and the serialized_encoding_parameters to form the key
1871 (which of course requires a full pass over the data). Uploadables can
1872 use the upload.ConvergentUploadMixin class to achieve this
1875 Uploadables that do not care about convergence (or do not wish to
1876 make multiple passes over the data) can simply return a
1877 strongly-random 16 byte string.
1879 get_encryption_key() may be called multiple times: the IUploadable is
1880 required to return the same value each time.
1884 """Return a Deferred that fires with a list of strings (perhaps with
1885 only a single element) that, when concatenated together, contain the
1886 next 'length' bytes of data. If EOF is near, this may provide fewer
1887 than 'length' bytes. The total number of bytes provided by read()
1888 before it signals EOF must equal the size provided by get_size().
1890 If the data must be acquired through multiple internal read
1891 operations, returning a list instead of a single string may help to
1892 reduce string copies. However, the length of the concatenated strings
1893 must equal the amount of data requested, unless EOF is encountered.
1894 Long reads, or short reads without EOF, are not allowed. read()
1895 should return the same amount of data as a local disk file read, just
1896 in a different shape and asynchronously.
1898 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1899 so a 10kB file means length=3kB, 100kB file means length=30kB,
1900 and >=1MB file means length=300kB.
1902 This method provides for a single full pass through the data. Later
1903 use cases may desire multiple passes or access to only parts of the
1904 data (such as a mutable file making small edits-in-place). This API
1905 will be expanded once those use cases are better understood.
1909 """The upload is finished, and whatever filehandle was in use may be
1913 class IMutableUploadable(Interface):
1915 I represent content that is due to be uploaded to a mutable filecap.
1917 # This is somewhat simpler than the IUploadable interface above
1918 # because mutable files do not need to be concerned with possibly
1919 # generating a CHK, nor with per-file keys. It is a subset of the
1920 # methods in IUploadable, though, so we could just as well implement
1921 # the mutable uploadables as IUploadables that don't happen to use
1922 # those methods (with the understanding that the unused methods will
1923 # never be called on such objects)
1926 Returns a Deferred that fires with the size of the content held
1932 Returns a list of strings that, when concatenated, are the next
1933 length bytes of the file, or fewer if there are fewer bytes
1934 between the current location and the end of the file.
1939 The process that used the Uploadable is finished using it, so
1940 the uploadable may be closed.
1944 class IUploadResults(Interface):
1945 """I am returned by immutable upload() methods and contain the results of
1948 Note that some of my methods return empty values (0 or an empty dict)
1949 when called for non-distributed LIT files."""
1951 def get_file_size():
1952 """Return the file size, in bytes."""
1955 """Return the (string) URI of the object uploaded, a CHK readcap."""
1957 def get_ciphertext_fetched():
1958 """Return the number of bytes fetched by the helpe for this upload,
1959 or 0 if the helper did not need to fetch any bytes (or if there was
1962 def get_preexisting_shares():
1963 """Return the number of shares that were already present in the grid."""
1965 def get_pushed_shares():
1966 """Return the number of shares that were uploaded."""
1969 """Return a dict mapping share identifier to set of IServer
1970 instances. This indicates which servers were given which shares. For
1971 immutable files, the shareid is an integer (the share number, from 0
1972 to N-1). For mutable files, it is a string of the form
1973 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1974 the share number."""
1976 def get_servermap():
1977 """Return dict mapping IServer instance to a set of share numbers."""
1980 """Return dict of timing information, mapping name to seconds. All
1982 total : total upload time, start to finish
1983 storage_index : time to compute the storage index
1984 peer_selection : time to decide which peers will be used
1985 contacting_helper : initial helper query to upload/no-upload decision
1986 helper_total : initial helper query to helper finished pushing
1987 cumulative_fetch : helper waiting for ciphertext requests
1988 total_fetch : helper start to last ciphertext response
1989 cumulative_encoding : just time spent in zfec
1990 cumulative_sending : just time spent waiting for storage servers
1991 hashes_and_close : last segment push to shareholder close
1992 total_encode_and_push : first encode to shareholder close
1995 def get_uri_extension_data():
1996 """Return the dict of UEB data created for this file."""
1998 def get_verifycapstr():
1999 """Return the (string) verify-cap URI for the uploaded object."""
2002 class IDownloadResults(Interface):
2003 """I am created internally by download() methods. I contain a number of
2004 public attributes that contain details about the download process.::
2006 .file_size : the size of the file, in bytes
2007 .servers_used : set of server peerids that were used during download
2008 .server_problems : dict mapping server peerid to a problem string. Only
2009 servers that had problems (bad hashes, disconnects)
2011 .servermap : dict mapping server peerid to a set of share numbers. Only
2012 servers that had any shares are listed here.
2013 .timings : dict of timing information, mapping name to seconds (float)
2014 peer_selection : time to ask servers about shares
2015 servers_peer_selection : dict of peerid to DYHB-query time
2016 uri_extension : time to fetch a copy of the URI extension block
2017 hashtrees : time to fetch the hash trees
2018 segments : time to fetch, decode, and deliver segments
2019 cumulative_fetch : time spent waiting for storage servers
2020 cumulative_decode : just time spent in zfec
2021 cumulative_decrypt : just time spent in decryption
2022 total : total download time, start to finish
2023 fetch_per_server : dict of server to list of per-segment fetch times
2027 class IUploader(Interface):
2028 def upload(uploadable):
2029 """Upload the file. 'uploadable' must impement IUploadable. This
2030 returns a Deferred that fires with an IUploadResults instance, from
2031 which the URI of the file can be obtained as results.uri ."""
2034 class ICheckable(Interface):
2035 def check(monitor, verify=False, add_lease=False):
2036 """Check up on my health, optionally repairing any problems.
2038 This returns a Deferred that fires with an instance that provides
2039 ICheckResults, or None if the object is non-distributed (i.e. LIT
2042 The monitor will be checked periodically to see if the operation has
2043 been cancelled. If so, no new queries will be sent, and the Deferred
2044 will fire (with a OperationCancelledError) immediately.
2046 Filenodes and dirnodes (which provide IFilesystemNode) are also
2047 checkable. Instances that represent verifier-caps will be checkable
2048 but not downloadable. Some objects (like LIT files) do not actually
2049 live in the grid, and their checkers return None (non-distributed
2050 files are always healthy).
2052 If verify=False, a relatively lightweight check will be performed: I
2053 will ask all servers if they have a share for me, and I will believe
2054 whatever they say. If there are at least N distinct shares on the
2055 grid, my results will indicate r.is_healthy()==True. This requires a
2056 roundtrip to each server, but does not transfer very much data, so
2057 the network bandwidth is fairly low.
2059 If verify=True, a more resource-intensive check will be performed:
2060 every share will be downloaded, and the hashes will be validated on
2061 every bit. I will ignore any shares that failed their hash checks. If
2062 there are at least N distinct valid shares on the grid, my results
2063 will indicate r.is_healthy()==True. This requires N/k times as much
2064 download bandwidth (and server disk IO) as a regular download. If a
2065 storage server is holding a corrupt share, or is experiencing memory
2066 failures during retrieval, or is malicious or buggy, then
2067 verification will detect the problem, but checking will not.
2069 If add_lease=True, I will ensure that an up-to-date lease is present
2070 on each share. The lease secrets will be derived from by node secret
2071 (in BASEDIR/private/secret), so either I will add a new lease to the
2072 share, or I will merely renew the lease that I already had. In a
2073 future version of the storage-server protocol (once Accounting has
2074 been implemented), there may be additional options here to define the
2075 kind of lease that is obtained (which account number to claim, etc).
2077 TODO: any problems seen during checking will be reported to the
2078 health-manager.furl, a centralized object that is responsible for
2079 figuring out why files are unhealthy so corrective action can be
2083 def check_and_repair(monitor, verify=False, add_lease=False):
2084 """Like check(), but if the file/directory is not healthy, attempt to
2087 Any non-healthy result will cause an immediate repair operation, to
2088 generate and upload new shares. After repair, the file will be as
2089 healthy as we can make it. Details about what sort of repair is done
2090 will be put in the check-and-repair results. The Deferred will not
2091 fire until the repair is complete.
2093 This returns a Deferred that fires with an instance of
2094 ICheckAndRepairResults."""
2097 class IDeepCheckable(Interface):
2098 def start_deep_check(verify=False, add_lease=False):
2099 """Check upon the health of me and everything I can reach.
2101 This is a recursive form of check(), useable only on dirnodes.
2103 I return a Monitor, with results that are an IDeepCheckResults
2106 TODO: If any of the directories I traverse are unrecoverable, the
2107 Monitor will report failure. If any of the files I check upon are
2108 unrecoverable, those problems will be reported in the
2109 IDeepCheckResults as usual, and the Monitor will not report a
2113 def start_deep_check_and_repair(verify=False, add_lease=False):
2114 """Check upon the health of me and everything I can reach. Repair
2115 anything that isn't healthy.
2117 This is a recursive form of check_and_repair(), useable only on
2120 I return a Monitor, with results that are an
2121 IDeepCheckAndRepairResults object.
2123 TODO: If any of the directories I traverse are unrecoverable, the
2124 Monitor will report failure. If any of the files I check upon are
2125 unrecoverable, those problems will be reported in the
2126 IDeepCheckResults as usual, and the Monitor will not report a
2131 class ICheckResults(Interface):
2132 """I contain the detailed results of a check/verify operation.
2135 def get_storage_index():
2136 """Return a string with the (binary) storage index."""
2138 def get_storage_index_string():
2139 """Return a string with the (printable) abbreviated storage index."""
2142 """Return the (string) URI of the object that was checked."""
2145 """Return a boolean, True if the file/dir is fully healthy, False if
2146 it is damaged in any way. Non-distributed LIT files always return
2149 def is_recoverable():
2150 """Return a boolean, True if the file/dir can be recovered, False if
2151 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2152 files always return True."""
2154 def needs_rebalancing():
2155 """Return a boolean, True if the file/dirs reliability could be
2156 improved by moving shares to new servers. Non-distributed LIT files
2157 always return False."""
2159 # the following methods all return None for non-distributed LIT files
2161 def get_encoding_needed():
2162 """Return 'k', the number of shares required for recovery"""
2164 def get_encoding_expected():
2165 """Return 'N', the number of total shares generated"""
2167 def get_share_counter_good():
2168 """Return the number of distinct good shares that were found. For
2169 mutable files, this counts shares for the 'best' version."""
2171 def get_share_counter_wrong():
2172 """For mutable files, return the number of shares for versions other
2173 than the 'best' one (which is defined as being the recoverable
2174 version with the highest sequence number, then the highest roothash).
2175 These are either leftover shares from an older version (perhaps on a
2176 server that was offline when an update occurred), shares from an
2177 unrecoverable newer version, or shares from an alternate current
2178 version that results from an uncoordinated write collision. For a
2179 healthy file, this will equal 0. For immutable files, this will
2182 def get_corrupt_shares():
2183 """Return a list of 'share locators', one for each share that was
2184 found to be corrupt (integrity failure). Each share locator is a list
2185 of (IServer, storage_index, sharenum)."""
2187 def get_incompatible_shares():
2188 """Return a list of 'share locators', one for each share that was
2189 found to be of an unknown format. Each share locator is a list of
2190 (IServer, storage_index, sharenum)."""
2192 def get_servers_responding():
2193 """Return a list of IServer objects, one for each server that
2194 responded to the share query (even if they said they didn't have
2195 shares, and even if they said they did have shares but then didn't
2196 send them when asked, or dropped the connection, or returned a
2197 Failure, and even if they said they did have shares and sent
2198 incorrect ones when asked)"""
2200 def get_host_counter_good_shares():
2201 """Return the number of distinct storage servers with good shares. If
2202 this number is less than get_share_counters()[good], then some shares
2203 are doubled up, increasing the correlation of failures. This
2204 indicates that one or more shares should be moved to an otherwise
2205 unused server, if one is available.
2208 def get_version_counter_recoverable():
2209 """Return the number of recoverable versions of the file. For a
2210 healthy file, this will equal 1."""
2212 def get_version_counter_unrecoverable():
2213 """Return the number of unrecoverable versions of the file. For a
2214 healthy file, this will be 0."""
2217 """Return a dict mapping share identifier to list of IServer objects.
2218 This indicates which servers are holding which shares. For immutable
2219 files, the shareid is an integer (the share number, from 0 to N-1).
2220 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2221 containing the sequence number, the roothash, and the share number."""
2224 """Return a string with a brief (one-line) summary of the results."""
2227 """Return a list of strings with more detailed results."""
2230 class ICheckAndRepairResults(Interface):
2231 """I contain the detailed results of a check/verify/repair operation.
2233 The IFilesystemNode.check()/verify()/repair() methods all return
2234 instances that provide ICheckAndRepairResults.
2237 def get_storage_index():
2238 """Return a string with the (binary) storage index."""
2240 def get_storage_index_string():
2241 """Return a string with the (printable) abbreviated storage index."""
2243 def get_repair_attempted():
2244 """Return a boolean, True if a repair was attempted. We might not
2245 attempt to repair the file because it was healthy, or healthy enough
2246 (i.e. some shares were missing but not enough to exceed some
2247 threshold), or because we don't know how to repair this object."""
2249 def get_repair_successful():
2250 """Return a boolean, True if repair was attempted and the file/dir
2251 was fully healthy afterwards. False if no repair was attempted or if
2252 a repair attempt failed."""
2254 def get_pre_repair_results():
2255 """Return an ICheckResults instance that describes the state of the
2256 file/dir before any repair was attempted."""
2258 def get_post_repair_results():
2259 """Return an ICheckResults instance that describes the state of the
2260 file/dir after any repair was attempted. If no repair was attempted,
2261 the pre-repair and post-repair results will be identical."""
2264 class IDeepCheckResults(Interface):
2265 """I contain the results of a deep-check operation.
2267 This is returned by a call to ICheckable.deep_check().
2270 def get_root_storage_index_string():
2271 """Return the storage index (abbreviated human-readable string) of
2272 the first object checked."""
2275 """Return a dictionary with the following keys::
2277 count-objects-checked: count of how many objects were checked
2278 count-objects-healthy: how many of those objects were completely
2280 count-objects-unhealthy: how many were damaged in some way
2281 count-objects-unrecoverable: how many were unrecoverable
2282 count-corrupt-shares: how many shares were found to have
2283 corruption, summed over all objects
2287 def get_corrupt_shares():
2288 """Return a set of (IServer, storage_index, sharenum) for all shares
2289 that were found to be corrupt. storage_index is binary."""
2291 def get_all_results():
2292 """Return a dictionary mapping pathname (a tuple of strings, ready to
2293 be slash-joined) to an ICheckResults instance, one for each object
2294 that was checked."""
2296 def get_results_for_storage_index(storage_index):
2297 """Retrive the ICheckResults instance for the given (binary)
2298 storage index. Raises KeyError if there are no results for that
2302 """Return a dictionary with the same keys as
2303 IDirectoryNode.deep_stats()."""
2306 class IDeepCheckAndRepairResults(Interface):
2307 """I contain the results of a deep-check-and-repair operation.
2309 This is returned by a call to ICheckable.deep_check_and_repair().
2312 def get_root_storage_index_string():
2313 """Return the storage index (abbreviated human-readable string) of
2314 the first object checked."""
2317 """Return a dictionary with the following keys::
2319 count-objects-checked: count of how many objects were checked
2320 count-objects-healthy-pre-repair: how many of those objects were
2321 completely healthy (before any
2323 count-objects-unhealthy-pre-repair: how many were damaged in
2325 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2326 count-objects-healthy-post-repair: how many of those objects were
2327 completely healthy (after any
2329 count-objects-unhealthy-post-repair: how many were damaged in
2331 count-objects-unrecoverable-post-repair: how many were
2333 count-repairs-attempted: repairs were attempted on this many
2334 objects. The count-repairs- keys will
2335 always be provided, however unless
2336 repair=true is present, they will all
2338 count-repairs-successful: how many repairs resulted in healthy
2340 count-repairs-unsuccessful: how many repairs resulted did not
2341 results in completely healthy objects
2342 count-corrupt-shares-pre-repair: how many shares were found to
2343 have corruption, summed over all
2344 objects examined (before any
2346 count-corrupt-shares-post-repair: how many shares were found to
2347 have corruption, summed over all
2348 objects examined (after any
2353 """Return a dictionary with the same keys as
2354 IDirectoryNode.deep_stats()."""
2356 def get_corrupt_shares():
2357 """Return a set of (IServer, storage_index, sharenum) for all shares
2358 that were found to be corrupt before any repair was attempted.
2359 storage_index is binary.
2361 def get_remaining_corrupt_shares():
2362 """Return a set of (IServer, storage_index, sharenum) for all shares
2363 that were found to be corrupt after any repair was completed.
2364 storage_index is binary. These are shares that need manual inspection
2365 and probably deletion.
2367 def get_all_results():
2368 """Return a dictionary mapping pathname (a tuple of strings, ready to
2369 be slash-joined) to an ICheckAndRepairResults instance, one for each
2370 object that was checked."""
2372 def get_results_for_storage_index(storage_index):
2373 """Retrive the ICheckAndRepairResults instance for the given (binary)
2374 storage index. Raises KeyError if there are no results for that
2378 class IRepairable(Interface):
2379 def repair(check_results):
2380 """Attempt to repair the given object. Returns a Deferred that fires
2381 with a IRepairResults object.
2383 I must be called with an object that implements ICheckResults, as
2384 proof that you have actually discovered a problem with this file. I
2385 will use the data in the checker results to guide the repair process,
2386 such as which servers provided bad data and should therefore be
2387 avoided. The ICheckResults object is inside the
2388 ICheckAndRepairResults object, which is returned by the
2389 ICheckable.check() method::
2391 d = filenode.check(repair=False)
2392 def _got_results(check_and_repair_results):
2393 check_results = check_and_repair_results.get_pre_repair_results()
2394 return filenode.repair(check_results)
2395 d.addCallback(_got_results)
2400 class IRepairResults(Interface):
2401 """I contain the results of a repair operation."""
2402 def get_successful():
2403 """Returns a boolean: True if the repair made the file healthy, False
2404 if not. Repair failure generally indicates a file that has been
2405 damaged beyond repair."""
2408 class IClient(Interface):
2409 def upload(uploadable):
2410 """Upload some data into a CHK, get back the UploadResults for it.
2411 @param uploadable: something that implements IUploadable
2412 @return: a Deferred that fires with the UploadResults instance.
2413 To get the URI for this file, use results.uri .
2416 def create_mutable_file(contents=""):
2417 """Create a new mutable file (with initial) contents, get back the
2420 @param contents: (bytestring, callable, or None): this provides the
2421 initial contents of the mutable file. If 'contents' is a bytestring,
2422 it will be used as-is. If 'contents' is a callable, it will be
2423 invoked with the new MutableFileNode instance and is expected to
2424 return a bytestring with the initial contents of the file (the
2425 callable can use node.get_writekey() to decide how to encrypt the
2426 initial contents, e.g. for a brand new dirnode with initial
2427 children). contents=None is equivalent to an empty string. Using
2428 content_maker= is more efficient than creating a mutable file and
2429 setting its contents in two separate operations.
2431 @return: a Deferred that fires with an IMutableFileNode instance.
2434 def create_dirnode(initial_children={}):
2435 """Create a new unattached dirnode, possibly with initial children.
2437 @param initial_children: dict with keys that are unicode child names,
2438 and values that are (childnode, metadata) tuples.
2440 @return: a Deferred that fires with the new IDirectoryNode instance.
2443 def create_node_from_uri(uri, rouri):
2444 """Create a new IFilesystemNode instance from the uri, synchronously.
2445 @param uri: a string or IURI-providing instance, or None. This could
2446 be for a LiteralFileNode, a CHK file node, a mutable file
2447 node, or a directory node
2448 @param rouri: a string or IURI-providing instance, or None. If the
2449 main uri is None, I will use the rouri instead. If I
2450 recognize the format of the main uri, I will ignore the
2451 rouri (because it can be derived from the writecap).
2453 @return: an instance that provides IFilesystemNode (or more usefully
2454 one of its subclasses). File-specifying URIs will result in
2455 IFileNode-providing instances, like ImmutableFileNode,
2456 LiteralFileNode, or MutableFileNode. Directory-specifying
2457 URIs will result in IDirectoryNode-providing instances, like
2462 class INodeMaker(Interface):
2463 """The NodeMaker is used to create IFilesystemNode instances. It can
2464 accept a filecap/dircap string and return the node right away. It can
2465 also create new nodes (i.e. upload a file, or create a mutable file)
2466 asynchronously. Once you have one of these nodes, you can use other
2467 methods to determine whether it is a file or directory, and to download
2468 or modify its contents.
2470 The NodeMaker encapsulates all the authorities that these
2471 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2472 Tahoe process will typically have a single NodeMaker, but unit tests may
2473 create simplified/mocked forms for testing purposes.
2476 def create_from_cap(writecap, readcap=None, deep_immutable=False, name=u"<unknown name>"):
2477 """I create an IFilesystemNode from the given writecap/readcap. I can
2478 only provide nodes for existing file/directory objects: use my other
2479 methods to create new objects. I return synchronously."""
2481 def create_mutable_file(contents=None, keysize=None):
2482 """I create a new mutable file, and return a Deferred that will fire
2483 with the IMutableFileNode instance when it is ready. If contents= is
2484 provided (a bytestring), it will be used as the initial contents of
2485 the new file, otherwise the file will contain zero bytes. keysize= is
2486 for use by unit tests, to create mutable files that are smaller than
2489 def create_new_mutable_directory(initial_children={}):
2490 """I create a new mutable directory, and return a Deferred that will
2491 fire with the IDirectoryNode instance when it is ready. If
2492 initial_children= is provided (a dict mapping unicode child name to
2493 (childnode, metadata_dict) tuples), the directory will be populated
2494 with those children, otherwise it will be empty."""
2497 class IClientStatus(Interface):
2498 def list_all_uploads():
2499 """Return a list of uploader objects, one for each upload that
2500 currently has an object available (tracked with weakrefs). This is
2501 intended for debugging purposes."""
2503 def list_active_uploads():
2504 """Return a list of active IUploadStatus objects."""
2506 def list_recent_uploads():
2507 """Return a list of IUploadStatus objects for the most recently
2510 def list_all_downloads():
2511 """Return a list of downloader objects, one for each download that
2512 currently has an object available (tracked with weakrefs). This is
2513 intended for debugging purposes."""
2515 def list_active_downloads():
2516 """Return a list of active IDownloadStatus objects."""
2518 def list_recent_downloads():
2519 """Return a list of IDownloadStatus objects for the most recently
2520 started downloads."""
2523 class IUploadStatus(Interface):
2525 """Return a timestamp (float with seconds since epoch) indicating
2526 when the operation was started."""
2528 def get_storage_index():
2529 """Return a string with the (binary) storage index in use on this
2530 upload. Returns None if the storage index has not yet been
2534 """Return an integer with the number of bytes that will eventually
2535 be uploaded for this file. Returns None if the size is not yet known.
2538 """Return True if this upload is using a Helper, False if not."""
2541 """Return a string describing the current state of the upload
2545 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2546 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2547 made towards hashing the file to determine a CHK encryption key: if
2548 non-convergent encryption is in use, this will be trivial, otherwise
2549 the whole file must be hashed. 'ciphertext' describes how much of the
2550 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2551 uploads. 'encode_and_push' describes how much of the encode-and-push
2552 process has finished: for helper uploads this is dependent upon the
2553 helper providing progress reports. It might be reasonable to add all
2554 three numbers and report the sum to the user."""
2557 """Return True if the upload is currently active, False if not."""
2560 """Return an instance of UploadResults (which contains timing and
2561 sharemap information). Might return None if the upload is not yet
2565 """Each upload status gets a unique number: this method returns that
2566 number. This provides a handle to this particular upload, so a web
2567 page can generate a suitable hyperlink."""
2570 class IDownloadStatus(Interface):
2572 """Return a timestamp (float with seconds since epoch) indicating
2573 when the operation was started."""
2575 def get_storage_index():
2576 """Return a string with the (binary) storage index in use on this
2577 download. This may be None if there is no storage index (i.e. LIT
2581 """Return an integer with the number of bytes that will eventually be
2582 retrieved for this file. Returns None if the size is not yet known.
2586 """Return True if this download is using a Helper, False if not."""
2589 """Return a string describing the current state of the download
2593 """Returns a float (from 0.0 to 1.0) describing the amount of the
2594 download that has completed. This value will remain at 0.0 until the
2595 first byte of plaintext is pushed to the download target."""
2598 """Return True if the download is currently active, False if not."""
2601 """Each download status gets a unique number: this method returns
2602 that number. This provides a handle to this particular download, so a
2603 web page can generate a suitable hyperlink."""
2606 class IServermapUpdaterStatus(Interface):
2609 class IPublishStatus(Interface):
2612 class IRetrieveStatus(Interface):
2616 class NotCapableError(Exception):
2617 """You have tried to write to a read-only node."""
2619 class BadWriteEnablerError(Exception):
2622 class RIControlClient(RemoteInterface):
2624 def wait_for_client_connections(num_clients=int):
2625 """Do not return until we have connections to at least NUM_CLIENTS
2629 def upload_from_file_to_uri(filename=str,
2630 convergence=ChoiceOf(None,
2631 StringConstraint(2**20))):
2632 """Upload a file to the grid. This accepts a filename (which must be
2633 absolute) that points to a file on the node's local disk. The node will
2634 read the contents of this file, upload it to the grid, then return the
2635 URI at which it was uploaded. If convergence is None then a random
2636 encryption key will be used, else the plaintext will be hashed, then
2637 that hash will be mixed together with the "convergence" string to form
2642 def download_from_uri_to_file(uri=URI, filename=str):
2643 """Download a file from the grid, placing it on the node's local disk
2644 at the given filename (which must be absolute[?]). Returns the
2645 absolute filename where the file was written."""
2650 def get_memory_usage():
2651 """Return a dict describes the amount of memory currently in use. The
2652 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2653 measuring memory consupmtion in bytes."""
2654 return DictOf(str, int)
2656 def speed_test(count=int, size=int, mutable=Any()):
2657 """Write 'count' tempfiles to disk, all of the given size. Measure
2658 how long (in seconds) it takes to upload them all to the servers.
2659 Then measure how long it takes to download all of them. If 'mutable'
2660 is 'create', time creation of mutable files. If 'mutable' is
2661 'upload', then time access to the same mutable file instead of
2664 Returns a tuple of (upload_time, download_time).
2666 return (float, float)
2668 def measure_peer_response_time():
2669 """Send a short message to each connected peer, and measure the time
2670 it takes for them to respond to it. This is a rough measure of the
2671 application-level round trip time.
2673 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2676 return DictOf(str, float)
2679 UploadResults = Any() #DictOf(str, str)
2682 class RIEncryptedUploadable(RemoteInterface):
2683 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2688 def get_all_encoding_parameters():
2689 return (int, int, int, long)
2691 def read_encrypted(offset=Offset, length=ReadSize):
2698 class RICHKUploadHelper(RemoteInterface):
2699 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2703 Return a dictionary of version information.
2705 return DictOf(str, Any())
2707 def upload(reader=RIEncryptedUploadable):
2708 return UploadResults
2711 class RIHelper(RemoteInterface):
2712 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2716 Return a dictionary of version information.
2718 return DictOf(str, Any())
2720 def upload_chk(si=StorageIndex):
2721 """See if a file with a given storage index needs uploading. The
2722 helper will ask the appropriate storage servers to see if the file
2723 has already been uploaded. If so, the helper will return a set of
2724 'upload results' that includes whatever hashes are needed to build
2725 the read-cap, and perhaps a truncated sharemap.
2727 If the file has not yet been uploaded (or if it was only partially
2728 uploaded), the helper will return an empty upload-results dictionary
2729 and also an RICHKUploadHelper object that will take care of the
2730 upload process. The client should call upload() on this object and
2731 pass it a reference to an RIEncryptedUploadable object that will
2732 provide ciphertext. When the upload is finished, the upload() method
2733 will finish and return the upload results.
2735 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2738 class RIStatsProvider(RemoteInterface):
2739 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2741 Provides access to statistics and monitoring information.
2746 returns a dictionary containing 'counters' and 'stats', each a
2747 dictionary with string counter/stat name keys, and numeric or None values.
2748 counters are monotonically increasing measures of work done, and
2749 stats are instantaneous measures (potentially time averaged
2752 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2755 class RIStatsGatherer(RemoteInterface):
2756 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2758 Provides a monitoring service for centralised collection of stats
2761 def provide(provider=RIStatsProvider, nickname=str):
2763 @param provider: a stats collector instance that should be polled
2764 periodically by the gatherer to collect stats.
2765 @param nickname: a name useful to identify the provided client
2770 class IStatsProducer(Interface):
2773 returns a dictionary, with str keys representing the names of stats
2774 to be monitored, and numeric values.
2777 class RIKeyGenerator(RemoteInterface):
2778 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2780 Provides a service offering to make RSA key pairs.
2783 def get_rsa_key_pair(key_size=int):
2785 @param key_size: the size of the signature key.
2786 @return: tuple(verifying_key, signing_key)
2788 return TupleOf(str, str)
2791 class FileTooLargeError(Exception):
2795 class IValidatedThingProxy(Interface):
2797 """ Acquire a thing and validate it. Return a deferred that is
2798 eventually fired with self if the thing is valid or errbacked if it
2799 can't be acquired or validated."""
2802 class InsufficientVersionError(Exception):
2803 def __init__(self, needed, got):
2804 self.needed = needed
2808 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2811 class EmptyPathnameComponentError(Exception):
2812 """The webapi disallows empty pathname components."""