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.
51 class RIBucketReader(RemoteInterface):
52 def read(offset=Offset, length=ReadSize):
55 def advise_corrupt_share(reason=str):
56 """Clients who discover hash failures in shares that they have
57 downloaded from me will use this method to inform me about the
58 failures. I will record their concern so that my operator can
59 manually inspect the shares in question. I return None.
61 This is a wrapper around RIStorageServer.advise_corrupt_share(),
62 which is tied to a specific share, and therefore does not need the
63 extra share-identifying arguments. Please see that method for full
67 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
68 # elements are (offset, length, operator, specimen)
69 # operator is one of "lt, le, eq, ne, ge, gt"
70 # nop always passes and is used to fetch data while writing.
71 # you should use length==len(specimen) for everything except nop
72 DataVector = ListOf(TupleOf(Offset, ShareData))
73 # (offset, data). This limits us to 30 writes of 1MiB each per call
74 TestAndWriteVectorsForShares = DictOf(int,
77 ChoiceOf(None, Offset), # new_length
79 ReadVector = ListOf(TupleOf(Offset, ReadSize))
80 ReadData = ListOf(ShareData)
81 # returns data[offset:offset+length] for each element of TestVector
83 class RIStorageServer(RemoteInterface):
84 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
88 Return a dictionary of version information.
90 return DictOf(str, Any())
92 def allocate_buckets(storage_index=StorageIndex,
93 renew_secret=LeaseRenewSecret,
94 cancel_secret=LeaseCancelSecret,
95 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
96 allocated_size=Offset, canary=Referenceable):
98 @param storage_index: the index of the bucket to be created or
100 @param sharenums: these are the share numbers (probably between 0 and
101 99) that the sender is proposing to store on this
103 @param renew_secret: This is the secret used to protect bucket refresh
104 This secret is generated by the client and
105 stored for later comparison by the server. Each
106 server is given a different secret.
107 @param cancel_secret: Like renew_secret, but protects bucket decref.
108 @param canary: If the canary is lost before close(), the bucket is
110 @return: tuple of (alreadygot, allocated), where alreadygot is what we
111 already have and allocated is what we hereby agree to accept.
112 New leases are added for shares in both lists.
114 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
115 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
117 def add_lease(storage_index=StorageIndex,
118 renew_secret=LeaseRenewSecret,
119 cancel_secret=LeaseCancelSecret):
121 Add a new lease on the given bucket. If the renew_secret matches an
122 existing lease, that lease will be renewed instead. If there is no
123 bucket for the given storage_index, return silently. (note that in
124 tahoe-1.3.0 and earlier, IndexError was raised if there was no
127 return Any() # returns None now, but future versions might change
129 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
131 Renew the lease on a given bucket, resetting the timer to 31 days.
132 Some networks will use this, some will not. If there is no bucket for
133 the given storage_index, IndexError will be raised.
135 For mutable shares, if the given renew_secret does not match an
136 existing lease, IndexError will be raised with a note listing the
137 server-nodeids on the existing leases, so leases on migrated shares
138 can be renewed or cancelled. For immutable shares, IndexError
139 (without the note) will be raised.
143 def get_buckets(storage_index=StorageIndex):
144 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
148 def slot_readv(storage_index=StorageIndex,
149 shares=ListOf(int), readv=ReadVector):
150 """Read a vector from the numbered shares associated with the given
151 storage index. An empty shares list means to return data from all
152 known shares. Returns a dictionary with one key per share."""
153 return DictOf(int, ReadData) # shnum -> results
155 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
156 secrets=TupleOf(WriteEnablerSecret,
159 tw_vectors=TestAndWriteVectorsForShares,
163 General-purpose test-read-and-set operation for mutable slots:
164 (1) For submitted shnums, compare the test vectors against extant
165 shares, or against an empty share for shnums that do not exist.
166 (2) Use the read vectors to extract "old data" from extant shares.
167 (3) If all tests in (1) passed, then apply the write vectors
168 (possibly creating new shares).
169 (4) Return whether the tests passed, and the "old data", which does
170 not include any modifications made by the writes.
172 The operation does not interleave with other operations on the same
175 This method is, um, large. The goal is to allow clients to update all
176 the shares associated with a mutable file in a single round trip.
178 @param storage_index: the index of the bucket to be created or
180 @param write_enabler: a secret that is stored along with the slot.
181 Writes are accepted from any caller who can
182 present the matching secret. A different secret
183 should be used for each slot*server pair.
184 @param renew_secret: This is the secret used to protect bucket refresh
185 This secret is generated by the client and
186 stored for later comparison by the server. Each
187 server is given a different secret.
188 @param cancel_secret: Like renew_secret, but protects bucket decref.
190 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
191 cancel_secret). The first is required to perform any write. The
192 latter two are used when allocating new shares. To simply acquire a
193 new lease on existing shares, use an empty testv and an empty writev.
195 Each share can have a separate test vector (i.e. a list of
196 comparisons to perform). If all vectors for all shares pass, then all
197 writes for all shares are recorded. Each comparison is a 4-tuple of
198 (offset, length, operator, specimen), which effectively does a
199 bool( (read(offset, length)) OPERATOR specimen ) and only performs
200 the write if all these evaluate to True. Basic test-and-set uses 'eq'.
201 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
202 Write-if-same-or-newer uses 'le'.
204 Reads from the end of the container are truncated, and missing shares
205 behave like empty ones, so to assert that a share doesn't exist (for
206 use when creating a new share), use (0, 1, 'eq', '').
208 The write vector will be applied to the given share, expanding it if
209 necessary. A write vector applied to a share number that did not
210 exist previously will cause that share to be created. Write vectors
211 must not overlap (if they do, this will either cause an error or
212 apply them in an unspecified order). Duplicate write vectors, with
213 the same offset and data, are currently tolerated but are not
216 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
217 vector whose offset is beyond the end of the current data, the space
218 between the end of the current data and the beginning of the write
219 vector will be filled with zero bytes. In earlier versions the
220 contents of this space was unspecified (and might end up containing
221 secrets). Storage servers with the new zero-filling behavior will
222 advertise a true value for the 'fills-holes-with-zero-bytes' key
223 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
226 Each write vector is accompanied by a 'new_length' argument, which
227 can be used to truncate the data. If new_length is not None and it is
228 less than the current size of the data (after applying all write
229 vectors), then the data will be truncated to new_length. If
230 new_length==0, the share will be deleted.
232 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
233 enlarge the file by sending a number larger than the size of the data
234 after applying all write vectors. That behavior was not used, and as
235 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
238 If a storage client knows that the server supports zero-filling, for
239 example from the 'fills-holes-with-zero-bytes' key in its version
240 information, it can extend the file efficiently by writing a single
241 zero byte just before the new end-of-file. Otherwise it must
242 explicitly write zeroes to all bytes between the old and new
243 end-of-file. In any case it should avoid sending new_length larger
244 than the size of the data after applying all write vectors.
246 The read vector is used to extract data from all known shares,
247 *before* any writes have been applied. The same read vector is used
248 for all shares. This captures the state that was tested by the test
249 vector, for extant shares.
251 This method returns two values: a boolean and a dict. The boolean is
252 True if the write vectors were applied, False if not. The dict is
253 keyed by share number, and each value contains a list of strings, one
254 for each element of the read vector.
256 If the write_enabler is wrong, this will raise BadWriteEnablerError.
257 To enable share migration (using update_write_enabler), the exception
258 will have the nodeid used for the old write enabler embedded in it,
259 in the following string::
261 The write enabler was recorded by nodeid '%s'.
263 Note that the nodeid here is encoded using the same base32 encoding
264 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
266 return TupleOf(bool, DictOf(int, ReadData))
268 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
269 shnum=int, reason=str):
270 """Clients who discover hash failures in shares that they have
271 downloaded from me will use this method to inform me about the
272 failures. I will record their concern so that my operator can
273 manually inspect the shares in question. I return None.
275 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
276 (binary) storage index string, and 'shnum' is the integer share
277 number. 'reason' is a human-readable explanation of the problem,
278 probably including some expected hash values and the computed ones
279 which did not match. Corruption advisories for mutable shares should
280 include a hash of the public key (the same value that appears in the
281 mutable-file verify-cap), since the current share format does not
285 class IStorageBucketWriter(Interface):
287 Objects of this kind live on the client side.
289 def put_block(segmentnum=int, data=ShareData):
290 """@param data: For most segments, this data will be 'blocksize'
291 bytes in length. The last segment might be shorter.
292 @return: a Deferred that fires (with None) when the operation completes
295 def put_plaintext_hashes(hashes=ListOf(Hash)):
297 @return: a Deferred that fires (with None) when the operation completes
300 def put_crypttext_hashes(hashes=ListOf(Hash)):
302 @return: a Deferred that fires (with None) when the operation completes
305 def put_block_hashes(blockhashes=ListOf(Hash)):
307 @return: a Deferred that fires (with None) when the operation completes
310 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
312 @return: a Deferred that fires (with None) when the operation completes
315 def put_uri_extension(data=URIExtensionData):
316 """This block of data contains integrity-checking information (hashes
317 of plaintext, crypttext, and shares), as well as encoding parameters
318 that are necessary to recover the data. This is a serialized dict
319 mapping strings to other strings. The hash of this data is kept in
320 the URI and verified before any of the data is used. All buckets for
321 a given file contain identical copies of this data.
323 The serialization format is specified with the following pseudocode:
324 for k in sorted(dict.keys()):
325 assert re.match(r'^[a-zA-Z_\-]+$', k)
326 write(k + ':' + netstring(dict[k]))
328 @return: a Deferred that fires (with None) when the operation completes
332 """Finish writing and close the bucket. The share is not finalized
333 until this method is called: if the uploading client disconnects
334 before calling close(), the partially-written share will be
337 @return: a Deferred that fires (with None) when the operation completes
340 class IStorageBucketReader(Interface):
342 def get_block_data(blocknum=int, blocksize=int, size=int):
343 """Most blocks will be the same size. The last block might be shorter
349 def get_crypttext_hashes():
351 @return: ListOf(Hash)
354 def get_block_hashes(at_least_these=SetOf(int)):
356 @return: ListOf(Hash)
359 def get_share_hashes(at_least_these=SetOf(int)):
361 @return: ListOf(TupleOf(int, Hash))
364 def get_uri_extension():
366 @return: URIExtensionData
369 class IStorageBroker(Interface):
370 def get_servers_for_psi(peer_selection_index):
372 @return: list of IServer instances
374 def get_connected_servers():
376 @return: frozenset of connected IServer instances
378 def get_known_servers():
380 @return: frozenset of IServer instances
382 def get_all_serverids():
384 @return: frozenset of serverid strings
386 def get_nickname_for_serverid(serverid):
388 @return: unicode nickname, or None
391 # methods moved from IntroducerClient, need review
392 def get_all_connections():
393 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
394 each active connection we've established to a remote service. This is
395 mostly useful for unit tests that need to wait until a certain number
396 of connections have been made."""
398 def get_all_connectors():
399 """Return a dict that maps from (nodeid, service_name) to a
400 RemoteServiceConnector instance for all services that we are actively
401 trying to connect to. Each RemoteServiceConnector has the following
404 service_name: the type of service provided, like 'storage'
405 announcement_time: when we first heard about this service
406 last_connect_time: when we last established a connection
407 last_loss_time: when we last lost a connection
409 version: the peer's version, from the most recent connection
410 oldest_supported: the peer's oldest supported version, same
412 rref: the RemoteReference, if connected, otherwise None
413 remote_host: the IAddress, if connected, otherwise None
415 This method is intended for monitoring interfaces, such as a web page
416 which describes connecting and connected peers.
419 def get_all_peerids():
420 """Return a frozenset of all peerids to whom we have a connection (to
421 one or more services) established. Mostly useful for unit tests."""
423 def get_all_connections_for(service_name):
424 """Return a frozenset of (nodeid, service_name, rref) tuples, one
425 for each active connection that provides the given SERVICE_NAME."""
427 def get_permuted_peers(service_name, key):
428 """Returns an ordered list of (peerid, rref) tuples, selecting from
429 the connections that provide SERVICE_NAME, using a hash-based
430 permutation keyed by KEY. This randomizes the service list in a
431 repeatable way, to distribute load over many peers.
434 class IDisplayableServer(Interface):
442 class IServer(IDisplayableServer):
443 """I live in the client, and represent a single server."""
444 def start_connecting(tub, trigger_cb):
450 class IMutableSlotWriter(Interface):
452 The interface for a writer around a mutable slot on a remote server.
454 def set_checkstring(checkstring, *args):
456 Set the checkstring that I will pass to the remote server when
459 @param checkstring A packed checkstring to use.
461 Note that implementations can differ in which semantics they
462 wish to support for set_checkstring -- they can, for example,
463 build the checkstring themselves from its constituents, or
467 def get_checkstring():
469 Get the checkstring that I think currently exists on the remote
473 def put_block(data, segnum, salt):
475 Add a block and salt to the share.
478 def put_encprivkey(encprivkey):
480 Add the encrypted private key to the share.
483 def put_blockhashes(blockhashes=list):
485 Add the block hash tree to the share.
488 def put_sharehashes(sharehashes=dict):
490 Add the share hash chain to the share.
495 Return the part of the share that needs to be signed.
498 def put_signature(signature):
500 Add the signature to the share.
503 def put_verification_key(verification_key):
505 Add the verification key to the share.
508 def finish_publishing():
510 Do anything necessary to finish writing the share to a remote
511 server. I require that no further publishing needs to take place
512 after this method has been called.
516 class IURI(Interface):
517 def init_from_string(uri):
518 """Accept a string (as created by my to_string() method) and populate
519 this instance with its data. I am not normally called directly,
520 please use the module-level uri.from_string() function to convert
521 arbitrary URI strings into IURI-providing instances."""
524 """Return False if this URI be used to modify the data. Return True
525 if this URI cannot be used to modify the data."""
528 """Return True if the data can be modified by *somebody* (perhaps
529 someone who has a more powerful URI than this one)."""
531 # TODO: rename to get_read_cap()
533 """Return another IURI instance, which represents a read-only form of
534 this one. If is_readonly() is True, this returns self."""
536 def get_verify_cap():
537 """Return an instance that provides IVerifierURI, which can be used
538 to check on the availability of the file or directory, without
539 providing enough capabilities to actually read or modify the
540 contents. This may return None if the file does not need checking or
541 verification (e.g. LIT URIs).
545 """Return a string of printable ASCII characters, suitable for
546 passing into init_from_string."""
548 class IVerifierURI(Interface, IURI):
549 def init_from_string(uri):
550 """Accept a string (as created by my to_string() method) and populate
551 this instance with its data. I am not normally called directly,
552 please use the module-level uri.from_string() function to convert
553 arbitrary URI strings into IURI-providing instances."""
556 """Return a string of printable ASCII characters, suitable for
557 passing into init_from_string."""
559 class IDirnodeURI(Interface):
560 """I am a URI which represents a dirnode."""
562 class IFileURI(Interface):
563 """I am a URI which represents a filenode."""
565 """Return the length (in bytes) of the file that I represent."""
567 class IImmutableFileURI(IFileURI):
570 class IMutableFileURI(Interface):
573 class IDirectoryURI(Interface):
576 class IReadonlyDirectoryURI(Interface):
579 class CapConstraintError(Exception):
580 """A constraint on a cap was violated."""
582 class MustBeDeepImmutableError(CapConstraintError):
583 """Mutable children cannot be added to an immutable directory.
584 Also, caps obtained from an immutable directory can trigger this error
585 if they are later found to refer to a mutable object and then used."""
587 class MustBeReadonlyError(CapConstraintError):
588 """Known write caps cannot be specified in a ro_uri field. Also,
589 caps obtained from a ro_uri field can trigger this error if they
590 are later found to be write caps and then used."""
592 class MustNotBeUnknownRWError(CapConstraintError):
593 """Cannot add an unknown child cap specified in a rw_uri field."""
596 class IReadable(Interface):
597 """I represent a readable object -- either an immutable file, or a
598 specific version of a mutable file.
602 """Return True if this reference provides mutable access to the given
603 file or directory (i.e. if you can modify it), or False if not. Note
604 that even if this reference is read-only, someone else may hold a
605 read-write reference to it.
607 For an IReadable returned by get_best_readable_version(), this will
608 always return True, but for instances of subinterfaces such as
609 IMutableFileVersion, it may return False."""
612 """Return True if this file or directory is mutable (by *somebody*,
613 not necessarily you), False if it is is immutable. Note that a file
614 might be mutable overall, but your reference to it might be
615 read-only. On the other hand, all references to an immutable file
616 will be read-only; there are no read-write references to an immutable
619 def get_storage_index():
620 """Return the storage index of the file."""
623 """Return the length (in bytes) of this readable object."""
625 def download_to_data():
626 """Download all of the file contents. I return a Deferred that fires
627 with the contents as a byte string."""
629 def read(consumer, offset=0, size=None):
630 """Download a portion (possibly all) of the file's contents, making
631 them available to the given IConsumer. Return a Deferred that fires
632 (with the consumer) when the consumer is unregistered (either because
633 the last byte has been given to it, or because the consumer threw an
634 exception during write(), possibly because it no longer wants to
635 receive data). The portion downloaded will start at 'offset' and
636 contain 'size' bytes (or the remainder of the file if size==None).
638 The consumer will be used in non-streaming mode: an IPullProducer
639 will be attached to it.
641 The consumer will not receive data right away: several network trips
642 must occur first. The order of events will be::
644 consumer.registerProducer(p, streaming)
645 (if streaming == False)::
646 consumer does p.resumeProducing()
648 consumer does p.resumeProducing()
649 consumer.write(data).. (repeat until all data is written)
650 consumer.unregisterProducer()
651 deferred.callback(consumer)
653 If a download error occurs, or an exception is raised by
654 consumer.registerProducer() or consumer.write(), I will call
655 consumer.unregisterProducer() and then deliver the exception via
656 deferred.errback(). To cancel the download, the consumer should call
657 p.stopProducing(), which will result in an exception being delivered
658 via deferred.errback().
660 See src/allmydata/util/consumer.py for an example of a simple
661 download-to-memory consumer.
665 class IWriteable(Interface):
667 I define methods that callers can use to update SDMF and MDMF
668 mutable files on a Tahoe-LAFS grid.
670 # XXX: For the moment, we have only this. It is possible that we
671 # want to move overwrite() and modify() in here too.
672 def update(data, offset):
674 I write the data from my data argument to the MDMF file,
675 starting at offset. I continue writing data until my data
676 argument is exhausted, appending data to the file as necessary.
678 # assert IMutableUploadable.providedBy(data)
679 # to append data: offset=node.get_size_of_best_version()
680 # do we want to support compacting MDMF?
681 # for an MDMF file, this can be done with O(data.get_size())
682 # memory. For an SDMF file, any modification takes
683 # O(node.get_size_of_best_version()).
686 class IMutableFileVersion(IReadable):
687 """I provide access to a particular version of a mutable file. The
688 access is read/write if I was obtained from a filenode derived from
689 a write cap, or read-only if the filenode was derived from a read cap.
692 def get_sequence_number():
693 """Return the sequence number of this version."""
696 """Return the IMutableFileServerMap instance that was used to create
701 """Return this filenode's writekey, or None if the node does not have
702 write-capability. This may be used to assist with data structures
703 that need to make certain data available only to writers, such as the
704 read-write child caps in dirnodes. The recommended process is to have
705 reader-visible data be submitted to the filenode in the clear (where
706 it will be encrypted by the filenode using the readkey), but encrypt
707 writer-visible data using this writekey.
710 # TODO: Can this be overwrite instead of replace?
711 def replace(new_contents):
712 """Replace the contents of the mutable file, provided that no other
713 node has published (or is attempting to publish, concurrently) a
714 newer version of the file than this one.
716 I will avoid modifying any share that is different than the version
717 given by get_sequence_number(). However, if another node is writing
718 to the file at the same time as me, I may manage to update some shares
719 while they update others. If I see any evidence of this, I will signal
720 UncoordinatedWriteError, and the file will be left in an inconsistent
721 state (possibly the version you provided, possibly the old version,
722 possibly somebody else's version, and possibly a mix of shares from
725 The recommended response to UncoordinatedWriteError is to either
726 return it to the caller (since they failed to coordinate their
727 writes), or to attempt some sort of recovery. It may be sufficient to
728 wait a random interval (with exponential backoff) and repeat your
729 operation. If I do not signal UncoordinatedWriteError, then I was
730 able to write the new version without incident.
732 I return a Deferred that fires (with a PublishStatus object) when the
733 update has completed.
736 def modify(modifier_cb):
737 """Modify the contents of the file, by downloading this version,
738 applying the modifier function (or bound method), then uploading
739 the new version. This will succeed as long as no other node
740 publishes a version between the download and the upload.
741 I return a Deferred that fires (with a PublishStatus object) when
742 the update is complete.
744 The modifier callable will be given three arguments: a string (with
745 the old contents), a 'first_time' boolean, and a servermap. As with
746 download_to_data(), the old contents will be from this version,
747 but the modifier can use the servermap to make other decisions
748 (such as refusing to apply the delta if there are multiple parallel
749 versions, or if there is evidence of a newer unrecoverable version).
750 'first_time' will be True the first time the modifier is called,
751 and False on any subsequent calls.
753 The callable should return a string with the new contents. The
754 callable must be prepared to be called multiple times, and must
755 examine the input string to see if the change that it wants to make
756 is already present in the old version. If it does not need to make
757 any changes, it can either return None, or return its input string.
759 If the modifier raises an exception, it will be returned in the
764 # The hierarchy looks like this:
771 class IFilesystemNode(Interface):
773 """Return the strongest 'cap instance' associated with this node.
774 (writecap for writeable-mutable files/directories, readcap for
775 immutable or readonly-mutable files/directories). To convert this
776 into a string, call .to_string() on the result."""
779 """Return a readonly cap instance for this node. For immutable or
780 readonly nodes, get_cap() and get_readcap() return the same thing."""
782 def get_repair_cap():
783 """Return an IURI instance that can be used to repair the file, or
784 None if this node cannot be repaired (either because it is not
785 distributed, like a LIT file, or because the node does not represent
786 sufficient authority to create a repair-cap, like a read-only RSA
787 mutable file node [which cannot create the correct write-enablers]).
790 def get_verify_cap():
791 """Return an IVerifierURI instance that represents the
792 'verifiy/refresh capability' for this node. The holder of this
793 capability will be able to renew the lease for this node, protecting
794 it from garbage-collection. They will also be able to ask a server if
795 it holds a share for the file or directory.
799 """Return the URI string corresponding to the strongest cap associated
800 with this node. If this node is read-only, the URI will only offer
801 read-only access. If this node is read-write, the URI will offer
804 If you have read-write access to a node and wish to share merely
805 read-only access with others, use get_readonly_uri().
809 """Return the URI string that can be used by others to get write
810 access to this node, if it is writeable. If this is a read-only node,
813 def get_readonly_uri():
814 """Return the URI string that can be used by others to get read-only
815 access to this node. The result is a read-only URI, regardless of
816 whether this node is read-only or read-write.
818 If you have merely read-only access to this node, get_readonly_uri()
819 will return the same thing as get_uri().
822 def get_storage_index():
823 """Return a string with the (binary) storage index in use on this
824 download. This may be None if there is no storage index (i.e. LIT
825 files and directories)."""
828 """Return True if this reference provides mutable access to the given
829 file or directory (i.e. if you can modify it), or False if not. Note
830 that even if this reference is read-only, someone else may hold a
831 read-write reference to it."""
834 """Return True if this file or directory is mutable (by *somebody*,
835 not necessarily you), False if it is is immutable. Note that a file
836 might be mutable overall, but your reference to it might be
837 read-only. On the other hand, all references to an immutable file
838 will be read-only; there are no read-write references to an immutable
843 """Return True if this is an unknown node."""
845 def is_allowed_in_immutable_directory():
846 """Return True if this node is allowed as a child of a deep-immutable
847 directory. This is true if either the node is of a known-immutable type,
848 or it is unknown and read-only.
852 """Raise any error associated with this node."""
854 # XXX: These may not be appropriate outside the context of an IReadable.
856 """Return the length (in bytes) of the data this node represents. For
857 directory nodes, I return the size of the backing store. I return
858 synchronously and do not consult the network, so for mutable objects,
859 I will return the most recently observed size for the object, or None
860 if I don't remember a size. Use get_current_size, which returns a
861 Deferred, if you want more up-to-date information."""
863 def get_current_size():
864 """I return a Deferred that fires with the length (in bytes) of the
865 data this node represents.
868 class IFileNode(IFilesystemNode):
869 """I am a node which represents a file: a sequence of bytes. I am not a
870 container, like IDirectoryNode."""
871 def get_best_readable_version():
872 """Return a Deferred that fires with an IReadable for the 'best'
873 available version of the file. The IReadable provides only read
874 access, even if this filenode was derived from a write cap.
876 For an immutable file, there is only one version. For a mutable
877 file, the 'best' version is the recoverable version with the
878 highest sequence number. If no uncoordinated writes have occurred,
879 and if enough shares are available, then this will be the most
880 recent version that has been uploaded. If no version is recoverable,
881 the Deferred will errback with an UnrecoverableFileError.
884 def download_best_version():
885 """Download the contents of the version that would be returned
886 by get_best_readable_version(). This is equivalent to calling
887 download_to_data() on the IReadable given by that method.
889 I return a Deferred that fires with a byte string when the file
890 has been fully downloaded. To support streaming download, use
891 the 'read' method of IReadable. If no version is recoverable,
892 the Deferred will errback with an UnrecoverableFileError.
895 def get_size_of_best_version():
896 """Find the size of the version that would be returned by
897 get_best_readable_version().
899 I return a Deferred that fires with an integer. If no version
900 is recoverable, the Deferred will errback with an
901 UnrecoverableFileError.
905 class IImmutableFileNode(IFileNode, IReadable):
906 """I am a node representing an immutable file. Immutable files have
910 class IMutableFileNode(IFileNode):
911 """I provide access to a 'mutable file', which retains its identity
912 regardless of what contents are put in it.
914 The consistency-vs-availability problem means that there might be
915 multiple versions of a file present in the grid, some of which might be
916 unrecoverable (i.e. have fewer than 'k' shares). These versions are
917 loosely ordered: each has a sequence number and a hash, and any version
918 with seqnum=N was uploaded by a node which has seen at least one version
921 The 'servermap' (an instance of IMutableFileServerMap) is used to
922 describe the versions that are known to be present in the grid, and which
923 servers are hosting their shares. It is used to represent the 'state of
924 the world', and is used for this purpose by my test-and-set operations.
925 Downloading the contents of the mutable file will also return a
926 servermap. Uploading a new version into the mutable file requires a
927 servermap as input, and the semantics of the replace operation is
928 'replace the file with my new version if it looks like nobody else has
929 changed the file since my previous download'. Because the file is
930 distributed, this is not a perfect test-and-set operation, but it will do
931 its best. If the replace process sees evidence of a simultaneous write,
932 it will signal an UncoordinatedWriteError, so that the caller can take
936 Most readers will want to use the 'best' current version of the file, and
937 should use my 'download_best_version()' method.
939 To unconditionally replace the file, callers should use overwrite(). This
940 is the mode that user-visible mutable files will probably use.
942 To apply some delta to the file, call modify() with a callable modifier
943 function that can apply the modification that you want to make. This is
944 the mode that dirnodes will use, since most directory modification
945 operations can be expressed in terms of deltas to the directory state.
948 Three methods are available for users who need to perform more complex
949 operations. The first is get_servermap(), which returns an up-to-date
950 servermap using a specified mode. The second is download_version(), which
951 downloads a specific version (not necessarily the 'best' one). The third
952 is 'upload', which accepts new contents and a servermap (which must have
953 been updated with MODE_WRITE). The upload method will attempt to apply
954 the new contents as long as no other node has modified the file since the
955 servermap was updated. This might be useful to a caller who wants to
956 merge multiple versions into a single new one.
958 Note that each time the servermap is updated, a specific 'mode' is used,
959 which determines how many peers are queried. To use a servermap for my
960 replace() method, that servermap must have been updated in MODE_WRITE.
961 These modes are defined in allmydata.mutable.common, and consist of
962 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
963 allmydata/mutable/servermap.py for details about the differences.
965 Mutable files are currently limited in size (about 3.5MB max) and can
966 only be retrieved and updated all-at-once, as a single big string. Future
967 versions of our mutable files will remove this restriction.
969 def get_best_mutable_version():
970 """Return a Deferred that fires with an IMutableFileVersion for
971 the 'best' available version of the file. The best version is
972 the recoverable version with the highest sequence number. If no
973 uncoordinated writes have occurred, and if enough shares are
974 available, then this will be the most recent version that has
977 If no version is recoverable, the Deferred will errback with an
978 UnrecoverableFileError.
981 def overwrite(new_contents):
982 """Unconditionally replace the contents of the mutable file with new
983 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
984 is only appropriate to use when the new contents of the file are
985 completely unrelated to the old ones, and you do not care about other
988 I return a Deferred that fires (with a PublishStatus object) when the
989 update has completed.
992 def modify(modifier_cb):
993 """Modify the contents of the file, by downloading the current
994 version, applying the modifier function (or bound method), then
995 uploading the new version. I return a Deferred that fires (with a
996 PublishStatus object) when the update is complete.
998 The modifier callable will be given three arguments: a string (with
999 the old contents), a 'first_time' boolean, and a servermap. As with
1000 download_best_version(), the old contents will be from the best
1001 recoverable version, but the modifier can use the servermap to make
1002 other decisions (such as refusing to apply the delta if there are
1003 multiple parallel versions, or if there is evidence of a newer
1004 unrecoverable version). 'first_time' will be True the first time the
1005 modifier is called, and False on any subsequent calls.
1007 The callable should return a string with the new contents. The
1008 callable must be prepared to be called multiple times, and must
1009 examine the input string to see if the change that it wants to make
1010 is already present in the old version. If it does not need to make
1011 any changes, it can either return None, or return its input string.
1013 If the modifier raises an exception, it will be returned in the
1017 def get_servermap(mode):
1018 """Return a Deferred that fires with an IMutableFileServerMap
1019 instance, updated using the given mode.
1022 def download_version(servermap, version):
1023 """Download a specific version of the file, using the servermap
1024 as a guide to where the shares are located.
1026 I return a Deferred that fires with the requested contents, or
1027 errbacks with UnrecoverableFileError. Note that a servermap which was
1028 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1029 all versions (those modes stop querying servers as soon as they can
1030 fulfil their goals), so you may want to use MODE_CHECK (which checks
1031 everything) to get increased visibility.
1034 def upload(new_contents, servermap):
1035 """Replace the contents of the file with new ones. This requires a
1036 servermap that was previously updated with MODE_WRITE.
1038 I attempt to provide test-and-set semantics, in that I will avoid
1039 modifying any share that is different than the version I saw in the
1040 servermap. However, if another node is writing to the file at the
1041 same time as me, I may manage to update some shares while they update
1042 others. If I see any evidence of this, I will signal
1043 UncoordinatedWriteError, and the file will be left in an inconsistent
1044 state (possibly the version you provided, possibly the old version,
1045 possibly somebody else's version, and possibly a mix of shares from
1048 The recommended response to UncoordinatedWriteError is to either
1049 return it to the caller (since they failed to coordinate their
1050 writes), or to attempt some sort of recovery. It may be sufficient to
1051 wait a random interval (with exponential backoff) and repeat your
1052 operation. If I do not signal UncoordinatedWriteError, then I was
1053 able to write the new version without incident.
1055 I return a Deferred that fires (with a PublishStatus object) when the
1056 publish has completed. I will update the servermap in-place with the
1057 location of all new shares.
1061 """Return this filenode's writekey, or None if the node does not have
1062 write-capability. This may be used to assist with data structures
1063 that need to make certain data available only to writers, such as the
1064 read-write child caps in dirnodes. The recommended process is to have
1065 reader-visible data be submitted to the filenode in the clear (where
1066 it will be encrypted by the filenode using the readkey), but encrypt
1067 writer-visible data using this writekey.
1071 """Returns the mutable file protocol version."""
1073 class NotEnoughSharesError(Exception):
1074 """Download was unable to get enough shares"""
1076 class NoSharesError(Exception):
1077 """Download was unable to get any shares at all."""
1079 class DownloadStopped(Exception):
1082 class UploadUnhappinessError(Exception):
1083 """Upload was unable to satisfy 'servers_of_happiness'"""
1085 class UnableToFetchCriticalDownloadDataError(Exception):
1086 """I was unable to fetch some piece of critical data which is supposed to
1087 be identically present in all shares."""
1089 class NoServersError(Exception):
1090 """Upload wasn't given any servers to work with, usually indicating a
1091 network or Introducer problem."""
1093 class ExistingChildError(Exception):
1094 """A directory node was asked to add or replace a child that already
1095 exists, and overwrite= was set to False."""
1097 class NoSuchChildError(Exception):
1098 """A directory node was asked to fetch a child which does not exist."""
1100 # avoid UnicodeEncodeErrors when converting to str
1101 return self.__repr__()
1103 class ChildOfWrongTypeError(Exception):
1104 """An operation was attempted on a child of the wrong type (file or directory)."""
1106 class IDirectoryNode(IFilesystemNode):
1107 """I represent a filesystem node that is a container, with a
1108 name-to-child mapping, holding the tahoe equivalent of a directory. All
1109 child names are unicode strings, and all children are some sort of
1110 IFilesystemNode (a file, subdirectory, or unknown node).
1115 The dirnode ('1') URI returned by this method can be used in
1116 set_uri() on a different directory ('2') to 'mount' a reference to
1117 this directory ('1') under the other ('2'). This URI is just a
1118 string, so it can be passed around through email or other out-of-band
1122 def get_readonly_uri():
1124 The dirnode ('1') URI returned by this method can be used in
1125 set_uri() on a different directory ('2') to 'mount' a reference to
1126 this directory ('1') under the other ('2'). This URI is just a
1127 string, so it can be passed around through email or other out-of-band
1132 """I return a Deferred that fires with a dictionary mapping child
1133 name (a unicode string) to (node, metadata_dict) tuples, in which
1134 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1137 def has_child(name):
1138 """I return a Deferred that fires with a boolean, True if there
1139 exists a child of the given name, False if not. The child name must
1140 be a unicode string."""
1143 """I return a Deferred that fires with a specific named child node,
1144 which is an IFilesystemNode. The child name must be a unicode string.
1145 I raise NoSuchChildError if I do not have a child by that name."""
1147 def get_metadata_for(name):
1148 """I return a Deferred that fires with the metadata dictionary for
1149 a specific named child node. The child name must be a unicode string.
1150 This metadata is stored in the *edge*, not in the child, so it is
1151 attached to the parent dirnode rather than the child node.
1152 I raise NoSuchChildError if I do not have a child by that name."""
1154 def set_metadata_for(name, metadata):
1155 """I replace any existing metadata for the named child with the new
1156 metadata. The child name must be a unicode string. This metadata is
1157 stored in the *edge*, not in the child, so it is attached to the
1158 parent dirnode rather than the child node. I return a Deferred
1159 (that fires with this dirnode) when the operation is complete.
1160 I raise NoSuchChildError if I do not have a child by that name."""
1162 def get_child_at_path(path):
1163 """Transform a child path into an IFilesystemNode.
1165 I perform a recursive series of 'get' operations to find the named
1166 descendant node. I return a Deferred that fires with the node, or
1167 errbacks with NoSuchChildError if the node could not be found.
1169 The path can be either a single string (slash-separated) or a list of
1170 path-name elements. All elements must be unicode strings.
1173 def get_child_and_metadata_at_path(path):
1174 """Transform a child path into an IFilesystemNode and metadata.
1176 I am like get_child_at_path(), but my Deferred fires with a tuple of
1177 (node, metadata). The metadata comes from the last edge. If the path
1178 is empty, the metadata will be an empty dictionary.
1181 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1182 """I add a child (by writecap+readcap) at the specific name. I return
1183 a Deferred that fires when the operation finishes. If overwrite= is
1184 True, I will replace any existing child of the same name, otherwise
1185 an existing child will cause me to return ExistingChildError. The
1186 child name must be a unicode string.
1188 The child caps could be for a file, or for a directory. If you have
1189 both the writecap and readcap, you should provide both arguments.
1190 If you have only one cap and don't know whether it is read-only,
1191 provide it as the writecap argument and leave the readcap as None.
1192 If you have only one cap that is known to be read-only, provide it
1193 as the readcap argument and leave the writecap as None.
1194 The filecaps are typically obtained from an IFilesystemNode with
1195 get_uri() and get_readonly_uri().
1197 If metadata= is provided, I will use it as the metadata for the named
1198 edge. This will replace any existing metadata. If metadata= is left
1199 as the default value of None, I will set ['mtime'] to the current
1200 time, and I will set ['ctime'] to the current time if there was not
1201 already a child by this name present. This roughly matches the
1202 ctime/mtime semantics of traditional filesystems. See the
1203 "About the metadata" section of webapi.txt for futher information.
1205 If this directory node is read-only, the Deferred will errback with a
1206 NotWriteableError."""
1208 def set_children(entries, overwrite=True):
1209 """Add multiple children (by writecap+readcap) to a directory node.
1210 Takes a dictionary, with childname as keys and (writecap, readcap)
1211 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1212 a Deferred that fires (with this dirnode) when the operation
1213 finishes. This is equivalent to calling set_uri() multiple times, but
1214 is much more efficient. All child names must be unicode strings.
1217 def set_node(name, child, metadata=None, overwrite=True):
1218 """I add a child at the specific name. I return a Deferred that fires
1219 when the operation finishes. This Deferred will fire with the child
1220 node that was just added. I will replace any existing child of the
1221 same name. The child name must be a unicode string. The 'child'
1222 instance must be an instance providing IFilesystemNode.
1224 If metadata= is provided, I will use it as the metadata for the named
1225 edge. This will replace any existing metadata. If metadata= is left
1226 as the default value of None, I will set ['mtime'] to the current
1227 time, and I will set ['ctime'] to the current time if there was not
1228 already a child by this name present. This roughly matches the
1229 ctime/mtime semantics of traditional filesystems. See the
1230 "About the metadata" section of webapi.txt for futher information.
1232 If this directory node is read-only, the Deferred will errback with a
1233 NotWriteableError."""
1235 def set_nodes(entries, overwrite=True):
1236 """Add multiple children to a directory node. Takes a dict mapping
1237 unicode childname to (child_node, metdata) tuples. If metdata=None,
1238 the original metadata is left unmodified. Returns a Deferred that
1239 fires (with this dirnode) when the operation finishes. This is
1240 equivalent to calling set_node() multiple times, but is much more
1243 def add_file(name, uploadable, metadata=None, overwrite=True):
1244 """I upload a file (using the given IUploadable), then attach the
1245 resulting ImmutableFileNode to the directory at the given name. I set
1246 metadata the same way as set_uri and set_node. The child name must be
1249 I return a Deferred that fires (with the IFileNode of the uploaded
1250 file) when the operation completes."""
1252 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1253 """I remove the child at the specific name. I return a Deferred that
1254 fires when the operation finishes. The child name must be a unicode
1255 string. If must_exist is True and I do not have a child by that name,
1256 I raise NoSuchChildError. If must_be_directory is True and the child
1257 is a file, or if must_be_file is True and the child is a directory,
1258 I raise ChildOfWrongTypeError."""
1260 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1261 """I create and attach a directory at the given name. The new
1262 directory can be empty, or it can be populated with children
1263 according to 'initial_children', which takes a dictionary in the same
1264 format as set_nodes (i.e. mapping unicode child name to (childnode,
1265 metadata) tuples). The child name must be a unicode string. I return
1266 a Deferred that fires (with the new directory node) when the
1267 operation finishes."""
1269 def move_child_to(current_child_name, new_parent, new_child_name=None,
1271 """I take one of my children and move them to a new parent. The child
1272 is referenced by name. On the new parent, the child will live under
1273 'new_child_name', which defaults to 'current_child_name'. TODO: what
1274 should we do about metadata? I return a Deferred that fires when the
1275 operation finishes. The child name must be a unicode string. I raise
1276 NoSuchChildError if I do not have a child by that name."""
1278 def build_manifest():
1279 """I generate a table of everything reachable from this directory.
1280 I also compute deep-stats as described below.
1282 I return a Monitor. The Monitor's results will be a dictionary with
1285 res['manifest']: a list of (path, cap) tuples for all nodes
1286 (directories and files) reachable from this one.
1287 'path' will be a tuple of unicode strings. The
1288 origin dirnode will be represented by an empty path
1290 res['verifycaps']: a list of (printable) verifycap strings, one for
1291 each reachable non-LIT node. This is a set:
1292 it will contain no duplicates.
1293 res['storage-index']: a list of (base32) storage index strings,
1294 one for each reachable non-LIT node. This is
1295 a set: it will contain no duplicates.
1296 res['stats']: a dictionary, the same that is generated by
1297 start_deep_stats() below.
1299 The Monitor will also have an .origin_si attribute with the (binary)
1300 storage index of the starting point.
1303 def start_deep_stats():
1304 """Return a Monitor, examining all nodes (directories and files)
1305 reachable from this one. The Monitor's results will be a dictionary
1306 with the following keys::
1308 count-immutable-files: count of how many CHK files are in the set
1309 count-mutable-files: same, for mutable files (does not include
1311 count-literal-files: same, for LIT files
1312 count-files: sum of the above three
1314 count-directories: count of directories
1316 size-immutable-files: total bytes for all CHK files in the set
1317 size-mutable-files (TODO): same, for current version of all mutable
1318 files, does not include directories
1319 size-literal-files: same, for LIT files
1320 size-directories: size of mutable files used by directories
1322 largest-directory: number of bytes in the largest directory
1323 largest-directory-children: number of children in the largest
1325 largest-immutable-file: number of bytes in the largest CHK file
1327 size-mutable-files is not yet implemented, because it would involve
1328 even more queries than deep_stats does.
1330 The Monitor will also have an .origin_si attribute with the (binary)
1331 storage index of the starting point.
1333 This operation will visit every directory node underneath this one,
1334 and can take a long time to run. On a typical workstation with good
1335 bandwidth, this can examine roughly 15 directories per second (and
1336 takes several minutes of 100% CPU for ~1700 directories).
1339 class ICodecEncoder(Interface):
1340 def set_params(data_size, required_shares, max_shares):
1341 """Set up the parameters of this encoder.
1343 This prepares the encoder to perform an operation that converts a
1344 single block of data into a number of shares, such that a future
1345 ICodecDecoder can use a subset of these shares to recover the
1346 original data. This operation is invoked by calling encode(). Once
1347 the encoding parameters are set up, the encode operation can be
1348 invoked multiple times.
1350 set_params() prepares the encoder to accept blocks of input data that
1351 are exactly 'data_size' bytes in length. The encoder will be prepared
1352 to produce 'max_shares' shares for each encode() operation (although
1353 see the 'desired_share_ids' to use less CPU). The encoding math will
1354 be chosen such that the decoder can get by with as few as
1355 'required_shares' of these shares and still reproduce the original
1356 data. For example, set_params(1000, 5, 5) offers no redundancy at
1357 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1359 Numerical Restrictions: 'data_size' is required to be an integral
1360 multiple of 'required_shares'. In general, the caller should choose
1361 required_shares and max_shares based upon their reliability
1362 requirements and the number of peers available (the total storage
1363 space used is roughly equal to max_shares*data_size/required_shares),
1364 then choose data_size to achieve the memory footprint desired (larger
1365 data_size means more efficient operation, smaller data_size means
1366 smaller memory footprint).
1368 In addition, 'max_shares' must be equal to or greater than
1369 'required_shares'. Of course, setting them to be equal causes
1370 encode() to degenerate into a particularly slow form of the 'split'
1373 See encode() for more details about how these parameters are used.
1375 set_params() must be called before any other ICodecEncoder methods
1380 """Return the 3-tuple of data_size, required_shares, max_shares"""
1382 def get_encoder_type():
1383 """Return a short string that describes the type of this encoder.
1385 There is required to be a global table of encoder classes. This method
1386 returns an index into this table; the value at this index is an
1387 encoder class, and this encoder is an instance of that class.
1390 def get_block_size():
1391 """Return the length of the shares that encode() will produce.
1394 def encode_proposal(data, desired_share_ids=None):
1395 """Encode some data.
1397 'data' must be a string (or other buffer object), and len(data) must
1398 be equal to the 'data_size' value passed earlier to set_params().
1400 This will return a Deferred that will fire with two lists. The first
1401 is a list of shares, each of which is a string (or other buffer
1402 object) such that len(share) is the same as what get_share_size()
1403 returned earlier. The second is a list of shareids, in which each is
1404 an integer. The lengths of the two lists will always be equal to each
1405 other. The user should take care to keep each share closely
1406 associated with its shareid, as one is useless without the other.
1408 The length of this output list will normally be the same as the value
1409 provided to the 'max_shares' parameter of set_params(). This may be
1410 different if 'desired_share_ids' is provided.
1412 'desired_share_ids', if provided, is required to be a sequence of
1413 ints, each of which is required to be >= 0 and < max_shares. If not
1414 provided, encode() will produce 'max_shares' shares, as if
1415 'desired_share_ids' were set to range(max_shares). You might use this
1416 if you initially thought you were going to use 10 peers, started
1417 encoding, and then two of the peers dropped out: you could use
1418 desired_share_ids= to skip the work (both memory and CPU) of
1419 producing shares for the peers which are no longer available.
1423 def encode(inshares, desired_share_ids=None):
1424 """Encode some data. This may be called multiple times. Each call is
1427 inshares is a sequence of length required_shares, containing buffers
1428 (i.e. strings), where each buffer contains the next contiguous
1429 non-overlapping segment of the input data. Each buffer is required to
1430 be the same length, and the sum of the lengths of the buffers is
1431 required to be exactly the data_size promised by set_params(). (This
1432 implies that the data has to be padded before being passed to
1433 encode(), unless of course it already happens to be an even multiple
1434 of required_shares in length.)
1436 Note: the requirement to break up your data into
1437 'required_shares' chunks of exactly the right length before
1438 calling encode() is surprising from point of view of a user
1439 who doesn't know how FEC works. It feels like an
1440 implementation detail that has leaked outside the abstraction
1441 barrier. Is there a use case in which the data to be encoded
1442 might already be available in pre-segmented chunks, such that
1443 it is faster or less work to make encode() take a list rather
1444 than splitting a single string?
1446 Yes, there is: suppose you are uploading a file with K=64,
1447 N=128, segsize=262,144. Then each in-share will be of size
1448 4096. If you use this .encode() API then your code could first
1449 read each successive 4096-byte chunk from the file and store
1450 each one in a Python string and store each such Python string
1451 in a Python list. Then you could call .encode(), passing that
1452 list as "inshares". The encoder would generate the other 64
1453 "secondary shares" and return to you a new list containing
1454 references to the same 64 Python strings that you passed in
1455 (as the primary shares) plus references to the new 64 Python
1458 (You could even imagine that your code could use readv() so
1459 that the operating system can arrange to get all of those
1460 bytes copied from the file into the Python list of Python
1461 strings as efficiently as possible instead of having a loop
1462 written in C or in Python to copy the next part of the file
1463 into the next string.)
1465 On the other hand if you instead use the .encode_proposal()
1466 API (above), then your code can first read in all of the
1467 262,144 bytes of the segment from the file into a Python
1468 string, then call .encode_proposal() passing the segment data
1469 as the "data" argument. The encoder would basically first
1470 split the "data" argument into a list of 64 in-shares of 4096
1471 byte each, and then do the same thing that .encode() does. So
1472 this would result in a little bit more copying of data and a
1473 little bit higher of a "maximum memory usage" during the
1474 process, although it might or might not make a practical
1475 difference for our current use cases.
1477 Note that "inshares" is a strange name for the parameter if
1478 you think of the parameter as being just for feeding in data
1479 to the codec. It makes more sense if you think of the result
1480 of this encoding as being the set of shares from inshares plus
1481 an extra set of "secondary shares" (or "check shares"). It is
1482 a surprising name! If the API is going to be surprising then
1483 the name should be surprising. If we switch to
1484 encode_proposal() above then we should also switch to an
1487 'desired_share_ids', if provided, is required to be a sequence of
1488 ints, each of which is required to be >= 0 and < max_shares. If not
1489 provided, encode() will produce 'max_shares' shares, as if
1490 'desired_share_ids' were set to range(max_shares). You might use this
1491 if you initially thought you were going to use 10 peers, started
1492 encoding, and then two of the peers dropped out: you could use
1493 desired_share_ids= to skip the work (both memory and CPU) of
1494 producing shares for the peers which are no longer available.
1496 For each call, encode() will return a Deferred that fires with two
1497 lists, one containing shares and the other containing the shareids.
1498 The get_share_size() method can be used to determine the length of
1499 the share strings returned by encode(). Each shareid is a small
1500 integer, exactly as passed into 'desired_share_ids' (or
1501 range(max_shares), if desired_share_ids was not provided).
1503 The shares and their corresponding shareids are required to be kept
1504 together during storage and retrieval. Specifically, the share data is
1505 useless by itself: the decoder needs to be told which share is which
1506 by providing it with both the shareid and the actual share data.
1508 This function will allocate an amount of memory roughly equal to::
1510 (max_shares - required_shares) * get_share_size()
1512 When combined with the memory that the caller must allocate to
1513 provide the input data, this leads to a memory footprint roughly
1514 equal to the size of the resulting encoded shares (i.e. the expansion
1515 factor times the size of the input segment).
1520 # returning a list of (shareidN,shareN) tuples instead of a pair of
1521 # lists (shareids..,shares..). Brian thought the tuples would
1522 # encourage users to keep the share and shareid together throughout
1523 # later processing, Zooko pointed out that the code to iterate
1524 # through two lists is not really more complicated than using a list
1525 # of tuples and there's also a performance improvement
1527 # having 'data_size' not required to be an integral multiple of
1528 # 'required_shares'. Doing this would require encode() to perform
1529 # padding internally, and we'd prefer to have any padding be done
1530 # explicitly by the caller. Yes, it is an abstraction leak, but
1531 # hopefully not an onerous one.
1534 class ICodecDecoder(Interface):
1535 def set_params(data_size, required_shares, max_shares):
1536 """Set the params. They have to be exactly the same ones that were
1537 used for encoding."""
1539 def get_needed_shares():
1540 """Return the number of shares needed to reconstruct the data.
1541 set_params() is required to be called before this."""
1543 def decode(some_shares, their_shareids):
1544 """Decode a partial list of shares into data.
1546 'some_shares' is required to be a sequence of buffers of sharedata, a
1547 subset of the shares returned by ICodecEncode.encode(). Each share is
1548 required to be of the same length. The i'th element of their_shareids
1549 is required to be the shareid of the i'th buffer in some_shares.
1551 This returns a Deferred which fires with a sequence of buffers. This
1552 sequence will contain all of the segments of the original data, in
1553 order. The sum of the lengths of all of the buffers will be the
1554 'data_size' value passed into the original ICodecEncode.set_params()
1555 call. To get back the single original input block of data, use
1556 ''.join(output_buffers), or you may wish to simply write them in
1557 order to an output file.
1559 Note that some of the elements in the result sequence may be
1560 references to the elements of the some_shares input sequence. In
1561 particular, this means that if those share objects are mutable (e.g.
1562 arrays) and if they are changed, then both the input (the
1563 'some_shares' parameter) and the output (the value given when the
1564 deferred is triggered) will change.
1566 The length of 'some_shares' is required to be exactly the value of
1567 'required_shares' passed into the original ICodecEncode.set_params()
1571 class IEncoder(Interface):
1572 """I take an object that provides IEncryptedUploadable, which provides
1573 encrypted data, and a list of shareholders. I then encode, hash, and
1574 deliver shares to those shareholders. I will compute all the necessary
1575 Merkle hash trees that are necessary to validate the crypttext that
1576 eventually comes back from the shareholders. I provide the URI Extension
1577 Block Hash, and the encoding parameters, both of which must be included
1580 I do not choose shareholders, that is left to the IUploader. I must be
1581 given a dict of RemoteReferences to storage buckets that are ready and
1582 willing to receive data.
1586 """Specify the number of bytes that will be encoded. This must be
1587 peformed before get_serialized_params() can be called.
1589 def set_params(params):
1590 """Override the default encoding parameters. 'params' is a tuple of
1591 (k,d,n), where 'k' is the number of required shares, 'd' is the
1592 servers_of_happiness, and 'n' is the total number of shares that will
1595 Encoding parameters can be set in three ways. 1: The Encoder class
1596 provides defaults (3/7/10). 2: the Encoder can be constructed with
1597 an 'options' dictionary, in which the
1598 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1599 set_params((k,d,n)) can be called.
1601 If you intend to use set_params(), you must call it before
1602 get_share_size or get_param are called.
1605 def set_encrypted_uploadable(u):
1606 """Provide a source of encrypted upload data. 'u' must implement
1607 IEncryptedUploadable.
1609 When this is called, the IEncryptedUploadable will be queried for its
1610 length and the storage_index that should be used.
1612 This returns a Deferred that fires with this Encoder instance.
1614 This must be performed before start() can be called.
1617 def get_param(name):
1618 """Return an encoding parameter, by name.
1620 'storage_index': return a string with the (16-byte truncated SHA-256
1621 hash) storage index to which these shares should be
1624 'share_counts': return a tuple describing how many shares are used:
1625 (needed_shares, servers_of_happiness, total_shares)
1627 'num_segments': return an int with the number of segments that
1630 'segment_size': return an int with the size of each segment.
1632 'block_size': return the size of the individual blocks that will
1633 be delivered to a shareholder's put_block() method. By
1634 knowing this, the shareholder will be able to keep all
1635 blocks in a single file and still provide random access
1636 when reading them. # TODO: can we avoid exposing this?
1638 'share_size': an int with the size of the data that will be stored
1639 on each shareholder. This is aggregate amount of data
1640 that will be sent to the shareholder, summed over all
1641 the put_block() calls I will ever make. It is useful to
1642 determine this size before asking potential
1643 shareholders whether they will grant a lease or not,
1644 since their answers will depend upon how much space we
1645 need. TODO: this might also include some amount of
1646 overhead, like the size of all the hashes. We need to
1647 decide whether this is useful or not.
1649 'serialized_params': a string with a concise description of the
1650 codec name and its parameters. This may be passed
1651 into the IUploadable to let it make sure that
1652 the same file encoded with different parameters
1653 will result in different storage indexes.
1655 Once this is called, set_size() and set_params() may not be called.
1658 def set_shareholders(shareholders, servermap):
1659 """Tell the encoder where to put the encoded shares. 'shareholders'
1660 must be a dictionary that maps share number (an integer ranging from
1661 0 to n-1) to an instance that provides IStorageBucketWriter.
1662 'servermap' is a dictionary that maps share number (as defined above)
1663 to a set of peerids. This must be performed before start() can be
1667 """Begin the encode/upload process. This involves reading encrypted
1668 data from the IEncryptedUploadable, encoding it, uploading the shares
1669 to the shareholders, then sending the hash trees.
1671 set_encrypted_uploadable() and set_shareholders() must be called
1672 before this can be invoked.
1674 This returns a Deferred that fires with a verify cap when the upload
1675 process is complete. The verifycap, plus the encryption key, is
1676 sufficient to construct the read cap.
1679 class IDecoder(Interface):
1680 """I take a list of shareholders and some setup information, then
1681 download, validate, decode, and decrypt data from them, writing the
1682 results to an output file.
1684 I do not locate the shareholders, that is left to the IDownloader. I must
1685 be given a dict of RemoteReferences to storage buckets that are ready to
1690 """I take a file-like object (providing write and close) to which all
1691 the plaintext data will be written.
1693 TODO: producer/consumer . Maybe write() should return a Deferred that
1694 indicates when it will accept more data? But probably having the
1695 IDecoder be a producer is easier to glue to IConsumer pieces.
1698 def set_shareholders(shareholders):
1699 """I take a dictionary that maps share identifiers (small integers)
1700 to RemoteReferences that provide RIBucketReader. This must be called
1704 """I start the download. This process involves retrieving data and
1705 hash chains from the shareholders, using the hashes to validate the
1706 data, decoding the shares into segments, decrypting the segments,
1707 then writing the resulting plaintext to the output file.
1709 I return a Deferred that will fire (with self) when the download is
1713 class IDownloadTarget(Interface):
1714 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1715 # will register itself as a producer. This allows the target to invoke
1716 # downloader.pauseProducing, resumeProducing, and stopProducing.
1718 """Called before any calls to write() or close(). If an error
1719 occurs before any data is available, fail() may be called without
1720 a previous call to open().
1722 'size' is the length of the file being downloaded, in bytes."""
1725 """Output some data to the target."""
1727 """Inform the target that there is no more data to be written."""
1729 """fail() is called to indicate that the download has failed. 'why'
1730 is a Failure object indicating what went wrong. No further methods
1731 will be invoked on the IDownloadTarget after fail()."""
1732 def register_canceller(cb):
1733 """The CiphertextDownloader uses this to register a no-argument function
1734 that the target can call to cancel the download. Once this canceller
1735 is invoked, no further calls to write() or close() will be made."""
1737 """When the CiphertextDownloader is done, this finish() function will be
1738 called. Whatever it returns will be returned to the invoker of
1739 Downloader.download.
1742 class IDownloader(Interface):
1743 def download(uri, target):
1744 """Perform a CHK download, sending the data to the given target.
1745 'target' must provide IDownloadTarget.
1747 Returns a Deferred that fires (with the results of target.finish)
1748 when the download is finished, or errbacks if something went wrong."""
1750 class IEncryptedUploadable(Interface):
1751 def set_upload_status(upload_status):
1752 """Provide an IUploadStatus object that should be filled with status
1753 information. The IEncryptedUploadable is responsible for setting
1754 key-determination progress ('chk'), size, storage_index, and
1755 ciphertext-fetch progress. It may delegate some of this
1756 responsibility to others, in particular to the IUploadable."""
1759 """This behaves just like IUploadable.get_size()."""
1761 def get_all_encoding_parameters():
1762 """Return a Deferred that fires with a tuple of
1763 (k,happy,n,segment_size). The segment_size will be used as-is, and
1764 must match the following constraints: it must be a multiple of k, and
1765 it shouldn't be unreasonably larger than the file size (if
1766 segment_size is larger than filesize, the difference must be stored
1769 This usually passes through to the IUploadable method of the same
1772 The encoder strictly obeys the values returned by this method. To
1773 make an upload use non-default encoding parameters, you must arrange
1774 to control the values that this method returns.
1777 def get_storage_index():
1778 """Return a Deferred that fires with a 16-byte storage index.
1781 def read_encrypted(length, hash_only):
1782 """This behaves just like IUploadable.read(), but returns crypttext
1783 instead of plaintext. If hash_only is True, then this discards the
1784 data (and returns an empty list); this improves efficiency when
1785 resuming an interrupted upload (where we need to compute the
1786 plaintext hashes, but don't need the redundant encrypted data)."""
1788 def get_plaintext_hashtree_leaves(first, last, num_segments):
1789 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1790 plaintext segments, i.e. get the tagged hashes of the given segments.
1791 The segment size is expected to be generated by the
1792 IEncryptedUploadable before any plaintext is read or ciphertext
1793 produced, so that the segment hashes can be generated with only a
1796 This returns a Deferred which fires with a sequence of hashes, using:
1798 tuple(segment_hashes[first:last])
1800 'num_segments' is used to assert that the number of segments that the
1801 IEncryptedUploadable handled matches the number of segments that the
1802 encoder was expecting.
1804 This method must not be called until the final byte has been read
1805 from read_encrypted(). Once this method is called, read_encrypted()
1806 can never be called again.
1809 def get_plaintext_hash():
1810 """OBSOLETE; Get the hash of the whole plaintext.
1812 This returns a Deferred which fires with a tagged SHA-256 hash of the
1813 whole plaintext, obtained from hashutil.plaintext_hash(data).
1817 """Just like IUploadable.close()."""
1819 class IUploadable(Interface):
1820 def set_upload_status(upload_status):
1821 """Provide an IUploadStatus object that should be filled with status
1822 information. The IUploadable is responsible for setting
1823 key-determination progress ('chk')."""
1825 def set_default_encoding_parameters(params):
1826 """Set the default encoding parameters, which must be a dict mapping
1827 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1828 'max_segment_size'. These might have an influence on the final
1829 encoding parameters returned by get_all_encoding_parameters(), if the
1830 Uploadable doesn't have more specific preferences.
1832 This call is optional: if it is not used, the Uploadable will use
1833 some built-in defaults. If used, this method must be called before
1834 any other IUploadable methods to have any effect.
1838 """Return a Deferred that will fire with the length of the data to be
1839 uploaded, in bytes. This will be called before the data is actually
1840 used, to compute encoding parameters.
1843 def get_all_encoding_parameters():
1844 """Return a Deferred that fires with a tuple of
1845 (k,happy,n,segment_size). The segment_size will be used as-is, and
1846 must match the following constraints: it must be a multiple of k, and
1847 it shouldn't be unreasonably larger than the file size (if
1848 segment_size is larger than filesize, the difference must be stored
1851 The relative values of k and n allow some IUploadables to request
1852 better redundancy than others (in exchange for consuming more space
1855 Larger values of segment_size reduce hash overhead, while smaller
1856 values reduce memory footprint and cause data to be delivered in
1857 smaller pieces (which may provide a smoother and more predictable
1858 download experience).
1860 The encoder strictly obeys the values returned by this method. To
1861 make an upload use non-default encoding parameters, you must arrange
1862 to control the values that this method returns. One way to influence
1863 them may be to call set_encoding_parameters() before calling
1864 get_all_encoding_parameters().
1867 def get_encryption_key():
1868 """Return a Deferred that fires with a 16-byte AES key. This key will
1869 be used to encrypt the data. The key will also be hashed to derive
1872 Uploadables which want to achieve convergence should hash their file
1873 contents and the serialized_encoding_parameters to form the key
1874 (which of course requires a full pass over the data). Uploadables can
1875 use the upload.ConvergentUploadMixin class to achieve this
1878 Uploadables which do not care about convergence (or do not wish to
1879 make multiple passes over the data) can simply return a
1880 strongly-random 16 byte string.
1882 get_encryption_key() may be called multiple times: the IUploadable is
1883 required to return the same value each time.
1887 """Return a Deferred that fires with a list of strings (perhaps with
1888 only a single element) which, when concatenated together, contain the
1889 next 'length' bytes of data. If EOF is near, this may provide fewer
1890 than 'length' bytes. The total number of bytes provided by read()
1891 before it signals EOF must equal the size provided by get_size().
1893 If the data must be acquired through multiple internal read
1894 operations, returning a list instead of a single string may help to
1895 reduce string copies. However, the length of the concatenated strings
1896 must equal the amount of data requested, unless EOF is encountered.
1897 Long reads, or short reads without EOF, are not allowed. read()
1898 should return the same amount of data as a local disk file read, just
1899 in a different shape and asynchronously.
1901 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1902 so a 10kB file means length=3kB, 100kB file means length=30kB,
1903 and >=1MB file means length=300kB.
1905 This method provides for a single full pass through the data. Later
1906 use cases may desire multiple passes or access to only parts of the
1907 data (such as a mutable file making small edits-in-place). This API
1908 will be expanded once those use cases are better understood.
1912 """The upload is finished, and whatever filehandle was in use may be
1916 class IMutableUploadable(Interface):
1918 I represent content that is due to be uploaded to a mutable filecap.
1920 # This is somewhat simpler than the IUploadable interface above
1921 # because mutable files do not need to be concerned with possibly
1922 # generating a CHK, nor with per-file keys. It is a subset of the
1923 # methods in IUploadable, though, so we could just as well implement
1924 # the mutable uploadables as IUploadables that don't happen to use
1925 # those methods (with the understanding that the unused methods will
1926 # never be called on such objects)
1929 Returns a Deferred that fires with the size of the content held
1935 Returns a list of strings which, when concatenated, are the next
1936 length bytes of the file, or fewer if there are fewer bytes
1937 between the current location and the end of the file.
1942 The process that used the Uploadable is finished using it, so
1943 the uploadable may be closed.
1946 class IUploadResults(Interface):
1947 """I am returned by immutable upload() methods and contain the results of
1950 Note that some of my methods return empty values (0 or an empty dict)
1951 when called for non-distributed LIT files."""
1953 def get_file_size():
1954 """Return the file size, in bytes."""
1956 """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
1961 def get_preexisting_shares():
1962 """Return the number of shares that were already present in the grid."""
1963 def get_pushed_shares():
1964 """Return the number of shares that were uploaded."""
1966 """Return a dict mapping share identifier to set of IServer
1967 instances. This indicates which servers were given which shares. For
1968 immutable files, the shareid is an integer (the share number, from 0
1969 to N-1). For mutable files, it is a string of the form
1970 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1971 the share number."""
1972 def get_servermap():
1973 """Return dict mapping IServer instance to a set of share numbers."""
1975 """Return dict of timing information, mapping name to seconds. All
1977 total : total upload time, start to finish
1978 storage_index : time to compute the storage index
1979 peer_selection : time to decide which peers will be used
1980 contacting_helper : initial helper query to upload/no-upload decision
1981 helper_total : initial helper query to helper finished pushing
1982 cumulative_fetch : helper waiting for ciphertext requests
1983 total_fetch : helper start to last ciphertext response
1984 cumulative_encoding : just time spent in zfec
1985 cumulative_sending : just time spent waiting for storage servers
1986 hashes_and_close : last segment push to shareholder close
1987 total_encode_and_push : first encode to shareholder close
1989 def get_uri_extension_data():
1990 """Return the dict of UEB data created for this file."""
1991 def get_verifycapstr():
1992 """Return the (string) verify-cap URI for the uploaded object."""
1994 class IDownloadResults(Interface):
1995 """I am created internally by download() methods. I contain a number of
1996 public attributes which contain details about the download process.::
1998 .file_size : the size of the file, in bytes
1999 .servers_used : set of server peerids that were used during download
2000 .server_problems : dict mapping server peerid to a problem string. Only
2001 servers that had problems (bad hashes, disconnects)
2003 .servermap : dict mapping server peerid to a set of share numbers. Only
2004 servers that had any shares are listed here.
2005 .timings : dict of timing information, mapping name to seconds (float)
2006 peer_selection : time to ask servers about shares
2007 servers_peer_selection : dict of peerid to DYHB-query time
2008 uri_extension : time to fetch a copy of the URI extension block
2009 hashtrees : time to fetch the hash trees
2010 segments : time to fetch, decode, and deliver segments
2011 cumulative_fetch : time spent waiting for storage servers
2012 cumulative_decode : just time spent in zfec
2013 cumulative_decrypt : just time spent in decryption
2014 total : total download time, start to finish
2015 fetch_per_server : dict of server to list of per-segment fetch times
2019 class IUploader(Interface):
2020 def upload(uploadable):
2021 """Upload the file. 'uploadable' must impement IUploadable. This
2022 returns a Deferred which fires with an IUploadResults instance, from
2023 which the URI of the file can be obtained as results.uri ."""
2025 def upload_ssk(write_capability, new_version, uploadable):
2026 """TODO: how should this work?"""
2028 class ICheckable(Interface):
2029 def check(monitor, verify=False, add_lease=False):
2030 """Check up on my health, optionally repairing any problems.
2032 This returns a Deferred that fires with an instance that provides
2033 ICheckResults, or None if the object is non-distributed (i.e. LIT
2036 The monitor will be checked periodically to see if the operation has
2037 been cancelled. If so, no new queries will be sent, and the Deferred
2038 will fire (with a OperationCancelledError) immediately.
2040 Filenodes and dirnodes (which provide IFilesystemNode) are also
2041 checkable. Instances that represent verifier-caps will be checkable
2042 but not downloadable. Some objects (like LIT files) do not actually
2043 live in the grid, and their checkers return None (non-distributed
2044 files are always healthy).
2046 If verify=False, a relatively lightweight check will be performed: I
2047 will ask all servers if they have a share for me, and I will believe
2048 whatever they say. If there are at least N distinct shares on the
2049 grid, my results will indicate r.is_healthy()==True. This requires a
2050 roundtrip to each server, but does not transfer very much data, so
2051 the network bandwidth is fairly low.
2053 If verify=True, a more resource-intensive check will be performed:
2054 every share will be downloaded, and the hashes will be validated on
2055 every bit. I will ignore any shares that failed their hash checks. If
2056 there are at least N distinct valid shares on the grid, my results
2057 will indicate r.is_healthy()==True. This requires N/k times as much
2058 download bandwidth (and server disk IO) as a regular download. If a
2059 storage server is holding a corrupt share, or is experiencing memory
2060 failures during retrieval, or is malicious or buggy, then
2061 verification will detect the problem, but checking will not.
2063 If add_lease=True, I will ensure that an up-to-date lease is present
2064 on each share. The lease secrets will be derived from by node secret
2065 (in BASEDIR/private/secret), so either I will add a new lease to the
2066 share, or I will merely renew the lease that I already had. In a
2067 future version of the storage-server protocol (once Accounting has
2068 been implemented), there may be additional options here to define the
2069 kind of lease that is obtained (which account number to claim, etc).
2071 TODO: any problems seen during checking will be reported to the
2072 health-manager.furl, a centralized object which is responsible for
2073 figuring out why files are unhealthy so corrective action can be
2077 def check_and_repair(monitor, verify=False, add_lease=False):
2078 """Like check(), but if the file/directory is not healthy, attempt to
2081 Any non-healthy result will cause an immediate repair operation, to
2082 generate and upload new shares. After repair, the file will be as
2083 healthy as we can make it. Details about what sort of repair is done
2084 will be put in the check-and-repair results. The Deferred will not
2085 fire until the repair is complete.
2087 This returns a Deferred which fires with an instance of
2088 ICheckAndRepairResults."""
2090 class IDeepCheckable(Interface):
2091 def start_deep_check(verify=False, add_lease=False):
2092 """Check upon the health of me and everything I can reach.
2094 This is a recursive form of check(), useable only on dirnodes.
2096 I return a Monitor, with results that are an IDeepCheckResults
2099 TODO: If any of the directories I traverse are unrecoverable, the
2100 Monitor will report failure. If any of the files I check upon are
2101 unrecoverable, those problems will be reported in the
2102 IDeepCheckResults as usual, and the Monitor will not report a
2106 def start_deep_check_and_repair(verify=False, add_lease=False):
2107 """Check upon the health of me and everything I can reach. Repair
2108 anything that isn't healthy.
2110 This is a recursive form of check_and_repair(), useable only on
2113 I return a Monitor, with results that are an
2114 IDeepCheckAndRepairResults object.
2116 TODO: If any of the directories I traverse are unrecoverable, the
2117 Monitor will report failure. If any of the files I check upon are
2118 unrecoverable, those problems will be reported in the
2119 IDeepCheckResults as usual, and the Monitor will not report a
2123 class ICheckResults(Interface):
2124 """I contain the detailed results of a check/verify operation.
2127 def get_storage_index():
2128 """Return a string with the (binary) storage index."""
2129 def get_storage_index_string():
2130 """Return a string with the (printable) abbreviated storage index."""
2132 """Return the (string) URI of the object that was checked."""
2135 """Return a boolean, True if the file/dir is fully healthy, False if
2136 it is damaged in any way. Non-distributed LIT files always return
2139 def is_recoverable():
2140 """Return a boolean, True if the file/dir can be recovered, False if
2141 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2142 files always return True."""
2144 def needs_rebalancing():
2145 """Return a boolean, True if the file/dirs reliability could be
2146 improved by moving shares to new servers. Non-distributed LIT files
2147 always return False."""
2149 # the following methods all return None for non-distributed LIT files
2151 def get_encoding_needed():
2152 """Return 'k', the number of shares required for recovery"""
2153 def get_encoding_expected():
2154 """Return 'N', the number of total shares generated"""
2156 def get_share_counter_good():
2157 """Return the number of distinct good shares that were found. For
2158 mutable files, this counts shares for the 'best' version."""
2159 def get_share_counter_wrong():
2160 """For mutable files, return the number of shares for versions other
2161 than the 'best' one (which is defined as being the recoverable
2162 version with the highest sequence number, then the highest roothash).
2163 These are either leftover shares from an older version (perhaps on a
2164 server that was offline when an update occurred), shares from an
2165 unrecoverable newer version, or shares from an alternate current
2166 version that results from an uncoordinated write collision. For a
2167 healthy file, this will equal 0. For immutable files, this will
2170 def get_corrupt_shares():
2171 """Return a list of 'share locators', one for each share that was
2172 found to be corrupt (integrity failure). Each share locator is a list
2173 of (IServer, storage_index, sharenum)."""
2175 def get_incompatible_shares():
2176 """Return a list of 'share locators', one for each share that was
2177 found to be of an unknown format. Each share locator is a list of
2178 (IServer, storage_index, sharenum)."""
2180 def get_servers_responding():
2181 """Return a list of IServer objects, one for each server which
2182 responded to the share query (even if they said they didn't have
2183 shares, and even if they said they did have shares but then didn't
2184 send them when asked, or dropped the connection, or returned a
2185 Failure, and even if they said they did have shares and sent
2186 incorrect ones when asked)"""
2188 def get_host_counter_good_shares():
2189 """Return the number of distinct storage servers with good shares. If
2190 this number is less than get_share_counters()[good], then some shares
2191 are doubled up, increasing the correlation of failures. This
2192 indicates that one or more shares should be moved to an otherwise
2193 unused server, if one is available.
2196 def get_version_counter_recoverable():
2197 """Return the number of recoverable versions of the file. For a
2198 healthy file, this will equal 1."""
2200 def get_version_counter_unrecoverable():
2201 """Return the number of unrecoverable versions of the file. For a
2202 healthy file, this will be 0."""
2205 """Return a dict mapping share identifier to list of IServer objects.
2206 This indicates which servers are holding which shares. For immutable
2207 files, the shareid is an integer (the share number, from 0 to N-1).
2208 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2209 containing the sequence number, the roothash, and the share number."""
2212 """Return a string with a brief (one-line) summary of the results."""
2215 """Return a list of strings with more detailed results."""
2217 class ICheckAndRepairResults(Interface):
2218 """I contain the detailed results of a check/verify/repair operation.
2220 The IFilesystemNode.check()/verify()/repair() methods all return
2221 instances that provide ICheckAndRepairResults.
2224 def get_storage_index():
2225 """Return a string with the (binary) storage index."""
2226 def get_storage_index_string():
2227 """Return a string with the (printable) abbreviated storage index."""
2228 def get_repair_attempted():
2229 """Return a boolean, True if a repair was attempted. We might not
2230 attempt to repair the file because it was healthy, or healthy enough
2231 (i.e. some shares were missing but not enough to exceed some
2232 threshold), or because we don't know how to repair this object."""
2233 def get_repair_successful():
2234 """Return a boolean, True if repair was attempted and the file/dir
2235 was fully healthy afterwards. False if no repair was attempted or if
2236 a repair attempt failed."""
2237 def get_pre_repair_results():
2238 """Return an ICheckResults instance that describes the state of the
2239 file/dir before any repair was attempted."""
2240 def get_post_repair_results():
2241 """Return an ICheckResults instance that describes the state of the
2242 file/dir after any repair was attempted. If no repair was attempted,
2243 the pre-repair and post-repair results will be identical."""
2246 class IDeepCheckResults(Interface):
2247 """I contain the results of a deep-check operation.
2249 This is returned by a call to ICheckable.deep_check().
2252 def get_root_storage_index_string():
2253 """Return the storage index (abbreviated human-readable string) of
2254 the first object checked."""
2256 """Return a dictionary with the following keys::
2258 count-objects-checked: count of how many objects were checked
2259 count-objects-healthy: how many of those objects were completely
2261 count-objects-unhealthy: how many were damaged in some way
2262 count-objects-unrecoverable: how many were unrecoverable
2263 count-corrupt-shares: how many shares were found to have
2264 corruption, summed over all objects
2268 def get_corrupt_shares():
2269 """Return a set of (IServer, storage_index, sharenum) for all shares
2270 that were found to be corrupt. storage_index is binary."""
2271 def get_all_results():
2272 """Return a dictionary mapping pathname (a tuple of strings, ready to
2273 be slash-joined) to an ICheckResults instance, one for each object
2274 that was checked."""
2276 def get_results_for_storage_index(storage_index):
2277 """Retrive the ICheckResults instance for the given (binary)
2278 storage index. Raises KeyError if there are no results for that
2282 """Return a dictionary with the same keys as
2283 IDirectoryNode.deep_stats()."""
2285 class IDeepCheckAndRepairResults(Interface):
2286 """I contain the results of a deep-check-and-repair operation.
2288 This is returned by a call to ICheckable.deep_check_and_repair().
2291 def get_root_storage_index_string():
2292 """Return the storage index (abbreviated human-readable string) of
2293 the first object checked."""
2295 """Return a dictionary with the following keys::
2297 count-objects-checked: count of how many objects were checked
2298 count-objects-healthy-pre-repair: how many of those objects were
2299 completely healthy (before any
2301 count-objects-unhealthy-pre-repair: how many were damaged in
2303 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2304 count-objects-healthy-post-repair: how many of those objects were
2305 completely healthy (after any
2307 count-objects-unhealthy-post-repair: how many were damaged in
2309 count-objects-unrecoverable-post-repair: how many were
2311 count-repairs-attempted: repairs were attempted on this many
2312 objects. The count-repairs- keys will
2313 always be provided, however unless
2314 repair=true is present, they will all
2316 count-repairs-successful: how many repairs resulted in healthy
2318 count-repairs-unsuccessful: how many repairs resulted did not
2319 results in completely healthy objects
2320 count-corrupt-shares-pre-repair: how many shares were found to
2321 have corruption, summed over all
2322 objects examined (before any
2324 count-corrupt-shares-post-repair: how many shares were found to
2325 have corruption, summed over all
2326 objects examined (after any
2331 """Return a dictionary with the same keys as
2332 IDirectoryNode.deep_stats()."""
2334 def get_corrupt_shares():
2335 """Return a set of (IServer, storage_index, sharenum) for all shares
2336 that were found to be corrupt before any repair was attempted.
2337 storage_index is binary.
2339 def get_remaining_corrupt_shares():
2340 """Return a set of (IServer, storage_index, sharenum) for all shares
2341 that were found to be corrupt after any repair was completed.
2342 storage_index is binary. These are shares that need manual inspection
2343 and probably deletion.
2345 def get_all_results():
2346 """Return a dictionary mapping pathname (a tuple of strings, ready to
2347 be slash-joined) to an ICheckAndRepairResults instance, one for each
2348 object that was checked."""
2350 def get_results_for_storage_index(storage_index):
2351 """Retrive the ICheckAndRepairResults instance for the given (binary)
2352 storage index. Raises KeyError if there are no results for that
2356 class IRepairable(Interface):
2357 def repair(check_results):
2358 """Attempt to repair the given object. Returns a Deferred that fires
2359 with a IRepairResults object.
2361 I must be called with an object that implements ICheckResults, as
2362 proof that you have actually discovered a problem with this file. I
2363 will use the data in the checker results to guide the repair process,
2364 such as which servers provided bad data and should therefore be
2365 avoided. The ICheckResults object is inside the
2366 ICheckAndRepairResults object, which is returned by the
2367 ICheckable.check() method::
2369 d = filenode.check(repair=False)
2370 def _got_results(check_and_repair_results):
2371 check_results = check_and_repair_results.get_pre_repair_results()
2372 return filenode.repair(check_results)
2373 d.addCallback(_got_results)
2377 class IRepairResults(Interface):
2378 """I contain the results of a repair operation."""
2379 def get_successful(self):
2380 """Returns a boolean: True if the repair made the file healthy, False
2381 if not. Repair failure generally indicates a file that has been
2382 damaged beyond repair."""
2385 class IClient(Interface):
2386 def upload(uploadable):
2387 """Upload some data into a CHK, get back the UploadResults for it.
2388 @param uploadable: something that implements IUploadable
2389 @return: a Deferred that fires with the UploadResults instance.
2390 To get the URI for this file, use results.uri .
2393 def create_mutable_file(contents=""):
2394 """Create a new mutable file (with initial) contents, get back the
2397 @param contents: (bytestring, callable, or None): this provides the
2398 initial contents of the mutable file. If 'contents' is a bytestring,
2399 it will be used as-is. If 'contents' is a callable, it will be
2400 invoked with the new MutableFileNode instance and is expected to
2401 return a bytestring with the initial contents of the file (the
2402 callable can use node.get_writekey() to decide how to encrypt the
2403 initial contents, e.g. for a brand new dirnode with initial
2404 children). contents=None is equivalent to an empty string. Using
2405 content_maker= is more efficient than creating a mutable file and
2406 setting its contents in two separate operations.
2408 @return: a Deferred that fires with an IMutableFileNode instance.
2411 def create_dirnode(initial_children={}):
2412 """Create a new unattached dirnode, possibly with initial children.
2414 @param initial_children: dict with keys that are unicode child names,
2415 and values that are (childnode, metadata) tuples.
2417 @return: a Deferred that fires with the new IDirectoryNode instance.
2420 def create_node_from_uri(uri, rouri):
2421 """Create a new IFilesystemNode instance from the uri, synchronously.
2422 @param uri: a string or IURI-providing instance, or None. This could
2423 be for a LiteralFileNode, a CHK file node, a mutable file
2424 node, or a directory node
2425 @param rouri: a string or IURI-providing instance, or None. If the
2426 main uri is None, I will use the rouri instead. If I
2427 recognize the format of the main uri, I will ignore the
2428 rouri (because it can be derived from the writecap).
2430 @return: an instance that provides IFilesystemNode (or more usefully
2431 one of its subclasses). File-specifying URIs will result in
2432 IFileNode-providing instances, like ImmutableFileNode,
2433 LiteralFileNode, or MutableFileNode. Directory-specifying
2434 URIs will result in IDirectoryNode-providing instances, like
2438 class INodeMaker(Interface):
2439 """The NodeMaker is used to create IFilesystemNode instances. It can
2440 accept a filecap/dircap string and return the node right away. It can
2441 also create new nodes (i.e. upload a file, or create a mutable file)
2442 asynchronously. Once you have one of these nodes, you can use other
2443 methods to determine whether it is a file or directory, and to download
2444 or modify its contents.
2446 The NodeMaker encapsulates all the authorities that these
2447 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2448 Tahoe process will typically have a single NodeMaker, but unit tests may
2449 create simplified/mocked forms for testing purposes.
2451 def create_from_cap(writecap, readcap=None, **kwargs):
2452 """I create an IFilesystemNode from the given writecap/readcap. I can
2453 only provide nodes for existing file/directory objects: use my other
2454 methods to create new objects. I return synchronously."""
2456 def create_mutable_file(contents=None, keysize=None):
2457 """I create a new mutable file, and return a Deferred which will fire
2458 with the IMutableFileNode instance when it is ready. If contents= is
2459 provided (a bytestring), it will be used as the initial contents of
2460 the new file, otherwise the file will contain zero bytes. keysize= is
2461 for use by unit tests, to create mutable files that are smaller than
2464 def create_new_mutable_directory(initial_children={}):
2465 """I create a new mutable directory, and return a Deferred which will
2466 fire with the IDirectoryNode instance when it is ready. If
2467 initial_children= is provided (a dict mapping unicode child name to
2468 (childnode, metadata_dict) tuples), the directory will be populated
2469 with those children, otherwise it will be empty."""
2471 class IClientStatus(Interface):
2472 def list_all_uploads():
2473 """Return a list of uploader objects, one for each upload which
2474 currently has an object available (tracked with weakrefs). This is
2475 intended for debugging purposes."""
2476 def list_active_uploads():
2477 """Return a list of active IUploadStatus objects."""
2478 def list_recent_uploads():
2479 """Return a list of IUploadStatus objects for the most recently
2482 def list_all_downloads():
2483 """Return a list of downloader objects, one for each download which
2484 currently has an object available (tracked with weakrefs). This is
2485 intended for debugging purposes."""
2486 def list_active_downloads():
2487 """Return a list of active IDownloadStatus objects."""
2488 def list_recent_downloads():
2489 """Return a list of IDownloadStatus objects for the most recently
2490 started downloads."""
2492 class IUploadStatus(Interface):
2494 """Return a timestamp (float with seconds since epoch) indicating
2495 when the operation was started."""
2496 def get_storage_index():
2497 """Return a string with the (binary) storage index in use on this
2498 upload. Returns None if the storage index has not yet been
2501 """Return an integer with the number of bytes that will eventually
2502 be uploaded for this file. Returns None if the size is not yet known.
2505 """Return True if this upload is using a Helper, False if not."""
2507 """Return a string describing the current state of the upload
2510 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2511 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2512 made towards hashing the file to determine a CHK encryption key: if
2513 non-convergent encryption is in use, this will be trivial, otherwise
2514 the whole file must be hashed. 'ciphertext' describes how much of the
2515 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2516 uploads. 'encode_and_push' describes how much of the encode-and-push
2517 process has finished: for helper uploads this is dependent upon the
2518 helper providing progress reports. It might be reasonable to add all
2519 three numbers and report the sum to the user."""
2521 """Return True if the upload is currently active, False if not."""
2523 """Return an instance of UploadResults (which contains timing and
2524 sharemap information). Might return None if the upload is not yet
2527 """Each upload status gets a unique number: this method returns that
2528 number. This provides a handle to this particular upload, so a web
2529 page can generate a suitable hyperlink."""
2531 class IDownloadStatus(Interface):
2533 """Return a timestamp (float with seconds since epoch) indicating
2534 when the operation was started."""
2535 def get_storage_index():
2536 """Return a string with the (binary) storage index in use on this
2537 download. This may be None if there is no storage index (i.e. LIT
2540 """Return an integer with the number of bytes that will eventually be
2541 retrieved for this file. Returns None if the size is not yet known.
2544 """Return True if this download is using a Helper, False if not."""
2546 """Return a string describing the current state of the download
2549 """Returns a float (from 0.0 to 1.0) describing the amount of the
2550 download that has completed. This value will remain at 0.0 until the
2551 first byte of plaintext is pushed to the download target."""
2553 """Return True if the download is currently active, False if not."""
2555 """Each download status gets a unique number: this method returns
2556 that number. This provides a handle to this particular download, so a
2557 web page can generate a suitable hyperlink."""
2559 class IServermapUpdaterStatus(Interface):
2561 class IPublishStatus(Interface):
2563 class IRetrieveStatus(Interface):
2566 class NotCapableError(Exception):
2567 """You have tried to write to a read-only node."""
2569 class BadWriteEnablerError(Exception):
2572 class RIControlClient(RemoteInterface):
2574 def wait_for_client_connections(num_clients=int):
2575 """Do not return until we have connections to at least NUM_CLIENTS
2579 def upload_from_file_to_uri(filename=str,
2580 convergence=ChoiceOf(None,
2581 StringConstraint(2**20))):
2582 """Upload a file to the grid. This accepts a filename (which must be
2583 absolute) that points to a file on the node's local disk. The node will
2584 read the contents of this file, upload it to the grid, then return the
2585 URI at which it was uploaded. If convergence is None then a random
2586 encryption key will be used, else the plaintext will be hashed, then
2587 that hash will be mixed together with the "convergence" string to form
2592 def download_from_uri_to_file(uri=URI, filename=str):
2593 """Download a file from the grid, placing it on the node's local disk
2594 at the given filename (which must be absolute[?]). Returns the
2595 absolute filename where the file was written."""
2600 def get_memory_usage():
2601 """Return a dict describes the amount of memory currently in use. The
2602 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2603 measuring memory consupmtion in bytes."""
2604 return DictOf(str, int)
2606 def speed_test(count=int, size=int, mutable=Any()):
2607 """Write 'count' tempfiles to disk, all of the given size. Measure
2608 how long (in seconds) it takes to upload them all to the servers.
2609 Then measure how long it takes to download all of them. If 'mutable'
2610 is 'create', time creation of mutable files. If 'mutable' is
2611 'upload', then time access to the same mutable file instead of
2614 Returns a tuple of (upload_time, download_time).
2616 return (float, float)
2618 def measure_peer_response_time():
2619 """Send a short message to each connected peer, and measure the time
2620 it takes for them to respond to it. This is a rough measure of the
2621 application-level round trip time.
2623 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2626 return DictOf(str, float)
2628 UploadResults = Any() #DictOf(str, str)
2630 class RIEncryptedUploadable(RemoteInterface):
2631 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2636 def get_all_encoding_parameters():
2637 return (int, int, int, long)
2639 def read_encrypted(offset=Offset, length=ReadSize):
2646 class RICHKUploadHelper(RemoteInterface):
2647 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2651 Return a dictionary of version information.
2653 return DictOf(str, Any())
2655 def upload(reader=RIEncryptedUploadable):
2656 return UploadResults
2659 class RIHelper(RemoteInterface):
2660 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2664 Return a dictionary of version information.
2666 return DictOf(str, Any())
2668 def upload_chk(si=StorageIndex):
2669 """See if a file with a given storage index needs uploading. The
2670 helper will ask the appropriate storage servers to see if the file
2671 has already been uploaded. If so, the helper will return a set of
2672 'upload results' that includes whatever hashes are needed to build
2673 the read-cap, and perhaps a truncated sharemap.
2675 If the file has not yet been uploaded (or if it was only partially
2676 uploaded), the helper will return an empty upload-results dictionary
2677 and also an RICHKUploadHelper object that will take care of the
2678 upload process. The client should call upload() on this object and
2679 pass it a reference to an RIEncryptedUploadable object that will
2680 provide ciphertext. When the upload is finished, the upload() method
2681 will finish and return the upload results.
2683 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2686 class RIStatsProvider(RemoteInterface):
2687 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2689 Provides access to statistics and monitoring information.
2694 returns a dictionary containing 'counters' and 'stats', each a
2695 dictionary with string counter/stat name keys, and numeric or None values.
2696 counters are monotonically increasing measures of work done, and
2697 stats are instantaneous measures (potentially time averaged
2700 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2702 class RIStatsGatherer(RemoteInterface):
2703 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2705 Provides a monitoring service for centralised collection of stats
2708 def provide(provider=RIStatsProvider, nickname=str):
2710 @param provider: a stats collector instance which should be polled
2711 periodically by the gatherer to collect stats.
2712 @param nickname: a name useful to identify the provided client
2717 class IStatsProducer(Interface):
2720 returns a dictionary, with str keys representing the names of stats
2721 to be monitored, and numeric values.
2724 class RIKeyGenerator(RemoteInterface):
2725 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2727 Provides a service offering to make RSA key pairs.
2730 def get_rsa_key_pair(key_size=int):
2732 @param key_size: the size of the signature key.
2733 @return: tuple(verifying_key, signing_key)
2735 return TupleOf(str, str)
2738 class FileTooLargeError(Exception):
2741 class IValidatedThingProxy(Interface):
2743 """ Acquire a thing and validate it. Return a deferred which is
2744 eventually fired with self if the thing is valid or errbacked if it
2745 can't be acquired or validated."""
2747 class InsufficientVersionError(Exception):
2748 def __init__(self, needed, got):
2749 self.needed = needed
2752 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2755 class EmptyPathnameComponentError(Exception):
2756 """The webapi disallows empty pathname components."""