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):
447 """Once a server is connected, I return a RemoteReference.
448 Before a server is connected for the first time, I return None.
450 Note that the rref I return will start producing DeadReferenceErrors
451 once the connection is lost.
455 class IMutableSlotWriter(Interface):
457 The interface for a writer around a mutable slot on a remote server.
459 def set_checkstring(checkstring, *args):
461 Set the checkstring that I will pass to the remote server when
464 @param checkstring A packed checkstring to use.
466 Note that implementations can differ in which semantics they
467 wish to support for set_checkstring -- they can, for example,
468 build the checkstring themselves from its constituents, or
472 def get_checkstring():
474 Get the checkstring that I think currently exists on the remote
478 def put_block(data, segnum, salt):
480 Add a block and salt to the share.
483 def put_encprivkey(encprivkey):
485 Add the encrypted private key to the share.
488 def put_blockhashes(blockhashes=list):
490 Add the block hash tree to the share.
493 def put_sharehashes(sharehashes=dict):
495 Add the share hash chain to the share.
500 Return the part of the share that needs to be signed.
503 def put_signature(signature):
505 Add the signature to the share.
508 def put_verification_key(verification_key):
510 Add the verification key to the share.
513 def finish_publishing():
515 Do anything necessary to finish writing the share to a remote
516 server. I require that no further publishing needs to take place
517 after this method has been called.
521 class IURI(Interface):
522 def init_from_string(uri):
523 """Accept a string (as created by my to_string() method) and populate
524 this instance with its data. I am not normally called directly,
525 please use the module-level uri.from_string() function to convert
526 arbitrary URI strings into IURI-providing instances."""
529 """Return False if this URI be used to modify the data. Return True
530 if this URI cannot be used to modify the data."""
533 """Return True if the data can be modified by *somebody* (perhaps
534 someone who has a more powerful URI than this one)."""
536 # TODO: rename to get_read_cap()
538 """Return another IURI instance, which represents a read-only form of
539 this one. If is_readonly() is True, this returns self."""
541 def get_verify_cap():
542 """Return an instance that provides IVerifierURI, which can be used
543 to check on the availability of the file or directory, without
544 providing enough capabilities to actually read or modify the
545 contents. This may return None if the file does not need checking or
546 verification (e.g. LIT URIs).
550 """Return a string of printable ASCII characters, suitable for
551 passing into init_from_string."""
553 class IVerifierURI(Interface, IURI):
554 def init_from_string(uri):
555 """Accept a string (as created by my to_string() method) and populate
556 this instance with its data. I am not normally called directly,
557 please use the module-level uri.from_string() function to convert
558 arbitrary URI strings into IURI-providing instances."""
561 """Return a string of printable ASCII characters, suitable for
562 passing into init_from_string."""
564 class IDirnodeURI(Interface):
565 """I am a URI which represents a dirnode."""
567 class IFileURI(Interface):
568 """I am a URI which represents a filenode."""
570 """Return the length (in bytes) of the file that I represent."""
572 class IImmutableFileURI(IFileURI):
575 class IMutableFileURI(Interface):
578 class IDirectoryURI(Interface):
581 class IReadonlyDirectoryURI(Interface):
584 class CapConstraintError(Exception):
585 """A constraint on a cap was violated."""
587 class MustBeDeepImmutableError(CapConstraintError):
588 """Mutable children cannot be added to an immutable directory.
589 Also, caps obtained from an immutable directory can trigger this error
590 if they are later found to refer to a mutable object and then used."""
592 class MustBeReadonlyError(CapConstraintError):
593 """Known write caps cannot be specified in a ro_uri field. Also,
594 caps obtained from a ro_uri field can trigger this error if they
595 are later found to be write caps and then used."""
597 class MustNotBeUnknownRWError(CapConstraintError):
598 """Cannot add an unknown child cap specified in a rw_uri field."""
601 class IReadable(Interface):
602 """I represent a readable object -- either an immutable file, or a
603 specific version of a mutable file.
607 """Return True if this reference provides mutable access to the given
608 file or directory (i.e. if you can modify it), or False if not. Note
609 that even if this reference is read-only, someone else may hold a
610 read-write reference to it.
612 For an IReadable returned by get_best_readable_version(), this will
613 always return True, but for instances of subinterfaces such as
614 IMutableFileVersion, it may return False."""
617 """Return True if this file or directory is mutable (by *somebody*,
618 not necessarily you), False if it is is immutable. Note that a file
619 might be mutable overall, but your reference to it might be
620 read-only. On the other hand, all references to an immutable file
621 will be read-only; there are no read-write references to an immutable
624 def get_storage_index():
625 """Return the storage index of the file."""
628 """Return the length (in bytes) of this readable object."""
630 def download_to_data():
631 """Download all of the file contents. I return a Deferred that fires
632 with the contents as a byte string."""
634 def read(consumer, offset=0, size=None):
635 """Download a portion (possibly all) of the file's contents, making
636 them available to the given IConsumer. Return a Deferred that fires
637 (with the consumer) when the consumer is unregistered (either because
638 the last byte has been given to it, or because the consumer threw an
639 exception during write(), possibly because it no longer wants to
640 receive data). The portion downloaded will start at 'offset' and
641 contain 'size' bytes (or the remainder of the file if size==None).
643 The consumer will be used in non-streaming mode: an IPullProducer
644 will be attached to it.
646 The consumer will not receive data right away: several network trips
647 must occur first. The order of events will be::
649 consumer.registerProducer(p, streaming)
650 (if streaming == False)::
651 consumer does p.resumeProducing()
653 consumer does p.resumeProducing()
654 consumer.write(data).. (repeat until all data is written)
655 consumer.unregisterProducer()
656 deferred.callback(consumer)
658 If a download error occurs, or an exception is raised by
659 consumer.registerProducer() or consumer.write(), I will call
660 consumer.unregisterProducer() and then deliver the exception via
661 deferred.errback(). To cancel the download, the consumer should call
662 p.stopProducing(), which will result in an exception being delivered
663 via deferred.errback().
665 See src/allmydata/util/consumer.py for an example of a simple
666 download-to-memory consumer.
670 class IWriteable(Interface):
672 I define methods that callers can use to update SDMF and MDMF
673 mutable files on a Tahoe-LAFS grid.
675 # XXX: For the moment, we have only this. It is possible that we
676 # want to move overwrite() and modify() in here too.
677 def update(data, offset):
679 I write the data from my data argument to the MDMF file,
680 starting at offset. I continue writing data until my data
681 argument is exhausted, appending data to the file as necessary.
683 # assert IMutableUploadable.providedBy(data)
684 # to append data: offset=node.get_size_of_best_version()
685 # do we want to support compacting MDMF?
686 # for an MDMF file, this can be done with O(data.get_size())
687 # memory. For an SDMF file, any modification takes
688 # O(node.get_size_of_best_version()).
691 class IMutableFileVersion(IReadable):
692 """I provide access to a particular version of a mutable file. The
693 access is read/write if I was obtained from a filenode derived from
694 a write cap, or read-only if the filenode was derived from a read cap.
697 def get_sequence_number():
698 """Return the sequence number of this version."""
701 """Return the IMutableFileServerMap instance that was used to create
706 """Return this filenode's writekey, or None if the node does not have
707 write-capability. This may be used to assist with data structures
708 that need to make certain data available only to writers, such as the
709 read-write child caps in dirnodes. The recommended process is to have
710 reader-visible data be submitted to the filenode in the clear (where
711 it will be encrypted by the filenode using the readkey), but encrypt
712 writer-visible data using this writekey.
715 # TODO: Can this be overwrite instead of replace?
716 def replace(new_contents):
717 """Replace the contents of the mutable file, provided that no other
718 node has published (or is attempting to publish, concurrently) a
719 newer version of the file than this one.
721 I will avoid modifying any share that is different than the version
722 given by get_sequence_number(). However, if another node is writing
723 to the file at the same time as me, I may manage to update some shares
724 while they update others. If I see any evidence of this, I will signal
725 UncoordinatedWriteError, and the file will be left in an inconsistent
726 state (possibly the version you provided, possibly the old version,
727 possibly somebody else's version, and possibly a mix of shares from
730 The recommended response to UncoordinatedWriteError is to either
731 return it to the caller (since they failed to coordinate their
732 writes), or to attempt some sort of recovery. It may be sufficient to
733 wait a random interval (with exponential backoff) and repeat your
734 operation. If I do not signal UncoordinatedWriteError, then I was
735 able to write the new version without incident.
737 I return a Deferred that fires (with a PublishStatus object) when the
738 update has completed.
741 def modify(modifier_cb):
742 """Modify the contents of the file, by downloading this version,
743 applying the modifier function (or bound method), then uploading
744 the new version. This will succeed as long as no other node
745 publishes a version between the download and the upload.
746 I return a Deferred that fires (with a PublishStatus object) when
747 the update is complete.
749 The modifier callable will be given three arguments: a string (with
750 the old contents), a 'first_time' boolean, and a servermap. As with
751 download_to_data(), the old contents will be from this version,
752 but the modifier can use the servermap to make other decisions
753 (such as refusing to apply the delta if there are multiple parallel
754 versions, or if there is evidence of a newer unrecoverable version).
755 'first_time' will be True the first time the modifier is called,
756 and False on any subsequent calls.
758 The callable should return a string with the new contents. The
759 callable must be prepared to be called multiple times, and must
760 examine the input string to see if the change that it wants to make
761 is already present in the old version. If it does not need to make
762 any changes, it can either return None, or return its input string.
764 If the modifier raises an exception, it will be returned in the
769 # The hierarchy looks like this:
776 class IFilesystemNode(Interface):
778 """Return the strongest 'cap instance' associated with this node.
779 (writecap for writeable-mutable files/directories, readcap for
780 immutable or readonly-mutable files/directories). To convert this
781 into a string, call .to_string() on the result."""
784 """Return a readonly cap instance for this node. For immutable or
785 readonly nodes, get_cap() and get_readcap() return the same thing."""
787 def get_repair_cap():
788 """Return an IURI instance that can be used to repair the file, or
789 None if this node cannot be repaired (either because it is not
790 distributed, like a LIT file, or because the node does not represent
791 sufficient authority to create a repair-cap, like a read-only RSA
792 mutable file node [which cannot create the correct write-enablers]).
795 def get_verify_cap():
796 """Return an IVerifierURI instance that represents the
797 'verifiy/refresh capability' for this node. The holder of this
798 capability will be able to renew the lease for this node, protecting
799 it from garbage-collection. They will also be able to ask a server if
800 it holds a share for the file or directory.
804 """Return the URI string corresponding to the strongest cap associated
805 with this node. If this node is read-only, the URI will only offer
806 read-only access. If this node is read-write, the URI will offer
809 If you have read-write access to a node and wish to share merely
810 read-only access with others, use get_readonly_uri().
814 """Return the URI string that can be used by others to get write
815 access to this node, if it is writeable. If this is a read-only node,
818 def get_readonly_uri():
819 """Return the URI string that can be used by others to get read-only
820 access to this node. The result is a read-only URI, regardless of
821 whether this node is read-only or read-write.
823 If you have merely read-only access to this node, get_readonly_uri()
824 will return the same thing as get_uri().
827 def get_storage_index():
828 """Return a string with the (binary) storage index in use on this
829 download. This may be None if there is no storage index (i.e. LIT
830 files and directories)."""
833 """Return True if this reference provides mutable access to the given
834 file or directory (i.e. if you can modify it), or False if not. Note
835 that even if this reference is read-only, someone else may hold a
836 read-write reference to it."""
839 """Return True if this file or directory is mutable (by *somebody*,
840 not necessarily you), False if it is is immutable. Note that a file
841 might be mutable overall, but your reference to it might be
842 read-only. On the other hand, all references to an immutable file
843 will be read-only; there are no read-write references to an immutable
848 """Return True if this is an unknown node."""
850 def is_allowed_in_immutable_directory():
851 """Return True if this node is allowed as a child of a deep-immutable
852 directory. This is true if either the node is of a known-immutable type,
853 or it is unknown and read-only.
857 """Raise any error associated with this node."""
859 # XXX: These may not be appropriate outside the context of an IReadable.
861 """Return the length (in bytes) of the data this node represents. For
862 directory nodes, I return the size of the backing store. I return
863 synchronously and do not consult the network, so for mutable objects,
864 I will return the most recently observed size for the object, or None
865 if I don't remember a size. Use get_current_size, which returns a
866 Deferred, if you want more up-to-date information."""
868 def get_current_size():
869 """I return a Deferred that fires with the length (in bytes) of the
870 data this node represents.
873 class IFileNode(IFilesystemNode):
874 """I am a node which represents a file: a sequence of bytes. I am not a
875 container, like IDirectoryNode."""
876 def get_best_readable_version():
877 """Return a Deferred that fires with an IReadable for the 'best'
878 available version of the file. The IReadable provides only read
879 access, even if this filenode was derived from a write cap.
881 For an immutable file, there is only one version. For a mutable
882 file, the 'best' version is the recoverable version with the
883 highest sequence number. If no uncoordinated writes have occurred,
884 and if enough shares are available, then this will be the most
885 recent version that has been uploaded. If no version is recoverable,
886 the Deferred will errback with an UnrecoverableFileError.
889 def download_best_version():
890 """Download the contents of the version that would be returned
891 by get_best_readable_version(). This is equivalent to calling
892 download_to_data() on the IReadable given by that method.
894 I return a Deferred that fires with a byte string when the file
895 has been fully downloaded. To support streaming download, use
896 the 'read' method of IReadable. If no version is recoverable,
897 the Deferred will errback with an UnrecoverableFileError.
900 def get_size_of_best_version():
901 """Find the size of the version that would be returned by
902 get_best_readable_version().
904 I return a Deferred that fires with an integer. If no version
905 is recoverable, the Deferred will errback with an
906 UnrecoverableFileError.
910 class IImmutableFileNode(IFileNode, IReadable):
911 """I am a node representing an immutable file. Immutable files have
915 class IMutableFileNode(IFileNode):
916 """I provide access to a 'mutable file', which retains its identity
917 regardless of what contents are put in it.
919 The consistency-vs-availability problem means that there might be
920 multiple versions of a file present in the grid, some of which might be
921 unrecoverable (i.e. have fewer than 'k' shares). These versions are
922 loosely ordered: each has a sequence number and a hash, and any version
923 with seqnum=N was uploaded by a node which has seen at least one version
926 The 'servermap' (an instance of IMutableFileServerMap) is used to
927 describe the versions that are known to be present in the grid, and which
928 servers are hosting their shares. It is used to represent the 'state of
929 the world', and is used for this purpose by my test-and-set operations.
930 Downloading the contents of the mutable file will also return a
931 servermap. Uploading a new version into the mutable file requires a
932 servermap as input, and the semantics of the replace operation is
933 'replace the file with my new version if it looks like nobody else has
934 changed the file since my previous download'. Because the file is
935 distributed, this is not a perfect test-and-set operation, but it will do
936 its best. If the replace process sees evidence of a simultaneous write,
937 it will signal an UncoordinatedWriteError, so that the caller can take
941 Most readers will want to use the 'best' current version of the file, and
942 should use my 'download_best_version()' method.
944 To unconditionally replace the file, callers should use overwrite(). This
945 is the mode that user-visible mutable files will probably use.
947 To apply some delta to the file, call modify() with a callable modifier
948 function that can apply the modification that you want to make. This is
949 the mode that dirnodes will use, since most directory modification
950 operations can be expressed in terms of deltas to the directory state.
953 Three methods are available for users who need to perform more complex
954 operations. The first is get_servermap(), which returns an up-to-date
955 servermap using a specified mode. The second is download_version(), which
956 downloads a specific version (not necessarily the 'best' one). The third
957 is 'upload', which accepts new contents and a servermap (which must have
958 been updated with MODE_WRITE). The upload method will attempt to apply
959 the new contents as long as no other node has modified the file since the
960 servermap was updated. This might be useful to a caller who wants to
961 merge multiple versions into a single new one.
963 Note that each time the servermap is updated, a specific 'mode' is used,
964 which determines how many peers are queried. To use a servermap for my
965 replace() method, that servermap must have been updated in MODE_WRITE.
966 These modes are defined in allmydata.mutable.common, and consist of
967 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
968 allmydata/mutable/servermap.py for details about the differences.
970 Mutable files are currently limited in size (about 3.5MB max) and can
971 only be retrieved and updated all-at-once, as a single big string. Future
972 versions of our mutable files will remove this restriction.
974 def get_best_mutable_version():
975 """Return a Deferred that fires with an IMutableFileVersion for
976 the 'best' available version of the file. The best version is
977 the recoverable version with the highest sequence number. If no
978 uncoordinated writes have occurred, and if enough shares are
979 available, then this will be the most recent version that has
982 If no version is recoverable, the Deferred will errback with an
983 UnrecoverableFileError.
986 def overwrite(new_contents):
987 """Unconditionally replace the contents of the mutable file with new
988 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
989 is only appropriate to use when the new contents of the file are
990 completely unrelated to the old ones, and you do not care about other
993 I return a Deferred that fires (with a PublishStatus object) when the
994 update has completed.
997 def modify(modifier_cb):
998 """Modify the contents of the file, by downloading the current
999 version, applying the modifier function (or bound method), then
1000 uploading the new version. I return a Deferred that fires (with a
1001 PublishStatus object) when the update is complete.
1003 The modifier callable will be given three arguments: a string (with
1004 the old contents), a 'first_time' boolean, and a servermap. As with
1005 download_best_version(), the old contents will be from the best
1006 recoverable version, but the modifier can use the servermap to make
1007 other decisions (such as refusing to apply the delta if there are
1008 multiple parallel versions, or if there is evidence of a newer
1009 unrecoverable version). 'first_time' will be True the first time the
1010 modifier is called, and False on any subsequent calls.
1012 The callable should return a string with the new contents. The
1013 callable must be prepared to be called multiple times, and must
1014 examine the input string to see if the change that it wants to make
1015 is already present in the old version. If it does not need to make
1016 any changes, it can either return None, or return its input string.
1018 If the modifier raises an exception, it will be returned in the
1022 def get_servermap(mode):
1023 """Return a Deferred that fires with an IMutableFileServerMap
1024 instance, updated using the given mode.
1027 def download_version(servermap, version):
1028 """Download a specific version of the file, using the servermap
1029 as a guide to where the shares are located.
1031 I return a Deferred that fires with the requested contents, or
1032 errbacks with UnrecoverableFileError. Note that a servermap which was
1033 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1034 all versions (those modes stop querying servers as soon as they can
1035 fulfil their goals), so you may want to use MODE_CHECK (which checks
1036 everything) to get increased visibility.
1039 def upload(new_contents, servermap):
1040 """Replace the contents of the file with new ones. This requires a
1041 servermap that was previously updated with MODE_WRITE.
1043 I attempt to provide test-and-set semantics, in that I will avoid
1044 modifying any share that is different than the version I saw in the
1045 servermap. However, if another node is writing to the file at the
1046 same time as me, I may manage to update some shares while they update
1047 others. If I see any evidence of this, I will signal
1048 UncoordinatedWriteError, and the file will be left in an inconsistent
1049 state (possibly the version you provided, possibly the old version,
1050 possibly somebody else's version, and possibly a mix of shares from
1053 The recommended response to UncoordinatedWriteError is to either
1054 return it to the caller (since they failed to coordinate their
1055 writes), or to attempt some sort of recovery. It may be sufficient to
1056 wait a random interval (with exponential backoff) and repeat your
1057 operation. If I do not signal UncoordinatedWriteError, then I was
1058 able to write the new version without incident.
1060 I return a Deferred that fires (with a PublishStatus object) when the
1061 publish has completed. I will update the servermap in-place with the
1062 location of all new shares.
1066 """Return this filenode's writekey, or None if the node does not have
1067 write-capability. This may be used to assist with data structures
1068 that need to make certain data available only to writers, such as the
1069 read-write child caps in dirnodes. The recommended process is to have
1070 reader-visible data be submitted to the filenode in the clear (where
1071 it will be encrypted by the filenode using the readkey), but encrypt
1072 writer-visible data using this writekey.
1076 """Returns the mutable file protocol version."""
1078 class NotEnoughSharesError(Exception):
1079 """Download was unable to get enough shares"""
1081 class NoSharesError(Exception):
1082 """Download was unable to get any shares at all."""
1084 class DownloadStopped(Exception):
1087 class UploadUnhappinessError(Exception):
1088 """Upload was unable to satisfy 'servers_of_happiness'"""
1090 class UnableToFetchCriticalDownloadDataError(Exception):
1091 """I was unable to fetch some piece of critical data which is supposed to
1092 be identically present in all shares."""
1094 class NoServersError(Exception):
1095 """Upload wasn't given any servers to work with, usually indicating a
1096 network or Introducer problem."""
1098 class ExistingChildError(Exception):
1099 """A directory node was asked to add or replace a child that already
1100 exists, and overwrite= was set to False."""
1102 class NoSuchChildError(Exception):
1103 """A directory node was asked to fetch a child which does not exist."""
1105 # avoid UnicodeEncodeErrors when converting to str
1106 return self.__repr__()
1108 class ChildOfWrongTypeError(Exception):
1109 """An operation was attempted on a child of the wrong type (file or directory)."""
1111 class IDirectoryNode(IFilesystemNode):
1112 """I represent a filesystem node that is a container, with a
1113 name-to-child mapping, holding the tahoe equivalent of a directory. All
1114 child names are unicode strings, and all children are some sort of
1115 IFilesystemNode (a file, subdirectory, or unknown node).
1120 The dirnode ('1') URI returned by this method can be used in
1121 set_uri() on a different directory ('2') to 'mount' a reference to
1122 this directory ('1') under the other ('2'). This URI is just a
1123 string, so it can be passed around through email or other out-of-band
1127 def get_readonly_uri():
1129 The dirnode ('1') URI returned by this method can be used in
1130 set_uri() on a different directory ('2') to 'mount' a reference to
1131 this directory ('1') under the other ('2'). This URI is just a
1132 string, so it can be passed around through email or other out-of-band
1137 """I return a Deferred that fires with a dictionary mapping child
1138 name (a unicode string) to (node, metadata_dict) tuples, in which
1139 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1142 def has_child(name):
1143 """I return a Deferred that fires with a boolean, True if there
1144 exists a child of the given name, False if not. The child name must
1145 be a unicode string."""
1148 """I return a Deferred that fires with a specific named child node,
1149 which is an IFilesystemNode. The child name must be a unicode string.
1150 I raise NoSuchChildError if I do not have a child by that name."""
1152 def get_metadata_for(name):
1153 """I return a Deferred that fires with the metadata dictionary for
1154 a specific named child node. The child name must be a unicode string.
1155 This metadata is stored in the *edge*, not in the child, so it is
1156 attached to the parent dirnode rather than the child node.
1157 I raise NoSuchChildError if I do not have a child by that name."""
1159 def set_metadata_for(name, metadata):
1160 """I replace any existing metadata for the named child with the new
1161 metadata. The child name must be a unicode string. This metadata is
1162 stored in the *edge*, not in the child, so it is attached to the
1163 parent dirnode rather than the child node. I return a Deferred
1164 (that fires with this dirnode) when the operation is complete.
1165 I raise NoSuchChildError if I do not have a child by that name."""
1167 def get_child_at_path(path):
1168 """Transform a child path into an IFilesystemNode.
1170 I perform a recursive series of 'get' operations to find the named
1171 descendant node. I return a Deferred that fires with the node, or
1172 errbacks with NoSuchChildError if the node could not be found.
1174 The path can be either a single string (slash-separated) or a list of
1175 path-name elements. All elements must be unicode strings.
1178 def get_child_and_metadata_at_path(path):
1179 """Transform a child path into an IFilesystemNode and metadata.
1181 I am like get_child_at_path(), but my Deferred fires with a tuple of
1182 (node, metadata). The metadata comes from the last edge. If the path
1183 is empty, the metadata will be an empty dictionary.
1186 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1187 """I add a child (by writecap+readcap) at the specific name. I return
1188 a Deferred that fires when the operation finishes. If overwrite= is
1189 True, I will replace any existing child of the same name, otherwise
1190 an existing child will cause me to return ExistingChildError. The
1191 child name must be a unicode string.
1193 The child caps could be for a file, or for a directory. If you have
1194 both the writecap and readcap, you should provide both arguments.
1195 If you have only one cap and don't know whether it is read-only,
1196 provide it as the writecap argument and leave the readcap as None.
1197 If you have only one cap that is known to be read-only, provide it
1198 as the readcap argument and leave the writecap as None.
1199 The filecaps are typically obtained from an IFilesystemNode with
1200 get_uri() and get_readonly_uri().
1202 If metadata= is provided, I will use it as the metadata for the named
1203 edge. This will replace any existing metadata. If metadata= is left
1204 as the default value of None, I will set ['mtime'] to the current
1205 time, and I will set ['ctime'] to the current time if there was not
1206 already a child by this name present. This roughly matches the
1207 ctime/mtime semantics of traditional filesystems. See the
1208 "About the metadata" section of webapi.txt for futher information.
1210 If this directory node is read-only, the Deferred will errback with a
1211 NotWriteableError."""
1213 def set_children(entries, overwrite=True):
1214 """Add multiple children (by writecap+readcap) to a directory node.
1215 Takes a dictionary, with childname as keys and (writecap, readcap)
1216 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1217 a Deferred that fires (with this dirnode) when the operation
1218 finishes. This is equivalent to calling set_uri() multiple times, but
1219 is much more efficient. All child names must be unicode strings.
1222 def set_node(name, child, metadata=None, overwrite=True):
1223 """I add a child at the specific name. I return a Deferred that fires
1224 when the operation finishes. This Deferred will fire with the child
1225 node that was just added. I will replace any existing child of the
1226 same name. The child name must be a unicode string. The 'child'
1227 instance must be an instance providing IFilesystemNode.
1229 If metadata= is provided, I will use it as the metadata for the named
1230 edge. This will replace any existing metadata. If metadata= is left
1231 as the default value of None, I will set ['mtime'] to the current
1232 time, and I will set ['ctime'] to the current time if there was not
1233 already a child by this name present. This roughly matches the
1234 ctime/mtime semantics of traditional filesystems. See the
1235 "About the metadata" section of webapi.txt for futher information.
1237 If this directory node is read-only, the Deferred will errback with a
1238 NotWriteableError."""
1240 def set_nodes(entries, overwrite=True):
1241 """Add multiple children to a directory node. Takes a dict mapping
1242 unicode childname to (child_node, metdata) tuples. If metdata=None,
1243 the original metadata is left unmodified. Returns a Deferred that
1244 fires (with this dirnode) when the operation finishes. This is
1245 equivalent to calling set_node() multiple times, but is much more
1248 def add_file(name, uploadable, metadata=None, overwrite=True):
1249 """I upload a file (using the given IUploadable), then attach the
1250 resulting ImmutableFileNode to the directory at the given name. I set
1251 metadata the same way as set_uri and set_node. The child name must be
1254 I return a Deferred that fires (with the IFileNode of the uploaded
1255 file) when the operation completes."""
1257 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1258 """I remove the child at the specific name. I return a Deferred that
1259 fires when the operation finishes. The child name must be a unicode
1260 string. If must_exist is True and I do not have a child by that name,
1261 I raise NoSuchChildError. If must_be_directory is True and the child
1262 is a file, or if must_be_file is True and the child is a directory,
1263 I raise ChildOfWrongTypeError."""
1265 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1266 """I create and attach a directory at the given name. The new
1267 directory can be empty, or it can be populated with children
1268 according to 'initial_children', which takes a dictionary in the same
1269 format as set_nodes (i.e. mapping unicode child name to (childnode,
1270 metadata) tuples). The child name must be a unicode string. I return
1271 a Deferred that fires (with the new directory node) when the
1272 operation finishes."""
1274 def move_child_to(current_child_name, new_parent, new_child_name=None,
1276 """I take one of my children and move them to a new parent. The child
1277 is referenced by name. On the new parent, the child will live under
1278 'new_child_name', which defaults to 'current_child_name'. TODO: what
1279 should we do about metadata? I return a Deferred that fires when the
1280 operation finishes. The child name must be a unicode string. I raise
1281 NoSuchChildError if I do not have a child by that name."""
1283 def build_manifest():
1284 """I generate a table of everything reachable from this directory.
1285 I also compute deep-stats as described below.
1287 I return a Monitor. The Monitor's results will be a dictionary with
1290 res['manifest']: a list of (path, cap) tuples for all nodes
1291 (directories and files) reachable from this one.
1292 'path' will be a tuple of unicode strings. The
1293 origin dirnode will be represented by an empty path
1295 res['verifycaps']: a list of (printable) verifycap strings, one for
1296 each reachable non-LIT node. This is a set:
1297 it will contain no duplicates.
1298 res['storage-index']: a list of (base32) storage index strings,
1299 one for each reachable non-LIT node. This is
1300 a set: it will contain no duplicates.
1301 res['stats']: a dictionary, the same that is generated by
1302 start_deep_stats() below.
1304 The Monitor will also have an .origin_si attribute with the (binary)
1305 storage index of the starting point.
1308 def start_deep_stats():
1309 """Return a Monitor, examining all nodes (directories and files)
1310 reachable from this one. The Monitor's results will be a dictionary
1311 with the following keys::
1313 count-immutable-files: count of how many CHK files are in the set
1314 count-mutable-files: same, for mutable files (does not include
1316 count-literal-files: same, for LIT files
1317 count-files: sum of the above three
1319 count-directories: count of directories
1321 size-immutable-files: total bytes for all CHK files in the set
1322 size-mutable-files (TODO): same, for current version of all mutable
1323 files, does not include directories
1324 size-literal-files: same, for LIT files
1325 size-directories: size of mutable files used by directories
1327 largest-directory: number of bytes in the largest directory
1328 largest-directory-children: number of children in the largest
1330 largest-immutable-file: number of bytes in the largest CHK file
1332 size-mutable-files is not yet implemented, because it would involve
1333 even more queries than deep_stats does.
1335 The Monitor will also have an .origin_si attribute with the (binary)
1336 storage index of the starting point.
1338 This operation will visit every directory node underneath this one,
1339 and can take a long time to run. On a typical workstation with good
1340 bandwidth, this can examine roughly 15 directories per second (and
1341 takes several minutes of 100% CPU for ~1700 directories).
1344 class ICodecEncoder(Interface):
1345 def set_params(data_size, required_shares, max_shares):
1346 """Set up the parameters of this encoder.
1348 This prepares the encoder to perform an operation that converts a
1349 single block of data into a number of shares, such that a future
1350 ICodecDecoder can use a subset of these shares to recover the
1351 original data. This operation is invoked by calling encode(). Once
1352 the encoding parameters are set up, the encode operation can be
1353 invoked multiple times.
1355 set_params() prepares the encoder to accept blocks of input data that
1356 are exactly 'data_size' bytes in length. The encoder will be prepared
1357 to produce 'max_shares' shares for each encode() operation (although
1358 see the 'desired_share_ids' to use less CPU). The encoding math will
1359 be chosen such that the decoder can get by with as few as
1360 'required_shares' of these shares and still reproduce the original
1361 data. For example, set_params(1000, 5, 5) offers no redundancy at
1362 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1364 Numerical Restrictions: 'data_size' is required to be an integral
1365 multiple of 'required_shares'. In general, the caller should choose
1366 required_shares and max_shares based upon their reliability
1367 requirements and the number of peers available (the total storage
1368 space used is roughly equal to max_shares*data_size/required_shares),
1369 then choose data_size to achieve the memory footprint desired (larger
1370 data_size means more efficient operation, smaller data_size means
1371 smaller memory footprint).
1373 In addition, 'max_shares' must be equal to or greater than
1374 'required_shares'. Of course, setting them to be equal causes
1375 encode() to degenerate into a particularly slow form of the 'split'
1378 See encode() for more details about how these parameters are used.
1380 set_params() must be called before any other ICodecEncoder methods
1385 """Return the 3-tuple of data_size, required_shares, max_shares"""
1387 def get_encoder_type():
1388 """Return a short string that describes the type of this encoder.
1390 There is required to be a global table of encoder classes. This method
1391 returns an index into this table; the value at this index is an
1392 encoder class, and this encoder is an instance of that class.
1395 def get_block_size():
1396 """Return the length of the shares that encode() will produce.
1399 def encode_proposal(data, desired_share_ids=None):
1400 """Encode some data.
1402 'data' must be a string (or other buffer object), and len(data) must
1403 be equal to the 'data_size' value passed earlier to set_params().
1405 This will return a Deferred that will fire with two lists. The first
1406 is a list of shares, each of which is a string (or other buffer
1407 object) such that len(share) is the same as what get_share_size()
1408 returned earlier. The second is a list of shareids, in which each is
1409 an integer. The lengths of the two lists will always be equal to each
1410 other. The user should take care to keep each share closely
1411 associated with its shareid, as one is useless without the other.
1413 The length of this output list will normally be the same as the value
1414 provided to the 'max_shares' parameter of set_params(). This may be
1415 different if 'desired_share_ids' is provided.
1417 'desired_share_ids', if provided, is required to be a sequence of
1418 ints, each of which is required to be >= 0 and < max_shares. If not
1419 provided, encode() will produce 'max_shares' shares, as if
1420 'desired_share_ids' were set to range(max_shares). You might use this
1421 if you initially thought you were going to use 10 peers, started
1422 encoding, and then two of the peers dropped out: you could use
1423 desired_share_ids= to skip the work (both memory and CPU) of
1424 producing shares for the peers which are no longer available.
1428 def encode(inshares, desired_share_ids=None):
1429 """Encode some data. This may be called multiple times. Each call is
1432 inshares is a sequence of length required_shares, containing buffers
1433 (i.e. strings), where each buffer contains the next contiguous
1434 non-overlapping segment of the input data. Each buffer is required to
1435 be the same length, and the sum of the lengths of the buffers is
1436 required to be exactly the data_size promised by set_params(). (This
1437 implies that the data has to be padded before being passed to
1438 encode(), unless of course it already happens to be an even multiple
1439 of required_shares in length.)
1441 Note: the requirement to break up your data into
1442 'required_shares' chunks of exactly the right length before
1443 calling encode() is surprising from point of view of a user
1444 who doesn't know how FEC works. It feels like an
1445 implementation detail that has leaked outside the abstraction
1446 barrier. Is there a use case in which the data to be encoded
1447 might already be available in pre-segmented chunks, such that
1448 it is faster or less work to make encode() take a list rather
1449 than splitting a single string?
1451 Yes, there is: suppose you are uploading a file with K=64,
1452 N=128, segsize=262,144. Then each in-share will be of size
1453 4096. If you use this .encode() API then your code could first
1454 read each successive 4096-byte chunk from the file and store
1455 each one in a Python string and store each such Python string
1456 in a Python list. Then you could call .encode(), passing that
1457 list as "inshares". The encoder would generate the other 64
1458 "secondary shares" and return to you a new list containing
1459 references to the same 64 Python strings that you passed in
1460 (as the primary shares) plus references to the new 64 Python
1463 (You could even imagine that your code could use readv() so
1464 that the operating system can arrange to get all of those
1465 bytes copied from the file into the Python list of Python
1466 strings as efficiently as possible instead of having a loop
1467 written in C or in Python to copy the next part of the file
1468 into the next string.)
1470 On the other hand if you instead use the .encode_proposal()
1471 API (above), then your code can first read in all of the
1472 262,144 bytes of the segment from the file into a Python
1473 string, then call .encode_proposal() passing the segment data
1474 as the "data" argument. The encoder would basically first
1475 split the "data" argument into a list of 64 in-shares of 4096
1476 byte each, and then do the same thing that .encode() does. So
1477 this would result in a little bit more copying of data and a
1478 little bit higher of a "maximum memory usage" during the
1479 process, although it might or might not make a practical
1480 difference for our current use cases.
1482 Note that "inshares" is a strange name for the parameter if
1483 you think of the parameter as being just for feeding in data
1484 to the codec. It makes more sense if you think of the result
1485 of this encoding as being the set of shares from inshares plus
1486 an extra set of "secondary shares" (or "check shares"). It is
1487 a surprising name! If the API is going to be surprising then
1488 the name should be surprising. If we switch to
1489 encode_proposal() above then we should also switch to an
1492 'desired_share_ids', if provided, is required to be a sequence of
1493 ints, each of which is required to be >= 0 and < max_shares. If not
1494 provided, encode() will produce 'max_shares' shares, as if
1495 'desired_share_ids' were set to range(max_shares). You might use this
1496 if you initially thought you were going to use 10 peers, started
1497 encoding, and then two of the peers dropped out: you could use
1498 desired_share_ids= to skip the work (both memory and CPU) of
1499 producing shares for the peers which are no longer available.
1501 For each call, encode() will return a Deferred that fires with two
1502 lists, one containing shares and the other containing the shareids.
1503 The get_share_size() method can be used to determine the length of
1504 the share strings returned by encode(). Each shareid is a small
1505 integer, exactly as passed into 'desired_share_ids' (or
1506 range(max_shares), if desired_share_ids was not provided).
1508 The shares and their corresponding shareids are required to be kept
1509 together during storage and retrieval. Specifically, the share data is
1510 useless by itself: the decoder needs to be told which share is which
1511 by providing it with both the shareid and the actual share data.
1513 This function will allocate an amount of memory roughly equal to::
1515 (max_shares - required_shares) * get_share_size()
1517 When combined with the memory that the caller must allocate to
1518 provide the input data, this leads to a memory footprint roughly
1519 equal to the size of the resulting encoded shares (i.e. the expansion
1520 factor times the size of the input segment).
1525 # returning a list of (shareidN,shareN) tuples instead of a pair of
1526 # lists (shareids..,shares..). Brian thought the tuples would
1527 # encourage users to keep the share and shareid together throughout
1528 # later processing, Zooko pointed out that the code to iterate
1529 # through two lists is not really more complicated than using a list
1530 # of tuples and there's also a performance improvement
1532 # having 'data_size' not required to be an integral multiple of
1533 # 'required_shares'. Doing this would require encode() to perform
1534 # padding internally, and we'd prefer to have any padding be done
1535 # explicitly by the caller. Yes, it is an abstraction leak, but
1536 # hopefully not an onerous one.
1539 class ICodecDecoder(Interface):
1540 def set_params(data_size, required_shares, max_shares):
1541 """Set the params. They have to be exactly the same ones that were
1542 used for encoding."""
1544 def get_needed_shares():
1545 """Return the number of shares needed to reconstruct the data.
1546 set_params() is required to be called before this."""
1548 def decode(some_shares, their_shareids):
1549 """Decode a partial list of shares into data.
1551 'some_shares' is required to be a sequence of buffers of sharedata, a
1552 subset of the shares returned by ICodecEncode.encode(). Each share is
1553 required to be of the same length. The i'th element of their_shareids
1554 is required to be the shareid of the i'th buffer in some_shares.
1556 This returns a Deferred which fires with a sequence of buffers. This
1557 sequence will contain all of the segments of the original data, in
1558 order. The sum of the lengths of all of the buffers will be the
1559 'data_size' value passed into the original ICodecEncode.set_params()
1560 call. To get back the single original input block of data, use
1561 ''.join(output_buffers), or you may wish to simply write them in
1562 order to an output file.
1564 Note that some of the elements in the result sequence may be
1565 references to the elements of the some_shares input sequence. In
1566 particular, this means that if those share objects are mutable (e.g.
1567 arrays) and if they are changed, then both the input (the
1568 'some_shares' parameter) and the output (the value given when the
1569 deferred is triggered) will change.
1571 The length of 'some_shares' is required to be exactly the value of
1572 'required_shares' passed into the original ICodecEncode.set_params()
1576 class IEncoder(Interface):
1577 """I take an object that provides IEncryptedUploadable, which provides
1578 encrypted data, and a list of shareholders. I then encode, hash, and
1579 deliver shares to those shareholders. I will compute all the necessary
1580 Merkle hash trees that are necessary to validate the crypttext that
1581 eventually comes back from the shareholders. I provide the URI Extension
1582 Block Hash, and the encoding parameters, both of which must be included
1585 I do not choose shareholders, that is left to the IUploader. I must be
1586 given a dict of RemoteReferences to storage buckets that are ready and
1587 willing to receive data.
1591 """Specify the number of bytes that will be encoded. This must be
1592 peformed before get_serialized_params() can be called.
1594 def set_params(params):
1595 """Override the default encoding parameters. 'params' is a tuple of
1596 (k,d,n), where 'k' is the number of required shares, 'd' is the
1597 servers_of_happiness, and 'n' is the total number of shares that will
1600 Encoding parameters can be set in three ways. 1: The Encoder class
1601 provides defaults (3/7/10). 2: the Encoder can be constructed with
1602 an 'options' dictionary, in which the
1603 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1604 set_params((k,d,n)) can be called.
1606 If you intend to use set_params(), you must call it before
1607 get_share_size or get_param are called.
1610 def set_encrypted_uploadable(u):
1611 """Provide a source of encrypted upload data. 'u' must implement
1612 IEncryptedUploadable.
1614 When this is called, the IEncryptedUploadable will be queried for its
1615 length and the storage_index that should be used.
1617 This returns a Deferred that fires with this Encoder instance.
1619 This must be performed before start() can be called.
1622 def get_param(name):
1623 """Return an encoding parameter, by name.
1625 'storage_index': return a string with the (16-byte truncated SHA-256
1626 hash) storage index to which these shares should be
1629 'share_counts': return a tuple describing how many shares are used:
1630 (needed_shares, servers_of_happiness, total_shares)
1632 'num_segments': return an int with the number of segments that
1635 'segment_size': return an int with the size of each segment.
1637 'block_size': return the size of the individual blocks that will
1638 be delivered to a shareholder's put_block() method. By
1639 knowing this, the shareholder will be able to keep all
1640 blocks in a single file and still provide random access
1641 when reading them. # TODO: can we avoid exposing this?
1643 'share_size': an int with the size of the data that will be stored
1644 on each shareholder. This is aggregate amount of data
1645 that will be sent to the shareholder, summed over all
1646 the put_block() calls I will ever make. It is useful to
1647 determine this size before asking potential
1648 shareholders whether they will grant a lease or not,
1649 since their answers will depend upon how much space we
1650 need. TODO: this might also include some amount of
1651 overhead, like the size of all the hashes. We need to
1652 decide whether this is useful or not.
1654 'serialized_params': a string with a concise description of the
1655 codec name and its parameters. This may be passed
1656 into the IUploadable to let it make sure that
1657 the same file encoded with different parameters
1658 will result in different storage indexes.
1660 Once this is called, set_size() and set_params() may not be called.
1663 def set_shareholders(shareholders, servermap):
1664 """Tell the encoder where to put the encoded shares. 'shareholders'
1665 must be a dictionary that maps share number (an integer ranging from
1666 0 to n-1) to an instance that provides IStorageBucketWriter.
1667 'servermap' is a dictionary that maps share number (as defined above)
1668 to a set of peerids. This must be performed before start() can be
1672 """Begin the encode/upload process. This involves reading encrypted
1673 data from the IEncryptedUploadable, encoding it, uploading the shares
1674 to the shareholders, then sending the hash trees.
1676 set_encrypted_uploadable() and set_shareholders() must be called
1677 before this can be invoked.
1679 This returns a Deferred that fires with a verify cap when the upload
1680 process is complete. The verifycap, plus the encryption key, is
1681 sufficient to construct the read cap.
1684 class IDecoder(Interface):
1685 """I take a list of shareholders and some setup information, then
1686 download, validate, decode, and decrypt data from them, writing the
1687 results to an output file.
1689 I do not locate the shareholders, that is left to the IDownloader. I must
1690 be given a dict of RemoteReferences to storage buckets that are ready to
1695 """I take a file-like object (providing write and close) to which all
1696 the plaintext data will be written.
1698 TODO: producer/consumer . Maybe write() should return a Deferred that
1699 indicates when it will accept more data? But probably having the
1700 IDecoder be a producer is easier to glue to IConsumer pieces.
1703 def set_shareholders(shareholders):
1704 """I take a dictionary that maps share identifiers (small integers)
1705 to RemoteReferences that provide RIBucketReader. This must be called
1709 """I start the download. This process involves retrieving data and
1710 hash chains from the shareholders, using the hashes to validate the
1711 data, decoding the shares into segments, decrypting the segments,
1712 then writing the resulting plaintext to the output file.
1714 I return a Deferred that will fire (with self) when the download is
1718 class IDownloadTarget(Interface):
1719 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1720 # will register itself as a producer. This allows the target to invoke
1721 # downloader.pauseProducing, resumeProducing, and stopProducing.
1723 """Called before any calls to write() or close(). If an error
1724 occurs before any data is available, fail() may be called without
1725 a previous call to open().
1727 'size' is the length of the file being downloaded, in bytes."""
1730 """Output some data to the target."""
1732 """Inform the target that there is no more data to be written."""
1734 """fail() is called to indicate that the download has failed. 'why'
1735 is a Failure object indicating what went wrong. No further methods
1736 will be invoked on the IDownloadTarget after fail()."""
1737 def register_canceller(cb):
1738 """The CiphertextDownloader uses this to register a no-argument function
1739 that the target can call to cancel the download. Once this canceller
1740 is invoked, no further calls to write() or close() will be made."""
1742 """When the CiphertextDownloader is done, this finish() function will be
1743 called. Whatever it returns will be returned to the invoker of
1744 Downloader.download.
1747 class IDownloader(Interface):
1748 def download(uri, target):
1749 """Perform a CHK download, sending the data to the given target.
1750 'target' must provide IDownloadTarget.
1752 Returns a Deferred that fires (with the results of target.finish)
1753 when the download is finished, or errbacks if something went wrong."""
1755 class IEncryptedUploadable(Interface):
1756 def set_upload_status(upload_status):
1757 """Provide an IUploadStatus object that should be filled with status
1758 information. The IEncryptedUploadable is responsible for setting
1759 key-determination progress ('chk'), size, storage_index, and
1760 ciphertext-fetch progress. It may delegate some of this
1761 responsibility to others, in particular to the IUploadable."""
1764 """This behaves just like IUploadable.get_size()."""
1766 def get_all_encoding_parameters():
1767 """Return a Deferred that fires with a tuple of
1768 (k,happy,n,segment_size). The segment_size will be used as-is, and
1769 must match the following constraints: it must be a multiple of k, and
1770 it shouldn't be unreasonably larger than the file size (if
1771 segment_size is larger than filesize, the difference must be stored
1774 This usually passes through to the IUploadable method of the same
1777 The encoder strictly obeys the values returned by this method. To
1778 make an upload use non-default encoding parameters, you must arrange
1779 to control the values that this method returns.
1782 def get_storage_index():
1783 """Return a Deferred that fires with a 16-byte storage index.
1786 def read_encrypted(length, hash_only):
1787 """This behaves just like IUploadable.read(), but returns crypttext
1788 instead of plaintext. If hash_only is True, then this discards the
1789 data (and returns an empty list); this improves efficiency when
1790 resuming an interrupted upload (where we need to compute the
1791 plaintext hashes, but don't need the redundant encrypted data)."""
1793 def get_plaintext_hashtree_leaves(first, last, num_segments):
1794 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1795 plaintext segments, i.e. get the tagged hashes of the given segments.
1796 The segment size is expected to be generated by the
1797 IEncryptedUploadable before any plaintext is read or ciphertext
1798 produced, so that the segment hashes can be generated with only a
1801 This returns a Deferred which fires with a sequence of hashes, using:
1803 tuple(segment_hashes[first:last])
1805 'num_segments' is used to assert that the number of segments that the
1806 IEncryptedUploadable handled matches the number of segments that the
1807 encoder was expecting.
1809 This method must not be called until the final byte has been read
1810 from read_encrypted(). Once this method is called, read_encrypted()
1811 can never be called again.
1814 def get_plaintext_hash():
1815 """OBSOLETE; Get the hash of the whole plaintext.
1817 This returns a Deferred which fires with a tagged SHA-256 hash of the
1818 whole plaintext, obtained from hashutil.plaintext_hash(data).
1822 """Just like IUploadable.close()."""
1824 class IUploadable(Interface):
1825 def set_upload_status(upload_status):
1826 """Provide an IUploadStatus object that should be filled with status
1827 information. The IUploadable is responsible for setting
1828 key-determination progress ('chk')."""
1830 def set_default_encoding_parameters(params):
1831 """Set the default encoding parameters, which must be a dict mapping
1832 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1833 'max_segment_size'. These might have an influence on the final
1834 encoding parameters returned by get_all_encoding_parameters(), if the
1835 Uploadable doesn't have more specific preferences.
1837 This call is optional: if it is not used, the Uploadable will use
1838 some built-in defaults. If used, this method must be called before
1839 any other IUploadable methods to have any effect.
1843 """Return a Deferred that will fire with the length of the data to be
1844 uploaded, in bytes. This will be called before the data is actually
1845 used, to compute encoding parameters.
1848 def get_all_encoding_parameters():
1849 """Return a Deferred that fires with a tuple of
1850 (k,happy,n,segment_size). The segment_size will be used as-is, and
1851 must match the following constraints: it must be a multiple of k, and
1852 it shouldn't be unreasonably larger than the file size (if
1853 segment_size is larger than filesize, the difference must be stored
1856 The relative values of k and n allow some IUploadables to request
1857 better redundancy than others (in exchange for consuming more space
1860 Larger values of segment_size reduce hash overhead, while smaller
1861 values reduce memory footprint and cause data to be delivered in
1862 smaller pieces (which may provide a smoother and more predictable
1863 download experience).
1865 The encoder strictly obeys the values returned by this method. To
1866 make an upload use non-default encoding parameters, you must arrange
1867 to control the values that this method returns. One way to influence
1868 them may be to call set_encoding_parameters() before calling
1869 get_all_encoding_parameters().
1872 def get_encryption_key():
1873 """Return a Deferred that fires with a 16-byte AES key. This key will
1874 be used to encrypt the data. The key will also be hashed to derive
1877 Uploadables which want to achieve convergence should hash their file
1878 contents and the serialized_encoding_parameters to form the key
1879 (which of course requires a full pass over the data). Uploadables can
1880 use the upload.ConvergentUploadMixin class to achieve this
1883 Uploadables which do not care about convergence (or do not wish to
1884 make multiple passes over the data) can simply return a
1885 strongly-random 16 byte string.
1887 get_encryption_key() may be called multiple times: the IUploadable is
1888 required to return the same value each time.
1892 """Return a Deferred that fires with a list of strings (perhaps with
1893 only a single element) which, when concatenated together, contain the
1894 next 'length' bytes of data. If EOF is near, this may provide fewer
1895 than 'length' bytes. The total number of bytes provided by read()
1896 before it signals EOF must equal the size provided by get_size().
1898 If the data must be acquired through multiple internal read
1899 operations, returning a list instead of a single string may help to
1900 reduce string copies. However, the length of the concatenated strings
1901 must equal the amount of data requested, unless EOF is encountered.
1902 Long reads, or short reads without EOF, are not allowed. read()
1903 should return the same amount of data as a local disk file read, just
1904 in a different shape and asynchronously.
1906 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1907 so a 10kB file means length=3kB, 100kB file means length=30kB,
1908 and >=1MB file means length=300kB.
1910 This method provides for a single full pass through the data. Later
1911 use cases may desire multiple passes or access to only parts of the
1912 data (such as a mutable file making small edits-in-place). This API
1913 will be expanded once those use cases are better understood.
1917 """The upload is finished, and whatever filehandle was in use may be
1921 class IMutableUploadable(Interface):
1923 I represent content that is due to be uploaded to a mutable filecap.
1925 # This is somewhat simpler than the IUploadable interface above
1926 # because mutable files do not need to be concerned with possibly
1927 # generating a CHK, nor with per-file keys. It is a subset of the
1928 # methods in IUploadable, though, so we could just as well implement
1929 # the mutable uploadables as IUploadables that don't happen to use
1930 # those methods (with the understanding that the unused methods will
1931 # never be called on such objects)
1934 Returns a Deferred that fires with the size of the content held
1940 Returns a list of strings which, when concatenated, are the next
1941 length bytes of the file, or fewer if there are fewer bytes
1942 between the current location and the end of the file.
1947 The process that used the Uploadable is finished using it, so
1948 the uploadable may be closed.
1951 class IUploadResults(Interface):
1952 """I am returned by immutable upload() methods and contain the results of
1955 Note that some of my methods return empty values (0 or an empty dict)
1956 when called for non-distributed LIT files."""
1958 def get_file_size():
1959 """Return the file size, in bytes."""
1961 """Return the (string) URI of the object uploaded, a CHK readcap."""
1962 def get_ciphertext_fetched():
1963 """Return the number of bytes fetched by the helpe for this upload,
1964 or 0 if the helper did not need to fetch any bytes (or if there was
1966 def get_preexisting_shares():
1967 """Return the number of shares that were already present in the grid."""
1968 def get_pushed_shares():
1969 """Return the number of shares that were uploaded."""
1971 """Return a dict mapping share identifier to set of IServer
1972 instances. This indicates which servers were given which shares. For
1973 immutable files, the shareid is an integer (the share number, from 0
1974 to N-1). For mutable files, it is a string of the form
1975 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1976 the share number."""
1977 def get_servermap():
1978 """Return dict mapping IServer instance to a set of share numbers."""
1980 """Return dict of timing information, mapping name to seconds. All
1982 total : total upload time, start to finish
1983 storage_index : time to compute the storage index
1984 peer_selection : time to decide which peers will be used
1985 contacting_helper : initial helper query to upload/no-upload decision
1986 helper_total : initial helper query to helper finished pushing
1987 cumulative_fetch : helper waiting for ciphertext requests
1988 total_fetch : helper start to last ciphertext response
1989 cumulative_encoding : just time spent in zfec
1990 cumulative_sending : just time spent waiting for storage servers
1991 hashes_and_close : last segment push to shareholder close
1992 total_encode_and_push : first encode to shareholder close
1994 def get_uri_extension_data():
1995 """Return the dict of UEB data created for this file."""
1996 def get_verifycapstr():
1997 """Return the (string) verify-cap URI for the uploaded object."""
1999 class IDownloadResults(Interface):
2000 """I am created internally by download() methods. I contain a number of
2001 public attributes which contain details about the download process.::
2003 .file_size : the size of the file, in bytes
2004 .servers_used : set of server peerids that were used during download
2005 .server_problems : dict mapping server peerid to a problem string. Only
2006 servers that had problems (bad hashes, disconnects)
2008 .servermap : dict mapping server peerid to a set of share numbers. Only
2009 servers that had any shares are listed here.
2010 .timings : dict of timing information, mapping name to seconds (float)
2011 peer_selection : time to ask servers about shares
2012 servers_peer_selection : dict of peerid to DYHB-query time
2013 uri_extension : time to fetch a copy of the URI extension block
2014 hashtrees : time to fetch the hash trees
2015 segments : time to fetch, decode, and deliver segments
2016 cumulative_fetch : time spent waiting for storage servers
2017 cumulative_decode : just time spent in zfec
2018 cumulative_decrypt : just time spent in decryption
2019 total : total download time, start to finish
2020 fetch_per_server : dict of server to list of per-segment fetch times
2024 class IUploader(Interface):
2025 def upload(uploadable):
2026 """Upload the file. 'uploadable' must impement IUploadable. This
2027 returns a Deferred which fires with an IUploadResults instance, from
2028 which the URI of the file can be obtained as results.uri ."""
2030 def upload_ssk(write_capability, new_version, uploadable):
2031 """TODO: how should this work?"""
2033 class ICheckable(Interface):
2034 def check(monitor, verify=False, add_lease=False):
2035 """Check up on my health, optionally repairing any problems.
2037 This returns a Deferred that fires with an instance that provides
2038 ICheckResults, or None if the object is non-distributed (i.e. LIT
2041 The monitor will be checked periodically to see if the operation has
2042 been cancelled. If so, no new queries will be sent, and the Deferred
2043 will fire (with a OperationCancelledError) immediately.
2045 Filenodes and dirnodes (which provide IFilesystemNode) are also
2046 checkable. Instances that represent verifier-caps will be checkable
2047 but not downloadable. Some objects (like LIT files) do not actually
2048 live in the grid, and their checkers return None (non-distributed
2049 files are always healthy).
2051 If verify=False, a relatively lightweight check will be performed: I
2052 will ask all servers if they have a share for me, and I will believe
2053 whatever they say. If there are at least N distinct shares on the
2054 grid, my results will indicate r.is_healthy()==True. This requires a
2055 roundtrip to each server, but does not transfer very much data, so
2056 the network bandwidth is fairly low.
2058 If verify=True, a more resource-intensive check will be performed:
2059 every share will be downloaded, and the hashes will be validated on
2060 every bit. I will ignore any shares that failed their hash checks. If
2061 there are at least N distinct valid shares on the grid, my results
2062 will indicate r.is_healthy()==True. This requires N/k times as much
2063 download bandwidth (and server disk IO) as a regular download. If a
2064 storage server is holding a corrupt share, or is experiencing memory
2065 failures during retrieval, or is malicious or buggy, then
2066 verification will detect the problem, but checking will not.
2068 If add_lease=True, I will ensure that an up-to-date lease is present
2069 on each share. The lease secrets will be derived from by node secret
2070 (in BASEDIR/private/secret), so either I will add a new lease to the
2071 share, or I will merely renew the lease that I already had. In a
2072 future version of the storage-server protocol (once Accounting has
2073 been implemented), there may be additional options here to define the
2074 kind of lease that is obtained (which account number to claim, etc).
2076 TODO: any problems seen during checking will be reported to the
2077 health-manager.furl, a centralized object which is responsible for
2078 figuring out why files are unhealthy so corrective action can be
2082 def check_and_repair(monitor, verify=False, add_lease=False):
2083 """Like check(), but if the file/directory is not healthy, attempt to
2086 Any non-healthy result will cause an immediate repair operation, to
2087 generate and upload new shares. After repair, the file will be as
2088 healthy as we can make it. Details about what sort of repair is done
2089 will be put in the check-and-repair results. The Deferred will not
2090 fire until the repair is complete.
2092 This returns a Deferred which fires with an instance of
2093 ICheckAndRepairResults."""
2095 class IDeepCheckable(Interface):
2096 def start_deep_check(verify=False, add_lease=False):
2097 """Check upon the health of me and everything I can reach.
2099 This is a recursive form of check(), useable only on dirnodes.
2101 I return a Monitor, with results that are an IDeepCheckResults
2104 TODO: If any of the directories I traverse are unrecoverable, the
2105 Monitor will report failure. If any of the files I check upon are
2106 unrecoverable, those problems will be reported in the
2107 IDeepCheckResults as usual, and the Monitor will not report a
2111 def start_deep_check_and_repair(verify=False, add_lease=False):
2112 """Check upon the health of me and everything I can reach. Repair
2113 anything that isn't healthy.
2115 This is a recursive form of check_and_repair(), useable only on
2118 I return a Monitor, with results that are an
2119 IDeepCheckAndRepairResults object.
2121 TODO: If any of the directories I traverse are unrecoverable, the
2122 Monitor will report failure. If any of the files I check upon are
2123 unrecoverable, those problems will be reported in the
2124 IDeepCheckResults as usual, and the Monitor will not report a
2128 class ICheckResults(Interface):
2129 """I contain the detailed results of a check/verify operation.
2132 def get_storage_index():
2133 """Return a string with the (binary) storage index."""
2134 def get_storage_index_string():
2135 """Return a string with the (printable) abbreviated storage index."""
2137 """Return the (string) URI of the object that was checked."""
2140 """Return a boolean, True if the file/dir is fully healthy, False if
2141 it is damaged in any way. Non-distributed LIT files always return
2144 def is_recoverable():
2145 """Return a boolean, True if the file/dir can be recovered, False if
2146 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2147 files always return True."""
2149 def needs_rebalancing():
2150 """Return a boolean, True if the file/dirs reliability could be
2151 improved by moving shares to new servers. Non-distributed LIT files
2152 always return False."""
2154 # the following methods all return None for non-distributed LIT files
2156 def get_encoding_needed():
2157 """Return 'k', the number of shares required for recovery"""
2158 def get_encoding_expected():
2159 """Return 'N', the number of total shares generated"""
2161 def get_share_counter_good():
2162 """Return the number of distinct good shares that were found. For
2163 mutable files, this counts shares for the 'best' version."""
2164 def get_share_counter_wrong():
2165 """For mutable files, return the number of shares for versions other
2166 than the 'best' one (which is defined as being the recoverable
2167 version with the highest sequence number, then the highest roothash).
2168 These are either leftover shares from an older version (perhaps on a
2169 server that was offline when an update occurred), shares from an
2170 unrecoverable newer version, or shares from an alternate current
2171 version that results from an uncoordinated write collision. For a
2172 healthy file, this will equal 0. For immutable files, this will
2175 def get_corrupt_shares():
2176 """Return a list of 'share locators', one for each share that was
2177 found to be corrupt (integrity failure). Each share locator is a list
2178 of (IServer, storage_index, sharenum)."""
2180 def get_incompatible_shares():
2181 """Return a list of 'share locators', one for each share that was
2182 found to be of an unknown format. Each share locator is a list of
2183 (IServer, storage_index, sharenum)."""
2185 def get_servers_responding():
2186 """Return a list of IServer objects, one for each server which
2187 responded to the share query (even if they said they didn't have
2188 shares, and even if they said they did have shares but then didn't
2189 send them when asked, or dropped the connection, or returned a
2190 Failure, and even if they said they did have shares and sent
2191 incorrect ones when asked)"""
2193 def get_host_counter_good_shares():
2194 """Return the number of distinct storage servers with good shares. If
2195 this number is less than get_share_counters()[good], then some shares
2196 are doubled up, increasing the correlation of failures. This
2197 indicates that one or more shares should be moved to an otherwise
2198 unused server, if one is available.
2201 def get_version_counter_recoverable():
2202 """Return the number of recoverable versions of the file. For a
2203 healthy file, this will equal 1."""
2205 def get_version_counter_unrecoverable():
2206 """Return the number of unrecoverable versions of the file. For a
2207 healthy file, this will be 0."""
2210 """Return a dict mapping share identifier to list of IServer objects.
2211 This indicates which servers are holding which shares. For immutable
2212 files, the shareid is an integer (the share number, from 0 to N-1).
2213 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2214 containing the sequence number, the roothash, and the share number."""
2217 """Return a string with a brief (one-line) summary of the results."""
2220 """Return a list of strings with more detailed results."""
2222 class ICheckAndRepairResults(Interface):
2223 """I contain the detailed results of a check/verify/repair operation.
2225 The IFilesystemNode.check()/verify()/repair() methods all return
2226 instances that provide ICheckAndRepairResults.
2229 def get_storage_index():
2230 """Return a string with the (binary) storage index."""
2231 def get_storage_index_string():
2232 """Return a string with the (printable) abbreviated storage index."""
2233 def get_repair_attempted():
2234 """Return a boolean, True if a repair was attempted. We might not
2235 attempt to repair the file because it was healthy, or healthy enough
2236 (i.e. some shares were missing but not enough to exceed some
2237 threshold), or because we don't know how to repair this object."""
2238 def get_repair_successful():
2239 """Return a boolean, True if repair was attempted and the file/dir
2240 was fully healthy afterwards. False if no repair was attempted or if
2241 a repair attempt failed."""
2242 def get_pre_repair_results():
2243 """Return an ICheckResults instance that describes the state of the
2244 file/dir before any repair was attempted."""
2245 def get_post_repair_results():
2246 """Return an ICheckResults instance that describes the state of the
2247 file/dir after any repair was attempted. If no repair was attempted,
2248 the pre-repair and post-repair results will be identical."""
2251 class IDeepCheckResults(Interface):
2252 """I contain the results of a deep-check operation.
2254 This is returned by a call to ICheckable.deep_check().
2257 def get_root_storage_index_string():
2258 """Return the storage index (abbreviated human-readable string) of
2259 the first object checked."""
2261 """Return a dictionary with the following keys::
2263 count-objects-checked: count of how many objects were checked
2264 count-objects-healthy: how many of those objects were completely
2266 count-objects-unhealthy: how many were damaged in some way
2267 count-objects-unrecoverable: how many were unrecoverable
2268 count-corrupt-shares: how many shares were found to have
2269 corruption, summed over all objects
2273 def get_corrupt_shares():
2274 """Return a set of (IServer, storage_index, sharenum) for all shares
2275 that were found to be corrupt. storage_index is binary."""
2276 def get_all_results():
2277 """Return a dictionary mapping pathname (a tuple of strings, ready to
2278 be slash-joined) to an ICheckResults instance, one for each object
2279 that was checked."""
2281 def get_results_for_storage_index(storage_index):
2282 """Retrive the ICheckResults instance for the given (binary)
2283 storage index. Raises KeyError if there are no results for that
2287 """Return a dictionary with the same keys as
2288 IDirectoryNode.deep_stats()."""
2290 class IDeepCheckAndRepairResults(Interface):
2291 """I contain the results of a deep-check-and-repair operation.
2293 This is returned by a call to ICheckable.deep_check_and_repair().
2296 def get_root_storage_index_string():
2297 """Return the storage index (abbreviated human-readable string) of
2298 the first object checked."""
2300 """Return a dictionary with the following keys::
2302 count-objects-checked: count of how many objects were checked
2303 count-objects-healthy-pre-repair: how many of those objects were
2304 completely healthy (before any
2306 count-objects-unhealthy-pre-repair: how many were damaged in
2308 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2309 count-objects-healthy-post-repair: how many of those objects were
2310 completely healthy (after any
2312 count-objects-unhealthy-post-repair: how many were damaged in
2314 count-objects-unrecoverable-post-repair: how many were
2316 count-repairs-attempted: repairs were attempted on this many
2317 objects. The count-repairs- keys will
2318 always be provided, however unless
2319 repair=true is present, they will all
2321 count-repairs-successful: how many repairs resulted in healthy
2323 count-repairs-unsuccessful: how many repairs resulted did not
2324 results in completely healthy objects
2325 count-corrupt-shares-pre-repair: how many shares were found to
2326 have corruption, summed over all
2327 objects examined (before any
2329 count-corrupt-shares-post-repair: how many shares were found to
2330 have corruption, summed over all
2331 objects examined (after any
2336 """Return a dictionary with the same keys as
2337 IDirectoryNode.deep_stats()."""
2339 def get_corrupt_shares():
2340 """Return a set of (IServer, storage_index, sharenum) for all shares
2341 that were found to be corrupt before any repair was attempted.
2342 storage_index is binary.
2344 def get_remaining_corrupt_shares():
2345 """Return a set of (IServer, storage_index, sharenum) for all shares
2346 that were found to be corrupt after any repair was completed.
2347 storage_index is binary. These are shares that need manual inspection
2348 and probably deletion.
2350 def get_all_results():
2351 """Return a dictionary mapping pathname (a tuple of strings, ready to
2352 be slash-joined) to an ICheckAndRepairResults instance, one for each
2353 object that was checked."""
2355 def get_results_for_storage_index(storage_index):
2356 """Retrive the ICheckAndRepairResults instance for the given (binary)
2357 storage index. Raises KeyError if there are no results for that
2361 class IRepairable(Interface):
2362 def repair(check_results):
2363 """Attempt to repair the given object. Returns a Deferred that fires
2364 with a IRepairResults object.
2366 I must be called with an object that implements ICheckResults, as
2367 proof that you have actually discovered a problem with this file. I
2368 will use the data in the checker results to guide the repair process,
2369 such as which servers provided bad data and should therefore be
2370 avoided. The ICheckResults object is inside the
2371 ICheckAndRepairResults object, which is returned by the
2372 ICheckable.check() method::
2374 d = filenode.check(repair=False)
2375 def _got_results(check_and_repair_results):
2376 check_results = check_and_repair_results.get_pre_repair_results()
2377 return filenode.repair(check_results)
2378 d.addCallback(_got_results)
2382 class IRepairResults(Interface):
2383 """I contain the results of a repair operation."""
2384 def get_successful(self):
2385 """Returns a boolean: True if the repair made the file healthy, False
2386 if not. Repair failure generally indicates a file that has been
2387 damaged beyond repair."""
2390 class IClient(Interface):
2391 def upload(uploadable):
2392 """Upload some data into a CHK, get back the UploadResults for it.
2393 @param uploadable: something that implements IUploadable
2394 @return: a Deferred that fires with the UploadResults instance.
2395 To get the URI for this file, use results.uri .
2398 def create_mutable_file(contents=""):
2399 """Create a new mutable file (with initial) contents, get back the
2402 @param contents: (bytestring, callable, or None): this provides the
2403 initial contents of the mutable file. If 'contents' is a bytestring,
2404 it will be used as-is. If 'contents' is a callable, it will be
2405 invoked with the new MutableFileNode instance and is expected to
2406 return a bytestring with the initial contents of the file (the
2407 callable can use node.get_writekey() to decide how to encrypt the
2408 initial contents, e.g. for a brand new dirnode with initial
2409 children). contents=None is equivalent to an empty string. Using
2410 content_maker= is more efficient than creating a mutable file and
2411 setting its contents in two separate operations.
2413 @return: a Deferred that fires with an IMutableFileNode instance.
2416 def create_dirnode(initial_children={}):
2417 """Create a new unattached dirnode, possibly with initial children.
2419 @param initial_children: dict with keys that are unicode child names,
2420 and values that are (childnode, metadata) tuples.
2422 @return: a Deferred that fires with the new IDirectoryNode instance.
2425 def create_node_from_uri(uri, rouri):
2426 """Create a new IFilesystemNode instance from the uri, synchronously.
2427 @param uri: a string or IURI-providing instance, or None. This could
2428 be for a LiteralFileNode, a CHK file node, a mutable file
2429 node, or a directory node
2430 @param rouri: a string or IURI-providing instance, or None. If the
2431 main uri is None, I will use the rouri instead. If I
2432 recognize the format of the main uri, I will ignore the
2433 rouri (because it can be derived from the writecap).
2435 @return: an instance that provides IFilesystemNode (or more usefully
2436 one of its subclasses). File-specifying URIs will result in
2437 IFileNode-providing instances, like ImmutableFileNode,
2438 LiteralFileNode, or MutableFileNode. Directory-specifying
2439 URIs will result in IDirectoryNode-providing instances, like
2443 class INodeMaker(Interface):
2444 """The NodeMaker is used to create IFilesystemNode instances. It can
2445 accept a filecap/dircap string and return the node right away. It can
2446 also create new nodes (i.e. upload a file, or create a mutable file)
2447 asynchronously. Once you have one of these nodes, you can use other
2448 methods to determine whether it is a file or directory, and to download
2449 or modify its contents.
2451 The NodeMaker encapsulates all the authorities that these
2452 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2453 Tahoe process will typically have a single NodeMaker, but unit tests may
2454 create simplified/mocked forms for testing purposes.
2456 def create_from_cap(writecap, readcap=None, **kwargs):
2457 """I create an IFilesystemNode from the given writecap/readcap. I can
2458 only provide nodes for existing file/directory objects: use my other
2459 methods to create new objects. I return synchronously."""
2461 def create_mutable_file(contents=None, keysize=None):
2462 """I create a new mutable file, and return a Deferred which will fire
2463 with the IMutableFileNode instance when it is ready. If contents= is
2464 provided (a bytestring), it will be used as the initial contents of
2465 the new file, otherwise the file will contain zero bytes. keysize= is
2466 for use by unit tests, to create mutable files that are smaller than
2469 def create_new_mutable_directory(initial_children={}):
2470 """I create a new mutable directory, and return a Deferred which will
2471 fire with the IDirectoryNode instance when it is ready. If
2472 initial_children= is provided (a dict mapping unicode child name to
2473 (childnode, metadata_dict) tuples), the directory will be populated
2474 with those children, otherwise it will be empty."""
2476 class IClientStatus(Interface):
2477 def list_all_uploads():
2478 """Return a list of uploader objects, one for each upload which
2479 currently has an object available (tracked with weakrefs). This is
2480 intended for debugging purposes."""
2481 def list_active_uploads():
2482 """Return a list of active IUploadStatus objects."""
2483 def list_recent_uploads():
2484 """Return a list of IUploadStatus objects for the most recently
2487 def list_all_downloads():
2488 """Return a list of downloader objects, one for each download which
2489 currently has an object available (tracked with weakrefs). This is
2490 intended for debugging purposes."""
2491 def list_active_downloads():
2492 """Return a list of active IDownloadStatus objects."""
2493 def list_recent_downloads():
2494 """Return a list of IDownloadStatus objects for the most recently
2495 started downloads."""
2497 class IUploadStatus(Interface):
2499 """Return a timestamp (float with seconds since epoch) indicating
2500 when the operation was started."""
2501 def get_storage_index():
2502 """Return a string with the (binary) storage index in use on this
2503 upload. Returns None if the storage index has not yet been
2506 """Return an integer with the number of bytes that will eventually
2507 be uploaded for this file. Returns None if the size is not yet known.
2510 """Return True if this upload is using a Helper, False if not."""
2512 """Return a string describing the current state of the upload
2515 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2516 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2517 made towards hashing the file to determine a CHK encryption key: if
2518 non-convergent encryption is in use, this will be trivial, otherwise
2519 the whole file must be hashed. 'ciphertext' describes how much of the
2520 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2521 uploads. 'encode_and_push' describes how much of the encode-and-push
2522 process has finished: for helper uploads this is dependent upon the
2523 helper providing progress reports. It might be reasonable to add all
2524 three numbers and report the sum to the user."""
2526 """Return True if the upload is currently active, False if not."""
2528 """Return an instance of UploadResults (which contains timing and
2529 sharemap information). Might return None if the upload is not yet
2532 """Each upload status gets a unique number: this method returns that
2533 number. This provides a handle to this particular upload, so a web
2534 page can generate a suitable hyperlink."""
2536 class IDownloadStatus(Interface):
2538 """Return a timestamp (float with seconds since epoch) indicating
2539 when the operation was started."""
2540 def get_storage_index():
2541 """Return a string with the (binary) storage index in use on this
2542 download. This may be None if there is no storage index (i.e. LIT
2545 """Return an integer with the number of bytes that will eventually be
2546 retrieved for this file. Returns None if the size is not yet known.
2549 """Return True if this download is using a Helper, False if not."""
2551 """Return a string describing the current state of the download
2554 """Returns a float (from 0.0 to 1.0) describing the amount of the
2555 download that has completed. This value will remain at 0.0 until the
2556 first byte of plaintext is pushed to the download target."""
2558 """Return True if the download is currently active, False if not."""
2560 """Each download status gets a unique number: this method returns
2561 that number. This provides a handle to this particular download, so a
2562 web page can generate a suitable hyperlink."""
2564 class IServermapUpdaterStatus(Interface):
2566 class IPublishStatus(Interface):
2568 class IRetrieveStatus(Interface):
2571 class NotCapableError(Exception):
2572 """You have tried to write to a read-only node."""
2574 class BadWriteEnablerError(Exception):
2577 class RIControlClient(RemoteInterface):
2579 def wait_for_client_connections(num_clients=int):
2580 """Do not return until we have connections to at least NUM_CLIENTS
2584 def upload_from_file_to_uri(filename=str,
2585 convergence=ChoiceOf(None,
2586 StringConstraint(2**20))):
2587 """Upload a file to the grid. This accepts a filename (which must be
2588 absolute) that points to a file on the node's local disk. The node will
2589 read the contents of this file, upload it to the grid, then return the
2590 URI at which it was uploaded. If convergence is None then a random
2591 encryption key will be used, else the plaintext will be hashed, then
2592 that hash will be mixed together with the "convergence" string to form
2597 def download_from_uri_to_file(uri=URI, filename=str):
2598 """Download a file from the grid, placing it on the node's local disk
2599 at the given filename (which must be absolute[?]). Returns the
2600 absolute filename where the file was written."""
2605 def get_memory_usage():
2606 """Return a dict describes the amount of memory currently in use. The
2607 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2608 measuring memory consupmtion in bytes."""
2609 return DictOf(str, int)
2611 def speed_test(count=int, size=int, mutable=Any()):
2612 """Write 'count' tempfiles to disk, all of the given size. Measure
2613 how long (in seconds) it takes to upload them all to the servers.
2614 Then measure how long it takes to download all of them. If 'mutable'
2615 is 'create', time creation of mutable files. If 'mutable' is
2616 'upload', then time access to the same mutable file instead of
2619 Returns a tuple of (upload_time, download_time).
2621 return (float, float)
2623 def measure_peer_response_time():
2624 """Send a short message to each connected peer, and measure the time
2625 it takes for them to respond to it. This is a rough measure of the
2626 application-level round trip time.
2628 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2631 return DictOf(str, float)
2633 UploadResults = Any() #DictOf(str, str)
2635 class RIEncryptedUploadable(RemoteInterface):
2636 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2641 def get_all_encoding_parameters():
2642 return (int, int, int, long)
2644 def read_encrypted(offset=Offset, length=ReadSize):
2651 class RICHKUploadHelper(RemoteInterface):
2652 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2656 Return a dictionary of version information.
2658 return DictOf(str, Any())
2660 def upload(reader=RIEncryptedUploadable):
2661 return UploadResults
2664 class RIHelper(RemoteInterface):
2665 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2669 Return a dictionary of version information.
2671 return DictOf(str, Any())
2673 def upload_chk(si=StorageIndex):
2674 """See if a file with a given storage index needs uploading. The
2675 helper will ask the appropriate storage servers to see if the file
2676 has already been uploaded. If so, the helper will return a set of
2677 'upload results' that includes whatever hashes are needed to build
2678 the read-cap, and perhaps a truncated sharemap.
2680 If the file has not yet been uploaded (or if it was only partially
2681 uploaded), the helper will return an empty upload-results dictionary
2682 and also an RICHKUploadHelper object that will take care of the
2683 upload process. The client should call upload() on this object and
2684 pass it a reference to an RIEncryptedUploadable object that will
2685 provide ciphertext. When the upload is finished, the upload() method
2686 will finish and return the upload results.
2688 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2691 class RIStatsProvider(RemoteInterface):
2692 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2694 Provides access to statistics and monitoring information.
2699 returns a dictionary containing 'counters' and 'stats', each a
2700 dictionary with string counter/stat name keys, and numeric or None values.
2701 counters are monotonically increasing measures of work done, and
2702 stats are instantaneous measures (potentially time averaged
2705 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2707 class RIStatsGatherer(RemoteInterface):
2708 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2710 Provides a monitoring service for centralised collection of stats
2713 def provide(provider=RIStatsProvider, nickname=str):
2715 @param provider: a stats collector instance which should be polled
2716 periodically by the gatherer to collect stats.
2717 @param nickname: a name useful to identify the provided client
2722 class IStatsProducer(Interface):
2725 returns a dictionary, with str keys representing the names of stats
2726 to be monitored, and numeric values.
2729 class RIKeyGenerator(RemoteInterface):
2730 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2732 Provides a service offering to make RSA key pairs.
2735 def get_rsa_key_pair(key_size=int):
2737 @param key_size: the size of the signature key.
2738 @return: tuple(verifying_key, signing_key)
2740 return TupleOf(str, str)
2743 class FileTooLargeError(Exception):
2746 class IValidatedThingProxy(Interface):
2748 """ Acquire a thing and validate it. Return a deferred which is
2749 eventually fired with self if the thing is valid or errbacked if it
2750 can't be acquired or validated."""
2752 class InsufficientVersionError(Exception):
2753 def __init__(self, needed, got):
2754 self.needed = needed
2757 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2760 class EmptyPathnameComponentError(Exception):
2761 """The webapi disallows empty pathname components."""