2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIStubClient(RemoteInterface):
34 """Each client publishes a service announcement for a dummy object called
35 the StubClient. This object doesn't actually offer any services, but the
36 announcement helps the Introducer keep track of which clients are
37 subscribed (so the grid admin can keep track of things like the size of
38 the grid and the client versions in use. This is the (empty)
39 RemoteInterface for the StubClient."""
41 class RIBucketWriter(RemoteInterface):
42 """ Objects of this kind live on the server side. """
43 def write(offset=Offset, data=ShareData):
48 If the data that has been written is incomplete or inconsistent then
49 the server will throw the data away, else it will store it for future
55 """Abandon all the data that has been written.
59 class RIBucketReader(RemoteInterface):
60 def read(offset=Offset, length=ReadSize):
63 def advise_corrupt_share(reason=str):
64 """Clients who discover hash failures in shares that they have
65 downloaded from me will use this method to inform me about the
66 failures. I will record their concern so that my operator can
67 manually inspect the shares in question. I return None.
69 This is a wrapper around RIStorageServer.advise_corrupt_share(),
70 which is tied to a specific share, and therefore does not need the
71 extra share-identifying arguments. Please see that method for full
75 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
76 # elements are (offset, length, operator, specimen)
77 # operator is one of "lt, le, eq, ne, ge, gt"
78 # nop always passes and is used to fetch data while writing.
79 # you should use length==len(specimen) for everything except nop
80 DataVector = ListOf(TupleOf(Offset, ShareData))
81 # (offset, data). This limits us to 30 writes of 1MiB each per call
82 TestAndWriteVectorsForShares = DictOf(int,
85 ChoiceOf(None, Offset), # new_length
87 ReadVector = ListOf(TupleOf(Offset, ReadSize))
88 ReadData = ListOf(ShareData)
89 # returns data[offset:offset+length] for each element of TestVector
91 class RIStorageServer(RemoteInterface):
92 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
96 Return a dictionary of version information.
98 return DictOf(str, Any())
100 def allocate_buckets(storage_index=StorageIndex,
101 renew_secret=LeaseRenewSecret,
102 cancel_secret=LeaseCancelSecret,
103 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
104 allocated_size=Offset, canary=Referenceable):
106 @param storage_index: the index of the bucket to be created or
108 @param sharenums: these are the share numbers (probably between 0 and
109 99) that the sender is proposing to store on this
111 @param renew_secret: This is the secret used to protect bucket refresh
112 This secret is generated by the client and
113 stored for later comparison by the server. Each
114 server is given a different secret.
115 @param cancel_secret: Like renew_secret, but protects bucket decref.
116 @param canary: If the canary is lost before close(), the bucket is
118 @return: tuple of (alreadygot, allocated), where alreadygot is what we
119 already have and allocated is what we hereby agree to accept.
120 New leases are added for shares in both lists.
122 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
123 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
125 def add_lease(storage_index=StorageIndex,
126 renew_secret=LeaseRenewSecret,
127 cancel_secret=LeaseCancelSecret):
129 Add a new lease on the given bucket. If the renew_secret matches an
130 existing lease, that lease will be renewed instead. If there is no
131 bucket for the given storage_index, return silently. (note that in
132 tahoe-1.3.0 and earlier, IndexError was raised if there was no
135 return Any() # returns None now, but future versions might change
137 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
139 Renew the lease on a given bucket, resetting the timer to 31 days.
140 Some networks will use this, some will not. If there is no bucket for
141 the given storage_index, IndexError will be raised.
143 For mutable shares, if the given renew_secret does not match an
144 existing lease, IndexError will be raised with a note listing the
145 server-nodeids on the existing leases, so leases on migrated shares
146 can be renewed or cancelled. For immutable shares, IndexError
147 (without the note) will be raised.
151 def get_buckets(storage_index=StorageIndex):
152 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
156 def slot_readv(storage_index=StorageIndex,
157 shares=ListOf(int), readv=ReadVector):
158 """Read a vector from the numbered shares associated with the given
159 storage index. An empty shares list means to return data from all
160 known shares. Returns a dictionary with one key per share."""
161 return DictOf(int, ReadData) # shnum -> results
163 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
164 secrets=TupleOf(WriteEnablerSecret,
167 tw_vectors=TestAndWriteVectorsForShares,
170 """General-purpose test-and-set operation for mutable slots. Perform
171 a bunch of comparisons against the existing shares. If they all pass,
172 then apply a bunch of write vectors to those shares. Then use the
173 read vectors to extract data from all the shares and return the data.
175 This method is, um, large. The goal is to allow clients to update all
176 the shares associated with a mutable file in a single round trip.
178 @param storage_index: the index of the bucket to be created or
180 @param write_enabler: a secret that is stored along with the slot.
181 Writes are accepted from any caller who can
182 present the matching secret. A different secret
183 should be used for each slot*server pair.
184 @param renew_secret: This is the secret used to protect bucket refresh
185 This secret is generated by the client and
186 stored for later comparison by the server. Each
187 server is given a different secret.
188 @param cancel_secret: Like renew_secret, but protects bucket decref.
190 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
191 cancel_secret). The first is required to perform any write. The
192 latter two are used when allocating new shares. To simply acquire a
193 new lease on existing shares, use an empty testv and an empty writev.
195 Each share can have a separate test vector (i.e. a list of
196 comparisons to perform). If all vectors for all shares pass, then all
197 writes for all shares are recorded. Each comparison is a 4-tuple of
198 (offset, length, operator, specimen), which effectively does a bool(
199 (read(offset, length)) OPERATOR specimen ) and only performs the
200 write if all these evaluate to True. Basic test-and-set uses 'eq'.
201 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
202 Write-if-same-or-newer uses 'le'.
204 Reads from the end of the container are truncated, and missing shares
205 behave like empty ones, so to assert that a share doesn't exist (for
206 use when creating a new share), use (0, 1, 'eq', '').
208 The write vector will be applied to the given share, expanding it if
209 necessary. A write vector applied to a share number that did not
210 exist previously will cause that share to be created.
212 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
213 vector whose offset is beyond the end of the current data, the space
214 between the end of the current data and the beginning of the write
215 vector will be filled with zero bytes. In earlier versions the
216 contents of this space was unspecified (and might end up containing
217 secrets). Storage servers with the new zero-filling behavior will
218 advertise a true value for the 'fills-holes-with-zero-bytes' key
219 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
222 Each write vector is accompanied by a 'new_length' argument, which
223 can be used to truncate the data. If new_length is not None and it is
224 less than the current size of the data (after applying all write
225 vectors), then the data will be truncated to new_length. If
226 new_length==0, the share will be deleted.
228 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
229 enlarge the file by sending a number larger than the size of the data
230 after applying all write vectors. That behavior was not used, and as
231 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
234 If a storage client knows that the server supports zero-filling, for
235 example from the 'fills-holes-with-zero-bytes' key in its version
236 information, it can extend the file efficiently by writing a single
237 zero byte just before the new end-of-file. Otherwise it must
238 explicitly write zeroes to all bytes between the old and new
239 end-of-file. In any case it should avoid sending new_length larger
240 than the size of the data after applying all write vectors.
242 The read vector is used to extract data from all known shares,
243 *before* any writes have been applied. The same vector is used for
244 all shares. This captures the state that was tested by the test
247 This method returns two values: a boolean and a dict. The boolean is
248 True if the write vectors were applied, False if not. The dict is
249 keyed by share number, and each value contains a list of strings, one
250 for each element of the read vector.
252 If the write_enabler is wrong, this will raise BadWriteEnablerError.
253 To enable share migration (using update_write_enabler), the exception
254 will have the nodeid used for the old write enabler embedded in it,
255 in the following string::
257 The write enabler was recorded by nodeid '%s'.
259 Note that the nodeid here is encoded using the same base32 encoding
260 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
263 return TupleOf(bool, DictOf(int, ReadData))
265 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
266 shnum=int, reason=str):
267 """Clients who discover hash failures in shares that they have
268 downloaded from me will use this method to inform me about the
269 failures. I will record their concern so that my operator can
270 manually inspect the shares in question. I return None.
272 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
273 (binary) storage index string, and 'shnum' is the integer share
274 number. 'reason' is a human-readable explanation of the problem,
275 probably including some expected hash values and the computed ones
276 which did not match. Corruption advisories for mutable shares should
277 include a hash of the public key (the same value that appears in the
278 mutable-file verify-cap), since the current share format does not
282 class IStorageBucketWriter(Interface):
284 Objects of this kind live on the client side.
286 def put_block(segmentnum=int, data=ShareData):
287 """@param data: For most segments, this data will be 'blocksize'
288 bytes in length. The last segment might be shorter.
289 @return: a Deferred that fires (with None) when the operation completes
292 def put_plaintext_hashes(hashes=ListOf(Hash)):
294 @return: a Deferred that fires (with None) when the operation completes
297 def put_crypttext_hashes(hashes=ListOf(Hash)):
299 @return: a Deferred that fires (with None) when the operation completes
302 def put_block_hashes(blockhashes=ListOf(Hash)):
304 @return: a Deferred that fires (with None) when the operation completes
307 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
309 @return: a Deferred that fires (with None) when the operation completes
312 def put_uri_extension(data=URIExtensionData):
313 """This block of data contains integrity-checking information (hashes
314 of plaintext, crypttext, and shares), as well as encoding parameters
315 that are necessary to recover the data. This is a serialized dict
316 mapping strings to other strings. The hash of this data is kept in
317 the URI and verified before any of the data is used. All buckets for
318 a given file contain identical copies of this data.
320 The serialization format is specified with the following pseudocode:
321 for k in sorted(dict.keys()):
322 assert re.match(r'^[a-zA-Z_\-]+$', k)
323 write(k + ':' + netstring(dict[k]))
325 @return: a Deferred that fires (with None) when the operation completes
329 """Finish writing and close the bucket. The share is not finalized
330 until this method is called: if the uploading client disconnects
331 before calling close(), the partially-written share will be
334 @return: a Deferred that fires (with None) when the operation completes
337 class IStorageBucketReader(Interface):
339 def get_block_data(blocknum=int, blocksize=int, size=int):
340 """Most blocks will be the same size. The last block might be shorter
346 def get_crypttext_hashes():
348 @return: ListOf(Hash)
351 def get_block_hashes(at_least_these=SetOf(int)):
353 @return: ListOf(Hash)
356 def get_share_hashes(at_least_these=SetOf(int)):
358 @return: ListOf(TupleOf(int, Hash))
361 def get_uri_extension():
363 @return: URIExtensionData
366 class IStorageBroker(Interface):
367 def get_servers_for_psi(peer_selection_index):
369 @return: list of IServer instances
371 def get_connected_servers():
373 @return: frozenset of connected IServer instances
375 def get_known_servers():
377 @return: frozenset of IServer instances
379 def get_all_serverids():
381 @return: frozenset of serverid strings
383 def get_nickname_for_serverid(serverid):
385 @return: unicode nickname, or None
388 # methods moved from IntroducerClient, need review
389 def get_all_connections():
390 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
391 each active connection we've established to a remote service. This is
392 mostly useful for unit tests that need to wait until a certain number
393 of connections have been made."""
395 def get_all_connectors():
396 """Return a dict that maps from (nodeid, service_name) to a
397 RemoteServiceConnector instance for all services that we are actively
398 trying to connect to. Each RemoteServiceConnector has the following
401 service_name: the type of service provided, like 'storage'
402 announcement_time: when we first heard about this service
403 last_connect_time: when we last established a connection
404 last_loss_time: when we last lost a connection
406 version: the peer's version, from the most recent connection
407 oldest_supported: the peer's oldest supported version, same
409 rref: the RemoteReference, if connected, otherwise None
410 remote_host: the IAddress, if connected, otherwise None
412 This method is intended for monitoring interfaces, such as a web page
413 which describes connecting and connected peers.
416 def get_all_peerids():
417 """Return a frozenset of all peerids to whom we have a connection (to
418 one or more services) established. Mostly useful for unit tests."""
420 def get_all_connections_for(service_name):
421 """Return a frozenset of (nodeid, service_name, rref) tuples, one
422 for each active connection that provides the given SERVICE_NAME."""
424 def get_permuted_peers(service_name, key):
425 """Returns an ordered list of (peerid, rref) tuples, selecting from
426 the connections that provide SERVICE_NAME, using a hash-based
427 permutation keyed by KEY. This randomizes the service list in a
428 repeatable way, to distribute load over many peers.
432 class IMutableSlotWriter(Interface):
434 The interface for a writer around a mutable slot on a remote server.
436 def set_checkstring(checkstring, *args):
438 Set the checkstring that I will pass to the remote server when
441 @param checkstring A packed checkstring to use.
443 Note that implementations can differ in which semantics they
444 wish to support for set_checkstring -- they can, for example,
445 build the checkstring themselves from its constituents, or
449 def get_checkstring():
451 Get the checkstring that I think currently exists on the remote
455 def put_block(data, segnum, salt):
457 Add a block and salt to the share.
460 def put_encprivkey(encprivkey):
462 Add the encrypted private key to the share.
465 def put_blockhashes(blockhashes=list):
467 Add the block hash tree to the share.
470 def put_sharehashes(sharehashes=dict):
472 Add the share hash chain to the share.
477 Return the part of the share that needs to be signed.
480 def put_signature(signature):
482 Add the signature to the share.
485 def put_verification_key(verification_key):
487 Add the verification key to the share.
490 def finish_publishing():
492 Do anything necessary to finish writing the share to a remote
493 server. I require that no further publishing needs to take place
494 after this method has been called.
498 class IURI(Interface):
499 def init_from_string(uri):
500 """Accept a string (as created by my to_string() method) and populate
501 this instance with its data. I am not normally called directly,
502 please use the module-level uri.from_string() function to convert
503 arbitrary URI strings into IURI-providing instances."""
506 """Return False if this URI be used to modify the data. Return True
507 if this URI cannot be used to modify the data."""
510 """Return True if the data can be modified by *somebody* (perhaps
511 someone who has a more powerful URI than this one)."""
513 # TODO: rename to get_read_cap()
515 """Return another IURI instance, which represents a read-only form of
516 this one. If is_readonly() is True, this returns self."""
518 def get_verify_cap():
519 """Return an instance that provides IVerifierURI, which can be used
520 to check on the availability of the file or directory, without
521 providing enough capabilities to actually read or modify the
522 contents. This may return None if the file does not need checking or
523 verification (e.g. LIT URIs).
527 """Return a string of printable ASCII characters, suitable for
528 passing into init_from_string."""
530 class IVerifierURI(Interface, IURI):
531 def init_from_string(uri):
532 """Accept a string (as created by my to_string() method) and populate
533 this instance with its data. I am not normally called directly,
534 please use the module-level uri.from_string() function to convert
535 arbitrary URI strings into IURI-providing instances."""
538 """Return a string of printable ASCII characters, suitable for
539 passing into init_from_string."""
541 class IDirnodeURI(Interface):
542 """I am a URI which represents a dirnode."""
544 class IFileURI(Interface):
545 """I am a URI which represents a filenode."""
547 """Return the length (in bytes) of the file that I represent."""
549 class IImmutableFileURI(IFileURI):
552 class IMutableFileURI(Interface):
555 class IDirectoryURI(Interface):
558 class IReadonlyDirectoryURI(Interface):
561 class CapConstraintError(Exception):
562 """A constraint on a cap was violated."""
564 class MustBeDeepImmutableError(CapConstraintError):
565 """Mutable children cannot be added to an immutable directory.
566 Also, caps obtained from an immutable directory can trigger this error
567 if they are later found to refer to a mutable object and then used."""
569 class MustBeReadonlyError(CapConstraintError):
570 """Known write caps cannot be specified in a ro_uri field. Also,
571 caps obtained from a ro_uri field can trigger this error if they
572 are later found to be write caps and then used."""
574 class MustNotBeUnknownRWError(CapConstraintError):
575 """Cannot add an unknown child cap specified in a rw_uri field."""
578 class IReadable(Interface):
579 """I represent a readable object -- either an immutable file, or a
580 specific version of a mutable file.
584 """Return True if this reference provides mutable access to the given
585 file or directory (i.e. if you can modify it), or False if not. Note
586 that even if this reference is read-only, someone else may hold a
587 read-write reference to it.
589 For an IReadable returned by get_best_readable_version(), this will
590 always return True, but for instances of subinterfaces such as
591 IMutableFileVersion, it may return False."""
594 """Return True if this file or directory is mutable (by *somebody*,
595 not necessarily you), False if it is is immutable. Note that a file
596 might be mutable overall, but your reference to it might be
597 read-only. On the other hand, all references to an immutable file
598 will be read-only; there are no read-write references to an immutable
601 def get_storage_index():
602 """Return the storage index of the file."""
605 """Return the length (in bytes) of this readable object."""
607 def download_to_data():
608 """Download all of the file contents. I return a Deferred that fires
609 with the contents as a byte string."""
611 def read(consumer, offset=0, size=None):
612 """Download a portion (possibly all) of the file's contents, making
613 them available to the given IConsumer. Return a Deferred that fires
614 (with the consumer) when the consumer is unregistered (either because
615 the last byte has been given to it, or because the consumer threw an
616 exception during write(), possibly because it no longer wants to
617 receive data). The portion downloaded will start at 'offset' and
618 contain 'size' bytes (or the remainder of the file if size==None).
620 The consumer will be used in non-streaming mode: an IPullProducer
621 will be attached to it.
623 The consumer will not receive data right away: several network trips
624 must occur first. The order of events will be::
626 consumer.registerProducer(p, streaming)
627 (if streaming == False)::
628 consumer does p.resumeProducing()
630 consumer does p.resumeProducing()
631 consumer.write(data).. (repeat until all data is written)
632 consumer.unregisterProducer()
633 deferred.callback(consumer)
635 If a download error occurs, or an exception is raised by
636 consumer.registerProducer() or consumer.write(), I will call
637 consumer.unregisterProducer() and then deliver the exception via
638 deferred.errback(). To cancel the download, the consumer should call
639 p.stopProducing(), which will result in an exception being delivered
640 via deferred.errback().
642 See src/allmydata/util/consumer.py for an example of a simple
643 download-to-memory consumer.
647 class IWriteable(Interface):
649 I define methods that callers can use to update SDMF and MDMF
650 mutable files on a Tahoe-LAFS grid.
652 # XXX: For the moment, we have only this. It is possible that we
653 # want to move overwrite() and modify() in here too.
654 def update(data, offset):
656 I write the data from my data argument to the MDMF file,
657 starting at offset. I continue writing data until my data
658 argument is exhausted, appending data to the file as necessary.
660 # assert IMutableUploadable.providedBy(data)
661 # to append data: offset=node.get_size_of_best_version()
662 # do we want to support compacting MDMF?
663 # for an MDMF file, this can be done with O(data.get_size())
664 # memory. For an SDMF file, any modification takes
665 # O(node.get_size_of_best_version()).
668 class IMutableFileVersion(IReadable):
669 """I provide access to a particular version of a mutable file. The
670 access is read/write if I was obtained from a filenode derived from
671 a write cap, or read-only if the filenode was derived from a read cap.
674 def get_sequence_number():
675 """Return the sequence number of this version."""
678 """Return the IMutableFileServerMap instance that was used to create
683 """Return this filenode's writekey, or None if the node does not have
684 write-capability. This may be used to assist with data structures
685 that need to make certain data available only to writers, such as the
686 read-write child caps in dirnodes. The recommended process is to have
687 reader-visible data be submitted to the filenode in the clear (where
688 it will be encrypted by the filenode using the readkey), but encrypt
689 writer-visible data using this writekey.
692 # TODO: Can this be overwrite instead of replace?
693 def replace(new_contents):
694 """Replace the contents of the mutable file, provided that no other
695 node has published (or is attempting to publish, concurrently) a
696 newer version of the file than this one.
698 I will avoid modifying any share that is different than the version
699 given by get_sequence_number(). However, if another node is writing
700 to the file at the same time as me, I may manage to update some shares
701 while they update others. If I see any evidence of this, I will signal
702 UncoordinatedWriteError, and the file will be left in an inconsistent
703 state (possibly the version you provided, possibly the old version,
704 possibly somebody else's version, and possibly a mix of shares from
707 The recommended response to UncoordinatedWriteError is to either
708 return it to the caller (since they failed to coordinate their
709 writes), or to attempt some sort of recovery. It may be sufficient to
710 wait a random interval (with exponential backoff) and repeat your
711 operation. If I do not signal UncoordinatedWriteError, then I was
712 able to write the new version without incident.
714 I return a Deferred that fires (with a PublishStatus object) when the
715 update has completed.
718 def modify(modifier_cb):
719 """Modify the contents of the file, by downloading this version,
720 applying the modifier function (or bound method), then uploading
721 the new version. This will succeed as long as no other node
722 publishes a version between the download and the upload.
723 I return a Deferred that fires (with a PublishStatus object) when
724 the update is complete.
726 The modifier callable will be given three arguments: a string (with
727 the old contents), a 'first_time' boolean, and a servermap. As with
728 download_to_data(), the old contents will be from this version,
729 but the modifier can use the servermap to make other decisions
730 (such as refusing to apply the delta if there are multiple parallel
731 versions, or if there is evidence of a newer unrecoverable version).
732 'first_time' will be True the first time the modifier is called,
733 and False on any subsequent calls.
735 The callable should return a string with the new contents. The
736 callable must be prepared to be called multiple times, and must
737 examine the input string to see if the change that it wants to make
738 is already present in the old version. If it does not need to make
739 any changes, it can either return None, or return its input string.
741 If the modifier raises an exception, it will be returned in the
746 # The hierarchy looks like this:
753 class IFilesystemNode(Interface):
755 """Return the strongest 'cap instance' associated with this node.
756 (writecap for writeable-mutable files/directories, readcap for
757 immutable or readonly-mutable files/directories). To convert this
758 into a string, call .to_string() on the result."""
761 """Return a readonly cap instance for this node. For immutable or
762 readonly nodes, get_cap() and get_readcap() return the same thing."""
764 def get_repair_cap():
765 """Return an IURI instance that can be used to repair the file, or
766 None if this node cannot be repaired (either because it is not
767 distributed, like a LIT file, or because the node does not represent
768 sufficient authority to create a repair-cap, like a read-only RSA
769 mutable file node [which cannot create the correct write-enablers]).
772 def get_verify_cap():
773 """Return an IVerifierURI instance that represents the
774 'verifiy/refresh capability' for this node. The holder of this
775 capability will be able to renew the lease for this node, protecting
776 it from garbage-collection. They will also be able to ask a server if
777 it holds a share for the file or directory.
781 """Return the URI string corresponding to the strongest cap associated
782 with this node. If this node is read-only, the URI will only offer
783 read-only access. If this node is read-write, the URI will offer
786 If you have read-write access to a node and wish to share merely
787 read-only access with others, use get_readonly_uri().
791 """Return the URI string that can be used by others to get write
792 access to this node, if it is writeable. If this is a read-only node,
795 def get_readonly_uri():
796 """Return the URI string that can be used by others to get read-only
797 access to this node. The result is a read-only URI, regardless of
798 whether this node is read-only or read-write.
800 If you have merely read-only access to this node, get_readonly_uri()
801 will return the same thing as get_uri().
804 def get_storage_index():
805 """Return a string with the (binary) storage index in use on this
806 download. This may be None if there is no storage index (i.e. LIT
807 files and directories)."""
810 """Return True if this reference provides mutable access to the given
811 file or directory (i.e. if you can modify it), or False if not. Note
812 that even if this reference is read-only, someone else may hold a
813 read-write reference to it."""
816 """Return True if this file or directory is mutable (by *somebody*,
817 not necessarily you), False if it is is immutable. Note that a file
818 might be mutable overall, but your reference to it might be
819 read-only. On the other hand, all references to an immutable file
820 will be read-only; there are no read-write references to an immutable
825 """Return True if this is an unknown node."""
827 def is_allowed_in_immutable_directory():
828 """Return True if this node is allowed as a child of a deep-immutable
829 directory. This is true if either the node is of a known-immutable type,
830 or it is unknown and read-only.
834 """Raise any error associated with this node."""
836 # XXX: These may not be appropriate outside the context of an IReadable.
838 """Return the length (in bytes) of the data this node represents. For
839 directory nodes, I return the size of the backing store. I return
840 synchronously and do not consult the network, so for mutable objects,
841 I will return the most recently observed size for the object, or None
842 if I don't remember a size. Use get_current_size, which returns a
843 Deferred, if you want more up-to-date information."""
845 def get_current_size():
846 """I return a Deferred that fires with the length (in bytes) of the
847 data this node represents.
850 class IFileNode(IFilesystemNode):
851 """I am a node which represents a file: a sequence of bytes. I am not a
852 container, like IDirectoryNode."""
853 def get_best_readable_version():
854 """Return a Deferred that fires with an IReadable for the 'best'
855 available version of the file. The IReadable provides only read
856 access, even if this filenode was derived from a write cap.
858 For an immutable file, there is only one version. For a mutable
859 file, the 'best' version is the recoverable version with the
860 highest sequence number. If no uncoordinated writes have occurred,
861 and if enough shares are available, then this will be the most
862 recent version that has been uploaded. If no version is recoverable,
863 the Deferred will errback with an UnrecoverableFileError.
866 def download_best_version():
867 """Download the contents of the version that would be returned
868 by get_best_readable_version(). This is equivalent to calling
869 download_to_data() on the IReadable given by that method.
871 I return a Deferred that fires with a byte string when the file
872 has been fully downloaded. To support streaming download, use
873 the 'read' method of IReadable. If no version is recoverable,
874 the Deferred will errback with an UnrecoverableFileError.
877 def get_size_of_best_version():
878 """Find the size of the version that would be returned by
879 get_best_readable_version().
881 I return a Deferred that fires with an integer. If no version
882 is recoverable, the Deferred will errback with an
883 UnrecoverableFileError.
887 class IImmutableFileNode(IFileNode, IReadable):
888 """I am a node representing an immutable file. Immutable files have
892 class IMutableFileNode(IFileNode):
893 """I provide access to a 'mutable file', which retains its identity
894 regardless of what contents are put in it.
896 The consistency-vs-availability problem means that there might be
897 multiple versions of a file present in the grid, some of which might be
898 unrecoverable (i.e. have fewer than 'k' shares). These versions are
899 loosely ordered: each has a sequence number and a hash, and any version
900 with seqnum=N was uploaded by a node which has seen at least one version
903 The 'servermap' (an instance of IMutableFileServerMap) is used to
904 describe the versions that are known to be present in the grid, and which
905 servers are hosting their shares. It is used to represent the 'state of
906 the world', and is used for this purpose by my test-and-set operations.
907 Downloading the contents of the mutable file will also return a
908 servermap. Uploading a new version into the mutable file requires a
909 servermap as input, and the semantics of the replace operation is
910 'replace the file with my new version if it looks like nobody else has
911 changed the file since my previous download'. Because the file is
912 distributed, this is not a perfect test-and-set operation, but it will do
913 its best. If the replace process sees evidence of a simultaneous write,
914 it will signal an UncoordinatedWriteError, so that the caller can take
918 Most readers will want to use the 'best' current version of the file, and
919 should use my 'download_best_version()' method.
921 To unconditionally replace the file, callers should use overwrite(). This
922 is the mode that user-visible mutable files will probably use.
924 To apply some delta to the file, call modify() with a callable modifier
925 function that can apply the modification that you want to make. This is
926 the mode that dirnodes will use, since most directory modification
927 operations can be expressed in terms of deltas to the directory state.
930 Three methods are available for users who need to perform more complex
931 operations. The first is get_servermap(), which returns an up-to-date
932 servermap using a specified mode. The second is download_version(), which
933 downloads a specific version (not necessarily the 'best' one). The third
934 is 'upload', which accepts new contents and a servermap (which must have
935 been updated with MODE_WRITE). The upload method will attempt to apply
936 the new contents as long as no other node has modified the file since the
937 servermap was updated. This might be useful to a caller who wants to
938 merge multiple versions into a single new one.
940 Note that each time the servermap is updated, a specific 'mode' is used,
941 which determines how many peers are queried. To use a servermap for my
942 replace() method, that servermap must have been updated in MODE_WRITE.
943 These modes are defined in allmydata.mutable.common, and consist of
944 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
945 allmydata/mutable/servermap.py for details about the differences.
947 Mutable files are currently limited in size (about 3.5MB max) and can
948 only be retrieved and updated all-at-once, as a single big string. Future
949 versions of our mutable files will remove this restriction.
951 def get_best_mutable_version():
952 """Return a Deferred that fires with an IMutableFileVersion for
953 the 'best' available version of the file. The best version is
954 the recoverable version with the highest sequence number. If no
955 uncoordinated writes have occurred, and if enough shares are
956 available, then this will be the most recent version that has
959 If no version is recoverable, the Deferred will errback with an
960 UnrecoverableFileError.
963 def overwrite(new_contents):
964 """Unconditionally replace the contents of the mutable file with new
965 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
966 is only appropriate to use when the new contents of the file are
967 completely unrelated to the old ones, and you do not care about other
970 I return a Deferred that fires (with a PublishStatus object) when the
971 update has completed.
974 def modify(modifier_cb):
975 """Modify the contents of the file, by downloading the current
976 version, applying the modifier function (or bound method), then
977 uploading the new version. I return a Deferred that fires (with a
978 PublishStatus object) when the update is complete.
980 The modifier callable will be given three arguments: a string (with
981 the old contents), a 'first_time' boolean, and a servermap. As with
982 download_best_version(), the old contents will be from the best
983 recoverable version, but the modifier can use the servermap to make
984 other decisions (such as refusing to apply the delta if there are
985 multiple parallel versions, or if there is evidence of a newer
986 unrecoverable version). 'first_time' will be True the first time the
987 modifier is called, and False on any subsequent calls.
989 The callable should return a string with the new contents. The
990 callable must be prepared to be called multiple times, and must
991 examine the input string to see if the change that it wants to make
992 is already present in the old version. If it does not need to make
993 any changes, it can either return None, or return its input string.
995 If the modifier raises an exception, it will be returned in the
999 def get_servermap(mode):
1000 """Return a Deferred that fires with an IMutableFileServerMap
1001 instance, updated using the given mode.
1004 def download_version(servermap, version):
1005 """Download a specific version of the file, using the servermap
1006 as a guide to where the shares are located.
1008 I return a Deferred that fires with the requested contents, or
1009 errbacks with UnrecoverableFileError. Note that a servermap which was
1010 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1011 all versions (those modes stop querying servers as soon as they can
1012 fulfil their goals), so you may want to use MODE_CHECK (which checks
1013 everything) to get increased visibility.
1016 def upload(new_contents, servermap):
1017 """Replace the contents of the file with new ones. This requires a
1018 servermap that was previously updated with MODE_WRITE.
1020 I attempt to provide test-and-set semantics, in that I will avoid
1021 modifying any share that is different than the version I saw in the
1022 servermap. However, if another node is writing to the file at the
1023 same time as me, I may manage to update some shares while they update
1024 others. If I see any evidence of this, I will signal
1025 UncoordinatedWriteError, and the file will be left in an inconsistent
1026 state (possibly the version you provided, possibly the old version,
1027 possibly somebody else's version, and possibly a mix of shares from
1030 The recommended response to UncoordinatedWriteError is to either
1031 return it to the caller (since they failed to coordinate their
1032 writes), or to attempt some sort of recovery. It may be sufficient to
1033 wait a random interval (with exponential backoff) and repeat your
1034 operation. If I do not signal UncoordinatedWriteError, then I was
1035 able to write the new version without incident.
1037 I return a Deferred that fires (with a PublishStatus object) when the
1038 publish has completed. I will update the servermap in-place with the
1039 location of all new shares.
1043 """Return this filenode's writekey, or None if the node does not have
1044 write-capability. This may be used to assist with data structures
1045 that need to make certain data available only to writers, such as the
1046 read-write child caps in dirnodes. The recommended process is to have
1047 reader-visible data be submitted to the filenode in the clear (where
1048 it will be encrypted by the filenode using the readkey), but encrypt
1049 writer-visible data using this writekey.
1053 """Returns the mutable file protocol version."""
1055 class NotEnoughSharesError(Exception):
1056 """Download was unable to get enough shares"""
1058 class NoSharesError(Exception):
1059 """Download was unable to get any shares at all."""
1061 class DownloadStopped(Exception):
1064 class UploadUnhappinessError(Exception):
1065 """Upload was unable to satisfy 'servers_of_happiness'"""
1067 class UnableToFetchCriticalDownloadDataError(Exception):
1068 """I was unable to fetch some piece of critical data which is supposed to
1069 be identically present in all shares."""
1071 class NoServersError(Exception):
1072 """Upload wasn't given any servers to work with, usually indicating a
1073 network or Introducer problem."""
1075 class ExistingChildError(Exception):
1076 """A directory node was asked to add or replace a child that already
1077 exists, and overwrite= was set to False."""
1079 class NoSuchChildError(Exception):
1080 """A directory node was asked to fetch a child which does not exist."""
1082 class ChildOfWrongTypeError(Exception):
1083 """An operation was attempted on a child of the wrong type (file or directory)."""
1085 class IDirectoryNode(IFilesystemNode):
1086 """I represent a filesystem node that is a container, with a
1087 name-to-child mapping, holding the tahoe equivalent of a directory. All
1088 child names are unicode strings, and all children are some sort of
1089 IFilesystemNode (a file, subdirectory, or unknown node).
1094 The dirnode ('1') URI returned by this method can be used in
1095 set_uri() on a different directory ('2') to 'mount' a reference to
1096 this directory ('1') under the other ('2'). This URI is just a
1097 string, so it can be passed around through email or other out-of-band
1101 def get_readonly_uri():
1103 The dirnode ('1') URI returned by this method can be used in
1104 set_uri() on a different directory ('2') to 'mount' a reference to
1105 this directory ('1') under the other ('2'). This URI is just a
1106 string, so it can be passed around through email or other out-of-band
1111 """I return a Deferred that fires with a dictionary mapping child
1112 name (a unicode string) to (node, metadata_dict) tuples, in which
1113 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1116 def has_child(name):
1117 """I return a Deferred that fires with a boolean, True if there
1118 exists a child of the given name, False if not. The child name must
1119 be a unicode string."""
1122 """I return a Deferred that fires with a specific named child node,
1123 which is an IFilesystemNode. The child name must be a unicode string.
1124 I raise NoSuchChildError if I do not have a child by that name."""
1126 def get_metadata_for(name):
1127 """I return a Deferred that fires with the metadata dictionary for
1128 a specific named child node. The child name must be a unicode string.
1129 This metadata is stored in the *edge*, not in the child, so it is
1130 attached to the parent dirnode rather than the child node.
1131 I raise NoSuchChildError if I do not have a child by that name."""
1133 def set_metadata_for(name, metadata):
1134 """I replace any existing metadata for the named child with the new
1135 metadata. The child name must be a unicode string. This metadata is
1136 stored in the *edge*, not in the child, so it is attached to the
1137 parent dirnode rather than the child node. I return a Deferred
1138 (that fires with this dirnode) when the operation is complete.
1139 I raise NoSuchChildError if I do not have a child by that name."""
1141 def get_child_at_path(path):
1142 """Transform a child path into an IFilesystemNode.
1144 I perform a recursive series of 'get' operations to find the named
1145 descendant node. I return a Deferred that fires with the node, or
1146 errbacks with NoSuchChildError if the node could not be found.
1148 The path can be either a single string (slash-separated) or a list of
1149 path-name elements. All elements must be unicode strings.
1152 def get_child_and_metadata_at_path(path):
1153 """Transform a child path into an IFilesystemNode and metadata.
1155 I am like get_child_at_path(), but my Deferred fires with a tuple of
1156 (node, metadata). The metadata comes from the last edge. If the path
1157 is empty, the metadata will be an empty dictionary.
1160 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1161 """I add a child (by writecap+readcap) at the specific name. I return
1162 a Deferred that fires when the operation finishes. If overwrite= is
1163 True, I will replace any existing child of the same name, otherwise
1164 an existing child will cause me to return ExistingChildError. The
1165 child name must be a unicode string.
1167 The child caps could be for a file, or for a directory. If you have
1168 both the writecap and readcap, you should provide both arguments.
1169 If you have only one cap and don't know whether it is read-only,
1170 provide it as the writecap argument and leave the readcap as None.
1171 If you have only one cap that is known to be read-only, provide it
1172 as the readcap argument and leave the writecap as None.
1173 The filecaps are typically obtained from an IFilesystemNode with
1174 get_uri() and get_readonly_uri().
1176 If metadata= is provided, I will use it as the metadata for the named
1177 edge. This will replace any existing metadata. If metadata= is left
1178 as the default value of None, I will set ['mtime'] to the current
1179 time, and I will set ['ctime'] to the current time if there was not
1180 already a child by this name present. This roughly matches the
1181 ctime/mtime semantics of traditional filesystems. See the
1182 "About the metadata" section of webapi.txt for futher information.
1184 If this directory node is read-only, the Deferred will errback with a
1185 NotWriteableError."""
1187 def set_children(entries, overwrite=True):
1188 """Add multiple children (by writecap+readcap) to a directory node.
1189 Takes a dictionary, with childname as keys and (writecap, readcap)
1190 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1191 a Deferred that fires (with this dirnode) when the operation
1192 finishes. This is equivalent to calling set_uri() multiple times, but
1193 is much more efficient. All child names must be unicode strings.
1196 def set_node(name, child, metadata=None, overwrite=True):
1197 """I add a child at the specific name. I return a Deferred that fires
1198 when the operation finishes. This Deferred will fire with the child
1199 node that was just added. I will replace any existing child of the
1200 same name. The child name must be a unicode string. The 'child'
1201 instance must be an instance providing IFilesystemNode.
1203 If metadata= is provided, I will use it as the metadata for the named
1204 edge. This will replace any existing metadata. If metadata= is left
1205 as the default value of None, I will set ['mtime'] to the current
1206 time, and I will set ['ctime'] to the current time if there was not
1207 already a child by this name present. This roughly matches the
1208 ctime/mtime semantics of traditional filesystems. See the
1209 "About the metadata" section of webapi.txt for futher information.
1211 If this directory node is read-only, the Deferred will errback with a
1212 NotWriteableError."""
1214 def set_nodes(entries, overwrite=True):
1215 """Add multiple children to a directory node. Takes a dict mapping
1216 unicode childname to (child_node, metdata) tuples. If metdata=None,
1217 the original metadata is left unmodified. Returns a Deferred that
1218 fires (with this dirnode) when the operation finishes. This is
1219 equivalent to calling set_node() multiple times, but is much more
1222 def add_file(name, uploadable, metadata=None, overwrite=True):
1223 """I upload a file (using the given IUploadable), then attach the
1224 resulting ImmutableFileNode to the directory at the given name. I set
1225 metadata the same way as set_uri and set_node. The child name must be
1228 I return a Deferred that fires (with the IFileNode of the uploaded
1229 file) when the operation completes."""
1231 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1232 """I remove the child at the specific name. I return a Deferred that
1233 fires when the operation finishes. The child name must be a unicode
1234 string. If must_exist is True and I do not have a child by that name,
1235 I raise NoSuchChildError. If must_be_directory is True and the child
1236 is a file, or if must_be_file is True and the child is a directory,
1237 I raise ChildOfWrongTypeError."""
1239 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1240 """I create and attach a directory at the given name. The new
1241 directory can be empty, or it can be populated with children
1242 according to 'initial_children', which takes a dictionary in the same
1243 format as set_nodes (i.e. mapping unicode child name to (childnode,
1244 metadata) tuples). The child name must be a unicode string. I return
1245 a Deferred that fires (with the new directory node) when the
1246 operation finishes."""
1248 def move_child_to(current_child_name, new_parent, new_child_name=None,
1250 """I take one of my children and move them to a new parent. The child
1251 is referenced by name. On the new parent, the child will live under
1252 'new_child_name', which defaults to 'current_child_name'. TODO: what
1253 should we do about metadata? I return a Deferred that fires when the
1254 operation finishes. The child name must be a unicode string. I raise
1255 NoSuchChildError if I do not have a child by that name."""
1257 def build_manifest():
1258 """I generate a table of everything reachable from this directory.
1259 I also compute deep-stats as described below.
1261 I return a Monitor. The Monitor's results will be a dictionary with
1264 res['manifest']: a list of (path, cap) tuples for all nodes
1265 (directories and files) reachable from this one.
1266 'path' will be a tuple of unicode strings. The
1267 origin dirnode will be represented by an empty path
1269 res['verifycaps']: a list of (printable) verifycap strings, one for
1270 each reachable non-LIT node. This is a set:
1271 it will contain no duplicates.
1272 res['storage-index']: a list of (base32) storage index strings,
1273 one for each reachable non-LIT node. This is
1274 a set: it will contain no duplicates.
1275 res['stats']: a dictionary, the same that is generated by
1276 start_deep_stats() below.
1278 The Monitor will also have an .origin_si attribute with the (binary)
1279 storage index of the starting point.
1282 def start_deep_stats():
1283 """Return a Monitor, examining all nodes (directories and files)
1284 reachable from this one. The Monitor's results will be a dictionary
1285 with the following keys::
1287 count-immutable-files: count of how many CHK files are in the set
1288 count-mutable-files: same, for mutable files (does not include
1290 count-literal-files: same, for LIT files
1291 count-files: sum of the above three
1293 count-directories: count of directories
1295 size-immutable-files: total bytes for all CHK files in the set
1296 size-mutable-files (TODO): same, for current version of all mutable
1297 files, does not include directories
1298 size-literal-files: same, for LIT files
1299 size-directories: size of mutable files used by directories
1301 largest-directory: number of bytes in the largest directory
1302 largest-directory-children: number of children in the largest
1304 largest-immutable-file: number of bytes in the largest CHK file
1306 size-mutable-files is not yet implemented, because it would involve
1307 even more queries than deep_stats does.
1309 The Monitor will also have an .origin_si attribute with the (binary)
1310 storage index of the starting point.
1312 This operation will visit every directory node underneath this one,
1313 and can take a long time to run. On a typical workstation with good
1314 bandwidth, this can examine roughly 15 directories per second (and
1315 takes several minutes of 100% CPU for ~1700 directories).
1318 class ICodecEncoder(Interface):
1319 def set_params(data_size, required_shares, max_shares):
1320 """Set up the parameters of this encoder.
1322 This prepares the encoder to perform an operation that converts a
1323 single block of data into a number of shares, such that a future
1324 ICodecDecoder can use a subset of these shares to recover the
1325 original data. This operation is invoked by calling encode(). Once
1326 the encoding parameters are set up, the encode operation can be
1327 invoked multiple times.
1329 set_params() prepares the encoder to accept blocks of input data that
1330 are exactly 'data_size' bytes in length. The encoder will be prepared
1331 to produce 'max_shares' shares for each encode() operation (although
1332 see the 'desired_share_ids' to use less CPU). The encoding math will
1333 be chosen such that the decoder can get by with as few as
1334 'required_shares' of these shares and still reproduce the original
1335 data. For example, set_params(1000, 5, 5) offers no redundancy at
1336 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1338 Numerical Restrictions: 'data_size' is required to be an integral
1339 multiple of 'required_shares'. In general, the caller should choose
1340 required_shares and max_shares based upon their reliability
1341 requirements and the number of peers available (the total storage
1342 space used is roughly equal to max_shares*data_size/required_shares),
1343 then choose data_size to achieve the memory footprint desired (larger
1344 data_size means more efficient operation, smaller data_size means
1345 smaller memory footprint).
1347 In addition, 'max_shares' must be equal to or greater than
1348 'required_shares'. Of course, setting them to be equal causes
1349 encode() to degenerate into a particularly slow form of the 'split'
1352 See encode() for more details about how these parameters are used.
1354 set_params() must be called before any other ICodecEncoder methods
1359 """Return the 3-tuple of data_size, required_shares, max_shares"""
1361 def get_encoder_type():
1362 """Return a short string that describes the type of this encoder.
1364 There is required to be a global table of encoder classes. This method
1365 returns an index into this table; the value at this index is an
1366 encoder class, and this encoder is an instance of that class.
1369 def get_block_size():
1370 """Return the length of the shares that encode() will produce.
1373 def encode_proposal(data, desired_share_ids=None):
1374 """Encode some data.
1376 'data' must be a string (or other buffer object), and len(data) must
1377 be equal to the 'data_size' value passed earlier to set_params().
1379 This will return a Deferred that will fire with two lists. The first
1380 is a list of shares, each of which is a string (or other buffer
1381 object) such that len(share) is the same as what get_share_size()
1382 returned earlier. The second is a list of shareids, in which each is
1383 an integer. The lengths of the two lists will always be equal to each
1384 other. The user should take care to keep each share closely
1385 associated with its shareid, as one is useless without the other.
1387 The length of this output list will normally be the same as the value
1388 provided to the 'max_shares' parameter of set_params(). This may be
1389 different if 'desired_share_ids' is provided.
1391 'desired_share_ids', if provided, is required to be a sequence of
1392 ints, each of which is required to be >= 0 and < max_shares. If not
1393 provided, encode() will produce 'max_shares' shares, as if
1394 'desired_share_ids' were set to range(max_shares). You might use this
1395 if you initially thought you were going to use 10 peers, started
1396 encoding, and then two of the peers dropped out: you could use
1397 desired_share_ids= to skip the work (both memory and CPU) of
1398 producing shares for the peers which are no longer available.
1402 def encode(inshares, desired_share_ids=None):
1403 """Encode some data. This may be called multiple times. Each call is
1406 inshares is a sequence of length required_shares, containing buffers
1407 (i.e. strings), where each buffer contains the next contiguous
1408 non-overlapping segment of the input data. Each buffer is required to
1409 be the same length, and the sum of the lengths of the buffers is
1410 required to be exactly the data_size promised by set_params(). (This
1411 implies that the data has to be padded before being passed to
1412 encode(), unless of course it already happens to be an even multiple
1413 of required_shares in length.)
1415 Note: the requirement to break up your data into
1416 'required_shares' chunks of exactly the right length before
1417 calling encode() is surprising from point of view of a user
1418 who doesn't know how FEC works. It feels like an
1419 implementation detail that has leaked outside the abstraction
1420 barrier. Is there a use case in which the data to be encoded
1421 might already be available in pre-segmented chunks, such that
1422 it is faster or less work to make encode() take a list rather
1423 than splitting a single string?
1425 Yes, there is: suppose you are uploading a file with K=64,
1426 N=128, segsize=262,144. Then each in-share will be of size
1427 4096. If you use this .encode() API then your code could first
1428 read each successive 4096-byte chunk from the file and store
1429 each one in a Python string and store each such Python string
1430 in a Python list. Then you could call .encode(), passing that
1431 list as "inshares". The encoder would generate the other 64
1432 "secondary shares" and return to you a new list containing
1433 references to the same 64 Python strings that you passed in
1434 (as the primary shares) plus references to the new 64 Python
1437 (You could even imagine that your code could use readv() so
1438 that the operating system can arrange to get all of those
1439 bytes copied from the file into the Python list of Python
1440 strings as efficiently as possible instead of having a loop
1441 written in C or in Python to copy the next part of the file
1442 into the next string.)
1444 On the other hand if you instead use the .encode_proposal()
1445 API (above), then your code can first read in all of the
1446 262,144 bytes of the segment from the file into a Python
1447 string, then call .encode_proposal() passing the segment data
1448 as the "data" argument. The encoder would basically first
1449 split the "data" argument into a list of 64 in-shares of 4096
1450 byte each, and then do the same thing that .encode() does. So
1451 this would result in a little bit more copying of data and a
1452 little bit higher of a "maximum memory usage" during the
1453 process, although it might or might not make a practical
1454 difference for our current use cases.
1456 Note that "inshares" is a strange name for the parameter if
1457 you think of the parameter as being just for feeding in data
1458 to the codec. It makes more sense if you think of the result
1459 of this encoding as being the set of shares from inshares plus
1460 an extra set of "secondary shares" (or "check shares"). It is
1461 a surprising name! If the API is going to be surprising then
1462 the name should be surprising. If we switch to
1463 encode_proposal() above then we should also switch to an
1466 'desired_share_ids', if provided, is required to be a sequence of
1467 ints, each of which is required to be >= 0 and < max_shares. If not
1468 provided, encode() will produce 'max_shares' shares, as if
1469 'desired_share_ids' were set to range(max_shares). You might use this
1470 if you initially thought you were going to use 10 peers, started
1471 encoding, and then two of the peers dropped out: you could use
1472 desired_share_ids= to skip the work (both memory and CPU) of
1473 producing shares for the peers which are no longer available.
1475 For each call, encode() will return a Deferred that fires with two
1476 lists, one containing shares and the other containing the shareids.
1477 The get_share_size() method can be used to determine the length of
1478 the share strings returned by encode(). Each shareid is a small
1479 integer, exactly as passed into 'desired_share_ids' (or
1480 range(max_shares), if desired_share_ids was not provided).
1482 The shares and their corresponding shareids are required to be kept
1483 together during storage and retrieval. Specifically, the share data is
1484 useless by itself: the decoder needs to be told which share is which
1485 by providing it with both the shareid and the actual share data.
1487 This function will allocate an amount of memory roughly equal to::
1489 (max_shares - required_shares) * get_share_size()
1491 When combined with the memory that the caller must allocate to
1492 provide the input data, this leads to a memory footprint roughly
1493 equal to the size of the resulting encoded shares (i.e. the expansion
1494 factor times the size of the input segment).
1499 # returning a list of (shareidN,shareN) tuples instead of a pair of
1500 # lists (shareids..,shares..). Brian thought the tuples would
1501 # encourage users to keep the share and shareid together throughout
1502 # later processing, Zooko pointed out that the code to iterate
1503 # through two lists is not really more complicated than using a list
1504 # of tuples and there's also a performance improvement
1506 # having 'data_size' not required to be an integral multiple of
1507 # 'required_shares'. Doing this would require encode() to perform
1508 # padding internally, and we'd prefer to have any padding be done
1509 # explicitly by the caller. Yes, it is an abstraction leak, but
1510 # hopefully not an onerous one.
1513 class ICodecDecoder(Interface):
1514 def set_params(data_size, required_shares, max_shares):
1515 """Set the params. They have to be exactly the same ones that were
1516 used for encoding."""
1518 def get_needed_shares():
1519 """Return the number of shares needed to reconstruct the data.
1520 set_params() is required to be called before this."""
1522 def decode(some_shares, their_shareids):
1523 """Decode a partial list of shares into data.
1525 'some_shares' is required to be a sequence of buffers of sharedata, a
1526 subset of the shares returned by ICodecEncode.encode(). Each share is
1527 required to be of the same length. The i'th element of their_shareids
1528 is required to be the shareid of the i'th buffer in some_shares.
1530 This returns a Deferred which fires with a sequence of buffers. This
1531 sequence will contain all of the segments of the original data, in
1532 order. The sum of the lengths of all of the buffers will be the
1533 'data_size' value passed into the original ICodecEncode.set_params()
1534 call. To get back the single original input block of data, use
1535 ''.join(output_buffers), or you may wish to simply write them in
1536 order to an output file.
1538 Note that some of the elements in the result sequence may be
1539 references to the elements of the some_shares input sequence. In
1540 particular, this means that if those share objects are mutable (e.g.
1541 arrays) and if they are changed, then both the input (the
1542 'some_shares' parameter) and the output (the value given when the
1543 deferred is triggered) will change.
1545 The length of 'some_shares' is required to be exactly the value of
1546 'required_shares' passed into the original ICodecEncode.set_params()
1550 class IEncoder(Interface):
1551 """I take an object that provides IEncryptedUploadable, which provides
1552 encrypted data, and a list of shareholders. I then encode, hash, and
1553 deliver shares to those shareholders. I will compute all the necessary
1554 Merkle hash trees that are necessary to validate the crypttext that
1555 eventually comes back from the shareholders. I provide the URI Extension
1556 Block Hash, and the encoding parameters, both of which must be included
1559 I do not choose shareholders, that is left to the IUploader. I must be
1560 given a dict of RemoteReferences to storage buckets that are ready and
1561 willing to receive data.
1565 """Specify the number of bytes that will be encoded. This must be
1566 peformed before get_serialized_params() can be called.
1568 def set_params(params):
1569 """Override the default encoding parameters. 'params' is a tuple of
1570 (k,d,n), where 'k' is the number of required shares, 'd' is the
1571 servers_of_happiness, and 'n' is the total number of shares that will
1574 Encoding parameters can be set in three ways. 1: The Encoder class
1575 provides defaults (3/7/10). 2: the Encoder can be constructed with
1576 an 'options' dictionary, in which the
1577 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1578 set_params((k,d,n)) can be called.
1580 If you intend to use set_params(), you must call it before
1581 get_share_size or get_param are called.
1584 def set_encrypted_uploadable(u):
1585 """Provide a source of encrypted upload data. 'u' must implement
1586 IEncryptedUploadable.
1588 When this is called, the IEncryptedUploadable will be queried for its
1589 length and the storage_index that should be used.
1591 This returns a Deferred that fires with this Encoder instance.
1593 This must be performed before start() can be called.
1596 def get_param(name):
1597 """Return an encoding parameter, by name.
1599 'storage_index': return a string with the (16-byte truncated SHA-256
1600 hash) storage index to which these shares should be
1603 'share_counts': return a tuple describing how many shares are used:
1604 (needed_shares, servers_of_happiness, total_shares)
1606 'num_segments': return an int with the number of segments that
1609 'segment_size': return an int with the size of each segment.
1611 'block_size': return the size of the individual blocks that will
1612 be delivered to a shareholder's put_block() method. By
1613 knowing this, the shareholder will be able to keep all
1614 blocks in a single file and still provide random access
1615 when reading them. # TODO: can we avoid exposing this?
1617 'share_size': an int with the size of the data that will be stored
1618 on each shareholder. This is aggregate amount of data
1619 that will be sent to the shareholder, summed over all
1620 the put_block() calls I will ever make. It is useful to
1621 determine this size before asking potential
1622 shareholders whether they will grant a lease or not,
1623 since their answers will depend upon how much space we
1624 need. TODO: this might also include some amount of
1625 overhead, like the size of all the hashes. We need to
1626 decide whether this is useful or not.
1628 'serialized_params': a string with a concise description of the
1629 codec name and its parameters. This may be passed
1630 into the IUploadable to let it make sure that
1631 the same file encoded with different parameters
1632 will result in different storage indexes.
1634 Once this is called, set_size() and set_params() may not be called.
1637 def set_shareholders(shareholders, servermap):
1638 """Tell the encoder where to put the encoded shares. 'shareholders'
1639 must be a dictionary that maps share number (an integer ranging from
1640 0 to n-1) to an instance that provides IStorageBucketWriter.
1641 'servermap' is a dictionary that maps share number (as defined above)
1642 to a set of peerids. This must be performed before start() can be
1646 """Begin the encode/upload process. This involves reading encrypted
1647 data from the IEncryptedUploadable, encoding it, uploading the shares
1648 to the shareholders, then sending the hash trees.
1650 set_encrypted_uploadable() and set_shareholders() must be called
1651 before this can be invoked.
1653 This returns a Deferred that fires with a verify cap when the upload
1654 process is complete. The verifycap, plus the encryption key, is
1655 sufficient to construct the read cap.
1658 class IDecoder(Interface):
1659 """I take a list of shareholders and some setup information, then
1660 download, validate, decode, and decrypt data from them, writing the
1661 results to an output file.
1663 I do not locate the shareholders, that is left to the IDownloader. I must
1664 be given a dict of RemoteReferences to storage buckets that are ready to
1669 """I take a file-like object (providing write and close) to which all
1670 the plaintext data will be written.
1672 TODO: producer/consumer . Maybe write() should return a Deferred that
1673 indicates when it will accept more data? But probably having the
1674 IDecoder be a producer is easier to glue to IConsumer pieces.
1677 def set_shareholders(shareholders):
1678 """I take a dictionary that maps share identifiers (small integers)
1679 to RemoteReferences that provide RIBucketReader. This must be called
1683 """I start the download. This process involves retrieving data and
1684 hash chains from the shareholders, using the hashes to validate the
1685 data, decoding the shares into segments, decrypting the segments,
1686 then writing the resulting plaintext to the output file.
1688 I return a Deferred that will fire (with self) when the download is
1692 class IDownloadTarget(Interface):
1693 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1694 # will register itself as a producer. This allows the target to invoke
1695 # downloader.pauseProducing, resumeProducing, and stopProducing.
1697 """Called before any calls to write() or close(). If an error
1698 occurs before any data is available, fail() may be called without
1699 a previous call to open().
1701 'size' is the length of the file being downloaded, in bytes."""
1704 """Output some data to the target."""
1706 """Inform the target that there is no more data to be written."""
1708 """fail() is called to indicate that the download has failed. 'why'
1709 is a Failure object indicating what went wrong. No further methods
1710 will be invoked on the IDownloadTarget after fail()."""
1711 def register_canceller(cb):
1712 """The CiphertextDownloader uses this to register a no-argument function
1713 that the target can call to cancel the download. Once this canceller
1714 is invoked, no further calls to write() or close() will be made."""
1716 """When the CiphertextDownloader is done, this finish() function will be
1717 called. Whatever it returns will be returned to the invoker of
1718 Downloader.download.
1721 class IDownloader(Interface):
1722 def download(uri, target):
1723 """Perform a CHK download, sending the data to the given target.
1724 'target' must provide IDownloadTarget.
1726 Returns a Deferred that fires (with the results of target.finish)
1727 when the download is finished, or errbacks if something went wrong."""
1729 class IEncryptedUploadable(Interface):
1730 def set_upload_status(upload_status):
1731 """Provide an IUploadStatus object that should be filled with status
1732 information. The IEncryptedUploadable is responsible for setting
1733 key-determination progress ('chk'), size, storage_index, and
1734 ciphertext-fetch progress. It may delegate some of this
1735 responsibility to others, in particular to the IUploadable."""
1738 """This behaves just like IUploadable.get_size()."""
1740 def get_all_encoding_parameters():
1741 """Return a Deferred that fires with a tuple of
1742 (k,happy,n,segment_size). The segment_size will be used as-is, and
1743 must match the following constraints: it must be a multiple of k, and
1744 it shouldn't be unreasonably larger than the file size (if
1745 segment_size is larger than filesize, the difference must be stored
1748 This usually passes through to the IUploadable method of the same
1751 The encoder strictly obeys the values returned by this method. To
1752 make an upload use non-default encoding parameters, you must arrange
1753 to control the values that this method returns.
1756 def get_storage_index():
1757 """Return a Deferred that fires with a 16-byte storage index.
1760 def read_encrypted(length, hash_only):
1761 """This behaves just like IUploadable.read(), but returns crypttext
1762 instead of plaintext. If hash_only is True, then this discards the
1763 data (and returns an empty list); this improves efficiency when
1764 resuming an interrupted upload (where we need to compute the
1765 plaintext hashes, but don't need the redundant encrypted data)."""
1767 def get_plaintext_hashtree_leaves(first, last, num_segments):
1768 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1769 plaintext segments, i.e. get the tagged hashes of the given segments.
1770 The segment size is expected to be generated by the
1771 IEncryptedUploadable before any plaintext is read or ciphertext
1772 produced, so that the segment hashes can be generated with only a
1775 This returns a Deferred which fires with a sequence of hashes, using:
1777 tuple(segment_hashes[first:last])
1779 'num_segments' is used to assert that the number of segments that the
1780 IEncryptedUploadable handled matches the number of segments that the
1781 encoder was expecting.
1783 This method must not be called until the final byte has been read
1784 from read_encrypted(). Once this method is called, read_encrypted()
1785 can never be called again.
1788 def get_plaintext_hash():
1789 """OBSOLETE; Get the hash of the whole plaintext.
1791 This returns a Deferred which fires with a tagged SHA-256 hash of the
1792 whole plaintext, obtained from hashutil.plaintext_hash(data).
1796 """Just like IUploadable.close()."""
1798 class IUploadable(Interface):
1799 def set_upload_status(upload_status):
1800 """Provide an IUploadStatus object that should be filled with status
1801 information. The IUploadable is responsible for setting
1802 key-determination progress ('chk')."""
1804 def set_default_encoding_parameters(params):
1805 """Set the default encoding parameters, which must be a dict mapping
1806 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1807 'max_segment_size'. These might have an influence on the final
1808 encoding parameters returned by get_all_encoding_parameters(), if the
1809 Uploadable doesn't have more specific preferences.
1811 This call is optional: if it is not used, the Uploadable will use
1812 some built-in defaults. If used, this method must be called before
1813 any other IUploadable methods to have any effect.
1817 """Return a Deferred that will fire with the length of the data to be
1818 uploaded, in bytes. This will be called before the data is actually
1819 used, to compute encoding parameters.
1822 def get_all_encoding_parameters():
1823 """Return a Deferred that fires with a tuple of
1824 (k,happy,n,segment_size). The segment_size will be used as-is, and
1825 must match the following constraints: it must be a multiple of k, and
1826 it shouldn't be unreasonably larger than the file size (if
1827 segment_size is larger than filesize, the difference must be stored
1830 The relative values of k and n allow some IUploadables to request
1831 better redundancy than others (in exchange for consuming more space
1834 Larger values of segment_size reduce hash overhead, while smaller
1835 values reduce memory footprint and cause data to be delivered in
1836 smaller pieces (which may provide a smoother and more predictable
1837 download experience).
1839 The encoder strictly obeys the values returned by this method. To
1840 make an upload use non-default encoding parameters, you must arrange
1841 to control the values that this method returns. One way to influence
1842 them may be to call set_encoding_parameters() before calling
1843 get_all_encoding_parameters().
1846 def get_encryption_key():
1847 """Return a Deferred that fires with a 16-byte AES key. This key will
1848 be used to encrypt the data. The key will also be hashed to derive
1851 Uploadables which want to achieve convergence should hash their file
1852 contents and the serialized_encoding_parameters to form the key
1853 (which of course requires a full pass over the data). Uploadables can
1854 use the upload.ConvergentUploadMixin class to achieve this
1857 Uploadables which do not care about convergence (or do not wish to
1858 make multiple passes over the data) can simply return a
1859 strongly-random 16 byte string.
1861 get_encryption_key() may be called multiple times: the IUploadable is
1862 required to return the same value each time.
1866 """Return a Deferred that fires with a list of strings (perhaps with
1867 only a single element) which, when concatenated together, contain the
1868 next 'length' bytes of data. If EOF is near, this may provide fewer
1869 than 'length' bytes. The total number of bytes provided by read()
1870 before it signals EOF must equal the size provided by get_size().
1872 If the data must be acquired through multiple internal read
1873 operations, returning a list instead of a single string may help to
1874 reduce string copies. However, the length of the concatenated strings
1875 must equal the amount of data requested, unless EOF is encountered.
1876 Long reads, or short reads without EOF, are not allowed. read()
1877 should return the same amount of data as a local disk file read, just
1878 in a different shape and asynchronously.
1880 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1881 so a 10kB file means length=3kB, 100kB file means length=30kB,
1882 and >=1MB file means length=300kB.
1884 This method provides for a single full pass through the data. Later
1885 use cases may desire multiple passes or access to only parts of the
1886 data (such as a mutable file making small edits-in-place). This API
1887 will be expanded once those use cases are better understood.
1891 """The upload is finished, and whatever filehandle was in use may be
1895 class IMutableUploadable(Interface):
1897 I represent content that is due to be uploaded to a mutable filecap.
1899 # This is somewhat simpler than the IUploadable interface above
1900 # because mutable files do not need to be concerned with possibly
1901 # generating a CHK, nor with per-file keys. It is a subset of the
1902 # methods in IUploadable, though, so we could just as well implement
1903 # the mutable uploadables as IUploadables that don't happen to use
1904 # those methods (with the understanding that the unused methods will
1905 # never be called on such objects)
1908 Returns a Deferred that fires with the size of the content held
1914 Returns a list of strings which, when concatenated, are the next
1915 length bytes of the file, or fewer if there are fewer bytes
1916 between the current location and the end of the file.
1921 The process that used the Uploadable is finished using it, so
1922 the uploadable may be closed.
1925 class IUploadResults(Interface):
1926 """I am returned by upload() methods. I contain a number of public
1927 attributes which can be read to determine the results of the upload. Some
1928 of these are functional, some are timing information. All of these may be
1931 .file_size : the size of the file, in bytes
1932 .uri : the CHK read-cap for the file
1933 .ciphertext_fetched : how many bytes were fetched by the helper
1934 .sharemap: dict mapping share identifier to set of serverids
1935 (binary strings). This indicates which servers were given
1936 which shares. For immutable files, the shareid is an
1937 integer (the share number, from 0 to N-1). For mutable
1938 files, it is a string of the form 'seq%d-%s-sh%d',
1939 containing the sequence number, the roothash, and the
1941 .servermap : dict mapping server peerid to a set of share numbers
1942 .timings : dict of timing information, mapping name to seconds (float)
1943 total : total upload time, start to finish
1944 storage_index : time to compute the storage index
1945 peer_selection : time to decide which peers will be used
1946 contacting_helper : initial helper query to upload/no-upload decision
1947 existence_check : helper pre-upload existence check
1948 helper_total : initial helper query to helper finished pushing
1949 cumulative_fetch : helper waiting for ciphertext requests
1950 total_fetch : helper start to last ciphertext response
1951 cumulative_encoding : just time spent in zfec
1952 cumulative_sending : just time spent waiting for storage servers
1953 hashes_and_close : last segment push to shareholder close
1954 total_encode_and_push : first encode to shareholder close
1958 class IDownloadResults(Interface):
1959 """I am created internally by download() methods. I contain a number of
1960 public attributes which contain details about the download process.::
1962 .file_size : the size of the file, in bytes
1963 .servers_used : set of server peerids that were used during download
1964 .server_problems : dict mapping server peerid to a problem string. Only
1965 servers that had problems (bad hashes, disconnects)
1967 .servermap : dict mapping server peerid to a set of share numbers. Only
1968 servers that had any shares are listed here.
1969 .timings : dict of timing information, mapping name to seconds (float)
1970 peer_selection : time to ask servers about shares
1971 servers_peer_selection : dict of peerid to DYHB-query time
1972 uri_extension : time to fetch a copy of the URI extension block
1973 hashtrees : time to fetch the hash trees
1974 segments : time to fetch, decode, and deliver segments
1975 cumulative_fetch : time spent waiting for storage servers
1976 cumulative_decode : just time spent in zfec
1977 cumulative_decrypt : just time spent in decryption
1978 total : total download time, start to finish
1979 fetch_per_server : dict of peerid to list of per-segment fetch times
1983 class IUploader(Interface):
1984 def upload(uploadable):
1985 """Upload the file. 'uploadable' must impement IUploadable. This
1986 returns a Deferred which fires with an IUploadResults instance, from
1987 which the URI of the file can be obtained as results.uri ."""
1989 def upload_ssk(write_capability, new_version, uploadable):
1990 """TODO: how should this work?"""
1992 class ICheckable(Interface):
1993 def check(monitor, verify=False, add_lease=False):
1994 """Check up on my health, optionally repairing any problems.
1996 This returns a Deferred that fires with an instance that provides
1997 ICheckResults, or None if the object is non-distributed (i.e. LIT
2000 The monitor will be checked periodically to see if the operation has
2001 been cancelled. If so, no new queries will be sent, and the Deferred
2002 will fire (with a OperationCancelledError) immediately.
2004 Filenodes and dirnodes (which provide IFilesystemNode) are also
2005 checkable. Instances that represent verifier-caps will be checkable
2006 but not downloadable. Some objects (like LIT files) do not actually
2007 live in the grid, and their checkers return None (non-distributed
2008 files are always healthy).
2010 If verify=False, a relatively lightweight check will be performed: I
2011 will ask all servers if they have a share for me, and I will believe
2012 whatever they say. If there are at least N distinct shares on the
2013 grid, my results will indicate r.is_healthy()==True. This requires a
2014 roundtrip to each server, but does not transfer very much data, so
2015 the network bandwidth is fairly low.
2017 If verify=True, a more resource-intensive check will be performed:
2018 every share will be downloaded, and the hashes will be validated on
2019 every bit. I will ignore any shares that failed their hash checks. If
2020 there are at least N distinct valid shares on the grid, my results
2021 will indicate r.is_healthy()==True. This requires N/k times as much
2022 download bandwidth (and server disk IO) as a regular download. If a
2023 storage server is holding a corrupt share, or is experiencing memory
2024 failures during retrieval, or is malicious or buggy, then
2025 verification will detect the problem, but checking will not.
2027 If add_lease=True, I will ensure that an up-to-date lease is present
2028 on each share. The lease secrets will be derived from by node secret
2029 (in BASEDIR/private/secret), so either I will add a new lease to the
2030 share, or I will merely renew the lease that I already had. In a
2031 future version of the storage-server protocol (once Accounting has
2032 been implemented), there may be additional options here to define the
2033 kind of lease that is obtained (which account number to claim, etc).
2035 TODO: any problems seen during checking will be reported to the
2036 health-manager.furl, a centralized object which is responsible for
2037 figuring out why files are unhealthy so corrective action can be
2041 def check_and_repair(monitor, verify=False, add_lease=False):
2042 """Like check(), but if the file/directory is not healthy, attempt to
2045 Any non-healthy result will cause an immediate repair operation, to
2046 generate and upload new shares. After repair, the file will be as
2047 healthy as we can make it. Details about what sort of repair is done
2048 will be put in the check-and-repair results. The Deferred will not
2049 fire until the repair is complete.
2051 This returns a Deferred which fires with an instance of
2052 ICheckAndRepairResults."""
2054 class IDeepCheckable(Interface):
2055 def start_deep_check(verify=False, add_lease=False):
2056 """Check upon the health of me and everything I can reach.
2058 This is a recursive form of check(), useable only on dirnodes.
2060 I return a Monitor, with results that are an IDeepCheckResults
2063 TODO: If any of the directories I traverse are unrecoverable, the
2064 Monitor will report failure. If any of the files I check upon are
2065 unrecoverable, those problems will be reported in the
2066 IDeepCheckResults as usual, and the Monitor will not report a
2070 def start_deep_check_and_repair(verify=False, add_lease=False):
2071 """Check upon the health of me and everything I can reach. Repair
2072 anything that isn't healthy.
2074 This is a recursive form of check_and_repair(), useable only on
2077 I return a Monitor, with results that are an
2078 IDeepCheckAndRepairResults object.
2080 TODO: If any of the directories I traverse are unrecoverable, the
2081 Monitor will report failure. If any of the files I check upon are
2082 unrecoverable, those problems will be reported in the
2083 IDeepCheckResults as usual, and the Monitor will not report a
2087 class ICheckResults(Interface):
2088 """I contain the detailed results of a check/verify operation.
2091 def get_storage_index():
2092 """Return a string with the (binary) storage index."""
2093 def get_storage_index_string():
2094 """Return a string with the (printable) abbreviated storage index."""
2096 """Return the (string) URI of the object that was checked."""
2099 """Return a boolean, True if the file/dir is fully healthy, False if
2100 it is damaged in any way. Non-distributed LIT files always return
2103 def is_recoverable():
2104 """Return a boolean, True if the file/dir can be recovered, False if
2105 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2106 files always return True."""
2108 def needs_rebalancing():
2109 """Return a boolean, True if the file/dir's reliability could be
2110 improved by moving shares to new servers. Non-distributed LIT files
2111 always return False."""
2115 """Return a dictionary that describes the state of the file/dir. LIT
2116 files always return an empty dictionary. Normal files and directories
2117 return a dictionary with the following keys (note that these use
2118 binary strings rather than base32-encoded ones) (also note that for
2119 mutable files, these counts are for the 'best' version):
2121 count-shares-good: the number of distinct good shares that were found
2122 count-shares-needed: 'k', the number of shares required for recovery
2123 count-shares-expected: 'N', the number of total shares generated
2124 count-good-share-hosts: the number of distinct storage servers with
2125 good shares. If this number is less than
2126 count-shares-good, then some shares are
2127 doubled up, increasing the correlation of
2128 failures. This indicates that one or more
2129 shares should be moved to an otherwise unused
2130 server, if one is available.
2131 count-corrupt-shares: the number of shares with integrity failures
2132 list-corrupt-shares: a list of 'share locators', one for each share
2133 that was found to be corrupt. Each share
2134 locator is a list of (serverid, storage_index,
2136 count-incompatible-shares: the number of shares which are of a share
2137 format unknown to this checker
2138 list-incompatible-shares: a list of 'share locators', one for each
2139 share that was found to be of an unknown
2140 format. Each share locator is a list of
2141 (serverid, storage_index, sharenum).
2142 servers-responding: list of (binary) storage server identifiers,
2143 one for each server which responded to the share
2144 query (even if they said they didn't have
2145 shares, and even if they said they did have
2146 shares but then didn't send them when asked, or
2147 dropped the connection, or returned a Failure,
2148 and even if they said they did have shares and
2149 sent incorrect ones when asked)
2150 sharemap: dict mapping share identifier to list of serverids
2151 (binary strings). This indicates which servers are holding
2152 which shares. For immutable files, the shareid is an
2153 integer (the share number, from 0 to N-1). For mutable
2154 files, it is a string of the form 'seq%d-%s-sh%d',
2155 containing the sequence number, the roothash, and the
2158 The following keys are most relevant for mutable files, but immutable
2159 files will provide sensible values too::
2161 count-wrong-shares: the number of shares for versions other than the
2162 'best' one (which is defined as being the
2163 recoverable version with the highest sequence
2164 number, then the highest roothash). These are
2165 either leftover shares from an older version
2166 (perhaps on a server that was offline when an
2167 update occurred), shares from an unrecoverable
2168 newer version, or shares from an alternate
2169 current version that results from an
2170 uncoordinated write collision. For a healthy
2171 file, this will equal 0.
2173 count-recoverable-versions: the number of recoverable versions of
2174 the file. For a healthy file, this will
2177 count-unrecoverable-versions: the number of unrecoverable versions
2178 of the file. For a healthy file, this
2184 """Return a string with a brief (one-line) summary of the results."""
2187 """Return a list of strings with more detailed results."""
2189 class ICheckAndRepairResults(Interface):
2190 """I contain the detailed results of a check/verify/repair operation.
2192 The IFilesystemNode.check()/verify()/repair() methods all return
2193 instances that provide ICheckAndRepairResults.
2196 def get_storage_index():
2197 """Return a string with the (binary) storage index."""
2198 def get_storage_index_string():
2199 """Return a string with the (printable) abbreviated storage index."""
2200 def get_repair_attempted():
2201 """Return a boolean, True if a repair was attempted. We might not
2202 attempt to repair the file because it was healthy, or healthy enough
2203 (i.e. some shares were missing but not enough to exceed some
2204 threshold), or because we don't know how to repair this object."""
2205 def get_repair_successful():
2206 """Return a boolean, True if repair was attempted and the file/dir
2207 was fully healthy afterwards. False if no repair was attempted or if
2208 a repair attempt failed."""
2209 def get_pre_repair_results():
2210 """Return an ICheckResults instance that describes the state of the
2211 file/dir before any repair was attempted."""
2212 def get_post_repair_results():
2213 """Return an ICheckResults instance that describes the state of the
2214 file/dir after any repair was attempted. If no repair was attempted,
2215 the pre-repair and post-repair results will be identical."""
2218 class IDeepCheckResults(Interface):
2219 """I contain the results of a deep-check operation.
2221 This is returned by a call to ICheckable.deep_check().
2224 def get_root_storage_index_string():
2225 """Return the storage index (abbreviated human-readable string) of
2226 the first object checked."""
2228 """Return a dictionary with the following keys::
2230 count-objects-checked: count of how many objects were checked
2231 count-objects-healthy: how many of those objects were completely
2233 count-objects-unhealthy: how many were damaged in some way
2234 count-objects-unrecoverable: how many were unrecoverable
2235 count-corrupt-shares: how many shares were found to have
2236 corruption, summed over all objects
2240 def get_corrupt_shares():
2241 """Return a set of (serverid, storage_index, sharenum) for all shares
2242 that were found to be corrupt. Both serverid and storage_index are
2245 def get_all_results():
2246 """Return a dictionary mapping pathname (a tuple of strings, ready to
2247 be slash-joined) to an ICheckResults instance, one for each object
2248 that was checked."""
2250 def get_results_for_storage_index(storage_index):
2251 """Retrive the ICheckResults instance for the given (binary)
2252 storage index. Raises KeyError if there are no results for that
2256 """Return a dictionary with the same keys as
2257 IDirectoryNode.deep_stats()."""
2259 class IDeepCheckAndRepairResults(Interface):
2260 """I contain the results of a deep-check-and-repair operation.
2262 This is returned by a call to ICheckable.deep_check_and_repair().
2265 def get_root_storage_index_string():
2266 """Return the storage index (abbreviated human-readable string) of
2267 the first object checked."""
2269 """Return a dictionary with the following keys::
2271 count-objects-checked: count of how many objects were checked
2272 count-objects-healthy-pre-repair: how many of those objects were
2273 completely healthy (before any
2275 count-objects-unhealthy-pre-repair: how many were damaged in
2277 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2278 count-objects-healthy-post-repair: how many of those objects were
2279 completely healthy (after any
2281 count-objects-unhealthy-post-repair: how many were damaged in
2283 count-objects-unrecoverable-post-repair: how many were
2285 count-repairs-attempted: repairs were attempted on this many
2286 objects. The count-repairs- keys will
2287 always be provided, however unless
2288 repair=true is present, they will all
2290 count-repairs-successful: how many repairs resulted in healthy
2292 count-repairs-unsuccessful: how many repairs resulted did not
2293 results in completely healthy objects
2294 count-corrupt-shares-pre-repair: how many shares were found to
2295 have corruption, summed over all
2296 objects examined (before any
2298 count-corrupt-shares-post-repair: how many shares were found to
2299 have corruption, summed over all
2300 objects examined (after any
2305 """Return a dictionary with the same keys as
2306 IDirectoryNode.deep_stats()."""
2308 def get_corrupt_shares():
2309 """Return a set of (serverid, storage_index, sharenum) for all shares
2310 that were found to be corrupt before any repair was attempted. Both
2311 serverid and storage_index are binary.
2313 def get_remaining_corrupt_shares():
2314 """Return a set of (serverid, storage_index, sharenum) for all shares
2315 that were found to be corrupt after any repair was completed. Both
2316 serverid and storage_index are binary. These are shares that need
2317 manual inspection and probably deletion.
2319 def get_all_results():
2320 """Return a dictionary mapping pathname (a tuple of strings, ready to
2321 be slash-joined) to an ICheckAndRepairResults instance, one for each
2322 object that was checked."""
2324 def get_results_for_storage_index(storage_index):
2325 """Retrive the ICheckAndRepairResults instance for the given (binary)
2326 storage index. Raises KeyError if there are no results for that
2330 class IRepairable(Interface):
2331 def repair(check_results):
2332 """Attempt to repair the given object. Returns a Deferred that fires
2333 with a IRepairResults object.
2335 I must be called with an object that implements ICheckResults, as
2336 proof that you have actually discovered a problem with this file. I
2337 will use the data in the checker results to guide the repair process,
2338 such as which servers provided bad data and should therefore be
2339 avoided. The ICheckResults object is inside the
2340 ICheckAndRepairResults object, which is returned by the
2341 ICheckable.check() method::
2343 d = filenode.check(repair=False)
2344 def _got_results(check_and_repair_results):
2345 check_results = check_and_repair_results.get_pre_repair_results()
2346 return filenode.repair(check_results)
2347 d.addCallback(_got_results)
2351 class IRepairResults(Interface):
2352 """I contain the results of a repair operation."""
2353 def get_successful(self):
2354 """Returns a boolean: True if the repair made the file healthy, False
2355 if not. Repair failure generally indicates a file that has been
2356 damaged beyond repair."""
2359 class IClient(Interface):
2360 def upload(uploadable):
2361 """Upload some data into a CHK, get back the UploadResults for it.
2362 @param uploadable: something that implements IUploadable
2363 @return: a Deferred that fires with the UploadResults instance.
2364 To get the URI for this file, use results.uri .
2367 def create_mutable_file(contents=""):
2368 """Create a new mutable file (with initial) contents, get back the
2371 @param contents: (bytestring, callable, or None): this provides the
2372 initial contents of the mutable file. If 'contents' is a bytestring,
2373 it will be used as-is. If 'contents' is a callable, it will be
2374 invoked with the new MutableFileNode instance and is expected to
2375 return a bytestring with the initial contents of the file (the
2376 callable can use node.get_writekey() to decide how to encrypt the
2377 initial contents, e.g. for a brand new dirnode with initial
2378 children). contents=None is equivalent to an empty string. Using
2379 content_maker= is more efficient than creating a mutable file and
2380 setting its contents in two separate operations.
2382 @return: a Deferred that fires with an IMutableFileNode instance.
2385 def create_dirnode(initial_children={}):
2386 """Create a new unattached dirnode, possibly with initial children.
2388 @param initial_children: dict with keys that are unicode child names,
2389 and values that are (childnode, metadata) tuples.
2391 @return: a Deferred that fires with the new IDirectoryNode instance.
2394 def create_node_from_uri(uri, rouri):
2395 """Create a new IFilesystemNode instance from the uri, synchronously.
2396 @param uri: a string or IURI-providing instance, or None. This could
2397 be for a LiteralFileNode, a CHK file node, a mutable file
2398 node, or a directory node
2399 @param rouri: a string or IURI-providing instance, or None. If the
2400 main uri is None, I will use the rouri instead. If I
2401 recognize the format of the main uri, I will ignore the
2402 rouri (because it can be derived from the writecap).
2404 @return: an instance that provides IFilesystemNode (or more usefully
2405 one of its subclasses). File-specifying URIs will result in
2406 IFileNode-providing instances, like ImmutableFileNode,
2407 LiteralFileNode, or MutableFileNode. Directory-specifying
2408 URIs will result in IDirectoryNode-providing instances, like
2412 class INodeMaker(Interface):
2413 """The NodeMaker is used to create IFilesystemNode instances. It can
2414 accept a filecap/dircap string and return the node right away. It can
2415 also create new nodes (i.e. upload a file, or create a mutable file)
2416 asynchronously. Once you have one of these nodes, you can use other
2417 methods to determine whether it is a file or directory, and to download
2418 or modify its contents.
2420 The NodeMaker encapsulates all the authorities that these
2421 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2422 Tahoe process will typically have a single NodeMaker, but unit tests may
2423 create simplified/mocked forms for testing purposes.
2425 def create_from_cap(writecap, readcap=None, **kwargs):
2426 """I create an IFilesystemNode from the given writecap/readcap. I can
2427 only provide nodes for existing file/directory objects: use my other
2428 methods to create new objects. I return synchronously."""
2430 def create_mutable_file(contents=None, keysize=None):
2431 """I create a new mutable file, and return a Deferred which will fire
2432 with the IMutableFileNode instance when it is ready. If contents= is
2433 provided (a bytestring), it will be used as the initial contents of
2434 the new file, otherwise the file will contain zero bytes. keysize= is
2435 for use by unit tests, to create mutable files that are smaller than
2438 def create_new_mutable_directory(initial_children={}):
2439 """I create a new mutable directory, and return a Deferred which will
2440 fire with the IDirectoryNode instance when it is ready. If
2441 initial_children= is provided (a dict mapping unicode child name to
2442 (childnode, metadata_dict) tuples), the directory will be populated
2443 with those children, otherwise it will be empty."""
2445 class IClientStatus(Interface):
2446 def list_all_uploads():
2447 """Return a list of uploader objects, one for each upload which
2448 currently has an object available (tracked with weakrefs). This is
2449 intended for debugging purposes."""
2450 def list_active_uploads():
2451 """Return a list of active IUploadStatus objects."""
2452 def list_recent_uploads():
2453 """Return a list of IUploadStatus objects for the most recently
2456 def list_all_downloads():
2457 """Return a list of downloader objects, one for each download which
2458 currently has an object available (tracked with weakrefs). This is
2459 intended for debugging purposes."""
2460 def list_active_downloads():
2461 """Return a list of active IDownloadStatus objects."""
2462 def list_recent_downloads():
2463 """Return a list of IDownloadStatus objects for the most recently
2464 started downloads."""
2466 class IUploadStatus(Interface):
2468 """Return a timestamp (float with seconds since epoch) indicating
2469 when the operation was started."""
2470 def get_storage_index():
2471 """Return a string with the (binary) storage index in use on this
2472 upload. Returns None if the storage index has not yet been
2475 """Return an integer with the number of bytes that will eventually
2476 be uploaded for this file. Returns None if the size is not yet known.
2479 """Return True if this upload is using a Helper, False if not."""
2481 """Return a string describing the current state of the upload
2484 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2485 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2486 made towards hashing the file to determine a CHK encryption key: if
2487 non-convergent encryption is in use, this will be trivial, otherwise
2488 the whole file must be hashed. 'ciphertext' describes how much of the
2489 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2490 uploads. 'encode_and_push' describes how much of the encode-and-push
2491 process has finished: for helper uploads this is dependent upon the
2492 helper providing progress reports. It might be reasonable to add all
2493 three numbers and report the sum to the user."""
2495 """Return True if the upload is currently active, False if not."""
2497 """Return an instance of UploadResults (which contains timing and
2498 sharemap information). Might return None if the upload is not yet
2501 """Each upload status gets a unique number: this method returns that
2502 number. This provides a handle to this particular upload, so a web
2503 page can generate a suitable hyperlink."""
2505 class IDownloadStatus(Interface):
2507 """Return a timestamp (float with seconds since epoch) indicating
2508 when the operation was started."""
2509 def get_storage_index():
2510 """Return a string with the (binary) storage index in use on this
2511 download. This may be None if there is no storage index (i.e. LIT
2514 """Return an integer with the number of bytes that will eventually be
2515 retrieved for this file. Returns None if the size is not yet known.
2518 """Return True if this download is using a Helper, False if not."""
2520 """Return a string describing the current state of the download
2523 """Returns a float (from 0.0 to 1.0) describing the amount of the
2524 download that has completed. This value will remain at 0.0 until the
2525 first byte of plaintext is pushed to the download target."""
2527 """Return True if the download is currently active, False if not."""
2529 """Each download status gets a unique number: this method returns
2530 that number. This provides a handle to this particular download, so a
2531 web page can generate a suitable hyperlink."""
2533 class IServermapUpdaterStatus(Interface):
2535 class IPublishStatus(Interface):
2537 class IRetrieveStatus(Interface):
2540 class NotCapableError(Exception):
2541 """You have tried to write to a read-only node."""
2543 class BadWriteEnablerError(Exception):
2546 class RIControlClient(RemoteInterface):
2548 def wait_for_client_connections(num_clients=int):
2549 """Do not return until we have connections to at least NUM_CLIENTS
2553 def upload_from_file_to_uri(filename=str,
2554 convergence=ChoiceOf(None,
2555 StringConstraint(2**20))):
2556 """Upload a file to the grid. This accepts a filename (which must be
2557 absolute) that points to a file on the node's local disk. The node will
2558 read the contents of this file, upload it to the grid, then return the
2559 URI at which it was uploaded. If convergence is None then a random
2560 encryption key will be used, else the plaintext will be hashed, then
2561 that hash will be mixed together with the "convergence" string to form
2566 def download_from_uri_to_file(uri=URI, filename=str):
2567 """Download a file from the grid, placing it on the node's local disk
2568 at the given filename (which must be absolute[?]). Returns the
2569 absolute filename where the file was written."""
2574 def get_memory_usage():
2575 """Return a dict describes the amount of memory currently in use. The
2576 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2577 measuring memory consupmtion in bytes."""
2578 return DictOf(str, int)
2580 def speed_test(count=int, size=int, mutable=Any()):
2581 """Write 'count' tempfiles to disk, all of the given size. Measure
2582 how long (in seconds) it takes to upload them all to the servers.
2583 Then measure how long it takes to download all of them. If 'mutable'
2584 is 'create', time creation of mutable files. If 'mutable' is
2585 'upload', then time access to the same mutable file instead of
2588 Returns a tuple of (upload_time, download_time).
2590 return (float, float)
2592 def measure_peer_response_time():
2593 """Send a short message to each connected peer, and measure the time
2594 it takes for them to respond to it. This is a rough measure of the
2595 application-level round trip time.
2597 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2600 return DictOf(str, float)
2602 UploadResults = Any() #DictOf(str, str)
2604 class RIEncryptedUploadable(RemoteInterface):
2605 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2610 def get_all_encoding_parameters():
2611 return (int, int, int, long)
2613 def read_encrypted(offset=Offset, length=ReadSize):
2620 class RICHKUploadHelper(RemoteInterface):
2621 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2625 Return a dictionary of version information.
2627 return DictOf(str, Any())
2629 def upload(reader=RIEncryptedUploadable):
2630 return UploadResults
2633 class RIHelper(RemoteInterface):
2634 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2638 Return a dictionary of version information.
2640 return DictOf(str, Any())
2642 def upload_chk(si=StorageIndex):
2643 """See if a file with a given storage index needs uploading. The
2644 helper will ask the appropriate storage servers to see if the file
2645 has already been uploaded. If so, the helper will return a set of
2646 'upload results' that includes whatever hashes are needed to build
2647 the read-cap, and perhaps a truncated sharemap.
2649 If the file has not yet been uploaded (or if it was only partially
2650 uploaded), the helper will return an empty upload-results dictionary
2651 and also an RICHKUploadHelper object that will take care of the
2652 upload process. The client should call upload() on this object and
2653 pass it a reference to an RIEncryptedUploadable object that will
2654 provide ciphertext. When the upload is finished, the upload() method
2655 will finish and return the upload results.
2657 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2660 class RIStatsProvider(RemoteInterface):
2661 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2663 Provides access to statistics and monitoring information.
2668 returns a dictionary containing 'counters' and 'stats', each a
2669 dictionary with string counter/stat name keys, and numeric or None values.
2670 counters are monotonically increasing measures of work done, and
2671 stats are instantaneous measures (potentially time averaged
2674 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2676 class RIStatsGatherer(RemoteInterface):
2677 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2679 Provides a monitoring service for centralised collection of stats
2682 def provide(provider=RIStatsProvider, nickname=str):
2684 @param provider: a stats collector instance which should be polled
2685 periodically by the gatherer to collect stats.
2686 @param nickname: a name useful to identify the provided client
2691 class IStatsProducer(Interface):
2694 returns a dictionary, with str keys representing the names of stats
2695 to be monitored, and numeric values.
2698 class RIKeyGenerator(RemoteInterface):
2699 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2701 Provides a service offering to make RSA key pairs.
2704 def get_rsa_key_pair(key_size=int):
2706 @param key_size: the size of the signature key.
2707 @return: tuple(verifying_key, signing_key)
2709 return TupleOf(str, str)
2712 class FileTooLargeError(Exception):
2715 class IValidatedThingProxy(Interface):
2717 """ Acquire a thing and validate it. Return a deferred which is
2718 eventually fired with self if the thing is valid or errbacked if it
2719 can't be acquired or validated."""
2721 class InsufficientVersionError(Exception):
2722 def __init__(self, needed, got):
2723 self.needed = needed
2726 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2729 class EmptyPathnameComponentError(Exception):
2730 """The webapi disallows empty pathname components."""