2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
8 Hash = StringConstraint(maxLength=HASH_SIZE,
9 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
10 Nodeid = StringConstraint(maxLength=20,
11 minLength=20) # binary format 20-byte SHA1 hash
12 FURL = StringConstraint(1000)
13 StorageIndex = StringConstraint(16)
14 URI = StringConstraint(300) # kind of arbitrary
16 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
18 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
20 ShareData = StringConstraint(None)
21 URIExtensionData = StringConstraint(1000)
22 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
24 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
25 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
26 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
27 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
29 class RIStubClient(RemoteInterface):
30 """Each client publishes a service announcement for a dummy object called
31 the StubClient. This object doesn't actually offer any services, but the
32 announcement helps the Introducer keep track of which clients are
33 subscribed (so the grid admin can keep track of things like the size of
34 the grid and the client versions in use. This is the (empty)
35 RemoteInterface for the StubClient."""
37 class RIBucketWriter(RemoteInterface):
38 """ Objects of this kind live on the server side. """
39 def write(offset=Offset, data=ShareData):
44 If the data that has been written is incomplete or inconsistent then
45 the server will throw the data away, else it will store it for future
51 """Abandon all the data that has been written.
55 class RIBucketReader(RemoteInterface):
56 def read(offset=Offset, length=ReadSize):
59 def advise_corrupt_share(reason=str):
60 """Clients who discover hash failures in shares that they have
61 downloaded from me will use this method to inform me about the
62 failures. I will record their concern so that my operator can
63 manually inspect the shares in question. I return None.
65 This is a wrapper around RIStorageServer.advise_corrupt_share(),
66 which is tied to a specific share, and therefore does not need the
67 extra share-identifying arguments. Please see that method for full
71 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
72 # elements are (offset, length, operator, specimen)
73 # operator is one of "lt, le, eq, ne, ge, gt"
74 # nop always passes and is used to fetch data while writing.
75 # you should use length==len(specimen) for everything except nop
76 DataVector = ListOf(TupleOf(Offset, ShareData))
77 # (offset, data). This limits us to 30 writes of 1MiB each per call
78 TestAndWriteVectorsForShares = DictOf(int,
81 ChoiceOf(None, Offset), # new_length
83 ReadVector = ListOf(TupleOf(Offset, ReadSize))
84 ReadData = ListOf(ShareData)
85 # returns data[offset:offset+length] for each element of TestVector
87 class RIStorageServer(RemoteInterface):
88 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
92 Return a dictionary of version information.
94 return DictOf(str, Any())
96 def allocate_buckets(storage_index=StorageIndex,
97 renew_secret=LeaseRenewSecret,
98 cancel_secret=LeaseCancelSecret,
99 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
100 allocated_size=Offset, canary=Referenceable):
102 @param storage_index: the index of the bucket to be created or
104 @param sharenums: these are the share numbers (probably between 0 and
105 99) that the sender is proposing to store on this
107 @param renew_secret: This is the secret used to protect bucket refresh
108 This secret is generated by the client and
109 stored for later comparison by the server. Each
110 server is given a different secret.
111 @param cancel_secret: Like renew_secret, but protects bucket decref.
112 @param canary: If the canary is lost before close(), the bucket is
114 @return: tuple of (alreadygot, allocated), where alreadygot is what we
115 already have and allocated is what we hereby agree to accept.
116 New leases are added for shares in both lists.
118 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
119 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
121 def add_lease(storage_index=StorageIndex,
122 renew_secret=LeaseRenewSecret,
123 cancel_secret=LeaseCancelSecret):
125 Add a new lease on the given bucket. If the renew_secret matches an
126 existing lease, that lease will be renewed instead. If there is no
127 bucket for the given storage_index, return silently. (note that in
128 tahoe-1.3.0 and earlier, IndexError was raised if there was no
131 return Any() # returns None now, but future versions might change
133 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
135 Renew the lease on a given bucket, resetting the timer to 31 days.
136 Some networks will use this, some will not. If there is no bucket for
137 the given storage_index, IndexError will be raised.
139 For mutable shares, if the given renew_secret does not match an
140 existing lease, IndexError will be raised with a note listing the
141 server-nodeids on the existing leases, so leases on migrated shares
142 can be renewed or cancelled. For immutable shares, IndexError
143 (without the note) will be raised.
147 def cancel_lease(storage_index=StorageIndex,
148 cancel_secret=LeaseCancelSecret):
150 Cancel the lease on a given bucket. If this was the last lease on the
151 bucket, the bucket will be deleted. If there is no bucket for the
152 given storage_index, IndexError will be raised.
154 For mutable shares, if the given cancel_secret does not match an
155 existing lease, IndexError will be raised with a note listing the
156 server-nodeids on the existing leases, so leases on migrated shares
157 can be renewed or cancelled. For immutable shares, IndexError
158 (without the note) will be raised.
162 def get_buckets(storage_index=StorageIndex):
163 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
167 def slot_readv(storage_index=StorageIndex,
168 shares=ListOf(int), readv=ReadVector):
169 """Read a vector from the numbered shares associated with the given
170 storage index. An empty shares list means to return data from all
171 known shares. Returns a dictionary with one key per share."""
172 return DictOf(int, ReadData) # shnum -> results
174 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
175 secrets=TupleOf(WriteEnablerSecret,
178 tw_vectors=TestAndWriteVectorsForShares,
181 """General-purpose test-and-set operation for mutable slots. Perform
182 a bunch of comparisons against the existing shares. If they all pass,
183 then apply a bunch of write vectors to those shares. Then use the
184 read vectors to extract data from all the shares and return the data.
186 This method is, um, large. The goal is to allow clients to update all
187 the shares associated with a mutable file in a single round trip.
189 @param storage_index: the index of the bucket to be created or
191 @param write_enabler: a secret that is stored along with the slot.
192 Writes are accepted from any caller who can
193 present the matching secret. A different secret
194 should be used for each slot*server pair.
195 @param renew_secret: This is the secret used to protect bucket refresh
196 This secret is generated by the client and
197 stored for later comparison by the server. Each
198 server is given a different secret.
199 @param cancel_secret: Like renew_secret, but protects bucket decref.
201 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
202 cancel_secret). The first is required to perform any write. The
203 latter two are used when allocating new shares. To simply acquire a
204 new lease on existing shares, use an empty testv and an empty writev.
206 Each share can have a separate test vector (i.e. a list of
207 comparisons to perform). If all vectors for all shares pass, then all
208 writes for all shares are recorded. Each comparison is a 4-tuple of
209 (offset, length, operator, specimen), which effectively does a bool(
210 (read(offset, length)) OPERATOR specimen ) and only performs the
211 write if all these evaluate to True. Basic test-and-set uses 'eq'.
212 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
213 Write-if-same-or-newer uses 'le'.
215 Reads from the end of the container are truncated, and missing shares
216 behave like empty ones, so to assert that a share doesn't exist (for
217 use when creating a new share), use (0, 1, 'eq', '').
219 The write vector will be applied to the given share, expanding it if
220 necessary. A write vector applied to a share number that did not
221 exist previously will cause that share to be created.
223 Each write vector is accompanied by a 'new_length' argument. If
224 new_length is not None, use it to set the size of the container. This
225 can be used to pre-allocate space for a series of upcoming writes, or
226 truncate existing data. If the container is growing, new_length will
227 be applied before datav. If the container is shrinking, it will be
228 applied afterwards. If new_length==0, the share will be deleted.
230 The read vector is used to extract data from all known shares,
231 *before* any writes have been applied. The same vector is used for
232 all shares. This captures the state that was tested by the test
235 This method returns two values: a boolean and a dict. The boolean is
236 True if the write vectors were applied, False if not. The dict is
237 keyed by share number, and each value contains a list of strings, one
238 for each element of the read vector.
240 If the write_enabler is wrong, this will raise BadWriteEnablerError.
241 To enable share migration (using update_write_enabler), the exception
242 will have the nodeid used for the old write enabler embedded in it,
243 in the following string::
245 The write enabler was recorded by nodeid '%s'.
247 Note that the nodeid here is encoded using the same base32 encoding
248 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
251 return TupleOf(bool, DictOf(int, ReadData))
253 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
254 shnum=int, reason=str):
255 """Clients who discover hash failures in shares that they have
256 downloaded from me will use this method to inform me about the
257 failures. I will record their concern so that my operator can
258 manually inspect the shares in question. I return None.
260 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
261 (binary) storage index string, and 'shnum' is the integer share
262 number. 'reason' is a human-readable explanation of the problem,
263 probably including some expected hash values and the computed ones
264 which did not match. Corruption advisories for mutable shares should
265 include a hash of the public key (the same value that appears in the
266 mutable-file verify-cap), since the current share format does not
270 class IStorageBucketWriter(Interface):
272 Objects of this kind live on the client side.
274 def put_block(segmentnum=int, data=ShareData):
275 """@param data: For most segments, this data will be 'blocksize'
276 bytes in length. The last segment might be shorter.
277 @return: a Deferred that fires (with None) when the operation completes
280 def put_plaintext_hashes(hashes=ListOf(Hash)):
282 @return: a Deferred that fires (with None) when the operation completes
285 def put_crypttext_hashes(hashes=ListOf(Hash)):
287 @return: a Deferred that fires (with None) when the operation completes
290 def put_block_hashes(blockhashes=ListOf(Hash)):
292 @return: a Deferred that fires (with None) when the operation completes
295 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
297 @return: a Deferred that fires (with None) when the operation completes
300 def put_uri_extension(data=URIExtensionData):
301 """This block of data contains integrity-checking information (hashes
302 of plaintext, crypttext, and shares), as well as encoding parameters
303 that are necessary to recover the data. This is a serialized dict
304 mapping strings to other strings. The hash of this data is kept in
305 the URI and verified before any of the data is used. All buckets for
306 a given file contain identical copies of this data.
308 The serialization format is specified with the following pseudocode:
309 for k in sorted(dict.keys()):
310 assert re.match(r'^[a-zA-Z_\-]+$', k)
311 write(k + ':' + netstring(dict[k]))
313 @return: a Deferred that fires (with None) when the operation completes
317 """Finish writing and close the bucket. The share is not finalized
318 until this method is called: if the uploading client disconnects
319 before calling close(), the partially-written share will be
322 @return: a Deferred that fires (with None) when the operation completes
325 class IStorageBucketReader(Interface):
327 def get_block_data(blocknum=int, blocksize=int, size=int):
328 """Most blocks will be the same size. The last block might be shorter
334 def get_crypttext_hashes():
336 @return: ListOf(Hash)
339 def get_block_hashes(at_least_these=SetOf(int)):
341 @return: ListOf(Hash)
344 def get_share_hashes(at_least_these=SetOf(int)):
346 @return: ListOf(TupleOf(int, Hash))
349 def get_uri_extension():
351 @return: URIExtensionData
354 class IStorageBroker(Interface):
355 def get_servers_for_index(peer_selection_index):
357 @return: list of (peerid, versioned-rref) tuples
359 def get_all_servers():
361 @return: frozenset of (peerid, versioned-rref) tuples
363 def get_all_serverids():
365 @return: frozenset of serverid strings
367 def get_nickname_for_serverid(serverid):
369 @return: unicode nickname, or None
372 # methods moved from IntroducerClient, need review
373 def get_all_connections():
374 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
375 each active connection we've established to a remote service. This is
376 mostly useful for unit tests that need to wait until a certain number
377 of connections have been made."""
379 def get_all_connectors():
380 """Return a dict that maps from (nodeid, service_name) to a
381 RemoteServiceConnector instance for all services that we are actively
382 trying to connect to. Each RemoteServiceConnector has the following
385 service_name: the type of service provided, like 'storage'
386 announcement_time: when we first heard about this service
387 last_connect_time: when we last established a connection
388 last_loss_time: when we last lost a connection
390 version: the peer's version, from the most recent connection
391 oldest_supported: the peer's oldest supported version, same
393 rref: the RemoteReference, if connected, otherwise None
394 remote_host: the IAddress, if connected, otherwise None
396 This method is intended for monitoring interfaces, such as a web page
397 which describes connecting and connected peers.
400 def get_all_peerids():
401 """Return a frozenset of all peerids to whom we have a connection (to
402 one or more services) established. Mostly useful for unit tests."""
404 def get_all_connections_for(service_name):
405 """Return a frozenset of (nodeid, service_name, rref) tuples, one
406 for each active connection that provides the given SERVICE_NAME."""
408 def get_permuted_peers(service_name, key):
409 """Returns an ordered list of (peerid, rref) tuples, selecting from
410 the connections that provide SERVICE_NAME, using a hash-based
411 permutation keyed by KEY. This randomizes the service list in a
412 repeatable way, to distribute load over many peers.
416 class IURI(Interface):
417 def init_from_string(uri):
418 """Accept a string (as created by my to_string() method) and populate
419 this instance with its data. I am not normally called directly,
420 please use the module-level uri.from_string() function to convert
421 arbitrary URI strings into IURI-providing instances."""
424 """Return False if this URI be used to modify the data. Return True
425 if this URI cannot be used to modify the data."""
428 """Return True if the data can be modified by *somebody* (perhaps
429 someone who has a more powerful URI than this one)."""
431 # TODO: rename to get_read_cap()
433 """Return another IURI instance, which represents a read-only form of
434 this one. If is_readonly() is True, this returns self."""
436 def get_verify_cap():
437 """Return an instance that provides IVerifierURI, which can be used
438 to check on the availability of the file or directory, without
439 providing enough capabilities to actually read or modify the
440 contents. This may return None if the file does not need checking or
441 verification (e.g. LIT URIs).
445 """Return a string of printable ASCII characters, suitable for
446 passing into init_from_string."""
448 class IVerifierURI(Interface, IURI):
449 def init_from_string(uri):
450 """Accept a string (as created by my to_string() method) and populate
451 this instance with its data. I am not normally called directly,
452 please use the module-level uri.from_string() function to convert
453 arbitrary URI strings into IURI-providing instances."""
456 """Return a string of printable ASCII characters, suitable for
457 passing into init_from_string."""
459 class IDirnodeURI(Interface):
460 """I am a URI which represents a dirnode."""
462 class IFileURI(Interface):
463 """I am a URI which represents a filenode."""
465 """Return the length (in bytes) of the file that I represent."""
467 class IImmutableFileURI(IFileURI):
470 class IMutableFileURI(Interface):
471 """I am a URI which represents a mutable filenode."""
473 class IDirectoryURI(Interface):
476 class IReadonlyDirectoryURI(Interface):
479 class CapConstraintError(Exception):
480 """A constraint on a cap was violated."""
482 class MustBeDeepImmutableError(CapConstraintError):
483 """Mutable children cannot be added to an immutable directory.
484 Also, caps obtained from an immutable directory can trigger this error
485 if they are later found to refer to a mutable object and then used."""
487 class MustBeReadonlyError(CapConstraintError):
488 """Known write caps cannot be specified in a ro_uri field. Also,
489 caps obtained from a ro_uri field can trigger this error if they
490 are later found to be write caps and then used."""
492 class MustNotBeUnknownRWError(CapConstraintError):
493 """Cannot add an unknown child cap specified in a rw_uri field."""
495 # The hierarchy looks like this:
502 class IFilesystemNode(Interface):
504 """Return the strongest 'cap instance' associated with this node.
505 (writecap for writeable-mutable files/directories, readcap for
506 immutable or readonly-mutable files/directories). To convert this
507 into a string, call .to_string() on the result."""
510 """Return a readonly cap instance for this node. For immutable or
511 readonly nodes, get_cap() and get_readcap() return the same thing."""
513 def get_repair_cap():
514 """Return an IURI instance that can be used to repair the file, or
515 None if this node cannot be repaired (either because it is not
516 distributed, like a LIT file, or because the node does not represent
517 sufficient authority to create a repair-cap, like a read-only RSA
518 mutable file node [which cannot create the correct write-enablers]).
521 def get_verify_cap():
522 """Return an IVerifierURI instance that represents the
523 'verifiy/refresh capability' for this node. The holder of this
524 capability will be able to renew the lease for this node, protecting
525 it from garbage-collection. They will also be able to ask a server if
526 it holds a share for the file or directory.
530 """Return the URI string corresponding to the strongest cap associated
531 with this node. If this node is read-only, the URI will only offer
532 read-only access. If this node is read-write, the URI will offer
535 If you have read-write access to a node and wish to share merely
536 read-only access with others, use get_readonly_uri().
539 def get_write_uri(n):
540 """Return the URI string that can be used by others to get write
541 access to this node, if it is writeable. If this is a read-only node,
544 def get_readonly_uri():
545 """Return the URI string that can be used by others to get read-only
546 access to this node. The result is a read-only URI, regardless of
547 whether this node is read-only or read-write.
549 If you have merely read-only access to this node, get_readonly_uri()
550 will return the same thing as get_uri().
553 def get_storage_index():
554 """Return a string with the (binary) storage index in use on this
555 download. This may be None if there is no storage index (i.e. LIT
556 files and directories)."""
559 """Return True if this reference provides mutable access to the given
560 file or directory (i.e. if you can modify it), or False if not. Note
561 that even if this reference is read-only, someone else may hold a
562 read-write reference to it."""
565 """Return True if this file or directory is mutable (by *somebody*,
566 not necessarily you), False if it is is immutable. Note that a file
567 might be mutable overall, but your reference to it might be
568 read-only. On the other hand, all references to an immutable file
569 will be read-only; there are no read-write references to an immutable
574 """Return True if this is an unknown node."""
576 def is_allowed_in_immutable_directory():
577 """Return True if this node is allowed as a child of a deep-immutable
578 directory. This is true if either the node is of a known-immutable type,
579 or it is unknown and read-only.
583 """Raise any error associated with this node."""
586 """Return the length (in bytes) of the data this node represents. For
587 directory nodes, I return the size of the backing store. I return
588 synchronously and do not consult the network, so for mutable objects,
589 I will return the most recently observed size for the object, or None
590 if I don't remember a size. Use get_current_size, which returns a
591 Deferred, if you want more up-to-date information."""
593 def get_current_size():
594 """I return a Deferred that fires with the length (in bytes) of the
595 data this node represents.
598 class IFileNode(IFilesystemNode):
599 """I am a node which represents a file: a sequence of bytes. I am not a
600 container, like IDirectoryNode."""
602 class IImmutableFileNode(IFileNode):
603 def read(consumer, offset=0, size=None):
604 """Download a portion (possibly all) of the file's contents, making
605 them available to the given IConsumer. Return a Deferred that fires
606 (with the consumer) when the consumer is unregistered (either because
607 the last byte has been given to it, or because the consumer threw an
608 exception during write(), possibly because it no longer wants to
609 receive data). The portion downloaded will start at 'offset' and
610 contain 'size' bytes (or the remainder of the file if size==None).
612 The consumer will be used in non-streaming mode: an IPullProducer
613 will be attached to it.
615 The consumer will not receive data right away: several network trips
616 must occur first. The order of events will be::
618 consumer.registerProducer(p, streaming)
619 (if streaming == False)::
620 consumer does p.resumeProducing()
622 consumer does p.resumeProducing()
623 consumer.write(data).. (repeat until all data is written)
624 consumer.unregisterProducer()
625 deferred.callback(consumer)
627 If a download error occurs, or an exception is raised by
628 consumer.registerProducer() or consumer.write(), I will call
629 consumer.unregisterProducer() and then deliver the exception via
630 deferred.errback(). To cancel the download, the consumer should call
631 p.stopProducing(), which will result in an exception being delivered
632 via deferred.errback().
634 See src/allmydata/util/consumer.py for an example of a simple
635 download-to-memory consumer.
638 class IMutableFileNode(IFileNode):
639 """I provide access to a 'mutable file', which retains its identity
640 regardless of what contents are put in it.
642 The consistency-vs-availability problem means that there might be
643 multiple versions of a file present in the grid, some of which might be
644 unrecoverable (i.e. have fewer than 'k' shares). These versions are
645 loosely ordered: each has a sequence number and a hash, and any version
646 with seqnum=N was uploaded by a node which has seen at least one version
649 The 'servermap' (an instance of IMutableFileServerMap) is used to
650 describe the versions that are known to be present in the grid, and which
651 servers are hosting their shares. It is used to represent the 'state of
652 the world', and is used for this purpose by my test-and-set operations.
653 Downloading the contents of the mutable file will also return a
654 servermap. Uploading a new version into the mutable file requires a
655 servermap as input, and the semantics of the replace operation is
656 'replace the file with my new version if it looks like nobody else has
657 changed the file since my previous download'. Because the file is
658 distributed, this is not a perfect test-and-set operation, but it will do
659 its best. If the replace process sees evidence of a simultaneous write,
660 it will signal an UncoordinatedWriteError, so that the caller can take
664 Most readers will want to use the 'best' current version of the file, and
665 should use my 'download_best_version()' method.
667 To unconditionally replace the file, callers should use overwrite(). This
668 is the mode that user-visible mutable files will probably use.
670 To apply some delta to the file, call modify() with a callable modifier
671 function that can apply the modification that you want to make. This is
672 the mode that dirnodes will use, since most directory modification
673 operations can be expressed in terms of deltas to the directory state.
676 Three methods are available for users who need to perform more complex
677 operations. The first is get_servermap(), which returns an up-to-date
678 servermap using a specified mode. The second is download_version(), which
679 downloads a specific version (not necessarily the 'best' one). The third
680 is 'upload', which accepts new contents and a servermap (which must have
681 been updated with MODE_WRITE). The upload method will attempt to apply
682 the new contents as long as no other node has modified the file since the
683 servermap was updated. This might be useful to a caller who wants to
684 merge multiple versions into a single new one.
686 Note that each time the servermap is updated, a specific 'mode' is used,
687 which determines how many peers are queried. To use a servermap for my
688 replace() method, that servermap must have been updated in MODE_WRITE.
689 These modes are defined in allmydata.mutable.common, and consist of
690 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
691 allmydata/mutable/servermap.py for details about the differences.
693 Mutable files are currently limited in size (about 3.5MB max) and can
694 only be retrieved and updated all-at-once, as a single big string. Future
695 versions of our mutable files will remove this restriction.
698 def download_best_version():
699 """Download the 'best' available version of the file, meaning one of
700 the recoverable versions with the highest sequence number. If no
701 uncoordinated writes have occurred, and if enough shares are
702 available, then this will be the most recent version that has been
705 I update an internal servermap with MODE_READ, determine which
706 version of the file is indicated by
707 servermap.best_recoverable_version(), and return a Deferred that
708 fires with its contents. If no version is recoverable, the Deferred
709 will errback with UnrecoverableFileError.
712 def get_size_of_best_version():
713 """Find the size of the version that would be downloaded with
714 download_best_version(), without actually downloading the whole file.
716 I return a Deferred that fires with an integer.
719 def overwrite(new_contents):
720 """Unconditionally replace the contents of the mutable file with new
721 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
722 is only appropriate to use when the new contents of the file are
723 completely unrelated to the old ones, and you do not care about other
726 I return a Deferred that fires (with a PublishStatus object) when the
727 update has completed.
730 def modify(modifier_cb):
731 """Modify the contents of the file, by downloading the current
732 version, applying the modifier function (or bound method), then
733 uploading the new version. I return a Deferred that fires (with a
734 PublishStatus object) when the update is complete.
736 The modifier callable will be given three arguments: a string (with
737 the old contents), a 'first_time' boolean, and a servermap. As with
738 download_best_version(), the old contents will be from the best
739 recoverable version, but the modifier can use the servermap to make
740 other decisions (such as refusing to apply the delta if there are
741 multiple parallel versions, or if there is evidence of a newer
742 unrecoverable version). 'first_time' will be True the first time the
743 modifier is called, and False on any subsequent calls.
745 The callable should return a string with the new contents. The
746 callable must be prepared to be called multiple times, and must
747 examine the input string to see if the change that it wants to make
748 is already present in the old version. If it does not need to make
749 any changes, it can either return None, or return its input string.
751 If the modifier raises an exception, it will be returned in the
756 def get_servermap(mode):
757 """Return a Deferred that fires with an IMutableFileServerMap
758 instance, updated using the given mode.
761 def download_version(servermap, version):
762 """Download a specific version of the file, using the servermap
763 as a guide to where the shares are located.
765 I return a Deferred that fires with the requested contents, or
766 errbacks with UnrecoverableFileError. Note that a servermap which was
767 updated with MODE_ANYTHING or MODE_READ may not know about shares for
768 all versions (those modes stop querying servers as soon as they can
769 fulfil their goals), so you may want to use MODE_CHECK (which checks
770 everything) to get increased visibility.
773 def upload(new_contents, servermap):
774 """Replace the contents of the file with new ones. This requires a
775 servermap that was previously updated with MODE_WRITE.
777 I attempt to provide test-and-set semantics, in that I will avoid
778 modifying any share that is different than the version I saw in the
779 servermap. However, if another node is writing to the file at the
780 same time as me, I may manage to update some shares while they update
781 others. If I see any evidence of this, I will signal
782 UncoordinatedWriteError, and the file will be left in an inconsistent
783 state (possibly the version you provided, possibly the old version,
784 possibly somebody else's version, and possibly a mix of shares from
787 The recommended response to UncoordinatedWriteError is to either
788 return it to the caller (since they failed to coordinate their
789 writes), or to attempt some sort of recovery. It may be sufficient to
790 wait a random interval (with exponential backoff) and repeat your
791 operation. If I do not signal UncoordinatedWriteError, then I was
792 able to write the new version without incident.
794 I return a Deferred that fires (with a PublishStatus object) when the
795 publish has completed. I will update the servermap in-place with the
796 location of all new shares.
800 """Return this filenode's writekey, or None if the node does not have
801 write-capability. This may be used to assist with data structures
802 that need to make certain data available only to writers, such as the
803 read-write child caps in dirnodes. The recommended process is to have
804 reader-visible data be submitted to the filenode in the clear (where
805 it will be encrypted by the filenode using the readkey), but encrypt
806 writer-visible data using this writekey.
809 class NotEnoughSharesError(Exception):
810 """Download was unable to get enough shares"""
812 class NoSharesError(Exception):
813 """Download was unable to get any shares at all."""
815 class UploadUnhappinessError(Exception):
816 """Upload was unable to satisfy 'servers_of_happiness'"""
818 class UnableToFetchCriticalDownloadDataError(Exception):
819 """I was unable to fetch some piece of critical data which is supposed to
820 be identically present in all shares."""
822 class NoServersError(Exception):
823 """Upload wasn't given any servers to work with, usually indicating a
824 network or Introducer problem."""
826 class ExistingChildError(Exception):
827 """A directory node was asked to add or replace a child that already
828 exists, and overwrite= was set to False."""
830 class NoSuchChildError(Exception):
831 """A directory node was asked to fetch a child which does not exist."""
833 class ChildOfWrongTypeError(Exception):
834 """An operation was attempted on a child of the wrong type (file or directory)."""
836 class IDirectoryNode(IFilesystemNode):
837 """I represent a filesystem node that is a container, with a
838 name-to-child mapping, holding the tahoe equivalent of a directory. All
839 child names are unicode strings, and all children are some sort of
840 IFilesystemNode (a file, subdirectory, or unknown node).
845 The dirnode ('1') URI returned by this method can be used in
846 set_uri() on a different directory ('2') to 'mount' a reference to
847 this directory ('1') under the other ('2'). This URI is just a
848 string, so it can be passed around through email or other out-of-band
852 def get_readonly_uri():
854 The dirnode ('1') URI returned by this method can be used in
855 set_uri() on a different directory ('2') to 'mount' a reference to
856 this directory ('1') under the other ('2'). This URI is just a
857 string, so it can be passed around through email or other out-of-band
862 """I return a Deferred that fires with a dictionary mapping child
863 name (a unicode string) to (node, metadata_dict) tuples, in which
864 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
868 """I return a Deferred that fires with a boolean, True if there
869 exists a child of the given name, False if not. The child name must
870 be a unicode string."""
873 """I return a Deferred that fires with a specific named child node,
874 which is an IFilesystemNode. The child name must be a unicode string.
875 I raise NoSuchChildError if I do not have a child by that name."""
877 def get_metadata_for(name):
878 """I return a Deferred that fires with the metadata dictionary for
879 a specific named child node. The child name must be a unicode string.
880 This metadata is stored in the *edge*, not in the child, so it is
881 attached to the parent dirnode rather than the child node.
882 I raise NoSuchChildError if I do not have a child by that name."""
884 def set_metadata_for(name, metadata):
885 """I replace any existing metadata for the named child with the new
886 metadata. The child name must be a unicode string. This metadata is
887 stored in the *edge*, not in the child, so it is attached to the
888 parent dirnode rather than the child node. I return a Deferred
889 (that fires with this dirnode) when the operation is complete.
890 I raise NoSuchChildError if I do not have a child by that name."""
892 def get_child_at_path(path):
893 """Transform a child path into an IFilesystemNode.
895 I perform a recursive series of 'get' operations to find the named
896 descendant node. I return a Deferred that fires with the node, or
897 errbacks with NoSuchChildError if the node could not be found.
899 The path can be either a single string (slash-separated) or a list of
900 path-name elements. All elements must be unicode strings.
903 def get_child_and_metadata_at_path(path):
904 """Transform a child path into an IFilesystemNode and metadata.
906 I am like get_child_at_path(), but my Deferred fires with a tuple of
907 (node, metadata). The metadata comes from the last edge. If the path
908 is empty, the metadata will be an empty dictionary.
911 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
912 """I add a child (by writecap+readcap) at the specific name. I return
913 a Deferred that fires when the operation finishes. If overwrite= is
914 True, I will replace any existing child of the same name, otherwise
915 an existing child will cause me to return ExistingChildError. The
916 child name must be a unicode string.
918 The child caps could be for a file, or for a directory. If you have
919 both the writecap and readcap, you should provide both arguments.
920 If you have only one cap and don't know whether it is read-only,
921 provide it as the writecap argument and leave the readcap as None.
922 If you have only one cap that is known to be read-only, provide it
923 as the readcap argument and leave the writecap as None.
924 The filecaps are typically obtained from an IFilesystemNode with
925 get_uri() and get_readonly_uri().
927 If metadata= is provided, I will use it as the metadata for the named
928 edge. This will replace any existing metadata. If metadata= is left
929 as the default value of None, I will set ['mtime'] to the current
930 time, and I will set ['ctime'] to the current time if there was not
931 already a child by this name present. This roughly matches the
932 ctime/mtime semantics of traditional filesystems. See the
933 "About the metadata" section of webapi.txt for futher information.
935 If this directory node is read-only, the Deferred will errback with a
936 NotWriteableError."""
938 def set_children(entries, overwrite=True):
939 """Add multiple children (by writecap+readcap) to a directory node.
940 Takes a dictionary, with childname as keys and (writecap, readcap)
941 tuples (or (writecap, readcap, metadata) triples) as values. Returns
942 a Deferred that fires (with this dirnode) when the operation
943 finishes. This is equivalent to calling set_uri() multiple times, but
944 is much more efficient. All child names must be unicode strings.
947 def set_node(name, child, metadata=None, overwrite=True):
948 """I add a child at the specific name. I return a Deferred that fires
949 when the operation finishes. This Deferred will fire with the child
950 node that was just added. I will replace any existing child of the
951 same name. The child name must be a unicode string. The 'child'
952 instance must be an instance providing IFilesystemNode.
954 If metadata= is provided, I will use it as the metadata for the named
955 edge. This will replace any existing metadata. If metadata= is left
956 as the default value of None, I will set ['mtime'] to the current
957 time, and I will set ['ctime'] to the current time if there was not
958 already a child by this name present. This roughly matches the
959 ctime/mtime semantics of traditional filesystems. See the
960 "About the metadata" section of webapi.txt for futher information.
962 If this directory node is read-only, the Deferred will errback with a
963 NotWriteableError."""
965 def set_nodes(entries, overwrite=True):
966 """Add multiple children to a directory node. Takes a dict mapping
967 unicode childname to (child_node, metdata) tuples. If metdata=None,
968 the original metadata is left unmodified. Returns a Deferred that
969 fires (with this dirnode) when the operation finishes. This is
970 equivalent to calling set_node() multiple times, but is much more
973 def add_file(name, uploadable, metadata=None, overwrite=True):
974 """I upload a file (using the given IUploadable), then attach the
975 resulting ImmutableFileNode to the directory at the given name. I set
976 metadata the same way as set_uri and set_node. The child name must be
979 I return a Deferred that fires (with the IFileNode of the uploaded
980 file) when the operation completes."""
982 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
983 """I remove the child at the specific name. I return a Deferred that
984 fires when the operation finishes. The child name must be a unicode
985 string. If must_exist is True and I do not have a child by that name,
986 I raise NoSuchChildError. If must_be_directory is True and the child
987 is a file, or if must_be_file is True and the child is a directory,
988 I raise ChildOfWrongTypeError."""
990 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
991 """I create and attach a directory at the given name. The new
992 directory can be empty, or it can be populated with children
993 according to 'initial_children', which takes a dictionary in the same
994 format as set_nodes (i.e. mapping unicode child name to (childnode,
995 metadata) tuples). The child name must be a unicode string. I return
996 a Deferred that fires (with the new directory node) when the
997 operation finishes."""
999 def move_child_to(current_child_name, new_parent, new_child_name=None,
1001 """I take one of my children and move them to a new parent. The child
1002 is referenced by name. On the new parent, the child will live under
1003 'new_child_name', which defaults to 'current_child_name'. TODO: what
1004 should we do about metadata? I return a Deferred that fires when the
1005 operation finishes. The child name must be a unicode string. I raise
1006 NoSuchChildError if I do not have a child by that name."""
1008 def build_manifest():
1009 """I generate a table of everything reachable from this directory.
1010 I also compute deep-stats as described below.
1012 I return a Monitor. The Monitor's results will be a dictionary with
1015 res['manifest']: a list of (path, cap) tuples for all nodes
1016 (directories and files) reachable from this one.
1017 'path' will be a tuple of unicode strings. The
1018 origin dirnode will be represented by an empty path
1020 res['verifycaps']: a list of (printable) verifycap strings, one for
1021 each reachable non-LIT node. This is a set:
1022 it will contain no duplicates.
1023 res['storage-index']: a list of (base32) storage index strings,
1024 one for each reachable non-LIT node. This is
1025 a set: it will contain no duplicates.
1026 res['stats']: a dictionary, the same that is generated by
1027 start_deep_stats() below.
1029 The Monitor will also have an .origin_si attribute with the (binary)
1030 storage index of the starting point.
1033 def start_deep_stats():
1034 """Return a Monitor, examining all nodes (directories and files)
1035 reachable from this one. The Monitor's results will be a dictionary
1036 with the following keys::
1038 count-immutable-files: count of how many CHK files are in the set
1039 count-mutable-files: same, for mutable files (does not include
1041 count-literal-files: same, for LIT files
1042 count-files: sum of the above three
1044 count-directories: count of directories
1046 size-immutable-files: total bytes for all CHK files in the set
1047 size-mutable-files (TODO): same, for current version of all mutable
1048 files, does not include directories
1049 size-literal-files: same, for LIT files
1050 size-directories: size of mutable files used by directories
1052 largest-directory: number of bytes in the largest directory
1053 largest-directory-children: number of children in the largest
1055 largest-immutable-file: number of bytes in the largest CHK file
1057 size-mutable-files is not yet implemented, because it would involve
1058 even more queries than deep_stats does.
1060 The Monitor will also have an .origin_si attribute with the (binary)
1061 storage index of the starting point.
1063 This operation will visit every directory node underneath this one,
1064 and can take a long time to run. On a typical workstation with good
1065 bandwidth, this can examine roughly 15 directories per second (and
1066 takes several minutes of 100% CPU for ~1700 directories).
1069 class ICodecEncoder(Interface):
1070 def set_params(data_size, required_shares, max_shares):
1071 """Set up the parameters of this encoder.
1073 This prepares the encoder to perform an operation that converts a
1074 single block of data into a number of shares, such that a future
1075 ICodecDecoder can use a subset of these shares to recover the
1076 original data. This operation is invoked by calling encode(). Once
1077 the encoding parameters are set up, the encode operation can be
1078 invoked multiple times.
1080 set_params() prepares the encoder to accept blocks of input data that
1081 are exactly 'data_size' bytes in length. The encoder will be prepared
1082 to produce 'max_shares' shares for each encode() operation (although
1083 see the 'desired_share_ids' to use less CPU). The encoding math will
1084 be chosen such that the decoder can get by with as few as
1085 'required_shares' of these shares and still reproduce the original
1086 data. For example, set_params(1000, 5, 5) offers no redundancy at
1087 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1089 Numerical Restrictions: 'data_size' is required to be an integral
1090 multiple of 'required_shares'. In general, the caller should choose
1091 required_shares and max_shares based upon their reliability
1092 requirements and the number of peers available (the total storage
1093 space used is roughly equal to max_shares*data_size/required_shares),
1094 then choose data_size to achieve the memory footprint desired (larger
1095 data_size means more efficient operation, smaller data_size means
1096 smaller memory footprint).
1098 In addition, 'max_shares' must be equal to or greater than
1099 'required_shares'. Of course, setting them to be equal causes
1100 encode() to degenerate into a particularly slow form of the 'split'
1103 See encode() for more details about how these parameters are used.
1105 set_params() must be called before any other ICodecEncoder methods
1110 """Return the 3-tuple of data_size, required_shares, max_shares"""
1112 def get_encoder_type():
1113 """Return a short string that describes the type of this encoder.
1115 There is required to be a global table of encoder classes. This method
1116 returns an index into this table; the value at this index is an
1117 encoder class, and this encoder is an instance of that class.
1120 def get_block_size():
1121 """Return the length of the shares that encode() will produce.
1124 def encode_proposal(data, desired_share_ids=None):
1125 """Encode some data.
1127 'data' must be a string (or other buffer object), and len(data) must
1128 be equal to the 'data_size' value passed earlier to set_params().
1130 This will return a Deferred that will fire with two lists. The first
1131 is a list of shares, each of which is a string (or other buffer
1132 object) such that len(share) is the same as what get_share_size()
1133 returned earlier. The second is a list of shareids, in which each is
1134 an integer. The lengths of the two lists will always be equal to each
1135 other. The user should take care to keep each share closely
1136 associated with its shareid, as one is useless without the other.
1138 The length of this output list will normally be the same as the value
1139 provided to the 'max_shares' parameter of set_params(). This may be
1140 different if 'desired_share_ids' is provided.
1142 'desired_share_ids', if provided, is required to be a sequence of
1143 ints, each of which is required to be >= 0 and < max_shares. If not
1144 provided, encode() will produce 'max_shares' shares, as if
1145 'desired_share_ids' were set to range(max_shares). You might use this
1146 if you initially thought you were going to use 10 peers, started
1147 encoding, and then two of the peers dropped out: you could use
1148 desired_share_ids= to skip the work (both memory and CPU) of
1149 producing shares for the peers which are no longer available.
1153 def encode(inshares, desired_share_ids=None):
1154 """Encode some data. This may be called multiple times. Each call is
1157 inshares is a sequence of length required_shares, containing buffers
1158 (i.e. strings), where each buffer contains the next contiguous
1159 non-overlapping segment of the input data. Each buffer is required to
1160 be the same length, and the sum of the lengths of the buffers is
1161 required to be exactly the data_size promised by set_params(). (This
1162 implies that the data has to be padded before being passed to
1163 encode(), unless of course it already happens to be an even multiple
1164 of required_shares in length.)
1166 Note: the requirement to break up your data into
1167 'required_shares' chunks of exactly the right length before
1168 calling encode() is surprising from point of view of a user
1169 who doesn't know how FEC works. It feels like an
1170 implementation detail that has leaked outside the abstraction
1171 barrier. Is there a use case in which the data to be encoded
1172 might already be available in pre-segmented chunks, such that
1173 it is faster or less work to make encode() take a list rather
1174 than splitting a single string?
1176 Yes, there is: suppose you are uploading a file with K=64,
1177 N=128, segsize=262,144. Then each in-share will be of size
1178 4096. If you use this .encode() API then your code could first
1179 read each successive 4096-byte chunk from the file and store
1180 each one in a Python string and store each such Python string
1181 in a Python list. Then you could call .encode(), passing that
1182 list as "inshares". The encoder would generate the other 64
1183 "secondary shares" and return to you a new list containing
1184 references to the same 64 Python strings that you passed in
1185 (as the primary shares) plus references to the new 64 Python
1188 (You could even imagine that your code could use readv() so
1189 that the operating system can arrange to get all of those
1190 bytes copied from the file into the Python list of Python
1191 strings as efficiently as possible instead of having a loop
1192 written in C or in Python to copy the next part of the file
1193 into the next string.)
1195 On the other hand if you instead use the .encode_proposal()
1196 API (above), then your code can first read in all of the
1197 262,144 bytes of the segment from the file into a Python
1198 string, then call .encode_proposal() passing the segment data
1199 as the "data" argument. The encoder would basically first
1200 split the "data" argument into a list of 64 in-shares of 4096
1201 byte each, and then do the same thing that .encode() does. So
1202 this would result in a little bit more copying of data and a
1203 little bit higher of a "maximum memory usage" during the
1204 process, although it might or might not make a practical
1205 difference for our current use cases.
1207 Note that "inshares" is a strange name for the parameter if
1208 you think of the parameter as being just for feeding in data
1209 to the codec. It makes more sense if you think of the result
1210 of this encoding as being the set of shares from inshares plus
1211 an extra set of "secondary shares" (or "check shares"). It is
1212 a surprising name! If the API is going to be surprising then
1213 the name should be surprising. If we switch to
1214 encode_proposal() above then we should also switch to an
1217 'desired_share_ids', if provided, is required to be a sequence of
1218 ints, each of which is required to be >= 0 and < max_shares. If not
1219 provided, encode() will produce 'max_shares' shares, as if
1220 'desired_share_ids' were set to range(max_shares). You might use this
1221 if you initially thought you were going to use 10 peers, started
1222 encoding, and then two of the peers dropped out: you could use
1223 desired_share_ids= to skip the work (both memory and CPU) of
1224 producing shares for the peers which are no longer available.
1226 For each call, encode() will return a Deferred that fires with two
1227 lists, one containing shares and the other containing the shareids.
1228 The get_share_size() method can be used to determine the length of
1229 the share strings returned by encode(). Each shareid is a small
1230 integer, exactly as passed into 'desired_share_ids' (or
1231 range(max_shares), if desired_share_ids was not provided).
1233 The shares and their corresponding shareids are required to be kept
1234 together during storage and retrieval. Specifically, the share data is
1235 useless by itself: the decoder needs to be told which share is which
1236 by providing it with both the shareid and the actual share data.
1238 This function will allocate an amount of memory roughly equal to::
1240 (max_shares - required_shares) * get_share_size()
1242 When combined with the memory that the caller must allocate to
1243 provide the input data, this leads to a memory footprint roughly
1244 equal to the size of the resulting encoded shares (i.e. the expansion
1245 factor times the size of the input segment).
1250 # returning a list of (shareidN,shareN) tuples instead of a pair of
1251 # lists (shareids..,shares..). Brian thought the tuples would
1252 # encourage users to keep the share and shareid together throughout
1253 # later processing, Zooko pointed out that the code to iterate
1254 # through two lists is not really more complicated than using a list
1255 # of tuples and there's also a performance improvement
1257 # having 'data_size' not required to be an integral multiple of
1258 # 'required_shares'. Doing this would require encode() to perform
1259 # padding internally, and we'd prefer to have any padding be done
1260 # explicitly by the caller. Yes, it is an abstraction leak, but
1261 # hopefully not an onerous one.
1264 class ICodecDecoder(Interface):
1265 def set_params(data_size, required_shares, max_shares):
1266 """Set the params. They have to be exactly the same ones that were
1267 used for encoding."""
1269 def get_needed_shares():
1270 """Return the number of shares needed to reconstruct the data.
1271 set_params() is required to be called before this."""
1273 def decode(some_shares, their_shareids):
1274 """Decode a partial list of shares into data.
1276 'some_shares' is required to be a sequence of buffers of sharedata, a
1277 subset of the shares returned by ICodecEncode.encode(). Each share is
1278 required to be of the same length. The i'th element of their_shareids
1279 is required to be the shareid of the i'th buffer in some_shares.
1281 This returns a Deferred which fires with a sequence of buffers. This
1282 sequence will contain all of the segments of the original data, in
1283 order. The sum of the lengths of all of the buffers will be the
1284 'data_size' value passed into the original ICodecEncode.set_params()
1285 call. To get back the single original input block of data, use
1286 ''.join(output_buffers), or you may wish to simply write them in
1287 order to an output file.
1289 Note that some of the elements in the result sequence may be
1290 references to the elements of the some_shares input sequence. In
1291 particular, this means that if those share objects are mutable (e.g.
1292 arrays) and if they are changed, then both the input (the
1293 'some_shares' parameter) and the output (the value given when the
1294 deferred is triggered) will change.
1296 The length of 'some_shares' is required to be exactly the value of
1297 'required_shares' passed into the original ICodecEncode.set_params()
1301 class IEncoder(Interface):
1302 """I take an object that provides IEncryptedUploadable, which provides
1303 encrypted data, and a list of shareholders. I then encode, hash, and
1304 deliver shares to those shareholders. I will compute all the necessary
1305 Merkle hash trees that are necessary to validate the crypttext that
1306 eventually comes back from the shareholders. I provide the URI Extension
1307 Block Hash, and the encoding parameters, both of which must be included
1310 I do not choose shareholders, that is left to the IUploader. I must be
1311 given a dict of RemoteReferences to storage buckets that are ready and
1312 willing to receive data.
1316 """Specify the number of bytes that will be encoded. This must be
1317 peformed before get_serialized_params() can be called.
1319 def set_params(params):
1320 """Override the default encoding parameters. 'params' is a tuple of
1321 (k,d,n), where 'k' is the number of required shares, 'd' is the
1322 servers_of_happiness, and 'n' is the total number of shares that will
1325 Encoding parameters can be set in three ways. 1: The Encoder class
1326 provides defaults (3/7/10). 2: the Encoder can be constructed with
1327 an 'options' dictionary, in which the
1328 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1329 set_params((k,d,n)) can be called.
1331 If you intend to use set_params(), you must call it before
1332 get_share_size or get_param are called.
1335 def set_encrypted_uploadable(u):
1336 """Provide a source of encrypted upload data. 'u' must implement
1337 IEncryptedUploadable.
1339 When this is called, the IEncryptedUploadable will be queried for its
1340 length and the storage_index that should be used.
1342 This returns a Deferred that fires with this Encoder instance.
1344 This must be performed before start() can be called.
1347 def get_param(name):
1348 """Return an encoding parameter, by name.
1350 'storage_index': return a string with the (16-byte truncated SHA-256
1351 hash) storage index to which these shares should be
1354 'share_counts': return a tuple describing how many shares are used:
1355 (needed_shares, servers_of_happiness, total_shares)
1357 'num_segments': return an int with the number of segments that
1360 'segment_size': return an int with the size of each segment.
1362 'block_size': return the size of the individual blocks that will
1363 be delivered to a shareholder's put_block() method. By
1364 knowing this, the shareholder will be able to keep all
1365 blocks in a single file and still provide random access
1366 when reading them. # TODO: can we avoid exposing this?
1368 'share_size': an int with the size of the data that will be stored
1369 on each shareholder. This is aggregate amount of data
1370 that will be sent to the shareholder, summed over all
1371 the put_block() calls I will ever make. It is useful to
1372 determine this size before asking potential
1373 shareholders whether they will grant a lease or not,
1374 since their answers will depend upon how much space we
1375 need. TODO: this might also include some amount of
1376 overhead, like the size of all the hashes. We need to
1377 decide whether this is useful or not.
1379 'serialized_params': a string with a concise description of the
1380 codec name and its parameters. This may be passed
1381 into the IUploadable to let it make sure that
1382 the same file encoded with different parameters
1383 will result in different storage indexes.
1385 Once this is called, set_size() and set_params() may not be called.
1388 def set_shareholders(shareholders, servermap):
1389 """Tell the encoder where to put the encoded shares. 'shareholders'
1390 must be a dictionary that maps share number (an integer ranging from
1391 0 to n-1) to an instance that provides IStorageBucketWriter.
1392 'servermap' is a dictionary that maps share number (as defined above)
1393 to a set of peerids. This must be performed before start() can be
1397 """Begin the encode/upload process. This involves reading encrypted
1398 data from the IEncryptedUploadable, encoding it, uploading the shares
1399 to the shareholders, then sending the hash trees.
1401 set_encrypted_uploadable() and set_shareholders() must be called
1402 before this can be invoked.
1404 This returns a Deferred that fires with a verify cap when the upload
1405 process is complete. The verifycap, plus the encryption key, is
1406 sufficient to construct the read cap.
1409 class IDecoder(Interface):
1410 """I take a list of shareholders and some setup information, then
1411 download, validate, decode, and decrypt data from them, writing the
1412 results to an output file.
1414 I do not locate the shareholders, that is left to the IDownloader. I must
1415 be given a dict of RemoteReferences to storage buckets that are ready to
1420 """I take a file-like object (providing write and close) to which all
1421 the plaintext data will be written.
1423 TODO: producer/consumer . Maybe write() should return a Deferred that
1424 indicates when it will accept more data? But probably having the
1425 IDecoder be a producer is easier to glue to IConsumer pieces.
1428 def set_shareholders(shareholders):
1429 """I take a dictionary that maps share identifiers (small integers)
1430 to RemoteReferences that provide RIBucketReader. This must be called
1434 """I start the download. This process involves retrieving data and
1435 hash chains from the shareholders, using the hashes to validate the
1436 data, decoding the shares into segments, decrypting the segments,
1437 then writing the resulting plaintext to the output file.
1439 I return a Deferred that will fire (with self) when the download is
1443 class IDownloadTarget(Interface):
1444 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1445 # will register itself as a producer. This allows the target to invoke
1446 # downloader.pauseProducing, resumeProducing, and stopProducing.
1448 """Called before any calls to write() or close(). If an error
1449 occurs before any data is available, fail() may be called without
1450 a previous call to open().
1452 'size' is the length of the file being downloaded, in bytes."""
1455 """Output some data to the target."""
1457 """Inform the target that there is no more data to be written."""
1459 """fail() is called to indicate that the download has failed. 'why'
1460 is a Failure object indicating what went wrong. No further methods
1461 will be invoked on the IDownloadTarget after fail()."""
1462 def register_canceller(cb):
1463 """The CiphertextDownloader uses this to register a no-argument function
1464 that the target can call to cancel the download. Once this canceller
1465 is invoked, no further calls to write() or close() will be made."""
1467 """When the CiphertextDownloader is done, this finish() function will be
1468 called. Whatever it returns will be returned to the invoker of
1469 Downloader.download.
1472 class IDownloader(Interface):
1473 def download(uri, target):
1474 """Perform a CHK download, sending the data to the given target.
1475 'target' must provide IDownloadTarget.
1477 Returns a Deferred that fires (with the results of target.finish)
1478 when the download is finished, or errbacks if something went wrong."""
1480 class IEncryptedUploadable(Interface):
1481 def set_upload_status(upload_status):
1482 """Provide an IUploadStatus object that should be filled with status
1483 information. The IEncryptedUploadable is responsible for setting
1484 key-determination progress ('chk'), size, storage_index, and
1485 ciphertext-fetch progress. It may delegate some of this
1486 responsibility to others, in particular to the IUploadable."""
1489 """This behaves just like IUploadable.get_size()."""
1491 def get_all_encoding_parameters():
1492 """Return a Deferred that fires with a tuple of
1493 (k,happy,n,segment_size). The segment_size will be used as-is, and
1494 must match the following constraints: it must be a multiple of k, and
1495 it shouldn't be unreasonably larger than the file size (if
1496 segment_size is larger than filesize, the difference must be stored
1499 This usually passes through to the IUploadable method of the same
1502 The encoder strictly obeys the values returned by this method. To
1503 make an upload use non-default encoding parameters, you must arrange
1504 to control the values that this method returns.
1507 def get_storage_index():
1508 """Return a Deferred that fires with a 16-byte storage index.
1511 def read_encrypted(length, hash_only):
1512 """This behaves just like IUploadable.read(), but returns crypttext
1513 instead of plaintext. If hash_only is True, then this discards the
1514 data (and returns an empty list); this improves efficiency when
1515 resuming an interrupted upload (where we need to compute the
1516 plaintext hashes, but don't need the redundant encrypted data)."""
1518 def get_plaintext_hashtree_leaves(first, last, num_segments):
1519 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1520 plaintext segments, i.e. get the tagged hashes of the given segments.
1521 The segment size is expected to be generated by the
1522 IEncryptedUploadable before any plaintext is read or ciphertext
1523 produced, so that the segment hashes can be generated with only a
1526 This returns a Deferred which fires with a sequence of hashes, using:
1528 tuple(segment_hashes[first:last])
1530 'num_segments' is used to assert that the number of segments that the
1531 IEncryptedUploadable handled matches the number of segments that the
1532 encoder was expecting.
1534 This method must not be called until the final byte has been read
1535 from read_encrypted(). Once this method is called, read_encrypted()
1536 can never be called again.
1539 def get_plaintext_hash():
1540 """OBSOLETE; Get the hash of the whole plaintext.
1542 This returns a Deferred which fires with a tagged SHA-256 hash of the
1543 whole plaintext, obtained from hashutil.plaintext_hash(data).
1547 """Just like IUploadable.close()."""
1549 class IUploadable(Interface):
1550 def set_upload_status(upload_status):
1551 """Provide an IUploadStatus object that should be filled with status
1552 information. The IUploadable is responsible for setting
1553 key-determination progress ('chk')."""
1555 def set_default_encoding_parameters(params):
1556 """Set the default encoding parameters, which must be a dict mapping
1557 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1558 'max_segment_size'. These might have an influence on the final
1559 encoding parameters returned by get_all_encoding_parameters(), if the
1560 Uploadable doesn't have more specific preferences.
1562 This call is optional: if it is not used, the Uploadable will use
1563 some built-in defaults. If used, this method must be called before
1564 any other IUploadable methods to have any effect.
1568 """Return a Deferred that will fire with the length of the data to be
1569 uploaded, in bytes. This will be called before the data is actually
1570 used, to compute encoding parameters.
1573 def get_all_encoding_parameters():
1574 """Return a Deferred that fires with a tuple of
1575 (k,happy,n,segment_size). The segment_size will be used as-is, and
1576 must match the following constraints: it must be a multiple of k, and
1577 it shouldn't be unreasonably larger than the file size (if
1578 segment_size is larger than filesize, the difference must be stored
1581 The relative values of k and n allow some IUploadables to request
1582 better redundancy than others (in exchange for consuming more space
1585 Larger values of segment_size reduce hash overhead, while smaller
1586 values reduce memory footprint and cause data to be delivered in
1587 smaller pieces (which may provide a smoother and more predictable
1588 download experience).
1590 The encoder strictly obeys the values returned by this method. To
1591 make an upload use non-default encoding parameters, you must arrange
1592 to control the values that this method returns. One way to influence
1593 them may be to call set_encoding_parameters() before calling
1594 get_all_encoding_parameters().
1597 def get_encryption_key():
1598 """Return a Deferred that fires with a 16-byte AES key. This key will
1599 be used to encrypt the data. The key will also be hashed to derive
1602 Uploadables which want to achieve convergence should hash their file
1603 contents and the serialized_encoding_parameters to form the key
1604 (which of course requires a full pass over the data). Uploadables can
1605 use the upload.ConvergentUploadMixin class to achieve this
1608 Uploadables which do not care about convergence (or do not wish to
1609 make multiple passes over the data) can simply return a
1610 strongly-random 16 byte string.
1612 get_encryption_key() may be called multiple times: the IUploadable is
1613 required to return the same value each time.
1617 """Return a Deferred that fires with a list of strings (perhaps with
1618 only a single element) which, when concatenated together, contain the
1619 next 'length' bytes of data. If EOF is near, this may provide fewer
1620 than 'length' bytes. The total number of bytes provided by read()
1621 before it signals EOF must equal the size provided by get_size().
1623 If the data must be acquired through multiple internal read
1624 operations, returning a list instead of a single string may help to
1625 reduce string copies.
1627 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1628 so a 10kB file means length=3kB, 100kB file means length=30kB,
1629 and >=1MB file means length=300kB.
1631 This method provides for a single full pass through the data. Later
1632 use cases may desire multiple passes or access to only parts of the
1633 data (such as a mutable file making small edits-in-place). This API
1634 will be expanded once those use cases are better understood.
1638 """The upload is finished, and whatever filehandle was in use may be
1641 class IUploadResults(Interface):
1642 """I am returned by upload() methods. I contain a number of public
1643 attributes which can be read to determine the results of the upload. Some
1644 of these are functional, some are timing information. All of these may be
1647 .file_size : the size of the file, in bytes
1648 .uri : the CHK read-cap for the file
1649 .ciphertext_fetched : how many bytes were fetched by the helper
1650 .sharemap: dict mapping share identifier to set of serverids
1651 (binary strings). This indicates which servers were given
1652 which shares. For immutable files, the shareid is an
1653 integer (the share number, from 0 to N-1). For mutable
1654 files, it is a string of the form 'seq%d-%s-sh%d',
1655 containing the sequence number, the roothash, and the
1657 .servermap : dict mapping server peerid to a set of share numbers
1658 .timings : dict of timing information, mapping name to seconds (float)
1659 total : total upload time, start to finish
1660 storage_index : time to compute the storage index
1661 peer_selection : time to decide which peers will be used
1662 contacting_helper : initial helper query to upload/no-upload decision
1663 existence_check : helper pre-upload existence check
1664 helper_total : initial helper query to helper finished pushing
1665 cumulative_fetch : helper waiting for ciphertext requests
1666 total_fetch : helper start to last ciphertext response
1667 cumulative_encoding : just time spent in zfec
1668 cumulative_sending : just time spent waiting for storage servers
1669 hashes_and_close : last segment push to shareholder close
1670 total_encode_and_push : first encode to shareholder close
1674 class IDownloadResults(Interface):
1675 """I am created internally by download() methods. I contain a number of
1676 public attributes which contain details about the download process.::
1678 .file_size : the size of the file, in bytes
1679 .servers_used : set of server peerids that were used during download
1680 .server_problems : dict mapping server peerid to a problem string. Only
1681 servers that had problems (bad hashes, disconnects)
1683 .servermap : dict mapping server peerid to a set of share numbers. Only
1684 servers that had any shares are listed here.
1685 .timings : dict of timing information, mapping name to seconds (float)
1686 peer_selection : time to ask servers about shares
1687 servers_peer_selection : dict of peerid to DYHB-query time
1688 uri_extension : time to fetch a copy of the URI extension block
1689 hashtrees : time to fetch the hash trees
1690 segments : time to fetch, decode, and deliver segments
1691 cumulative_fetch : time spent waiting for storage servers
1692 cumulative_decode : just time spent in zfec
1693 cumulative_decrypt : just time spent in decryption
1694 total : total download time, start to finish
1695 fetch_per_server : dict of peerid to list of per-segment fetch times
1699 class IUploader(Interface):
1700 def upload(uploadable):
1701 """Upload the file. 'uploadable' must impement IUploadable. This
1702 returns a Deferred which fires with an IUploadResults instance, from
1703 which the URI of the file can be obtained as results.uri ."""
1705 def upload_ssk(write_capability, new_version, uploadable):
1706 """TODO: how should this work?"""
1708 class ICheckable(Interface):
1709 def check(monitor, verify=False, add_lease=False):
1710 """Check upon my health, optionally repairing any problems.
1712 This returns a Deferred that fires with an instance that provides
1713 ICheckResults, or None if the object is non-distributed (i.e. LIT
1716 The monitor will be checked periodically to see if the operation has
1717 been cancelled. If so, no new queries will be sent, and the Deferred
1718 will fire (with a OperationCancelledError) immediately.
1720 Filenodes and dirnodes (which provide IFilesystemNode) are also
1721 checkable. Instances that represent verifier-caps will be checkable
1722 but not downloadable. Some objects (like LIT files) do not actually
1723 live in the grid, and their checkers return None (non-distributed
1724 files are always healthy).
1726 If verify=False, a relatively lightweight check will be performed: I
1727 will ask all servers if they have a share for me, and I will believe
1728 whatever they say. If there are at least N distinct shares on the
1729 grid, my results will indicate r.is_healthy()==True. This requires a
1730 roundtrip to each server, but does not transfer very much data, so
1731 the network bandwidth is fairly low.
1733 If verify=True, a more resource-intensive check will be performed:
1734 every share will be downloaded, and the hashes will be validated on
1735 every bit. I will ignore any shares that failed their hash checks. If
1736 there are at least N distinct valid shares on the grid, my results
1737 will indicate r.is_healthy()==True. This requires N/k times as much
1738 download bandwidth (and server disk IO) as a regular download. If a
1739 storage server is holding a corrupt share, or is experiencing memory
1740 failures during retrieval, or is malicious or buggy, then
1741 verification will detect the problem, but checking will not.
1743 If add_lease=True, I will ensure that an up-to-date lease is present
1744 on each share. The lease secrets will be derived from by node secret
1745 (in BASEDIR/private/secret), so either I will add a new lease to the
1746 share, or I will merely renew the lease that I already had. In a
1747 future version of the storage-server protocol (once Accounting has
1748 been implemented), there may be additional options here to define the
1749 kind of lease that is obtained (which account number to claim, etc).
1751 TODO: any problems seen during checking will be reported to the
1752 health-manager.furl, a centralized object which is responsible for
1753 figuring out why files are unhealthy so corrective action can be
1757 def check_and_repair(monitor, verify=False, add_lease=False):
1758 """Like check(), but if the file/directory is not healthy, attempt to
1761 Any non-healthy result will cause an immediate repair operation, to
1762 generate and upload new shares. After repair, the file will be as
1763 healthy as we can make it. Details about what sort of repair is done
1764 will be put in the check-and-repair results. The Deferred will not
1765 fire until the repair is complete.
1767 This returns a Deferred which fires with an instance of
1768 ICheckAndRepairResults."""
1770 class IDeepCheckable(Interface):
1771 def start_deep_check(verify=False, add_lease=False):
1772 """Check upon the health of me and everything I can reach.
1774 This is a recursive form of check(), useable only on dirnodes.
1776 I return a Monitor, with results that are an IDeepCheckResults
1779 TODO: If any of the directories I traverse are unrecoverable, the
1780 Monitor will report failure. If any of the files I check upon are
1781 unrecoverable, those problems will be reported in the
1782 IDeepCheckResults as usual, and the Monitor will not report a
1786 def start_deep_check_and_repair(verify=False, add_lease=False):
1787 """Check upon the health of me and everything I can reach. Repair
1788 anything that isn't healthy.
1790 This is a recursive form of check_and_repair(), useable only on
1793 I return a Monitor, with results that are an
1794 IDeepCheckAndRepairResults object.
1796 TODO: If any of the directories I traverse are unrecoverable, the
1797 Monitor will report failure. If any of the files I check upon are
1798 unrecoverable, those problems will be reported in the
1799 IDeepCheckResults as usual, and the Monitor will not report a
1803 class ICheckResults(Interface):
1804 """I contain the detailed results of a check/verify operation.
1807 def get_storage_index():
1808 """Return a string with the (binary) storage index."""
1809 def get_storage_index_string():
1810 """Return a string with the (printable) abbreviated storage index."""
1812 """Return the (string) URI of the object that was checked."""
1815 """Return a boolean, True if the file/dir is fully healthy, False if
1816 it is damaged in any way. Non-distributed LIT files always return
1819 def is_recoverable():
1820 """Return a boolean, True if the file/dir can be recovered, False if
1821 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1822 files always return True."""
1824 def needs_rebalancing():
1825 """Return a boolean, True if the file/dir's reliability could be
1826 improved by moving shares to new servers. Non-distributed LIT files
1827 always return False."""
1831 """Return a dictionary that describes the state of the file/dir. LIT
1832 files always return an empty dictionary. Normal files and directories
1833 return a dictionary with the following keys (note that these use
1834 binary strings rather than base32-encoded ones) (also note that for
1835 mutable files, these counts are for the 'best' version):
1837 count-shares-good: the number of distinct good shares that were found
1838 count-shares-needed: 'k', the number of shares required for recovery
1839 count-shares-expected: 'N', the number of total shares generated
1840 count-good-share-hosts: the number of distinct storage servers with
1841 good shares. If this number is less than
1842 count-shares-good, then some shares are
1843 doubled up, increasing the correlation of
1844 failures. This indicates that one or more
1845 shares should be moved to an otherwise unused
1846 server, if one is available.
1847 count-corrupt-shares: the number of shares with integrity failures
1848 list-corrupt-shares: a list of 'share locators', one for each share
1849 that was found to be corrupt. Each share
1850 locator is a list of (serverid, storage_index,
1852 count-incompatible-shares: the number of shares which are of a share
1853 format unknown to this checker
1854 list-incompatible-shares: a list of 'share locators', one for each
1855 share that was found to be of an unknown
1856 format. Each share locator is a list of
1857 (serverid, storage_index, sharenum).
1858 servers-responding: list of (binary) storage server identifiers,
1859 one for each server which responded to the share
1860 query (even if they said they didn't have
1861 shares, and even if they said they did have
1862 shares but then didn't send them when asked, or
1863 dropped the connection, or returned a Failure,
1864 and even if they said they did have shares and
1865 sent incorrect ones when asked)
1866 sharemap: dict mapping share identifier to list of serverids
1867 (binary strings). This indicates which servers are holding
1868 which shares. For immutable files, the shareid is an
1869 integer (the share number, from 0 to N-1). For mutable
1870 files, it is a string of the form 'seq%d-%s-sh%d',
1871 containing the sequence number, the roothash, and the
1874 The following keys are most relevant for mutable files, but immutable
1875 files will provide sensible values too::
1877 count-wrong-shares: the number of shares for versions other than the
1878 'best' one (which is defined as being the
1879 recoverable version with the highest sequence
1880 number, then the highest roothash). These are
1881 either leftover shares from an older version
1882 (perhaps on a server that was offline when an
1883 update occurred), shares from an unrecoverable
1884 newer version, or shares from an alternate
1885 current version that results from an
1886 uncoordinated write collision. For a healthy
1887 file, this will equal 0.
1889 count-recoverable-versions: the number of recoverable versions of
1890 the file. For a healthy file, this will
1893 count-unrecoverable-versions: the number of unrecoverable versions
1894 of the file. For a healthy file, this
1900 """Return a string with a brief (one-line) summary of the results."""
1903 """Return a list of strings with more detailed results."""
1905 class ICheckAndRepairResults(Interface):
1906 """I contain the detailed results of a check/verify/repair operation.
1908 The IFilesystemNode.check()/verify()/repair() methods all return
1909 instances that provide ICheckAndRepairResults.
1912 def get_storage_index():
1913 """Return a string with the (binary) storage index."""
1914 def get_storage_index_string():
1915 """Return a string with the (printable) abbreviated storage index."""
1916 def get_repair_attempted():
1917 """Return a boolean, True if a repair was attempted. We might not
1918 attempt to repair the file because it was healthy, or healthy enough
1919 (i.e. some shares were missing but not enough to exceed some
1920 threshold), or because we don't know how to repair this object."""
1921 def get_repair_successful():
1922 """Return a boolean, True if repair was attempted and the file/dir
1923 was fully healthy afterwards. False if no repair was attempted or if
1924 a repair attempt failed."""
1925 def get_pre_repair_results():
1926 """Return an ICheckResults instance that describes the state of the
1927 file/dir before any repair was attempted."""
1928 def get_post_repair_results():
1929 """Return an ICheckResults instance that describes the state of the
1930 file/dir after any repair was attempted. If no repair was attempted,
1931 the pre-repair and post-repair results will be identical."""
1934 class IDeepCheckResults(Interface):
1935 """I contain the results of a deep-check operation.
1937 This is returned by a call to ICheckable.deep_check().
1940 def get_root_storage_index_string():
1941 """Return the storage index (abbreviated human-readable string) of
1942 the first object checked."""
1944 """Return a dictionary with the following keys::
1946 count-objects-checked: count of how many objects were checked
1947 count-objects-healthy: how many of those objects were completely
1949 count-objects-unhealthy: how many were damaged in some way
1950 count-objects-unrecoverable: how many were unrecoverable
1951 count-corrupt-shares: how many shares were found to have
1952 corruption, summed over all objects
1956 def get_corrupt_shares():
1957 """Return a set of (serverid, storage_index, sharenum) for all shares
1958 that were found to be corrupt. Both serverid and storage_index are
1961 def get_all_results():
1962 """Return a dictionary mapping pathname (a tuple of strings, ready to
1963 be slash-joined) to an ICheckResults instance, one for each object
1964 that was checked."""
1966 def get_results_for_storage_index(storage_index):
1967 """Retrive the ICheckResults instance for the given (binary)
1968 storage index. Raises KeyError if there are no results for that
1972 """Return a dictionary with the same keys as
1973 IDirectoryNode.deep_stats()."""
1975 class IDeepCheckAndRepairResults(Interface):
1976 """I contain the results of a deep-check-and-repair operation.
1978 This is returned by a call to ICheckable.deep_check_and_repair().
1981 def get_root_storage_index_string():
1982 """Return the storage index (abbreviated human-readable string) of
1983 the first object checked."""
1985 """Return a dictionary with the following keys::
1987 count-objects-checked: count of how many objects were checked
1988 count-objects-healthy-pre-repair: how many of those objects were
1989 completely healthy (before any
1991 count-objects-unhealthy-pre-repair: how many were damaged in
1993 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1994 count-objects-healthy-post-repair: how many of those objects were
1995 completely healthy (after any
1997 count-objects-unhealthy-post-repair: how many were damaged in
1999 count-objects-unrecoverable-post-repair: how many were
2001 count-repairs-attempted: repairs were attempted on this many
2002 objects. The count-repairs- keys will
2003 always be provided, however unless
2004 repair=true is present, they will all
2006 count-repairs-successful: how many repairs resulted in healthy
2008 count-repairs-unsuccessful: how many repairs resulted did not
2009 results in completely healthy objects
2010 count-corrupt-shares-pre-repair: how many shares were found to
2011 have corruption, summed over all
2012 objects examined (before any
2014 count-corrupt-shares-post-repair: how many shares were found to
2015 have corruption, summed over all
2016 objects examined (after any
2021 """Return a dictionary with the same keys as
2022 IDirectoryNode.deep_stats()."""
2024 def get_corrupt_shares():
2025 """Return a set of (serverid, storage_index, sharenum) for all shares
2026 that were found to be corrupt before any repair was attempted. Both
2027 serverid and storage_index are binary.
2029 def get_remaining_corrupt_shares():
2030 """Return a set of (serverid, storage_index, sharenum) for all shares
2031 that were found to be corrupt after any repair was completed. Both
2032 serverid and storage_index are binary. These are shares that need
2033 manual inspection and probably deletion.
2035 def get_all_results():
2036 """Return a dictionary mapping pathname (a tuple of strings, ready to
2037 be slash-joined) to an ICheckAndRepairResults instance, one for each
2038 object that was checked."""
2040 def get_results_for_storage_index(storage_index):
2041 """Retrive the ICheckAndRepairResults instance for the given (binary)
2042 storage index. Raises KeyError if there are no results for that
2046 class IRepairable(Interface):
2047 def repair(check_results):
2048 """Attempt to repair the given object. Returns a Deferred that fires
2049 with a IRepairResults object.
2051 I must be called with an object that implements ICheckResults, as
2052 proof that you have actually discovered a problem with this file. I
2053 will use the data in the checker results to guide the repair process,
2054 such as which servers provided bad data and should therefore be
2055 avoided. The ICheckResults object is inside the
2056 ICheckAndRepairResults object, which is returned by the
2057 ICheckable.check() method::
2059 d = filenode.check(repair=False)
2060 def _got_results(check_and_repair_results):
2061 check_results = check_and_repair_results.get_pre_repair_results()
2062 return filenode.repair(check_results)
2063 d.addCallback(_got_results)
2067 class IRepairResults(Interface):
2068 """I contain the results of a repair operation."""
2069 def get_successful(self):
2070 """Returns a boolean: True if the repair made the file healthy, False
2071 if not. Repair failure generally indicates a file that has been
2072 damaged beyond repair."""
2075 class IClient(Interface):
2076 def upload(uploadable):
2077 """Upload some data into a CHK, get back the UploadResults for it.
2078 @param uploadable: something that implements IUploadable
2079 @return: a Deferred that fires with the UploadResults instance.
2080 To get the URI for this file, use results.uri .
2083 def create_mutable_file(contents=""):
2084 """Create a new mutable file (with initial) contents, get back the
2087 @param contents: (bytestring, callable, or None): this provides the
2088 initial contents of the mutable file. If 'contents' is a bytestring,
2089 it will be used as-is. If 'contents' is a callable, it will be
2090 invoked with the new MutableFileNode instance and is expected to
2091 return a bytestring with the initial contents of the file (the
2092 callable can use node.get_writekey() to decide how to encrypt the
2093 initial contents, e.g. for a brand new dirnode with initial
2094 children). contents=None is equivalent to an empty string. Using
2095 content_maker= is more efficient than creating a mutable file and
2096 setting its contents in two separate operations.
2098 @return: a Deferred that fires with an IMutableFileNode instance.
2101 def create_dirnode(initial_children={}):
2102 """Create a new unattached dirnode, possibly with initial children.
2104 @param initial_children: dict with keys that are unicode child names,
2105 and values that are (childnode, metadata) tuples.
2107 @return: a Deferred that fires with the new IDirectoryNode instance.
2110 def create_node_from_uri(uri, rouri):
2111 """Create a new IFilesystemNode instance from the uri, synchronously.
2112 @param uri: a string or IURI-providing instance, or None. This could
2113 be for a LiteralFileNode, a CHK file node, a mutable file
2114 node, or a directory node
2115 @param rouri: a string or IURI-providing instance, or None. If the
2116 main uri is None, I will use the rouri instead. If I
2117 recognize the format of the main uri, I will ignore the
2118 rouri (because it can be derived from the writecap).
2120 @return: an instance that provides IFilesystemNode (or more usefully
2121 one of its subclasses). File-specifying URIs will result in
2122 IFileNode-providing instances, like ImmutableFileNode,
2123 LiteralFileNode, or MutableFileNode. Directory-specifying
2124 URIs will result in IDirectoryNode-providing instances, like
2128 class INodeMaker(Interface):
2129 """The NodeMaker is used to create IFilesystemNode instances. It can
2130 accept a filecap/dircap string and return the node right away. It can
2131 also create new nodes (i.e. upload a file, or create a mutable file)
2132 asynchronously. Once you have one of these nodes, you can use other
2133 methods to determine whether it is a file or directory, and to download
2134 or modify its contents.
2136 The NodeMaker encapsulates all the authorities that these
2137 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2138 Tahoe process will typically have a single NodeMaker, but unit tests may
2139 create simplified/mocked forms for testing purposes.
2141 def create_from_cap(writecap, readcap=None, **kwargs):
2142 """I create an IFilesystemNode from the given writecap/readcap. I can
2143 only provide nodes for existing file/directory objects: use my other
2144 methods to create new objects. I return synchronously."""
2146 def create_mutable_file(contents=None, keysize=None):
2147 """I create a new mutable file, and return a Deferred which will fire
2148 with the IMutableFileNode instance when it is ready. If contents= is
2149 provided (a bytestring), it will be used as the initial contents of
2150 the new file, otherwise the file will contain zero bytes. keysize= is
2151 for use by unit tests, to create mutable files that are smaller than
2154 def create_new_mutable_directory(initial_children={}):
2155 """I create a new mutable directory, and return a Deferred which will
2156 fire with the IDirectoryNode instance when it is ready. If
2157 initial_children= is provided (a dict mapping unicode child name to
2158 (childnode, metadata_dict) tuples), the directory will be populated
2159 with those children, otherwise it will be empty."""
2161 class IClientStatus(Interface):
2162 def list_all_uploads():
2163 """Return a list of uploader objects, one for each upload which
2164 currently has an object available (tracked with weakrefs). This is
2165 intended for debugging purposes."""
2166 def list_active_uploads():
2167 """Return a list of active IUploadStatus objects."""
2168 def list_recent_uploads():
2169 """Return a list of IUploadStatus objects for the most recently
2172 def list_all_downloads():
2173 """Return a list of downloader objects, one for each download which
2174 currently has an object available (tracked with weakrefs). This is
2175 intended for debugging purposes."""
2176 def list_active_downloads():
2177 """Return a list of active IDownloadStatus objects."""
2178 def list_recent_downloads():
2179 """Return a list of IDownloadStatus objects for the most recently
2180 started downloads."""
2182 class IUploadStatus(Interface):
2184 """Return a timestamp (float with seconds since epoch) indicating
2185 when the operation was started."""
2186 def get_storage_index():
2187 """Return a string with the (binary) storage index in use on this
2188 upload. Returns None if the storage index has not yet been
2191 """Return an integer with the number of bytes that will eventually
2192 be uploaded for this file. Returns None if the size is not yet known.
2195 """Return True if this upload is using a Helper, False if not."""
2197 """Return a string describing the current state of the upload
2200 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2201 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2202 made towards hashing the file to determine a CHK encryption key: if
2203 non-convergent encryption is in use, this will be trivial, otherwise
2204 the whole file must be hashed. 'ciphertext' describes how much of the
2205 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2206 uploads. 'encode_and_push' describes how much of the encode-and-push
2207 process has finished: for helper uploads this is dependent upon the
2208 helper providing progress reports. It might be reasonable to add all
2209 three numbers and report the sum to the user."""
2211 """Return True if the upload is currently active, False if not."""
2213 """Return an instance of UploadResults (which contains timing and
2214 sharemap information). Might return None if the upload is not yet
2217 """Each upload status gets a unique number: this method returns that
2218 number. This provides a handle to this particular upload, so a web
2219 page can generate a suitable hyperlink."""
2221 class IDownloadStatus(Interface):
2223 """Return a timestamp (float with seconds since epoch) indicating
2224 when the operation was started."""
2225 def get_storage_index():
2226 """Return a string with the (binary) storage index in use on this
2227 download. This may be None if there is no storage index (i.e. LIT
2230 """Return an integer with the number of bytes that will eventually be
2231 retrieved for this file. Returns None if the size is not yet known.
2234 """Return True if this download is using a Helper, False if not."""
2236 """Return a string describing the current state of the download
2239 """Returns a float (from 0.0 to 1.0) describing the amount of the
2240 download that has completed. This value will remain at 0.0 until the
2241 first byte of plaintext is pushed to the download target."""
2243 """Return True if the download is currently active, False if not."""
2245 """Each download status gets a unique number: this method returns
2246 that number. This provides a handle to this particular download, so a
2247 web page can generate a suitable hyperlink."""
2249 class IServermapUpdaterStatus(Interface):
2251 class IPublishStatus(Interface):
2253 class IRetrieveStatus(Interface):
2256 class NotCapableError(Exception):
2257 """You have tried to write to a read-only node."""
2259 class BadWriteEnablerError(Exception):
2262 class RIControlClient(RemoteInterface):
2264 def wait_for_client_connections(num_clients=int):
2265 """Do not return until we have connections to at least NUM_CLIENTS
2269 def upload_from_file_to_uri(filename=str,
2270 convergence=ChoiceOf(None,
2271 StringConstraint(2**20))):
2272 """Upload a file to the grid. This accepts a filename (which must be
2273 absolute) that points to a file on the node's local disk. The node will
2274 read the contents of this file, upload it to the grid, then return the
2275 URI at which it was uploaded. If convergence is None then a random
2276 encryption key will be used, else the plaintext will be hashed, then
2277 that hash will be mixed together with the "convergence" string to form
2282 def download_from_uri_to_file(uri=URI, filename=str):
2283 """Download a file from the grid, placing it on the node's local disk
2284 at the given filename (which must be absolute[?]). Returns the
2285 absolute filename where the file was written."""
2290 def get_memory_usage():
2291 """Return a dict describes the amount of memory currently in use. The
2292 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2293 measuring memory consupmtion in bytes."""
2294 return DictOf(str, int)
2296 def speed_test(count=int, size=int, mutable=Any()):
2297 """Write 'count' tempfiles to disk, all of the given size. Measure
2298 how long (in seconds) it takes to upload them all to the servers.
2299 Then measure how long it takes to download all of them. If 'mutable'
2300 is 'create', time creation of mutable files. If 'mutable' is
2301 'upload', then time access to the same mutable file instead of
2304 Returns a tuple of (upload_time, download_time).
2306 return (float, float)
2308 def measure_peer_response_time():
2309 """Send a short message to each connected peer, and measure the time
2310 it takes for them to respond to it. This is a rough measure of the
2311 application-level round trip time.
2313 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2316 return DictOf(Nodeid, float)
2318 UploadResults = Any() #DictOf(str, str)
2320 class RIEncryptedUploadable(RemoteInterface):
2321 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2326 def get_all_encoding_parameters():
2327 return (int, int, int, long)
2329 def read_encrypted(offset=Offset, length=ReadSize):
2336 class RICHKUploadHelper(RemoteInterface):
2337 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2341 Return a dictionary of version information.
2343 return DictOf(str, Any())
2345 def upload(reader=RIEncryptedUploadable):
2346 return UploadResults
2349 class RIHelper(RemoteInterface):
2350 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2354 Return a dictionary of version information.
2356 return DictOf(str, Any())
2358 def upload_chk(si=StorageIndex):
2359 """See if a file with a given storage index needs uploading. The
2360 helper will ask the appropriate storage servers to see if the file
2361 has already been uploaded. If so, the helper will return a set of
2362 'upload results' that includes whatever hashes are needed to build
2363 the read-cap, and perhaps a truncated sharemap.
2365 If the file has not yet been uploaded (or if it was only partially
2366 uploaded), the helper will return an empty upload-results dictionary
2367 and also an RICHKUploadHelper object that will take care of the
2368 upload process. The client should call upload() on this object and
2369 pass it a reference to an RIEncryptedUploadable object that will
2370 provide ciphertext. When the upload is finished, the upload() method
2371 will finish and return the upload results.
2373 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2376 class RIStatsProvider(RemoteInterface):
2377 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2379 Provides access to statistics and monitoring information.
2384 returns a dictionary containing 'counters' and 'stats', each a
2385 dictionary with string counter/stat name keys, and numeric values.
2386 counters are monotonically increasing measures of work done, and
2387 stats are instantaneous measures (potentially time averaged
2390 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2392 class RIStatsGatherer(RemoteInterface):
2393 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2395 Provides a monitoring service for centralised collection of stats
2398 def provide(provider=RIStatsProvider, nickname=str):
2400 @param provider: a stats collector instance which should be polled
2401 periodically by the gatherer to collect stats.
2402 @param nickname: a name useful to identify the provided client
2407 class IStatsProducer(Interface):
2410 returns a dictionary, with str keys representing the names of stats
2411 to be monitored, and numeric values.
2414 class RIKeyGenerator(RemoteInterface):
2415 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2417 Provides a service offering to make RSA key pairs.
2420 def get_rsa_key_pair(key_size=int):
2422 @param key_size: the size of the signature key.
2423 @return: tuple(verifying_key, signing_key)
2425 return TupleOf(str, str)
2428 class FileTooLargeError(Exception):
2431 class IValidatedThingProxy(Interface):
2433 """ Acquire a thing and validate it. Return a deferred which is
2434 eventually fired with self if the thing is valid or errbacked if it
2435 can't be acquired or validated."""
2437 class InsufficientVersionError(Exception):
2438 def __init__(self, needed, got):
2439 self.needed = needed
2442 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2445 class EmptyPathnameComponentError(Exception):
2446 """The webapi disallows empty pathname components."""