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 ShareData = StringConstraint(None)
19 URIExtensionData = StringConstraint(1000)
20 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
22 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
23 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
24 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
25 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
27 class RIStubClient(RemoteInterface):
28 """Each client publishes a service announcement for a dummy object called
29 the StubClient. This object doesn't actually offer any services, but the
30 announcement helps the Introducer keep track of which clients are
31 subscribed (so the grid admin can keep track of things like the size of
32 the grid and the client versions in use. This is the (empty)
33 RemoteInterface for the StubClient."""
35 class RIBucketWriter(RemoteInterface):
36 """ Objects of this kind live on the server side. """
37 def write(offset=Offset, data=ShareData):
42 If the data that has been written is incomplete or inconsistent then
43 the server will throw the data away, else it will store it for future
49 """Abandon all the data that has been written.
53 class RIBucketReader(RemoteInterface):
54 def read(offset=Offset, length=ReadSize):
57 def advise_corrupt_share(reason=str):
58 """Clients who discover hash failures in shares that they have
59 downloaded from me will use this method to inform me about the
60 failures. I will record their concern so that my operator can
61 manually inspect the shares in question. I return None.
63 This is a wrapper around RIStorageServer.advise_corrupt_share(),
64 which is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
85 class RIStorageServer(RemoteInterface):
86 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
90 Return a dictionary of version information.
92 return DictOf(str, Any())
94 def allocate_buckets(storage_index=StorageIndex,
95 renew_secret=LeaseRenewSecret,
96 cancel_secret=LeaseCancelSecret,
97 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
98 allocated_size=Offset, canary=Referenceable):
100 @param storage_index: the index of the bucket to be created or
102 @param sharenums: these are the share numbers (probably between 0 and
103 99) that the sender is proposing to store on this
105 @param renew_secret: This is the secret used to protect bucket refresh
106 This secret is generated by the client and
107 stored for later comparison by the server. Each
108 server is given a different secret.
109 @param cancel_secret: Like renew_secret, but protects bucket decref.
110 @param canary: If the canary is lost before close(), the bucket is
112 @return: tuple of (alreadygot, allocated), where alreadygot is what we
113 already have and allocated is what we hereby agree to accept.
114 New leases are added for shares in both lists.
116 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
117 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
119 def add_lease(storage_index=StorageIndex,
120 renew_secret=LeaseRenewSecret,
121 cancel_secret=LeaseCancelSecret):
123 Add a new lease on the given bucket. If the renew_secret matches an
124 existing lease, that lease will be renewed instead. If there is no
125 bucket for the given storage_index, return silently. (note that in
126 tahoe-1.3.0 and earlier, IndexError was raised if there was no
129 return Any() # returns None now, but future versions might change
131 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
133 Renew the lease on a given bucket, resetting the timer to 31 days.
134 Some networks will use this, some will not. If there is no bucket for
135 the given storage_index, IndexError will be raised.
137 For mutable shares, if the given renew_secret does not match an
138 existing lease, IndexError will be raised with a note listing the
139 server-nodeids on the existing leases, so leases on migrated shares
140 can be renewed or cancelled. For immutable shares, IndexError
141 (without the note) will be raised.
145 def cancel_lease(storage_index=StorageIndex,
146 cancel_secret=LeaseCancelSecret):
148 Cancel the lease on a given bucket. If this was the last lease on the
149 bucket, the bucket will be deleted. If there is no bucket for the
150 given storage_index, IndexError will be raised.
152 For mutable shares, if the given cancel_secret does not match an
153 existing lease, IndexError will be raised with a note listing the
154 server-nodeids on the existing leases, so leases on migrated shares
155 can be renewed or cancelled. For immutable shares, IndexError
156 (without the note) will be raised.
160 def get_buckets(storage_index=StorageIndex):
161 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
165 def slot_readv(storage_index=StorageIndex,
166 shares=ListOf(int), readv=ReadVector):
167 """Read a vector from the numbered shares associated with the given
168 storage index. An empty shares list means to return data from all
169 known shares. Returns a dictionary with one key per share."""
170 return DictOf(int, ReadData) # shnum -> results
172 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
173 secrets=TupleOf(WriteEnablerSecret,
176 tw_vectors=TestAndWriteVectorsForShares,
179 """General-purpose test-and-set operation for mutable slots. Perform
180 a bunch of comparisons against the existing shares. If they all pass,
181 then apply a bunch of write vectors to those shares. Then use the
182 read vectors to extract data from all the shares and return the data.
184 This method is, um, large. The goal is to allow clients to update all
185 the shares associated with a mutable file in a single round trip.
187 @param storage_index: the index of the bucket to be created or
189 @param write_enabler: a secret that is stored along with the slot.
190 Writes are accepted from any caller who can
191 present the matching secret. A different secret
192 should be used for each slot*server pair.
193 @param renew_secret: This is the secret used to protect bucket refresh
194 This secret is generated by the client and
195 stored for later comparison by the server. Each
196 server is given a different secret.
197 @param cancel_secret: Like renew_secret, but protects bucket decref.
199 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
200 cancel_secret). The first is required to perform any write. The
201 latter two are used when allocating new shares. To simply acquire a
202 new lease on existing shares, use an empty testv and an empty writev.
204 Each share can have a separate test vector (i.e. a list of
205 comparisons to perform). If all vectors for all shares pass, then all
206 writes for all shares are recorded. Each comparison is a 4-tuple of
207 (offset, length, operator, specimen), which effectively does a bool(
208 (read(offset, length)) OPERATOR specimen ) and only performs the
209 write if all these evaluate to True. Basic test-and-set uses 'eq'.
210 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
211 Write-if-same-or-newer uses 'le'.
213 Reads from the end of the container are truncated, and missing shares
214 behave like empty ones, so to assert that a share doesn't exist (for
215 use when creating a new share), use (0, 1, 'eq', '').
217 The write vector will be applied to the given share, expanding it if
218 necessary. A write vector applied to a share number that did not
219 exist previously will cause that share to be created.
221 Each write vector is accompanied by a 'new_length' argument. If
222 new_length is not None, use it to set the size of the container. This
223 can be used to pre-allocate space for a series of upcoming writes, or
224 truncate existing data. If the container is growing, new_length will
225 be applied before datav. If the container is shrinking, it will be
226 applied afterwards. If new_length==0, the share will be deleted.
228 The read vector is used to extract data from all known shares,
229 *before* any writes have been applied. The same vector is used for
230 all shares. This captures the state that was tested by the test
233 This method returns two values: a boolean and a dict. The boolean is
234 True if the write vectors were applied, False if not. The dict is
235 keyed by share number, and each value contains a list of strings, one
236 for each element of the read vector.
238 If the write_enabler is wrong, this will raise BadWriteEnablerError.
239 To enable share migration (using update_write_enabler), the exception
240 will have the nodeid used for the old write enabler embedded in it,
241 in the following string::
243 The write enabler was recorded by nodeid '%s'.
245 Note that the nodeid here is encoded using the same base32 encoding
246 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
249 return TupleOf(bool, DictOf(int, ReadData))
251 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
252 shnum=int, reason=str):
253 """Clients who discover hash failures in shares that they have
254 downloaded from me will use this method to inform me about the
255 failures. I will record their concern so that my operator can
256 manually inspect the shares in question. I return None.
258 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
259 (binary) storage index string, and 'shnum' is the integer share
260 number. 'reason' is a human-readable explanation of the problem,
261 probably including some expected hash values and the computed ones
262 which did not match. Corruption advisories for mutable shares should
263 include a hash of the public key (the same value that appears in the
264 mutable-file verify-cap), since the current share format does not
268 class IStorageBucketWriter(Interface):
270 Objects of this kind live on the client side.
272 def put_block(segmentnum=int, data=ShareData):
273 """@param data: For most segments, this data will be 'blocksize'
274 bytes in length. The last segment might be shorter.
275 @return: a Deferred that fires (with None) when the operation completes
278 def put_plaintext_hashes(hashes=ListOf(Hash)):
280 @return: a Deferred that fires (with None) when the operation completes
283 def put_crypttext_hashes(hashes=ListOf(Hash)):
285 @return: a Deferred that fires (with None) when the operation completes
288 def put_block_hashes(blockhashes=ListOf(Hash)):
290 @return: a Deferred that fires (with None) when the operation completes
293 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
295 @return: a Deferred that fires (with None) when the operation completes
298 def put_uri_extension(data=URIExtensionData):
299 """This block of data contains integrity-checking information (hashes
300 of plaintext, crypttext, and shares), as well as encoding parameters
301 that are necessary to recover the data. This is a serialized dict
302 mapping strings to other strings. The hash of this data is kept in
303 the URI and verified before any of the data is used. All buckets for
304 a given file contain identical copies of this data.
306 The serialization format is specified with the following pseudocode:
307 for k in sorted(dict.keys()):
308 assert re.match(r'^[a-zA-Z_\-]+$', k)
309 write(k + ':' + netstring(dict[k]))
311 @return: a Deferred that fires (with None) when the operation completes
315 """Finish writing and close the bucket. The share is not finalized
316 until this method is called: if the uploading client disconnects
317 before calling close(), the partially-written share will be
320 @return: a Deferred that fires (with None) when the operation completes
323 class IStorageBucketReader(Interface):
325 def get_block_data(blocknum=int, blocksize=int, size=int):
326 """Most blocks will be the same size. The last block might be shorter
332 def get_crypttext_hashes():
334 @return: ListOf(Hash)
337 def get_block_hashes(at_least_these=SetOf(int)):
339 @return: ListOf(Hash)
342 def get_share_hashes(at_least_these=SetOf(int)):
344 @return: ListOf(TupleOf(int, Hash))
347 def get_uri_extension():
349 @return: URIExtensionData
352 class IStorageBroker(Interface):
353 def get_servers_for_index(peer_selection_index):
355 @return: list of (peerid, versioned-rref) tuples
357 def get_all_servers():
359 @return: frozenset of (peerid, versioned-rref) tuples
361 def get_all_serverids():
363 @return: frozenset of serverid strings
365 def get_nickname_for_serverid(serverid):
367 @return: unicode nickname, or None
370 # methods moved from IntroducerClient, need review
371 def get_all_connections():
372 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
373 each active connection we've established to a remote service. This is
374 mostly useful for unit tests that need to wait until a certain number
375 of connections have been made."""
377 def get_all_connectors():
378 """Return a dict that maps from (nodeid, service_name) to a
379 RemoteServiceConnector instance for all services that we are actively
380 trying to connect to. Each RemoteServiceConnector has the following
383 service_name: the type of service provided, like 'storage'
384 announcement_time: when we first heard about this service
385 last_connect_time: when we last established a connection
386 last_loss_time: when we last lost a connection
388 version: the peer's version, from the most recent connection
389 oldest_supported: the peer's oldest supported version, same
391 rref: the RemoteReference, if connected, otherwise None
392 remote_host: the IAddress, if connected, otherwise None
394 This method is intended for monitoring interfaces, such as a web page
395 which describes connecting and connected peers.
398 def get_all_peerids():
399 """Return a frozenset of all peerids to whom we have a connection (to
400 one or more services) established. Mostly useful for unit tests."""
402 def get_all_connections_for(service_name):
403 """Return a frozenset of (nodeid, service_name, rref) tuples, one
404 for each active connection that provides the given SERVICE_NAME."""
406 def get_permuted_peers(service_name, key):
407 """Returns an ordered list of (peerid, rref) tuples, selecting from
408 the connections that provide SERVICE_NAME, using a hash-based
409 permutation keyed by KEY. This randomizes the service list in a
410 repeatable way, to distribute load over many peers.
414 class IURI(Interface):
415 def init_from_string(uri):
416 """Accept a string (as created by my to_string() method) and populate
417 this instance with its data. I am not normally called directly,
418 please use the module-level uri.from_string() function to convert
419 arbitrary URI strings into IURI-providing instances."""
422 """Return False if this URI be used to modify the data. Return True
423 if this URI cannot be used to modify the data."""
426 """Return True if the data can be modified by *somebody* (perhaps
427 someone who has a more powerful URI than this one)."""
430 """Return another IURI instance, which represents a read-only form of
431 this one. If is_readonly() is True, this returns self."""
433 def get_verify_cap():
434 """Return an instance that provides IVerifierURI, which can be used
435 to check on the availability of the file or directory, without
436 providing enough capabilities to actually read or modify the
437 contents. This may return None if the file does not need checking or
438 verification (e.g. LIT URIs).
442 """Return a string of printable ASCII characters, suitable for
443 passing into init_from_string."""
445 class IVerifierURI(Interface, IURI):
446 def init_from_string(uri):
447 """Accept a string (as created by my to_string() method) and populate
448 this instance with its data. I am not normally called directly,
449 please use the module-level uri.from_string() function to convert
450 arbitrary URI strings into IURI-providing instances."""
453 """Return a string of printable ASCII characters, suitable for
454 passing into init_from_string."""
456 class IDirnodeURI(Interface):
457 """I am a URI which represents a dirnode."""
460 class IFileURI(Interface):
461 """I am a URI which represents a filenode."""
463 """Return the length (in bytes) of the file that I represent."""
465 class IImmutableFileURI(IFileURI):
468 class IMutableFileURI(Interface):
469 """I am a URI which represents a mutable filenode."""
470 class IDirectoryURI(Interface):
472 class IReadonlyDirectoryURI(Interface):
475 class CannotPackUnknownNodeError(Exception):
476 """UnknownNodes (using filecaps from the future that we don't understand)
477 cannot yet be copied safely, so I refuse to copy them."""
479 class UnhandledCapTypeError(Exception):
480 """I recognize the cap/URI, but I cannot create an IFilesystemNode for
483 class NotDeepImmutableError(Exception):
484 """Deep-immutable directories can only contain deep-immutable children"""
486 # The hierarchy looks like this:
493 class IFilesystemNode(Interface):
495 """Return the strongest 'cap instance' associated with this node.
496 (writecap for writeable-mutable files/directories, readcap for
497 immutable or readonly-mutable files/directories). To convert this
498 into a string, call .to_string() on the result."""
501 """Return a readonly cap instance for this node. For immutable or
502 readonly nodes, get_cap() and get_readcap() return the same thing."""
504 def get_repair_cap():
505 """Return an IURI instance that can be used to repair the file, or
506 None if this node cannot be repaired (either because it is not
507 distributed, like a LIT file, or because the node does not represent
508 sufficient authority to create a repair-cap, like a read-only RSA
509 mutable file node [which cannot create the correct write-enablers]).
512 def get_verify_cap():
513 """Return an IVerifierURI instance that represents the
514 'verifiy/refresh capability' for this node. The holder of this
515 capability will be able to renew the lease for this node, protecting
516 it from garbage-collection. They will also be able to ask a server if
517 it holds a share for the file or directory.
522 Return the URI string that can be used by others to get access to
523 this node. If this node is read-only, the URI will only offer
524 read-only access. If this node is read-write, the URI will offer
527 If you have read-write access to a node and wish to share merely
528 read-only access with others, use get_readonly_uri().
531 def get_readonly_uri():
532 """Return the URI string that can be used by others to get read-only
533 access to this node. The result is a read-only URI, regardless of
534 whether this node is read-only or read-write.
536 If you have merely read-only access to this node, get_readonly_uri()
537 will return the same thing as get_uri().
540 def get_storage_index():
541 """Return a string with the (binary) storage index in use on this
542 download. This may be None if there is no storage index (i.e. LIT
546 """Return True if this reference provides mutable access to the given
547 file or directory (i.e. if you can modify it), or False if not. Note
548 that even if this reference is read-only, someone else may hold a
549 read-write reference to it."""
552 """Return True if this file or directory is mutable (by *somebody*,
553 not necessarily you), False if it is is immutable. Note that a file
554 might be mutable overall, but your reference to it might be
555 read-only. On the other hand, all references to an immutable file
556 will be read-only; there are no read-write references to an immutable
561 """Return the length (in bytes) of the data this node represents. For
562 directory nodes, I return the size of the backing store. I return
563 synchronously and do not consult the network, so for mutable objects,
564 I will return the most recently observed size for the object, or None
565 if I don't remember a size. Use get_current_size, which returns a
566 Deferred, if you want more up-to-date information."""
568 def get_current_size():
569 """I return a Deferred that fires with the length (in bytes) of the
570 data this node represents.
573 class IFileNode(IFilesystemNode):
574 """I am a node which represents a file: a sequence of bytes. I am not a
575 container, like IDirectoryNode."""
577 class IImmutableFileNode(IFileNode):
578 def read(consumer, offset=0, size=None):
579 """Download a portion (possibly all) of the file's contents, making
580 them available to the given IConsumer. Return a Deferred that fires
581 (with the consumer) when the consumer is unregistered (either because
582 the last byte has been given to it, or because the consumer threw an
583 exception during write(), possibly because it no longer wants to
584 receive data). The portion downloaded will start at 'offset' and
585 contain 'size' bytes (or the remainder of the file if size==None).
587 The consumer will be used in non-streaming mode: an IPullProducer
588 will be attached to it.
590 The consumer will not receive data right away: several network trips
591 must occur first. The order of events will be::
593 consumer.registerProducer(p, streaming)
594 (if streaming == False)::
595 consumer does p.resumeProducing()
597 consumer does p.resumeProducing()
598 consumer.write(data).. (repeat until all data is written)
599 consumer.unregisterProducer()
600 deferred.callback(consumer)
602 If a download error occurs, or an exception is raised by
603 consumer.registerProducer() or consumer.write(), I will call
604 consumer.unregisterProducer() and then deliver the exception via
605 deferred.errback(). To cancel the download, the consumer should call
606 p.stopProducing(), which will result in an exception being delivered
607 via deferred.errback().
609 See src/allmydata/util/consumer.py for an example of a simple
610 download-to-memory consumer.
613 class IMutableFileNode(IFileNode):
614 """I provide access to a 'mutable file', which retains its identity
615 regardless of what contents are put in it.
617 The consistency-vs-availability problem means that there might be
618 multiple versions of a file present in the grid, some of which might be
619 unrecoverable (i.e. have fewer than 'k' shares). These versions are
620 loosely ordered: each has a sequence number and a hash, and any version
621 with seqnum=N was uploaded by a node which has seen at least one version
624 The 'servermap' (an instance of IMutableFileServerMap) is used to
625 describe the versions that are known to be present in the grid, and which
626 servers are hosting their shares. It is used to represent the 'state of
627 the world', and is used for this purpose by my test-and-set operations.
628 Downloading the contents of the mutable file will also return a
629 servermap. Uploading a new version into the mutable file requires a
630 servermap as input, and the semantics of the replace operation is
631 'replace the file with my new version if it looks like nobody else has
632 changed the file since my previous download'. Because the file is
633 distributed, this is not a perfect test-and-set operation, but it will do
634 its best. If the replace process sees evidence of a simultaneous write,
635 it will signal an UncoordinatedWriteError, so that the caller can take
639 Most readers will want to use the 'best' current version of the file, and
640 should use my 'download_best_version()' method.
642 To unconditionally replace the file, callers should use overwrite(). This
643 is the mode that user-visible mutable files will probably use.
645 To apply some delta to the file, call modify() with a callable modifier
646 function that can apply the modification that you want to make. This is
647 the mode that dirnodes will use, since most directory modification
648 operations can be expressed in terms of deltas to the directory state.
651 Three methods are available for users who need to perform more complex
652 operations. The first is get_servermap(), which returns an up-to-date
653 servermap using a specified mode. The second is download_version(), which
654 downloads a specific version (not necessarily the 'best' one). The third
655 is 'upload', which accepts new contents and a servermap (which must have
656 been updated with MODE_WRITE). The upload method will attempt to apply
657 the new contents as long as no other node has modified the file since the
658 servermap was updated. This might be useful to a caller who wants to
659 merge multiple versions into a single new one.
661 Note that each time the servermap is updated, a specific 'mode' is used,
662 which determines how many peers are queried. To use a servermap for my
663 replace() method, that servermap must have been updated in MODE_WRITE.
664 These modes are defined in allmydata.mutable.common, and consist of
665 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
666 allmydata/mutable/servermap.py for details about the differences.
668 Mutable files are currently limited in size (about 3.5MB max) and can
669 only be retrieved and updated all-at-once, as a single big string. Future
670 versions of our mutable files will remove this restriction.
673 def download_best_version():
674 """Download the 'best' available version of the file, meaning one of
675 the recoverable versions with the highest sequence number. If no
676 uncoordinated writes have occurred, and if enough shares are
677 available, then this will be the most recent version that has been
680 I update an internal servermap with MODE_READ, determine which
681 version of the file is indicated by
682 servermap.best_recoverable_version(), and return a Deferred that
683 fires with its contents. If no version is recoverable, the Deferred
684 will errback with UnrecoverableFileError.
687 def get_size_of_best_version():
688 """Find the size of the version that would be downloaded with
689 download_best_version(), without actually downloading the whole file.
691 I return a Deferred that fires with an integer.
694 def overwrite(new_contents):
695 """Unconditionally replace the contents of the mutable file with new
696 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
697 is only appropriate to use when the new contents of the file are
698 completely unrelated to the old ones, and you do not care about other
701 I return a Deferred that fires (with a PublishStatus object) when the
702 update has completed.
705 def modify(modifier_cb):
706 """Modify the contents of the file, by downloading the current
707 version, applying the modifier function (or bound method), then
708 uploading the new version. I return a Deferred that fires (with a
709 PublishStatus object) when the update is complete.
711 The modifier callable will be given three arguments: a string (with
712 the old contents), a 'first_time' boolean, and a servermap. As with
713 download_best_version(), the old contents will be from the best
714 recoverable version, but the modifier can use the servermap to make
715 other decisions (such as refusing to apply the delta if there are
716 multiple parallel versions, or if there is evidence of a newer
717 unrecoverable version). 'first_time' will be True the first time the
718 modifier is called, and False on any subsequent calls.
720 The callable should return a string with the new contents. The
721 callable must be prepared to be called multiple times, and must
722 examine the input string to see if the change that it wants to make
723 is already present in the old version. If it does not need to make
724 any changes, it can either return None, or return its input string.
726 If the modifier raises an exception, it will be returned in the
731 def get_servermap(mode):
732 """Return a Deferred that fires with an IMutableFileServerMap
733 instance, updated using the given mode.
736 def download_version(servermap, version):
737 """Download a specific version of the file, using the servermap
738 as a guide to where the shares are located.
740 I return a Deferred that fires with the requested contents, or
741 errbacks with UnrecoverableFileError. Note that a servermap which was
742 updated with MODE_ANYTHING or MODE_READ may not know about shares for
743 all versions (those modes stop querying servers as soon as they can
744 fulfil their goals), so you may want to use MODE_CHECK (which checks
745 everything) to get increased visibility.
748 def upload(new_contents, servermap):
749 """Replace the contents of the file with new ones. This requires a
750 servermap that was previously updated with MODE_WRITE.
752 I attempt to provide test-and-set semantics, in that I will avoid
753 modifying any share that is different than the version I saw in the
754 servermap. However, if another node is writing to the file at the
755 same time as me, I may manage to update some shares while they update
756 others. If I see any evidence of this, I will signal
757 UncoordinatedWriteError, and the file will be left in an inconsistent
758 state (possibly the version you provided, possibly the old version,
759 possibly somebody else's version, and possibly a mix of shares from
762 The recommended response to UncoordinatedWriteError is to either
763 return it to the caller (since they failed to coordinate their
764 writes), or to attempt some sort of recovery. It may be sufficient to
765 wait a random interval (with exponential backoff) and repeat your
766 operation. If I do not signal UncoordinatedWriteError, then I was
767 able to write the new version without incident.
769 I return a Deferred that fires (with a PublishStatus object) when the
770 publish has completed. I will update the servermap in-place with the
771 location of all new shares.
775 """Return this filenode's writekey, or None if the node does not have
776 write-capability. This may be used to assist with data structures
777 that need to make certain data available only to writers, such as the
778 read-write child caps in dirnodes. The recommended process is to have
779 reader-visible data be submitted to the filenode in the clear (where
780 it will be encrypted by the filenode using the readkey), but encrypt
781 writer-visible data using this writekey.
784 class NotEnoughSharesError(Exception):
785 """Download was unable to get enough shares, or upload was unable to
786 place 'shares_of_happiness' shares."""
788 class NoSharesError(Exception):
789 """Upload or Download was unable to get any shares at all."""
791 class UnableToFetchCriticalDownloadDataError(Exception):
792 """I was unable to fetch some piece of critical data which is supposed to
793 be identically present in all shares."""
795 class NoServersError(Exception):
796 """Upload wasn't given any servers to work with, usually indicating a
797 network or Introducer problem."""
799 class ExistingChildError(Exception):
800 """A directory node was asked to add or replace a child that already
801 exists, and overwrite= was set to False."""
803 class NoSuchChildError(Exception):
804 """A directory node was asked to fetch a child which does not exist."""
806 class IDirectoryNode(IFilesystemNode):
807 """I represent a filesystem node that is a container, with a
808 name-to-child mapping, holding the tahoe equivalent of a directory. All
809 child names are unicode strings, and all children are some sort of
810 IFilesystemNode (either files or subdirectories).
815 The dirnode ('1') URI returned by this method can be used in
816 set_uri() on a different directory ('2') to 'mount' a reference to
817 this directory ('1') under the other ('2'). This URI is just a
818 string, so it can be passed around through email or other out-of-band
822 def get_readonly_uri():
824 The dirnode ('1') URI returned by this method can be used in
825 set_uri() on a different directory ('2') to 'mount' a reference to
826 this directory ('1') under the other ('2'). This URI is just a
827 string, so it can be passed around through email or other out-of-band
832 """I return a Deferred that fires with a dictionary mapping child
833 name (a unicode string) to (node, metadata_dict) tuples, in which
834 'node' is an IFilesystemNode (either IFileNode or IDirectoryNode),
835 and 'metadata_dict' is a dictionary of metadata."""
838 """I return a Deferred that fires with a boolean, True if there
839 exists a child of the given name, False if not. The child name must
840 be a unicode string."""
843 """I return a Deferred that fires with a specific named child node,
844 either an IFileNode or an IDirectoryNode. The child name must be a
845 unicode string. I raise NoSuchChildError if I do not have a child by
848 def get_metadata_for(name):
849 """I return a Deferred that fires with the metadata dictionary for a
850 specific named child node. This metadata is stored in the *edge*, not
851 in the child, so it is attached to the parent dirnode rather than the
852 child dir-or-file-node. The child name must be a unicode string. I
853 raise NoSuchChildError if I do not have a child by that name."""
855 def set_metadata_for(name, metadata):
856 """I replace any existing metadata for the named child with the new
857 metadata. The child name must be a unicode string. This metadata is
858 stored in the *edge*, not in the child, so it is attached to the
859 parent dirnode rather than the child dir-or-file-node. I return a
860 Deferred (that fires with this dirnode) when the operation is
861 complete. I raise NoSuchChildError if I do not have a child by that
864 def get_child_at_path(path):
865 """Transform a child path into an IFilesystemNode.
867 I perform a recursive series of 'get' operations to find the named
868 descendant node. I return a Deferred that fires with the node, or
869 errbacks with NoSuchChildError if the node could not be found.
871 The path can be either a single string (slash-separated) or a list of
872 path-name elements. All elements must be unicode strings.
875 def get_child_and_metadata_at_path(path):
876 """Transform a child path into an IFilesystemNode and metadata.
878 I am like get_child_at_path(), but my Deferred fires with a tuple of
879 (node, metadata). The metadata comes from the last edge. If the path
880 is empty, the metadata will be an empty dictionary.
883 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
884 """I add a child (by writecap+readcap) at the specific name. I return
885 a Deferred that fires when the operation finishes. If overwrite= is
886 True, I will replace any existing child of the same name, otherwise
887 an existing child will cause me to return ExistingChildError. The
888 child name must be a unicode string.
890 The child caps could be for a file, or for a directory. If the new
891 child is read/write, you will provide both writecap and readcap. If
892 the child is read-only, you will provide the readcap write (i.e. the
893 writecap= and readcap= arguments will both be the child's readcap).
894 The filecaps are typically obtained from an IFilesystemNode with
895 get_uri() and get_readonly_uri().
897 If metadata= is provided, I will use it as the metadata for the named
898 edge. This will replace any existing metadata. If metadata= is left
899 as the default value of None, I will set ['mtime'] to the current
900 time, and I will set ['ctime'] to the current time if there was not
901 already a child by this name present. This roughly matches the
902 ctime/mtime semantics of traditional filesystems.
904 If this directory node is read-only, the Deferred will errback with a
907 def set_children(entries, overwrite=True):
908 """Add multiple children (by writecap+readcap) to a directory node.
909 Takes a dictionary, with childname as keys and (writecap, readcap)
910 tuples (or (writecap, readcap, metadata) triples) as values. Returns
911 a Deferred that fires (with this dirnode) when the operation
912 finishes. This is equivalent to calling set_uri() multiple times, but
913 is much more efficient. All child names must be unicode strings.
916 def set_node(name, child, metadata=None, overwrite=True):
917 """I add a child at the specific name. I return a Deferred that fires
918 when the operation finishes. This Deferred will fire with the child
919 node that was just added. I will replace any existing child of the
920 same name. The child name must be a unicode string. The 'child'
921 instance must be an instance providing IFilesystemNode.
923 If metadata= is provided, I will use it as the metadata for the named
924 edge. This will replace any existing metadata. If metadata= is left
925 as the default value of None, I will set ['mtime'] to the current
926 time, and I will set ['ctime'] to the current time if there was not
927 already a child by this name present. This roughly matches the
928 ctime/mtime semantics of traditional filesystems.
930 If this directory node is read-only, the Deferred will errback with a
933 def set_nodes(entries, overwrite=True):
934 """Add multiple children to a directory node. Takes a dict mapping
935 unicode childname to (child_node, metdata) tuples. If metdata=None,
936 the original metadata is left unmodified. Returns a Deferred that
937 fires (with this dirnode) when the operation finishes. This is
938 equivalent to calling set_node() multiple times, but is much more
941 def add_file(name, uploadable, metadata=None, overwrite=True):
942 """I upload a file (using the given IUploadable), then attach the
943 resulting ImmutableFileNode to the directory at the given name. I set
944 metadata the same way as set_uri and set_node. The child name must be
947 I return a Deferred that fires (with the IFileNode of the uploaded
948 file) when the operation completes."""
951 """I remove the child at the specific name. I return a Deferred that
952 fires when the operation finishes. The child name must be a unicode
953 string. I raise NoSuchChildError if I do not have a child by that
956 def create_subdirectory(name, initial_children={}, overwrite=True):
957 """I create and attach a directory at the given name. The new
958 directory can be empty, or it can be populated with children
959 according to 'initial_children', which takes a dictionary in the same
960 format as set_nodes (i.e. mapping unicode child name to (childnode,
961 metadata) tuples). The child name must be a unicode string. I return
962 a Deferred that fires (with the new directory node) when the
963 operation finishes."""
965 def move_child_to(current_child_name, new_parent, new_child_name=None,
967 """I take one of my children and move them to a new parent. The child
968 is referenced by name. On the new parent, the child will live under
969 'new_child_name', which defaults to 'current_child_name'. TODO: what
970 should we do about metadata? I return a Deferred that fires when the
971 operation finishes. The child name must be a unicode string. I raise
972 NoSuchChildError if I do not have a child by that name."""
974 def build_manifest():
975 """I generate a table of everything reachable from this directory.
976 I also compute deep-stats as described below.
978 I return a Monitor. The Monitor's results will be a dictionary with
981 res['manifest']: a list of (path, cap) tuples for all nodes
982 (directories and files) reachable from this one.
983 'path' will be a tuple of unicode strings. The
984 origin dirnode will be represented by an empty path
986 res['verifycaps']: a list of (printable) verifycap strings, one for
987 each reachable non-LIT node. This is a set:
988 it will contain no duplicates.
989 res['storage-index']: a list of (base32) storage index strings,
990 one for each reachable non-LIT node. This is
991 a set: it will contain no duplicates.
992 res['stats']: a dictionary, the same that is generated by
993 start_deep_stats() below.
995 The Monitor will also have an .origin_si attribute with the (binary)
996 storage index of the starting point.
999 def start_deep_stats():
1000 """Return a Monitor, examining all nodes (directories and files)
1001 reachable from this one. The Monitor's results will be a dictionary
1002 with the following keys::
1004 count-immutable-files: count of how many CHK files are in the set
1005 count-mutable-files: same, for mutable files (does not include
1007 count-literal-files: same, for LIT files
1008 count-files: sum of the above three
1010 count-directories: count of directories
1012 size-immutable-files: total bytes for all CHK files in the set
1013 size-mutable-files (TODO): same, for current version of all mutable
1014 files, does not include directories
1015 size-literal-files: same, for LIT files
1016 size-directories: size of mutable files used by directories
1018 largest-directory: number of bytes in the largest directory
1019 largest-directory-children: number of children in the largest
1021 largest-immutable-file: number of bytes in the largest CHK file
1023 size-mutable-files is not yet implemented, because it would involve
1024 even more queries than deep_stats does.
1026 The Monitor will also have an .origin_si attribute with the (binary)
1027 storage index of the starting point.
1029 This operation will visit every directory node underneath this one,
1030 and can take a long time to run. On a typical workstation with good
1031 bandwidth, this can examine roughly 15 directories per second (and
1032 takes several minutes of 100% CPU for ~1700 directories).
1035 class ICodecEncoder(Interface):
1036 def set_params(data_size, required_shares, max_shares):
1037 """Set up the parameters of this encoder.
1039 This prepares the encoder to perform an operation that converts a
1040 single block of data into a number of shares, such that a future
1041 ICodecDecoder can use a subset of these shares to recover the
1042 original data. This operation is invoked by calling encode(). Once
1043 the encoding parameters are set up, the encode operation can be
1044 invoked multiple times.
1046 set_params() prepares the encoder to accept blocks of input data that
1047 are exactly 'data_size' bytes in length. The encoder will be prepared
1048 to produce 'max_shares' shares for each encode() operation (although
1049 see the 'desired_share_ids' to use less CPU). The encoding math will
1050 be chosen such that the decoder can get by with as few as
1051 'required_shares' of these shares and still reproduce the original
1052 data. For example, set_params(1000, 5, 5) offers no redundancy at
1053 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1055 Numerical Restrictions: 'data_size' is required to be an integral
1056 multiple of 'required_shares'. In general, the caller should choose
1057 required_shares and max_shares based upon their reliability
1058 requirements and the number of peers available (the total storage
1059 space used is roughly equal to max_shares*data_size/required_shares),
1060 then choose data_size to achieve the memory footprint desired (larger
1061 data_size means more efficient operation, smaller data_size means
1062 smaller memory footprint).
1064 In addition, 'max_shares' must be equal to or greater than
1065 'required_shares'. Of course, setting them to be equal causes
1066 encode() to degenerate into a particularly slow form of the 'split'
1069 See encode() for more details about how these parameters are used.
1071 set_params() must be called before any other ICodecEncoder methods
1076 """Return the 3-tuple of data_size, required_shares, max_shares"""
1078 def get_encoder_type():
1079 """Return a short string that describes the type of this encoder.
1081 There is required to be a global table of encoder classes. This method
1082 returns an index into this table; the value at this index is an
1083 encoder class, and this encoder is an instance of that class.
1086 def get_block_size():
1087 """Return the length of the shares that encode() will produce.
1090 def encode_proposal(data, desired_share_ids=None):
1091 """Encode some data.
1093 'data' must be a string (or other buffer object), and len(data) must
1094 be equal to the 'data_size' value passed earlier to set_params().
1096 This will return a Deferred that will fire with two lists. The first
1097 is a list of shares, each of which is a string (or other buffer
1098 object) such that len(share) is the same as what get_share_size()
1099 returned earlier. The second is a list of shareids, in which each is
1100 an integer. The lengths of the two lists will always be equal to each
1101 other. The user should take care to keep each share closely
1102 associated with its shareid, as one is useless without the other.
1104 The length of this output list will normally be the same as the value
1105 provided to the 'max_shares' parameter of set_params(). This may be
1106 different if 'desired_share_ids' is provided.
1108 'desired_share_ids', if provided, is required to be a sequence of
1109 ints, each of which is required to be >= 0 and < max_shares. If not
1110 provided, encode() will produce 'max_shares' shares, as if
1111 'desired_share_ids' were set to range(max_shares). You might use this
1112 if you initially thought you were going to use 10 peers, started
1113 encoding, and then two of the peers dropped out: you could use
1114 desired_share_ids= to skip the work (both memory and CPU) of
1115 producing shares for the peers which are no longer available.
1119 def encode(inshares, desired_share_ids=None):
1120 """Encode some data. This may be called multiple times. Each call is
1123 inshares is a sequence of length required_shares, containing buffers
1124 (i.e. strings), where each buffer contains the next contiguous
1125 non-overlapping segment of the input data. Each buffer is required to
1126 be the same length, and the sum of the lengths of the buffers is
1127 required to be exactly the data_size promised by set_params(). (This
1128 implies that the data has to be padded before being passed to
1129 encode(), unless of course it already happens to be an even multiple
1130 of required_shares in length.)
1132 ALSO: the requirement to break up your data into 'required_shares'
1133 chunks before calling encode() feels a bit surprising, at least from
1134 the point of view of a user who doesn't know how FEC works. It feels
1135 like an implementation detail that has leaked outside the
1136 abstraction barrier. Can you imagine a use case in which the data to
1137 be encoded might already be available in pre-segmented chunks, such
1138 that it is faster or less work to make encode() take a list rather
1139 than splitting a single string?
1141 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1142 is supposed to *produce* shares, so what it *accepts* should be
1143 something other than shares. Other places in this interface use the
1144 word 'data' for that-which-is-not-shares.. maybe we should use that
1147 'desired_share_ids', if provided, is required to be a sequence of
1148 ints, each of which is required to be >= 0 and < max_shares. If not
1149 provided, encode() will produce 'max_shares' shares, as if
1150 'desired_share_ids' were set to range(max_shares). You might use this
1151 if you initially thought you were going to use 10 peers, started
1152 encoding, and then two of the peers dropped out: you could use
1153 desired_share_ids= to skip the work (both memory and CPU) of
1154 producing shares for the peers which are no longer available.
1156 For each call, encode() will return a Deferred that fires with two
1157 lists, one containing shares and the other containing the shareids.
1158 The get_share_size() method can be used to determine the length of
1159 the share strings returned by encode(). Each shareid is a small
1160 integer, exactly as passed into 'desired_share_ids' (or
1161 range(max_shares), if desired_share_ids was not provided).
1163 The shares and their corresponding shareids are required to be kept
1164 together during storage and retrieval. Specifically, the share data is
1165 useless by itself: the decoder needs to be told which share is which
1166 by providing it with both the shareid and the actual share data.
1168 This function will allocate an amount of memory roughly equal to::
1170 (max_shares - required_shares) * get_share_size()
1172 When combined with the memory that the caller must allocate to
1173 provide the input data, this leads to a memory footprint roughly
1174 equal to the size of the resulting encoded shares (i.e. the expansion
1175 factor times the size of the input segment).
1180 # returning a list of (shareidN,shareN) tuples instead of a pair of
1181 # lists (shareids..,shares..). Brian thought the tuples would
1182 # encourage users to keep the share and shareid together throughout
1183 # later processing, Zooko pointed out that the code to iterate
1184 # through two lists is not really more complicated than using a list
1185 # of tuples and there's also a performance improvement
1187 # having 'data_size' not required to be an integral multiple of
1188 # 'required_shares'. Doing this would require encode() to perform
1189 # padding internally, and we'd prefer to have any padding be done
1190 # explicitly by the caller. Yes, it is an abstraction leak, but
1191 # hopefully not an onerous one.
1194 class ICodecDecoder(Interface):
1195 def set_params(data_size, required_shares, max_shares):
1196 """Set the params. They have to be exactly the same ones that were
1197 used for encoding."""
1199 def get_needed_shares():
1200 """Return the number of shares needed to reconstruct the data.
1201 set_params() is required to be called before this."""
1203 def decode(some_shares, their_shareids):
1204 """Decode a partial list of shares into data.
1206 'some_shares' is required to be a sequence of buffers of sharedata, a
1207 subset of the shares returned by ICodecEncode.encode(). Each share is
1208 required to be of the same length. The i'th element of their_shareids
1209 is required to be the shareid of the i'th buffer in some_shares.
1211 This returns a Deferred which fires with a sequence of buffers. This
1212 sequence will contain all of the segments of the original data, in
1213 order. The sum of the lengths of all of the buffers will be the
1214 'data_size' value passed into the original ICodecEncode.set_params()
1215 call. To get back the single original input block of data, use
1216 ''.join(output_buffers), or you may wish to simply write them in
1217 order to an output file.
1219 Note that some of the elements in the result sequence may be
1220 references to the elements of the some_shares input sequence. In
1221 particular, this means that if those share objects are mutable (e.g.
1222 arrays) and if they are changed, then both the input (the
1223 'some_shares' parameter) and the output (the value given when the
1224 deferred is triggered) will change.
1226 The length of 'some_shares' is required to be exactly the value of
1227 'required_shares' passed into the original ICodecEncode.set_params()
1231 class IEncoder(Interface):
1232 """I take an object that provides IEncryptedUploadable, which provides
1233 encrypted data, and a list of shareholders. I then encode, hash, and
1234 deliver shares to those shareholders. I will compute all the necessary
1235 Merkle hash trees that are necessary to validate the crypttext that
1236 eventually comes back from the shareholders. I provide the URI Extension
1237 Block Hash, and the encoding parameters, both of which must be included
1240 I do not choose shareholders, that is left to the IUploader. I must be
1241 given a dict of RemoteReferences to storage buckets that are ready and
1242 willing to receive data.
1246 """Specify the number of bytes that will be encoded. This must be
1247 peformed before get_serialized_params() can be called.
1249 def set_params(params):
1250 """Override the default encoding parameters. 'params' is a tuple of
1251 (k,d,n), where 'k' is the number of required shares, 'd' is the
1252 shares_of_happiness, and 'n' is the total number of shares that will
1255 Encoding parameters can be set in three ways. 1: The Encoder class
1256 provides defaults (3/7/10). 2: the Encoder can be constructed with
1257 an 'options' dictionary, in which the
1258 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1259 set_params((k,d,n)) can be called.
1261 If you intend to use set_params(), you must call it before
1262 get_share_size or get_param are called.
1265 def set_encrypted_uploadable(u):
1266 """Provide a source of encrypted upload data. 'u' must implement
1267 IEncryptedUploadable.
1269 When this is called, the IEncryptedUploadable will be queried for its
1270 length and the storage_index that should be used.
1272 This returns a Deferred that fires with this Encoder instance.
1274 This must be performed before start() can be called.
1277 def get_param(name):
1278 """Return an encoding parameter, by name.
1280 'storage_index': return a string with the (16-byte truncated SHA-256
1281 hash) storage index to which these shares should be
1284 'share_counts': return a tuple describing how many shares are used:
1285 (needed_shares, shares_of_happiness, total_shares)
1287 'num_segments': return an int with the number of segments that
1290 'segment_size': return an int with the size of each segment.
1292 'block_size': return the size of the individual blocks that will
1293 be delivered to a shareholder's put_block() method. By
1294 knowing this, the shareholder will be able to keep all
1295 blocks in a single file and still provide random access
1296 when reading them. # TODO: can we avoid exposing this?
1298 'share_size': an int with the size of the data that will be stored
1299 on each shareholder. This is aggregate amount of data
1300 that will be sent to the shareholder, summed over all
1301 the put_block() calls I will ever make. It is useful to
1302 determine this size before asking potential
1303 shareholders whether they will grant a lease or not,
1304 since their answers will depend upon how much space we
1305 need. TODO: this might also include some amount of
1306 overhead, like the size of all the hashes. We need to
1307 decide whether this is useful or not.
1309 'serialized_params': a string with a concise description of the
1310 codec name and its parameters. This may be passed
1311 into the IUploadable to let it make sure that
1312 the same file encoded with different parameters
1313 will result in different storage indexes.
1315 Once this is called, set_size() and set_params() may not be called.
1318 def set_shareholders(shareholders):
1319 """Tell the encoder where to put the encoded shares. 'shareholders'
1320 must be a dictionary that maps share number (an integer ranging from
1321 0 to n-1) to an instance that provides IStorageBucketWriter. This
1322 must be performed before start() can be called."""
1325 """Begin the encode/upload process. This involves reading encrypted
1326 data from the IEncryptedUploadable, encoding it, uploading the shares
1327 to the shareholders, then sending the hash trees.
1329 set_encrypted_uploadable() and set_shareholders() must be called
1330 before this can be invoked.
1332 This returns a Deferred that fires with a verify cap when the upload
1333 process is complete. The verifycap, plus the encryption key, is
1334 sufficient to construct the read cap.
1337 class IDecoder(Interface):
1338 """I take a list of shareholders and some setup information, then
1339 download, validate, decode, and decrypt data from them, writing the
1340 results to an output file.
1342 I do not locate the shareholders, that is left to the IDownloader. I must
1343 be given a dict of RemoteReferences to storage buckets that are ready to
1348 """I take a file-like object (providing write and close) to which all
1349 the plaintext data will be written.
1351 TODO: producer/consumer . Maybe write() should return a Deferred that
1352 indicates when it will accept more data? But probably having the
1353 IDecoder be a producer is easier to glue to IConsumer pieces.
1356 def set_shareholders(shareholders):
1357 """I take a dictionary that maps share identifiers (small integers)
1358 to RemoteReferences that provide RIBucketReader. This must be called
1362 """I start the download. This process involves retrieving data and
1363 hash chains from the shareholders, using the hashes to validate the
1364 data, decoding the shares into segments, decrypting the segments,
1365 then writing the resulting plaintext to the output file.
1367 I return a Deferred that will fire (with self) when the download is
1371 class IDownloadTarget(Interface):
1372 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1373 # will register itself as a producer. This allows the target to invoke
1374 # downloader.pauseProducing, resumeProducing, and stopProducing.
1376 """Called before any calls to write() or close(). If an error
1377 occurs before any data is available, fail() may be called without
1378 a previous call to open().
1380 'size' is the length of the file being downloaded, in bytes."""
1383 """Output some data to the target."""
1385 """Inform the target that there is no more data to be written."""
1387 """fail() is called to indicate that the download has failed. 'why'
1388 is a Failure object indicating what went wrong. No further methods
1389 will be invoked on the IDownloadTarget after fail()."""
1390 def register_canceller(cb):
1391 """The CiphertextDownloader uses this to register a no-argument function
1392 that the target can call to cancel the download. Once this canceller
1393 is invoked, no further calls to write() or close() will be made."""
1395 """When the CiphertextDownloader is done, this finish() function will be
1396 called. Whatever it returns will be returned to the invoker of
1397 Downloader.download.
1399 # The following methods are just because that target might be a
1400 # repairer.DownUpConnector, and just because the current CHKUpload object
1401 # expects to find the storage index and encoding parameters in its
1403 def set_storageindex(storageindex):
1404 """ Set the storage index. """
1405 def set_encodingparams(encodingparams):
1406 """ Set the encoding parameters. """
1408 class IDownloader(Interface):
1409 def download(uri, target):
1410 """Perform a CHK download, sending the data to the given target.
1411 'target' must provide IDownloadTarget.
1413 Returns a Deferred that fires (with the results of target.finish)
1414 when the download is finished, or errbacks if something went wrong."""
1416 class IEncryptedUploadable(Interface):
1417 def set_upload_status(upload_status):
1418 """Provide an IUploadStatus object that should be filled with status
1419 information. The IEncryptedUploadable is responsible for setting
1420 key-determination progress ('chk'), size, storage_index, and
1421 ciphertext-fetch progress. It may delegate some of this
1422 responsibility to others, in particular to the IUploadable."""
1425 """This behaves just like IUploadable.get_size()."""
1427 def get_all_encoding_parameters():
1428 """Return a Deferred that fires with a tuple of
1429 (k,happy,n,segment_size). The segment_size will be used as-is, and
1430 must match the following constraints: it must be a multiple of k, and
1431 it shouldn't be unreasonably larger than the file size (if
1432 segment_size is larger than filesize, the difference must be stored
1435 This usually passes through to the IUploadable method of the same
1438 The encoder strictly obeys the values returned by this method. To
1439 make an upload use non-default encoding parameters, you must arrange
1440 to control the values that this method returns.
1443 def get_storage_index():
1444 """Return a Deferred that fires with a 16-byte storage index.
1447 def read_encrypted(length, hash_only):
1448 """This behaves just like IUploadable.read(), but returns crypttext
1449 instead of plaintext. If hash_only is True, then this discards the
1450 data (and returns an empty list); this improves efficiency when
1451 resuming an interrupted upload (where we need to compute the
1452 plaintext hashes, but don't need the redundant encrypted data)."""
1454 def get_plaintext_hashtree_leaves(first, last, num_segments):
1455 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1456 plaintext segments, i.e. get the tagged hashes of the given segments.
1457 The segment size is expected to be generated by the
1458 IEncryptedUploadable before any plaintext is read or ciphertext
1459 produced, so that the segment hashes can be generated with only a
1462 This returns a Deferred which fires with a sequence of hashes, using:
1464 tuple(segment_hashes[first:last])
1466 'num_segments' is used to assert that the number of segments that the
1467 IEncryptedUploadable handled matches the number of segments that the
1468 encoder was expecting.
1470 This method must not be called until the final byte has been read
1471 from read_encrypted(). Once this method is called, read_encrypted()
1472 can never be called again.
1475 def get_plaintext_hash():
1476 """OBSOLETE; Get the hash of the whole plaintext.
1478 This returns a Deferred which fires with a tagged SHA-256 hash of the
1479 whole plaintext, obtained from hashutil.plaintext_hash(data).
1483 """Just like IUploadable.close()."""
1485 class IUploadable(Interface):
1486 def set_upload_status(upload_status):
1487 """Provide an IUploadStatus object that should be filled with status
1488 information. The IUploadable is responsible for setting
1489 key-determination progress ('chk')."""
1491 def set_default_encoding_parameters(params):
1492 """Set the default encoding parameters, which must be a dict mapping
1493 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1494 'max_segment_size'. These might have an influence on the final
1495 encoding parameters returned by get_all_encoding_parameters(), if the
1496 Uploadable doesn't have more specific preferences.
1498 This call is optional: if it is not used, the Uploadable will use
1499 some built-in defaults. If used, this method must be called before
1500 any other IUploadable methods to have any effect.
1504 """Return a Deferred that will fire with the length of the data to be
1505 uploaded, in bytes. This will be called before the data is actually
1506 used, to compute encoding parameters.
1509 def get_all_encoding_parameters():
1510 """Return a Deferred that fires with a tuple of
1511 (k,happy,n,segment_size). The segment_size will be used as-is, and
1512 must match the following constraints: it must be a multiple of k, and
1513 it shouldn't be unreasonably larger than the file size (if
1514 segment_size is larger than filesize, the difference must be stored
1517 The relative values of k and n allow some IUploadables to request
1518 better redundancy than others (in exchange for consuming more space
1521 Larger values of segment_size reduce hash overhead, while smaller
1522 values reduce memory footprint and cause data to be delivered in
1523 smaller pieces (which may provide a smoother and more predictable
1524 download experience).
1526 The encoder strictly obeys the values returned by this method. To
1527 make an upload use non-default encoding parameters, you must arrange
1528 to control the values that this method returns. One way to influence
1529 them may be to call set_encoding_parameters() before calling
1530 get_all_encoding_parameters().
1533 def get_encryption_key():
1534 """Return a Deferred that fires with a 16-byte AES key. This key will
1535 be used to encrypt the data. The key will also be hashed to derive
1538 Uploadables which want to achieve convergence should hash their file
1539 contents and the serialized_encoding_parameters to form the key
1540 (which of course requires a full pass over the data). Uploadables can
1541 use the upload.ConvergentUploadMixin class to achieve this
1544 Uploadables which do not care about convergence (or do not wish to
1545 make multiple passes over the data) can simply return a
1546 strongly-random 16 byte string.
1548 get_encryption_key() may be called multiple times: the IUploadable is
1549 required to return the same value each time.
1553 """Return a Deferred that fires with a list of strings (perhaps with
1554 only a single element) which, when concatenated together, contain the
1555 next 'length' bytes of data. If EOF is near, this may provide fewer
1556 than 'length' bytes. The total number of bytes provided by read()
1557 before it signals EOF must equal the size provided by get_size().
1559 If the data must be acquired through multiple internal read
1560 operations, returning a list instead of a single string may help to
1561 reduce string copies.
1563 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1564 so a 10kB file means length=3kB, 100kB file means length=30kB,
1565 and >=1MB file means length=300kB.
1567 This method provides for a single full pass through the data. Later
1568 use cases may desire multiple passes or access to only parts of the
1569 data (such as a mutable file making small edits-in-place). This API
1570 will be expanded once those use cases are better understood.
1574 """The upload is finished, and whatever filehandle was in use may be
1577 class IUploadResults(Interface):
1578 """I am returned by upload() methods. I contain a number of public
1579 attributes which can be read to determine the results of the upload. Some
1580 of these are functional, some are timing information. All of these may be
1583 .file_size : the size of the file, in bytes
1584 .uri : the CHK read-cap for the file
1585 .ciphertext_fetched : how many bytes were fetched by the helper
1586 .sharemap: dict mapping share identifier to set of serverids
1587 (binary strings). This indicates which servers were given
1588 which shares. For immutable files, the shareid is an
1589 integer (the share number, from 0 to N-1). For mutable
1590 files, it is a string of the form 'seq%d-%s-sh%d',
1591 containing the sequence number, the roothash, and the
1593 .servermap : dict mapping server peerid to a set of share numbers
1594 .timings : dict of timing information, mapping name to seconds (float)
1595 total : total upload time, start to finish
1596 storage_index : time to compute the storage index
1597 peer_selection : time to decide which peers will be used
1598 contacting_helper : initial helper query to upload/no-upload decision
1599 existence_check : helper pre-upload existence check
1600 helper_total : initial helper query to helper finished pushing
1601 cumulative_fetch : helper waiting for ciphertext requests
1602 total_fetch : helper start to last ciphertext response
1603 cumulative_encoding : just time spent in zfec
1604 cumulative_sending : just time spent waiting for storage servers
1605 hashes_and_close : last segment push to shareholder close
1606 total_encode_and_push : first encode to shareholder close
1610 class IDownloadResults(Interface):
1611 """I am created internally by download() methods. I contain a number of
1612 public attributes which contain details about the download process.::
1614 .file_size : the size of the file, in bytes
1615 .servers_used : set of server peerids that were used during download
1616 .server_problems : dict mapping server peerid to a problem string. Only
1617 servers that had problems (bad hashes, disconnects)
1619 .servermap : dict mapping server peerid to a set of share numbers. Only
1620 servers that had any shares are listed here.
1621 .timings : dict of timing information, mapping name to seconds (float)
1622 peer_selection : time to ask servers about shares
1623 servers_peer_selection : dict of peerid to DYHB-query time
1624 uri_extension : time to fetch a copy of the URI extension block
1625 hashtrees : time to fetch the hash trees
1626 segments : time to fetch, decode, and deliver segments
1627 cumulative_fetch : time spent waiting for storage servers
1628 cumulative_decode : just time spent in zfec
1629 cumulative_decrypt : just time spent in decryption
1630 total : total download time, start to finish
1631 fetch_per_server : dict of peerid to list of per-segment fetch times
1635 class IUploader(Interface):
1636 def upload(uploadable):
1637 """Upload the file. 'uploadable' must impement IUploadable. This
1638 returns a Deferred which fires with an IUploadResults instance, from
1639 which the URI of the file can be obtained as results.uri ."""
1641 def upload_ssk(write_capability, new_version, uploadable):
1642 """TODO: how should this work?"""
1644 class ICheckable(Interface):
1645 def check(monitor, verify=False, add_lease=False):
1646 """Check upon my health, optionally repairing any problems.
1648 This returns a Deferred that fires with an instance that provides
1649 ICheckResults, or None if the object is non-distributed (i.e. LIT
1652 The monitor will be checked periodically to see if the operation has
1653 been cancelled. If so, no new queries will be sent, and the Deferred
1654 will fire (with a OperationCancelledError) immediately.
1656 Filenodes and dirnodes (which provide IFilesystemNode) are also
1657 checkable. Instances that represent verifier-caps will be checkable
1658 but not downloadable. Some objects (like LIT files) do not actually
1659 live in the grid, and their checkers return None (non-distributed
1660 files are always healthy).
1662 If verify=False, a relatively lightweight check will be performed: I
1663 will ask all servers if they have a share for me, and I will believe
1664 whatever they say. If there are at least N distinct shares on the
1665 grid, my results will indicate r.is_healthy()==True. This requires a
1666 roundtrip to each server, but does not transfer very much data, so
1667 the network bandwidth is fairly low.
1669 If verify=True, a more resource-intensive check will be performed:
1670 every share will be downloaded, and the hashes will be validated on
1671 every bit. I will ignore any shares that failed their hash checks. If
1672 there are at least N distinct valid shares on the grid, my results
1673 will indicate r.is_healthy()==True. This requires N/k times as much
1674 download bandwidth (and server disk IO) as a regular download. If a
1675 storage server is holding a corrupt share, or is experiencing memory
1676 failures during retrieval, or is malicious or buggy, then
1677 verification will detect the problem, but checking will not.
1679 If add_lease=True, I will ensure that an up-to-date lease is present
1680 on each share. The lease secrets will be derived from by node secret
1681 (in BASEDIR/private/secret), so either I will add a new lease to the
1682 share, or I will merely renew the lease that I already had. In a
1683 future version of the storage-server protocol (once Accounting has
1684 been implemented), there may be additional options here to define the
1685 kind of lease that is obtained (which account number to claim, etc).
1687 TODO: any problems seen during checking will be reported to the
1688 health-manager.furl, a centralized object which is responsible for
1689 figuring out why files are unhealthy so corrective action can be
1693 def check_and_repair(monitor, verify=False, add_lease=False):
1694 """Like check(), but if the file/directory is not healthy, attempt to
1697 Any non-healthy result will cause an immediate repair operation, to
1698 generate and upload new shares. After repair, the file will be as
1699 healthy as we can make it. Details about what sort of repair is done
1700 will be put in the check-and-repair results. The Deferred will not
1701 fire until the repair is complete.
1703 This returns a Deferred which fires with an instance of
1704 ICheckAndRepairResults."""
1706 class IDeepCheckable(Interface):
1707 def start_deep_check(verify=False, add_lease=False):
1708 """Check upon the health of me and everything I can reach.
1710 This is a recursive form of check(), useable only on dirnodes.
1712 I return a Monitor, with results that are an IDeepCheckResults
1715 TODO: If any of the directories I traverse are unrecoverable, the
1716 Monitor will report failure. If any of the files I check upon are
1717 unrecoverable, those problems will be reported in the
1718 IDeepCheckResults as usual, and the Monitor will not report a
1722 def start_deep_check_and_repair(verify=False, add_lease=False):
1723 """Check upon the health of me and everything I can reach. Repair
1724 anything that isn't healthy.
1726 This is a recursive form of check_and_repair(), useable only on
1729 I return a Monitor, with results that are an
1730 IDeepCheckAndRepairResults object.
1732 TODO: If any of the directories I traverse are unrecoverable, the
1733 Monitor will report failure. If any of the files I check upon are
1734 unrecoverable, those problems will be reported in the
1735 IDeepCheckResults as usual, and the Monitor will not report a
1739 class ICheckResults(Interface):
1740 """I contain the detailed results of a check/verify operation.
1743 def get_storage_index():
1744 """Return a string with the (binary) storage index."""
1745 def get_storage_index_string():
1746 """Return a string with the (printable) abbreviated storage index."""
1748 """Return the (string) URI of the object that was checked."""
1751 """Return a boolean, True if the file/dir is fully healthy, False if
1752 it is damaged in any way. Non-distributed LIT files always return
1755 def is_recoverable():
1756 """Return a boolean, True if the file/dir can be recovered, False if
1757 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1758 files always return True."""
1760 def needs_rebalancing():
1761 """Return a boolean, True if the file/dir's reliability could be
1762 improved by moving shares to new servers. Non-distributed LIT files
1763 always return False."""
1767 """Return a dictionary that describes the state of the file/dir. LIT
1768 files always return an empty dictionary. Normal files and directories
1769 return a dictionary with the following keys (note that these use
1770 binary strings rather than base32-encoded ones) (also note that for
1771 mutable files, these counts are for the 'best' version):
1773 count-shares-good: the number of distinct good shares that were found
1774 count-shares-needed: 'k', the number of shares required for recovery
1775 count-shares-expected: 'N', the number of total shares generated
1776 count-good-share-hosts: the number of distinct storage servers with
1777 good shares. If this number is less than
1778 count-shares-good, then some shares are
1779 doubled up, increasing the correlation of
1780 failures. This indicates that one or more
1781 shares should be moved to an otherwise unused
1782 server, if one is available.
1783 count-corrupt-shares: the number of shares with integrity failures
1784 list-corrupt-shares: a list of 'share locators', one for each share
1785 that was found to be corrupt. Each share
1786 locator is a list of (serverid, storage_index,
1788 count-incompatible-shares: the number of shares which are of a share
1789 format unknown to this checker
1790 list-incompatible-shares: a list of 'share locators', one for each
1791 share that was found to be of an unknown
1792 format. Each share locator is a list of
1793 (serverid, storage_index, sharenum).
1794 servers-responding: list of (binary) storage server identifiers,
1795 one for each server which responded to the share
1796 query (even if they said they didn't have
1797 shares, and even if they said they did have
1798 shares but then didn't send them when asked, or
1799 dropped the connection, or returned a Failure,
1800 and even if they said they did have shares and
1801 sent incorrect ones when asked)
1802 sharemap: dict mapping share identifier to list of serverids
1803 (binary strings). This indicates which servers are holding
1804 which shares. For immutable files, the shareid is an
1805 integer (the share number, from 0 to N-1). For mutable
1806 files, it is a string of the form 'seq%d-%s-sh%d',
1807 containing the sequence number, the roothash, and the
1810 The following keys are most relevant for mutable files, but immutable
1811 files will provide sensible values too::
1813 count-wrong-shares: the number of shares for versions other than the
1814 'best' one (which is defined as being the
1815 recoverable version with the highest sequence
1816 number, then the highest roothash). These are
1817 either leftover shares from an older version
1818 (perhaps on a server that was offline when an
1819 update occurred), shares from an unrecoverable
1820 newer version, or shares from an alternate
1821 current version that results from an
1822 uncoordinated write collision. For a healthy
1823 file, this will equal 0.
1825 count-recoverable-versions: the number of recoverable versions of
1826 the file. For a healthy file, this will
1829 count-unrecoverable-versions: the number of unrecoverable versions
1830 of the file. For a healthy file, this
1836 """Return a string with a brief (one-line) summary of the results."""
1839 """Return a list of strings with more detailed results."""
1841 class ICheckAndRepairResults(Interface):
1842 """I contain the detailed results of a check/verify/repair operation.
1844 The IFilesystemNode.check()/verify()/repair() methods all return
1845 instances that provide ICheckAndRepairResults.
1848 def get_storage_index():
1849 """Return a string with the (binary) storage index."""
1850 def get_storage_index_string():
1851 """Return a string with the (printable) abbreviated storage index."""
1852 def get_repair_attempted():
1853 """Return a boolean, True if a repair was attempted. We might not
1854 attempt to repair the file because it was healthy, or healthy enough
1855 (i.e. some shares were missing but not enough to exceed some
1856 threshold), or because we don't know how to repair this object."""
1857 def get_repair_successful():
1858 """Return a boolean, True if repair was attempted and the file/dir
1859 was fully healthy afterwards. False if no repair was attempted or if
1860 a repair attempt failed."""
1861 def get_pre_repair_results():
1862 """Return an ICheckResults instance that describes the state of the
1863 file/dir before any repair was attempted."""
1864 def get_post_repair_results():
1865 """Return an ICheckResults instance that describes the state of the
1866 file/dir after any repair was attempted. If no repair was attempted,
1867 the pre-repair and post-repair results will be identical."""
1870 class IDeepCheckResults(Interface):
1871 """I contain the results of a deep-check operation.
1873 This is returned by a call to ICheckable.deep_check().
1876 def get_root_storage_index_string():
1877 """Return the storage index (abbreviated human-readable string) of
1878 the first object checked."""
1880 """Return a dictionary with the following keys::
1882 count-objects-checked: count of how many objects were checked
1883 count-objects-healthy: how many of those objects were completely
1885 count-objects-unhealthy: how many were damaged in some way
1886 count-objects-unrecoverable: how many were unrecoverable
1887 count-corrupt-shares: how many shares were found to have
1888 corruption, summed over all objects
1892 def get_corrupt_shares():
1893 """Return a set of (serverid, storage_index, sharenum) for all shares
1894 that were found to be corrupt. Both serverid and storage_index are
1897 def get_all_results():
1898 """Return a dictionary mapping pathname (a tuple of strings, ready to
1899 be slash-joined) to an ICheckResults instance, one for each object
1900 that was checked."""
1902 def get_results_for_storage_index(storage_index):
1903 """Retrive the ICheckResults instance for the given (binary)
1904 storage index. Raises KeyError if there are no results for that
1908 """Return a dictionary with the same keys as
1909 IDirectoryNode.deep_stats()."""
1911 class IDeepCheckAndRepairResults(Interface):
1912 """I contain the results of a deep-check-and-repair operation.
1914 This is returned by a call to ICheckable.deep_check_and_repair().
1917 def get_root_storage_index_string():
1918 """Return the storage index (abbreviated human-readable string) of
1919 the first object checked."""
1921 """Return a dictionary with the following keys::
1923 count-objects-checked: count of how many objects were checked
1924 count-objects-healthy-pre-repair: how many of those objects were
1925 completely healthy (before any
1927 count-objects-unhealthy-pre-repair: how many were damaged in
1929 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1930 count-objects-healthy-post-repair: how many of those objects were
1931 completely healthy (after any
1933 count-objects-unhealthy-post-repair: how many were damaged in
1935 count-objects-unrecoverable-post-repair: how many were
1937 count-repairs-attempted: repairs were attempted on this many
1938 objects. The count-repairs- keys will
1939 always be provided, however unless
1940 repair=true is present, they will all
1942 count-repairs-successful: how many repairs resulted in healthy
1944 count-repairs-unsuccessful: how many repairs resulted did not
1945 results in completely healthy objects
1946 count-corrupt-shares-pre-repair: how many shares were found to
1947 have corruption, summed over all
1948 objects examined (before any
1950 count-corrupt-shares-post-repair: how many shares were found to
1951 have corruption, summed over all
1952 objects examined (after any
1957 """Return a dictionary with the same keys as
1958 IDirectoryNode.deep_stats()."""
1960 def get_corrupt_shares():
1961 """Return a set of (serverid, storage_index, sharenum) for all shares
1962 that were found to be corrupt before any repair was attempted. Both
1963 serverid and storage_index are binary.
1965 def get_remaining_corrupt_shares():
1966 """Return a set of (serverid, storage_index, sharenum) for all shares
1967 that were found to be corrupt after any repair was completed. Both
1968 serverid and storage_index are binary. These are shares that need
1969 manual inspection and probably deletion.
1971 def get_all_results():
1972 """Return a dictionary mapping pathname (a tuple of strings, ready to
1973 be slash-joined) to an ICheckAndRepairResults instance, one for each
1974 object that was checked."""
1976 def get_results_for_storage_index(storage_index):
1977 """Retrive the ICheckAndRepairResults instance for the given (binary)
1978 storage index. Raises KeyError if there are no results for that
1982 class IRepairable(Interface):
1983 def repair(check_results):
1984 """Attempt to repair the given object. Returns a Deferred that fires
1985 with a IRepairResults object.
1987 I must be called with an object that implements ICheckResults, as
1988 proof that you have actually discovered a problem with this file. I
1989 will use the data in the checker results to guide the repair process,
1990 such as which servers provided bad data and should therefore be
1991 avoided. The ICheckResults object is inside the
1992 ICheckAndRepairResults object, which is returned by the
1993 ICheckable.check() method::
1995 d = filenode.check(repair=False)
1996 def _got_results(check_and_repair_results):
1997 check_results = check_and_repair_results.get_pre_repair_results()
1998 return filenode.repair(check_results)
1999 d.addCallback(_got_results)
2003 class IRepairResults(Interface):
2004 """I contain the results of a repair operation."""
2007 class IClient(Interface):
2008 def upload(uploadable):
2009 """Upload some data into a CHK, get back the UploadResults for it.
2010 @param uploadable: something that implements IUploadable
2011 @return: a Deferred that fires with the UploadResults instance.
2012 To get the URI for this file, use results.uri .
2015 def create_mutable_file(contents=""):
2016 """Create a new mutable file (with initial) contents, get back the
2019 @param contents: (bytestring, callable, or None): this provides the
2020 initial contents of the mutable file. If 'contents' is a bytestring,
2021 it will be used as-is. If 'contents' is a callable, it will be
2022 invoked with the new MutableFileNode instance and is expected to
2023 return a bytestring with the initial contents of the file (the
2024 callable can use node.get_writekey() to decide how to encrypt the
2025 initial contents, e.g. for a brand new dirnode with initial
2026 children). contents=None is equivalent to an empty string. Using
2027 content_maker= is more efficient than creating a mutable file and
2028 setting its contents in two separate operations.
2030 @return: a Deferred that fires with tne (string) SSK URI for the new
2034 def create_dirnode(initial_children={}):
2035 """Create a new unattached dirnode, possibly with initial children.
2037 @param initial_children: dict with keys that are unicode child names,
2038 and values that are (childnode, metadata) tuples.
2040 @return: a Deferred that fires with the new IDirectoryNode instance.
2043 def create_node_from_uri(uri, rouri):
2044 """Create a new IFilesystemNode instance from the uri, synchronously.
2045 @param uri: a string or IURI-providing instance, or None. This could
2046 be for a LiteralFileNode, a CHK file node, a mutable file
2047 node, or a directory node
2048 @param rouri: a string or IURI-providing instance, or None. If the
2049 main uri is None, I will use the rouri instead. If I
2050 recognize the format of the main uri, I will ignore the
2051 rouri (because it can be derived from the writecap).
2053 @return: an instance that provides IFilesystemNode (or more usefully
2054 one of its subclasses). File-specifying URIs will result in
2055 IFileNode-providing instances, like ImmutableFileNode,
2056 LiteralFileNode, or MutableFileNode. Directory-specifying
2057 URIs will result in IDirectoryNode-providing instances, like
2061 class INodeMaker(Interface):
2062 """The NodeMaker is used to create IFilesystemNode instances. It can
2063 accept a filecap/dircap string and return the node right away. It can
2064 also create new nodes (i.e. upload a file, or create a mutable file)
2065 asynchronously. Once you have one of these nodes, you can use other
2066 methods to determine whether it is a file or directory, and to download
2067 or modify its contents.
2069 The NodeMaker encapsulates all the authorities that these
2070 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2071 Tahoe process will typically have a single NodeMaker, but unit tests may
2072 create simplified/mocked forms for testing purposes.
2074 def create_from_cap(writecap, readcap=None):
2075 """I create an IFilesystemNode from the given writecap/readcap. I can
2076 only provide nodes for existing file/directory objects: use my other
2077 methods to create new objects. I return synchronously."""
2079 def create_mutable_file(contents=None, keysize=None):
2080 """I create a new mutable file, and return a Deferred which will fire
2081 with the IMutableFileNode instance when it is ready. If contents= is
2082 provided (a bytestring), it will be used as the initial contents of
2083 the new file, otherwise the file will contain zero bytes. keysize= is
2084 for use by unit tests, to create mutable files that are smaller than
2087 def create_new_mutable_directory(initial_children={}):
2088 """I create a new mutable directory, and return a Deferred which will
2089 fire with the IDirectoryNode instance when it is ready. If
2090 initial_children= is provided (a dict mapping unicode child name to
2091 (childnode, metadata_dict) tuples), the directory will be populated
2092 with those children, otherwise it will be empty."""
2094 class IClientStatus(Interface):
2095 def list_all_uploads():
2096 """Return a list of uploader objects, one for each upload which
2097 currently has an object available (tracked with weakrefs). This is
2098 intended for debugging purposes."""
2099 def list_active_uploads():
2100 """Return a list of active IUploadStatus objects."""
2101 def list_recent_uploads():
2102 """Return a list of IUploadStatus objects for the most recently
2105 def list_all_downloads():
2106 """Return a list of downloader objects, one for each download which
2107 currently has an object available (tracked with weakrefs). This is
2108 intended for debugging purposes."""
2109 def list_active_downloads():
2110 """Return a list of active IDownloadStatus objects."""
2111 def list_recent_downloads():
2112 """Return a list of IDownloadStatus objects for the most recently
2113 started downloads."""
2115 class IUploadStatus(Interface):
2117 """Return a timestamp (float with seconds since epoch) indicating
2118 when the operation was started."""
2119 def get_storage_index():
2120 """Return a string with the (binary) storage index in use on this
2121 upload. Returns None if the storage index has not yet been
2124 """Return an integer with the number of bytes that will eventually
2125 be uploaded for this file. Returns None if the size is not yet known.
2128 """Return True if this upload is using a Helper, False if not."""
2130 """Return a string describing the current state of the upload
2133 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2134 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2135 made towards hashing the file to determine a CHK encryption key: if
2136 non-convergent encryption is in use, this will be trivial, otherwise
2137 the whole file must be hashed. 'ciphertext' describes how much of the
2138 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2139 uploads. 'encode_and_push' describes how much of the encode-and-push
2140 process has finished: for helper uploads this is dependent upon the
2141 helper providing progress reports. It might be reasonable to add all
2142 three numbers and report the sum to the user."""
2144 """Return True if the upload is currently active, False if not."""
2146 """Return an instance of UploadResults (which contains timing and
2147 sharemap information). Might return None if the upload is not yet
2150 """Each upload status gets a unique number: this method returns that
2151 number. This provides a handle to this particular upload, so a web
2152 page can generate a suitable hyperlink."""
2154 class IDownloadStatus(Interface):
2156 """Return a timestamp (float with seconds since epoch) indicating
2157 when the operation was started."""
2158 def get_storage_index():
2159 """Return a string with the (binary) storage index in use on this
2160 download. This may be None if there is no storage index (i.e. LIT
2163 """Return an integer with the number of bytes that will eventually be
2164 retrieved for this file. Returns None if the size is not yet known.
2167 """Return True if this download is using a Helper, False if not."""
2169 """Return a string describing the current state of the download
2172 """Returns a float (from 0.0 to 1.0) describing the amount of the
2173 download that has completed. This value will remain at 0.0 until the
2174 first byte of plaintext is pushed to the download target."""
2176 """Return True if the download is currently active, False if not."""
2178 """Each download status gets a unique number: this method returns
2179 that number. This provides a handle to this particular download, so a
2180 web page can generate a suitable hyperlink."""
2182 class IServermapUpdaterStatus(Interface):
2184 class IPublishStatus(Interface):
2186 class IRetrieveStatus(Interface):
2189 class NotCapableError(Exception):
2190 """You have tried to write to a read-only node."""
2192 class BadWriteEnablerError(Exception):
2195 class RIControlClient(RemoteInterface):
2197 def wait_for_client_connections(num_clients=int):
2198 """Do not return until we have connections to at least NUM_CLIENTS
2202 def upload_from_file_to_uri(filename=str,
2203 convergence=ChoiceOf(None,
2204 StringConstraint(2**20))):
2205 """Upload a file to the grid. This accepts a filename (which must be
2206 absolute) that points to a file on the node's local disk. The node will
2207 read the contents of this file, upload it to the grid, then return the
2208 URI at which it was uploaded. If convergence is None then a random
2209 encryption key will be used, else the plaintext will be hashed, then
2210 that hash will be mixed together with the "convergence" string to form
2215 def download_from_uri_to_file(uri=URI, filename=str):
2216 """Download a file from the grid, placing it on the node's local disk
2217 at the given filename (which must be absolute[?]). Returns the
2218 absolute filename where the file was written."""
2223 def get_memory_usage():
2224 """Return a dict describes the amount of memory currently in use. The
2225 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2226 measuring memory consupmtion in bytes."""
2227 return DictOf(str, int)
2229 def speed_test(count=int, size=int, mutable=Any()):
2230 """Write 'count' tempfiles to disk, all of the given size. Measure
2231 how long (in seconds) it takes to upload them all to the servers.
2232 Then measure how long it takes to download all of them. If 'mutable'
2233 is 'create', time creation of mutable files. If 'mutable' is
2234 'upload', then time access to the same mutable file instead of
2237 Returns a tuple of (upload_time, download_time).
2239 return (float, float)
2241 def measure_peer_response_time():
2242 """Send a short message to each connected peer, and measure the time
2243 it takes for them to respond to it. This is a rough measure of the
2244 application-level round trip time.
2246 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2249 return DictOf(Nodeid, float)
2251 UploadResults = Any() #DictOf(str, str)
2253 class RIEncryptedUploadable(RemoteInterface):
2254 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2259 def get_all_encoding_parameters():
2260 return (int, int, int, long)
2262 def read_encrypted(offset=Offset, length=ReadSize):
2269 class RICHKUploadHelper(RemoteInterface):
2270 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2274 Return a dictionary of version information.
2276 return DictOf(str, Any())
2278 def upload(reader=RIEncryptedUploadable):
2279 return UploadResults
2282 class RIHelper(RemoteInterface):
2283 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2287 Return a dictionary of version information.
2289 return DictOf(str, Any())
2291 def upload_chk(si=StorageIndex):
2292 """See if a file with a given storage index needs uploading. The
2293 helper will ask the appropriate storage servers to see if the file
2294 has already been uploaded. If so, the helper will return a set of
2295 'upload results' that includes whatever hashes are needed to build
2296 the read-cap, and perhaps a truncated sharemap.
2298 If the file has not yet been uploaded (or if it was only partially
2299 uploaded), the helper will return an empty upload-results dictionary
2300 and also an RICHKUploadHelper object that will take care of the
2301 upload process. The client should call upload() on this object and
2302 pass it a reference to an RIEncryptedUploadable object that will
2303 provide ciphertext. When the upload is finished, the upload() method
2304 will finish and return the upload results.
2306 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2309 class RIStatsProvider(RemoteInterface):
2310 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2312 Provides access to statistics and monitoring information.
2317 returns a dictionary containing 'counters' and 'stats', each a
2318 dictionary with string counter/stat name keys, and numeric values.
2319 counters are monotonically increasing measures of work done, and
2320 stats are instantaneous measures (potentially time averaged
2323 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2325 class RIStatsGatherer(RemoteInterface):
2326 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2328 Provides a monitoring service for centralised collection of stats
2331 def provide(provider=RIStatsProvider, nickname=str):
2333 @param provider: a stats collector instance which should be polled
2334 periodically by the gatherer to collect stats.
2335 @param nickname: a name useful to identify the provided client
2340 class IStatsProducer(Interface):
2343 returns a dictionary, with str keys representing the names of stats
2344 to be monitored, and numeric values.
2347 class RIKeyGenerator(RemoteInterface):
2348 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2350 Provides a service offering to make RSA key pairs.
2353 def get_rsa_key_pair(key_size=int):
2355 @param key_size: the size of the signature key.
2356 @return: tuple(verifying_key, signing_key)
2358 return TupleOf(str, str)
2361 class FileTooLargeError(Exception):
2364 class IValidatedThingProxy(Interface):
2366 """ Acquire a thing and validate it. Return a deferred which is
2367 eventually fired with self if the thing is valid or errbacked if it
2368 can't be acquired or validated."""
2370 class InsufficientVersionError(Exception):
2371 def __init__(self, needed, got):
2372 self.needed = needed
2375 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2378 class EmptyPathnameComponentError(Exception):
2379 """The webapi disallows empty pathname components."""