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 IFilesystemNode(Interface):
486 Return the URI string that can be used by others to get access to
487 this node. If this node is read-only, the URI will only offer
488 read-only access. If this node is read-write, the URI will offer
491 If you have read-write access to a node and wish to share merely
492 read-only access with others, use get_readonly_uri().
495 def get_readonly_uri():
496 """Return the URI string that can be used by others to get read-only
497 access to this node. The result is a read-only URI, regardless of
498 whether this node is read-only or read-write.
500 If you have merely read-only access to this node, get_readonly_uri()
501 will return the same thing as get_uri().
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.
520 def get_storage_index():
521 """Return a string with the (binary) storage index in use on this
522 download. This may be None if there is no storage index (i.e. LIT
526 """Return True if this reference provides mutable access to the given
527 file or directory (i.e. if you can modify it), or False if not. Note
528 that even if this reference is read-only, someone else may hold a
529 read-write reference to it."""
532 """Return True if this file or directory is mutable (by *somebody*,
533 not necessarily you), False if it is is immutable. Note that a file
534 might be mutable overall, but your reference to it might be
535 read-only. On the other hand, all references to an immutable file
536 will be read-only; there are no read-write references to an immutable
540 class IMutableFilesystemNode(IFilesystemNode):
543 class IFileNode(IFilesystemNode):
544 def download(target):
545 """Download the file's contents to a given IDownloadTarget"""
547 def download_to_data():
548 """Download the file's contents. Return a Deferred that fires
549 with those contents."""
552 """Return the length (in bytes) of the data this node represents."""
554 def read(consumer, offset=0, size=None):
555 """Download a portion (possibly all) of the file's contents, making
556 them available to the given IConsumer. Return a Deferred that fires
557 (with the consumer) when the consumer is unregistered (either because
558 the last byte has been given to it, or because the consumer threw an
559 exception during write(), possibly because it no longer wants to
560 receive data). The portion downloaded will start at 'offset' and
561 contain 'size' bytes (or the remainder of the file if size==None).
563 The consumer will be used in non-streaming mode: an IPullProducer
564 will be attached to it.
566 The consumer will not receive data right away: several network trips
567 must occur first. The order of events will be::
569 consumer.registerProducer(p, streaming)
570 (if streaming == False)::
571 consumer does p.resumeProducing()
573 consumer does p.resumeProducing()
574 consumer.write(data).. (repeat until all data is written)
575 consumer.unregisterProducer()
576 deferred.callback(consumer)
578 If a download error occurs, or an exception is raised by
579 consumer.registerProducer() or consumer.write(), I will call
580 consumer.unregisterProducer() and then deliver the exception via
581 deferred.errback(). To cancel the download, the consumer should call
582 p.stopProducing(), which will result in an exception being delivered
583 via deferred.errback().
585 A simple download-to-memory consumer example would look like this::
587 class MemoryConsumer:
588 implements(IConsumer)
592 def registerProducer(self, p, streaming):
593 assert streaming == False
596 def write(self, data):
597 self.chunks.append(data)
598 def unregisterProducer(self):
600 d = filenode.read(MemoryConsumer())
601 d.addCallback(lambda mc: "".join(mc.chunks))
606 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
607 """I provide access to a 'mutable file', which retains its identity
608 regardless of what contents are put in it.
610 The consistency-vs-availability problem means that there might be
611 multiple versions of a file present in the grid, some of which might be
612 unrecoverable (i.e. have fewer than 'k' shares). These versions are
613 loosely ordered: each has a sequence number and a hash, and any version
614 with seqnum=N was uploaded by a node which has seen at least one version
617 The 'servermap' (an instance of IMutableFileServerMap) is used to
618 describe the versions that are known to be present in the grid, and which
619 servers are hosting their shares. It is used to represent the 'state of
620 the world', and is used for this purpose by my test-and-set operations.
621 Downloading the contents of the mutable file will also return a
622 servermap. Uploading a new version into the mutable file requires a
623 servermap as input, and the semantics of the replace operation is
624 'replace the file with my new version if it looks like nobody else has
625 changed the file since my previous download'. Because the file is
626 distributed, this is not a perfect test-and-set operation, but it will do
627 its best. If the replace process sees evidence of a simultaneous write,
628 it will signal an UncoordinatedWriteError, so that the caller can take
632 Most readers will want to use the 'best' current version of the file, and
633 should use my 'download_best_version()' method.
635 To unconditionally replace the file, callers should use overwrite(). This
636 is the mode that user-visible mutable files will probably use.
638 To apply some delta to the file, call modify() with a callable modifier
639 function that can apply the modification that you want to make. This is
640 the mode that dirnodes will use, since most directory modification
641 operations can be expressed in terms of deltas to the directory state.
644 Three methods are available for users who need to perform more complex
645 operations. The first is get_servermap(), which returns an up-to-date
646 servermap using a specified mode. The second is download_version(), which
647 downloads a specific version (not necessarily the 'best' one). The third
648 is 'upload', which accepts new contents and a servermap (which must have
649 been updated with MODE_WRITE). The upload method will attempt to apply
650 the new contents as long as no other node has modified the file since the
651 servermap was updated. This might be useful to a caller who wants to
652 merge multiple versions into a single new one.
654 Note that each time the servermap is updated, a specific 'mode' is used,
655 which determines how many peers are queried. To use a servermap for my
656 replace() method, that servermap must have been updated in MODE_WRITE.
657 These modes are defined in allmydata.mutable.common, and consist of
658 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
659 allmydata/mutable/servermap.py for details about the differences.
661 Mutable files are currently limited in size (about 3.5MB max) and can
662 only be retrieved and updated all-at-once, as a single big string. Future
663 versions of our mutable files will remove this restriction.
666 def download_best_version():
667 """Download the 'best' available version of the file, meaning one of
668 the recoverable versions with the highest sequence number. If no
669 uncoordinated writes have occurred, and if enough shares are
670 available, then this will be the most recent version that has been
673 I update an internal servermap with MODE_READ, determine which
674 version of the file is indicated by
675 servermap.best_recoverable_version(), and return a Deferred that
676 fires with its contents. If no version is recoverable, the Deferred
677 will errback with UnrecoverableFileError.
680 def get_size_of_best_version():
681 """Find the size of the version that would be downloaded with
682 download_best_version(), without actually downloading the whole file.
684 I return a Deferred that fires with an integer.
687 def overwrite(new_contents):
688 """Unconditionally replace the contents of the mutable file with new
689 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
690 is only appropriate to use when the new contents of the file are
691 completely unrelated to the old ones, and you do not care about other
694 I return a Deferred that fires (with a PublishStatus object) when the
695 update has completed.
698 def modify(modifier_cb):
699 """Modify the contents of the file, by downloading the current
700 version, applying the modifier function (or bound method), then
701 uploading the new version. I return a Deferred that fires (with a
702 PublishStatus object) when the update is complete.
704 The modifier callable will be given three arguments: a string (with
705 the old contents), a 'first_time' boolean, and a servermap. As with
706 download_best_version(), the old contents will be from the best
707 recoverable version, but the modifier can use the servermap to make
708 other decisions (such as refusing to apply the delta if there are
709 multiple parallel versions, or if there is evidence of a newer
710 unrecoverable version). 'first_time' will be True the first time the
711 modifier is called, and False on any subsequent calls.
713 The callable should return a string with the new contents. The
714 callable must be prepared to be called multiple times, and must
715 examine the input string to see if the change that it wants to make
716 is already present in the old version. If it does not need to make
717 any changes, it can either return None, or return its input string.
719 If the modifier raises an exception, it will be returned in the
724 def get_servermap(mode):
725 """Return a Deferred that fires with an IMutableFileServerMap
726 instance, updated using the given mode.
729 def download_version(servermap, version):
730 """Download a specific version of the file, using the servermap
731 as a guide to where the shares are located.
733 I return a Deferred that fires with the requested contents, or
734 errbacks with UnrecoverableFileError. Note that a servermap which was
735 updated with MODE_ANYTHING or MODE_READ may not know about shares for
736 all versions (those modes stop querying servers as soon as they can
737 fulfil their goals), so you may want to use MODE_CHECK (which checks
738 everything) to get increased visibility.
741 def upload(new_contents, servermap):
742 """Replace the contents of the file with new ones. This requires a
743 servermap that was previously updated with MODE_WRITE.
745 I attempt to provide test-and-set semantics, in that I will avoid
746 modifying any share that is different than the version I saw in the
747 servermap. However, if another node is writing to the file at the
748 same time as me, I may manage to update some shares while they update
749 others. If I see any evidence of this, I will signal
750 UncoordinatedWriteError, and the file will be left in an inconsistent
751 state (possibly the version you provided, possibly the old version,
752 possibly somebody else's version, and possibly a mix of shares from
755 The recommended response to UncoordinatedWriteError is to either
756 return it to the caller (since they failed to coordinate their
757 writes), or to attempt some sort of recovery. It may be sufficient to
758 wait a random interval (with exponential backoff) and repeat your
759 operation. If I do not signal UncoordinatedWriteError, then I was
760 able to write the new version without incident.
762 I return a Deferred that fires (with a PublishStatus object) when the
763 publish has completed. I will update the servermap in-place with the
764 location of all new shares.
768 """Return this filenode's writekey, or None if the node does not have
769 write-capability. This may be used to assist with data structures
770 that need to make certain data available only to writers, such as the
771 read-write child caps in dirnodes. The recommended process is to have
772 reader-visible data be submitted to the filenode in the clear (where
773 it will be encrypted by the filenode using the readkey), but encrypt
774 writer-visible data using this writekey.
777 class NotEnoughSharesError(Exception):
778 """Download was unable to get enough shares, or upload was unable to
779 place 'shares_of_happiness' shares."""
781 class NoSharesError(Exception):
782 """Upload or Download was unable to get any shares at all."""
784 class UnableToFetchCriticalDownloadDataError(Exception):
785 """I was unable to fetch some piece of critical data which is supposed to
786 be identically present in all shares."""
788 class NoServersError(Exception):
789 """Upload wasn't given any servers to work with, usually indicating a
790 network or Introducer problem."""
792 class ExistingChildError(Exception):
793 """A directory node was asked to add or replace a child that already
794 exists, and overwrite= was set to False."""
796 class NoSuchChildError(Exception):
797 """A directory node was asked to fetch a child which does not exist."""
799 class IDirectoryNode(IMutableFilesystemNode):
800 """I represent a name-to-child mapping, holding the tahoe equivalent of a
801 directory. All child names are unicode strings, and all children are some
802 sort of IFilesystemNode (either files or subdirectories).
807 The dirnode ('1') URI returned by this method can be used in
808 set_uri() on a different directory ('2') to 'mount' a reference to
809 this directory ('1') under the other ('2'). This URI is just a
810 string, so it can be passed around through email or other out-of-band
814 def get_readonly_uri():
816 The dirnode ('1') URI returned by this method can be used in
817 set_uri() on a different directory ('2') to 'mount' a reference to
818 this directory ('1') under the other ('2'). This URI is just a
819 string, so it can be passed around through email or other out-of-band
824 """I return a Deferred that fires with a dictionary mapping child
825 name (a unicode string) to (node, metadata_dict) tuples, in which
826 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
827 is a dictionary of metadata."""
830 """I return a Deferred that fires with a boolean, True if there
831 exists a child of the given name, False if not. The child name must
832 be a unicode string."""
835 """I return a Deferred that fires with a specific named child node,
836 either an IFileNode or an IDirectoryNode. The child name must be a
837 unicode string. I raise NoSuchChildError if I do not have a child by
840 def get_metadata_for(name):
841 """I return a Deferred that fires with the metadata dictionary for a
842 specific named child node. This metadata is stored in the *edge*, not
843 in the child, so it is attached to the parent dirnode rather than the
844 child dir-or-file-node. The child name must be a unicode string. I
845 raise NoSuchChildError if I do not have a child by that name."""
847 def set_metadata_for(name, metadata):
848 """I replace any existing metadata for the named child with the new
849 metadata. The child name must be a unicode string. This metadata is
850 stored in the *edge*, not in the child, so it is attached to the
851 parent dirnode rather than the child dir-or-file-node. I return a
852 Deferred (that fires with this dirnode) when the operation is
853 complete. I raise NoSuchChildError if I do not have a child by that
856 def get_child_at_path(path):
857 """Transform a child path into an IDirectoryNode or IFileNode.
859 I perform a recursive series of 'get' operations to find the named
860 descendant node. I return a Deferred that fires with the node, or
861 errbacks with NoSuchChildError if the node could not be found.
863 The path can be either a single string (slash-separated) or a list of
864 path-name elements. All elements must be unicode strings.
867 def get_child_and_metadata_at_path(path):
868 """Transform a child path into an IDirectoryNode/IFileNode and
871 I am like get_child_at_path(), but my Deferred fires with a tuple of
872 (node, metadata). The metadata comes from the last edge. If the path
873 is empty, the metadata will be an empty dictionary.
876 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
877 """I add a child (by writecap+readcap) at the specific name. I return
878 a Deferred that fires when the operation finishes. If overwrite= is
879 True, I will replace any existing child of the same name, otherwise
880 an existing child will cause me to return ExistingChildError. The
881 child name must be a unicode string.
883 The child caps could be for a file, or for a directory. If the new
884 child is read/write, you will provide both writecap and readcap. If
885 the child is read-only, you will provide the readcap write (i.e. the
886 writecap= and readcap= arguments will both be the child's readcap).
887 The filecaps are typically obtained from an IFilesystemNode with
888 get_uri() and get_readonly_uri().
890 If metadata= is provided, I will use it as the metadata for the named
891 edge. This will replace any existing metadata. If metadata= is left
892 as the default value of None, I will set ['mtime'] to the current
893 time, and I will set ['ctime'] to the current time if there was not
894 already a child by this name present. This roughly matches the
895 ctime/mtime semantics of traditional filesystems.
897 If this directory node is read-only, the Deferred will errback with a
900 def set_children(entries, overwrite=True):
901 """Add multiple children (by writecap+readcap) to a directory node.
902 Takes a dictionary, with childname as keys and (writecap, readcap)
903 tuples (or (writecap, readcap, metadata) triples) as values. Returns
904 a Deferred that fires (with this dirnode) when the operation
905 finishes. This is equivalent to calling set_uri() multiple times, but
906 is much more efficient. All child names must be unicode strings.
909 def set_node(name, child, metadata=None, overwrite=True):
910 """I add a child at the specific name. I return a Deferred that fires
911 when the operation finishes. This Deferred will fire with the child
912 node that was just added. I will replace any existing child of the
913 same name. The child name must be a unicode string. The 'child'
914 instance must be an instance providing IDirectoryNode or IFileNode.
916 If metadata= is provided, I will use it as the metadata for the named
917 edge. This will replace any existing metadata. If metadata= is left
918 as the default value of None, I will set ['mtime'] to the current
919 time, and I will set ['ctime'] to the current time if there was not
920 already a child by this name present. This roughly matches the
921 ctime/mtime semantics of traditional filesystems.
923 If this directory node is read-only, the Deferred will errback with a
926 def set_nodes(entries, overwrite=True):
927 """Add multiple children to a directory node. Takes a dict mapping
928 unicode childname to (child_node, metdata) tuples. If metdata=None,
929 the original metadata is left unmodified. Returns a Deferred that
930 fires (with this dirnode) when the operation finishes. This is
931 equivalent to calling set_node() multiple times, but is much more
934 def add_file(name, uploadable, metadata=None, overwrite=True):
935 """I upload a file (using the given IUploadable), then attach the
936 resulting FileNode to the directory at the given name. I set metadata
937 the same way as set_uri and set_node. The child name must be a
940 I return a Deferred that fires (with the IFileNode of the uploaded
941 file) when the operation completes."""
944 """I remove the child at the specific name. I return a Deferred that
945 fires when the operation finishes. The child name must be a unicode
946 string. I raise NoSuchChildError if I do not have a child by that
949 def create_subdirectory(name, initial_children={}, overwrite=True):
950 """I create and attach a directory at the given name. The new
951 directory can be empty, or it can be populated with children
952 according to 'initial_children', which takes a dictionary in the same
953 format as set_nodes (i.e. mapping unicode child name to (childnode,
954 metadata) tuples). The child name must be a unicode string. I return
955 a Deferred that fires (with the new directory node) when the
956 operation finishes."""
958 def move_child_to(current_child_name, new_parent, new_child_name=None,
960 """I take one of my children and move them to a new parent. The child
961 is referenced by name. On the new parent, the child will live under
962 'new_child_name', which defaults to 'current_child_name'. TODO: what
963 should we do about metadata? I return a Deferred that fires when the
964 operation finishes. The child name must be a unicode string. I raise
965 NoSuchChildError if I do not have a child by that name."""
967 def build_manifest():
968 """I generate a table of everything reachable from this directory.
969 I also compute deep-stats as described below.
971 I return a Monitor. The Monitor's results will be a dictionary with
974 res['manifest']: a list of (path, cap) tuples for all nodes
975 (directories and files) reachable from this one.
976 'path' will be a tuple of unicode strings. The
977 origin dirnode will be represented by an empty path
979 res['verifycaps']: a list of (printable) verifycap strings, one for
980 each reachable non-LIT node. This is a set:
981 it will contain no duplicates.
982 res['storage-index']: a list of (base32) storage index strings,
983 one for each reachable non-LIT node. This is
984 a set: it will contain no duplicates.
985 res['stats']: a dictionary, the same that is generated by
986 start_deep_stats() below.
988 The Monitor will also have an .origin_si attribute with the (binary)
989 storage index of the starting point.
992 def start_deep_stats():
993 """Return a Monitor, examining all nodes (directories and files)
994 reachable from this one. The Monitor's results will be a dictionary
995 with the following keys::
997 count-immutable-files: count of how many CHK files are in the set
998 count-mutable-files: same, for mutable files (does not include
1000 count-literal-files: same, for LIT files
1001 count-files: sum of the above three
1003 count-directories: count of directories
1005 size-immutable-files: total bytes for all CHK files in the set
1006 size-mutable-files (TODO): same, for current version of all mutable
1007 files, does not include directories
1008 size-literal-files: same, for LIT files
1009 size-directories: size of mutable files used by directories
1011 largest-directory: number of bytes in the largest directory
1012 largest-directory-children: number of children in the largest
1014 largest-immutable-file: number of bytes in the largest CHK file
1016 size-mutable-files is not yet implemented, because it would involve
1017 even more queries than deep_stats does.
1019 The Monitor will also have an .origin_si attribute with the (binary)
1020 storage index of the starting point.
1022 This operation will visit every directory node underneath this one,
1023 and can take a long time to run. On a typical workstation with good
1024 bandwidth, this can examine roughly 15 directories per second (and
1025 takes several minutes of 100% CPU for ~1700 directories).
1028 class ICodecEncoder(Interface):
1029 def set_params(data_size, required_shares, max_shares):
1030 """Set up the parameters of this encoder.
1032 This prepares the encoder to perform an operation that converts a
1033 single block of data into a number of shares, such that a future
1034 ICodecDecoder can use a subset of these shares to recover the
1035 original data. This operation is invoked by calling encode(). Once
1036 the encoding parameters are set up, the encode operation can be
1037 invoked multiple times.
1039 set_params() prepares the encoder to accept blocks of input data that
1040 are exactly 'data_size' bytes in length. The encoder will be prepared
1041 to produce 'max_shares' shares for each encode() operation (although
1042 see the 'desired_share_ids' to use less CPU). The encoding math will
1043 be chosen such that the decoder can get by with as few as
1044 'required_shares' of these shares and still reproduce the original
1045 data. For example, set_params(1000, 5, 5) offers no redundancy at
1046 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1048 Numerical Restrictions: 'data_size' is required to be an integral
1049 multiple of 'required_shares'. In general, the caller should choose
1050 required_shares and max_shares based upon their reliability
1051 requirements and the number of peers available (the total storage
1052 space used is roughly equal to max_shares*data_size/required_shares),
1053 then choose data_size to achieve the memory footprint desired (larger
1054 data_size means more efficient operation, smaller data_size means
1055 smaller memory footprint).
1057 In addition, 'max_shares' must be equal to or greater than
1058 'required_shares'. Of course, setting them to be equal causes
1059 encode() to degenerate into a particularly slow form of the 'split'
1062 See encode() for more details about how these parameters are used.
1064 set_params() must be called before any other ICodecEncoder methods
1069 """Return the 3-tuple of data_size, required_shares, max_shares"""
1071 def get_encoder_type():
1072 """Return a short string that describes the type of this encoder.
1074 There is required to be a global table of encoder classes. This method
1075 returns an index into this table; the value at this index is an
1076 encoder class, and this encoder is an instance of that class.
1079 def get_block_size():
1080 """Return the length of the shares that encode() will produce.
1083 def encode_proposal(data, desired_share_ids=None):
1084 """Encode some data.
1086 'data' must be a string (or other buffer object), and len(data) must
1087 be equal to the 'data_size' value passed earlier to set_params().
1089 This will return a Deferred that will fire with two lists. The first
1090 is a list of shares, each of which is a string (or other buffer
1091 object) such that len(share) is the same as what get_share_size()
1092 returned earlier. The second is a list of shareids, in which each is
1093 an integer. The lengths of the two lists will always be equal to each
1094 other. The user should take care to keep each share closely
1095 associated with its shareid, as one is useless without the other.
1097 The length of this output list will normally be the same as the value
1098 provided to the 'max_shares' parameter of set_params(). This may be
1099 different if 'desired_share_ids' is provided.
1101 'desired_share_ids', if provided, is required to be a sequence of
1102 ints, each of which is required to be >= 0 and < max_shares. If not
1103 provided, encode() will produce 'max_shares' shares, as if
1104 'desired_share_ids' were set to range(max_shares). You might use this
1105 if you initially thought you were going to use 10 peers, started
1106 encoding, and then two of the peers dropped out: you could use
1107 desired_share_ids= to skip the work (both memory and CPU) of
1108 producing shares for the peers which are no longer available.
1112 def encode(inshares, desired_share_ids=None):
1113 """Encode some data. This may be called multiple times. Each call is
1116 inshares is a sequence of length required_shares, containing buffers
1117 (i.e. strings), where each buffer contains the next contiguous
1118 non-overlapping segment of the input data. Each buffer is required to
1119 be the same length, and the sum of the lengths of the buffers is
1120 required to be exactly the data_size promised by set_params(). (This
1121 implies that the data has to be padded before being passed to
1122 encode(), unless of course it already happens to be an even multiple
1123 of required_shares in length.)
1125 ALSO: the requirement to break up your data into 'required_shares'
1126 chunks before calling encode() feels a bit surprising, at least from
1127 the point of view of a user who doesn't know how FEC works. It feels
1128 like an implementation detail that has leaked outside the
1129 abstraction barrier. Can you imagine a use case in which the data to
1130 be encoded might already be available in pre-segmented chunks, such
1131 that it is faster or less work to make encode() take a list rather
1132 than splitting a single string?
1134 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1135 is supposed to *produce* shares, so what it *accepts* should be
1136 something other than shares. Other places in this interface use the
1137 word 'data' for that-which-is-not-shares.. maybe we should use that
1140 'desired_share_ids', if provided, is required to be a sequence of
1141 ints, each of which is required to be >= 0 and < max_shares. If not
1142 provided, encode() will produce 'max_shares' shares, as if
1143 'desired_share_ids' were set to range(max_shares). You might use this
1144 if you initially thought you were going to use 10 peers, started
1145 encoding, and then two of the peers dropped out: you could use
1146 desired_share_ids= to skip the work (both memory and CPU) of
1147 producing shares for the peers which are no longer available.
1149 For each call, encode() will return a Deferred that fires with two
1150 lists, one containing shares and the other containing the shareids.
1151 The get_share_size() method can be used to determine the length of
1152 the share strings returned by encode(). Each shareid is a small
1153 integer, exactly as passed into 'desired_share_ids' (or
1154 range(max_shares), if desired_share_ids was not provided).
1156 The shares and their corresponding shareids are required to be kept
1157 together during storage and retrieval. Specifically, the share data is
1158 useless by itself: the decoder needs to be told which share is which
1159 by providing it with both the shareid and the actual share data.
1161 This function will allocate an amount of memory roughly equal to::
1163 (max_shares - required_shares) * get_share_size()
1165 When combined with the memory that the caller must allocate to
1166 provide the input data, this leads to a memory footprint roughly
1167 equal to the size of the resulting encoded shares (i.e. the expansion
1168 factor times the size of the input segment).
1173 # returning a list of (shareidN,shareN) tuples instead of a pair of
1174 # lists (shareids..,shares..). Brian thought the tuples would
1175 # encourage users to keep the share and shareid together throughout
1176 # later processing, Zooko pointed out that the code to iterate
1177 # through two lists is not really more complicated than using a list
1178 # of tuples and there's also a performance improvement
1180 # having 'data_size' not required to be an integral multiple of
1181 # 'required_shares'. Doing this would require encode() to perform
1182 # padding internally, and we'd prefer to have any padding be done
1183 # explicitly by the caller. Yes, it is an abstraction leak, but
1184 # hopefully not an onerous one.
1187 class ICodecDecoder(Interface):
1188 def set_params(data_size, required_shares, max_shares):
1189 """Set the params. They have to be exactly the same ones that were
1190 used for encoding."""
1192 def get_needed_shares():
1193 """Return the number of shares needed to reconstruct the data.
1194 set_params() is required to be called before this."""
1196 def decode(some_shares, their_shareids):
1197 """Decode a partial list of shares into data.
1199 'some_shares' is required to be a sequence of buffers of sharedata, a
1200 subset of the shares returned by ICodecEncode.encode(). Each share is
1201 required to be of the same length. The i'th element of their_shareids
1202 is required to be the shareid of the i'th buffer in some_shares.
1204 This returns a Deferred which fires with a sequence of buffers. This
1205 sequence will contain all of the segments of the original data, in
1206 order. The sum of the lengths of all of the buffers will be the
1207 'data_size' value passed into the original ICodecEncode.set_params()
1208 call. To get back the single original input block of data, use
1209 ''.join(output_buffers), or you may wish to simply write them in
1210 order to an output file.
1212 Note that some of the elements in the result sequence may be
1213 references to the elements of the some_shares input sequence. In
1214 particular, this means that if those share objects are mutable (e.g.
1215 arrays) and if they are changed, then both the input (the
1216 'some_shares' parameter) and the output (the value given when the
1217 deferred is triggered) will change.
1219 The length of 'some_shares' is required to be exactly the value of
1220 'required_shares' passed into the original ICodecEncode.set_params()
1224 class IEncoder(Interface):
1225 """I take an object that provides IEncryptedUploadable, which provides
1226 encrypted data, and a list of shareholders. I then encode, hash, and
1227 deliver shares to those shareholders. I will compute all the necessary
1228 Merkle hash trees that are necessary to validate the crypttext that
1229 eventually comes back from the shareholders. I provide the URI Extension
1230 Block Hash, and the encoding parameters, both of which must be included
1233 I do not choose shareholders, that is left to the IUploader. I must be
1234 given a dict of RemoteReferences to storage buckets that are ready and
1235 willing to receive data.
1239 """Specify the number of bytes that will be encoded. This must be
1240 peformed before get_serialized_params() can be called.
1242 def set_params(params):
1243 """Override the default encoding parameters. 'params' is a tuple of
1244 (k,d,n), where 'k' is the number of required shares, 'd' is the
1245 shares_of_happiness, and 'n' is the total number of shares that will
1248 Encoding parameters can be set in three ways. 1: The Encoder class
1249 provides defaults (3/7/10). 2: the Encoder can be constructed with
1250 an 'options' dictionary, in which the
1251 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1252 set_params((k,d,n)) can be called.
1254 If you intend to use set_params(), you must call it before
1255 get_share_size or get_param are called.
1258 def set_encrypted_uploadable(u):
1259 """Provide a source of encrypted upload data. 'u' must implement
1260 IEncryptedUploadable.
1262 When this is called, the IEncryptedUploadable will be queried for its
1263 length and the storage_index that should be used.
1265 This returns a Deferred that fires with this Encoder instance.
1267 This must be performed before start() can be called.
1270 def get_param(name):
1271 """Return an encoding parameter, by name.
1273 'storage_index': return a string with the (16-byte truncated SHA-256
1274 hash) storage index to which these shares should be
1277 'share_counts': return a tuple describing how many shares are used:
1278 (needed_shares, shares_of_happiness, total_shares)
1280 'num_segments': return an int with the number of segments that
1283 'segment_size': return an int with the size of each segment.
1285 'block_size': return the size of the individual blocks that will
1286 be delivered to a shareholder's put_block() method. By
1287 knowing this, the shareholder will be able to keep all
1288 blocks in a single file and still provide random access
1289 when reading them. # TODO: can we avoid exposing this?
1291 'share_size': an int with the size of the data that will be stored
1292 on each shareholder. This is aggregate amount of data
1293 that will be sent to the shareholder, summed over all
1294 the put_block() calls I will ever make. It is useful to
1295 determine this size before asking potential
1296 shareholders whether they will grant a lease or not,
1297 since their answers will depend upon how much space we
1298 need. TODO: this might also include some amount of
1299 overhead, like the size of all the hashes. We need to
1300 decide whether this is useful or not.
1302 'serialized_params': a string with a concise description of the
1303 codec name and its parameters. This may be passed
1304 into the IUploadable to let it make sure that
1305 the same file encoded with different parameters
1306 will result in different storage indexes.
1308 Once this is called, set_size() and set_params() may not be called.
1311 def set_shareholders(shareholders):
1312 """Tell the encoder where to put the encoded shares. 'shareholders'
1313 must be a dictionary that maps share number (an integer ranging from
1314 0 to n-1) to an instance that provides IStorageBucketWriter. This
1315 must be performed before start() can be called."""
1318 """Begin the encode/upload process. This involves reading encrypted
1319 data from the IEncryptedUploadable, encoding it, uploading the shares
1320 to the shareholders, then sending the hash trees.
1322 set_encrypted_uploadable() and set_shareholders() must be called
1323 before this can be invoked.
1325 This returns a Deferred that fires with a verify cap when the upload
1326 process is complete. The verifycap, plus the encryption key, is
1327 sufficient to construct the read cap.
1330 class IDecoder(Interface):
1331 """I take a list of shareholders and some setup information, then
1332 download, validate, decode, and decrypt data from them, writing the
1333 results to an output file.
1335 I do not locate the shareholders, that is left to the IDownloader. I must
1336 be given a dict of RemoteReferences to storage buckets that are ready to
1341 """I take a file-like object (providing write and close) to which all
1342 the plaintext data will be written.
1344 TODO: producer/consumer . Maybe write() should return a Deferred that
1345 indicates when it will accept more data? But probably having the
1346 IDecoder be a producer is easier to glue to IConsumer pieces.
1349 def set_shareholders(shareholders):
1350 """I take a dictionary that maps share identifiers (small integers)
1351 to RemoteReferences that provide RIBucketReader. This must be called
1355 """I start the download. This process involves retrieving data and
1356 hash chains from the shareholders, using the hashes to validate the
1357 data, decoding the shares into segments, decrypting the segments,
1358 then writing the resulting plaintext to the output file.
1360 I return a Deferred that will fire (with self) when the download is
1364 class IDownloadTarget(Interface):
1365 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1366 # will register itself as a producer. This allows the target to invoke
1367 # downloader.pauseProducing, resumeProducing, and stopProducing.
1369 """Called before any calls to write() or close(). If an error
1370 occurs before any data is available, fail() may be called without
1371 a previous call to open().
1373 'size' is the length of the file being downloaded, in bytes."""
1376 """Output some data to the target."""
1378 """Inform the target that there is no more data to be written."""
1380 """fail() is called to indicate that the download has failed. 'why'
1381 is a Failure object indicating what went wrong. No further methods
1382 will be invoked on the IDownloadTarget after fail()."""
1383 def register_canceller(cb):
1384 """The CiphertextDownloader uses this to register a no-argument function
1385 that the target can call to cancel the download. Once this canceller
1386 is invoked, no further calls to write() or close() will be made."""
1388 """When the CiphertextDownloader is done, this finish() function will be
1389 called. Whatever it returns will be returned to the invoker of
1390 Downloader.download.
1392 # The following methods are just because that target might be a
1393 # repairer.DownUpConnector, and just because the current CHKUpload object
1394 # expects to find the storage index and encoding parameters in its
1396 def set_storageindex(storageindex):
1397 """ Set the storage index. """
1398 def set_encodingparams(encodingparams):
1399 """ Set the encoding parameters. """
1401 class IDownloader(Interface):
1402 def download(uri, target):
1403 """Perform a CHK download, sending the data to the given target.
1404 'target' must provide IDownloadTarget.
1406 Returns a Deferred that fires (with the results of target.finish)
1407 when the download is finished, or errbacks if something went wrong."""
1409 class IEncryptedUploadable(Interface):
1410 def set_upload_status(upload_status):
1411 """Provide an IUploadStatus object that should be filled with status
1412 information. The IEncryptedUploadable is responsible for setting
1413 key-determination progress ('chk'), size, storage_index, and
1414 ciphertext-fetch progress. It may delegate some of this
1415 responsibility to others, in particular to the IUploadable."""
1418 """This behaves just like IUploadable.get_size()."""
1420 def get_all_encoding_parameters():
1421 """Return a Deferred that fires with a tuple of
1422 (k,happy,n,segment_size). The segment_size will be used as-is, and
1423 must match the following constraints: it must be a multiple of k, and
1424 it shouldn't be unreasonably larger than the file size (if
1425 segment_size is larger than filesize, the difference must be stored
1428 This usually passes through to the IUploadable method of the same
1431 The encoder strictly obeys the values returned by this method. To
1432 make an upload use non-default encoding parameters, you must arrange
1433 to control the values that this method returns.
1436 def get_storage_index():
1437 """Return a Deferred that fires with a 16-byte storage index.
1440 def read_encrypted(length, hash_only):
1441 """This behaves just like IUploadable.read(), but returns crypttext
1442 instead of plaintext. If hash_only is True, then this discards the
1443 data (and returns an empty list); this improves efficiency when
1444 resuming an interrupted upload (where we need to compute the
1445 plaintext hashes, but don't need the redundant encrypted data)."""
1447 def get_plaintext_hashtree_leaves(first, last, num_segments):
1448 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1449 plaintext segments, i.e. get the tagged hashes of the given segments.
1450 The segment size is expected to be generated by the
1451 IEncryptedUploadable before any plaintext is read or ciphertext
1452 produced, so that the segment hashes can be generated with only a
1455 This returns a Deferred which fires with a sequence of hashes, using:
1457 tuple(segment_hashes[first:last])
1459 'num_segments' is used to assert that the number of segments that the
1460 IEncryptedUploadable handled matches the number of segments that the
1461 encoder was expecting.
1463 This method must not be called until the final byte has been read
1464 from read_encrypted(). Once this method is called, read_encrypted()
1465 can never be called again.
1468 def get_plaintext_hash():
1469 """OBSOLETE; Get the hash of the whole plaintext.
1471 This returns a Deferred which fires with a tagged SHA-256 hash of the
1472 whole plaintext, obtained from hashutil.plaintext_hash(data).
1476 """Just like IUploadable.close()."""
1478 class IUploadable(Interface):
1479 def set_upload_status(upload_status):
1480 """Provide an IUploadStatus object that should be filled with status
1481 information. The IUploadable is responsible for setting
1482 key-determination progress ('chk')."""
1484 def set_default_encoding_parameters(params):
1485 """Set the default encoding parameters, which must be a dict mapping
1486 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1487 'max_segment_size'. These might have an influence on the final
1488 encoding parameters returned by get_all_encoding_parameters(), if the
1489 Uploadable doesn't have more specific preferences.
1491 This call is optional: if it is not used, the Uploadable will use
1492 some built-in defaults. If used, this method must be called before
1493 any other IUploadable methods to have any effect.
1497 """Return a Deferred that will fire with the length of the data to be
1498 uploaded, in bytes. This will be called before the data is actually
1499 used, to compute encoding parameters.
1502 def get_all_encoding_parameters():
1503 """Return a Deferred that fires with a tuple of
1504 (k,happy,n,segment_size). The segment_size will be used as-is, and
1505 must match the following constraints: it must be a multiple of k, and
1506 it shouldn't be unreasonably larger than the file size (if
1507 segment_size is larger than filesize, the difference must be stored
1510 The relative values of k and n allow some IUploadables to request
1511 better redundancy than others (in exchange for consuming more space
1514 Larger values of segment_size reduce hash overhead, while smaller
1515 values reduce memory footprint and cause data to be delivered in
1516 smaller pieces (which may provide a smoother and more predictable
1517 download experience).
1519 The encoder strictly obeys the values returned by this method. To
1520 make an upload use non-default encoding parameters, you must arrange
1521 to control the values that this method returns. One way to influence
1522 them may be to call set_encoding_parameters() before calling
1523 get_all_encoding_parameters().
1526 def get_encryption_key():
1527 """Return a Deferred that fires with a 16-byte AES key. This key will
1528 be used to encrypt the data. The key will also be hashed to derive
1531 Uploadables which want to achieve convergence should hash their file
1532 contents and the serialized_encoding_parameters to form the key
1533 (which of course requires a full pass over the data). Uploadables can
1534 use the upload.ConvergentUploadMixin class to achieve this
1537 Uploadables which do not care about convergence (or do not wish to
1538 make multiple passes over the data) can simply return a
1539 strongly-random 16 byte string.
1541 get_encryption_key() may be called multiple times: the IUploadable is
1542 required to return the same value each time.
1546 """Return a Deferred that fires with a list of strings (perhaps with
1547 only a single element) which, when concatenated together, contain the
1548 next 'length' bytes of data. If EOF is near, this may provide fewer
1549 than 'length' bytes. The total number of bytes provided by read()
1550 before it signals EOF must equal the size provided by get_size().
1552 If the data must be acquired through multiple internal read
1553 operations, returning a list instead of a single string may help to
1554 reduce string copies.
1556 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1557 so a 10kB file means length=3kB, 100kB file means length=30kB,
1558 and >=1MB file means length=300kB.
1560 This method provides for a single full pass through the data. Later
1561 use cases may desire multiple passes or access to only parts of the
1562 data (such as a mutable file making small edits-in-place). This API
1563 will be expanded once those use cases are better understood.
1567 """The upload is finished, and whatever filehandle was in use may be
1570 class IUploadResults(Interface):
1571 """I am returned by upload() methods. I contain a number of public
1572 attributes which can be read to determine the results of the upload. Some
1573 of these are functional, some are timing information. All of these may be
1576 .file_size : the size of the file, in bytes
1577 .uri : the CHK read-cap for the file
1578 .ciphertext_fetched : how many bytes were fetched by the helper
1579 .sharemap: dict mapping share identifier to set of serverids
1580 (binary strings). This indicates which servers were given
1581 which shares. For immutable files, the shareid is an
1582 integer (the share number, from 0 to N-1). For mutable
1583 files, it is a string of the form 'seq%d-%s-sh%d',
1584 containing the sequence number, the roothash, and the
1586 .servermap : dict mapping server peerid to a set of share numbers
1587 .timings : dict of timing information, mapping name to seconds (float)
1588 total : total upload time, start to finish
1589 storage_index : time to compute the storage index
1590 peer_selection : time to decide which peers will be used
1591 contacting_helper : initial helper query to upload/no-upload decision
1592 existence_check : helper pre-upload existence check
1593 helper_total : initial helper query to helper finished pushing
1594 cumulative_fetch : helper waiting for ciphertext requests
1595 total_fetch : helper start to last ciphertext response
1596 cumulative_encoding : just time spent in zfec
1597 cumulative_sending : just time spent waiting for storage servers
1598 hashes_and_close : last segment push to shareholder close
1599 total_encode_and_push : first encode to shareholder close
1603 class IDownloadResults(Interface):
1604 """I am created internally by download() methods. I contain a number of
1605 public attributes which contain details about the download process.::
1607 .file_size : the size of the file, in bytes
1608 .servers_used : set of server peerids that were used during download
1609 .server_problems : dict mapping server peerid to a problem string. Only
1610 servers that had problems (bad hashes, disconnects)
1612 .servermap : dict mapping server peerid to a set of share numbers. Only
1613 servers that had any shares are listed here.
1614 .timings : dict of timing information, mapping name to seconds (float)
1615 peer_selection : time to ask servers about shares
1616 servers_peer_selection : dict of peerid to DYHB-query time
1617 uri_extension : time to fetch a copy of the URI extension block
1618 hashtrees : time to fetch the hash trees
1619 segments : time to fetch, decode, and deliver segments
1620 cumulative_fetch : time spent waiting for storage servers
1621 cumulative_decode : just time spent in zfec
1622 cumulative_decrypt : just time spent in decryption
1623 total : total download time, start to finish
1624 fetch_per_server : dict of peerid to list of per-segment fetch times
1628 class IUploader(Interface):
1629 def upload(uploadable):
1630 """Upload the file. 'uploadable' must impement IUploadable. This
1631 returns a Deferred which fires with an UploadResults instance, from
1632 which the URI of the file can be obtained as results.uri ."""
1634 def upload_ssk(write_capability, new_version, uploadable):
1635 """TODO: how should this work?"""
1637 class ICheckable(Interface):
1638 def check(monitor, verify=False, add_lease=False):
1639 """Check upon my health, optionally repairing any problems.
1641 This returns a Deferred that fires with an instance that provides
1642 ICheckResults, or None if the object is non-distributed (i.e. LIT
1645 The monitor will be checked periodically to see if the operation has
1646 been cancelled. If so, no new queries will be sent, and the Deferred
1647 will fire (with a OperationCancelledError) immediately.
1649 Filenodes and dirnodes (which provide IFilesystemNode) are also
1650 checkable. Instances that represent verifier-caps will be checkable
1651 but not downloadable. Some objects (like LIT files) do not actually
1652 live in the grid, and their checkers return None (non-distributed
1653 files are always healthy).
1655 If verify=False, a relatively lightweight check will be performed: I
1656 will ask all servers if they have a share for me, and I will believe
1657 whatever they say. If there are at least N distinct shares on the
1658 grid, my results will indicate r.is_healthy()==True. This requires a
1659 roundtrip to each server, but does not transfer very much data, so
1660 the network bandwidth is fairly low.
1662 If verify=True, a more resource-intensive check will be performed:
1663 every share will be downloaded, and the hashes will be validated on
1664 every bit. I will ignore any shares that failed their hash checks. If
1665 there are at least N distinct valid shares on the grid, my results
1666 will indicate r.is_healthy()==True. This requires N/k times as much
1667 download bandwidth (and server disk IO) as a regular download. If a
1668 storage server is holding a corrupt share, or is experiencing memory
1669 failures during retrieval, or is malicious or buggy, then
1670 verification will detect the problem, but checking will not.
1672 If add_lease=True, I will ensure that an up-to-date lease is present
1673 on each share. The lease secrets will be derived from by node secret
1674 (in BASEDIR/private/secret), so either I will add a new lease to the
1675 share, or I will merely renew the lease that I already had. In a
1676 future version of the storage-server protocol (once Accounting has
1677 been implemented), there may be additional options here to define the
1678 kind of lease that is obtained (which account number to claim, etc).
1680 TODO: any problems seen during checking will be reported to the
1681 health-manager.furl, a centralized object which is responsible for
1682 figuring out why files are unhealthy so corrective action can be
1686 def check_and_repair(monitor, verify=False, add_lease=False):
1687 """Like check(), but if the file/directory is not healthy, attempt to
1690 Any non-healthy result will cause an immediate repair operation, to
1691 generate and upload new shares. After repair, the file will be as
1692 healthy as we can make it. Details about what sort of repair is done
1693 will be put in the check-and-repair results. The Deferred will not
1694 fire until the repair is complete.
1696 This returns a Deferred which fires with an instance of
1697 ICheckAndRepairResults."""
1699 class IDeepCheckable(Interface):
1700 def start_deep_check(verify=False, add_lease=False):
1701 """Check upon the health of me and everything I can reach.
1703 This is a recursive form of check(), useable only on dirnodes.
1705 I return a Monitor, with results that are an IDeepCheckResults
1708 TODO: If any of the directories I traverse are unrecoverable, the
1709 Monitor will report failure. If any of the files I check upon are
1710 unrecoverable, those problems will be reported in the
1711 IDeepCheckResults as usual, and the Monitor will not report a
1715 def start_deep_check_and_repair(verify=False, add_lease=False):
1716 """Check upon the health of me and everything I can reach. Repair
1717 anything that isn't healthy.
1719 This is a recursive form of check_and_repair(), useable only on
1722 I return a Monitor, with results that are an
1723 IDeepCheckAndRepairResults object.
1725 TODO: If any of the directories I traverse are unrecoverable, the
1726 Monitor will report failure. If any of the files I check upon are
1727 unrecoverable, those problems will be reported in the
1728 IDeepCheckResults as usual, and the Monitor will not report a
1732 class ICheckResults(Interface):
1733 """I contain the detailed results of a check/verify operation.
1736 def get_storage_index():
1737 """Return a string with the (binary) storage index."""
1738 def get_storage_index_string():
1739 """Return a string with the (printable) abbreviated storage index."""
1741 """Return the (string) URI of the object that was checked."""
1744 """Return a boolean, True if the file/dir is fully healthy, False if
1745 it is damaged in any way. Non-distributed LIT files always return
1748 def is_recoverable():
1749 """Return a boolean, True if the file/dir can be recovered, False if
1750 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1751 files always return True."""
1753 def needs_rebalancing():
1754 """Return a boolean, True if the file/dir's reliability could be
1755 improved by moving shares to new servers. Non-distributed LIT files
1756 always return False."""
1760 """Return a dictionary that describes the state of the file/dir. LIT
1761 files always return an empty dictionary. Normal files and directories
1762 return a dictionary with the following keys (note that these use
1763 binary strings rather than base32-encoded ones) (also note that for
1764 mutable files, these counts are for the 'best' version):
1766 count-shares-good: the number of distinct good shares that were found
1767 count-shares-needed: 'k', the number of shares required for recovery
1768 count-shares-expected: 'N', the number of total shares generated
1769 count-good-share-hosts: the number of distinct storage servers with
1770 good shares. If this number is less than
1771 count-shares-good, then some shares are
1772 doubled up, increasing the correlation of
1773 failures. This indicates that one or more
1774 shares should be moved to an otherwise unused
1775 server, if one is available.
1776 count-corrupt-shares: the number of shares with integrity failures
1777 list-corrupt-shares: a list of 'share locators', one for each share
1778 that was found to be corrupt. Each share
1779 locator is a list of (serverid, storage_index,
1781 count-incompatible-shares: the number of shares which are of a share
1782 format unknown to this checker
1783 list-incompatible-shares: a list of 'share locators', one for each
1784 share that was found to be of an unknown
1785 format. Each share locator is a list of
1786 (serverid, storage_index, sharenum).
1787 servers-responding: list of (binary) storage server identifiers,
1788 one for each server which responded to the share
1789 query (even if they said they didn't have
1790 shares, and even if they said they did have
1791 shares but then didn't send them when asked, or
1792 dropped the connection, or returned a Failure,
1793 and even if they said they did have shares and
1794 sent incorrect ones when asked)
1795 sharemap: dict mapping share identifier to list of serverids
1796 (binary strings). This indicates which servers are holding
1797 which shares. For immutable files, the shareid is an
1798 integer (the share number, from 0 to N-1). For mutable
1799 files, it is a string of the form 'seq%d-%s-sh%d',
1800 containing the sequence number, the roothash, and the
1803 The following keys are most relevant for mutable files, but immutable
1804 files will provide sensible values too::
1806 count-wrong-shares: the number of shares for versions other than the
1807 'best' one (which is defined as being the
1808 recoverable version with the highest sequence
1809 number, then the highest roothash). These are
1810 either leftover shares from an older version
1811 (perhaps on a server that was offline when an
1812 update occurred), shares from an unrecoverable
1813 newer version, or shares from an alternate
1814 current version that results from an
1815 uncoordinated write collision. For a healthy
1816 file, this will equal 0.
1818 count-recoverable-versions: the number of recoverable versions of
1819 the file. For a healthy file, this will
1822 count-unrecoverable-versions: the number of unrecoverable versions
1823 of the file. For a healthy file, this
1829 """Return a string with a brief (one-line) summary of the results."""
1832 """Return a list of strings with more detailed results."""
1834 class ICheckAndRepairResults(Interface):
1835 """I contain the detailed results of a check/verify/repair operation.
1837 The IFilesystemNode.check()/verify()/repair() methods all return
1838 instances that provide ICheckAndRepairResults.
1841 def get_storage_index():
1842 """Return a string with the (binary) storage index."""
1843 def get_storage_index_string():
1844 """Return a string with the (printable) abbreviated storage index."""
1845 def get_repair_attempted():
1846 """Return a boolean, True if a repair was attempted. We might not
1847 attempt to repair the file because it was healthy, or healthy enough
1848 (i.e. some shares were missing but not enough to exceed some
1849 threshold), or because we don't know how to repair this object."""
1850 def get_repair_successful():
1851 """Return a boolean, True if repair was attempted and the file/dir
1852 was fully healthy afterwards. False if no repair was attempted or if
1853 a repair attempt failed."""
1854 def get_pre_repair_results():
1855 """Return an ICheckResults instance that describes the state of the
1856 file/dir before any repair was attempted."""
1857 def get_post_repair_results():
1858 """Return an ICheckResults instance that describes the state of the
1859 file/dir after any repair was attempted. If no repair was attempted,
1860 the pre-repair and post-repair results will be identical."""
1863 class IDeepCheckResults(Interface):
1864 """I contain the results of a deep-check operation.
1866 This is returned by a call to ICheckable.deep_check().
1869 def get_root_storage_index_string():
1870 """Return the storage index (abbreviated human-readable string) of
1871 the first object checked."""
1873 """Return a dictionary with the following keys::
1875 count-objects-checked: count of how many objects were checked
1876 count-objects-healthy: how many of those objects were completely
1878 count-objects-unhealthy: how many were damaged in some way
1879 count-objects-unrecoverable: how many were unrecoverable
1880 count-corrupt-shares: how many shares were found to have
1881 corruption, summed over all objects
1885 def get_corrupt_shares():
1886 """Return a set of (serverid, storage_index, sharenum) for all shares
1887 that were found to be corrupt. Both serverid and storage_index are
1890 def get_all_results():
1891 """Return a dictionary mapping pathname (a tuple of strings, ready to
1892 be slash-joined) to an ICheckResults instance, one for each object
1893 that was checked."""
1895 def get_results_for_storage_index(storage_index):
1896 """Retrive the ICheckResults instance for the given (binary)
1897 storage index. Raises KeyError if there are no results for that
1901 """Return a dictionary with the same keys as
1902 IDirectoryNode.deep_stats()."""
1904 class IDeepCheckAndRepairResults(Interface):
1905 """I contain the results of a deep-check-and-repair operation.
1907 This is returned by a call to ICheckable.deep_check_and_repair().
1910 def get_root_storage_index_string():
1911 """Return the storage index (abbreviated human-readable string) of
1912 the first object checked."""
1914 """Return a dictionary with the following keys::
1916 count-objects-checked: count of how many objects were checked
1917 count-objects-healthy-pre-repair: how many of those objects were
1918 completely healthy (before any
1920 count-objects-unhealthy-pre-repair: how many were damaged in
1922 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1923 count-objects-healthy-post-repair: how many of those objects were
1924 completely healthy (after any
1926 count-objects-unhealthy-post-repair: how many were damaged in
1928 count-objects-unrecoverable-post-repair: how many were
1930 count-repairs-attempted: repairs were attempted on this many
1931 objects. The count-repairs- keys will
1932 always be provided, however unless
1933 repair=true is present, they will all
1935 count-repairs-successful: how many repairs resulted in healthy
1937 count-repairs-unsuccessful: how many repairs resulted did not
1938 results in completely healthy objects
1939 count-corrupt-shares-pre-repair: how many shares were found to
1940 have corruption, summed over all
1941 objects examined (before any
1943 count-corrupt-shares-post-repair: how many shares were found to
1944 have corruption, summed over all
1945 objects examined (after any
1950 """Return a dictionary with the same keys as
1951 IDirectoryNode.deep_stats()."""
1953 def get_corrupt_shares():
1954 """Return a set of (serverid, storage_index, sharenum) for all shares
1955 that were found to be corrupt before any repair was attempted. Both
1956 serverid and storage_index are binary.
1958 def get_remaining_corrupt_shares():
1959 """Return a set of (serverid, storage_index, sharenum) for all shares
1960 that were found to be corrupt after any repair was completed. Both
1961 serverid and storage_index are binary. These are shares that need
1962 manual inspection and probably deletion.
1964 def get_all_results():
1965 """Return a dictionary mapping pathname (a tuple of strings, ready to
1966 be slash-joined) to an ICheckAndRepairResults instance, one for each
1967 object that was checked."""
1969 def get_results_for_storage_index(storage_index):
1970 """Retrive the ICheckAndRepairResults instance for the given (binary)
1971 storage index. Raises KeyError if there are no results for that
1975 class IRepairable(Interface):
1976 def repair(check_results):
1977 """Attempt to repair the given object. Returns a Deferred that fires
1978 with a IRepairResults object.
1980 I must be called with an object that implements ICheckResults, as
1981 proof that you have actually discovered a problem with this file. I
1982 will use the data in the checker results to guide the repair process,
1983 such as which servers provided bad data and should therefore be
1984 avoided. The ICheckResults object is inside the
1985 ICheckAndRepairResults object, which is returned by the
1986 ICheckable.check() method::
1988 d = filenode.check(repair=False)
1989 def _got_results(check_and_repair_results):
1990 check_results = check_and_repair_results.get_pre_repair_results()
1991 return filenode.repair(check_results)
1992 d.addCallback(_got_results)
1996 class IRepairResults(Interface):
1997 """I contain the results of a repair operation."""
2000 class IClient(Interface):
2001 def upload(uploadable):
2002 """Upload some data into a CHK, get back the UploadResults for it.
2003 @param uploadable: something that implements IUploadable
2004 @return: a Deferred that fires with the UploadResults instance.
2005 To get the URI for this file, use results.uri .
2008 def create_mutable_file(contents=""):
2009 """Create a new mutable file (with initial) contents, get back the
2012 @param contents: (bytestring, callable, or None): this provides the
2013 initial contents of the mutable file. If 'contents' is a bytestring,
2014 it will be used as-is. If 'contents' is a callable, it will be
2015 invoked with the new MutableFileNode instance and is expected to
2016 return a bytestring with the initial contents of the file (the
2017 callable can use node.get_writekey() to decide how to encrypt the
2018 initial contents, e.g. for a brand new dirnode with initial
2019 children). contents=None is equivalent to an empty string. Using
2020 content_maker= is more efficient than creating a mutable file and
2021 setting its contents in two separate operations.
2023 @return: a Deferred that fires with tne (string) SSK URI for the new
2027 def create_dirnode(initial_children={}):
2028 """Create a new unattached dirnode, possibly with initial children.
2030 @param initial_children: dict with keys that are unicode child names,
2031 and values that are (childnode, metadata) tuples.
2033 @return: a Deferred that fires with the new IDirectoryNode instance.
2036 def create_node_from_uri(uri, rouri):
2037 """Create a new IFilesystemNode instance from the uri, synchronously.
2038 @param uri: a string or IURI-providing instance, or None. This could
2039 be for a LiteralFileNode, a CHK file node, a mutable file
2040 node, or a directory node
2041 @param rouri: a string or IURI-providing instance, or None. If the
2042 main uri is None, I will use the rouri instead. If I
2043 recognize the format of the main uri, I will ignore the
2044 rouri (because it can be derived from the writecap).
2046 @return: an instance that provides IFilesystemNode (or more usefully
2047 one of its subclasses). File-specifying URIs will result in
2048 IFileNode or IMutableFileNode -providing instances, like
2049 FileNode, LiteralFileNode, or MutableFileNode.
2050 Directory-specifying URIs will result in
2051 IDirectoryNode-providing instances, like DirectoryNode.
2054 class INodeMaker(Interface):
2055 """The NodeMaker is used to create IFilesystemNode instances. It can
2056 accept a filecap/dircap string and return the node right away. It can
2057 also create new nodes (i.e. upload a file, or create a mutable file)
2058 asynchronously. Once you have one of these nodes, you can use other
2059 methods to determine whether it is a file or directory, and to download
2060 or modify its contents.
2062 The NodeMaker encapsulates all the authorities that these
2063 IfilesystemNodes require (like references to the StorageFarmBroker). Each
2064 Tahoe process will typically have a single NodeMaker, but unit tests may
2065 create simplified/mocked forms for testing purposes.
2067 def create_from_cap(writecap, readcap=None):
2068 """I create an IFilesystemNode from the given writecap/readcap. I can
2069 only provide nodes for existing file/directory objects: use my other
2070 methods to create new objects. I return synchronously."""
2072 def create_mutable_file(contents=None, keysize=None):
2073 """I create a new mutable file, and return a Deferred which will fire
2074 with the IMutableFileNode instance when it is ready. If contents= is
2075 provided (a bytestring), it will be used as the initial contents of
2076 the new file, otherwise the file will contain zero bytes. keysize= is
2077 for use by unit tests, to create mutable files that are smaller than
2080 def create_new_mutable_directory(initial_children={}):
2081 """I create a new mutable directory, and return a Deferred which will
2082 fire with the IDirectoryNode instance when it is ready. If
2083 initial_children= is provided (a dict mapping unicode child name to
2084 (childnode, metadata_dict) tuples), the directory will be populated
2085 with those children, otherwise it will be empty."""
2087 class IClientStatus(Interface):
2088 def list_all_uploads():
2089 """Return a list of uploader objects, one for each upload which
2090 currently has an object available (tracked with weakrefs). This is
2091 intended for debugging purposes."""
2092 def list_active_uploads():
2093 """Return a list of active IUploadStatus objects."""
2094 def list_recent_uploads():
2095 """Return a list of IUploadStatus objects for the most recently
2098 def list_all_downloads():
2099 """Return a list of downloader objects, one for each download which
2100 currently has an object available (tracked with weakrefs). This is
2101 intended for debugging purposes."""
2102 def list_active_downloads():
2103 """Return a list of active IDownloadStatus objects."""
2104 def list_recent_downloads():
2105 """Return a list of IDownloadStatus objects for the most recently
2106 started downloads."""
2108 class IUploadStatus(Interface):
2110 """Return a timestamp (float with seconds since epoch) indicating
2111 when the operation was started."""
2112 def get_storage_index():
2113 """Return a string with the (binary) storage index in use on this
2114 upload. Returns None if the storage index has not yet been
2117 """Return an integer with the number of bytes that will eventually
2118 be uploaded for this file. Returns None if the size is not yet known.
2121 """Return True if this upload is using a Helper, False if not."""
2123 """Return a string describing the current state of the upload
2126 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2127 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2128 made towards hashing the file to determine a CHK encryption key: if
2129 non-convergent encryption is in use, this will be trivial, otherwise
2130 the whole file must be hashed. 'ciphertext' describes how much of the
2131 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2132 uploads. 'encode_and_push' describes how much of the encode-and-push
2133 process has finished: for helper uploads this is dependent upon the
2134 helper providing progress reports. It might be reasonable to add all
2135 three numbers and report the sum to the user."""
2137 """Return True if the upload is currently active, False if not."""
2139 """Return an instance of UploadResults (which contains timing and
2140 sharemap information). Might return None if the upload is not yet
2143 """Each upload status gets a unique number: this method returns that
2144 number. This provides a handle to this particular upload, so a web
2145 page can generate a suitable hyperlink."""
2147 class IDownloadStatus(Interface):
2149 """Return a timestamp (float with seconds since epoch) indicating
2150 when the operation was started."""
2151 def get_storage_index():
2152 """Return a string with the (binary) storage index in use on this
2153 download. This may be None if there is no storage index (i.e. LIT
2156 """Return an integer with the number of bytes that will eventually be
2157 retrieved for this file. Returns None if the size is not yet known.
2160 """Return True if this download is using a Helper, False if not."""
2162 """Return a string describing the current state of the download
2165 """Returns a float (from 0.0 to 1.0) describing the amount of the
2166 download that has completed. This value will remain at 0.0 until the
2167 first byte of plaintext is pushed to the download target."""
2169 """Return True if the download is currently active, False if not."""
2171 """Each download status gets a unique number: this method returns
2172 that number. This provides a handle to this particular download, so a
2173 web page can generate a suitable hyperlink."""
2175 class IServermapUpdaterStatus(Interface):
2177 class IPublishStatus(Interface):
2179 class IRetrieveStatus(Interface):
2182 class NotCapableError(Exception):
2183 """You have tried to write to a read-only node."""
2185 class BadWriteEnablerError(Exception):
2188 class RIControlClient(RemoteInterface):
2190 def wait_for_client_connections(num_clients=int):
2191 """Do not return until we have connections to at least NUM_CLIENTS
2195 def upload_from_file_to_uri(filename=str,
2196 convergence=ChoiceOf(None,
2197 StringConstraint(2**20))):
2198 """Upload a file to the grid. This accepts a filename (which must be
2199 absolute) that points to a file on the node's local disk. The node will
2200 read the contents of this file, upload it to the grid, then return the
2201 URI at which it was uploaded. If convergence is None then a random
2202 encryption key will be used, else the plaintext will be hashed, then
2203 that hash will be mixed together with the "convergence" string to form
2208 def download_from_uri_to_file(uri=URI, filename=str):
2209 """Download a file from the grid, placing it on the node's local disk
2210 at the given filename (which must be absolute[?]). Returns the
2211 absolute filename where the file was written."""
2216 def get_memory_usage():
2217 """Return a dict describes the amount of memory currently in use. The
2218 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2219 measuring memory consupmtion in bytes."""
2220 return DictOf(str, int)
2222 def speed_test(count=int, size=int, mutable=Any()):
2223 """Write 'count' tempfiles to disk, all of the given size. Measure
2224 how long (in seconds) it takes to upload them all to the servers.
2225 Then measure how long it takes to download all of them. If 'mutable'
2226 is 'create', time creation of mutable files. If 'mutable' is
2227 'upload', then time access to the same mutable file instead of
2230 Returns a tuple of (upload_time, download_time).
2232 return (float, float)
2234 def measure_peer_response_time():
2235 """Send a short message to each connected peer, and measure the time
2236 it takes for them to respond to it. This is a rough measure of the
2237 application-level round trip time.
2239 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2242 return DictOf(Nodeid, float)
2244 UploadResults = Any() #DictOf(str, str)
2246 class RIEncryptedUploadable(RemoteInterface):
2247 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2252 def get_all_encoding_parameters():
2253 return (int, int, int, long)
2255 def read_encrypted(offset=Offset, length=ReadSize):
2262 class RICHKUploadHelper(RemoteInterface):
2263 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2267 Return a dictionary of version information.
2269 return DictOf(str, Any())
2271 def upload(reader=RIEncryptedUploadable):
2272 return UploadResults
2275 class RIHelper(RemoteInterface):
2276 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2280 Return a dictionary of version information.
2282 return DictOf(str, Any())
2284 def upload_chk(si=StorageIndex):
2285 """See if a file with a given storage index needs uploading. The
2286 helper will ask the appropriate storage servers to see if the file
2287 has already been uploaded. If so, the helper will return a set of
2288 'upload results' that includes whatever hashes are needed to build
2289 the read-cap, and perhaps a truncated sharemap.
2291 If the file has not yet been uploaded (or if it was only partially
2292 uploaded), the helper will return an empty upload-results dictionary
2293 and also an RICHKUploadHelper object that will take care of the
2294 upload process. The client should call upload() on this object and
2295 pass it a reference to an RIEncryptedUploadable object that will
2296 provide ciphertext. When the upload is finished, the upload() method
2297 will finish and return the upload results.
2299 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2302 class RIStatsProvider(RemoteInterface):
2303 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2305 Provides access to statistics and monitoring information.
2310 returns a dictionary containing 'counters' and 'stats', each a
2311 dictionary with string counter/stat name keys, and numeric values.
2312 counters are monotonically increasing measures of work done, and
2313 stats are instantaneous measures (potentially time averaged
2316 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2318 class RIStatsGatherer(RemoteInterface):
2319 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2321 Provides a monitoring service for centralised collection of stats
2324 def provide(provider=RIStatsProvider, nickname=str):
2326 @param provider: a stats collector instance which should be polled
2327 periodically by the gatherer to collect stats.
2328 @param nickname: a name useful to identify the provided client
2333 class IStatsProducer(Interface):
2336 returns a dictionary, with str keys representing the names of stats
2337 to be monitored, and numeric values.
2340 class RIKeyGenerator(RemoteInterface):
2341 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2343 Provides a service offering to make RSA key pairs.
2346 def get_rsa_key_pair(key_size=int):
2348 @param key_size: the size of the signature key.
2349 @return: tuple(verifying_key, signing_key)
2351 return TupleOf(str, str)
2354 class FileTooLargeError(Exception):
2357 class IValidatedThingProxy(Interface):
2359 """ Acquire a thing and validate it. Return a deferred which is
2360 eventually fired with self if the thing is valid or errbacked if it
2361 can't be acquired or validated."""
2363 class InsufficientVersionError(Exception):
2364 def __init__(self, needed, got):
2365 self.needed = needed
2368 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,