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
354 # hm, we need a solution for forward references in schemas
355 FileNode_ = Any() # TODO: foolscap needs constraints on copyables
356 DirectoryNode_ = Any() # TODO: same
357 AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
360 class IURI(Interface):
361 def init_from_string(uri):
362 """Accept a string (as created by my to_string() method) and populate
363 this instance with its data. I am not normally called directly,
364 please use the module-level uri.from_string() function to convert
365 arbitrary URI strings into IURI-providing instances."""
368 """Return False if this URI be used to modify the data. Return True
369 if this URI cannot be used to modify the data."""
372 """Return True if the data can be modified by *somebody* (perhaps
373 someone who has a more powerful URI than this one)."""
376 """Return another IURI instance, which represents a read-only form of
377 this one. If is_readonly() is True, this returns self."""
379 def get_verify_cap():
380 """Return an instance that provides IVerifierURI, which can be used
381 to check on the availability of the file or directory, without
382 providing enough capabilities to actually read or modify the
383 contents. This may return None if the file does not need checking or
384 verification (e.g. LIT URIs).
388 """Return a string of printable ASCII characters, suitable for
389 passing into init_from_string."""
391 class IVerifierURI(Interface, IURI):
392 def init_from_string(uri):
393 """Accept a string (as created by my to_string() method) and populate
394 this instance with its data. I am not normally called directly,
395 please use the module-level uri.from_string() function to convert
396 arbitrary URI strings into IURI-providing instances."""
399 """Return a string of printable ASCII characters, suitable for
400 passing into init_from_string."""
402 class IDirnodeURI(Interface):
403 """I am a URI which represents a dirnode."""
406 class IFileURI(Interface):
407 """I am a URI which represents a filenode."""
409 """Return the length (in bytes) of the file that I represent."""
411 class IImmutableFileURI(IFileURI):
414 class IMutableFileURI(Interface):
415 """I am a URI which represents a mutable filenode."""
416 class INewDirectoryURI(Interface):
418 class IReadonlyNewDirectoryURI(Interface):
422 class IFilesystemNode(Interface):
425 Return the URI that can be used by others to get access to this
426 node. If this node is read-only, the URI will only offer read-only
427 access. If this node is read-write, the URI will offer read-write
430 If you have read-write access to a node and wish to share merely
431 read-only access with others, use get_readonly_uri().
434 def get_readonly_uri():
435 """Return the directory URI that can be used by others to get
436 read-only access to this directory node. The result is a read-only
437 URI, regardless of whether this dirnode is read-only or read-write.
439 If you have merely read-only access to this dirnode,
440 get_readonly_uri() will return the same thing as get_uri().
443 def get_repair_cap():
444 """Return an IURI instance that can be used to repair the file, or
445 None if this node cannot be repaired (either because it is not
446 distributed, like a LIT file, or because the node does not represent
447 sufficient authority to create a repair-cap, like a read-only RSA
448 mutable file node [which cannot create the correct write-enablers]).
451 def get_verify_cap():
452 """Return an IVerifierURI instance that represents the
453 'verifiy/refresh capability' for this node. The holder of this
454 capability will be able to renew the lease for this node, protecting
455 it from garbage-collection. They will also be able to ask a server if
456 it holds a share for the file or directory.
459 def get_storage_index():
460 """Return a string with the (binary) storage index in use on this
461 download. This may be None if there is no storage index (i.e. LIT
465 """Return True if this reference provides mutable access to the given
466 file or directory (i.e. if you can modify it), or False if not. Note
467 that even if this reference is read-only, someone else may hold a
468 read-write reference to it."""
471 """Return True if this file or directory is mutable (by *somebody*,
472 not necessarily you), False if it is is immutable. Note that a file
473 might be mutable overall, but your reference to it might be
474 read-only. On the other hand, all references to an immutable file
475 will be read-only; there are no read-write references to an immutable
479 class IMutableFilesystemNode(IFilesystemNode):
482 class IFileNode(IFilesystemNode):
483 def download(target):
484 """Download the file's contents to a given IDownloadTarget"""
486 def download_to_data():
487 """Download the file's contents. Return a Deferred that fires
488 with those contents."""
491 """Return the length (in bytes) of the data this node represents."""
493 def read(consumer, offset=0, size=None):
494 """Download a portion (possibly all) of the file's contents, making
495 them available to the given IConsumer. Return a Deferred that fires
496 (with the consumer) when the consumer is unregistered (either because
497 the last byte has been given to it, or because the consumer threw an
498 exception during write(), possibly because it no longer wants to
499 receive data). The portion downloaded will start at 'offset' and
500 contain 'size' bytes (or the remainder of the file if size==None).
502 The consumer will be used in non-streaming mode: an IPullProducer
503 will be attached to it.
505 The consumer will not receive data right away: several network trips
506 must occur first. The order of events will be::
508 consumer.registerProducer(p, streaming)
509 (if streaming == False)::
510 consumer does p.resumeProducing()
512 consumer does p.resumeProducing()
513 consumer.write(data).. (repeat until all data is written)
514 consumer.unregisterProducer()
515 deferred.callback(consumer)
517 If a download error occurs, or an exception is raised by
518 consumer.registerProducer() or consumer.write(), I will call
519 consumer.unregisterProducer() and then deliver the exception via
520 deferred.errback(). To cancel the download, the consumer should call
521 p.stopProducing(), which will result in an exception being delivered
522 via deferred.errback().
524 A simple download-to-memory consumer example would look like this::
526 class MemoryConsumer:
527 implements(IConsumer)
531 def registerProducer(self, p, streaming):
532 assert streaming == False
535 def write(self, data):
536 self.chunks.append(data)
537 def unregisterProducer(self):
539 d = filenode.read(MemoryConsumer())
540 d.addCallback(lambda mc: "".join(mc.chunks))
545 class IMutableFileNode(IFileNode, IMutableFilesystemNode):
546 """I provide access to a 'mutable file', which retains its identity
547 regardless of what contents are put in it.
549 The consistency-vs-availability problem means that there might be
550 multiple versions of a file present in the grid, some of which might be
551 unrecoverable (i.e. have fewer than 'k' shares). These versions are
552 loosely ordered: each has a sequence number and a hash, and any version
553 with seqnum=N was uploaded by a node which has seen at least one version
556 The 'servermap' (an instance of IMutableFileServerMap) is used to
557 describe the versions that are known to be present in the grid, and which
558 servers are hosting their shares. It is used to represent the 'state of
559 the world', and is used for this purpose by my test-and-set operations.
560 Downloading the contents of the mutable file will also return a
561 servermap. Uploading a new version into the mutable file requires a
562 servermap as input, and the semantics of the replace operation is
563 'replace the file with my new version if it looks like nobody else has
564 changed the file since my previous download'. Because the file is
565 distributed, this is not a perfect test-and-set operation, but it will do
566 its best. If the replace process sees evidence of a simultaneous write,
567 it will signal an UncoordinatedWriteError, so that the caller can take
571 Most readers will want to use the 'best' current version of the file, and
572 should use my 'download_best_version()' method.
574 To unconditionally replace the file, callers should use overwrite(). This
575 is the mode that user-visible mutable files will probably use.
577 To apply some delta to the file, call modify() with a callable modifier
578 function that can apply the modification that you want to make. This is
579 the mode that dirnodes will use, since most directory modification
580 operations can be expressed in terms of deltas to the directory state.
583 Three methods are available for users who need to perform more complex
584 operations. The first is get_servermap(), which returns an up-to-date
585 servermap using a specified mode. The second is download_version(), which
586 downloads a specific version (not necessarily the 'best' one). The third
587 is 'upload', which accepts new contents and a servermap (which must have
588 been updated with MODE_WRITE). The upload method will attempt to apply
589 the new contents as long as no other node has modified the file since the
590 servermap was updated. This might be useful to a caller who wants to
591 merge multiple versions into a single new one.
593 Note that each time the servermap is updated, a specific 'mode' is used,
594 which determines how many peers are queried. To use a servermap for my
595 replace() method, that servermap must have been updated in MODE_WRITE.
596 These modes are defined in allmydata.mutable.common, and consist of
597 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
598 allmydata/mutable/servermap.py for details about the differences.
600 Mutable files are currently limited in size (about 3.5MB max) and can
601 only be retrieved and updated all-at-once, as a single big string. Future
602 versions of our mutable files will remove this restriction.
605 def download_best_version():
606 """Download the 'best' available version of the file, meaning one of
607 the recoverable versions with the highest sequence number. If no
608 uncoordinated writes have occurred, and if enough shares are
609 available, then this will be the most recent version that has been
612 I update an internal servermap with MODE_READ, determine which
613 version of the file is indicated by
614 servermap.best_recoverable_version(), and return a Deferred that
615 fires with its contents. If no version is recoverable, the Deferred
616 will errback with UnrecoverableFileError.
619 def get_size_of_best_version():
620 """Find the size of the version that would be downloaded with
621 download_best_version(), without actually downloading the whole file.
623 I return a Deferred that fires with an integer.
626 def overwrite(new_contents):
627 """Unconditionally replace the contents of the mutable file with new
628 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
629 is only appropriate to use when the new contents of the file are
630 completely unrelated to the old ones, and you do not care about other
633 I return a Deferred that fires (with a PublishStatus object) when the
634 update has completed.
637 def modify(modifier_cb):
638 """Modify the contents of the file, by downloading the current
639 version, applying the modifier function (or bound method), then
640 uploading the new version. I return a Deferred that fires (with a
641 PublishStatus object) when the update is complete.
643 The modifier callable will be given three arguments: a string (with
644 the old contents), a 'first_time' boolean, and a servermap. As with
645 download_best_version(), the old contents will be from the best
646 recoverable version, but the modifier can use the servermap to make
647 other decisions (such as refusing to apply the delta if there are
648 multiple parallel versions, or if there is evidence of a newer
649 unrecoverable version). 'first_time' will be True the first time the
650 modifier is called, and False on any subsequent calls.
652 The callable should return a string with the new contents. The
653 callable must be prepared to be called multiple times, and must
654 examine the input string to see if the change that it wants to make
655 is already present in the old version. If it does not need to make
656 any changes, it can either return None, or return its input string.
658 If the modifier raises an exception, it will be returned in the
663 def get_servermap(mode):
664 """Return a Deferred that fires with an IMutableFileServerMap
665 instance, updated using the given mode.
668 def download_version(servermap, version):
669 """Download a specific version of the file, using the servermap
670 as a guide to where the shares are located.
672 I return a Deferred that fires with the requested contents, or
673 errbacks with UnrecoverableFileError. Note that a servermap which was
674 updated with MODE_ANYTHING or MODE_READ may not know about shares for
675 all versions (those modes stop querying servers as soon as they can
676 fulfil their goals), so you may want to use MODE_CHECK (which checks
677 everything) to get increased visibility.
680 def upload(new_contents, servermap):
681 """Replace the contents of the file with new ones. This requires a
682 servermap that was previously updated with MODE_WRITE.
684 I attempt to provide test-and-set semantics, in that I will avoid
685 modifying any share that is different than the version I saw in the
686 servermap. However, if another node is writing to the file at the
687 same time as me, I may manage to update some shares while they update
688 others. If I see any evidence of this, I will signal
689 UncoordinatedWriteError, and the file will be left in an inconsistent
690 state (possibly the version you provided, possibly the old version,
691 possibly somebody else's version, and possibly a mix of shares from
694 The recommended response to UncoordinatedWriteError is to either
695 return it to the caller (since they failed to coordinate their
696 writes), or to attempt some sort of recovery. It may be sufficient to
697 wait a random interval (with exponential backoff) and repeat your
698 operation. If I do not signal UncoordinatedWriteError, then I was
699 able to write the new version without incident.
701 I return a Deferred that fires (with a PublishStatus object) when the
702 publish has completed. I will update the servermap in-place with the
703 location of all new shares.
707 """Return this filenode's writekey, or None if the node does not have
708 write-capability. This may be used to assist with data structures
709 that need to make certain data available only to writers, such as the
710 read-write child caps in dirnodes. The recommended process is to have
711 reader-visible data be submitted to the filenode in the clear (where
712 it will be encrypted by the filenode using the readkey), but encrypt
713 writer-visible data using this writekey.
716 class NotEnoughSharesError(Exception):
717 def __init__(self, msg, got, needed):
718 Exception.__init__(self, msg)
721 self.servermap = None
723 class UnableToFetchCriticalDownloadDataError(Exception):
724 """I was unable to fetch some piece of critical data which is supposed to
725 be identically present in all shares."""
727 class NoServersError(Exception):
728 """Upload wasn't given any servers to work with, usually indicating a
729 network or Introducer problem."""
731 class ExistingChildError(Exception):
732 """A directory node was asked to add or replace a child that already
733 exists, and overwrite= was set to False."""
735 class NoSuchChildError(Exception):
736 """A directory node was asked to fetch a child which does not exist."""
738 class IDirectoryNode(IMutableFilesystemNode):
739 """I represent a name-to-child mapping, holding the tahoe equivalent of a
740 directory. All child names are unicode strings, and all children are some
741 sort of IFilesystemNode (either files or subdirectories).
746 The dirnode ('1') URI returned by this method can be used in
747 set_uri() on a different directory ('2') to 'mount' a reference to
748 this directory ('1') under the other ('2'). This URI is just a
749 string, so it can be passed around through email or other out-of-band
753 def get_readonly_uri():
755 The dirnode ('1') URI returned by this method can be used in
756 set_uri() on a different directory ('2') to 'mount' a reference to
757 this directory ('1') under the other ('2'). This URI is just a
758 string, so it can be passed around through email or other out-of-band
763 """I return a Deferred that fires with a dictionary mapping child
764 name (a unicode string) to (node, metadata_dict) tuples, in which
765 'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
766 is a dictionary of metadata."""
769 """I return a Deferred that fires with a boolean, True if there
770 exists a child of the given name, False if not. The child name must
771 be a unicode string."""
774 """I return a Deferred that fires with a specific named child node,
775 either an IFileNode or an IDirectoryNode. The child name must be a
776 unicode string. I raise NoSuchChildError if I do not have a child by
779 def get_metadata_for(name):
780 """I return a Deferred that fires with the metadata dictionary for a
781 specific named child node. This metadata is stored in the *edge*, not
782 in the child, so it is attached to the parent dirnode rather than the
783 child dir-or-file-node. The child name must be a unicode string. I
784 raise NoSuchChildError if I do not have a child by that name."""
786 def set_metadata_for(name, metadata):
787 """I replace any existing metadata for the named child with the new
788 metadata. The child name must be a unicode string. This metadata is
789 stored in the *edge*, not in the child, so it is attached to the
790 parent dirnode rather than the child dir-or-file-node. I return a
791 Deferred (that fires with this dirnode) when the operation is
792 complete. I raise NoSuchChildError if I do not have a child by that
795 def get_child_at_path(path):
796 """Transform a child path into an IDirectoryNode or IFileNode.
798 I perform a recursive series of 'get' operations to find the named
799 descendant node. I return a Deferred that fires with the node, or
800 errbacks with NoSuchChildError if the node could not be found.
802 The path can be either a single string (slash-separated) or a list of
803 path-name elements. All elements must be unicode strings.
806 def get_child_and_metadata_at_path(path):
807 """Transform a child path into an IDirectoryNode/IFileNode and
810 I am like get_child_at_path(), but my Deferred fires with a tuple of
811 (node, metadata). The metadata comes from the last edge. If the path
812 is empty, the metadata will be an empty dictionary.
815 def set_uri(name, child_uri, metadata=None, overwrite=True):
816 """I add a child (by URI) at the specific name. I return a Deferred
817 that fires when the operation finishes. If overwrite= is True, I will
818 replace any existing child of the same name, otherwise an existing
819 child will cause me to return ExistingChildError. The child name must
822 The child_uri could be for a file, or for a directory (either
823 read-write or read-only, using a URI that came from get_uri() ).
825 If metadata= is provided, I will use it as the metadata for the named
826 edge. This will replace any existing metadata. If metadata= is left
827 as the default value of None, I will set ['mtime'] to the current
828 time, and I will set ['ctime'] to the current time if there was not
829 already a child by this name present. This roughly matches the
830 ctime/mtime semantics of traditional filesystems.
832 If this directory node is read-only, the Deferred will errback with a
835 def set_children(entries, overwrite=True):
836 """Add multiple (name, child_uri) pairs (or (name, child_uri,
837 metadata) triples) to a directory node. Returns a Deferred that fires
838 (with None) when the operation finishes. This is equivalent to
839 calling set_uri() multiple times, but is much more efficient. All
840 child names must be unicode strings.
843 def set_node(name, child, metadata=None, overwrite=True):
844 """I add a child at the specific name. I return a Deferred that fires
845 when the operation finishes. This Deferred will fire with the child
846 node that was just added. I will replace any existing child of the
847 same name. The child name must be a unicode string. The 'child'
848 instance must be an instance providing IDirectoryNode or IFileNode.
850 If metadata= is provided, I will use it as the metadata for the named
851 edge. This will replace any existing metadata. If metadata= is left
852 as the default value of None, I will set ['mtime'] to the current
853 time, and I will set ['ctime'] to the current time if there was not
854 already a child by this name present. This roughly matches the
855 ctime/mtime semantics of traditional filesystems.
857 If this directory node is read-only, the Deferred will errback with a
860 def set_nodes(entries, overwrite=True):
861 """Add multiple (name, child_node) pairs (or (name, child_node,
862 metadata) triples) to a directory node. Returns a Deferred that fires
863 (with None) when the operation finishes. This is equivalent to
864 calling set_node() multiple times, but is much more efficient. All
865 child names must be unicode strings."""
868 def add_file(name, uploadable, metadata=None, overwrite=True):
869 """I upload a file (using the given IUploadable), then attach the
870 resulting FileNode to the directory at the given name. I set metadata
871 the same way as set_uri and set_node. The child name must be a
874 I return a Deferred that fires (with the IFileNode of the uploaded
875 file) when the operation completes."""
878 """I remove the child at the specific name. I return a Deferred that
879 fires when the operation finishes. The child name must be a unicode
880 string. I raise NoSuchChildError if I do not have a child by that
883 def create_empty_directory(name, overwrite=True):
884 """I create and attach an empty directory at the given name. The
885 child name must be a unicode string. I return a Deferred that fires
886 when the operation finishes."""
888 def move_child_to(current_child_name, new_parent, new_child_name=None,
890 """I take one of my children and move them to a new parent. The child
891 is referenced by name. On the new parent, the child will live under
892 'new_child_name', which defaults to 'current_child_name'. TODO: what
893 should we do about metadata? I return a Deferred that fires when the
894 operation finishes. The child name must be a unicode string. I raise
895 NoSuchChildError if I do not have a child by that name."""
897 def build_manifest():
898 """I generate a table of everything reachable from this directory.
899 I also compute deep-stats as described below.
901 I return a Monitor. The Monitor's results will be a dictionary with
904 res['manifest']: a list of (path, cap) tuples for all nodes
905 (directories and files) reachable from this one.
906 'path' will be a tuple of unicode strings. The
907 origin dirnode will be represented by an empty path
909 res['verifycaps']: a list of (printable) verifycap strings, one for
910 each reachable non-LIT node. This is a set:
911 it will contain no duplicates.
912 res['storage-index']: a list of (base32) storage index strings,
913 one for each reachable non-LIT node. This is
914 a set: it will contain no duplicates.
915 res['stats']: a dictionary, the same that is generated by
916 start_deep_stats() below.
918 The Monitor will also have an .origin_si attribute with the (binary)
919 storage index of the starting point.
922 def start_deep_stats():
923 """Return a Monitor, examining all nodes (directories and files)
924 reachable from this one. The Monitor's results will be a dictionary
925 with the following keys::
927 count-immutable-files: count of how many CHK files are in the set
928 count-mutable-files: same, for mutable files (does not include
930 count-literal-files: same, for LIT files
931 count-files: sum of the above three
933 count-directories: count of directories
935 size-immutable-files: total bytes for all CHK files in the set
936 size-mutable-files (TODO): same, for current version of all mutable
937 files, does not include directories
938 size-literal-files: same, for LIT files
939 size-directories: size of mutable files used by directories
941 largest-directory: number of bytes in the largest directory
942 largest-directory-children: number of children in the largest
944 largest-immutable-file: number of bytes in the largest CHK file
946 size-mutable-files is not yet implemented, because it would involve
947 even more queries than deep_stats does.
949 The Monitor will also have an .origin_si attribute with the (binary)
950 storage index of the starting point.
952 This operation will visit every directory node underneath this one,
953 and can take a long time to run. On a typical workstation with good
954 bandwidth, this can examine roughly 15 directories per second (and
955 takes several minutes of 100% CPU for ~1700 directories).
958 class ICodecEncoder(Interface):
959 def set_params(data_size, required_shares, max_shares):
960 """Set up the parameters of this encoder.
962 This prepares the encoder to perform an operation that converts a
963 single block of data into a number of shares, such that a future
964 ICodecDecoder can use a subset of these shares to recover the
965 original data. This operation is invoked by calling encode(). Once
966 the encoding parameters are set up, the encode operation can be
967 invoked multiple times.
969 set_params() prepares the encoder to accept blocks of input data that
970 are exactly 'data_size' bytes in length. The encoder will be prepared
971 to produce 'max_shares' shares for each encode() operation (although
972 see the 'desired_share_ids' to use less CPU). The encoding math will
973 be chosen such that the decoder can get by with as few as
974 'required_shares' of these shares and still reproduce the original
975 data. For example, set_params(1000, 5, 5) offers no redundancy at
976 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
978 Numerical Restrictions: 'data_size' is required to be an integral
979 multiple of 'required_shares'. In general, the caller should choose
980 required_shares and max_shares based upon their reliability
981 requirements and the number of peers available (the total storage
982 space used is roughly equal to max_shares*data_size/required_shares),
983 then choose data_size to achieve the memory footprint desired (larger
984 data_size means more efficient operation, smaller data_size means
985 smaller memory footprint).
987 In addition, 'max_shares' must be equal to or greater than
988 'required_shares'. Of course, setting them to be equal causes
989 encode() to degenerate into a particularly slow form of the 'split'
992 See encode() for more details about how these parameters are used.
994 set_params() must be called before any other ICodecEncoder methods
999 """Return the 3-tuple of data_size, required_shares, max_shares"""
1001 def get_encoder_type():
1002 """Return a short string that describes the type of this encoder.
1004 There is required to be a global table of encoder classes. This method
1005 returns an index into this table; the value at this index is an
1006 encoder class, and this encoder is an instance of that class.
1009 def get_block_size():
1010 """Return the length of the shares that encode() will produce.
1013 def encode_proposal(data, desired_share_ids=None):
1014 """Encode some data.
1016 'data' must be a string (or other buffer object), and len(data) must
1017 be equal to the 'data_size' value passed earlier to set_params().
1019 This will return a Deferred that will fire with two lists. The first
1020 is a list of shares, each of which is a string (or other buffer
1021 object) such that len(share) is the same as what get_share_size()
1022 returned earlier. The second is a list of shareids, in which each is
1023 an integer. The lengths of the two lists will always be equal to each
1024 other. The user should take care to keep each share closely
1025 associated with its shareid, as one is useless without the other.
1027 The length of this output list will normally be the same as the value
1028 provided to the 'max_shares' parameter of set_params(). This may be
1029 different if 'desired_share_ids' is provided.
1031 'desired_share_ids', if provided, is required to be a sequence of
1032 ints, each of which is required to be >= 0 and < max_shares. If not
1033 provided, encode() will produce 'max_shares' shares, as if
1034 'desired_share_ids' were set to range(max_shares). You might use this
1035 if you initially thought you were going to use 10 peers, started
1036 encoding, and then two of the peers dropped out: you could use
1037 desired_share_ids= to skip the work (both memory and CPU) of
1038 producing shares for the peers which are no longer available.
1042 def encode(inshares, desired_share_ids=None):
1043 """Encode some data. This may be called multiple times. Each call is
1046 inshares is a sequence of length required_shares, containing buffers
1047 (i.e. strings), where each buffer contains the next contiguous
1048 non-overlapping segment of the input data. Each buffer is required to
1049 be the same length, and the sum of the lengths of the buffers is
1050 required to be exactly the data_size promised by set_params(). (This
1051 implies that the data has to be padded before being passed to
1052 encode(), unless of course it already happens to be an even multiple
1053 of required_shares in length.)
1055 ALSO: the requirement to break up your data into 'required_shares'
1056 chunks before calling encode() feels a bit surprising, at least from
1057 the point of view of a user who doesn't know how FEC works. It feels
1058 like an implementation detail that has leaked outside the
1059 abstraction barrier. Can you imagine a use case in which the data to
1060 be encoded might already be available in pre-segmented chunks, such
1061 that it is faster or less work to make encode() take a list rather
1062 than splitting a single string?
1064 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1065 is supposed to *produce* shares, so what it *accepts* should be
1066 something other than shares. Other places in this interface use the
1067 word 'data' for that-which-is-not-shares.. maybe we should use that
1070 'desired_share_ids', if provided, is required to be a sequence of
1071 ints, each of which is required to be >= 0 and < max_shares. If not
1072 provided, encode() will produce 'max_shares' shares, as if
1073 'desired_share_ids' were set to range(max_shares). You might use this
1074 if you initially thought you were going to use 10 peers, started
1075 encoding, and then two of the peers dropped out: you could use
1076 desired_share_ids= to skip the work (both memory and CPU) of
1077 producing shares for the peers which are no longer available.
1079 For each call, encode() will return a Deferred that fires with two
1080 lists, one containing shares and the other containing the shareids.
1081 The get_share_size() method can be used to determine the length of
1082 the share strings returned by encode(). Each shareid is a small
1083 integer, exactly as passed into 'desired_share_ids' (or
1084 range(max_shares), if desired_share_ids was not provided).
1086 The shares and their corresponding shareids are required to be kept
1087 together during storage and retrieval. Specifically, the share data is
1088 useless by itself: the decoder needs to be told which share is which
1089 by providing it with both the shareid and the actual share data.
1091 This function will allocate an amount of memory roughly equal to::
1093 (max_shares - required_shares) * get_share_size()
1095 When combined with the memory that the caller must allocate to
1096 provide the input data, this leads to a memory footprint roughly
1097 equal to the size of the resulting encoded shares (i.e. the expansion
1098 factor times the size of the input segment).
1103 # returning a list of (shareidN,shareN) tuples instead of a pair of
1104 # lists (shareids..,shares..). Brian thought the tuples would
1105 # encourage users to keep the share and shareid together throughout
1106 # later processing, Zooko pointed out that the code to iterate
1107 # through two lists is not really more complicated than using a list
1108 # of tuples and there's also a performance improvement
1110 # having 'data_size' not required to be an integral multiple of
1111 # 'required_shares'. Doing this would require encode() to perform
1112 # padding internally, and we'd prefer to have any padding be done
1113 # explicitly by the caller. Yes, it is an abstraction leak, but
1114 # hopefully not an onerous one.
1117 class ICodecDecoder(Interface):
1118 def set_params(data_size, required_shares, max_shares):
1119 """Set the params. They have to be exactly the same ones that were used for encoding. """
1121 def get_needed_shares():
1122 """Return the number of shares needed to reconstruct the data.
1123 set_params() is required to be called before this."""
1125 def decode(some_shares, their_shareids):
1126 """Decode a partial list of shares into data.
1128 'some_shares' is required to be a sequence of buffers of sharedata, a
1129 subset of the shares returned by ICodecEncode.encode(). Each share is
1130 required to be of the same length. The i'th element of their_shareids
1131 is required to be the shareid of the i'th buffer in some_shares.
1133 This returns a Deferred which fires with a sequence of buffers. This
1134 sequence will contain all of the segments of the original data, in
1135 order. The sum of the lengths of all of the buffers will be the
1136 'data_size' value passed into the original ICodecEncode.set_params()
1137 call. To get back the single original input block of data, use
1138 ''.join(output_buffers), or you may wish to simply write them in
1139 order to an output file.
1141 Note that some of the elements in the result sequence may be
1142 references to the elements of the some_shares input sequence. In
1143 particular, this means that if those share objects are mutable (e.g.
1144 arrays) and if they are changed, then both the input (the
1145 'some_shares' parameter) and the output (the value given when the
1146 deferred is triggered) will change.
1148 The length of 'some_shares' is required to be exactly the value of
1149 'required_shares' passed into the original ICodecEncode.set_params()
1153 class IEncoder(Interface):
1154 """I take an object that provides IEncryptedUploadable, which provides
1155 encrypted data, and a list of shareholders. I then encode, hash, and
1156 deliver shares to those shareholders. I will compute all the necessary
1157 Merkle hash trees that are necessary to validate the crypttext that
1158 eventually comes back from the shareholders. I provide the URI Extension
1159 Block Hash, and the encoding parameters, both of which must be included
1162 I do not choose shareholders, that is left to the IUploader. I must be
1163 given a dict of RemoteReferences to storage buckets that are ready and
1164 willing to receive data.
1168 """Specify the number of bytes that will be encoded. This must be
1169 peformed before get_serialized_params() can be called.
1171 def set_params(params):
1172 """Override the default encoding parameters. 'params' is a tuple of
1173 (k,d,n), where 'k' is the number of required shares, 'd' is the
1174 shares_of_happiness, and 'n' is the total number of shares that will
1177 Encoding parameters can be set in three ways. 1: The Encoder class
1178 provides defaults (3/7/10). 2: the Encoder can be constructed with
1179 an 'options' dictionary, in which the
1180 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1181 set_params((k,d,n)) can be called.
1183 If you intend to use set_params(), you must call it before
1184 get_share_size or get_param are called.
1187 def set_encrypted_uploadable(u):
1188 """Provide a source of encrypted upload data. 'u' must implement
1189 IEncryptedUploadable.
1191 When this is called, the IEncryptedUploadable will be queried for its
1192 length and the storage_index that should be used.
1194 This returns a Deferred that fires with this Encoder instance.
1196 This must be performed before start() can be called.
1199 def get_param(name):
1200 """Return an encoding parameter, by name.
1202 'storage_index': return a string with the (16-byte truncated SHA-256
1203 hash) storage index to which these shares should be
1206 'share_counts': return a tuple describing how many shares are used:
1207 (needed_shares, shares_of_happiness, total_shares)
1209 'num_segments': return an int with the number of segments that
1212 'segment_size': return an int with the size of each segment.
1214 'block_size': return the size of the individual blocks that will
1215 be delivered to a shareholder's put_block() method. By
1216 knowing this, the shareholder will be able to keep all
1217 blocks in a single file and still provide random access
1218 when reading them. # TODO: can we avoid exposing this?
1220 'share_size': an int with the size of the data that will be stored
1221 on each shareholder. This is aggregate amount of data
1222 that will be sent to the shareholder, summed over all
1223 the put_block() calls I will ever make. It is useful to
1224 determine this size before asking potential
1225 shareholders whether they will grant a lease or not,
1226 since their answers will depend upon how much space we
1227 need. TODO: this might also include some amount of
1228 overhead, like the size of all the hashes. We need to
1229 decide whether this is useful or not.
1231 'serialized_params': a string with a concise description of the
1232 codec name and its parameters. This may be passed
1233 into the IUploadable to let it make sure that
1234 the same file encoded with different parameters
1235 will result in different storage indexes.
1237 Once this is called, set_size() and set_params() may not be called.
1240 def set_shareholders(shareholders):
1241 """Tell the encoder where to put the encoded shares. 'shareholders'
1242 must be a dictionary that maps share number (an integer ranging from
1243 0 to n-1) to an instance that provides IStorageBucketWriter. This
1244 must be performed before start() can be called."""
1247 """Begin the encode/upload process. This involves reading encrypted
1248 data from the IEncryptedUploadable, encoding it, uploading the shares
1249 to the shareholders, then sending the hash trees.
1251 set_encrypted_uploadable() and set_shareholders() must be called
1252 before this can be invoked.
1254 This returns a Deferred that fires with a verify cap when the upload process is
1255 complete. The verifycap, plus the encryption key, is sufficient to construct the read
1259 class IDecoder(Interface):
1260 """I take a list of shareholders and some setup information, then
1261 download, validate, decode, and decrypt data from them, writing the
1262 results to an output file.
1264 I do not locate the shareholders, that is left to the IDownloader. I must
1265 be given a dict of RemoteReferences to storage buckets that are ready to
1270 """I take a file-like object (providing write and close) to which all
1271 the plaintext data will be written.
1273 TODO: producer/consumer . Maybe write() should return a Deferred that
1274 indicates when it will accept more data? But probably having the
1275 IDecoder be a producer is easier to glue to IConsumer pieces.
1278 def set_shareholders(shareholders):
1279 """I take a dictionary that maps share identifiers (small integers)
1280 to RemoteReferences that provide RIBucketReader. This must be called
1284 """I start the download. This process involves retrieving data and
1285 hash chains from the shareholders, using the hashes to validate the
1286 data, decoding the shares into segments, decrypting the segments,
1287 then writing the resulting plaintext to the output file.
1289 I return a Deferred that will fire (with self) when the download is
1293 class IDownloadTarget(Interface):
1294 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1295 # will register itself as a producer. This allows the target to invoke
1296 # downloader.pauseProducing, resumeProducing, and stopProducing.
1298 """Called before any calls to write() or close(). If an error
1299 occurs before any data is available, fail() may be called without
1300 a previous call to open().
1302 'size' is the length of the file being downloaded, in bytes."""
1305 """Output some data to the target."""
1307 """Inform the target that there is no more data to be written."""
1309 """fail() is called to indicate that the download has failed. 'why'
1310 is a Failure object indicating what went wrong. No further methods
1311 will be invoked on the IDownloadTarget after fail()."""
1312 def register_canceller(cb):
1313 """The CiphertextDownloader uses this to register a no-argument function
1314 that the target can call to cancel the download. Once this canceller
1315 is invoked, no further calls to write() or close() will be made."""
1317 """When the CiphertextDownloader is done, this finish() function will be
1318 called. Whatever it returns will be returned to the invoker of
1319 Downloader.download.
1321 # The following methods are just because that target might be a repairer.DownUpConnector,
1322 # and just because the current CHKUpload object expects to find the storage index and
1323 # encoding parameters in its Uploadable.
1324 def set_storageindex(storageindex):
1325 """ Set the storage index. """
1326 def set_encodingparams(encodingparams):
1327 """ Set the encoding parameters. """
1329 class IDownloader(Interface):
1330 def download(uri, target):
1331 """Perform a CHK download, sending the data to the given target.
1332 'target' must provide IDownloadTarget.
1334 Returns a Deferred that fires (with the results of target.finish)
1335 when the download is finished, or errbacks if something went wrong."""
1337 class IEncryptedUploadable(Interface):
1338 def set_upload_status(upload_status):
1339 """Provide an IUploadStatus object that should be filled with status
1340 information. The IEncryptedUploadable is responsible for setting
1341 key-determination progress ('chk'), size, storage_index, and
1342 ciphertext-fetch progress. It may delegate some of this
1343 responsibility to others, in particular to the IUploadable."""
1346 """This behaves just like IUploadable.get_size()."""
1348 def get_all_encoding_parameters():
1349 """Return a Deferred that fires with a tuple of
1350 (k,happy,n,segment_size). The segment_size will be used as-is, and
1351 must match the following constraints: it must be a multiple of k, and
1352 it shouldn't be unreasonably larger than the file size (if
1353 segment_size is larger than filesize, the difference must be stored
1356 This usually passes through to the IUploadable method of the same
1359 The encoder strictly obeys the values returned by this method. To
1360 make an upload use non-default encoding parameters, you must arrange
1361 to control the values that this method returns.
1364 def get_storage_index():
1365 """Return a Deferred that fires with a 16-byte storage index.
1368 def read_encrypted(length, hash_only):
1369 """This behaves just like IUploadable.read(), but returns crypttext
1370 instead of plaintext. If hash_only is True, then this discards the
1371 data (and returns an empty list); this improves efficiency when
1372 resuming an interrupted upload (where we need to compute the
1373 plaintext hashes, but don't need the redundant encrypted data)."""
1375 def get_plaintext_hashtree_leaves(first, last, num_segments):
1376 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the plaintext
1377 segments, i.e. get the tagged hashes of the given segments. The
1378 segment size is expected to be generated by the IEncryptedUploadable
1379 before any plaintext is read or ciphertext produced, so that the
1380 segment hashes can be generated with only a single pass.
1382 This returns a Deferred which fires with a sequence of hashes, using:
1384 tuple(segment_hashes[first:last])
1386 'num_segments' is used to assert that the number of segments that the
1387 IEncryptedUploadable handled matches the number of segments that the
1388 encoder was expecting.
1390 This method must not be called until the final byte has been read
1391 from read_encrypted(). Once this method is called, read_encrypted()
1392 can never be called again.
1395 def get_plaintext_hash():
1396 """OBSOLETE; Get the hash of the whole plaintext.
1398 This returns a Deferred which fires with a tagged SHA-256 hash of the
1399 whole plaintext, obtained from hashutil.plaintext_hash(data).
1403 """Just like IUploadable.close()."""
1405 class IUploadable(Interface):
1406 def set_upload_status(upload_status):
1407 """Provide an IUploadStatus object that should be filled with status
1408 information. The IUploadable is responsible for setting
1409 key-determination progress ('chk')."""
1411 def set_default_encoding_parameters(params):
1412 """Set the default encoding parameters, which must be a dict mapping
1413 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1414 'max_segment_size'. These might have an influence on the final
1415 encoding parameters returned by get_all_encoding_parameters(), if the
1416 Uploadable doesn't have more specific preferences.
1418 This call is optional: if it is not used, the Uploadable will use
1419 some built-in defaults. If used, this method must be called before
1420 any other IUploadable methods to have any effect.
1424 """Return a Deferred that will fire with the length of the data to be
1425 uploaded, in bytes. This will be called before the data is actually
1426 used, to compute encoding parameters.
1429 def get_all_encoding_parameters():
1430 """Return a Deferred that fires with a tuple of
1431 (k,happy,n,segment_size). The segment_size will be used as-is, and
1432 must match the following constraints: it must be a multiple of k, and
1433 it shouldn't be unreasonably larger than the file size (if
1434 segment_size is larger than filesize, the difference must be stored
1437 The relative values of k and n allow some IUploadables to request
1438 better redundancy than others (in exchange for consuming more space
1441 Larger values of segment_size reduce hash overhead, while smaller
1442 values reduce memory footprint and cause data to be delivered in
1443 smaller pieces (which may provide a smoother and more predictable
1444 download experience).
1446 The encoder strictly obeys the values returned by this method. To
1447 make an upload use non-default encoding parameters, you must arrange
1448 to control the values that this method returns. One way to influence
1449 them may be to call set_encoding_parameters() before calling
1450 get_all_encoding_parameters().
1453 def get_encryption_key():
1454 """Return a Deferred that fires with a 16-byte AES key. This key will
1455 be used to encrypt the data. The key will also be hashed to derive
1458 Uploadables which want to achieve convergence should hash their file
1459 contents and the serialized_encoding_parameters to form the key
1460 (which of course requires a full pass over the data). Uploadables can
1461 use the upload.ConvergentUploadMixin class to achieve this
1464 Uploadables which do not care about convergence (or do not wish to
1465 make multiple passes over the data) can simply return a
1466 strongly-random 16 byte string.
1468 get_encryption_key() may be called multiple times: the IUploadable is
1469 required to return the same value each time.
1473 """Return a Deferred that fires with a list of strings (perhaps with
1474 only a single element) which, when concatenated together, contain the
1475 next 'length' bytes of data. If EOF is near, this may provide fewer
1476 than 'length' bytes. The total number of bytes provided by read()
1477 before it signals EOF must equal the size provided by get_size().
1479 If the data must be acquired through multiple internal read
1480 operations, returning a list instead of a single string may help to
1481 reduce string copies.
1483 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1484 so a 10kB file means length=3kB, 100kB file means length=30kB,
1485 and >=1MB file means length=300kB.
1487 This method provides for a single full pass through the data. Later
1488 use cases may desire multiple passes or access to only parts of the
1489 data (such as a mutable file making small edits-in-place). This API
1490 will be expanded once those use cases are better understood.
1494 """The upload is finished, and whatever filehandle was in use may be
1497 class IUploadResults(Interface):
1498 """I am returned by upload() methods. I contain a number of public
1499 attributes which can be read to determine the results of the upload. Some
1500 of these are functional, some are timing information. All of these may be
1503 .file_size : the size of the file, in bytes
1504 .uri : the CHK read-cap for the file
1505 .ciphertext_fetched : how many bytes were fetched by the helper
1506 .sharemap: dict mapping share identifier to set of serverids
1507 (binary strings). This indicates which servers were given
1508 which shares. For immutable files, the shareid is an
1509 integer (the share number, from 0 to N-1). For mutable
1510 files, it is a string of the form 'seq%d-%s-sh%d',
1511 containing the sequence number, the roothash, and the
1513 .servermap : dict mapping server peerid to a set of share numbers
1514 .timings : dict of timing information, mapping name to seconds (float)
1515 total : total upload time, start to finish
1516 storage_index : time to compute the storage index
1517 peer_selection : time to decide which peers will be used
1518 contacting_helper : initial helper query to upload/no-upload decision
1519 existence_check : helper pre-upload existence check
1520 helper_total : initial helper query to helper finished pushing
1521 cumulative_fetch : helper waiting for ciphertext requests
1522 total_fetch : helper start to last ciphertext response
1523 cumulative_encoding : just time spent in zfec
1524 cumulative_sending : just time spent waiting for storage servers
1525 hashes_and_close : last segment push to shareholder close
1526 total_encode_and_push : first encode to shareholder close
1530 class IDownloadResults(Interface):
1531 """I am created internally by download() methods. I contain a number of
1532 public attributes which contain details about the download process.::
1534 .file_size : the size of the file, in bytes
1535 .servers_used : set of server peerids that were used during download
1536 .server_problems : dict mapping server peerid to a problem string. Only
1537 servers that had problems (bad hashes, disconnects) are
1539 .servermap : dict mapping server peerid to a set of share numbers. Only
1540 servers that had any shares are listed here.
1541 .timings : dict of timing information, mapping name to seconds (float)
1542 peer_selection : time to ask servers about shares
1543 servers_peer_selection : dict of peerid to DYHB-query time
1544 uri_extension : time to fetch a copy of the URI extension block
1545 hashtrees : time to fetch the hash trees
1546 segments : time to fetch, decode, and deliver segments
1547 cumulative_fetch : time spent waiting for storage servers
1548 cumulative_decode : just time spent in zfec
1549 cumulative_decrypt : just time spent in decryption
1550 total : total download time, start to finish
1551 fetch_per_server : dict of peerid to list of per-segment fetch times
1555 class IUploader(Interface):
1556 def upload(uploadable):
1557 """Upload the file. 'uploadable' must impement IUploadable. This
1558 returns a Deferred which fires with an UploadResults instance, from
1559 which the URI of the file can be obtained as results.uri ."""
1561 def upload_ssk(write_capability, new_version, uploadable):
1562 """TODO: how should this work?"""
1564 class ICheckable(Interface):
1565 def check(monitor, verify=False, add_lease=False):
1566 """Check upon my health, optionally repairing any problems.
1568 This returns a Deferred that fires with an instance that provides
1569 ICheckResults, or None if the object is non-distributed (i.e. LIT
1572 The monitor will be checked periodically to see if the operation has
1573 been cancelled. If so, no new queries will be sent, and the Deferred
1574 will fire (with a OperationCancelledError) immediately.
1576 Filenodes and dirnodes (which provide IFilesystemNode) are also
1577 checkable. Instances that represent verifier-caps will be checkable
1578 but not downloadable. Some objects (like LIT files) do not actually
1579 live in the grid, and their checkers return None (non-distributed
1580 files are always healthy).
1582 If verify=False, a relatively lightweight check will be performed: I
1583 will ask all servers if they have a share for me, and I will believe
1584 whatever they say. If there are at least N distinct shares on the
1585 grid, my results will indicate r.is_healthy()==True. This requires a
1586 roundtrip to each server, but does not transfer very much data, so
1587 the network bandwidth is fairly low.
1589 If verify=True, a more resource-intensive check will be performed:
1590 every share will be downloaded, and the hashes will be validated on
1591 every bit. I will ignore any shares that failed their hash checks. If
1592 there are at least N distinct valid shares on the grid, my results
1593 will indicate r.is_healthy()==True. This requires N/k times as much
1594 download bandwidth (and server disk IO) as a regular download. If a
1595 storage server is holding a corrupt share, or is experiencing memory
1596 failures during retrieval, or is malicious or buggy, then
1597 verification will detect the problem, but checking will not.
1599 If add_lease=True, I will ensure that an up-to-date lease is present
1600 on each share. The lease secrets will be derived from by node secret
1601 (in BASEDIR/private/secret), so either I will add a new lease to the
1602 share, or I will merely renew the lease that I already had. In a
1603 future version of the storage-server protocol (once Accounting has
1604 been implemented), there may be additional options here to define the
1605 kind of lease that is obtained (which account number to claim, etc).
1607 TODO: any problems seen during checking will be reported to the
1608 health-manager.furl, a centralized object which is responsible for
1609 figuring out why files are unhealthy so corrective action can be
1613 def check_and_repair(monitor, verify=False, add_lease=False):
1614 """Like check(), but if the file/directory is not healthy, attempt to
1617 Any non-healthy result will cause an immediate repair operation, to
1618 generate and upload new shares. After repair, the file will be as
1619 healthy as we can make it. Details about what sort of repair is done
1620 will be put in the check-and-repair results. The Deferred will not
1621 fire until the repair is complete.
1623 This returns a Deferred which fires with an instance of
1624 ICheckAndRepairResults."""
1626 class IDeepCheckable(Interface):
1627 def start_deep_check(verify=False, add_lease=False):
1628 """Check upon the health of me and everything I can reach.
1630 This is a recursive form of check(), useable only on dirnodes.
1632 I return a Monitor, with results that are an IDeepCheckResults
1635 TODO: If any of the directories I traverse are unrecoverable, the
1636 Monitor will report failure. If any of the files I check upon are
1637 unrecoverable, those problems will be reported in the
1638 IDeepCheckResults as usual, and the Monitor will not report a
1642 def start_deep_check_and_repair(verify=False, add_lease=False):
1643 """Check upon the health of me and everything I can reach. Repair
1644 anything that isn't healthy.
1646 This is a recursive form of check_and_repair(), useable only on
1649 I return a Monitor, with results that are an
1650 IDeepCheckAndRepairResults object.
1652 TODO: If any of the directories I traverse are unrecoverable, the
1653 Monitor will report failure. If any of the files I check upon are
1654 unrecoverable, those problems will be reported in the
1655 IDeepCheckResults as usual, and the Monitor will not report a
1659 class ICheckResults(Interface):
1660 """I contain the detailed results of a check/verify operation.
1663 def get_storage_index():
1664 """Return a string with the (binary) storage index."""
1665 def get_storage_index_string():
1666 """Return a string with the (printable) abbreviated storage index."""
1668 """Return the (string) URI of the object that was checked."""
1671 """Return a boolean, True if the file/dir is fully healthy, False if
1672 it is damaged in any way. Non-distributed LIT files always return
1675 def is_recoverable():
1676 """Return a boolean, True if the file/dir can be recovered, False if
1677 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1678 files always return True."""
1680 def needs_rebalancing():
1681 """Return a boolean, True if the file/dir's reliability could be
1682 improved by moving shares to new servers. Non-distributed LIT files
1683 always return False."""
1687 """Return a dictionary that describes the state of the file/dir. LIT
1688 files always return an empty dictionary. Normal files and directories return a
1689 dictionary with the following keys (note that these use binary strings rather than
1690 base32-encoded ones) (also note that for mutable files, these counts are for the 'best'
1693 count-shares-good: the number of distinct good shares that were found
1694 count-shares-needed: 'k', the number of shares required for recovery
1695 count-shares-expected: 'N', the number of total shares generated
1696 count-good-share-hosts: the number of distinct storage servers with
1697 good shares. If this number is less than
1698 count-shares-good, then some shares are
1699 doubled up, increasing the correlation of
1700 failures. This indicates that one or more
1701 shares should be moved to an otherwise unused
1702 server, if one is available.
1703 count-corrupt-shares: the number of shares with integrity failures
1704 list-corrupt-shares: a list of 'share locators', one for each share
1705 that was found to be corrupt. Each share
1706 locator is a list of (serverid, storage_index,
1708 count-incompatible-shares: the number of shares which are of a share format unknown to
1710 list-incompatible-shares: a list of 'share locators', one for each share that was found
1711 to be of an unknown format. Each share locator is a list of
1712 (serverid, storage_index, sharenum).
1713 servers-responding: list of (binary) storage server identifiers,
1714 one for each server which responded to the share
1715 query (even if they said they didn't have shares,
1716 and even if they said they did have shares but then
1717 didn't send them when asked, or dropped the
1718 connection, or returned a Failure, and even if they
1719 said they did have shares and sent incorrect ones
1721 sharemap: dict mapping share identifier to list of serverids
1722 (binary strings). This indicates which servers are holding
1723 which shares. For immutable files, the shareid is an
1724 integer (the share number, from 0 to N-1). For mutable
1725 files, it is a string of the form 'seq%d-%s-sh%d',
1726 containing the sequence number, the roothash, and the
1729 The following keys are most relevant for mutable files, but immutable
1730 files will provide sensible values too::
1732 count-wrong-shares: the number of shares for versions other than the
1733 'best' one (which is defined as being the
1734 recoverable version with the highest sequence
1735 number, then the highest roothash). These are
1736 either leftover shares from an older version
1737 (perhaps on a server that was offline when an
1738 update occurred), shares from an unrecoverable
1739 newer version, or shares from an alternate
1740 current version that results from an
1741 uncoordinated write collision. For a healthy
1742 file, this will equal 0.
1744 count-recoverable-versions: the number of recoverable versions of
1745 the file. For a healthy file, this will
1748 count-unrecoverable-versions: the number of unrecoverable versions
1749 of the file. For a healthy file, this
1755 """Return a string with a brief (one-line) summary of the results."""
1758 """Return a list of strings with more detailed results."""
1760 class ICheckAndRepairResults(Interface):
1761 """I contain the detailed results of a check/verify/repair operation.
1763 The IFilesystemNode.check()/verify()/repair() methods all return
1764 instances that provide ICheckAndRepairResults.
1767 def get_storage_index():
1768 """Return a string with the (binary) storage index."""
1769 def get_storage_index_string():
1770 """Return a string with the (printable) abbreviated storage index."""
1771 def get_repair_attempted():
1772 """Return a boolean, True if a repair was attempted."""
1773 def get_repair_successful():
1774 """Return a boolean, True if repair was attempted and the file/dir
1775 was fully healthy afterwards. False if no repair was attempted or if
1776 a repair attempt failed."""
1777 def get_pre_repair_results():
1778 """Return an ICheckResults instance that describes the state of the
1779 file/dir before any repair was attempted."""
1780 def get_post_repair_results():
1781 """Return an ICheckResults instance that describes the state of the
1782 file/dir after any repair was attempted. If no repair was attempted,
1783 the pre-repair and post-repair results will be identical."""
1786 class IDeepCheckResults(Interface):
1787 """I contain the results of a deep-check operation.
1789 This is returned by a call to ICheckable.deep_check().
1792 def get_root_storage_index_string():
1793 """Return the storage index (abbreviated human-readable string) of
1794 the first object checked."""
1796 """Return a dictionary with the following keys::
1798 count-objects-checked: count of how many objects were checked
1799 count-objects-healthy: how many of those objects were completely
1801 count-objects-unhealthy: how many were damaged in some way
1802 count-objects-unrecoverable: how many were unrecoverable
1803 count-corrupt-shares: how many shares were found to have
1804 corruption, summed over all objects
1808 def get_corrupt_shares():
1809 """Return a set of (serverid, storage_index, sharenum) for all shares
1810 that were found to be corrupt. Both serverid and storage_index are
1813 def get_all_results():
1814 """Return a dictionary mapping pathname (a tuple of strings, ready to
1815 be slash-joined) to an ICheckResults instance, one for each object
1816 that was checked."""
1818 def get_results_for_storage_index(storage_index):
1819 """Retrive the ICheckResults instance for the given (binary)
1820 storage index. Raises KeyError if there are no results for that
1824 """Return a dictionary with the same keys as
1825 IDirectoryNode.deep_stats()."""
1827 class IDeepCheckAndRepairResults(Interface):
1828 """I contain the results of a deep-check-and-repair operation.
1830 This is returned by a call to ICheckable.deep_check_and_repair().
1833 def get_root_storage_index_string():
1834 """Return the storage index (abbreviated human-readable string) of
1835 the first object checked."""
1837 """Return a dictionary with the following keys::
1839 count-objects-checked: count of how many objects were checked
1840 count-objects-healthy-pre-repair: how many of those objects were
1841 completely healthy (before any
1843 count-objects-unhealthy-pre-repair: how many were damaged in
1845 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1846 count-objects-healthy-post-repair: how many of those objects were
1847 completely healthy (after any
1849 count-objects-unhealthy-post-repair: how many were damaged in
1851 count-objects-unrecoverable-post-repair: how many were
1853 count-repairs-attempted: repairs were attempted on this many
1854 objects. The count-repairs- keys will
1855 always be provided, however unless
1856 repair=true is present, they will all
1858 count-repairs-successful: how many repairs resulted in healthy
1860 count-repairs-unsuccessful: how many repairs resulted did not
1861 results in completely healthy objects
1862 count-corrupt-shares-pre-repair: how many shares were found to
1863 have corruption, summed over all
1864 objects examined (before any
1866 count-corrupt-shares-post-repair: how many shares were found to
1867 have corruption, summed over all
1868 objects examined (after any
1873 """Return a dictionary with the same keys as
1874 IDirectoryNode.deep_stats()."""
1876 def get_corrupt_shares():
1877 """Return a set of (serverid, storage_index, sharenum) for all shares
1878 that were found to be corrupt before any repair was attempted. Both
1879 serverid and storage_index are binary.
1881 def get_remaining_corrupt_shares():
1882 """Return a set of (serverid, storage_index, sharenum) for all shares
1883 that were found to be corrupt after any repair was completed. Both
1884 serverid and storage_index are binary. These are shares that need
1885 manual inspection and probably deletion.
1887 def get_all_results():
1888 """Return a dictionary mapping pathname (a tuple of strings, ready to
1889 be slash-joined) to an ICheckAndRepairResults instance, one for each
1890 object that was checked."""
1892 def get_results_for_storage_index(storage_index):
1893 """Retrive the ICheckAndRepairResults instance for the given (binary)
1894 storage index. Raises KeyError if there are no results for that
1898 class IRepairable(Interface):
1899 def repair(check_results):
1900 """Attempt to repair the given object. Returns a Deferred that fires
1901 with a IRepairResults object.
1903 I must be called with an object that implements ICheckResults, as
1904 proof that you have actually discovered a problem with this file. I
1905 will use the data in the checker results to guide the repair process,
1906 such as which servers provided bad data and should therefore be
1907 avoided. The ICheckResults object is inside the
1908 ICheckAndRepairResults object, which is returned by the
1909 ICheckable.check() method::
1911 d = filenode.check(repair=False)
1912 def _got_results(check_and_repair_results):
1913 check_results = check_and_repair_results.get_pre_repair_results()
1914 return filenode.repair(check_results)
1915 d.addCallback(_got_results)
1919 class IRepairResults(Interface):
1920 """I contain the results of a repair operation."""
1923 class IClient(Interface):
1924 def upload(uploadable):
1925 """Upload some data into a CHK, get back the UploadResults for it.
1926 @param uploadable: something that implements IUploadable
1927 @return: a Deferred that fires with the UploadResults instance.
1928 To get the URI for this file, use results.uri .
1931 def create_mutable_file(contents=""):
1932 """Create a new mutable file with contents, get back the URI string.
1933 @param contents: the initial contents to place in the file.
1934 @return: a Deferred that fires with tne (string) SSK URI for the new
1938 def create_empty_dirnode():
1939 """Create a new dirnode, empty and unattached.
1940 @return: a Deferred that fires with the new IDirectoryNode instance.
1943 def create_node_from_uri(uri):
1944 """Create a new IFilesystemNode instance from the uri, synchronously.
1945 @param uri: a string or IURI-providing instance. This could be for a
1946 LiteralFileNode, a CHK file node, a mutable file node, or
1948 @return: an instance that provides IFilesystemNode (or more usefully one
1949 of its subclasses). File-specifying URIs will result in
1950 IFileNode or IMutableFileNode -providing instances, like
1951 FileNode, LiteralFileNode, or MutableFileNode.
1952 Directory-specifying URIs will result in
1953 IDirectoryNode-providing instances, like NewDirectoryNode.
1956 class IClientStatus(Interface):
1957 def list_all_uploads():
1958 """Return a list of uploader objects, one for each upload which
1959 currently has an object available (tracked with weakrefs). This is
1960 intended for debugging purposes."""
1961 def list_active_uploads():
1962 """Return a list of active IUploadStatus objects."""
1963 def list_recent_uploads():
1964 """Return a list of IUploadStatus objects for the most recently
1967 def list_all_downloads():
1968 """Return a list of downloader objects, one for each download which
1969 currently has an object available (tracked with weakrefs). This is
1970 intended for debugging purposes."""
1971 def list_active_downloads():
1972 """Return a list of active IDownloadStatus objects."""
1973 def list_recent_downloads():
1974 """Return a list of IDownloadStatus objects for the most recently
1975 started downloads."""
1977 class IUploadStatus(Interface):
1979 """Return a timestamp (float with seconds since epoch) indicating
1980 when the operation was started."""
1981 def get_storage_index():
1982 """Return a string with the (binary) storage index in use on this
1983 upload. Returns None if the storage index has not yet been
1986 """Return an integer with the number of bytes that will eventually
1987 be uploaded for this file. Returns None if the size is not yet known.
1990 """Return True if this upload is using a Helper, False if not."""
1992 """Return a string describing the current state of the upload
1995 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
1996 each from 0.0 to 1.0 . 'chk' describes how much progress has been
1997 made towards hashing the file to determine a CHK encryption key: if
1998 non-convergent encryption is in use, this will be trivial, otherwise
1999 the whole file must be hashed. 'ciphertext' describes how much of the
2000 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2001 uploads. 'encode_and_push' describes how much of the encode-and-push
2002 process has finished: for helper uploads this is dependent upon the
2003 helper providing progress reports. It might be reasonable to add all
2004 three numbers and report the sum to the user."""
2006 """Return True if the upload is currently active, False if not."""
2008 """Return an instance of UploadResults (which contains timing and
2009 sharemap information). Might return None if the upload is not yet
2012 """Each upload status gets a unique number: this method returns that
2013 number. This provides a handle to this particular upload, so a web
2014 page can generate a suitable hyperlink."""
2016 class IDownloadStatus(Interface):
2018 """Return a timestamp (float with seconds since epoch) indicating
2019 when the operation was started."""
2020 def get_storage_index():
2021 """Return a string with the (binary) storage index in use on this
2022 download. This may be None if there is no storage index (i.e. LIT
2025 """Return an integer with the number of bytes that will eventually be
2026 retrieved for this file. Returns None if the size is not yet known.
2029 """Return True if this download is using a Helper, False if not."""
2031 """Return a string describing the current state of the download
2034 """Returns a float (from 0.0 to 1.0) describing the amount of the
2035 download that has completed. This value will remain at 0.0 until the
2036 first byte of plaintext is pushed to the download target."""
2038 """Return True if the download is currently active, False if not."""
2040 """Each download status gets a unique number: this method returns
2041 that number. This provides a handle to this particular download, so a
2042 web page can generate a suitable hyperlink."""
2044 class IServermapUpdaterStatus(Interface):
2046 class IPublishStatus(Interface):
2048 class IRetrieveStatus(Interface):
2051 class NotCapableError(Exception):
2052 """You have tried to write to a read-only node."""
2054 class BadWriteEnablerError(Exception):
2057 class RIControlClient(RemoteInterface):
2059 def wait_for_client_connections(num_clients=int):
2060 """Do not return until we have connections to at least NUM_CLIENTS
2064 def upload_from_file_to_uri(filename=str, convergence=ChoiceOf(None, StringConstraint(2**20))):
2065 """Upload a file to the grid. This accepts a filename (which must be
2066 absolute) that points to a file on the node's local disk. The node will
2067 read the contents of this file, upload it to the grid, then return the
2068 URI at which it was uploaded. If convergence is None then a random
2069 encryption key will be used, else the plaintext will be hashed, then
2070 that hash will be mixed together with the "convergence" string to form
2075 def download_from_uri_to_file(uri=URI, filename=str):
2076 """Download a file from the grid, placing it on the node's local disk
2077 at the given filename (which must be absolute[?]). Returns the
2078 absolute filename where the file was written."""
2083 def get_memory_usage():
2084 """Return a dict describes the amount of memory currently in use. The
2085 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2086 measuring memory consupmtion in bytes."""
2087 return DictOf(str, int)
2089 def speed_test(count=int, size=int, mutable=Any()):
2090 """Write 'count' tempfiles to disk, all of the given size. Measure
2091 how long (in seconds) it takes to upload them all to the servers.
2092 Then measure how long it takes to download all of them. If 'mutable'
2093 is 'create', time creation of mutable files. If 'mutable' is
2094 'upload', then time access to the same mutable file instead of
2097 Returns a tuple of (upload_time, download_time).
2099 return (float, float)
2101 def measure_peer_response_time():
2102 """Send a short message to each connected peer, and measure the time
2103 it takes for them to respond to it. This is a rough measure of the
2104 application-level round trip time.
2106 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2109 return DictOf(Nodeid, float)
2111 UploadResults = Any() #DictOf(str, str)
2113 class RIEncryptedUploadable(RemoteInterface):
2114 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2119 def get_all_encoding_parameters():
2120 return (int, int, int, long)
2122 def read_encrypted(offset=Offset, length=ReadSize):
2129 class RICHKUploadHelper(RemoteInterface):
2130 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2134 Return a dictionary of version information.
2136 return DictOf(str, Any())
2138 def upload(reader=RIEncryptedUploadable):
2139 return UploadResults
2142 class RIHelper(RemoteInterface):
2143 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2147 Return a dictionary of version information.
2149 return DictOf(str, Any())
2151 def upload_chk(si=StorageIndex):
2152 """See if a file with a given storage index needs uploading. The
2153 helper will ask the appropriate storage servers to see if the file
2154 has already been uploaded. If so, the helper will return a set of
2155 'upload results' that includes whatever hashes are needed to build
2156 the read-cap, and perhaps a truncated sharemap.
2158 If the file has not yet been uploaded (or if it was only partially
2159 uploaded), the helper will return an empty upload-results dictionary
2160 and also an RICHKUploadHelper object that will take care of the
2161 upload process. The client should call upload() on this object and
2162 pass it a reference to an RIEncryptedUploadable object that will
2163 provide ciphertext. When the upload is finished, the upload() method
2164 will finish and return the upload results.
2166 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2169 class RIStatsProvider(RemoteInterface):
2170 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2172 Provides access to statistics and monitoring information.
2177 returns a dictionary containing 'counters' and 'stats', each a dictionary
2178 with string counter/stat name keys, and numeric values. counters are
2179 monotonically increasing measures of work done, and stats are instantaneous
2180 measures (potentially time averaged internally)
2182 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2184 class RIStatsGatherer(RemoteInterface):
2185 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2187 Provides a monitoring service for centralised collection of stats
2190 def provide(provider=RIStatsProvider, nickname=str):
2192 @param provider: a stats collector instance which should be polled
2193 periodically by the gatherer to collect stats.
2194 @param nickname: a name useful to identify the provided client
2199 class IStatsProducer(Interface):
2202 returns a dictionary, with str keys representing the names of stats
2203 to be monitored, and numeric values.
2206 class RIKeyGenerator(RemoteInterface):
2207 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2209 Provides a service offering to make RSA key pairs.
2212 def get_rsa_key_pair(key_size=int):
2214 @param key_size: the size of the signature key.
2215 @return: tuple(verifying_key, signing_key)
2217 return TupleOf(str, str)
2220 class FileTooLargeError(Exception):
2223 class IValidatedThingProxy(Interface):
2225 """ Acquire a thing and validate it. Return a deferred which is eventually fired with
2226 self if the thing is valid or errbacked if it can't be acquired or validated. """
2228 class InsufficientVersionError(Exception):
2229 def __init__(self, needed, got):
2230 self.needed = needed
2233 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,