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)."""
429 # TODO: rename to get_read_cap()
431 """Return another IURI instance, which represents a read-only form of
432 this one. If is_readonly() is True, this returns self."""
434 def get_verify_cap():
435 """Return an instance that provides IVerifierURI, which can be used
436 to check on the availability of the file or directory, without
437 providing enough capabilities to actually read or modify the
438 contents. This may return None if the file does not need checking or
439 verification (e.g. LIT URIs).
443 """Return a string of printable ASCII characters, suitable for
444 passing into init_from_string."""
446 class IVerifierURI(Interface, IURI):
447 def init_from_string(uri):
448 """Accept a string (as created by my to_string() method) and populate
449 this instance with its data. I am not normally called directly,
450 please use the module-level uri.from_string() function to convert
451 arbitrary URI strings into IURI-providing instances."""
454 """Return a string of printable ASCII characters, suitable for
455 passing into init_from_string."""
457 class IDirnodeURI(Interface):
458 """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."""
471 class IDirectoryURI(Interface):
474 class IReadonlyDirectoryURI(Interface):
477 class CapConstraintError(Exception):
478 """A constraint on a cap was violated."""
480 class MustBeDeepImmutableError(CapConstraintError):
481 """Mutable children cannot be added to an immutable directory.
482 Also, caps obtained from an immutable directory can trigger this error
483 if they are later found to refer to a mutable object and then used."""
485 class MustBeReadonlyError(CapConstraintError):
486 """Known write caps cannot be specified in a ro_uri field. Also,
487 caps obtained from a ro_uri field can trigger this error if they
488 are later found to be write caps and then used."""
490 class MustNotBeUnknownRWError(CapConstraintError):
491 """Cannot add an unknown child cap specified in a rw_uri field."""
493 # The hierarchy looks like this:
500 class IFilesystemNode(Interface):
502 """Return the strongest 'cap instance' associated with this node.
503 (writecap for writeable-mutable files/directories, readcap for
504 immutable or readonly-mutable files/directories). To convert this
505 into a string, call .to_string() on the result."""
508 """Return a readonly cap instance for this node. For immutable or
509 readonly nodes, get_cap() and get_readcap() return the same thing."""
511 def get_repair_cap():
512 """Return an IURI instance that can be used to repair the file, or
513 None if this node cannot be repaired (either because it is not
514 distributed, like a LIT file, or because the node does not represent
515 sufficient authority to create a repair-cap, like a read-only RSA
516 mutable file node [which cannot create the correct write-enablers]).
519 def get_verify_cap():
520 """Return an IVerifierURI instance that represents the
521 'verifiy/refresh capability' for this node. The holder of this
522 capability will be able to renew the lease for this node, protecting
523 it from garbage-collection. They will also be able to ask a server if
524 it holds a share for the file or directory.
528 """Return the URI string corresponding to the strongest cap associated
529 with this node. If this node is read-only, the URI will only offer
530 read-only access. If this node is read-write, the URI will offer
533 If you have read-write access to a node and wish to share merely
534 read-only access with others, use get_readonly_uri().
537 def get_write_uri(n):
538 """Return the URI string that can be used by others to get write
539 access to this node, if it is writeable. If this is a read-only node,
542 def get_readonly_uri():
543 """Return the URI string that can be used by others to get read-only
544 access to this node. The result is a read-only URI, regardless of
545 whether this node is read-only or read-write.
547 If you have merely read-only access to this node, get_readonly_uri()
548 will return the same thing as get_uri().
551 def get_storage_index():
552 """Return a string with the (binary) storage index in use on this
553 download. This may be None if there is no storage index (i.e. LIT
557 """Return True if this reference provides mutable access to the given
558 file or directory (i.e. if you can modify it), or False if not. Note
559 that even if this reference is read-only, someone else may hold a
560 read-write reference to it."""
563 """Return True if this file or directory is mutable (by *somebody*,
564 not necessarily you), False if it is is immutable. Note that a file
565 might be mutable overall, but your reference to it might be
566 read-only. On the other hand, all references to an immutable file
567 will be read-only; there are no read-write references to an immutable
572 """Return True if this is an unknown node."""
574 def is_allowed_in_immutable_directory():
575 """Return True if this node is allowed as a child of a deep-immutable
576 directory. This is true if either the node is of a known-immutable type,
577 or it is unknown and read-only.
581 """Raise any error associated with this node."""
584 """Return the length (in bytes) of the data this node represents. For
585 directory nodes, I return the size of the backing store. I return
586 synchronously and do not consult the network, so for mutable objects,
587 I will return the most recently observed size for the object, or None
588 if I don't remember a size. Use get_current_size, which returns a
589 Deferred, if you want more up-to-date information."""
591 def get_current_size():
592 """I return a Deferred that fires with the length (in bytes) of the
593 data this node represents.
596 class IFileNode(IFilesystemNode):
597 """I am a node which represents a file: a sequence of bytes. I am not a
598 container, like IDirectoryNode."""
600 class IImmutableFileNode(IFileNode):
601 def read(consumer, offset=0, size=None):
602 """Download a portion (possibly all) of the file's contents, making
603 them available to the given IConsumer. Return a Deferred that fires
604 (with the consumer) when the consumer is unregistered (either because
605 the last byte has been given to it, or because the consumer threw an
606 exception during write(), possibly because it no longer wants to
607 receive data). The portion downloaded will start at 'offset' and
608 contain 'size' bytes (or the remainder of the file if size==None).
610 The consumer will be used in non-streaming mode: an IPullProducer
611 will be attached to it.
613 The consumer will not receive data right away: several network trips
614 must occur first. The order of events will be::
616 consumer.registerProducer(p, streaming)
617 (if streaming == False)::
618 consumer does p.resumeProducing()
620 consumer does p.resumeProducing()
621 consumer.write(data).. (repeat until all data is written)
622 consumer.unregisterProducer()
623 deferred.callback(consumer)
625 If a download error occurs, or an exception is raised by
626 consumer.registerProducer() or consumer.write(), I will call
627 consumer.unregisterProducer() and then deliver the exception via
628 deferred.errback(). To cancel the download, the consumer should call
629 p.stopProducing(), which will result in an exception being delivered
630 via deferred.errback().
632 See src/allmydata/util/consumer.py for an example of a simple
633 download-to-memory consumer.
636 class IMutableFileNode(IFileNode):
637 """I provide access to a 'mutable file', which retains its identity
638 regardless of what contents are put in it.
640 The consistency-vs-availability problem means that there might be
641 multiple versions of a file present in the grid, some of which might be
642 unrecoverable (i.e. have fewer than 'k' shares). These versions are
643 loosely ordered: each has a sequence number and a hash, and any version
644 with seqnum=N was uploaded by a node which has seen at least one version
647 The 'servermap' (an instance of IMutableFileServerMap) is used to
648 describe the versions that are known to be present in the grid, and which
649 servers are hosting their shares. It is used to represent the 'state of
650 the world', and is used for this purpose by my test-and-set operations.
651 Downloading the contents of the mutable file will also return a
652 servermap. Uploading a new version into the mutable file requires a
653 servermap as input, and the semantics of the replace operation is
654 'replace the file with my new version if it looks like nobody else has
655 changed the file since my previous download'. Because the file is
656 distributed, this is not a perfect test-and-set operation, but it will do
657 its best. If the replace process sees evidence of a simultaneous write,
658 it will signal an UncoordinatedWriteError, so that the caller can take
662 Most readers will want to use the 'best' current version of the file, and
663 should use my 'download_best_version()' method.
665 To unconditionally replace the file, callers should use overwrite(). This
666 is the mode that user-visible mutable files will probably use.
668 To apply some delta to the file, call modify() with a callable modifier
669 function that can apply the modification that you want to make. This is
670 the mode that dirnodes will use, since most directory modification
671 operations can be expressed in terms of deltas to the directory state.
674 Three methods are available for users who need to perform more complex
675 operations. The first is get_servermap(), which returns an up-to-date
676 servermap using a specified mode. The second is download_version(), which
677 downloads a specific version (not necessarily the 'best' one). The third
678 is 'upload', which accepts new contents and a servermap (which must have
679 been updated with MODE_WRITE). The upload method will attempt to apply
680 the new contents as long as no other node has modified the file since the
681 servermap was updated. This might be useful to a caller who wants to
682 merge multiple versions into a single new one.
684 Note that each time the servermap is updated, a specific 'mode' is used,
685 which determines how many peers are queried. To use a servermap for my
686 replace() method, that servermap must have been updated in MODE_WRITE.
687 These modes are defined in allmydata.mutable.common, and consist of
688 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
689 allmydata/mutable/servermap.py for details about the differences.
691 Mutable files are currently limited in size (about 3.5MB max) and can
692 only be retrieved and updated all-at-once, as a single big string. Future
693 versions of our mutable files will remove this restriction.
696 def download_best_version():
697 """Download the 'best' available version of the file, meaning one of
698 the recoverable versions with the highest sequence number. If no
699 uncoordinated writes have occurred, and if enough shares are
700 available, then this will be the most recent version that has been
703 I update an internal servermap with MODE_READ, determine which
704 version of the file is indicated by
705 servermap.best_recoverable_version(), and return a Deferred that
706 fires with its contents. If no version is recoverable, the Deferred
707 will errback with UnrecoverableFileError.
710 def get_size_of_best_version():
711 """Find the size of the version that would be downloaded with
712 download_best_version(), without actually downloading the whole file.
714 I return a Deferred that fires with an integer.
717 def overwrite(new_contents):
718 """Unconditionally replace the contents of the mutable file with new
719 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
720 is only appropriate to use when the new contents of the file are
721 completely unrelated to the old ones, and you do not care about other
724 I return a Deferred that fires (with a PublishStatus object) when the
725 update has completed.
728 def modify(modifier_cb):
729 """Modify the contents of the file, by downloading the current
730 version, applying the modifier function (or bound method), then
731 uploading the new version. I return a Deferred that fires (with a
732 PublishStatus object) when the update is complete.
734 The modifier callable will be given three arguments: a string (with
735 the old contents), a 'first_time' boolean, and a servermap. As with
736 download_best_version(), the old contents will be from the best
737 recoverable version, but the modifier can use the servermap to make
738 other decisions (such as refusing to apply the delta if there are
739 multiple parallel versions, or if there is evidence of a newer
740 unrecoverable version). 'first_time' will be True the first time the
741 modifier is called, and False on any subsequent calls.
743 The callable should return a string with the new contents. The
744 callable must be prepared to be called multiple times, and must
745 examine the input string to see if the change that it wants to make
746 is already present in the old version. If it does not need to make
747 any changes, it can either return None, or return its input string.
749 If the modifier raises an exception, it will be returned in the
754 def get_servermap(mode):
755 """Return a Deferred that fires with an IMutableFileServerMap
756 instance, updated using the given mode.
759 def download_version(servermap, version):
760 """Download a specific version of the file, using the servermap
761 as a guide to where the shares are located.
763 I return a Deferred that fires with the requested contents, or
764 errbacks with UnrecoverableFileError. Note that a servermap which was
765 updated with MODE_ANYTHING or MODE_READ may not know about shares for
766 all versions (those modes stop querying servers as soon as they can
767 fulfil their goals), so you may want to use MODE_CHECK (which checks
768 everything) to get increased visibility.
771 def upload(new_contents, servermap):
772 """Replace the contents of the file with new ones. This requires a
773 servermap that was previously updated with MODE_WRITE.
775 I attempt to provide test-and-set semantics, in that I will avoid
776 modifying any share that is different than the version I saw in the
777 servermap. However, if another node is writing to the file at the
778 same time as me, I may manage to update some shares while they update
779 others. If I see any evidence of this, I will signal
780 UncoordinatedWriteError, and the file will be left in an inconsistent
781 state (possibly the version you provided, possibly the old version,
782 possibly somebody else's version, and possibly a mix of shares from
785 The recommended response to UncoordinatedWriteError is to either
786 return it to the caller (since they failed to coordinate their
787 writes), or to attempt some sort of recovery. It may be sufficient to
788 wait a random interval (with exponential backoff) and repeat your
789 operation. If I do not signal UncoordinatedWriteError, then I was
790 able to write the new version without incident.
792 I return a Deferred that fires (with a PublishStatus object) when the
793 publish has completed. I will update the servermap in-place with the
794 location of all new shares.
798 """Return this filenode's writekey, or None if the node does not have
799 write-capability. This may be used to assist with data structures
800 that need to make certain data available only to writers, such as the
801 read-write child caps in dirnodes. The recommended process is to have
802 reader-visible data be submitted to the filenode in the clear (where
803 it will be encrypted by the filenode using the readkey), but encrypt
804 writer-visible data using this writekey.
807 class NotEnoughSharesError(Exception):
808 """Download was unable to get enough shares"""
810 class NoSharesError(Exception):
811 """Download was unable to get any shares at all."""
813 class UploadUnhappinessError(Exception):
814 """Upload was unable to satisfy 'servers_of_happiness'"""
816 class UnableToFetchCriticalDownloadDataError(Exception):
817 """I was unable to fetch some piece of critical data which is supposed to
818 be identically present in all shares."""
820 class NoServersError(Exception):
821 """Upload wasn't given any servers to work with, usually indicating a
822 network or Introducer problem."""
824 class ExistingChildError(Exception):
825 """A directory node was asked to add or replace a child that already
826 exists, and overwrite= was set to False."""
828 class NoSuchChildError(Exception):
829 """A directory node was asked to fetch a child which does not exist."""
831 class IDirectoryNode(IFilesystemNode):
832 """I represent a filesystem node that is a container, with a
833 name-to-child mapping, holding the tahoe equivalent of a directory. All
834 child names are unicode strings, and all children are some sort of
835 IFilesystemNode (either files or subdirectories).
840 The dirnode ('1') URI returned by this method can be used in
841 set_uri() on a different directory ('2') to 'mount' a reference to
842 this directory ('1') under the other ('2'). This URI is just a
843 string, so it can be passed around through email or other out-of-band
847 def get_readonly_uri():
849 The dirnode ('1') URI returned by this method can be used in
850 set_uri() on a different directory ('2') to 'mount' a reference to
851 this directory ('1') under the other ('2'). This URI is just a
852 string, so it can be passed around through email or other out-of-band
857 """I return a Deferred that fires with a dictionary mapping child
858 name (a unicode string) to (node, metadata_dict) tuples, in which
859 'node' is an IFilesystemNode (either IFileNode or IDirectoryNode),
860 and 'metadata_dict' is a dictionary of metadata."""
863 """I return a Deferred that fires with a boolean, True if there
864 exists a child of the given name, False if not. The child name must
865 be a unicode string."""
868 """I return a Deferred that fires with a specific named child node,
869 either an IFileNode or an IDirectoryNode. The child name must be a
870 unicode string. I raise NoSuchChildError if I do not have a child by
873 def get_metadata_for(name):
874 """I return a Deferred that fires with the metadata dictionary for a
875 specific named child node. This metadata is stored in the *edge*, not
876 in the child, so it is attached to the parent dirnode rather than the
877 child dir-or-file-node. The child name must be a unicode string. I
878 raise NoSuchChildError if I do not have a child by that name."""
880 def set_metadata_for(name, metadata):
881 """I replace any existing metadata for the named child with the new
882 metadata. The child name must be a unicode string. This metadata is
883 stored in the *edge*, not in the child, so it is attached to the
884 parent dirnode rather than the child dir-or-file-node. I return a
885 Deferred (that fires with this dirnode) when the operation is
886 complete. I raise NoSuchChildError if I do not have a child by that
889 def get_child_at_path(path):
890 """Transform a child path into an IFilesystemNode.
892 I perform a recursive series of 'get' operations to find the named
893 descendant node. I return a Deferred that fires with the node, or
894 errbacks with NoSuchChildError if the node could not be found.
896 The path can be either a single string (slash-separated) or a list of
897 path-name elements. All elements must be unicode strings.
900 def get_child_and_metadata_at_path(path):
901 """Transform a child path into an IFilesystemNode and metadata.
903 I am like get_child_at_path(), but my Deferred fires with a tuple of
904 (node, metadata). The metadata comes from the last edge. If the path
905 is empty, the metadata will be an empty dictionary.
908 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
909 """I add a child (by writecap+readcap) at the specific name. I return
910 a Deferred that fires when the operation finishes. If overwrite= is
911 True, I will replace any existing child of the same name, otherwise
912 an existing child will cause me to return ExistingChildError. The
913 child name must be a unicode string.
915 The child caps could be for a file, or for a directory. If the new
916 child is read/write, you will provide both writecap and readcap. If
917 the child is read-only, you will provide the readcap write (i.e. the
918 writecap= and readcap= arguments will both be the child's readcap).
919 The filecaps are typically obtained from an IFilesystemNode with
920 get_uri() and get_readonly_uri().
922 If metadata= is provided, I will use it as the metadata for the named
923 edge. This will replace any existing metadata. If metadata= is left
924 as the default value of None, I will set ['mtime'] to the current
925 time, and I will set ['ctime'] to the current time if there was not
926 already a child by this name present. This roughly matches the
927 ctime/mtime semantics of traditional filesystems.
929 If this directory node is read-only, the Deferred will errback with a
930 NotWriteableError."""
932 def set_children(entries, overwrite=True):
933 """Add multiple children (by writecap+readcap) to a directory node.
934 Takes a dictionary, with childname as keys and (writecap, readcap)
935 tuples (or (writecap, readcap, metadata) triples) as values. Returns
936 a Deferred that fires (with this dirnode) when the operation
937 finishes. This is equivalent to calling set_uri() multiple times, but
938 is much more efficient. All child names must be unicode strings.
941 def set_node(name, child, metadata=None, overwrite=True):
942 """I add a child at the specific name. I return a Deferred that fires
943 when the operation finishes. This Deferred will fire with the child
944 node that was just added. I will replace any existing child of the
945 same name. The child name must be a unicode string. The 'child'
946 instance must be an instance providing IFilesystemNode.
948 If metadata= is provided, I will use it as the metadata for the named
949 edge. This will replace any existing metadata. If metadata= is left
950 as the default value of None, I will set ['mtime'] to the current
951 time, and I will set ['ctime'] to the current time if there was not
952 already a child by this name present. This roughly matches the
953 ctime/mtime semantics of traditional filesystems.
955 If this directory node is read-only, the Deferred will errback with a
956 NotWriteableError."""
958 def set_nodes(entries, overwrite=True):
959 """Add multiple children to a directory node. Takes a dict mapping
960 unicode childname to (child_node, metdata) tuples. If metdata=None,
961 the original metadata is left unmodified. Returns a Deferred that
962 fires (with this dirnode) when the operation finishes. This is
963 equivalent to calling set_node() multiple times, but is much more
966 def add_file(name, uploadable, metadata=None, overwrite=True):
967 """I upload a file (using the given IUploadable), then attach the
968 resulting ImmutableFileNode to the directory at the given name. I set
969 metadata the same way as set_uri and set_node. The child name must be
972 I return a Deferred that fires (with the IFileNode of the uploaded
973 file) when the operation completes."""
976 """I remove the child at the specific name. I return a Deferred that
977 fires when the operation finishes. The child name must be a unicode
978 string. I raise NoSuchChildError if I do not have a child by that
981 def create_subdirectory(name, initial_children={}, overwrite=True):
982 """I create and attach a directory at the given name. The new
983 directory can be empty, or it can be populated with children
984 according to 'initial_children', which takes a dictionary in the same
985 format as set_nodes (i.e. mapping unicode child name to (childnode,
986 metadata) tuples). The child name must be a unicode string. I return
987 a Deferred that fires (with the new directory node) when the
988 operation finishes."""
990 def move_child_to(current_child_name, new_parent, new_child_name=None,
992 """I take one of my children and move them to a new parent. The child
993 is referenced by name. On the new parent, the child will live under
994 'new_child_name', which defaults to 'current_child_name'. TODO: what
995 should we do about metadata? I return a Deferred that fires when the
996 operation finishes. The child name must be a unicode string. I raise
997 NoSuchChildError if I do not have a child by that name."""
999 def build_manifest():
1000 """I generate a table of everything reachable from this directory.
1001 I also compute deep-stats as described below.
1003 I return a Monitor. The Monitor's results will be a dictionary with
1006 res['manifest']: a list of (path, cap) tuples for all nodes
1007 (directories and files) reachable from this one.
1008 'path' will be a tuple of unicode strings. The
1009 origin dirnode will be represented by an empty path
1011 res['verifycaps']: a list of (printable) verifycap strings, one for
1012 each reachable non-LIT node. This is a set:
1013 it will contain no duplicates.
1014 res['storage-index']: a list of (base32) storage index strings,
1015 one for each reachable non-LIT node. This is
1016 a set: it will contain no duplicates.
1017 res['stats']: a dictionary, the same that is generated by
1018 start_deep_stats() below.
1020 The Monitor will also have an .origin_si attribute with the (binary)
1021 storage index of the starting point.
1024 def start_deep_stats():
1025 """Return a Monitor, examining all nodes (directories and files)
1026 reachable from this one. The Monitor's results will be a dictionary
1027 with the following keys::
1029 count-immutable-files: count of how many CHK files are in the set
1030 count-mutable-files: same, for mutable files (does not include
1032 count-literal-files: same, for LIT files
1033 count-files: sum of the above three
1035 count-directories: count of directories
1037 size-immutable-files: total bytes for all CHK files in the set
1038 size-mutable-files (TODO): same, for current version of all mutable
1039 files, does not include directories
1040 size-literal-files: same, for LIT files
1041 size-directories: size of mutable files used by directories
1043 largest-directory: number of bytes in the largest directory
1044 largest-directory-children: number of children in the largest
1046 largest-immutable-file: number of bytes in the largest CHK file
1048 size-mutable-files is not yet implemented, because it would involve
1049 even more queries than deep_stats does.
1051 The Monitor will also have an .origin_si attribute with the (binary)
1052 storage index of the starting point.
1054 This operation will visit every directory node underneath this one,
1055 and can take a long time to run. On a typical workstation with good
1056 bandwidth, this can examine roughly 15 directories per second (and
1057 takes several minutes of 100% CPU for ~1700 directories).
1060 class ICodecEncoder(Interface):
1061 def set_params(data_size, required_shares, max_shares):
1062 """Set up the parameters of this encoder.
1064 This prepares the encoder to perform an operation that converts a
1065 single block of data into a number of shares, such that a future
1066 ICodecDecoder can use a subset of these shares to recover the
1067 original data. This operation is invoked by calling encode(). Once
1068 the encoding parameters are set up, the encode operation can be
1069 invoked multiple times.
1071 set_params() prepares the encoder to accept blocks of input data that
1072 are exactly 'data_size' bytes in length. The encoder will be prepared
1073 to produce 'max_shares' shares for each encode() operation (although
1074 see the 'desired_share_ids' to use less CPU). The encoding math will
1075 be chosen such that the decoder can get by with as few as
1076 'required_shares' of these shares and still reproduce the original
1077 data. For example, set_params(1000, 5, 5) offers no redundancy at
1078 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1080 Numerical Restrictions: 'data_size' is required to be an integral
1081 multiple of 'required_shares'. In general, the caller should choose
1082 required_shares and max_shares based upon their reliability
1083 requirements and the number of peers available (the total storage
1084 space used is roughly equal to max_shares*data_size/required_shares),
1085 then choose data_size to achieve the memory footprint desired (larger
1086 data_size means more efficient operation, smaller data_size means
1087 smaller memory footprint).
1089 In addition, 'max_shares' must be equal to or greater than
1090 'required_shares'. Of course, setting them to be equal causes
1091 encode() to degenerate into a particularly slow form of the 'split'
1094 See encode() for more details about how these parameters are used.
1096 set_params() must be called before any other ICodecEncoder methods
1101 """Return the 3-tuple of data_size, required_shares, max_shares"""
1103 def get_encoder_type():
1104 """Return a short string that describes the type of this encoder.
1106 There is required to be a global table of encoder classes. This method
1107 returns an index into this table; the value at this index is an
1108 encoder class, and this encoder is an instance of that class.
1111 def get_block_size():
1112 """Return the length of the shares that encode() will produce.
1115 def encode_proposal(data, desired_share_ids=None):
1116 """Encode some data.
1118 'data' must be a string (or other buffer object), and len(data) must
1119 be equal to the 'data_size' value passed earlier to set_params().
1121 This will return a Deferred that will fire with two lists. The first
1122 is a list of shares, each of which is a string (or other buffer
1123 object) such that len(share) is the same as what get_share_size()
1124 returned earlier. The second is a list of shareids, in which each is
1125 an integer. The lengths of the two lists will always be equal to each
1126 other. The user should take care to keep each share closely
1127 associated with its shareid, as one is useless without the other.
1129 The length of this output list will normally be the same as the value
1130 provided to the 'max_shares' parameter of set_params(). This may be
1131 different if 'desired_share_ids' is provided.
1133 'desired_share_ids', if provided, is required to be a sequence of
1134 ints, each of which is required to be >= 0 and < max_shares. If not
1135 provided, encode() will produce 'max_shares' shares, as if
1136 'desired_share_ids' were set to range(max_shares). You might use this
1137 if you initially thought you were going to use 10 peers, started
1138 encoding, and then two of the peers dropped out: you could use
1139 desired_share_ids= to skip the work (both memory and CPU) of
1140 producing shares for the peers which are no longer available.
1144 def encode(inshares, desired_share_ids=None):
1145 """Encode some data. This may be called multiple times. Each call is
1148 inshares is a sequence of length required_shares, containing buffers
1149 (i.e. strings), where each buffer contains the next contiguous
1150 non-overlapping segment of the input data. Each buffer is required to
1151 be the same length, and the sum of the lengths of the buffers is
1152 required to be exactly the data_size promised by set_params(). (This
1153 implies that the data has to be padded before being passed to
1154 encode(), unless of course it already happens to be an even multiple
1155 of required_shares in length.)
1157 ALSO: the requirement to break up your data into 'required_shares'
1158 chunks before calling encode() feels a bit surprising, at least from
1159 the point of view of a user who doesn't know how FEC works. It feels
1160 like an implementation detail that has leaked outside the
1161 abstraction barrier. Can you imagine a use case in which the data to
1162 be encoded might already be available in pre-segmented chunks, such
1163 that it is faster or less work to make encode() take a list rather
1164 than splitting a single string?
1166 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1167 is supposed to *produce* shares, so what it *accepts* should be
1168 something other than shares. Other places in this interface use the
1169 word 'data' for that-which-is-not-shares.. maybe we should use that
1172 'desired_share_ids', if provided, is required to be a sequence of
1173 ints, each of which is required to be >= 0 and < max_shares. If not
1174 provided, encode() will produce 'max_shares' shares, as if
1175 'desired_share_ids' were set to range(max_shares). You might use this
1176 if you initially thought you were going to use 10 peers, started
1177 encoding, and then two of the peers dropped out: you could use
1178 desired_share_ids= to skip the work (both memory and CPU) of
1179 producing shares for the peers which are no longer available.
1181 For each call, encode() will return a Deferred that fires with two
1182 lists, one containing shares and the other containing the shareids.
1183 The get_share_size() method can be used to determine the length of
1184 the share strings returned by encode(). Each shareid is a small
1185 integer, exactly as passed into 'desired_share_ids' (or
1186 range(max_shares), if desired_share_ids was not provided).
1188 The shares and their corresponding shareids are required to be kept
1189 together during storage and retrieval. Specifically, the share data is
1190 useless by itself: the decoder needs to be told which share is which
1191 by providing it with both the shareid and the actual share data.
1193 This function will allocate an amount of memory roughly equal to::
1195 (max_shares - required_shares) * get_share_size()
1197 When combined with the memory that the caller must allocate to
1198 provide the input data, this leads to a memory footprint roughly
1199 equal to the size of the resulting encoded shares (i.e. the expansion
1200 factor times the size of the input segment).
1205 # returning a list of (shareidN,shareN) tuples instead of a pair of
1206 # lists (shareids..,shares..). Brian thought the tuples would
1207 # encourage users to keep the share and shareid together throughout
1208 # later processing, Zooko pointed out that the code to iterate
1209 # through two lists is not really more complicated than using a list
1210 # of tuples and there's also a performance improvement
1212 # having 'data_size' not required to be an integral multiple of
1213 # 'required_shares'. Doing this would require encode() to perform
1214 # padding internally, and we'd prefer to have any padding be done
1215 # explicitly by the caller. Yes, it is an abstraction leak, but
1216 # hopefully not an onerous one.
1219 class ICodecDecoder(Interface):
1220 def set_params(data_size, required_shares, max_shares):
1221 """Set the params. They have to be exactly the same ones that were
1222 used for encoding."""
1224 def get_needed_shares():
1225 """Return the number of shares needed to reconstruct the data.
1226 set_params() is required to be called before this."""
1228 def decode(some_shares, their_shareids):
1229 """Decode a partial list of shares into data.
1231 'some_shares' is required to be a sequence of buffers of sharedata, a
1232 subset of the shares returned by ICodecEncode.encode(). Each share is
1233 required to be of the same length. The i'th element of their_shareids
1234 is required to be the shareid of the i'th buffer in some_shares.
1236 This returns a Deferred which fires with a sequence of buffers. This
1237 sequence will contain all of the segments of the original data, in
1238 order. The sum of the lengths of all of the buffers will be the
1239 'data_size' value passed into the original ICodecEncode.set_params()
1240 call. To get back the single original input block of data, use
1241 ''.join(output_buffers), or you may wish to simply write them in
1242 order to an output file.
1244 Note that some of the elements in the result sequence may be
1245 references to the elements of the some_shares input sequence. In
1246 particular, this means that if those share objects are mutable (e.g.
1247 arrays) and if they are changed, then both the input (the
1248 'some_shares' parameter) and the output (the value given when the
1249 deferred is triggered) will change.
1251 The length of 'some_shares' is required to be exactly the value of
1252 'required_shares' passed into the original ICodecEncode.set_params()
1256 class IEncoder(Interface):
1257 """I take an object that provides IEncryptedUploadable, which provides
1258 encrypted data, and a list of shareholders. I then encode, hash, and
1259 deliver shares to those shareholders. I will compute all the necessary
1260 Merkle hash trees that are necessary to validate the crypttext that
1261 eventually comes back from the shareholders. I provide the URI Extension
1262 Block Hash, and the encoding parameters, both of which must be included
1265 I do not choose shareholders, that is left to the IUploader. I must be
1266 given a dict of RemoteReferences to storage buckets that are ready and
1267 willing to receive data.
1271 """Specify the number of bytes that will be encoded. This must be
1272 peformed before get_serialized_params() can be called.
1274 def set_params(params):
1275 """Override the default encoding parameters. 'params' is a tuple of
1276 (k,d,n), where 'k' is the number of required shares, 'd' is the
1277 servers_of_happiness, and 'n' is the total number of shares that will
1280 Encoding parameters can be set in three ways. 1: The Encoder class
1281 provides defaults (3/7/10). 2: the Encoder can be constructed with
1282 an 'options' dictionary, in which the
1283 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1284 set_params((k,d,n)) can be called.
1286 If you intend to use set_params(), you must call it before
1287 get_share_size or get_param are called.
1290 def set_encrypted_uploadable(u):
1291 """Provide a source of encrypted upload data. 'u' must implement
1292 IEncryptedUploadable.
1294 When this is called, the IEncryptedUploadable will be queried for its
1295 length and the storage_index that should be used.
1297 This returns a Deferred that fires with this Encoder instance.
1299 This must be performed before start() can be called.
1302 def get_param(name):
1303 """Return an encoding parameter, by name.
1305 'storage_index': return a string with the (16-byte truncated SHA-256
1306 hash) storage index to which these shares should be
1309 'share_counts': return a tuple describing how many shares are used:
1310 (needed_shares, servers_of_happiness, total_shares)
1312 'num_segments': return an int with the number of segments that
1315 'segment_size': return an int with the size of each segment.
1317 'block_size': return the size of the individual blocks that will
1318 be delivered to a shareholder's put_block() method. By
1319 knowing this, the shareholder will be able to keep all
1320 blocks in a single file and still provide random access
1321 when reading them. # TODO: can we avoid exposing this?
1323 'share_size': an int with the size of the data that will be stored
1324 on each shareholder. This is aggregate amount of data
1325 that will be sent to the shareholder, summed over all
1326 the put_block() calls I will ever make. It is useful to
1327 determine this size before asking potential
1328 shareholders whether they will grant a lease or not,
1329 since their answers will depend upon how much space we
1330 need. TODO: this might also include some amount of
1331 overhead, like the size of all the hashes. We need to
1332 decide whether this is useful or not.
1334 'serialized_params': a string with a concise description of the
1335 codec name and its parameters. This may be passed
1336 into the IUploadable to let it make sure that
1337 the same file encoded with different parameters
1338 will result in different storage indexes.
1340 Once this is called, set_size() and set_params() may not be called.
1343 def set_shareholders(shareholders, servermap):
1344 """Tell the encoder where to put the encoded shares. 'shareholders'
1345 must be a dictionary that maps share number (an integer ranging from
1346 0 to n-1) to an instance that provides IStorageBucketWriter.
1347 'servermap' is a dictionary that maps share number (as defined above)
1348 to a set of peerids. This must be performed before start() can be
1352 """Begin the encode/upload process. This involves reading encrypted
1353 data from the IEncryptedUploadable, encoding it, uploading the shares
1354 to the shareholders, then sending the hash trees.
1356 set_encrypted_uploadable() and set_shareholders() must be called
1357 before this can be invoked.
1359 This returns a Deferred that fires with a verify cap when the upload
1360 process is complete. The verifycap, plus the encryption key, is
1361 sufficient to construct the read cap.
1364 class IDecoder(Interface):
1365 """I take a list of shareholders and some setup information, then
1366 download, validate, decode, and decrypt data from them, writing the
1367 results to an output file.
1369 I do not locate the shareholders, that is left to the IDownloader. I must
1370 be given a dict of RemoteReferences to storage buckets that are ready to
1375 """I take a file-like object (providing write and close) to which all
1376 the plaintext data will be written.
1378 TODO: producer/consumer . Maybe write() should return a Deferred that
1379 indicates when it will accept more data? But probably having the
1380 IDecoder be a producer is easier to glue to IConsumer pieces.
1383 def set_shareholders(shareholders):
1384 """I take a dictionary that maps share identifiers (small integers)
1385 to RemoteReferences that provide RIBucketReader. This must be called
1389 """I start the download. This process involves retrieving data and
1390 hash chains from the shareholders, using the hashes to validate the
1391 data, decoding the shares into segments, decrypting the segments,
1392 then writing the resulting plaintext to the output file.
1394 I return a Deferred that will fire (with self) when the download is
1398 class IDownloadTarget(Interface):
1399 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1400 # will register itself as a producer. This allows the target to invoke
1401 # downloader.pauseProducing, resumeProducing, and stopProducing.
1403 """Called before any calls to write() or close(). If an error
1404 occurs before any data is available, fail() may be called without
1405 a previous call to open().
1407 'size' is the length of the file being downloaded, in bytes."""
1410 """Output some data to the target."""
1412 """Inform the target that there is no more data to be written."""
1414 """fail() is called to indicate that the download has failed. 'why'
1415 is a Failure object indicating what went wrong. No further methods
1416 will be invoked on the IDownloadTarget after fail()."""
1417 def register_canceller(cb):
1418 """The CiphertextDownloader uses this to register a no-argument function
1419 that the target can call to cancel the download. Once this canceller
1420 is invoked, no further calls to write() or close() will be made."""
1422 """When the CiphertextDownloader is done, this finish() function will be
1423 called. Whatever it returns will be returned to the invoker of
1424 Downloader.download.
1426 # The following methods are just because that target might be a
1427 # repairer.DownUpConnector, and just because the current CHKUpload object
1428 # expects to find the storage index and encoding parameters in its
1430 def set_storageindex(storageindex):
1431 """ Set the storage index. """
1432 def set_encodingparams(encodingparams):
1433 """ Set the encoding parameters. """
1435 class IDownloader(Interface):
1436 def download(uri, target):
1437 """Perform a CHK download, sending the data to the given target.
1438 'target' must provide IDownloadTarget.
1440 Returns a Deferred that fires (with the results of target.finish)
1441 when the download is finished, or errbacks if something went wrong."""
1443 class IEncryptedUploadable(Interface):
1444 def set_upload_status(upload_status):
1445 """Provide an IUploadStatus object that should be filled with status
1446 information. The IEncryptedUploadable is responsible for setting
1447 key-determination progress ('chk'), size, storage_index, and
1448 ciphertext-fetch progress. It may delegate some of this
1449 responsibility to others, in particular to the IUploadable."""
1452 """This behaves just like IUploadable.get_size()."""
1454 def get_all_encoding_parameters():
1455 """Return a Deferred that fires with a tuple of
1456 (k,happy,n,segment_size). The segment_size will be used as-is, and
1457 must match the following constraints: it must be a multiple of k, and
1458 it shouldn't be unreasonably larger than the file size (if
1459 segment_size is larger than filesize, the difference must be stored
1462 This usually passes through to the IUploadable method of the same
1465 The encoder strictly obeys the values returned by this method. To
1466 make an upload use non-default encoding parameters, you must arrange
1467 to control the values that this method returns.
1470 def get_storage_index():
1471 """Return a Deferred that fires with a 16-byte storage index.
1474 def read_encrypted(length, hash_only):
1475 """This behaves just like IUploadable.read(), but returns crypttext
1476 instead of plaintext. If hash_only is True, then this discards the
1477 data (and returns an empty list); this improves efficiency when
1478 resuming an interrupted upload (where we need to compute the
1479 plaintext hashes, but don't need the redundant encrypted data)."""
1481 def get_plaintext_hashtree_leaves(first, last, num_segments):
1482 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1483 plaintext segments, i.e. get the tagged hashes of the given segments.
1484 The segment size is expected to be generated by the
1485 IEncryptedUploadable before any plaintext is read or ciphertext
1486 produced, so that the segment hashes can be generated with only a
1489 This returns a Deferred which fires with a sequence of hashes, using:
1491 tuple(segment_hashes[first:last])
1493 'num_segments' is used to assert that the number of segments that the
1494 IEncryptedUploadable handled matches the number of segments that the
1495 encoder was expecting.
1497 This method must not be called until the final byte has been read
1498 from read_encrypted(). Once this method is called, read_encrypted()
1499 can never be called again.
1502 def get_plaintext_hash():
1503 """OBSOLETE; Get the hash of the whole plaintext.
1505 This returns a Deferred which fires with a tagged SHA-256 hash of the
1506 whole plaintext, obtained from hashutil.plaintext_hash(data).
1510 """Just like IUploadable.close()."""
1512 class IUploadable(Interface):
1513 def set_upload_status(upload_status):
1514 """Provide an IUploadStatus object that should be filled with status
1515 information. The IUploadable is responsible for setting
1516 key-determination progress ('chk')."""
1518 def set_default_encoding_parameters(params):
1519 """Set the default encoding parameters, which must be a dict mapping
1520 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1521 'max_segment_size'. These might have an influence on the final
1522 encoding parameters returned by get_all_encoding_parameters(), if the
1523 Uploadable doesn't have more specific preferences.
1525 This call is optional: if it is not used, the Uploadable will use
1526 some built-in defaults. If used, this method must be called before
1527 any other IUploadable methods to have any effect.
1531 """Return a Deferred that will fire with the length of the data to be
1532 uploaded, in bytes. This will be called before the data is actually
1533 used, to compute encoding parameters.
1536 def get_all_encoding_parameters():
1537 """Return a Deferred that fires with a tuple of
1538 (k,happy,n,segment_size). The segment_size will be used as-is, and
1539 must match the following constraints: it must be a multiple of k, and
1540 it shouldn't be unreasonably larger than the file size (if
1541 segment_size is larger than filesize, the difference must be stored
1544 The relative values of k and n allow some IUploadables to request
1545 better redundancy than others (in exchange for consuming more space
1548 Larger values of segment_size reduce hash overhead, while smaller
1549 values reduce memory footprint and cause data to be delivered in
1550 smaller pieces (which may provide a smoother and more predictable
1551 download experience).
1553 The encoder strictly obeys the values returned by this method. To
1554 make an upload use non-default encoding parameters, you must arrange
1555 to control the values that this method returns. One way to influence
1556 them may be to call set_encoding_parameters() before calling
1557 get_all_encoding_parameters().
1560 def get_encryption_key():
1561 """Return a Deferred that fires with a 16-byte AES key. This key will
1562 be used to encrypt the data. The key will also be hashed to derive
1565 Uploadables which want to achieve convergence should hash their file
1566 contents and the serialized_encoding_parameters to form the key
1567 (which of course requires a full pass over the data). Uploadables can
1568 use the upload.ConvergentUploadMixin class to achieve this
1571 Uploadables which do not care about convergence (or do not wish to
1572 make multiple passes over the data) can simply return a
1573 strongly-random 16 byte string.
1575 get_encryption_key() may be called multiple times: the IUploadable is
1576 required to return the same value each time.
1580 """Return a Deferred that fires with a list of strings (perhaps with
1581 only a single element) which, when concatenated together, contain the
1582 next 'length' bytes of data. If EOF is near, this may provide fewer
1583 than 'length' bytes. The total number of bytes provided by read()
1584 before it signals EOF must equal the size provided by get_size().
1586 If the data must be acquired through multiple internal read
1587 operations, returning a list instead of a single string may help to
1588 reduce string copies.
1590 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1591 so a 10kB file means length=3kB, 100kB file means length=30kB,
1592 and >=1MB file means length=300kB.
1594 This method provides for a single full pass through the data. Later
1595 use cases may desire multiple passes or access to only parts of the
1596 data (such as a mutable file making small edits-in-place). This API
1597 will be expanded once those use cases are better understood.
1601 """The upload is finished, and whatever filehandle was in use may be
1604 class IUploadResults(Interface):
1605 """I am returned by upload() methods. I contain a number of public
1606 attributes which can be read to determine the results of the upload. Some
1607 of these are functional, some are timing information. All of these may be
1610 .file_size : the size of the file, in bytes
1611 .uri : the CHK read-cap for the file
1612 .ciphertext_fetched : how many bytes were fetched by the helper
1613 .sharemap: dict mapping share identifier to set of serverids
1614 (binary strings). This indicates which servers were given
1615 which shares. For immutable files, the shareid is an
1616 integer (the share number, from 0 to N-1). For mutable
1617 files, it is a string of the form 'seq%d-%s-sh%d',
1618 containing the sequence number, the roothash, and the
1620 .servermap : dict mapping server peerid to a set of share numbers
1621 .timings : dict of timing information, mapping name to seconds (float)
1622 total : total upload time, start to finish
1623 storage_index : time to compute the storage index
1624 peer_selection : time to decide which peers will be used
1625 contacting_helper : initial helper query to upload/no-upload decision
1626 existence_check : helper pre-upload existence check
1627 helper_total : initial helper query to helper finished pushing
1628 cumulative_fetch : helper waiting for ciphertext requests
1629 total_fetch : helper start to last ciphertext response
1630 cumulative_encoding : just time spent in zfec
1631 cumulative_sending : just time spent waiting for storage servers
1632 hashes_and_close : last segment push to shareholder close
1633 total_encode_and_push : first encode to shareholder close
1637 class IDownloadResults(Interface):
1638 """I am created internally by download() methods. I contain a number of
1639 public attributes which contain details about the download process.::
1641 .file_size : the size of the file, in bytes
1642 .servers_used : set of server peerids that were used during download
1643 .server_problems : dict mapping server peerid to a problem string. Only
1644 servers that had problems (bad hashes, disconnects)
1646 .servermap : dict mapping server peerid to a set of share numbers. Only
1647 servers that had any shares are listed here.
1648 .timings : dict of timing information, mapping name to seconds (float)
1649 peer_selection : time to ask servers about shares
1650 servers_peer_selection : dict of peerid to DYHB-query time
1651 uri_extension : time to fetch a copy of the URI extension block
1652 hashtrees : time to fetch the hash trees
1653 segments : time to fetch, decode, and deliver segments
1654 cumulative_fetch : time spent waiting for storage servers
1655 cumulative_decode : just time spent in zfec
1656 cumulative_decrypt : just time spent in decryption
1657 total : total download time, start to finish
1658 fetch_per_server : dict of peerid to list of per-segment fetch times
1662 class IUploader(Interface):
1663 def upload(uploadable):
1664 """Upload the file. 'uploadable' must impement IUploadable. This
1665 returns a Deferred which fires with an IUploadResults instance, from
1666 which the URI of the file can be obtained as results.uri ."""
1668 def upload_ssk(write_capability, new_version, uploadable):
1669 """TODO: how should this work?"""
1671 class ICheckable(Interface):
1672 def check(monitor, verify=False, add_lease=False):
1673 """Check upon my health, optionally repairing any problems.
1675 This returns a Deferred that fires with an instance that provides
1676 ICheckResults, or None if the object is non-distributed (i.e. LIT
1679 The monitor will be checked periodically to see if the operation has
1680 been cancelled. If so, no new queries will be sent, and the Deferred
1681 will fire (with a OperationCancelledError) immediately.
1683 Filenodes and dirnodes (which provide IFilesystemNode) are also
1684 checkable. Instances that represent verifier-caps will be checkable
1685 but not downloadable. Some objects (like LIT files) do not actually
1686 live in the grid, and their checkers return None (non-distributed
1687 files are always healthy).
1689 If verify=False, a relatively lightweight check will be performed: I
1690 will ask all servers if they have a share for me, and I will believe
1691 whatever they say. If there are at least N distinct shares on the
1692 grid, my results will indicate r.is_healthy()==True. This requires a
1693 roundtrip to each server, but does not transfer very much data, so
1694 the network bandwidth is fairly low.
1696 If verify=True, a more resource-intensive check will be performed:
1697 every share will be downloaded, and the hashes will be validated on
1698 every bit. I will ignore any shares that failed their hash checks. If
1699 there are at least N distinct valid shares on the grid, my results
1700 will indicate r.is_healthy()==True. This requires N/k times as much
1701 download bandwidth (and server disk IO) as a regular download. If a
1702 storage server is holding a corrupt share, or is experiencing memory
1703 failures during retrieval, or is malicious or buggy, then
1704 verification will detect the problem, but checking will not.
1706 If add_lease=True, I will ensure that an up-to-date lease is present
1707 on each share. The lease secrets will be derived from by node secret
1708 (in BASEDIR/private/secret), so either I will add a new lease to the
1709 share, or I will merely renew the lease that I already had. In a
1710 future version of the storage-server protocol (once Accounting has
1711 been implemented), there may be additional options here to define the
1712 kind of lease that is obtained (which account number to claim, etc).
1714 TODO: any problems seen during checking will be reported to the
1715 health-manager.furl, a centralized object which is responsible for
1716 figuring out why files are unhealthy so corrective action can be
1720 def check_and_repair(monitor, verify=False, add_lease=False):
1721 """Like check(), but if the file/directory is not healthy, attempt to
1724 Any non-healthy result will cause an immediate repair operation, to
1725 generate and upload new shares. After repair, the file will be as
1726 healthy as we can make it. Details about what sort of repair is done
1727 will be put in the check-and-repair results. The Deferred will not
1728 fire until the repair is complete.
1730 This returns a Deferred which fires with an instance of
1731 ICheckAndRepairResults."""
1733 class IDeepCheckable(Interface):
1734 def start_deep_check(verify=False, add_lease=False):
1735 """Check upon the health of me and everything I can reach.
1737 This is a recursive form of check(), useable only on dirnodes.
1739 I return a Monitor, with results that are an IDeepCheckResults
1742 TODO: If any of the directories I traverse are unrecoverable, the
1743 Monitor will report failure. If any of the files I check upon are
1744 unrecoverable, those problems will be reported in the
1745 IDeepCheckResults as usual, and the Monitor will not report a
1749 def start_deep_check_and_repair(verify=False, add_lease=False):
1750 """Check upon the health of me and everything I can reach. Repair
1751 anything that isn't healthy.
1753 This is a recursive form of check_and_repair(), useable only on
1756 I return a Monitor, with results that are an
1757 IDeepCheckAndRepairResults object.
1759 TODO: If any of the directories I traverse are unrecoverable, the
1760 Monitor will report failure. If any of the files I check upon are
1761 unrecoverable, those problems will be reported in the
1762 IDeepCheckResults as usual, and the Monitor will not report a
1766 class ICheckResults(Interface):
1767 """I contain the detailed results of a check/verify operation.
1770 def get_storage_index():
1771 """Return a string with the (binary) storage index."""
1772 def get_storage_index_string():
1773 """Return a string with the (printable) abbreviated storage index."""
1775 """Return the (string) URI of the object that was checked."""
1778 """Return a boolean, True if the file/dir is fully healthy, False if
1779 it is damaged in any way. Non-distributed LIT files always return
1782 def is_recoverable():
1783 """Return a boolean, True if the file/dir can be recovered, False if
1784 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1785 files always return True."""
1787 def needs_rebalancing():
1788 """Return a boolean, True if the file/dir's reliability could be
1789 improved by moving shares to new servers. Non-distributed LIT files
1790 always return False."""
1794 """Return a dictionary that describes the state of the file/dir. LIT
1795 files always return an empty dictionary. Normal files and directories
1796 return a dictionary with the following keys (note that these use
1797 binary strings rather than base32-encoded ones) (also note that for
1798 mutable files, these counts are for the 'best' version):
1800 count-shares-good: the number of distinct good shares that were found
1801 count-shares-needed: 'k', the number of shares required for recovery
1802 count-shares-expected: 'N', the number of total shares generated
1803 count-good-share-hosts: the number of distinct storage servers with
1804 good shares. If this number is less than
1805 count-shares-good, then some shares are
1806 doubled up, increasing the correlation of
1807 failures. This indicates that one or more
1808 shares should be moved to an otherwise unused
1809 server, if one is available.
1810 count-corrupt-shares: the number of shares with integrity failures
1811 list-corrupt-shares: a list of 'share locators', one for each share
1812 that was found to be corrupt. Each share
1813 locator is a list of (serverid, storage_index,
1815 count-incompatible-shares: the number of shares which are of a share
1816 format unknown to this checker
1817 list-incompatible-shares: a list of 'share locators', one for each
1818 share that was found to be of an unknown
1819 format. Each share locator is a list of
1820 (serverid, storage_index, sharenum).
1821 servers-responding: list of (binary) storage server identifiers,
1822 one for each server which responded to the share
1823 query (even if they said they didn't have
1824 shares, and even if they said they did have
1825 shares but then didn't send them when asked, or
1826 dropped the connection, or returned a Failure,
1827 and even if they said they did have shares and
1828 sent incorrect ones when asked)
1829 sharemap: dict mapping share identifier to list of serverids
1830 (binary strings). This indicates which servers are holding
1831 which shares. For immutable files, the shareid is an
1832 integer (the share number, from 0 to N-1). For mutable
1833 files, it is a string of the form 'seq%d-%s-sh%d',
1834 containing the sequence number, the roothash, and the
1837 The following keys are most relevant for mutable files, but immutable
1838 files will provide sensible values too::
1840 count-wrong-shares: the number of shares for versions other than the
1841 'best' one (which is defined as being the
1842 recoverable version with the highest sequence
1843 number, then the highest roothash). These are
1844 either leftover shares from an older version
1845 (perhaps on a server that was offline when an
1846 update occurred), shares from an unrecoverable
1847 newer version, or shares from an alternate
1848 current version that results from an
1849 uncoordinated write collision. For a healthy
1850 file, this will equal 0.
1852 count-recoverable-versions: the number of recoverable versions of
1853 the file. For a healthy file, this will
1856 count-unrecoverable-versions: the number of unrecoverable versions
1857 of the file. For a healthy file, this
1863 """Return a string with a brief (one-line) summary of the results."""
1866 """Return a list of strings with more detailed results."""
1868 class ICheckAndRepairResults(Interface):
1869 """I contain the detailed results of a check/verify/repair operation.
1871 The IFilesystemNode.check()/verify()/repair() methods all return
1872 instances that provide ICheckAndRepairResults.
1875 def get_storage_index():
1876 """Return a string with the (binary) storage index."""
1877 def get_storage_index_string():
1878 """Return a string with the (printable) abbreviated storage index."""
1879 def get_repair_attempted():
1880 """Return a boolean, True if a repair was attempted. We might not
1881 attempt to repair the file because it was healthy, or healthy enough
1882 (i.e. some shares were missing but not enough to exceed some
1883 threshold), or because we don't know how to repair this object."""
1884 def get_repair_successful():
1885 """Return a boolean, True if repair was attempted and the file/dir
1886 was fully healthy afterwards. False if no repair was attempted or if
1887 a repair attempt failed."""
1888 def get_pre_repair_results():
1889 """Return an ICheckResults instance that describes the state of the
1890 file/dir before any repair was attempted."""
1891 def get_post_repair_results():
1892 """Return an ICheckResults instance that describes the state of the
1893 file/dir after any repair was attempted. If no repair was attempted,
1894 the pre-repair and post-repair results will be identical."""
1897 class IDeepCheckResults(Interface):
1898 """I contain the results of a deep-check operation.
1900 This is returned by a call to ICheckable.deep_check().
1903 def get_root_storage_index_string():
1904 """Return the storage index (abbreviated human-readable string) of
1905 the first object checked."""
1907 """Return a dictionary with the following keys::
1909 count-objects-checked: count of how many objects were checked
1910 count-objects-healthy: how many of those objects were completely
1912 count-objects-unhealthy: how many were damaged in some way
1913 count-objects-unrecoverable: how many were unrecoverable
1914 count-corrupt-shares: how many shares were found to have
1915 corruption, summed over all objects
1919 def get_corrupt_shares():
1920 """Return a set of (serverid, storage_index, sharenum) for all shares
1921 that were found to be corrupt. Both serverid and storage_index are
1924 def get_all_results():
1925 """Return a dictionary mapping pathname (a tuple of strings, ready to
1926 be slash-joined) to an ICheckResults instance, one for each object
1927 that was checked."""
1929 def get_results_for_storage_index(storage_index):
1930 """Retrive the ICheckResults instance for the given (binary)
1931 storage index. Raises KeyError if there are no results for that
1935 """Return a dictionary with the same keys as
1936 IDirectoryNode.deep_stats()."""
1938 class IDeepCheckAndRepairResults(Interface):
1939 """I contain the results of a deep-check-and-repair operation.
1941 This is returned by a call to ICheckable.deep_check_and_repair().
1944 def get_root_storage_index_string():
1945 """Return the storage index (abbreviated human-readable string) of
1946 the first object checked."""
1948 """Return a dictionary with the following keys::
1950 count-objects-checked: count of how many objects were checked
1951 count-objects-healthy-pre-repair: how many of those objects were
1952 completely healthy (before any
1954 count-objects-unhealthy-pre-repair: how many were damaged in
1956 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1957 count-objects-healthy-post-repair: how many of those objects were
1958 completely healthy (after any
1960 count-objects-unhealthy-post-repair: how many were damaged in
1962 count-objects-unrecoverable-post-repair: how many were
1964 count-repairs-attempted: repairs were attempted on this many
1965 objects. The count-repairs- keys will
1966 always be provided, however unless
1967 repair=true is present, they will all
1969 count-repairs-successful: how many repairs resulted in healthy
1971 count-repairs-unsuccessful: how many repairs resulted did not
1972 results in completely healthy objects
1973 count-corrupt-shares-pre-repair: how many shares were found to
1974 have corruption, summed over all
1975 objects examined (before any
1977 count-corrupt-shares-post-repair: how many shares were found to
1978 have corruption, summed over all
1979 objects examined (after any
1984 """Return a dictionary with the same keys as
1985 IDirectoryNode.deep_stats()."""
1987 def get_corrupt_shares():
1988 """Return a set of (serverid, storage_index, sharenum) for all shares
1989 that were found to be corrupt before any repair was attempted. Both
1990 serverid and storage_index are binary.
1992 def get_remaining_corrupt_shares():
1993 """Return a set of (serverid, storage_index, sharenum) for all shares
1994 that were found to be corrupt after any repair was completed. Both
1995 serverid and storage_index are binary. These are shares that need
1996 manual inspection and probably deletion.
1998 def get_all_results():
1999 """Return a dictionary mapping pathname (a tuple of strings, ready to
2000 be slash-joined) to an ICheckAndRepairResults instance, one for each
2001 object that was checked."""
2003 def get_results_for_storage_index(storage_index):
2004 """Retrive the ICheckAndRepairResults instance for the given (binary)
2005 storage index. Raises KeyError if there are no results for that
2009 class IRepairable(Interface):
2010 def repair(check_results):
2011 """Attempt to repair the given object. Returns a Deferred that fires
2012 with a IRepairResults object.
2014 I must be called with an object that implements ICheckResults, as
2015 proof that you have actually discovered a problem with this file. I
2016 will use the data in the checker results to guide the repair process,
2017 such as which servers provided bad data and should therefore be
2018 avoided. The ICheckResults object is inside the
2019 ICheckAndRepairResults object, which is returned by the
2020 ICheckable.check() method::
2022 d = filenode.check(repair=False)
2023 def _got_results(check_and_repair_results):
2024 check_results = check_and_repair_results.get_pre_repair_results()
2025 return filenode.repair(check_results)
2026 d.addCallback(_got_results)
2030 class IRepairResults(Interface):
2031 """I contain the results of a repair operation."""
2032 def get_successful(self):
2033 """Returns a boolean: True if the repair made the file healthy, False
2034 if not. Repair failure generally indicates a file that has been
2035 damaged beyond repair."""
2038 class IClient(Interface):
2039 def upload(uploadable):
2040 """Upload some data into a CHK, get back the UploadResults for it.
2041 @param uploadable: something that implements IUploadable
2042 @return: a Deferred that fires with the UploadResults instance.
2043 To get the URI for this file, use results.uri .
2046 def create_mutable_file(contents=""):
2047 """Create a new mutable file (with initial) contents, get back the
2050 @param contents: (bytestring, callable, or None): this provides the
2051 initial contents of the mutable file. If 'contents' is a bytestring,
2052 it will be used as-is. If 'contents' is a callable, it will be
2053 invoked with the new MutableFileNode instance and is expected to
2054 return a bytestring with the initial contents of the file (the
2055 callable can use node.get_writekey() to decide how to encrypt the
2056 initial contents, e.g. for a brand new dirnode with initial
2057 children). contents=None is equivalent to an empty string. Using
2058 content_maker= is more efficient than creating a mutable file and
2059 setting its contents in two separate operations.
2061 @return: a Deferred that fires with an IMutableFileNode instance.
2064 def create_dirnode(initial_children={}):
2065 """Create a new unattached dirnode, possibly with initial children.
2067 @param initial_children: dict with keys that are unicode child names,
2068 and values that are (childnode, metadata) tuples.
2070 @return: a Deferred that fires with the new IDirectoryNode instance.
2073 def create_node_from_uri(uri, rouri):
2074 """Create a new IFilesystemNode instance from the uri, synchronously.
2075 @param uri: a string or IURI-providing instance, or None. This could
2076 be for a LiteralFileNode, a CHK file node, a mutable file
2077 node, or a directory node
2078 @param rouri: a string or IURI-providing instance, or None. If the
2079 main uri is None, I will use the rouri instead. If I
2080 recognize the format of the main uri, I will ignore the
2081 rouri (because it can be derived from the writecap).
2083 @return: an instance that provides IFilesystemNode (or more usefully
2084 one of its subclasses). File-specifying URIs will result in
2085 IFileNode-providing instances, like ImmutableFileNode,
2086 LiteralFileNode, or MutableFileNode. Directory-specifying
2087 URIs will result in IDirectoryNode-providing instances, like
2091 class INodeMaker(Interface):
2092 """The NodeMaker is used to create IFilesystemNode instances. It can
2093 accept a filecap/dircap string and return the node right away. It can
2094 also create new nodes (i.e. upload a file, or create a mutable file)
2095 asynchronously. Once you have one of these nodes, you can use other
2096 methods to determine whether it is a file or directory, and to download
2097 or modify its contents.
2099 The NodeMaker encapsulates all the authorities that these
2100 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2101 Tahoe process will typically have a single NodeMaker, but unit tests may
2102 create simplified/mocked forms for testing purposes.
2104 def create_from_cap(writecap, readcap=None, **kwargs):
2105 """I create an IFilesystemNode from the given writecap/readcap. I can
2106 only provide nodes for existing file/directory objects: use my other
2107 methods to create new objects. I return synchronously."""
2109 def create_mutable_file(contents=None, keysize=None):
2110 """I create a new mutable file, and return a Deferred which will fire
2111 with the IMutableFileNode instance when it is ready. If contents= is
2112 provided (a bytestring), it will be used as the initial contents of
2113 the new file, otherwise the file will contain zero bytes. keysize= is
2114 for use by unit tests, to create mutable files that are smaller than
2117 def create_new_mutable_directory(initial_children={}):
2118 """I create a new mutable directory, and return a Deferred which will
2119 fire with the IDirectoryNode instance when it is ready. If
2120 initial_children= is provided (a dict mapping unicode child name to
2121 (childnode, metadata_dict) tuples), the directory will be populated
2122 with those children, otherwise it will be empty."""
2124 class IClientStatus(Interface):
2125 def list_all_uploads():
2126 """Return a list of uploader objects, one for each upload which
2127 currently has an object available (tracked with weakrefs). This is
2128 intended for debugging purposes."""
2129 def list_active_uploads():
2130 """Return a list of active IUploadStatus objects."""
2131 def list_recent_uploads():
2132 """Return a list of IUploadStatus objects for the most recently
2135 def list_all_downloads():
2136 """Return a list of downloader objects, one for each download which
2137 currently has an object available (tracked with weakrefs). This is
2138 intended for debugging purposes."""
2139 def list_active_downloads():
2140 """Return a list of active IDownloadStatus objects."""
2141 def list_recent_downloads():
2142 """Return a list of IDownloadStatus objects for the most recently
2143 started downloads."""
2145 class IUploadStatus(Interface):
2147 """Return a timestamp (float with seconds since epoch) indicating
2148 when the operation was started."""
2149 def get_storage_index():
2150 """Return a string with the (binary) storage index in use on this
2151 upload. Returns None if the storage index has not yet been
2154 """Return an integer with the number of bytes that will eventually
2155 be uploaded for this file. Returns None if the size is not yet known.
2158 """Return True if this upload is using a Helper, False if not."""
2160 """Return a string describing the current state of the upload
2163 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2164 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2165 made towards hashing the file to determine a CHK encryption key: if
2166 non-convergent encryption is in use, this will be trivial, otherwise
2167 the whole file must be hashed. 'ciphertext' describes how much of the
2168 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2169 uploads. 'encode_and_push' describes how much of the encode-and-push
2170 process has finished: for helper uploads this is dependent upon the
2171 helper providing progress reports. It might be reasonable to add all
2172 three numbers and report the sum to the user."""
2174 """Return True if the upload is currently active, False if not."""
2176 """Return an instance of UploadResults (which contains timing and
2177 sharemap information). Might return None if the upload is not yet
2180 """Each upload status gets a unique number: this method returns that
2181 number. This provides a handle to this particular upload, so a web
2182 page can generate a suitable hyperlink."""
2184 class IDownloadStatus(Interface):
2186 """Return a timestamp (float with seconds since epoch) indicating
2187 when the operation was started."""
2188 def get_storage_index():
2189 """Return a string with the (binary) storage index in use on this
2190 download. This may be None if there is no storage index (i.e. LIT
2193 """Return an integer with the number of bytes that will eventually be
2194 retrieved for this file. Returns None if the size is not yet known.
2197 """Return True if this download is using a Helper, False if not."""
2199 """Return a string describing the current state of the download
2202 """Returns a float (from 0.0 to 1.0) describing the amount of the
2203 download that has completed. This value will remain at 0.0 until the
2204 first byte of plaintext is pushed to the download target."""
2206 """Return True if the download is currently active, False if not."""
2208 """Each download status gets a unique number: this method returns
2209 that number. This provides a handle to this particular download, so a
2210 web page can generate a suitable hyperlink."""
2212 class IServermapUpdaterStatus(Interface):
2214 class IPublishStatus(Interface):
2216 class IRetrieveStatus(Interface):
2219 class NotCapableError(Exception):
2220 """You have tried to write to a read-only node."""
2222 class BadWriteEnablerError(Exception):
2225 class RIControlClient(RemoteInterface):
2227 def wait_for_client_connections(num_clients=int):
2228 """Do not return until we have connections to at least NUM_CLIENTS
2232 def upload_from_file_to_uri(filename=str,
2233 convergence=ChoiceOf(None,
2234 StringConstraint(2**20))):
2235 """Upload a file to the grid. This accepts a filename (which must be
2236 absolute) that points to a file on the node's local disk. The node will
2237 read the contents of this file, upload it to the grid, then return the
2238 URI at which it was uploaded. If convergence is None then a random
2239 encryption key will be used, else the plaintext will be hashed, then
2240 that hash will be mixed together with the "convergence" string to form
2245 def download_from_uri_to_file(uri=URI, filename=str):
2246 """Download a file from the grid, placing it on the node's local disk
2247 at the given filename (which must be absolute[?]). Returns the
2248 absolute filename where the file was written."""
2253 def get_memory_usage():
2254 """Return a dict describes the amount of memory currently in use. The
2255 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2256 measuring memory consupmtion in bytes."""
2257 return DictOf(str, int)
2259 def speed_test(count=int, size=int, mutable=Any()):
2260 """Write 'count' tempfiles to disk, all of the given size. Measure
2261 how long (in seconds) it takes to upload them all to the servers.
2262 Then measure how long it takes to download all of them. If 'mutable'
2263 is 'create', time creation of mutable files. If 'mutable' is
2264 'upload', then time access to the same mutable file instead of
2267 Returns a tuple of (upload_time, download_time).
2269 return (float, float)
2271 def measure_peer_response_time():
2272 """Send a short message to each connected peer, and measure the time
2273 it takes for them to respond to it. This is a rough measure of the
2274 application-level round trip time.
2276 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2279 return DictOf(Nodeid, float)
2281 UploadResults = Any() #DictOf(str, str)
2283 class RIEncryptedUploadable(RemoteInterface):
2284 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2289 def get_all_encoding_parameters():
2290 return (int, int, int, long)
2292 def read_encrypted(offset=Offset, length=ReadSize):
2299 class RICHKUploadHelper(RemoteInterface):
2300 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2304 Return a dictionary of version information.
2306 return DictOf(str, Any())
2308 def upload(reader=RIEncryptedUploadable):
2309 return UploadResults
2312 class RIHelper(RemoteInterface):
2313 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2317 Return a dictionary of version information.
2319 return DictOf(str, Any())
2321 def upload_chk(si=StorageIndex):
2322 """See if a file with a given storage index needs uploading. The
2323 helper will ask the appropriate storage servers to see if the file
2324 has already been uploaded. If so, the helper will return a set of
2325 'upload results' that includes whatever hashes are needed to build
2326 the read-cap, and perhaps a truncated sharemap.
2328 If the file has not yet been uploaded (or if it was only partially
2329 uploaded), the helper will return an empty upload-results dictionary
2330 and also an RICHKUploadHelper object that will take care of the
2331 upload process. The client should call upload() on this object and
2332 pass it a reference to an RIEncryptedUploadable object that will
2333 provide ciphertext. When the upload is finished, the upload() method
2334 will finish and return the upload results.
2336 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2339 class RIStatsProvider(RemoteInterface):
2340 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2342 Provides access to statistics and monitoring information.
2347 returns a dictionary containing 'counters' and 'stats', each a
2348 dictionary with string counter/stat name keys, and numeric values.
2349 counters are monotonically increasing measures of work done, and
2350 stats are instantaneous measures (potentially time averaged
2353 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2355 class RIStatsGatherer(RemoteInterface):
2356 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2358 Provides a monitoring service for centralised collection of stats
2361 def provide(provider=RIStatsProvider, nickname=str):
2363 @param provider: a stats collector instance which should be polled
2364 periodically by the gatherer to collect stats.
2365 @param nickname: a name useful to identify the provided client
2370 class IStatsProducer(Interface):
2373 returns a dictionary, with str keys representing the names of stats
2374 to be monitored, and numeric values.
2377 class RIKeyGenerator(RemoteInterface):
2378 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2380 Provides a service offering to make RSA key pairs.
2383 def get_rsa_key_pair(key_size=int):
2385 @param key_size: the size of the signature key.
2386 @return: tuple(verifying_key, signing_key)
2388 return TupleOf(str, str)
2391 class FileTooLargeError(Exception):
2394 class IValidatedThingProxy(Interface):
2396 """ Acquire a thing and validate it. Return a deferred which is
2397 eventually fired with self if the thing is valid or errbacked if it
2398 can't be acquired or validated."""
2400 class InsufficientVersionError(Exception):
2401 def __init__(self, needed, got):
2402 self.needed = needed
2405 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2408 class EmptyPathnameComponentError(Exception):
2409 """The webapi disallows empty pathname components."""