2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIStubClient(RemoteInterface):
34 """Each client publishes a service announcement for a dummy object called
35 the StubClient. This object doesn't actually offer any services, but the
36 announcement helps the Introducer keep track of which clients are
37 subscribed (so the grid admin can keep track of things like the size of
38 the grid and the client versions in use. This is the (empty)
39 RemoteInterface for the StubClient."""
41 class RIBucketWriter(RemoteInterface):
42 """ Objects of this kind live on the server side. """
43 def write(offset=Offset, data=ShareData):
48 If the data that has been written is incomplete or inconsistent then
49 the server will throw the data away, else it will store it for future
55 """Abandon all the data that has been written.
59 class RIBucketReader(RemoteInterface):
60 def read(offset=Offset, length=ReadSize):
63 def advise_corrupt_share(reason=str):
64 """Clients who discover hash failures in shares that they have
65 downloaded from me will use this method to inform me about the
66 failures. I will record their concern so that my operator can
67 manually inspect the shares in question. I return None.
69 This is a wrapper around RIStorageServer.advise_corrupt_share(),
70 which is tied to a specific share, and therefore does not need the
71 extra share-identifying arguments. Please see that method for full
75 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
76 # elements are (offset, length, operator, specimen)
77 # operator is one of "lt, le, eq, ne, ge, gt"
78 # nop always passes and is used to fetch data while writing.
79 # you should use length==len(specimen) for everything except nop
80 DataVector = ListOf(TupleOf(Offset, ShareData))
81 # (offset, data). This limits us to 30 writes of 1MiB each per call
82 TestAndWriteVectorsForShares = DictOf(int,
85 ChoiceOf(None, Offset), # new_length
87 ReadVector = ListOf(TupleOf(Offset, ReadSize))
88 ReadData = ListOf(ShareData)
89 # returns data[offset:offset+length] for each element of TestVector
91 class RIStorageServer(RemoteInterface):
92 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
96 Return a dictionary of version information.
98 return DictOf(str, Any())
100 def allocate_buckets(storage_index=StorageIndex,
101 renew_secret=LeaseRenewSecret,
102 cancel_secret=LeaseCancelSecret,
103 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
104 allocated_size=Offset, canary=Referenceable):
106 @param storage_index: the index of the bucket to be created or
108 @param sharenums: these are the share numbers (probably between 0 and
109 99) that the sender is proposing to store on this
111 @param renew_secret: This is the secret used to protect bucket refresh
112 This secret is generated by the client and
113 stored for later comparison by the server. Each
114 server is given a different secret.
115 @param cancel_secret: Like renew_secret, but protects bucket decref.
116 @param canary: If the canary is lost before close(), the bucket is
118 @return: tuple of (alreadygot, allocated), where alreadygot is what we
119 already have and allocated is what we hereby agree to accept.
120 New leases are added for shares in both lists.
122 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
123 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
125 def add_lease(storage_index=StorageIndex,
126 renew_secret=LeaseRenewSecret,
127 cancel_secret=LeaseCancelSecret):
129 Add a new lease on the given bucket. If the renew_secret matches an
130 existing lease, that lease will be renewed instead. If there is no
131 bucket for the given storage_index, return silently. (note that in
132 tahoe-1.3.0 and earlier, IndexError was raised if there was no
135 return Any() # returns None now, but future versions might change
137 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
139 Renew the lease on a given bucket, resetting the timer to 31 days.
140 Some networks will use this, some will not. If there is no bucket for
141 the given storage_index, IndexError will be raised.
143 For mutable shares, if the given renew_secret does not match an
144 existing lease, IndexError will be raised with a note listing the
145 server-nodeids on the existing leases, so leases on migrated shares
146 can be renewed or cancelled. For immutable shares, IndexError
147 (without the note) will be raised.
151 def get_buckets(storage_index=StorageIndex):
152 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
156 def slot_readv(storage_index=StorageIndex,
157 shares=ListOf(int), readv=ReadVector):
158 """Read a vector from the numbered shares associated with the given
159 storage index. An empty shares list means to return data from all
160 known shares. Returns a dictionary with one key per share."""
161 return DictOf(int, ReadData) # shnum -> results
163 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
164 secrets=TupleOf(WriteEnablerSecret,
167 tw_vectors=TestAndWriteVectorsForShares,
170 """General-purpose test-and-set operation for mutable slots. Perform
171 a bunch of comparisons against the existing shares. If they all pass,
172 then apply a bunch of write vectors to those shares. Then use the
173 read vectors to extract data from all the shares and return the data.
175 This method is, um, large. The goal is to allow clients to update all
176 the shares associated with a mutable file in a single round trip.
178 @param storage_index: the index of the bucket to be created or
180 @param write_enabler: a secret that is stored along with the slot.
181 Writes are accepted from any caller who can
182 present the matching secret. A different secret
183 should be used for each slot*server pair.
184 @param renew_secret: This is the secret used to protect bucket refresh
185 This secret is generated by the client and
186 stored for later comparison by the server. Each
187 server is given a different secret.
188 @param cancel_secret: Like renew_secret, but protects bucket decref.
190 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
191 cancel_secret). The first is required to perform any write. The
192 latter two are used when allocating new shares. To simply acquire a
193 new lease on existing shares, use an empty testv and an empty writev.
195 Each share can have a separate test vector (i.e. a list of
196 comparisons to perform). If all vectors for all shares pass, then all
197 writes for all shares are recorded. Each comparison is a 4-tuple of
198 (offset, length, operator, specimen), which effectively does a bool(
199 (read(offset, length)) OPERATOR specimen ) and only performs the
200 write if all these evaluate to True. Basic test-and-set uses 'eq'.
201 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
202 Write-if-same-or-newer uses 'le'.
204 Reads from the end of the container are truncated, and missing shares
205 behave like empty ones, so to assert that a share doesn't exist (for
206 use when creating a new share), use (0, 1, 'eq', '').
208 The write vector will be applied to the given share, expanding it if
209 necessary. A write vector applied to a share number that did not
210 exist previously will cause that share to be created.
212 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
213 vector whose offset is beyond the end of the current data, the space
214 between the end of the current data and the beginning of the write
215 vector will be filled with zero bytes. In earlier versions the
216 contents of this space was unspecified (and might end up containing
217 secrets). Storage servers with the new zero-filling behavior will
218 advertise a true value for the 'fills-holes-with-zero-bytes' key
219 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
222 Each write vector is accompanied by a 'new_length' argument, which
223 can be used to truncate the data. If new_length is not None and it is
224 less than the current size of the data (after applying all write
225 vectors), then the data will be truncated to new_length. If
226 new_length==0, the share will be deleted.
228 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
229 enlarge the file by sending a number larger than the size of the data
230 after applying all write vectors. That behavior was not used, and as
231 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
234 If a storage client knows that the server supports zero-filling, for
235 example from the 'fills-holes-with-zero-bytes' key in its version
236 information, it can extend the file efficiently by writing a single
237 zero byte just before the new end-of-file. Otherwise it must
238 explicitly write zeroes to all bytes between the old and new
239 end-of-file. In any case it should avoid sending new_length larger
240 than the size of the data after applying all write vectors.
242 The read vector is used to extract data from all known shares,
243 *before* any writes have been applied. The same vector is used for
244 all shares. This captures the state that was tested by the test
247 This method returns two values: a boolean and a dict. The boolean is
248 True if the write vectors were applied, False if not. The dict is
249 keyed by share number, and each value contains a list of strings, one
250 for each element of the read vector.
252 If the write_enabler is wrong, this will raise BadWriteEnablerError.
253 To enable share migration (using update_write_enabler), the exception
254 will have the nodeid used for the old write enabler embedded in it,
255 in the following string::
257 The write enabler was recorded by nodeid '%s'.
259 Note that the nodeid here is encoded using the same base32 encoding
260 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
263 return TupleOf(bool, DictOf(int, ReadData))
265 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
266 shnum=int, reason=str):
267 """Clients who discover hash failures in shares that they have
268 downloaded from me will use this method to inform me about the
269 failures. I will record their concern so that my operator can
270 manually inspect the shares in question. I return None.
272 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
273 (binary) storage index string, and 'shnum' is the integer share
274 number. 'reason' is a human-readable explanation of the problem,
275 probably including some expected hash values and the computed ones
276 which did not match. Corruption advisories for mutable shares should
277 include a hash of the public key (the same value that appears in the
278 mutable-file verify-cap), since the current share format does not
282 class IStorageBucketWriter(Interface):
284 Objects of this kind live on the client side.
286 def put_block(segmentnum=int, data=ShareData):
287 """@param data: For most segments, this data will be 'blocksize'
288 bytes in length. The last segment might be shorter.
289 @return: a Deferred that fires (with None) when the operation completes
292 def put_plaintext_hashes(hashes=ListOf(Hash)):
294 @return: a Deferred that fires (with None) when the operation completes
297 def put_crypttext_hashes(hashes=ListOf(Hash)):
299 @return: a Deferred that fires (with None) when the operation completes
302 def put_block_hashes(blockhashes=ListOf(Hash)):
304 @return: a Deferred that fires (with None) when the operation completes
307 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
309 @return: a Deferred that fires (with None) when the operation completes
312 def put_uri_extension(data=URIExtensionData):
313 """This block of data contains integrity-checking information (hashes
314 of plaintext, crypttext, and shares), as well as encoding parameters
315 that are necessary to recover the data. This is a serialized dict
316 mapping strings to other strings. The hash of this data is kept in
317 the URI and verified before any of the data is used. All buckets for
318 a given file contain identical copies of this data.
320 The serialization format is specified with the following pseudocode:
321 for k in sorted(dict.keys()):
322 assert re.match(r'^[a-zA-Z_\-]+$', k)
323 write(k + ':' + netstring(dict[k]))
325 @return: a Deferred that fires (with None) when the operation completes
329 """Finish writing and close the bucket. The share is not finalized
330 until this method is called: if the uploading client disconnects
331 before calling close(), the partially-written share will be
334 @return: a Deferred that fires (with None) when the operation completes
337 class IStorageBucketReader(Interface):
339 def get_block_data(blocknum=int, blocksize=int, size=int):
340 """Most blocks will be the same size. The last block might be shorter
346 def get_crypttext_hashes():
348 @return: ListOf(Hash)
351 def get_block_hashes(at_least_these=SetOf(int)):
353 @return: ListOf(Hash)
356 def get_share_hashes(at_least_these=SetOf(int)):
358 @return: ListOf(TupleOf(int, Hash))
361 def get_uri_extension():
363 @return: URIExtensionData
366 class IStorageBroker(Interface):
367 def get_servers_for_psi(peer_selection_index):
369 @return: list of IServer instances
371 def get_connected_servers():
373 @return: frozenset of connected IServer instances
375 def get_known_servers():
377 @return: frozenset of IServer instances
379 def get_all_serverids():
381 @return: frozenset of serverid strings
383 def get_nickname_for_serverid(serverid):
385 @return: unicode nickname, or None
388 # methods moved from IntroducerClient, need review
389 def get_all_connections():
390 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
391 each active connection we've established to a remote service. This is
392 mostly useful for unit tests that need to wait until a certain number
393 of connections have been made."""
395 def get_all_connectors():
396 """Return a dict that maps from (nodeid, service_name) to a
397 RemoteServiceConnector instance for all services that we are actively
398 trying to connect to. Each RemoteServiceConnector has the following
401 service_name: the type of service provided, like 'storage'
402 announcement_time: when we first heard about this service
403 last_connect_time: when we last established a connection
404 last_loss_time: when we last lost a connection
406 version: the peer's version, from the most recent connection
407 oldest_supported: the peer's oldest supported version, same
409 rref: the RemoteReference, if connected, otherwise None
410 remote_host: the IAddress, if connected, otherwise None
412 This method is intended for monitoring interfaces, such as a web page
413 which describes connecting and connected peers.
416 def get_all_peerids():
417 """Return a frozenset of all peerids to whom we have a connection (to
418 one or more services) established. Mostly useful for unit tests."""
420 def get_all_connections_for(service_name):
421 """Return a frozenset of (nodeid, service_name, rref) tuples, one
422 for each active connection that provides the given SERVICE_NAME."""
424 def get_permuted_peers(service_name, key):
425 """Returns an ordered list of (peerid, rref) tuples, selecting from
426 the connections that provide SERVICE_NAME, using a hash-based
427 permutation keyed by KEY. This randomizes the service list in a
428 repeatable way, to distribute load over many peers.
432 class IMutableSlotWriter(Interface):
434 The interface for a writer around a mutable slot on a remote server.
436 def set_checkstring(checkstring, *args):
438 Set the checkstring that I will pass to the remote server when
441 @param checkstring A packed checkstring to use.
443 Note that implementations can differ in which semantics they
444 wish to support for set_checkstring -- they can, for example,
445 build the checkstring themselves from its constituents, or
449 def get_checkstring():
451 Get the checkstring that I think currently exists on the remote
455 def put_block(data, segnum, salt):
457 Add a block and salt to the share.
460 def put_encprivey(encprivkey):
462 Add the encrypted private key to the share.
465 def put_blockhashes(blockhashes=list):
467 Add the block hash tree to the share.
470 def put_sharehashes(sharehashes=dict):
472 Add the share hash chain to the share.
477 Return the part of the share that needs to be signed.
480 def put_signature(signature):
482 Add the signature to the share.
485 def put_verification_key(verification_key):
487 Add the verification key to the share.
490 def finish_publishing():
492 Do anything necessary to finish writing the share to a remote
493 server. I require that no further publishing needs to take place
494 after this method has been called.
498 class IURI(Interface):
499 def init_from_string(uri):
500 """Accept a string (as created by my to_string() method) and populate
501 this instance with its data. I am not normally called directly,
502 please use the module-level uri.from_string() function to convert
503 arbitrary URI strings into IURI-providing instances."""
506 """Return False if this URI be used to modify the data. Return True
507 if this URI cannot be used to modify the data."""
510 """Return True if the data can be modified by *somebody* (perhaps
511 someone who has a more powerful URI than this one)."""
513 # TODO: rename to get_read_cap()
515 """Return another IURI instance, which represents a read-only form of
516 this one. If is_readonly() is True, this returns self."""
518 def get_verify_cap():
519 """Return an instance that provides IVerifierURI, which can be used
520 to check on the availability of the file or directory, without
521 providing enough capabilities to actually read or modify the
522 contents. This may return None if the file does not need checking or
523 verification (e.g. LIT URIs).
527 """Return a string of printable ASCII characters, suitable for
528 passing into init_from_string."""
530 class IVerifierURI(Interface, IURI):
531 def init_from_string(uri):
532 """Accept a string (as created by my to_string() method) and populate
533 this instance with its data. I am not normally called directly,
534 please use the module-level uri.from_string() function to convert
535 arbitrary URI strings into IURI-providing instances."""
538 """Return a string of printable ASCII characters, suitable for
539 passing into init_from_string."""
541 class IDirnodeURI(Interface):
542 """I am a URI which represents a dirnode."""
544 class IFileURI(Interface):
545 """I am a URI which represents a filenode."""
547 """Return the length (in bytes) of the file that I represent."""
549 class IImmutableFileURI(IFileURI):
552 class IMutableFileURI(Interface):
553 """I am a URI which represents a mutable filenode."""
554 def get_extension_params():
555 """Return the extension parameters in the URI"""
557 def set_extension_params():
558 """Set the extension parameters that should be in the URI"""
560 class IDirectoryURI(Interface):
563 class IReadonlyDirectoryURI(Interface):
566 class CapConstraintError(Exception):
567 """A constraint on a cap was violated."""
569 class MustBeDeepImmutableError(CapConstraintError):
570 """Mutable children cannot be added to an immutable directory.
571 Also, caps obtained from an immutable directory can trigger this error
572 if they are later found to refer to a mutable object and then used."""
574 class MustBeReadonlyError(CapConstraintError):
575 """Known write caps cannot be specified in a ro_uri field. Also,
576 caps obtained from a ro_uri field can trigger this error if they
577 are later found to be write caps and then used."""
579 class MustNotBeUnknownRWError(CapConstraintError):
580 """Cannot add an unknown child cap specified in a rw_uri field."""
583 class IReadable(Interface):
584 """I represent a readable object -- either an immutable file, or a
585 specific version of a mutable file.
589 """Return True if this reference provides mutable access to the given
590 file or directory (i.e. if you can modify it), or False if not. Note
591 that even if this reference is read-only, someone else may hold a
592 read-write reference to it.
594 For an IReadable returned by get_best_readable_version(), this will
595 always return True, but for instances of subinterfaces such as
596 IMutableFileVersion, it may return False."""
599 """Return True if this file or directory is mutable (by *somebody*,
600 not necessarily you), False if it is is immutable. Note that a file
601 might be mutable overall, but your reference to it might be
602 read-only. On the other hand, all references to an immutable file
603 will be read-only; there are no read-write references to an immutable
606 def get_storage_index():
607 """Return the storage index of the file."""
610 """Return the length (in bytes) of this readable object."""
612 def download_to_data():
613 """Download all of the file contents. I return a Deferred that fires
614 with the contents as a byte string."""
616 def read(consumer, offset=0, size=None):
617 """Download a portion (possibly all) of the file's contents, making
618 them available to the given IConsumer. Return a Deferred that fires
619 (with the consumer) when the consumer is unregistered (either because
620 the last byte has been given to it, or because the consumer threw an
621 exception during write(), possibly because it no longer wants to
622 receive data). The portion downloaded will start at 'offset' and
623 contain 'size' bytes (or the remainder of the file if size==None).
625 The consumer will be used in non-streaming mode: an IPullProducer
626 will be attached to it.
628 The consumer will not receive data right away: several network trips
629 must occur first. The order of events will be::
631 consumer.registerProducer(p, streaming)
632 (if streaming == False)::
633 consumer does p.resumeProducing()
635 consumer does p.resumeProducing()
636 consumer.write(data).. (repeat until all data is written)
637 consumer.unregisterProducer()
638 deferred.callback(consumer)
640 If a download error occurs, or an exception is raised by
641 consumer.registerProducer() or consumer.write(), I will call
642 consumer.unregisterProducer() and then deliver the exception via
643 deferred.errback(). To cancel the download, the consumer should call
644 p.stopProducing(), which will result in an exception being delivered
645 via deferred.errback().
647 See src/allmydata/util/consumer.py for an example of a simple
648 download-to-memory consumer.
652 class IWriteable(Interface):
654 I define methods that callers can use to update SDMF and MDMF
655 mutable files on a Tahoe-LAFS grid.
657 # XXX: For the moment, we have only this. It is possible that we
658 # want to move overwrite() and modify() in here too.
659 def update(data, offset):
661 I write the data from my data argument to the MDMF file,
662 starting at offset. I continue writing data until my data
663 argument is exhausted, appending data to the file as necessary.
665 # assert IMutableUploadable.providedBy(data)
666 # to append data: offset=node.get_size_of_best_version()
667 # do we want to support compacting MDMF?
668 # for an MDMF file, this can be done with O(data.get_size())
669 # memory. For an SDMF file, any modification takes
670 # O(node.get_size_of_best_version()).
673 class IMutableFileVersion(IReadable):
674 """I provide access to a particular version of a mutable file. The
675 access is read/write if I was obtained from a filenode derived from
676 a write cap, or read-only if the filenode was derived from a read cap.
679 def get_sequence_number():
680 """Return the sequence number of this version."""
683 """Return the IMutableFileServerMap instance that was used to create
688 """Return this filenode's writekey, or None if the node does not have
689 write-capability. This may be used to assist with data structures
690 that need to make certain data available only to writers, such as the
691 read-write child caps in dirnodes. The recommended process is to have
692 reader-visible data be submitted to the filenode in the clear (where
693 it will be encrypted by the filenode using the readkey), but encrypt
694 writer-visible data using this writekey.
697 # TODO: Can this be overwrite instead of replace?
698 def replace(new_contents):
699 """Replace the contents of the mutable file, provided that no other
700 node has published (or is attempting to publish, concurrently) a
701 newer version of the file than this one.
703 I will avoid modifying any share that is different than the version
704 given by get_sequence_number(). However, if another node is writing
705 to the file at the same time as me, I may manage to update some shares
706 while they update others. If I see any evidence of this, I will signal
707 UncoordinatedWriteError, and the file will be left in an inconsistent
708 state (possibly the version you provided, possibly the old version,
709 possibly somebody else's version, and possibly a mix of shares from
712 The recommended response to UncoordinatedWriteError is to either
713 return it to the caller (since they failed to coordinate their
714 writes), or to attempt some sort of recovery. It may be sufficient to
715 wait a random interval (with exponential backoff) and repeat your
716 operation. If I do not signal UncoordinatedWriteError, then I was
717 able to write the new version without incident.
719 I return a Deferred that fires (with a PublishStatus object) when the
720 update has completed.
723 def modify(modifier_cb):
724 """Modify the contents of the file, by downloading this version,
725 applying the modifier function (or bound method), then uploading
726 the new version. This will succeed as long as no other node
727 publishes a version between the download and the upload.
728 I return a Deferred that fires (with a PublishStatus object) when
729 the update is complete.
731 The modifier callable will be given three arguments: a string (with
732 the old contents), a 'first_time' boolean, and a servermap. As with
733 download_to_data(), the old contents will be from this version,
734 but the modifier can use the servermap to make other decisions
735 (such as refusing to apply the delta if there are multiple parallel
736 versions, or if there is evidence of a newer unrecoverable version).
737 'first_time' will be True the first time the modifier is called,
738 and False on any subsequent calls.
740 The callable should return a string with the new contents. The
741 callable must be prepared to be called multiple times, and must
742 examine the input string to see if the change that it wants to make
743 is already present in the old version. If it does not need to make
744 any changes, it can either return None, or return its input string.
746 If the modifier raises an exception, it will be returned in the
751 # The hierarchy looks like this:
758 class IFilesystemNode(Interface):
760 """Return the strongest 'cap instance' associated with this node.
761 (writecap for writeable-mutable files/directories, readcap for
762 immutable or readonly-mutable files/directories). To convert this
763 into a string, call .to_string() on the result."""
766 """Return a readonly cap instance for this node. For immutable or
767 readonly nodes, get_cap() and get_readcap() return the same thing."""
769 def get_repair_cap():
770 """Return an IURI instance that can be used to repair the file, or
771 None if this node cannot be repaired (either because it is not
772 distributed, like a LIT file, or because the node does not represent
773 sufficient authority to create a repair-cap, like a read-only RSA
774 mutable file node [which cannot create the correct write-enablers]).
777 def get_verify_cap():
778 """Return an IVerifierURI instance that represents the
779 'verifiy/refresh capability' for this node. The holder of this
780 capability will be able to renew the lease for this node, protecting
781 it from garbage-collection. They will also be able to ask a server if
782 it holds a share for the file or directory.
786 """Return the URI string corresponding to the strongest cap associated
787 with this node. If this node is read-only, the URI will only offer
788 read-only access. If this node is read-write, the URI will offer
791 If you have read-write access to a node and wish to share merely
792 read-only access with others, use get_readonly_uri().
796 """Return the URI string that can be used by others to get write
797 access to this node, if it is writeable. If this is a read-only node,
800 def get_readonly_uri():
801 """Return the URI string that can be used by others to get read-only
802 access to this node. The result is a read-only URI, regardless of
803 whether this node is read-only or read-write.
805 If you have merely read-only access to this node, get_readonly_uri()
806 will return the same thing as get_uri().
809 def get_storage_index():
810 """Return a string with the (binary) storage index in use on this
811 download. This may be None if there is no storage index (i.e. LIT
812 files and directories)."""
815 """Return True if this reference provides mutable access to the given
816 file or directory (i.e. if you can modify it), or False if not. Note
817 that even if this reference is read-only, someone else may hold a
818 read-write reference to it."""
821 """Return True if this file or directory is mutable (by *somebody*,
822 not necessarily you), False if it is is immutable. Note that a file
823 might be mutable overall, but your reference to it might be
824 read-only. On the other hand, all references to an immutable file
825 will be read-only; there are no read-write references to an immutable
830 """Return True if this is an unknown node."""
832 def is_allowed_in_immutable_directory():
833 """Return True if this node is allowed as a child of a deep-immutable
834 directory. This is true if either the node is of a known-immutable type,
835 or it is unknown and read-only.
839 """Raise any error associated with this node."""
841 # XXX: These may not be appropriate outside the context of an IReadable.
843 """Return the length (in bytes) of the data this node represents. For
844 directory nodes, I return the size of the backing store. I return
845 synchronously and do not consult the network, so for mutable objects,
846 I will return the most recently observed size for the object, or None
847 if I don't remember a size. Use get_current_size, which returns a
848 Deferred, if you want more up-to-date information."""
850 def get_current_size():
851 """I return a Deferred that fires with the length (in bytes) of the
852 data this node represents.
855 class IFileNode(IFilesystemNode):
856 """I am a node which represents a file: a sequence of bytes. I am not a
857 container, like IDirectoryNode."""
858 def get_best_readable_version():
859 """Return a Deferred that fires with an IReadable for the 'best'
860 available version of the file. The IReadable provides only read
861 access, even if this filenode was derived from a write cap.
863 For an immutable file, there is only one version. For a mutable
864 file, the 'best' version is the recoverable version with the
865 highest sequence number. If no uncoordinated writes have occurred,
866 and if enough shares are available, then this will be the most
867 recent version that has been uploaded. If no version is recoverable,
868 the Deferred will errback with an UnrecoverableFileError.
871 def download_best_version():
872 """Download the contents of the version that would be returned
873 by get_best_readable_version(). This is equivalent to calling
874 download_to_data() on the IReadable given by that method.
876 I return a Deferred that fires with a byte string when the file
877 has been fully downloaded. To support streaming download, use
878 the 'read' method of IReadable. If no version is recoverable,
879 the Deferred will errback with an UnrecoverableFileError.
882 def get_size_of_best_version():
883 """Find the size of the version that would be returned by
884 get_best_readable_version().
886 I return a Deferred that fires with an integer. If no version
887 is recoverable, the Deferred will errback with an
888 UnrecoverableFileError.
892 class IImmutableFileNode(IFileNode, IReadable):
893 """I am a node representing an immutable file. Immutable files have
897 class IMutableFileNode(IFileNode):
898 """I provide access to a 'mutable file', which retains its identity
899 regardless of what contents are put in it.
901 The consistency-vs-availability problem means that there might be
902 multiple versions of a file present in the grid, some of which might be
903 unrecoverable (i.e. have fewer than 'k' shares). These versions are
904 loosely ordered: each has a sequence number and a hash, and any version
905 with seqnum=N was uploaded by a node which has seen at least one version
908 The 'servermap' (an instance of IMutableFileServerMap) is used to
909 describe the versions that are known to be present in the grid, and which
910 servers are hosting their shares. It is used to represent the 'state of
911 the world', and is used for this purpose by my test-and-set operations.
912 Downloading the contents of the mutable file will also return a
913 servermap. Uploading a new version into the mutable file requires a
914 servermap as input, and the semantics of the replace operation is
915 'replace the file with my new version if it looks like nobody else has
916 changed the file since my previous download'. Because the file is
917 distributed, this is not a perfect test-and-set operation, but it will do
918 its best. If the replace process sees evidence of a simultaneous write,
919 it will signal an UncoordinatedWriteError, so that the caller can take
923 Most readers will want to use the 'best' current version of the file, and
924 should use my 'download_best_version()' method.
926 To unconditionally replace the file, callers should use overwrite(). This
927 is the mode that user-visible mutable files will probably use.
929 To apply some delta to the file, call modify() with a callable modifier
930 function that can apply the modification that you want to make. This is
931 the mode that dirnodes will use, since most directory modification
932 operations can be expressed in terms of deltas to the directory state.
935 Three methods are available for users who need to perform more complex
936 operations. The first is get_servermap(), which returns an up-to-date
937 servermap using a specified mode. The second is download_version(), which
938 downloads a specific version (not necessarily the 'best' one). The third
939 is 'upload', which accepts new contents and a servermap (which must have
940 been updated with MODE_WRITE). The upload method will attempt to apply
941 the new contents as long as no other node has modified the file since the
942 servermap was updated. This might be useful to a caller who wants to
943 merge multiple versions into a single new one.
945 Note that each time the servermap is updated, a specific 'mode' is used,
946 which determines how many peers are queried. To use a servermap for my
947 replace() method, that servermap must have been updated in MODE_WRITE.
948 These modes are defined in allmydata.mutable.common, and consist of
949 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
950 allmydata/mutable/servermap.py for details about the differences.
952 Mutable files are currently limited in size (about 3.5MB max) and can
953 only be retrieved and updated all-at-once, as a single big string. Future
954 versions of our mutable files will remove this restriction.
956 def get_best_mutable_version():
957 """Return a Deferred that fires with an IMutableFileVersion for
958 the 'best' available version of the file. The best version is
959 the recoverable version with the highest sequence number. If no
960 uncoordinated writes have occurred, and if enough shares are
961 available, then this will be the most recent version that has
964 If no version is recoverable, the Deferred will errback with an
965 UnrecoverableFileError.
968 def overwrite(new_contents):
969 """Unconditionally replace the contents of the mutable file with new
970 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
971 is only appropriate to use when the new contents of the file are
972 completely unrelated to the old ones, and you do not care about other
975 I return a Deferred that fires (with a PublishStatus object) when the
976 update has completed.
979 def modify(modifier_cb):
980 """Modify the contents of the file, by downloading the current
981 version, applying the modifier function (or bound method), then
982 uploading the new version. I return a Deferred that fires (with a
983 PublishStatus object) when the update is complete.
985 The modifier callable will be given three arguments: a string (with
986 the old contents), a 'first_time' boolean, and a servermap. As with
987 download_best_version(), the old contents will be from the best
988 recoverable version, but the modifier can use the servermap to make
989 other decisions (such as refusing to apply the delta if there are
990 multiple parallel versions, or if there is evidence of a newer
991 unrecoverable version). 'first_time' will be True the first time the
992 modifier is called, and False on any subsequent calls.
994 The callable should return a string with the new contents. The
995 callable must be prepared to be called multiple times, and must
996 examine the input string to see if the change that it wants to make
997 is already present in the old version. If it does not need to make
998 any changes, it can either return None, or return its input string.
1000 If the modifier raises an exception, it will be returned in the
1004 def get_servermap(mode):
1005 """Return a Deferred that fires with an IMutableFileServerMap
1006 instance, updated using the given mode.
1009 def download_version(servermap, version):
1010 """Download a specific version of the file, using the servermap
1011 as a guide to where the shares are located.
1013 I return a Deferred that fires with the requested contents, or
1014 errbacks with UnrecoverableFileError. Note that a servermap which was
1015 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1016 all versions (those modes stop querying servers as soon as they can
1017 fulfil their goals), so you may want to use MODE_CHECK (which checks
1018 everything) to get increased visibility.
1021 def upload(new_contents, servermap):
1022 """Replace the contents of the file with new ones. This requires a
1023 servermap that was previously updated with MODE_WRITE.
1025 I attempt to provide test-and-set semantics, in that I will avoid
1026 modifying any share that is different than the version I saw in the
1027 servermap. However, if another node is writing to the file at the
1028 same time as me, I may manage to update some shares while they update
1029 others. If I see any evidence of this, I will signal
1030 UncoordinatedWriteError, and the file will be left in an inconsistent
1031 state (possibly the version you provided, possibly the old version,
1032 possibly somebody else's version, and possibly a mix of shares from
1035 The recommended response to UncoordinatedWriteError is to either
1036 return it to the caller (since they failed to coordinate their
1037 writes), or to attempt some sort of recovery. It may be sufficient to
1038 wait a random interval (with exponential backoff) and repeat your
1039 operation. If I do not signal UncoordinatedWriteError, then I was
1040 able to write the new version without incident.
1042 I return a Deferred that fires (with a PublishStatus object) when the
1043 publish has completed. I will update the servermap in-place with the
1044 location of all new shares.
1048 """Return this filenode's writekey, or None if the node does not have
1049 write-capability. This may be used to assist with data structures
1050 that need to make certain data available only to writers, such as the
1051 read-write child caps in dirnodes. The recommended process is to have
1052 reader-visible data be submitted to the filenode in the clear (where
1053 it will be encrypted by the filenode using the readkey), but encrypt
1054 writer-visible data using this writekey.
1058 """Returns the mutable file protocol version."""
1060 class NotEnoughSharesError(Exception):
1061 """Download was unable to get enough shares"""
1063 class NoSharesError(Exception):
1064 """Download was unable to get any shares at all."""
1066 class DownloadStopped(Exception):
1069 class UploadUnhappinessError(Exception):
1070 """Upload was unable to satisfy 'servers_of_happiness'"""
1072 class UnableToFetchCriticalDownloadDataError(Exception):
1073 """I was unable to fetch some piece of critical data which is supposed to
1074 be identically present in all shares."""
1076 class NoServersError(Exception):
1077 """Upload wasn't given any servers to work with, usually indicating a
1078 network or Introducer problem."""
1080 class ExistingChildError(Exception):
1081 """A directory node was asked to add or replace a child that already
1082 exists, and overwrite= was set to False."""
1084 class NoSuchChildError(Exception):
1085 """A directory node was asked to fetch a child which does not exist."""
1087 class ChildOfWrongTypeError(Exception):
1088 """An operation was attempted on a child of the wrong type (file or directory)."""
1090 class IDirectoryNode(IFilesystemNode):
1091 """I represent a filesystem node that is a container, with a
1092 name-to-child mapping, holding the tahoe equivalent of a directory. All
1093 child names are unicode strings, and all children are some sort of
1094 IFilesystemNode (a file, subdirectory, or unknown node).
1099 The dirnode ('1') URI returned by this method can be used in
1100 set_uri() on a different directory ('2') to 'mount' a reference to
1101 this directory ('1') under the other ('2'). This URI is just a
1102 string, so it can be passed around through email or other out-of-band
1106 def get_readonly_uri():
1108 The dirnode ('1') URI returned by this method can be used in
1109 set_uri() on a different directory ('2') to 'mount' a reference to
1110 this directory ('1') under the other ('2'). This URI is just a
1111 string, so it can be passed around through email or other out-of-band
1116 """I return a Deferred that fires with a dictionary mapping child
1117 name (a unicode string) to (node, metadata_dict) tuples, in which
1118 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1121 def has_child(name):
1122 """I return a Deferred that fires with a boolean, True if there
1123 exists a child of the given name, False if not. The child name must
1124 be a unicode string."""
1127 """I return a Deferred that fires with a specific named child node,
1128 which is an IFilesystemNode. The child name must be a unicode string.
1129 I raise NoSuchChildError if I do not have a child by that name."""
1131 def get_metadata_for(name):
1132 """I return a Deferred that fires with the metadata dictionary for
1133 a specific named child node. The child name must be a unicode string.
1134 This metadata is stored in the *edge*, not in the child, so it is
1135 attached to the parent dirnode rather than the child node.
1136 I raise NoSuchChildError if I do not have a child by that name."""
1138 def set_metadata_for(name, metadata):
1139 """I replace any existing metadata for the named child with the new
1140 metadata. The child name must be a unicode string. This metadata is
1141 stored in the *edge*, not in the child, so it is attached to the
1142 parent dirnode rather than the child node. I return a Deferred
1143 (that fires with this dirnode) when the operation is complete.
1144 I raise NoSuchChildError if I do not have a child by that name."""
1146 def get_child_at_path(path):
1147 """Transform a child path into an IFilesystemNode.
1149 I perform a recursive series of 'get' operations to find the named
1150 descendant node. I return a Deferred that fires with the node, or
1151 errbacks with NoSuchChildError if the node could not be found.
1153 The path can be either a single string (slash-separated) or a list of
1154 path-name elements. All elements must be unicode strings.
1157 def get_child_and_metadata_at_path(path):
1158 """Transform a child path into an IFilesystemNode and metadata.
1160 I am like get_child_at_path(), but my Deferred fires with a tuple of
1161 (node, metadata). The metadata comes from the last edge. If the path
1162 is empty, the metadata will be an empty dictionary.
1165 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1166 """I add a child (by writecap+readcap) at the specific name. I return
1167 a Deferred that fires when the operation finishes. If overwrite= is
1168 True, I will replace any existing child of the same name, otherwise
1169 an existing child will cause me to return ExistingChildError. The
1170 child name must be a unicode string.
1172 The child caps could be for a file, or for a directory. If you have
1173 both the writecap and readcap, you should provide both arguments.
1174 If you have only one cap and don't know whether it is read-only,
1175 provide it as the writecap argument and leave the readcap as None.
1176 If you have only one cap that is known to be read-only, provide it
1177 as the readcap argument and leave the writecap as None.
1178 The filecaps are typically obtained from an IFilesystemNode with
1179 get_uri() and get_readonly_uri().
1181 If metadata= is provided, I will use it as the metadata for the named
1182 edge. This will replace any existing metadata. If metadata= is left
1183 as the default value of None, I will set ['mtime'] to the current
1184 time, and I will set ['ctime'] to the current time if there was not
1185 already a child by this name present. This roughly matches the
1186 ctime/mtime semantics of traditional filesystems. See the
1187 "About the metadata" section of webapi.txt for futher information.
1189 If this directory node is read-only, the Deferred will errback with a
1190 NotWriteableError."""
1192 def set_children(entries, overwrite=True):
1193 """Add multiple children (by writecap+readcap) to a directory node.
1194 Takes a dictionary, with childname as keys and (writecap, readcap)
1195 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1196 a Deferred that fires (with this dirnode) when the operation
1197 finishes. This is equivalent to calling set_uri() multiple times, but
1198 is much more efficient. All child names must be unicode strings.
1201 def set_node(name, child, metadata=None, overwrite=True):
1202 """I add a child at the specific name. I return a Deferred that fires
1203 when the operation finishes. This Deferred will fire with the child
1204 node that was just added. I will replace any existing child of the
1205 same name. The child name must be a unicode string. The 'child'
1206 instance must be an instance providing IFilesystemNode.
1208 If metadata= is provided, I will use it as the metadata for the named
1209 edge. This will replace any existing metadata. If metadata= is left
1210 as the default value of None, I will set ['mtime'] to the current
1211 time, and I will set ['ctime'] to the current time if there was not
1212 already a child by this name present. This roughly matches the
1213 ctime/mtime semantics of traditional filesystems. See the
1214 "About the metadata" section of webapi.txt for futher information.
1216 If this directory node is read-only, the Deferred will errback with a
1217 NotWriteableError."""
1219 def set_nodes(entries, overwrite=True):
1220 """Add multiple children to a directory node. Takes a dict mapping
1221 unicode childname to (child_node, metdata) tuples. If metdata=None,
1222 the original metadata is left unmodified. Returns a Deferred that
1223 fires (with this dirnode) when the operation finishes. This is
1224 equivalent to calling set_node() multiple times, but is much more
1227 def add_file(name, uploadable, metadata=None, overwrite=True):
1228 """I upload a file (using the given IUploadable), then attach the
1229 resulting ImmutableFileNode to the directory at the given name. I set
1230 metadata the same way as set_uri and set_node. The child name must be
1233 I return a Deferred that fires (with the IFileNode of the uploaded
1234 file) when the operation completes."""
1236 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1237 """I remove the child at the specific name. I return a Deferred that
1238 fires when the operation finishes. The child name must be a unicode
1239 string. If must_exist is True and I do not have a child by that name,
1240 I raise NoSuchChildError. If must_be_directory is True and the child
1241 is a file, or if must_be_file is True and the child is a directory,
1242 I raise ChildOfWrongTypeError."""
1244 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1245 """I create and attach a directory at the given name. The new
1246 directory can be empty, or it can be populated with children
1247 according to 'initial_children', which takes a dictionary in the same
1248 format as set_nodes (i.e. mapping unicode child name to (childnode,
1249 metadata) tuples). The child name must be a unicode string. I return
1250 a Deferred that fires (with the new directory node) when the
1251 operation finishes."""
1253 def move_child_to(current_child_name, new_parent, new_child_name=None,
1255 """I take one of my children and move them to a new parent. The child
1256 is referenced by name. On the new parent, the child will live under
1257 'new_child_name', which defaults to 'current_child_name'. TODO: what
1258 should we do about metadata? I return a Deferred that fires when the
1259 operation finishes. The child name must be a unicode string. I raise
1260 NoSuchChildError if I do not have a child by that name."""
1262 def build_manifest():
1263 """I generate a table of everything reachable from this directory.
1264 I also compute deep-stats as described below.
1266 I return a Monitor. The Monitor's results will be a dictionary with
1269 res['manifest']: a list of (path, cap) tuples for all nodes
1270 (directories and files) reachable from this one.
1271 'path' will be a tuple of unicode strings. The
1272 origin dirnode will be represented by an empty path
1274 res['verifycaps']: a list of (printable) verifycap strings, one for
1275 each reachable non-LIT node. This is a set:
1276 it will contain no duplicates.
1277 res['storage-index']: a list of (base32) storage index strings,
1278 one for each reachable non-LIT node. This is
1279 a set: it will contain no duplicates.
1280 res['stats']: a dictionary, the same that is generated by
1281 start_deep_stats() below.
1283 The Monitor will also have an .origin_si attribute with the (binary)
1284 storage index of the starting point.
1287 def start_deep_stats():
1288 """Return a Monitor, examining all nodes (directories and files)
1289 reachable from this one. The Monitor's results will be a dictionary
1290 with the following keys::
1292 count-immutable-files: count of how many CHK files are in the set
1293 count-mutable-files: same, for mutable files (does not include
1295 count-literal-files: same, for LIT files
1296 count-files: sum of the above three
1298 count-directories: count of directories
1300 size-immutable-files: total bytes for all CHK files in the set
1301 size-mutable-files (TODO): same, for current version of all mutable
1302 files, does not include directories
1303 size-literal-files: same, for LIT files
1304 size-directories: size of mutable files used by directories
1306 largest-directory: number of bytes in the largest directory
1307 largest-directory-children: number of children in the largest
1309 largest-immutable-file: number of bytes in the largest CHK file
1311 size-mutable-files is not yet implemented, because it would involve
1312 even more queries than deep_stats does.
1314 The Monitor will also have an .origin_si attribute with the (binary)
1315 storage index of the starting point.
1317 This operation will visit every directory node underneath this one,
1318 and can take a long time to run. On a typical workstation with good
1319 bandwidth, this can examine roughly 15 directories per second (and
1320 takes several minutes of 100% CPU for ~1700 directories).
1323 class ICodecEncoder(Interface):
1324 def set_params(data_size, required_shares, max_shares):
1325 """Set up the parameters of this encoder.
1327 This prepares the encoder to perform an operation that converts a
1328 single block of data into a number of shares, such that a future
1329 ICodecDecoder can use a subset of these shares to recover the
1330 original data. This operation is invoked by calling encode(). Once
1331 the encoding parameters are set up, the encode operation can be
1332 invoked multiple times.
1334 set_params() prepares the encoder to accept blocks of input data that
1335 are exactly 'data_size' bytes in length. The encoder will be prepared
1336 to produce 'max_shares' shares for each encode() operation (although
1337 see the 'desired_share_ids' to use less CPU). The encoding math will
1338 be chosen such that the decoder can get by with as few as
1339 'required_shares' of these shares and still reproduce the original
1340 data. For example, set_params(1000, 5, 5) offers no redundancy at
1341 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1343 Numerical Restrictions: 'data_size' is required to be an integral
1344 multiple of 'required_shares'. In general, the caller should choose
1345 required_shares and max_shares based upon their reliability
1346 requirements and the number of peers available (the total storage
1347 space used is roughly equal to max_shares*data_size/required_shares),
1348 then choose data_size to achieve the memory footprint desired (larger
1349 data_size means more efficient operation, smaller data_size means
1350 smaller memory footprint).
1352 In addition, 'max_shares' must be equal to or greater than
1353 'required_shares'. Of course, setting them to be equal causes
1354 encode() to degenerate into a particularly slow form of the 'split'
1357 See encode() for more details about how these parameters are used.
1359 set_params() must be called before any other ICodecEncoder methods
1364 """Return the 3-tuple of data_size, required_shares, max_shares"""
1366 def get_encoder_type():
1367 """Return a short string that describes the type of this encoder.
1369 There is required to be a global table of encoder classes. This method
1370 returns an index into this table; the value at this index is an
1371 encoder class, and this encoder is an instance of that class.
1374 def get_block_size():
1375 """Return the length of the shares that encode() will produce.
1378 def encode_proposal(data, desired_share_ids=None):
1379 """Encode some data.
1381 'data' must be a string (or other buffer object), and len(data) must
1382 be equal to the 'data_size' value passed earlier to set_params().
1384 This will return a Deferred that will fire with two lists. The first
1385 is a list of shares, each of which is a string (or other buffer
1386 object) such that len(share) is the same as what get_share_size()
1387 returned earlier. The second is a list of shareids, in which each is
1388 an integer. The lengths of the two lists will always be equal to each
1389 other. The user should take care to keep each share closely
1390 associated with its shareid, as one is useless without the other.
1392 The length of this output list will normally be the same as the value
1393 provided to the 'max_shares' parameter of set_params(). This may be
1394 different if 'desired_share_ids' is provided.
1396 'desired_share_ids', if provided, is required to be a sequence of
1397 ints, each of which is required to be >= 0 and < max_shares. If not
1398 provided, encode() will produce 'max_shares' shares, as if
1399 'desired_share_ids' were set to range(max_shares). You might use this
1400 if you initially thought you were going to use 10 peers, started
1401 encoding, and then two of the peers dropped out: you could use
1402 desired_share_ids= to skip the work (both memory and CPU) of
1403 producing shares for the peers which are no longer available.
1407 def encode(inshares, desired_share_ids=None):
1408 """Encode some data. This may be called multiple times. Each call is
1411 inshares is a sequence of length required_shares, containing buffers
1412 (i.e. strings), where each buffer contains the next contiguous
1413 non-overlapping segment of the input data. Each buffer is required to
1414 be the same length, and the sum of the lengths of the buffers is
1415 required to be exactly the data_size promised by set_params(). (This
1416 implies that the data has to be padded before being passed to
1417 encode(), unless of course it already happens to be an even multiple
1418 of required_shares in length.)
1420 Note: the requirement to break up your data into
1421 'required_shares' chunks of exactly the right length before
1422 calling encode() is surprising from point of view of a user
1423 who doesn't know how FEC works. It feels like an
1424 implementation detail that has leaked outside the abstraction
1425 barrier. Is there a use case in which the data to be encoded
1426 might already be available in pre-segmented chunks, such that
1427 it is faster or less work to make encode() take a list rather
1428 than splitting a single string?
1430 Yes, there is: suppose you are uploading a file with K=64,
1431 N=128, segsize=262,144. Then each in-share will be of size
1432 4096. If you use this .encode() API then your code could first
1433 read each successive 4096-byte chunk from the file and store
1434 each one in a Python string and store each such Python string
1435 in a Python list. Then you could call .encode(), passing that
1436 list as "inshares". The encoder would generate the other 64
1437 "secondary shares" and return to you a new list containing
1438 references to the same 64 Python strings that you passed in
1439 (as the primary shares) plus references to the new 64 Python
1442 (You could even imagine that your code could use readv() so
1443 that the operating system can arrange to get all of those
1444 bytes copied from the file into the Python list of Python
1445 strings as efficiently as possible instead of having a loop
1446 written in C or in Python to copy the next part of the file
1447 into the next string.)
1449 On the other hand if you instead use the .encode_proposal()
1450 API (above), then your code can first read in all of the
1451 262,144 bytes of the segment from the file into a Python
1452 string, then call .encode_proposal() passing the segment data
1453 as the "data" argument. The encoder would basically first
1454 split the "data" argument into a list of 64 in-shares of 4096
1455 byte each, and then do the same thing that .encode() does. So
1456 this would result in a little bit more copying of data and a
1457 little bit higher of a "maximum memory usage" during the
1458 process, although it might or might not make a practical
1459 difference for our current use cases.
1461 Note that "inshares" is a strange name for the parameter if
1462 you think of the parameter as being just for feeding in data
1463 to the codec. It makes more sense if you think of the result
1464 of this encoding as being the set of shares from inshares plus
1465 an extra set of "secondary shares" (or "check shares"). It is
1466 a surprising name! If the API is going to be surprising then
1467 the name should be surprising. If we switch to
1468 encode_proposal() above then we should also switch to an
1471 'desired_share_ids', if provided, is required to be a sequence of
1472 ints, each of which is required to be >= 0 and < max_shares. If not
1473 provided, encode() will produce 'max_shares' shares, as if
1474 'desired_share_ids' were set to range(max_shares). You might use this
1475 if you initially thought you were going to use 10 peers, started
1476 encoding, and then two of the peers dropped out: you could use
1477 desired_share_ids= to skip the work (both memory and CPU) of
1478 producing shares for the peers which are no longer available.
1480 For each call, encode() will return a Deferred that fires with two
1481 lists, one containing shares and the other containing the shareids.
1482 The get_share_size() method can be used to determine the length of
1483 the share strings returned by encode(). Each shareid is a small
1484 integer, exactly as passed into 'desired_share_ids' (or
1485 range(max_shares), if desired_share_ids was not provided).
1487 The shares and their corresponding shareids are required to be kept
1488 together during storage and retrieval. Specifically, the share data is
1489 useless by itself: the decoder needs to be told which share is which
1490 by providing it with both the shareid and the actual share data.
1492 This function will allocate an amount of memory roughly equal to::
1494 (max_shares - required_shares) * get_share_size()
1496 When combined with the memory that the caller must allocate to
1497 provide the input data, this leads to a memory footprint roughly
1498 equal to the size of the resulting encoded shares (i.e. the expansion
1499 factor times the size of the input segment).
1504 # returning a list of (shareidN,shareN) tuples instead of a pair of
1505 # lists (shareids..,shares..). Brian thought the tuples would
1506 # encourage users to keep the share and shareid together throughout
1507 # later processing, Zooko pointed out that the code to iterate
1508 # through two lists is not really more complicated than using a list
1509 # of tuples and there's also a performance improvement
1511 # having 'data_size' not required to be an integral multiple of
1512 # 'required_shares'. Doing this would require encode() to perform
1513 # padding internally, and we'd prefer to have any padding be done
1514 # explicitly by the caller. Yes, it is an abstraction leak, but
1515 # hopefully not an onerous one.
1518 class ICodecDecoder(Interface):
1519 def set_params(data_size, required_shares, max_shares):
1520 """Set the params. They have to be exactly the same ones that were
1521 used for encoding."""
1523 def get_needed_shares():
1524 """Return the number of shares needed to reconstruct the data.
1525 set_params() is required to be called before this."""
1527 def decode(some_shares, their_shareids):
1528 """Decode a partial list of shares into data.
1530 'some_shares' is required to be a sequence of buffers of sharedata, a
1531 subset of the shares returned by ICodecEncode.encode(). Each share is
1532 required to be of the same length. The i'th element of their_shareids
1533 is required to be the shareid of the i'th buffer in some_shares.
1535 This returns a Deferred which fires with a sequence of buffers. This
1536 sequence will contain all of the segments of the original data, in
1537 order. The sum of the lengths of all of the buffers will be the
1538 'data_size' value passed into the original ICodecEncode.set_params()
1539 call. To get back the single original input block of data, use
1540 ''.join(output_buffers), or you may wish to simply write them in
1541 order to an output file.
1543 Note that some of the elements in the result sequence may be
1544 references to the elements of the some_shares input sequence. In
1545 particular, this means that if those share objects are mutable (e.g.
1546 arrays) and if they are changed, then both the input (the
1547 'some_shares' parameter) and the output (the value given when the
1548 deferred is triggered) will change.
1550 The length of 'some_shares' is required to be exactly the value of
1551 'required_shares' passed into the original ICodecEncode.set_params()
1555 class IEncoder(Interface):
1556 """I take an object that provides IEncryptedUploadable, which provides
1557 encrypted data, and a list of shareholders. I then encode, hash, and
1558 deliver shares to those shareholders. I will compute all the necessary
1559 Merkle hash trees that are necessary to validate the crypttext that
1560 eventually comes back from the shareholders. I provide the URI Extension
1561 Block Hash, and the encoding parameters, both of which must be included
1564 I do not choose shareholders, that is left to the IUploader. I must be
1565 given a dict of RemoteReferences to storage buckets that are ready and
1566 willing to receive data.
1570 """Specify the number of bytes that will be encoded. This must be
1571 peformed before get_serialized_params() can be called.
1573 def set_params(params):
1574 """Override the default encoding parameters. 'params' is a tuple of
1575 (k,d,n), where 'k' is the number of required shares, 'd' is the
1576 servers_of_happiness, and 'n' is the total number of shares that will
1579 Encoding parameters can be set in three ways. 1: The Encoder class
1580 provides defaults (3/7/10). 2: the Encoder can be constructed with
1581 an 'options' dictionary, in which the
1582 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1583 set_params((k,d,n)) can be called.
1585 If you intend to use set_params(), you must call it before
1586 get_share_size or get_param are called.
1589 def set_encrypted_uploadable(u):
1590 """Provide a source of encrypted upload data. 'u' must implement
1591 IEncryptedUploadable.
1593 When this is called, the IEncryptedUploadable will be queried for its
1594 length and the storage_index that should be used.
1596 This returns a Deferred that fires with this Encoder instance.
1598 This must be performed before start() can be called.
1601 def get_param(name):
1602 """Return an encoding parameter, by name.
1604 'storage_index': return a string with the (16-byte truncated SHA-256
1605 hash) storage index to which these shares should be
1608 'share_counts': return a tuple describing how many shares are used:
1609 (needed_shares, servers_of_happiness, total_shares)
1611 'num_segments': return an int with the number of segments that
1614 'segment_size': return an int with the size of each segment.
1616 'block_size': return the size of the individual blocks that will
1617 be delivered to a shareholder's put_block() method. By
1618 knowing this, the shareholder will be able to keep all
1619 blocks in a single file and still provide random access
1620 when reading them. # TODO: can we avoid exposing this?
1622 'share_size': an int with the size of the data that will be stored
1623 on each shareholder. This is aggregate amount of data
1624 that will be sent to the shareholder, summed over all
1625 the put_block() calls I will ever make. It is useful to
1626 determine this size before asking potential
1627 shareholders whether they will grant a lease or not,
1628 since their answers will depend upon how much space we
1629 need. TODO: this might also include some amount of
1630 overhead, like the size of all the hashes. We need to
1631 decide whether this is useful or not.
1633 'serialized_params': a string with a concise description of the
1634 codec name and its parameters. This may be passed
1635 into the IUploadable to let it make sure that
1636 the same file encoded with different parameters
1637 will result in different storage indexes.
1639 Once this is called, set_size() and set_params() may not be called.
1642 def set_shareholders(shareholders, servermap):
1643 """Tell the encoder where to put the encoded shares. 'shareholders'
1644 must be a dictionary that maps share number (an integer ranging from
1645 0 to n-1) to an instance that provides IStorageBucketWriter.
1646 'servermap' is a dictionary that maps share number (as defined above)
1647 to a set of peerids. This must be performed before start() can be
1651 """Begin the encode/upload process. This involves reading encrypted
1652 data from the IEncryptedUploadable, encoding it, uploading the shares
1653 to the shareholders, then sending the hash trees.
1655 set_encrypted_uploadable() and set_shareholders() must be called
1656 before this can be invoked.
1658 This returns a Deferred that fires with a verify cap when the upload
1659 process is complete. The verifycap, plus the encryption key, is
1660 sufficient to construct the read cap.
1663 class IDecoder(Interface):
1664 """I take a list of shareholders and some setup information, then
1665 download, validate, decode, and decrypt data from them, writing the
1666 results to an output file.
1668 I do not locate the shareholders, that is left to the IDownloader. I must
1669 be given a dict of RemoteReferences to storage buckets that are ready to
1674 """I take a file-like object (providing write and close) to which all
1675 the plaintext data will be written.
1677 TODO: producer/consumer . Maybe write() should return a Deferred that
1678 indicates when it will accept more data? But probably having the
1679 IDecoder be a producer is easier to glue to IConsumer pieces.
1682 def set_shareholders(shareholders):
1683 """I take a dictionary that maps share identifiers (small integers)
1684 to RemoteReferences that provide RIBucketReader. This must be called
1688 """I start the download. This process involves retrieving data and
1689 hash chains from the shareholders, using the hashes to validate the
1690 data, decoding the shares into segments, decrypting the segments,
1691 then writing the resulting plaintext to the output file.
1693 I return a Deferred that will fire (with self) when the download is
1697 class IDownloadTarget(Interface):
1698 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1699 # will register itself as a producer. This allows the target to invoke
1700 # downloader.pauseProducing, resumeProducing, and stopProducing.
1702 """Called before any calls to write() or close(). If an error
1703 occurs before any data is available, fail() may be called without
1704 a previous call to open().
1706 'size' is the length of the file being downloaded, in bytes."""
1709 """Output some data to the target."""
1711 """Inform the target that there is no more data to be written."""
1713 """fail() is called to indicate that the download has failed. 'why'
1714 is a Failure object indicating what went wrong. No further methods
1715 will be invoked on the IDownloadTarget after fail()."""
1716 def register_canceller(cb):
1717 """The CiphertextDownloader uses this to register a no-argument function
1718 that the target can call to cancel the download. Once this canceller
1719 is invoked, no further calls to write() or close() will be made."""
1721 """When the CiphertextDownloader is done, this finish() function will be
1722 called. Whatever it returns will be returned to the invoker of
1723 Downloader.download.
1726 class IDownloader(Interface):
1727 def download(uri, target):
1728 """Perform a CHK download, sending the data to the given target.
1729 'target' must provide IDownloadTarget.
1731 Returns a Deferred that fires (with the results of target.finish)
1732 when the download is finished, or errbacks if something went wrong."""
1734 class IEncryptedUploadable(Interface):
1735 def set_upload_status(upload_status):
1736 """Provide an IUploadStatus object that should be filled with status
1737 information. The IEncryptedUploadable is responsible for setting
1738 key-determination progress ('chk'), size, storage_index, and
1739 ciphertext-fetch progress. It may delegate some of this
1740 responsibility to others, in particular to the IUploadable."""
1743 """This behaves just like IUploadable.get_size()."""
1745 def get_all_encoding_parameters():
1746 """Return a Deferred that fires with a tuple of
1747 (k,happy,n,segment_size). The segment_size will be used as-is, and
1748 must match the following constraints: it must be a multiple of k, and
1749 it shouldn't be unreasonably larger than the file size (if
1750 segment_size is larger than filesize, the difference must be stored
1753 This usually passes through to the IUploadable method of the same
1756 The encoder strictly obeys the values returned by this method. To
1757 make an upload use non-default encoding parameters, you must arrange
1758 to control the values that this method returns.
1761 def get_storage_index():
1762 """Return a Deferred that fires with a 16-byte storage index.
1765 def read_encrypted(length, hash_only):
1766 """This behaves just like IUploadable.read(), but returns crypttext
1767 instead of plaintext. If hash_only is True, then this discards the
1768 data (and returns an empty list); this improves efficiency when
1769 resuming an interrupted upload (where we need to compute the
1770 plaintext hashes, but don't need the redundant encrypted data)."""
1772 def get_plaintext_hashtree_leaves(first, last, num_segments):
1773 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1774 plaintext segments, i.e. get the tagged hashes of the given segments.
1775 The segment size is expected to be generated by the
1776 IEncryptedUploadable before any plaintext is read or ciphertext
1777 produced, so that the segment hashes can be generated with only a
1780 This returns a Deferred which fires with a sequence of hashes, using:
1782 tuple(segment_hashes[first:last])
1784 'num_segments' is used to assert that the number of segments that the
1785 IEncryptedUploadable handled matches the number of segments that the
1786 encoder was expecting.
1788 This method must not be called until the final byte has been read
1789 from read_encrypted(). Once this method is called, read_encrypted()
1790 can never be called again.
1793 def get_plaintext_hash():
1794 """OBSOLETE; Get the hash of the whole plaintext.
1796 This returns a Deferred which fires with a tagged SHA-256 hash of the
1797 whole plaintext, obtained from hashutil.plaintext_hash(data).
1801 """Just like IUploadable.close()."""
1803 class IUploadable(Interface):
1804 def set_upload_status(upload_status):
1805 """Provide an IUploadStatus object that should be filled with status
1806 information. The IUploadable is responsible for setting
1807 key-determination progress ('chk')."""
1809 def set_default_encoding_parameters(params):
1810 """Set the default encoding parameters, which must be a dict mapping
1811 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1812 'max_segment_size'. These might have an influence on the final
1813 encoding parameters returned by get_all_encoding_parameters(), if the
1814 Uploadable doesn't have more specific preferences.
1816 This call is optional: if it is not used, the Uploadable will use
1817 some built-in defaults. If used, this method must be called before
1818 any other IUploadable methods to have any effect.
1822 """Return a Deferred that will fire with the length of the data to be
1823 uploaded, in bytes. This will be called before the data is actually
1824 used, to compute encoding parameters.
1827 def get_all_encoding_parameters():
1828 """Return a Deferred that fires with a tuple of
1829 (k,happy,n,segment_size). The segment_size will be used as-is, and
1830 must match the following constraints: it must be a multiple of k, and
1831 it shouldn't be unreasonably larger than the file size (if
1832 segment_size is larger than filesize, the difference must be stored
1835 The relative values of k and n allow some IUploadables to request
1836 better redundancy than others (in exchange for consuming more space
1839 Larger values of segment_size reduce hash overhead, while smaller
1840 values reduce memory footprint and cause data to be delivered in
1841 smaller pieces (which may provide a smoother and more predictable
1842 download experience).
1844 The encoder strictly obeys the values returned by this method. To
1845 make an upload use non-default encoding parameters, you must arrange
1846 to control the values that this method returns. One way to influence
1847 them may be to call set_encoding_parameters() before calling
1848 get_all_encoding_parameters().
1851 def get_encryption_key():
1852 """Return a Deferred that fires with a 16-byte AES key. This key will
1853 be used to encrypt the data. The key will also be hashed to derive
1856 Uploadables which want to achieve convergence should hash their file
1857 contents and the serialized_encoding_parameters to form the key
1858 (which of course requires a full pass over the data). Uploadables can
1859 use the upload.ConvergentUploadMixin class to achieve this
1862 Uploadables which do not care about convergence (or do not wish to
1863 make multiple passes over the data) can simply return a
1864 strongly-random 16 byte string.
1866 get_encryption_key() may be called multiple times: the IUploadable is
1867 required to return the same value each time.
1871 """Return a Deferred that fires with a list of strings (perhaps with
1872 only a single element) which, when concatenated together, contain the
1873 next 'length' bytes of data. If EOF is near, this may provide fewer
1874 than 'length' bytes. The total number of bytes provided by read()
1875 before it signals EOF must equal the size provided by get_size().
1877 If the data must be acquired through multiple internal read
1878 operations, returning a list instead of a single string may help to
1879 reduce string copies. However, the length of the concatenated strings
1880 must equal the amount of data requested, unless EOF is encountered.
1881 Long reads, or short reads without EOF, are not allowed. read()
1882 should return the same amount of data as a local disk file read, just
1883 in a different shape and asynchronously.
1885 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1886 so a 10kB file means length=3kB, 100kB file means length=30kB,
1887 and >=1MB file means length=300kB.
1889 This method provides for a single full pass through the data. Later
1890 use cases may desire multiple passes or access to only parts of the
1891 data (such as a mutable file making small edits-in-place). This API
1892 will be expanded once those use cases are better understood.
1896 """The upload is finished, and whatever filehandle was in use may be
1900 class IMutableUploadable(Interface):
1902 I represent content that is due to be uploaded to a mutable filecap.
1904 # This is somewhat simpler than the IUploadable interface above
1905 # because mutable files do not need to be concerned with possibly
1906 # generating a CHK, nor with per-file keys. It is a subset of the
1907 # methods in IUploadable, though, so we could just as well implement
1908 # the mutable uploadables as IUploadables that don't happen to use
1909 # those methods (with the understanding that the unused methods will
1910 # never be called on such objects)
1913 Returns a Deferred that fires with the size of the content held
1919 Returns a list of strings which, when concatenated, are the next
1920 length bytes of the file, or fewer if there are fewer bytes
1921 between the current location and the end of the file.
1926 The process that used the Uploadable is finished using it, so
1927 the uploadable may be closed.
1930 class IUploadResults(Interface):
1931 """I am returned by upload() methods. I contain a number of public
1932 attributes which can be read to determine the results of the upload. Some
1933 of these are functional, some are timing information. All of these may be
1936 .file_size : the size of the file, in bytes
1937 .uri : the CHK read-cap for the file
1938 .ciphertext_fetched : how many bytes were fetched by the helper
1939 .sharemap: dict mapping share identifier to set of serverids
1940 (binary strings). This indicates which servers were given
1941 which shares. For immutable files, the shareid is an
1942 integer (the share number, from 0 to N-1). For mutable
1943 files, it is a string of the form 'seq%d-%s-sh%d',
1944 containing the sequence number, the roothash, and the
1946 .servermap : dict mapping server peerid to a set of share numbers
1947 .timings : dict of timing information, mapping name to seconds (float)
1948 total : total upload time, start to finish
1949 storage_index : time to compute the storage index
1950 peer_selection : time to decide which peers will be used
1951 contacting_helper : initial helper query to upload/no-upload decision
1952 existence_check : helper pre-upload existence check
1953 helper_total : initial helper query to helper finished pushing
1954 cumulative_fetch : helper waiting for ciphertext requests
1955 total_fetch : helper start to last ciphertext response
1956 cumulative_encoding : just time spent in zfec
1957 cumulative_sending : just time spent waiting for storage servers
1958 hashes_and_close : last segment push to shareholder close
1959 total_encode_and_push : first encode to shareholder close
1963 class IDownloadResults(Interface):
1964 """I am created internally by download() methods. I contain a number of
1965 public attributes which contain details about the download process.::
1967 .file_size : the size of the file, in bytes
1968 .servers_used : set of server peerids that were used during download
1969 .server_problems : dict mapping server peerid to a problem string. Only
1970 servers that had problems (bad hashes, disconnects)
1972 .servermap : dict mapping server peerid to a set of share numbers. Only
1973 servers that had any shares are listed here.
1974 .timings : dict of timing information, mapping name to seconds (float)
1975 peer_selection : time to ask servers about shares
1976 servers_peer_selection : dict of peerid to DYHB-query time
1977 uri_extension : time to fetch a copy of the URI extension block
1978 hashtrees : time to fetch the hash trees
1979 segments : time to fetch, decode, and deliver segments
1980 cumulative_fetch : time spent waiting for storage servers
1981 cumulative_decode : just time spent in zfec
1982 cumulative_decrypt : just time spent in decryption
1983 total : total download time, start to finish
1984 fetch_per_server : dict of peerid to list of per-segment fetch times
1988 class IUploader(Interface):
1989 def upload(uploadable):
1990 """Upload the file. 'uploadable' must impement IUploadable. This
1991 returns a Deferred which fires with an IUploadResults instance, from
1992 which the URI of the file can be obtained as results.uri ."""
1994 def upload_ssk(write_capability, new_version, uploadable):
1995 """TODO: how should this work?"""
1997 class ICheckable(Interface):
1998 def check(monitor, verify=False, add_lease=False):
1999 """Check up on my health, optionally repairing any problems.
2001 This returns a Deferred that fires with an instance that provides
2002 ICheckResults, or None if the object is non-distributed (i.e. LIT
2005 The monitor will be checked periodically to see if the operation has
2006 been cancelled. If so, no new queries will be sent, and the Deferred
2007 will fire (with a OperationCancelledError) immediately.
2009 Filenodes and dirnodes (which provide IFilesystemNode) are also
2010 checkable. Instances that represent verifier-caps will be checkable
2011 but not downloadable. Some objects (like LIT files) do not actually
2012 live in the grid, and their checkers return None (non-distributed
2013 files are always healthy).
2015 If verify=False, a relatively lightweight check will be performed: I
2016 will ask all servers if they have a share for me, and I will believe
2017 whatever they say. If there are at least N distinct shares on the
2018 grid, my results will indicate r.is_healthy()==True. This requires a
2019 roundtrip to each server, but does not transfer very much data, so
2020 the network bandwidth is fairly low.
2022 If verify=True, a more resource-intensive check will be performed:
2023 every share will be downloaded, and the hashes will be validated on
2024 every bit. I will ignore any shares that failed their hash checks. If
2025 there are at least N distinct valid shares on the grid, my results
2026 will indicate r.is_healthy()==True. This requires N/k times as much
2027 download bandwidth (and server disk IO) as a regular download. If a
2028 storage server is holding a corrupt share, or is experiencing memory
2029 failures during retrieval, or is malicious or buggy, then
2030 verification will detect the problem, but checking will not.
2032 If add_lease=True, I will ensure that an up-to-date lease is present
2033 on each share. The lease secrets will be derived from by node secret
2034 (in BASEDIR/private/secret), so either I will add a new lease to the
2035 share, or I will merely renew the lease that I already had. In a
2036 future version of the storage-server protocol (once Accounting has
2037 been implemented), there may be additional options here to define the
2038 kind of lease that is obtained (which account number to claim, etc).
2040 TODO: any problems seen during checking will be reported to the
2041 health-manager.furl, a centralized object which is responsible for
2042 figuring out why files are unhealthy so corrective action can be
2046 def check_and_repair(monitor, verify=False, add_lease=False):
2047 """Like check(), but if the file/directory is not healthy, attempt to
2050 Any non-healthy result will cause an immediate repair operation, to
2051 generate and upload new shares. After repair, the file will be as
2052 healthy as we can make it. Details about what sort of repair is done
2053 will be put in the check-and-repair results. The Deferred will not
2054 fire until the repair is complete.
2056 This returns a Deferred which fires with an instance of
2057 ICheckAndRepairResults."""
2059 class IDeepCheckable(Interface):
2060 def start_deep_check(verify=False, add_lease=False):
2061 """Check upon the health of me and everything I can reach.
2063 This is a recursive form of check(), useable only on dirnodes.
2065 I return a Monitor, with results that are an IDeepCheckResults
2068 TODO: If any of the directories I traverse are unrecoverable, the
2069 Monitor will report failure. If any of the files I check upon are
2070 unrecoverable, those problems will be reported in the
2071 IDeepCheckResults as usual, and the Monitor will not report a
2075 def start_deep_check_and_repair(verify=False, add_lease=False):
2076 """Check upon the health of me and everything I can reach. Repair
2077 anything that isn't healthy.
2079 This is a recursive form of check_and_repair(), useable only on
2082 I return a Monitor, with results that are an
2083 IDeepCheckAndRepairResults object.
2085 TODO: If any of the directories I traverse are unrecoverable, the
2086 Monitor will report failure. If any of the files I check upon are
2087 unrecoverable, those problems will be reported in the
2088 IDeepCheckResults as usual, and the Monitor will not report a
2092 class ICheckResults(Interface):
2093 """I contain the detailed results of a check/verify operation.
2096 def get_storage_index():
2097 """Return a string with the (binary) storage index."""
2098 def get_storage_index_string():
2099 """Return a string with the (printable) abbreviated storage index."""
2101 """Return the (string) URI of the object that was checked."""
2104 """Return a boolean, True if the file/dir is fully healthy, False if
2105 it is damaged in any way. Non-distributed LIT files always return
2108 def is_recoverable():
2109 """Return a boolean, True if the file/dir can be recovered, False if
2110 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2111 files always return True."""
2113 def needs_rebalancing():
2114 """Return a boolean, True if the file/dir's reliability could be
2115 improved by moving shares to new servers. Non-distributed LIT files
2116 always return False."""
2120 """Return a dictionary that describes the state of the file/dir. LIT
2121 files always return an empty dictionary. Normal files and directories
2122 return a dictionary with the following keys (note that these use
2123 binary strings rather than base32-encoded ones) (also note that for
2124 mutable files, these counts are for the 'best' version):
2126 count-shares-good: the number of distinct good shares that were found
2127 count-shares-needed: 'k', the number of shares required for recovery
2128 count-shares-expected: 'N', the number of total shares generated
2129 count-good-share-hosts: the number of distinct storage servers with
2130 good shares. If this number is less than
2131 count-shares-good, then some shares are
2132 doubled up, increasing the correlation of
2133 failures. This indicates that one or more
2134 shares should be moved to an otherwise unused
2135 server, if one is available.
2136 count-corrupt-shares: the number of shares with integrity failures
2137 list-corrupt-shares: a list of 'share locators', one for each share
2138 that was found to be corrupt. Each share
2139 locator is a list of (serverid, storage_index,
2141 count-incompatible-shares: the number of shares which are of a share
2142 format unknown to this checker
2143 list-incompatible-shares: a list of 'share locators', one for each
2144 share that was found to be of an unknown
2145 format. Each share locator is a list of
2146 (serverid, storage_index, sharenum).
2147 servers-responding: list of (binary) storage server identifiers,
2148 one for each server which responded to the share
2149 query (even if they said they didn't have
2150 shares, and even if they said they did have
2151 shares but then didn't send them when asked, or
2152 dropped the connection, or returned a Failure,
2153 and even if they said they did have shares and
2154 sent incorrect ones when asked)
2155 sharemap: dict mapping share identifier to list of serverids
2156 (binary strings). This indicates which servers are holding
2157 which shares. For immutable files, the shareid is an
2158 integer (the share number, from 0 to N-1). For mutable
2159 files, it is a string of the form 'seq%d-%s-sh%d',
2160 containing the sequence number, the roothash, and the
2163 The following keys are most relevant for mutable files, but immutable
2164 files will provide sensible values too::
2166 count-wrong-shares: the number of shares for versions other than the
2167 'best' one (which is defined as being the
2168 recoverable version with the highest sequence
2169 number, then the highest roothash). These are
2170 either leftover shares from an older version
2171 (perhaps on a server that was offline when an
2172 update occurred), shares from an unrecoverable
2173 newer version, or shares from an alternate
2174 current version that results from an
2175 uncoordinated write collision. For a healthy
2176 file, this will equal 0.
2178 count-recoverable-versions: the number of recoverable versions of
2179 the file. For a healthy file, this will
2182 count-unrecoverable-versions: the number of unrecoverable versions
2183 of the file. For a healthy file, this
2189 """Return a string with a brief (one-line) summary of the results."""
2192 """Return a list of strings with more detailed results."""
2194 class ICheckAndRepairResults(Interface):
2195 """I contain the detailed results of a check/verify/repair operation.
2197 The IFilesystemNode.check()/verify()/repair() methods all return
2198 instances that provide ICheckAndRepairResults.
2201 def get_storage_index():
2202 """Return a string with the (binary) storage index."""
2203 def get_storage_index_string():
2204 """Return a string with the (printable) abbreviated storage index."""
2205 def get_repair_attempted():
2206 """Return a boolean, True if a repair was attempted. We might not
2207 attempt to repair the file because it was healthy, or healthy enough
2208 (i.e. some shares were missing but not enough to exceed some
2209 threshold), or because we don't know how to repair this object."""
2210 def get_repair_successful():
2211 """Return a boolean, True if repair was attempted and the file/dir
2212 was fully healthy afterwards. False if no repair was attempted or if
2213 a repair attempt failed."""
2214 def get_pre_repair_results():
2215 """Return an ICheckResults instance that describes the state of the
2216 file/dir before any repair was attempted."""
2217 def get_post_repair_results():
2218 """Return an ICheckResults instance that describes the state of the
2219 file/dir after any repair was attempted. If no repair was attempted,
2220 the pre-repair and post-repair results will be identical."""
2223 class IDeepCheckResults(Interface):
2224 """I contain the results of a deep-check operation.
2226 This is returned by a call to ICheckable.deep_check().
2229 def get_root_storage_index_string():
2230 """Return the storage index (abbreviated human-readable string) of
2231 the first object checked."""
2233 """Return a dictionary with the following keys::
2235 count-objects-checked: count of how many objects were checked
2236 count-objects-healthy: how many of those objects were completely
2238 count-objects-unhealthy: how many were damaged in some way
2239 count-objects-unrecoverable: how many were unrecoverable
2240 count-corrupt-shares: how many shares were found to have
2241 corruption, summed over all objects
2245 def get_corrupt_shares():
2246 """Return a set of (serverid, storage_index, sharenum) for all shares
2247 that were found to be corrupt. Both serverid and storage_index are
2250 def get_all_results():
2251 """Return a dictionary mapping pathname (a tuple of strings, ready to
2252 be slash-joined) to an ICheckResults instance, one for each object
2253 that was checked."""
2255 def get_results_for_storage_index(storage_index):
2256 """Retrive the ICheckResults instance for the given (binary)
2257 storage index. Raises KeyError if there are no results for that
2261 """Return a dictionary with the same keys as
2262 IDirectoryNode.deep_stats()."""
2264 class IDeepCheckAndRepairResults(Interface):
2265 """I contain the results of a deep-check-and-repair operation.
2267 This is returned by a call to ICheckable.deep_check_and_repair().
2270 def get_root_storage_index_string():
2271 """Return the storage index (abbreviated human-readable string) of
2272 the first object checked."""
2274 """Return a dictionary with the following keys::
2276 count-objects-checked: count of how many objects were checked
2277 count-objects-healthy-pre-repair: how many of those objects were
2278 completely healthy (before any
2280 count-objects-unhealthy-pre-repair: how many were damaged in
2282 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2283 count-objects-healthy-post-repair: how many of those objects were
2284 completely healthy (after any
2286 count-objects-unhealthy-post-repair: how many were damaged in
2288 count-objects-unrecoverable-post-repair: how many were
2290 count-repairs-attempted: repairs were attempted on this many
2291 objects. The count-repairs- keys will
2292 always be provided, however unless
2293 repair=true is present, they will all
2295 count-repairs-successful: how many repairs resulted in healthy
2297 count-repairs-unsuccessful: how many repairs resulted did not
2298 results in completely healthy objects
2299 count-corrupt-shares-pre-repair: how many shares were found to
2300 have corruption, summed over all
2301 objects examined (before any
2303 count-corrupt-shares-post-repair: how many shares were found to
2304 have corruption, summed over all
2305 objects examined (after any
2310 """Return a dictionary with the same keys as
2311 IDirectoryNode.deep_stats()."""
2313 def get_corrupt_shares():
2314 """Return a set of (serverid, storage_index, sharenum) for all shares
2315 that were found to be corrupt before any repair was attempted. Both
2316 serverid and storage_index are binary.
2318 def get_remaining_corrupt_shares():
2319 """Return a set of (serverid, storage_index, sharenum) for all shares
2320 that were found to be corrupt after any repair was completed. Both
2321 serverid and storage_index are binary. These are shares that need
2322 manual inspection and probably deletion.
2324 def get_all_results():
2325 """Return a dictionary mapping pathname (a tuple of strings, ready to
2326 be slash-joined) to an ICheckAndRepairResults instance, one for each
2327 object that was checked."""
2329 def get_results_for_storage_index(storage_index):
2330 """Retrive the ICheckAndRepairResults instance for the given (binary)
2331 storage index. Raises KeyError if there are no results for that
2335 class IRepairable(Interface):
2336 def repair(check_results):
2337 """Attempt to repair the given object. Returns a Deferred that fires
2338 with a IRepairResults object.
2340 I must be called with an object that implements ICheckResults, as
2341 proof that you have actually discovered a problem with this file. I
2342 will use the data in the checker results to guide the repair process,
2343 such as which servers provided bad data and should therefore be
2344 avoided. The ICheckResults object is inside the
2345 ICheckAndRepairResults object, which is returned by the
2346 ICheckable.check() method::
2348 d = filenode.check(repair=False)
2349 def _got_results(check_and_repair_results):
2350 check_results = check_and_repair_results.get_pre_repair_results()
2351 return filenode.repair(check_results)
2352 d.addCallback(_got_results)
2356 class IRepairResults(Interface):
2357 """I contain the results of a repair operation."""
2358 def get_successful(self):
2359 """Returns a boolean: True if the repair made the file healthy, False
2360 if not. Repair failure generally indicates a file that has been
2361 damaged beyond repair."""
2364 class IClient(Interface):
2365 def upload(uploadable):
2366 """Upload some data into a CHK, get back the UploadResults for it.
2367 @param uploadable: something that implements IUploadable
2368 @return: a Deferred that fires with the UploadResults instance.
2369 To get the URI for this file, use results.uri .
2372 def create_mutable_file(contents=""):
2373 """Create a new mutable file (with initial) contents, get back the
2376 @param contents: (bytestring, callable, or None): this provides the
2377 initial contents of the mutable file. If 'contents' is a bytestring,
2378 it will be used as-is. If 'contents' is a callable, it will be
2379 invoked with the new MutableFileNode instance and is expected to
2380 return a bytestring with the initial contents of the file (the
2381 callable can use node.get_writekey() to decide how to encrypt the
2382 initial contents, e.g. for a brand new dirnode with initial
2383 children). contents=None is equivalent to an empty string. Using
2384 content_maker= is more efficient than creating a mutable file and
2385 setting its contents in two separate operations.
2387 @return: a Deferred that fires with an IMutableFileNode instance.
2390 def create_dirnode(initial_children={}):
2391 """Create a new unattached dirnode, possibly with initial children.
2393 @param initial_children: dict with keys that are unicode child names,
2394 and values that are (childnode, metadata) tuples.
2396 @return: a Deferred that fires with the new IDirectoryNode instance.
2399 def create_node_from_uri(uri, rouri):
2400 """Create a new IFilesystemNode instance from the uri, synchronously.
2401 @param uri: a string or IURI-providing instance, or None. This could
2402 be for a LiteralFileNode, a CHK file node, a mutable file
2403 node, or a directory node
2404 @param rouri: a string or IURI-providing instance, or None. If the
2405 main uri is None, I will use the rouri instead. If I
2406 recognize the format of the main uri, I will ignore the
2407 rouri (because it can be derived from the writecap).
2409 @return: an instance that provides IFilesystemNode (or more usefully
2410 one of its subclasses). File-specifying URIs will result in
2411 IFileNode-providing instances, like ImmutableFileNode,
2412 LiteralFileNode, or MutableFileNode. Directory-specifying
2413 URIs will result in IDirectoryNode-providing instances, like
2417 class INodeMaker(Interface):
2418 """The NodeMaker is used to create IFilesystemNode instances. It can
2419 accept a filecap/dircap string and return the node right away. It can
2420 also create new nodes (i.e. upload a file, or create a mutable file)
2421 asynchronously. Once you have one of these nodes, you can use other
2422 methods to determine whether it is a file or directory, and to download
2423 or modify its contents.
2425 The NodeMaker encapsulates all the authorities that these
2426 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2427 Tahoe process will typically have a single NodeMaker, but unit tests may
2428 create simplified/mocked forms for testing purposes.
2430 def create_from_cap(writecap, readcap=None, **kwargs):
2431 """I create an IFilesystemNode from the given writecap/readcap. I can
2432 only provide nodes for existing file/directory objects: use my other
2433 methods to create new objects. I return synchronously."""
2435 def create_mutable_file(contents=None, keysize=None):
2436 """I create a new mutable file, and return a Deferred which will fire
2437 with the IMutableFileNode instance when it is ready. If contents= is
2438 provided (a bytestring), it will be used as the initial contents of
2439 the new file, otherwise the file will contain zero bytes. keysize= is
2440 for use by unit tests, to create mutable files that are smaller than
2443 def create_new_mutable_directory(initial_children={}):
2444 """I create a new mutable directory, and return a Deferred which will
2445 fire with the IDirectoryNode instance when it is ready. If
2446 initial_children= is provided (a dict mapping unicode child name to
2447 (childnode, metadata_dict) tuples), the directory will be populated
2448 with those children, otherwise it will be empty."""
2450 class IClientStatus(Interface):
2451 def list_all_uploads():
2452 """Return a list of uploader objects, one for each upload which
2453 currently has an object available (tracked with weakrefs). This is
2454 intended for debugging purposes."""
2455 def list_active_uploads():
2456 """Return a list of active IUploadStatus objects."""
2457 def list_recent_uploads():
2458 """Return a list of IUploadStatus objects for the most recently
2461 def list_all_downloads():
2462 """Return a list of downloader objects, one for each download which
2463 currently has an object available (tracked with weakrefs). This is
2464 intended for debugging purposes."""
2465 def list_active_downloads():
2466 """Return a list of active IDownloadStatus objects."""
2467 def list_recent_downloads():
2468 """Return a list of IDownloadStatus objects for the most recently
2469 started downloads."""
2471 class IUploadStatus(Interface):
2473 """Return a timestamp (float with seconds since epoch) indicating
2474 when the operation was started."""
2475 def get_storage_index():
2476 """Return a string with the (binary) storage index in use on this
2477 upload. Returns None if the storage index has not yet been
2480 """Return an integer with the number of bytes that will eventually
2481 be uploaded for this file. Returns None if the size is not yet known.
2484 """Return True if this upload is using a Helper, False if not."""
2486 """Return a string describing the current state of the upload
2489 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2490 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2491 made towards hashing the file to determine a CHK encryption key: if
2492 non-convergent encryption is in use, this will be trivial, otherwise
2493 the whole file must be hashed. 'ciphertext' describes how much of the
2494 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2495 uploads. 'encode_and_push' describes how much of the encode-and-push
2496 process has finished: for helper uploads this is dependent upon the
2497 helper providing progress reports. It might be reasonable to add all
2498 three numbers and report the sum to the user."""
2500 """Return True if the upload is currently active, False if not."""
2502 """Return an instance of UploadResults (which contains timing and
2503 sharemap information). Might return None if the upload is not yet
2506 """Each upload status gets a unique number: this method returns that
2507 number. This provides a handle to this particular upload, so a web
2508 page can generate a suitable hyperlink."""
2510 class IDownloadStatus(Interface):
2512 """Return a timestamp (float with seconds since epoch) indicating
2513 when the operation was started."""
2514 def get_storage_index():
2515 """Return a string with the (binary) storage index in use on this
2516 download. This may be None if there is no storage index (i.e. LIT
2519 """Return an integer with the number of bytes that will eventually be
2520 retrieved for this file. Returns None if the size is not yet known.
2523 """Return True if this download is using a Helper, False if not."""
2525 """Return a string describing the current state of the download
2528 """Returns a float (from 0.0 to 1.0) describing the amount of the
2529 download that has completed. This value will remain at 0.0 until the
2530 first byte of plaintext is pushed to the download target."""
2532 """Return True if the download is currently active, False if not."""
2534 """Each download status gets a unique number: this method returns
2535 that number. This provides a handle to this particular download, so a
2536 web page can generate a suitable hyperlink."""
2538 class IServermapUpdaterStatus(Interface):
2540 class IPublishStatus(Interface):
2542 class IRetrieveStatus(Interface):
2545 class NotCapableError(Exception):
2546 """You have tried to write to a read-only node."""
2548 class BadWriteEnablerError(Exception):
2551 class RIControlClient(RemoteInterface):
2553 def wait_for_client_connections(num_clients=int):
2554 """Do not return until we have connections to at least NUM_CLIENTS
2558 def upload_from_file_to_uri(filename=str,
2559 convergence=ChoiceOf(None,
2560 StringConstraint(2**20))):
2561 """Upload a file to the grid. This accepts a filename (which must be
2562 absolute) that points to a file on the node's local disk. The node will
2563 read the contents of this file, upload it to the grid, then return the
2564 URI at which it was uploaded. If convergence is None then a random
2565 encryption key will be used, else the plaintext will be hashed, then
2566 that hash will be mixed together with the "convergence" string to form
2571 def download_from_uri_to_file(uri=URI, filename=str):
2572 """Download a file from the grid, placing it on the node's local disk
2573 at the given filename (which must be absolute[?]). Returns the
2574 absolute filename where the file was written."""
2579 def get_memory_usage():
2580 """Return a dict describes the amount of memory currently in use. The
2581 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2582 measuring memory consupmtion in bytes."""
2583 return DictOf(str, int)
2585 def speed_test(count=int, size=int, mutable=Any()):
2586 """Write 'count' tempfiles to disk, all of the given size. Measure
2587 how long (in seconds) it takes to upload them all to the servers.
2588 Then measure how long it takes to download all of them. If 'mutable'
2589 is 'create', time creation of mutable files. If 'mutable' is
2590 'upload', then time access to the same mutable file instead of
2593 Returns a tuple of (upload_time, download_time).
2595 return (float, float)
2597 def measure_peer_response_time():
2598 """Send a short message to each connected peer, and measure the time
2599 it takes for them to respond to it. This is a rough measure of the
2600 application-level round trip time.
2602 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2605 return DictOf(str, float)
2607 UploadResults = Any() #DictOf(str, str)
2609 class RIEncryptedUploadable(RemoteInterface):
2610 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2615 def get_all_encoding_parameters():
2616 return (int, int, int, long)
2618 def read_encrypted(offset=Offset, length=ReadSize):
2625 class RICHKUploadHelper(RemoteInterface):
2626 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2630 Return a dictionary of version information.
2632 return DictOf(str, Any())
2634 def upload(reader=RIEncryptedUploadable):
2635 return UploadResults
2638 class RIHelper(RemoteInterface):
2639 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2643 Return a dictionary of version information.
2645 return DictOf(str, Any())
2647 def upload_chk(si=StorageIndex):
2648 """See if a file with a given storage index needs uploading. The
2649 helper will ask the appropriate storage servers to see if the file
2650 has already been uploaded. If so, the helper will return a set of
2651 'upload results' that includes whatever hashes are needed to build
2652 the read-cap, and perhaps a truncated sharemap.
2654 If the file has not yet been uploaded (or if it was only partially
2655 uploaded), the helper will return an empty upload-results dictionary
2656 and also an RICHKUploadHelper object that will take care of the
2657 upload process. The client should call upload() on this object and
2658 pass it a reference to an RIEncryptedUploadable object that will
2659 provide ciphertext. When the upload is finished, the upload() method
2660 will finish and return the upload results.
2662 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2665 class RIStatsProvider(RemoteInterface):
2666 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2668 Provides access to statistics and monitoring information.
2673 returns a dictionary containing 'counters' and 'stats', each a
2674 dictionary with string counter/stat name keys, and numeric or None values.
2675 counters are monotonically increasing measures of work done, and
2676 stats are instantaneous measures (potentially time averaged
2679 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2681 class RIStatsGatherer(RemoteInterface):
2682 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2684 Provides a monitoring service for centralised collection of stats
2687 def provide(provider=RIStatsProvider, nickname=str):
2689 @param provider: a stats collector instance which should be polled
2690 periodically by the gatherer to collect stats.
2691 @param nickname: a name useful to identify the provided client
2696 class IStatsProducer(Interface):
2699 returns a dictionary, with str keys representing the names of stats
2700 to be monitored, and numeric values.
2703 class RIKeyGenerator(RemoteInterface):
2704 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2706 Provides a service offering to make RSA key pairs.
2709 def get_rsa_key_pair(key_size=int):
2711 @param key_size: the size of the signature key.
2712 @return: tuple(verifying_key, signing_key)
2714 return TupleOf(str, str)
2717 class FileTooLargeError(Exception):
2720 class IValidatedThingProxy(Interface):
2722 """ Acquire a thing and validate it. Return a deferred which is
2723 eventually fired with self if the thing is valid or errbacked if it
2724 can't be acquired or validated."""
2726 class InsufficientVersionError(Exception):
2727 def __init__(self, needed, got):
2728 self.needed = needed
2731 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2734 class EmptyPathnameComponentError(Exception):
2735 """The webapi disallows empty pathname components."""