ExternalVendorCode/tahoe-lafs
1
0
Fork 0
mirror of https://github.com/tahoe-lafs/tahoe-lafs.git synced 2024-12-21 22:07:51 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
bc53c24003
tahoe-lafs/src/allmydata/interfaces.py

2170 lines
98 KiB
Python
Raw Blame History

from zope.interface import Interface
from foolscap.schema import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
ChoiceOf, IntegerConstraint, Any
from foolscap import RemoteInterface, Referenceable
HASH_SIZE=32
Hash = StringConstraint(maxLength=HASH_SIZE,
minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
Nodeid = StringConstraint(maxLength=20,
minLength=20) # binary format 20-byte SHA1 hash
FURL = StringConstraint(1000)
StorageIndex = StringConstraint(16)
URI = StringConstraint(300) # kind of arbitrary
MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
ShareData = StringConstraint(None)
URIExtensionData = StringConstraint(1000)
Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
Offset = Number
ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
WriteEnablerSecret = Hash # used to protect mutable bucket modifications
LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
class RIStubClient(RemoteInterface):
"""Each client publishes a service announcement for a dummy object called
the StubClient. This object doesn't actually offer any services, but the
announcement helps the Introducer keep track of which clients are
subscribed (so the grid admin can keep track of things like the size of
the grid and the client versions in use. This is the (empty)
RemoteInterface for the StubClient."""
class RIBucketWriter(RemoteInterface):
""" Objects of this kind live on the server side. """
def write(offset=Offset, data=ShareData):
return None
def close():
"""
If the data that has been written is incomplete or inconsistent then
the server will throw the data away, else it will store it for future
retrieval.
"""
return None
def abort():
"""Abandon all the data that has been written.
"""
return None
class RIBucketReader(RemoteInterface):
def read(offset=Offset, length=ReadSize):
return ShareData
def advise_corrupt_share(reason=str):
"""Clients who discover hash failures in shares that they have
downloaded from me will use this method to inform me about the
failures. I will record their concern so that my operator can
manually inspect the shares in question. I return None.
This is a wrapper around RIStorageServer.advise_corrupt_share(),
which is tied to a specific share, and therefore does not need the
extra share-identifying arguments. Please see that method for full
documentation.
"""
TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
# elements are (offset, length, operator, specimen)
# operator is one of "lt, le, eq, ne, ge, gt"
# nop always passes and is used to fetch data while writing.
# you should use length==len(specimen) for everything except nop
DataVector = ListOf(TupleOf(Offset, ShareData))
# (offset, data). This limits us to 30 writes of 1MiB each per call
TestAndWriteVectorsForShares = DictOf(int,
TupleOf(TestVector,
DataVector,
ChoiceOf(None, Offset), # new_length
))
ReadVector = ListOf(TupleOf(Offset, ReadSize))
ReadData = ListOf(ShareData)
# returns data[offset:offset+length] for each element of TestVector
class RIStorageServer(RemoteInterface):
__remote_name__ = "RIStorageServer.tahoe.allmydata.com"
def get_version():
"""
Return a dictionary of version information.
"""
return DictOf(str, Any())
def allocate_buckets(storage_index=StorageIndex,
renew_secret=LeaseRenewSecret,
cancel_secret=LeaseCancelSecret,
sharenums=SetOf(int, maxLength=MAX_BUCKETS),
allocated_size=Offset, canary=Referenceable):
"""
@param storage_index: the index of the bucket to be created or
increfed.
@param sharenums: these are the share numbers (probably between 0 and
99) that the sender is proposing to store on this
server.
@param renew_secret: This is the secret used to protect bucket refresh
This secret is generated by the client and
stored for later comparison by the server. Each
server is given a different secret.
@param cancel_secret: Like renew_secret, but protects bucket decref.
@param canary: If the canary is lost before close(), the bucket is
deleted.
@return: tuple of (alreadygot, allocated), where alreadygot is what we
already have and is what we hereby agree to accept. New
leases are added for shares in both lists.
"""
return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
def add_lease(storage_index=StorageIndex,
renew_secret=LeaseRenewSecret,
cancel_secret=LeaseCancelSecret):
"""
Add a new lease on the given bucket. If the renew_secret matches an
existing lease, that lease will be renewed instead.
"""
return None
def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
"""
Renew the lease on a given bucket. Some networks will use this, some
will not.
"""
return None
def cancel_lease(storage_index=StorageIndex,
cancel_secret=LeaseCancelSecret):
"""
Cancel the lease on a given bucket. If this was the last lease on the
bucket, the bucket will be deleted.
"""
return None
def get_buckets(storage_index=StorageIndex):
return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
def slot_readv(storage_index=StorageIndex,
shares=ListOf(int), readv=ReadVector):
"""Read a vector from the numbered shares associated with the given
storage index. An empty shares list means to return data from all
known shares. Returns a dictionary with one key per share."""
return DictOf(int, ReadData) # shnum -> results
def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
secrets=TupleOf(WriteEnablerSecret,
LeaseRenewSecret,
LeaseCancelSecret),
tw_vectors=TestAndWriteVectorsForShares,
r_vector=ReadVector,
):
"""General-purpose test-and-set operation for mutable slots. Perform
a bunch of comparisons against the existing shares. If they all pass,
then apply a bunch of write vectors to those shares. Then use the
read vectors to extract data from all the shares and return the data.
This method is, um, large. The goal is to allow clients to update all
the shares associated with a mutable file in a single round trip.
@param storage_index: the index of the bucket to be created or
increfed.
@param write_enabler: a secret that is stored along with the slot.
Writes are accepted from any caller who can
present the matching secret. A different secret
should be used for each slot*server pair.
@param renew_secret: This is the secret used to protect bucket refresh
This secret is generated by the client and
stored for later comparison by the server. Each
server is given a different secret.
@param cancel_secret: Like renew_secret, but protects bucket decref.
The 'secrets' argument is a tuple of (write_enabler, renew_secret,
cancel_secret). The first is required to perform any write. The
latter two are used when allocating new shares. To simply acquire a
new lease on existing shares, use an empty testv and an empty writev.
Each share can have a separate test vector (i.e. a list of
comparisons to perform). If all vectors for all shares pass, then all
writes for all shares are recorded. Each comparison is a 4-tuple of
(offset, length, operator, specimen), which effectively does a bool(
(read(offset, length)) OPERATOR specimen ) and only performs the
write if all these evaluate to True. Basic test-and-set uses 'eq'.
Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
Write-if-same-or-newer uses 'le'.
Reads from the end of the container are truncated, and missing shares
behave like empty ones, so to assert that a share doesn't exist (for
use when creating a new share), use (0, 1, 'eq', '').
The write vector will be applied to the given share, expanding it if
necessary. A write vector applied to a share number that did not
exist previously will cause that share to be created.
Each write vector is accompanied by a 'new_length' argument. If
new_length is not None, use it to set the size of the container. This
can be used to pre-allocate space for a series of upcoming writes, or
truncate existing data. If the container is growing, new_length will
be applied before datav. If the container is shrinking, it will be
applied afterwards.
The read vector is used to extract data from all known shares,
*before* any writes have been applied. The same vector is used for
all shares. This captures the state that was tested by the test
vector.
This method returns two values: a boolean and a dict. The boolean is
True if the write vectors were applied, False if not. The dict is
keyed by share number, and each value contains a list of strings, one
for each element of the read vector.
If the write_enabler is wrong, this will raise BadWriteEnablerError.
To enable share migration (using update_write_enabler), the exception
will have the nodeid used for the old write enabler embedded in it,
in the following string::
The write enabler was recorded by nodeid '%s'.
Note that the nodeid here is encoded using the same base32 encoding
used by Foolscap and allmydata.util.idlib.nodeid_b2a().
"""
return TupleOf(bool, DictOf(int, ReadData))
def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
shnum=int, reason=str):
"""Clients who discover hash failures in shares that they have
downloaded from me will use this method to inform me about the
failures. I will record their concern so that my operator can
manually inspect the shares in question. I return None.
'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
(binary) storage index string, and 'shnum' is the integer share
number. 'reason' is a human-readable explanation of the problem,
probably including some expected hash values and the computed ones
which did not match. Corruption advisories for mutable shares should
include a hash of the public key (the same value that appears in the
mutable-file verify-cap), since the current share format does not
store that on disk.
"""
class IStorageBucketWriter(Interface):
"""
Objects of this kind live on the client side.
"""
def put_block(segmentnum=int, data=ShareData):
"""@param data: For most segments, this data will be 'blocksize'
bytes in length. The last segment might be shorter.
@return: a Deferred that fires (with None) when the operation completes
"""
def put_plaintext_hashes(hashes=ListOf(Hash)):
"""
@return: a Deferred that fires (with None) when the operation completes
"""
def put_crypttext_hashes(hashes=ListOf(Hash)):
"""
@return: a Deferred that fires (with None) when the operation completes
"""
def put_block_hashes(blockhashes=ListOf(Hash)):
"""
@return: a Deferred that fires (with None) when the operation completes
"""
def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
"""
@return: a Deferred that fires (with None) when the operation completes
"""
def put_uri_extension(data=URIExtensionData):
"""This block of data contains integrity-checking information (hashes
of plaintext, crypttext, and shares), as well as encoding parameters
that are necessary to recover the data. This is a serialized dict
mapping strings to other strings. The hash of this data is kept in
the URI and verified before any of the data is used. All buckets for
a given file contain identical copies of this data.
The serialization format is specified with the following pseudocode:
for k in sorted(dict.keys()):
assert re.match(r'^[a-zA-Z_\-]+$', k)
write(k + ':' + netstring(dict[k]))
@return: a Deferred that fires (with None) when the operation completes
"""
def close():
"""Finish writing and close the bucket. The share is not finalized
until this method is called: if the uploading client disconnects
before calling close(), the partially-written share will be
discarded.
@return: a Deferred that fires (with None) when the operation completes
"""
class IStorageBucketReader(Interface):
def get_block(blocknum=int):
"""Most blocks will be the same size. The last block might be shorter
than the others.
@return: ShareData
"""
def get_plaintext_hashes():
"""
@return: ListOf(Hash)
"""
def get_crypttext_hashes():
"""
@return: ListOf(Hash)
"""
def get_block_hashes():
"""
@return: ListOf(Hash)
"""
def get_share_hashes():
"""
@return: ListOf(TupleOf(int, Hash))
"""
def get_uri_extension():
"""
@return: URIExtensionData
"""
# hm, we need a solution for forward references in schemas
FileNode_ = Any() # TODO: foolscap needs constraints on copyables
DirectoryNode_ = Any() # TODO: same
AnyNode_ = ChoiceOf(FileNode_, DirectoryNode_)
EncryptedThing = str
class IURI(Interface):
def init_from_string(uri):
"""Accept a string (as created by my to_string() method) and populate
this instance with its data. I am not normally called directly,
please use the module-level uri.from_string() function to convert
arbitrary URI strings into IURI-providing instances."""
def is_readonly():
"""Return False if this URI be used to modify the data. Return True
if this URI cannot be used to modify the data."""
def is_mutable():
"""Return True if the data can be modified by *somebody* (perhaps
someone who has a more powerful URI than this one)."""
def get_readonly():
"""Return another IURI instance, which represents a read-only form of
this one. If is_readonly() is True, this returns self."""
def get_verifier():
"""Return an instance that provides IVerifierURI, which can be used
to check on the availability of the file or directory, without
providing enough capabilities to actually read or modify the
contents. This may return None if the file does not need checking or
verification (e.g. LIT URIs).
"""
def to_string():
"""Return a string of printable ASCII characters, suitable for
passing into init_from_string."""
class IVerifierURI(Interface):
def init_from_string(uri):
"""Accept a string (as created by my to_string() method) and populate
this instance with its data. I am not normally called directly,
please use the module-level uri.from_string() function to convert
arbitrary URI strings into IURI-providing instances."""
def to_string():
"""Return a string of printable ASCII characters, suitable for
passing into init_from_string."""
class IDirnodeURI(Interface):
"""I am a URI which represents a dirnode."""
class IFileURI(Interface):
"""I am a URI which represents a filenode."""
def get_size():
"""Return the length (in bytes) of the file that I represent."""
class IMutableFileURI(Interface):
"""I am a URI which represents a mutable filenode."""
class INewDirectoryURI(Interface):
pass
class IReadonlyNewDirectoryURI(Interface):
pass
class IFilesystemNode(Interface):
def get_uri():
"""
Return the URI that can be used by others to get access to this
node. If this node is read-only, the URI will only offer read-only
access. If this node is read-write, the URI will offer read-write
access.
If you have read-write access to a node and wish to share merely
read-only access with others, use get_readonly_uri().
"""
def get_readonly_uri():
"""Return the directory URI that can be used by others to get
read-only access to this directory node. The result is a read-only
URI, regardless of whether this dirnode is read-only or read-write.
If you have merely read-only access to this dirnode,
get_readonly_uri() will return the same thing as get_uri().
"""
def get_verifier():
"""Return an IVerifierURI instance that represents the
'verifiy/refresh capability' for this node. The holder of this
capability will be able to renew the lease for this node, protecting
it from garbage-collection. They will also be able to ask a server if
it holds a share for the file or directory.
"""
def get_storage_index():
"""Return a string with the (binary) storage index in use on this
download. This may be None if there is no storage index (i.e. LIT
files)."""
def is_readonly():
"""Return True if this reference provides mutable access to the given
file or directory (i.e. if you can modify it), or False if not. Note
that even if this reference is read-only, someone else may hold a
read-write reference to it."""
def is_mutable():
"""Return True if this file or directory is mutable (by *somebody*,
not necessarily you), False if it is is immutable. Note that a file
might be mutable overall, but your reference to it might be
read-only. On the other hand, all references to an immutable file
will be read-only; there are no read-write references to an immutable
file.
"""
class IMutableFilesystemNode(IFilesystemNode):
pass
class IFileNode(IFilesystemNode):
def download(target):
"""Download the file's contents to a given IDownloadTarget"""
def download_to_data():
"""Download the file's contents. Return a Deferred that fires
with those contents."""
def get_size():
"""Return the length (in bytes) of the data this node represents."""
def read(consumer, offset=0, size=None):
"""Download a portion (possibly all) of the file's contents, making
them available to the given IConsumer. Return a Deferred that fires
(with the consumer) when the consumer is unregistered (either because
the last byte has been given to it, or because the consumer threw an
exception during write(), possibly because it no longer wants to
receive data). The portion downloaded will start at 'offset' and
contain 'size' bytes (or the remainder of the file if size==None).
The consumer will be used in non-streaming mode: an IPullProducer
will be attached to it.
The consumer will not receive data right away: several network trips
must occur first. The order of events will be::
consumer.registerProducer(p, streaming)
(if streaming == False)::
consumer does p.resumeProducing()
consumer.write(data)
consumer does p.resumeProducing()
consumer.write(data).. (repeat until all data is written)
consumer.unregisterProducer()
deferred.callback(consumer)
If a download error occurs, or an exception is raised by
consumer.registerProducer() or consumer.write(), I will call
consumer.unregisterProducer() and then deliver the exception via
deferred.errback(). To cancel the download, the consumer should call
p.stopProducing(), which will result in an exception being delivered
via deferred.errback().
A simple download-to-memory consumer example would look like this::
class MemoryConsumer:
implements(IConsumer)
def __init__(self):
self.chunks = []
self.done = False
def registerProducer(self, p, streaming):
assert streaming == False
while not self.done:
p.resumeProducing()
def write(self, data):
self.chunks.append(data)
def unregisterProducer(self):
self.done = True
d = filenode.read(MemoryConsumer())
d.addCallback(lambda mc: "".join(mc.chunks))
return d
"""
class IMutableFileNode(IFileNode, IMutableFilesystemNode):
"""I provide access to a 'mutable file', which retains its identity
regardless of what contents are put in it.
The consistency-vs-availability problem means that there might be
multiple versions of a file present in the grid, some of which might be
unrecoverable (i.e. have fewer than 'k' shares). These versions are
loosely ordered: each has a sequence number and a hash, and any version
with seqnum=N was uploaded by a node which has seen at least one version
with seqnum=N-1.
The 'servermap' (an instance of IMutableFileServerMap) is used to
describe the versions that are known to be present in the grid, and which
servers are hosting their shares. It is used to represent the 'state of
the world', and is used for this purpose by my test-and-set operations.
Downloading the contents of the mutable file will also return a
servermap. Uploading a new version into the mutable file requires a
servermap as input, and the semantics of the replace operation is
'replace the file with my new version if it looks like nobody else has
changed the file since my previous download'. Because the file is
distributed, this is not a perfect test-and-set operation, but it will do
its best. If the replace process sees evidence of a simultaneous write,
it will signal an UncoordinatedWriteError, so that the caller can take
corrective action.
Most readers will want to use the 'best' current version of the file, and
should use my 'download_best_version()' method.
To unconditionally replace the file, callers should use overwrite(). This
is the mode that user-visible mutable files will probably use.
To apply some delta to the file, call modify() with a callable modifier
function that can apply the modification that you want to make. This is
the mode that dirnodes will use, since most directory modification
operations can be expressed in terms of deltas to the directory state.
Three methods are available for users who need to perform more complex
operations. The first is get_servermap(), which returns an up-to-date
servermap using a specified mode. The second is download_version(), which
downloads a specific version (not necessarily the 'best' one). The third
is 'upload', which accepts new contents and a servermap (which must have
been updated with MODE_WRITE). The upload method will attempt to apply
the new contents as long as no other node has modified the file since the
servermap was updated. This might be useful to a caller who wants to
merge multiple versions into a single new one.
Note that each time the servermap is updated, a specific 'mode' is used,
which determines how many peers are queried. To use a servermap for my
replace() method, that servermap must have been updated in MODE_WRITE.
These modes are defined in allmydata.mutable.common, and consist of
MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
allmydata/mutable/servermap.py for details about the differences.
Mutable files are currently limited in size (about 3.5MB max) and can
only be retrieved and updated all-at-once, as a single big string. Future
versions of our mutable files will remove this restriction.
"""
def download_best_version():
"""Download the 'best' available version of the file, meaning one of
the recoverable versions with the highest sequence number. If no
uncoordinated writes have occurred, and if enough shares are
available, then this will be the most recent version that has been
uploaded.
I update an internal servermap with MODE_READ, determine which
version of the file is indicated by
servermap.best_recoverable_version(), and return a Deferred that
fires with its contents. If no version is recoverable, the Deferred
will errback with UnrecoverableFileError.
"""
def get_size_of_best_version():
"""Find the size of the version that would be downloaded with
download_best_version(), without actually downloading the whole file.
I return a Deferred that fires with an integer.
"""
def overwrite(new_contents):
"""Unconditionally replace the contents of the mutable file with new
ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
is only appropriate to use when the new contents of the file are
completely unrelated to the old ones, and you do not care about other
clients' changes.
I return a Deferred that fires (with a PublishStatus object) when the
update has completed.
"""
def modify(modifier_cb):
"""Modify the contents of the file, by downloading the current
version, applying the modifier function (or bound method), then
uploading the new version. I return a Deferred that fires (with a
PublishStatus object) when the update is complete.
The modifier callable will be given two arguments: a string (with the
old contents) and a servermap. As with download_best_version(), the
old contents will be from the best recoverable version, but the
modifier can use the servermap to make other decisions (such as
refusing to apply the delta if there are multiple parallel versions,
or if there is evidence of a newer unrecoverable version).
The callable should return a string with the new contents. The
callable must be prepared to be called multiple times, and must
examine the input string to see if the change that it wants to make
is already present in the old version. If it does not need to make
any changes, it can either return None, or return its input string.
If the modifier raises an exception, it will be returned in the
errback.
"""
def get_servermap(mode):
"""Return a Deferred that fires with an IMutableFileServerMap
instance, updated using the given mode.
"""
def download_version(servermap, version):
"""Download a specific version of the file, using the servermap
as a guide to where the shares are located.
I return a Deferred that fires with the requested contents, or
errbacks with UnrecoverableFileError. Note that a servermap which was
updated with MODE_ANYTHING or MODE_READ may not know about shares for
all versions (those modes stop querying servers as soon as they can
fulfil their goals), so you may want to use MODE_CHECK (which checks
everything) to get increased visibility.
"""
def upload(new_contents, servermap):
"""Replace the contents of the file with new ones. This requires a
servermap that was previously updated with MODE_WRITE.
I attempt to provide test-and-set semantics, in that I will avoid
modifying any share that is different than the version I saw in the
servermap. However, if another node is writing to the file at the
same time as me, I may manage to update some shares while they update
others. If I see any evidence of this, I will signal
UncoordinatedWriteError, and the file will be left in an inconsistent
state (possibly the version you provided, possibly the old version,
possibly somebody else's version, and possibly a mix of shares from
all of these).
The recommended response to UncoordinatedWriteError is to either
return it to the caller (since they failed to coordinate their
writes), or to attempt some sort of recovery. It may be sufficient to
wait a random interval (with exponential backoff) and repeat your
operation. If I do not signal UncoordinatedWriteError, then I was
able to write the new version without incident.
I return a Deferred that fires (with a PublishStatus object) when the
publish has completed. I will update the servermap in-place with the
location of all new shares.
"""
def get_writekey():
"""Return this filenode's writekey, or None if the node does not have
write-capability. This may be used to assist with data structures
that need to make certain data available only to writers, such as the
read-write child caps in dirnodes. The recommended process is to have
reader-visible data be submitted to the filenode in the clear (where
it will be encrypted by the filenode using the readkey), but encrypt
writer-visible data using this writekey.
"""
class NotEnoughSharesError(Exception):
servermap = None
class ExistingChildError(Exception):
"""A directory node was asked to add or replace a child that already
exists, and overwrite= was set to False."""
class NoSuchChildError(Exception):
"""A directory node was asked to fetch a child which does not exist."""
class IDirectoryNode(IMutableFilesystemNode):
"""I represent a name-to-child mapping, holding the tahoe equivalent of a
directory. All child names are unicode strings, and all children are some
sort of IFilesystemNode (either files or subdirectories).
"""
def get_uri():
"""
The dirnode ('1') URI returned by this method can be used in
set_uri() on a different directory ('2') to 'mount' a reference to
this directory ('1') under the other ('2'). This URI is just a
string, so it can be passed around through email or other out-of-band
protocol.
"""
def get_readonly_uri():
"""
The dirnode ('1') URI returned by this method can be used in
set_uri() on a different directory ('2') to 'mount' a reference to
this directory ('1') under the other ('2'). This URI is just a
string, so it can be passed around through email or other out-of-band
protocol.
"""
def list():
"""I return a Deferred that fires with a dictionary mapping child
name (a unicode string) to (node, metadata_dict) tuples, in which
'node' is either an IFileNode or IDirectoryNode, and 'metadata_dict'
is a dictionary of metadata."""
def has_child(name):
"""I return a Deferred that fires with a boolean, True if there
exists a child of the given name, False if not. The child name must
be a unicode string."""
def get(name):
"""I return a Deferred that fires with a specific named child node,
either an IFileNode or an IDirectoryNode. The child name must be a
unicode string. I raise NoSuchChildError if I do not have a child by
that name."""
def get_metadata_for(name):
"""I return a Deferred that fires with the metadata dictionary for a
specific named child node. This metadata is stored in the *edge*, not
in the child, so it is attached to the parent dirnode rather than the
child dir-or-file-node. The child name must be a unicode string. I
raise NoSuchChildError if I do not have a child by that name."""
def set_metadata_for(name, metadata):
"""I replace any existing metadata for the named child with the new
metadata. The child name must be a unicode string. This metadata is
stored in the *edge*, not in the child, so it is attached to the
parent dirnode rather than the child dir-or-file-node. I return a
Deferred (that fires with this dirnode) when the operation is
complete. I raise NoSuchChildError if I do not have a child by that
name."""
def get_child_at_path(path):
"""Transform a child path into an IDirectoryNode or IFileNode.
I perform a recursive series of 'get' operations to find the named
descendant node. I return a Deferred that fires with the node, or
errbacks with NoSuchChildError if the node could not be found.
The path can be either a single string (slash-separated) or a list of
path-name elements. All elements must be unicode strings.
"""
def get_child_and_metadata_at_path(path):
"""Transform a child path into an IDirectoryNode/IFileNode and
metadata.
I am like get_child_at_path(), but my Deferred fires with a tuple of
(node, metadata). The metadata comes from the last edge. If the path
is empty, the metadata will be an empty dictionary.
"""
def set_uri(name, child_uri, metadata=None, overwrite=True):
"""I add a child (by URI) at the specific name. I return a Deferred
that fires when the operation finishes. If overwrite= is True, I will
replace any existing child of the same name, otherwise an existing
child will cause me to return ExistingChildError. The child name must
be a unicode string.
The child_uri could be for a file, or for a directory (either
read-write or read-only, using a URI that came from get_uri() ).
If metadata= is provided, I will use it as the metadata for the named
edge. This will replace any existing metadata. If metadata= is left
as the default value of None, I will set ['mtime'] to the current
time, and I will set ['ctime'] to the current time if there was not
already a child by this name present. This roughly matches the
ctime/mtime semantics of traditional filesystems.
If this directory node is read-only, the Deferred will errback with a
NotMutableError."""
def set_children(entries, overwrite=True):
"""Add multiple (name, child_uri) pairs (or (name, child_uri,
metadata) triples) to a directory node. Returns a Deferred that fires
(with None) when the operation finishes. This is equivalent to
calling set_uri() multiple times, but is much more efficient. All
child names must be unicode strings.
"""
def set_node(name, child, metadata=None, overwrite=True):
"""I add a child at the specific name. I return a Deferred that fires
when the operation finishes. This Deferred will fire with the child
node that was just added. I will replace any existing child of the
same name. The child name must be a unicode string. The 'child'
instance must be an instance providing IDirectoryNode or IFileNode.
If metadata= is provided, I will use it as the metadata for the named
edge. This will replace any existing metadata. If metadata= is left
as the default value of None, I will set ['mtime'] to the current
time, and I will set ['ctime'] to the current time if there was not
already a child by this name present. This roughly matches the
ctime/mtime semantics of traditional filesystems.
If this directory node is read-only, the Deferred will errback with a
NotMutableError."""
def set_nodes(entries, overwrite=True):
"""Add multiple (name, child_node) pairs (or (name, child_node,
metadata) triples) to a directory node. Returns a Deferred that fires
(with None) when the operation finishes. This is equivalent to
calling set_node() multiple times, but is much more efficient. All
child names must be unicode strings."""
def add_file(name, uploadable, metadata=None, overwrite=True):
"""I upload a file (using the given IUploadable), then attach the
resulting FileNode to the directory at the given name. I set metadata
the same way as set_uri and set_node. The child name must be a
unicode string.
I return a Deferred that fires (with the IFileNode of the uploaded
file) when the operation completes."""
def delete(name):
"""I remove the child at the specific name. I return a Deferred that
fires when the operation finishes. The child name must be a unicode
string. I raise NoSuchChildError if I do not have a child by that
name."""
def create_empty_directory(name, overwrite=True):
"""I create and attach an empty directory at the given name. The
child name must be a unicode string. I return a Deferred that fires
when the operation finishes."""
def move_child_to(current_child_name, new_parent, new_child_name=None,
overwrite=True):
"""I take one of my children and move them to a new parent. The child
is referenced by name. On the new parent, the child will live under
'new_child_name', which defaults to 'current_child_name'. TODO: what
should we do about metadata? I return a Deferred that fires when the
operation finishes. The child name must be a unicode string. I raise
NoSuchChildError if I do not have a child by that name."""
def build_manifest():
"""I generate a table of everything reachable from this directory.
I also compute deep-stats as described below.
I return a Monitor. The Monitor's results will be a dictionary with
four elements:
res['manifest']: a list of (path, cap) tuples for all nodes
(directories and files) reachable from this one.
'path' will be a tuple of unicode strings. The
origin dirnode will be represented by an empty path
tuple.
res['verifycaps']: a list of (printable) verifycap strings, one for
each reachable non-LIT node. This is a set:
it will contain no duplicates.
res['storage-index']: a list of (base32) storage index strings,
one for each reachable non-LIT node. This is
a set: it will contain no duplicates.
res['stats']: a dictionary, the same that is generated by
start_deep_stats() below.
The Monitor will also have an .origin_si attribute with the (binary)
storage index of the starting point.
"""
def start_deep_stats():
"""Return a Monitor, examining all nodes (directories and files)
reachable from this one. The Monitor's results will be a dictionary
with the following keys::
count-immutable-files: count of how many CHK files are in the set
count-mutable-files: same, for mutable files (does not include
directories)
count-literal-files: same, for LIT files
count-files: sum of the above three
count-directories: count of directories
size-immutable-files: total bytes for all CHK files in the set
size-mutable-files (TODO): same, for current version of all mutable
files, does not include directories
size-literal-files: same, for LIT files
size-directories: size of mutable files used by directories
largest-directory: number of bytes in the largest directory
largest-directory-children: number of children in the largest
directory
largest-immutable-file: number of bytes in the largest CHK file
size-mutable-files is not yet implemented, because it would involve
even more queries than deep_stats does.
The Monitor will also have an .origin_si attribute with the (binary)
storage index of the starting point.
This operation will visit every directory node underneath this one,
and can take a long time to run. On a typical workstation with good
bandwidth, this can examine roughly 15 directories per second (and
takes several minutes of 100% CPU for ~1700 directories).
"""
class ICodecEncoder(Interface):
def set_params(data_size, required_shares, max_shares):
"""Set up the parameters of this encoder.
This prepares the encoder to perform an operation that converts a
single block of data into a number of shares, such that a future
ICodecDecoder can use a subset of these shares to recover the
original data. This operation is invoked by calling encode(). Once
the encoding parameters are set up, the encode operation can be
invoked multiple times.
set_params() prepares the encoder to accept blocks of input data that
are exactly 'data_size' bytes in length. The encoder will be prepared
to produce 'max_shares' shares for each encode() operation (although
see the 'desired_share_ids' to use less CPU). The encoding math will
be chosen such that the decoder can get by with as few as
'required_shares' of these shares and still reproduce the original
data. For example, set_params(1000, 5, 5) offers no redundancy at
all, whereas set_params(1000, 1, 10) provides 10x redundancy.
Numerical Restrictions: 'data_size' is required to be an integral
multiple of 'required_shares'. In general, the caller should choose
required_shares and max_shares based upon their reliability
requirements and the number of peers available (the total storage
space used is roughly equal to max_shares*data_size/required_shares),
then choose data_size to achieve the memory footprint desired (larger
data_size means more efficient operation, smaller data_size means
smaller memory footprint).
In addition, 'max_shares' must be equal to or greater than
'required_shares'. Of course, setting them to be equal causes
encode() to degenerate into a particularly slow form of the 'split'
utility.
See encode() for more details about how these parameters are used.
set_params() must be called before any other ICodecEncoder methods
may be invoked.
"""
def get_encoder_type():
"""Return a short string that describes the type of this encoder.
There is required to be a global table of encoder classes. This method
returns an index into this table; the value at this index is an
encoder class, and this encoder is an instance of that class.
"""
def get_serialized_params(): # TODO: maybe, maybe not
"""Return a string that describes the parameters of this encoder.
This string can be passed to the decoder to prepare it for handling
the encoded shares we create. It might contain more information than
was presented to set_params(), if there is some flexibility of
parameter choice.
This string is intended to be embedded in the URI, so there are
several restrictions on its contents. At the moment I'm thinking that
this means it may contain hex digits and hyphens, and nothing else.
The idea is that the URI contains something like '%s:%s:%s' %
(encoder.get_encoder_name(), encoder.get_serialized_params(),
b2a(crypttext_hash)), and this is enough information to construct a
compatible decoder.
"""
def get_block_size():
"""Return the length of the shares that encode() will produce.
"""
def encode_proposal(data, desired_share_ids=None):
"""Encode some data.
'data' must be a string (or other buffer object), and len(data) must
be equal to the 'data_size' value passed earlier to set_params().
This will return a Deferred that will fire with two lists. The first
is a list of shares, each of which is a string (or other buffer
object) such that len(share) is the same as what get_share_size()
returned earlier. The second is a list of shareids, in which each is
an integer. The lengths of the two lists will always be equal to each
other. The user should take care to keep each share closely
associated with its shareid, as one is useless without the other.
The length of this output list will normally be the same as the value
provided to the 'max_shares' parameter of set_params(). This may be
different if 'desired_share_ids' is provided.
'desired_share_ids', if provided, is required to be a sequence of
ints, each of which is required to be >= 0 and < max_shares. If not
provided, encode() will produce 'max_shares' shares, as if
'desired_share_ids' were set to range(max_shares). You might use this
if you initially thought you were going to use 10 peers, started
encoding, and then two of the peers dropped out: you could use
desired_share_ids= to skip the work (both memory and CPU) of
producing shares for the peers which are no longer available.
"""
def encode(inshares, desired_share_ids=None):
"""Encode some data. This may be called multiple times. Each call is
independent.
inshares is a sequence of length required_shares, containing buffers
(i.e. strings), where each buffer contains the next contiguous
non-overlapping segment of the input data. Each buffer is required to
be the same length, and the sum of the lengths of the buffers is
required to be exactly the data_size promised by set_params(). (This
implies that the data has to be padded before being passed to
encode(), unless of course it already happens to be an even multiple
of required_shares in length.)
ALSO: the requirement to break up your data into 'required_shares'
chunks before calling encode() feels a bit surprising, at least from
the point of view of a user who doesn't know how FEC works. It feels
like an implementation detail that has leaked outside the
abstraction barrier. Can you imagine a use case in which the data to
be encoded might already be available in pre-segmented chunks, such
that it is faster or less work to make encode() take a list rather
than splitting a single string?
ALSO ALSO: I think 'inshares' is a misleading term, since encode()
is supposed to *produce* shares, so what it *accepts* should be
something other than shares. Other places in this interface use the
word 'data' for that-which-is-not-shares.. maybe we should use that
term?
'desired_share_ids', if provided, is required to be a sequence of
ints, each of which is required to be >= 0 and < max_shares. If not
provided, encode() will produce 'max_shares' shares, as if
'desired_share_ids' were set to range(max_shares). You might use this
if you initially thought you were going to use 10 peers, started
encoding, and then two of the peers dropped out: you could use
desired_share_ids= to skip the work (both memory and CPU) of
producing shares for the peers which are no longer available.
For each call, encode() will return a Deferred that fires with two
lists, one containing shares and the other containing the shareids.
The get_share_size() method can be used to determine the length of
the share strings returned by encode(). Each shareid is a small
integer, exactly as passed into 'desired_share_ids' (or
range(max_shares), if desired_share_ids was not provided).
The shares and their corresponding shareids are required to be kept
together during storage and retrieval. Specifically, the share data is
useless by itself: the decoder needs to be told which share is which
by providing it with both the shareid and the actual share data.
This function will allocate an amount of memory roughly equal to::
(max_shares - required_shares) * get_share_size()
When combined with the memory that the caller must allocate to
provide the input data, this leads to a memory footprint roughly
equal to the size of the resulting encoded shares (i.e. the expansion
factor times the size of the input segment).
"""
# rejected ideas:
#
# returning a list of (shareidN,shareN) tuples instead of a pair of
# lists (shareids..,shares..). Brian thought the tuples would
# encourage users to keep the share and shareid together throughout
# later processing, Zooko pointed out that the code to iterate
# through two lists is not really more complicated than using a list
# of tuples and there's also a performance improvement
#
# having 'data_size' not required to be an integral multiple of
# 'required_shares'. Doing this would require encode() to perform
# padding internally, and we'd prefer to have any padding be done
# explicitly by the caller. Yes, it is an abstraction leak, but
# hopefully not an onerous one.
class ICodecDecoder(Interface):
def set_serialized_params(params):
"""Set up the parameters of this encoder, from a string returned by
encoder.get_serialized_params()."""
def get_needed_shares():
"""Return the number of shares needed to reconstruct the data.
set_serialized_params() is required to be called before this."""
def decode(some_shares, their_shareids):
"""Decode a partial list of shares into data.
'some_shares' is required to be a sequence of buffers of sharedata, a
subset of the shares returned by ICodecEncode.encode(). Each share is
required to be of the same length. The i'th element of their_shareids
is required to be the shareid of the i'th buffer in some_shares.
This returns a Deferred which fires with a sequence of buffers. This
sequence will contain all of the segments of the original data, in
order. The sum of the lengths of all of the buffers will be the
'data_size' value passed into the original ICodecEncode.set_params()
call. To get back the single original input block of data, use
''.join(output_buffers), or you may wish to simply write them in
order to an output file.
Note that some of the elements in the result sequence may be
references to the elements of the some_shares input sequence. In
particular, this means that if those share objects are mutable (e.g.
arrays) and if they are changed, then both the input (the
'some_shares' parameter) and the output (the value given when the
deferred is triggered) will change.
The length of 'some_shares' is required to be exactly the value of
'required_shares' passed into the original ICodecEncode.set_params()
call.
"""
class IEncoder(Interface):
"""I take an object that provides IEncryptedUploadable, which provides
encrypted data, and a list of shareholders. I then encode, hash, and
deliver shares to those shareholders. I will compute all the necessary
Merkle hash trees that are necessary to validate the crypttext that
eventually comes back from the shareholders. I provide the URI Extension
Block Hash, and the encoding parameters, both of which must be included
in the URI.
I do not choose shareholders, that is left to the IUploader. I must be
given a dict of RemoteReferences to storage buckets that are ready and
willing to receive data.
"""
def set_size(size):
"""Specify the number of bytes that will be encoded. This must be
peformed before get_serialized_params() can be called.
"""
def set_params(params):
"""Override the default encoding parameters. 'params' is a tuple of
(k,d,n), where 'k' is the number of required shares, 'd' is the
shares_of_happiness, and 'n' is the total number of shares that will
be created.
Encoding parameters can be set in three ways. 1: The Encoder class
provides defaults (3/7/10). 2: the Encoder can be constructed with
an 'options' dictionary, in which the
needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
set_params((k,d,n)) can be called.
If you intend to use set_params(), you must call it before
get_share_size or get_param are called.
"""
def set_encrypted_uploadable(u):
"""Provide a source of encrypted upload data. 'u' must implement
IEncryptedUploadable.
When this is called, the IEncryptedUploadable will be queried for its
length and the storage_index that should be used.
This returns a Deferred that fires with this Encoder instance.
This must be performed before start() can be called.
"""
def get_param(name):
"""Return an encoding parameter, by name.
'storage_index': return a string with the (16-byte truncated SHA-256
hash) storage index to which these shares should be
pushed.
'share_counts': return a tuple describing how many shares are used:
(needed_shares, shares_of_happiness, total_shares)
'num_segments': return an int with the number of segments that
will be encoded.
'segment_size': return an int with the size of each segment.
'block_size': return the size of the individual blocks that will
be delivered to a shareholder's put_block() method. By
knowing this, the shareholder will be able to keep all
blocks in a single file and still provide random access
when reading them. # TODO: can we avoid exposing this?
'share_size': an int with the size of the data that will be stored
on each shareholder. This is aggregate amount of data
that will be sent to the shareholder, summed over all
the put_block() calls I will ever make. It is useful to
determine this size before asking potential
shareholders whether they will grant a lease or not,
since their answers will depend upon how much space we
need. TODO: this might also include some amount of
overhead, like the size of all the hashes. We need to
decide whether this is useful or not.
'serialized_params': a string with a concise description of the
codec name and its parameters. This may be passed
into the IUploadable to let it make sure that
the same file encoded with different parameters
will result in different storage indexes.
Once this is called, set_size() and set_params() may not be called.
"""
def set_shareholders(shareholders):
"""Tell the encoder where to put the encoded shares. 'shareholders'
must be a dictionary that maps share number (an integer ranging from
0 to n-1) to an instance that provides IStorageBucketWriter. This
must be performed before start() can be called."""
def start():
"""Begin the encode/upload process. This involves reading encrypted
data from the IEncryptedUploadable, encoding it, uploading the shares
to the shareholders, then sending the hash trees.
set_encrypted_uploadable() and set_shareholders() must be called
before this can be invoked.
This returns a Deferred that fires with a tuple of
(uri_extension_hash, needed_shares, total_shares, size) when the
upload process is complete. This information, plus the encryption
key, is sufficient to construct the URI.
"""
class IDecoder(Interface):
"""I take a list of shareholders and some setup information, then
download, validate, decode, and decrypt data from them, writing the
results to an output file.
I do not locate the shareholders, that is left to the IDownloader. I must
be given a dict of RemoteReferences to storage buckets that are ready to
send data.
"""
def setup(outfile):
"""I take a file-like object (providing write and close) to which all
the plaintext data will be written.
TODO: producer/consumer . Maybe write() should return a Deferred that
indicates when it will accept more data? But probably having the
IDecoder be a producer is easier to glue to IConsumer pieces.
"""
def set_shareholders(shareholders):
"""I take a dictionary that maps share identifiers (small integers)
to RemoteReferences that provide RIBucketReader. This must be called
before start()."""
def start():
"""I start the download. This process involves retrieving data and
hash chains from the shareholders, using the hashes to validate the
data, decoding the shares into segments, decrypting the segments,
then writing the resulting plaintext to the output file.
I return a Deferred that will fire (with self) when the download is
complete.
"""
class IDownloadTarget(Interface):
# Note that if the IDownloadTarget is also an IConsumer, the downloader
# will register itself as a producer. This allows the target to invoke
# downloader.pauseProducing, resumeProducing, and stopProducing.
def open(size):
"""Called before any calls to write() or close(). If an error
occurs before any data is available, fail() may be called without
a previous call to open().
'size' is the length of the file being downloaded, in bytes."""
def write(data):
"""Output some data to the target."""
def close():
"""Inform the target that there is no more data to be written."""
def fail(why):
"""fail() is called to indicate that the download has failed. 'why'
is a Failure object indicating what went wrong. No further methods
will be invoked on the IDownloadTarget after fail()."""
def register_canceller(cb):
"""The FileDownloader uses this to register a no-argument function
that the target can call to cancel the download. Once this canceller
is invoked, no further calls to write() or close() will be made."""
def finish():
"""When the FileDownloader is done, this finish() function will be
called. Whatever it returns will be returned to the invoker of
Downloader.download.
"""
class IDownloader(Interface):
def download(uri, target):
"""Perform a CHK download, sending the data to the given target.
'target' must provide IDownloadTarget.
Returns a Deferred that fires (with the results of target.finish)
when the download is finished, or errbacks if something went wrong."""
class IEncryptedUploadable(Interface):
def set_upload_status(upload_status):
"""Provide an IUploadStatus object that should be filled with status
information. The IEncryptedUploadable is responsible for setting
key-determination progress ('chk'), size, storage_index, and
ciphertext-fetch progress. It may delegate some of this
responsibility to others, in particular to the IUploadable."""
def get_size():
"""This behaves just like IUploadable.get_size()."""
def get_all_encoding_parameters():
"""Return a Deferred that fires with a tuple of
(k,happy,n,segment_size). The segment_size will be used as-is, and
must match the following constraints: it must be a multiple of k, and
it shouldn't be unreasonably larger than the file size (if
segment_size is larger than filesize, the difference must be stored
as padding).
This usually passes through to the IUploadable method of the same
name.
The encoder strictly obeys the values returned by this method. To
make an upload use non-default encoding parameters, you must arrange
to control the values that this method returns.
"""
def get_storage_index():
"""Return a Deferred that fires with a 16-byte storage index.
"""
def read_encrypted(length, hash_only):
"""This behaves just like IUploadable.read(), but returns crypttext
instead of plaintext. If hash_only is True, then this discards the
data (and returns an empty list); this improves efficiency when
resuming an interrupted upload (where we need to compute the
plaintext hashes, but don't need the redundant encrypted data)."""
def get_plaintext_hashtree_leaves(first, last, num_segments):
"""Get the leaf nodes of a merkle hash tree over the plaintext
segments, i.e. get the tagged hashes of the given segments. The
segment size is expected to be generated by the IEncryptedUploadable
before any plaintext is read or ciphertext produced, so that the
segment hashes can be generated with only a single pass.
This returns a Deferred which fires with a sequence of hashes, using:
tuple(segment_hashes[first:last])
'num_segments' is used to assert that the number of segments that the
IEncryptedUploadable handled matches the number of segments that the
encoder was expecting.
This method must not be called until the final byte has been read
from read_encrypted(). Once this method is called, read_encrypted()
can never be called again.
"""
def get_plaintext_hash():
"""Get the hash of the whole plaintext.
This returns a Deferred which fires with a tagged SHA-256 hash of the
whole plaintext, obtained from hashutil.plaintext_hash(data).
"""
def close():
"""Just like IUploadable.close()."""
class IUploadable(Interface):
def set_upload_status(upload_status):
"""Provide an IUploadStatus object that should be filled with status
information. The IUploadable is responsible for setting
key-determination progress ('chk')."""
def set_default_encoding_parameters(params):
"""Set the default encoding parameters, which must be a dict mapping
strings to ints. The meaningful keys are 'k', 'happy', 'n', and
'max_segment_size'. These might have an influence on the final
encoding parameters returned by get_all_encoding_parameters(), if the
Uploadable doesn't have more specific preferences.
This call is optional: if it is not used, the Uploadable will use
some built-in defaults. If used, this method must be called before
any other IUploadable methods to have any effect.
"""
def get_size():
"""Return a Deferred that will fire with the length of the data to be
uploaded, in bytes. This will be called before the data is actually
used, to compute encoding parameters.
"""
def get_all_encoding_parameters():
"""Return a Deferred that fires with a tuple of
(k,happy,n,segment_size). The segment_size will be used as-is, and
must match the following constraints: it must be a multiple of k, and
it shouldn't be unreasonably larger than the file size (if
segment_size is larger than filesize, the difference must be stored
as padding).
The relative values of k and n allow some IUploadables to request
better redundancy than others (in exchange for consuming more space
in the grid).
Larger values of segment_size reduce hash overhead, while smaller
values reduce memory footprint and cause data to be delivered in
smaller pieces (which may provide a smoother and more predictable
download experience).
The encoder strictly obeys the values returned by this method. To
make an upload use non-default encoding parameters, you must arrange
to control the values that this method returns. One way to influence
them may be to call set_encoding_parameters() before calling
get_all_encoding_parameters().
"""
def get_encryption_key():
"""Return a Deferred that fires with a 16-byte AES key. This key will
be used to encrypt the data. The key will also be hashed to derive
the StorageIndex.
Uploadables which want to achieve convergence should hash their file
contents and the serialized_encoding_parameters to form the key
(which of course requires a full pass over the data). Uploadables can
use the upload.ConvergentUploadMixin class to achieve this
automatically.
Uploadables which do not care about convergence (or do not wish to
make multiple passes over the data) can simply return a
strongly-random 16 byte string.
get_encryption_key() may be called multiple times: the IUploadable is
required to return the same value each time.
"""
def read(length):
"""Return a Deferred that fires with a list of strings (perhaps with
only a single element) which, when concatenated together, contain the
next 'length' bytes of data. If EOF is near, this may provide fewer
than 'length' bytes. The total number of bytes provided by read()
before it signals EOF must equal the size provided by get_size().
If the data must be acquired through multiple internal read
operations, returning a list instead of a single string may help to
reduce string copies.
'length' will typically be equal to (min(get_size(),1MB)/req_shares),
so a 10kB file means length=3kB, 100kB file means length=30kB,
and >=1MB file means length=300kB.
This method provides for a single full pass through the data. Later
use cases may desire multiple passes or access to only parts of the
data (such as a mutable file making small edits-in-place). This API
will be expanded once those use cases are better understood.
"""
def close():
"""The upload is finished, and whatever filehandle was in use may be
closed."""
class IUploadResults(Interface):
"""I am returned by upload() methods. I contain a number of public
attributes which can be read to determine the results of the upload. Some
of these are functional, some are timing information. All of these may be
None.::
.file_size : the size of the file, in bytes
.uri : the CHK read-cap for the file
.ciphertext_fetched : how many bytes were fetched by the helper
.sharemap : dict mapping share number to placement string
.servermap : dict mapping server peerid to a set of share numbers
.timings : dict of timing information, mapping name to seconds (float)
total : total upload time, start to finish
storage_index : time to compute the storage index
peer_selection : time to decide which peers will be used
contacting_helper : initial helper query to upload/no-upload decision
existence_check : helper pre-upload existence check
helper_total : initial helper query to helper finished pushing
cumulative_fetch : helper waiting for ciphertext requests
total_fetch : helper start to last ciphertext response
cumulative_encoding : just time spent in zfec
cumulative_sending : just time spent waiting for storage servers
hashes_and_close : last segment push to shareholder close
total_encode_and_push : first encode to shareholder close
"""
class IDownloadResults(Interface):
"""I am created internally by download() methods. I contain a number of
public attributes which contain details about the download process.::
.file_size : the size of the file, in bytes
.servers_used : set of server peerids that were used during download
.server_problems : dict mapping server peerid to a problem string. Only
servers that had problems (bad hashes, disconnects) are
listed here.
.servermap : dict mapping server peerid to a set of share numbers. Only
servers that had any shares are listed here.
.timings : dict of timing information, mapping name to seconds (float)
peer_selection : time to ask servers about shares
servers_peer_selection : dict of peerid to DYHB-query time
uri_extension : time to fetch a copy of the URI extension block
hashtrees : time to fetch the hash trees
segments : time to fetch, decode, and deliver segments
cumulative_fetch : time spent waiting for storage servers
cumulative_decode : just time spent in zfec
cumulative_decrypt : just time spent in decryption
total : total download time, start to finish
fetch_per_server : dict of peerid to list of per-segment fetch times
"""
class IUploader(Interface):
def upload(uploadable):
"""Upload the file. 'uploadable' must impement IUploadable. This
returns a Deferred which fires with an UploadResults instance, from
which the URI of the file can be obtained as results.uri ."""
def upload_ssk(write_capability, new_version, uploadable):
"""TODO: how should this work?"""
class ICheckable(Interface):
def check(monitor, verify=False):
"""Check upon my health, optionally repairing any problems.
This returns a Deferred that fires with an instance that provides
ICheckerResults, or None if the object is non-distributed (i.e. LIT
files).
The monitor will be checked periodically to see if the operation has
been cancelled. If so, no new queries will be sent, and the Deferred
will fire (with a OperationCancelledError) immediately.
Filenodes and dirnodes (which provide IFilesystemNode) are also
checkable. Instances that represent verifier-caps will be checkable
but not downloadable. Some objects (like LIT files) do not actually
live in the grid, and their checkers return None (non-distributed
files are always healthy).
If verify=False, a relatively lightweight check will be performed: I
will ask all servers if they have a share for me, and I will believe
whatever they say. If there are at least N distinct shares on the
grid, my results will indicate r.is_healthy()==True. This requires a
roundtrip to each server, but does not transfer very much data, so
the network bandwidth is fairly low.
If verify=True, a more resource-intensive check will be performed:
every share will be downloaded, and the hashes will be validated on
every bit. I will ignore any shares that failed their hash checks. If
there are at least N distinct valid shares on the grid, my results
will indicate r.is_healthy()==True. This requires N/k times as much
download bandwidth (and server disk IO) as a regular download. If a
storage server is holding a corrupt share, or is experiencing memory
failures during retrieval, or is malicious or buggy, then
verification will detect the problem, but checking will not.
TODO: any problems seen during checking will be reported to the
health-manager.furl, a centralized object which is responsible for
figuring out why files are unhealthy so corrective action can be
taken.
"""
def check_and_repair(monitor, verify=False):
"""Like check(), but if the file/directory is not healthy, attempt to
repair the damage.
Any non-healthy result will cause an immediate repair operation, to
generate and upload new shares. After repair, the file will be as
healthy as we can make it. Details about what sort of repair is done
will be put in the check-and-repair results. The Deferred will not
fire until the repair is complete.
This returns a Deferred which fires with an instance of
ICheckAndRepairResults."""
class IDeepCheckable(Interface):
def start_deep_check(verify=False):
"""Check upon the health of me and everything I can reach.
This is a recursive form of check(), useable only on dirnodes.
I return a Monitor, with results that are an IDeepCheckResults
object.
"""
def start_deep_check_and_repair(verify=False):
"""Check upon the health of me and everything I can reach. Repair
anything that isn't healthy.
This is a recursive form of check_and_repair(), useable only on
dirnodes.
I return a Monitor, with results that are an
IDeepCheckAndRepairResults object.
"""
class ICheckerResults(Interface):
"""I contain the detailed results of a check/verify operation.
"""
def get_storage_index():
"""Return a string with the (binary) storage index."""
def get_storage_index_string():
"""Return a string with the (printable) abbreviated storage index."""
def get_uri():
"""Return the (string) URI of the object that was checked."""
def is_healthy():
"""Return a boolean, True if the file/dir is fully healthy, False if
it is damaged in any way. Non-distributed LIT files always return
True."""
def is_recoverable():
"""Return a boolean, True if the file/dir can be recovered, False if
not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
files always return True."""
def needs_rebalancing():
"""Return a boolean, True if the file/dir's reliability could be
improved by moving shares to new servers. Non-distributed LIT files
always returne False."""
def get_data():
"""Return a dictionary that describes the state of the file/dir.
Non-distributed LIT files always return an empty dictionary. Normal
files and directories return a dictionary with the following keys
(note that these use base32-encoded strings rather than binary ones)
(also note that for mutable files, these counts are for the 'best'
version)::
count-shares-good: the number of distinct good shares that were found
count-shares-needed: 'k', the number of shares required for recovery
count-shares-expected: 'N', the number of total shares generated
count-good-share-hosts: the number of distinct storage servers with
good shares. If this number is less than
count-shares-good, then some shares are
doubled up, increasing the correlation of
failures. This indicates that one or more
shares should be moved to an otherwise unused
server, if one is available.
count-corrupt-shares: the number of shares with integrity failures
list-corrupt-shares: a list of 'share locators', one for each share
that was found to be corrupt. Each share
locator is a list of (serverid, storage_index,
sharenum).
servers-responding: list of (binary) storage server identifiers,
one for each server which responded to the share
query.
sharemap: dict mapping share identifier to list of serverids
(binary strings). This indicates which servers are holding
which shares. For immutable files, the shareid is an
integer (the share number, from 0 to N-1). For mutable
files, it is a string of the form 'seq%d-%s-sh%d',
containing the sequence number, the roothash, and the
share number.
The following keys are most relevant for mutable files, but immutable
files will provide sensible values too::
count-wrong-shares: the number of shares for versions other than the
'best' one (which is defined as being the
recoverable version with the highest sequence
number, then the highest roothash). These are
either leftover shares from an older version
(perhaps on a server that was offline when an
update occurred), shares from an unrecoverable
newer version, or shares from an alternate
current version that results from an
uncoordinated write collision. For a healthy
file, this will equal 0.
count-recoverable-versions: the number of recoverable versions of
the file. For a healthy file, this will
equal 1.
count-unrecoverable-versions: the number of unrecoverable versions
of the file. For a healthy file, this
will be 0.
"""
def get_summary():
"""Return a string with a brief (one-line) summary of the results."""
def get_report():
"""Return a list of strings with more detailed results."""
class ICheckAndRepairResults(Interface):
"""I contain the detailed results of a check/verify/repair operation.
The IFilesystemNode.check()/verify()/repair() methods all return
instances that provide ICheckAndRepairResults.
"""
def get_storage_index():
"""Return a string with the (binary) storage index."""
def get_storage_index_string():
"""Return a string with the (printable) abbreviated storage index."""
def get_repair_attempted():
"""Return a boolean, True if a repair was attempted."""
def get_repair_successful():
"""Return a boolean, True if repair was attempted and the file/dir
was fully healthy afterwards. False if no repair was attempted or if
a repair attempt failed."""
def get_pre_repair_results():
"""Return an ICheckerResults instance that describes the state of the
file/dir before any repair was attempted."""
def get_post_repair_results():
"""Return an ICheckerResults instance that describes the state of the
file/dir after any repair was attempted. If no repair was attempted,
the pre-repair and post-repair results will be identical."""
class IDeepCheckResults(Interface):
"""I contain the results of a deep-check operation.
This is returned by a call to ICheckable.deep_check().
"""
def get_root_storage_index_string():
"""Return the storage index (abbreviated human-readable string) of
the first object checked."""
def get_counters():
"""Return a dictionary with the following keys::
count-objects-checked: count of how many objects were checked
count-objects-healthy: how many of those objects were completely
healthy
count-objects-unhealthy: how many were damaged in some way
count-objects-unrecoverable: how many were unrecoverable
count-corrupt-shares: how many shares were found to have
corruption, summed over all objects
examined
"""
def get_corrupt_shares():
"""Return a set of (serverid, storage_index, sharenum) for all shares
that were found to be corrupt. Both serverid and storage_index are
binary.
"""
def get_all_results():
"""Return a dictionary mapping pathname (a tuple of strings, ready to
be slash-joined) to an ICheckerResults instance, one for each object
that was checked."""
def get_results_for_storage_index(storage_index):
"""Retrive the ICheckerResults instance for the given (binary)
storage index. Raises KeyError if there are no results for that
storage index."""
def get_stats():
"""Return a dictionary with the same keys as
IDirectoryNode.deep_stats()."""
class IDeepCheckAndRepairResults(Interface):
"""I contain the results of a deep-check-and-repair operation.
This is returned by a call to ICheckable.deep_check_and_repair().
"""
def get_root_storage_index_string():
"""Return the storage index (abbreviated human-readable string) of
the first object checked."""
def get_counters():
"""Return a dictionary with the following keys::
count-objects-checked: count of how many objects were checked
count-objects-healthy-pre-repair: how many of those objects were
completely healthy (before any
repair)
count-objects-unhealthy-pre-repair: how many were damaged in
some way
count-objects-unrecoverable-pre-repair: how many were unrecoverable
count-objects-healthy-post-repair: how many of those objects were
completely healthy (after any
repair)
count-objects-unhealthy-post-repair: how many were damaged in
some way
count-objects-unrecoverable-post-repair: how many were
unrecoverable
count-repairs-attempted: repairs were attempted on this many
objects. The count-repairs- keys will
always be provided, however unless
repair=true is present, they will all
be zero.
count-repairs-successful: how many repairs resulted in healthy
objects
count-repairs-unsuccessful: how many repairs resulted did not
results in completely healthy objects
count-corrupt-shares-pre-repair: how many shares were found to
have corruption, summed over all
objects examined (before any
repair)
count-corrupt-shares-post-repair: how many shares were found to
have corruption, summed over all
objects examined (after any
repair)
"""
def get_stats():
"""Return a dictionary with the same keys as
IDirectoryNode.deep_stats()."""
def get_corrupt_shares():
"""Return a set of (serverid, storage_index, sharenum) for all shares
that were found to be corrupt before any repair was attempted. Both
serverid and storage_index are binary.
"""
def get_remaining_corrupt_shares():
"""Return a set of (serverid, storage_index, sharenum) for all shares
that were found to be corrupt after any repair was completed. Both
serverid and storage_index are binary. These are shares that need
manual inspection and probably deletion.
"""
def get_all_results():
"""Return a dictionary mapping pathname (a tuple of strings, ready to
be slash-joined) to an ICheckAndRepairResults instance, one for each
object that was checked."""
class IRepairable(Interface):
def repair(checker_results):
"""Attempt to repair the given object. Returns a Deferred that fires
with a IRepairResults object.
I must be called with an object that implements ICheckerResults, as
proof that you have actually discovered a problem with this file. I
will use the data in the checker results to guide the repair process,
such as which servers provided bad data and should therefore be
avoided. The ICheckerResults object is inside the
ICheckAndRepairResults object, which is returned by the
ICheckable.check() method::
d = filenode.check(repair=False)
def _got_results(check_and_repair_results):
check_results = check_and_repair_results.get_pre_repair_results()
return filenode.repair(check_results)
d.addCallback(_got_results)
return d
"""
class IRepairResults(Interface):
"""I contain the results of a repair operation."""
class IClient(Interface):
def upload(uploadable):
"""Upload some data into a CHK, get back the UploadResults for it.
@param uploadable: something that implements IUploadable
@return: a Deferred that fires with the UploadResults instance.
To get the URI for this file, use results.uri .
"""
def create_mutable_file(contents=""):
"""Create a new mutable file with contents, get back the URI string.
@param contents: the initial contents to place in the file.
@return: a Deferred that fires with tne (string) SSK URI for the new
file.
"""
def create_empty_dirnode():
"""Create a new dirnode, empty and unattached.
@return: a Deferred that fires with the new IDirectoryNode instance.
"""
def create_node_from_uri(uri):
"""Create a new IFilesystemNode instance from the uri, synchronously.
@param uri: a string or IURI-providing instance. This could be for a
LiteralFileNode, a CHK file node, a mutable file node, or
a directory node
@return: an instance that provides IFilesystemNode (or more usefully one
of its subclasses). File-specifying URIs will result in
IFileNode or IMutableFileNode -providing instances, like
FileNode, LiteralFileNode, or MutableFileNode.
Directory-specifying URIs will result in
IDirectoryNode-providing instances, like NewDirectoryNode.
"""
class IClientStatus(Interface):
def list_all_uploads():
"""Return a list of uploader objects, one for each upload which
currently has an object available (tracked with weakrefs). This is
intended for debugging purposes."""
def list_active_uploads():
"""Return a list of active IUploadStatus objects."""
def list_recent_uploads():
"""Return a list of IUploadStatus objects for the most recently
started uploads."""
def list_all_downloads():
"""Return a list of downloader objects, one for each download which
currently has an object available (tracked with weakrefs). This is
intended for debugging purposes."""
def list_active_downloads():
"""Return a list of active IDownloadStatus objects."""
def list_recent_downloads():
"""Return a list of IDownloadStatus objects for the most recently
started downloads."""
class IUploadStatus(Interface):
def get_started():
"""Return a timestamp (float with seconds since epoch) indicating
when the operation was started."""
def get_storage_index():
"""Return a string with the (binary) storage index in use on this
upload. Returns None if the storage index has not yet been
calculated."""
def get_size():
"""Return an integer with the number of bytes that will eventually
be uploaded for this file. Returns None if the size is not yet known.
"""
def using_helper():
"""Return True if this upload is using a Helper, False if not."""
def get_status():
"""Return a string describing the current state of the upload
process."""
def get_progress():
"""Returns a tuple of floats, (chk, ciphertext, encode_and_push),
each from 0.0 to 1.0 . 'chk' describes how much progress has been
made towards hashing the file to determine a CHK encryption key: if
non-convergent encryption is in use, this will be trivial, otherwise
the whole file must be hashed. 'ciphertext' describes how much of the
ciphertext has been pushed to the helper, and is '1.0' for non-helper
uploads. 'encode_and_push' describes how much of the encode-and-push
process has finished: for helper uploads this is dependent upon the
helper providing progress reports. It might be reasonable to add all
three numbers and report the sum to the user."""
def get_active():
"""Return True if the upload is currently active, False if not."""
def get_results():
"""Return an instance of UploadResults (which contains timing and
sharemap information). Might return None if the upload is not yet
finished."""
def get_counter():
"""Each upload status gets a unique number: this method returns that
number. This provides a handle to this particular upload, so a web
page can generate a suitable hyperlink."""
class IDownloadStatus(Interface):
def get_started():
"""Return a timestamp (float with seconds since epoch) indicating
when the operation was started."""
def get_storage_index():
"""Return a string with the (binary) storage index in use on this
download. This may be None if there is no storage index (i.e. LIT
files)."""
def get_size():
"""Return an integer with the number of bytes that will eventually be
retrieved for this file. Returns None if the size is not yet known.
"""
def using_helper():
"""Return True if this download is using a Helper, False if not."""
def get_status():
"""Return a string describing the current state of the download
process."""
def get_progress():
"""Returns a float (from 0.0 to 1.0) describing the amount of the
download that has completed. This value will remain at 0.0 until the
first byte of plaintext is pushed to the download target."""
def get_active():
"""Return True if the download is currently active, False if not."""
def get_counter():
"""Each download status gets a unique number: this method returns
that number. This provides a handle to this particular download, so a
web page can generate a suitable hyperlink."""
class IServermapUpdaterStatus(Interface):
pass
class IPublishStatus(Interface):
pass
class IRetrieveStatus(Interface):
pass
class NotCapableError(Exception):
"""You have tried to write to a read-only node."""
class BadWriteEnablerError(Exception):
pass
class RIControlClient(RemoteInterface):
def wait_for_client_connections(num_clients=int):
"""Do not return until we have connections to at least NUM_CLIENTS
storage servers.
"""
def upload_from_file_to_uri(filename=str, convergence=ChoiceOf(None, StringConstraint(2**20))):
"""Upload a file to the grid. This accepts a filename (which must be
absolute) that points to a file on the node's local disk. The node will
read the contents of this file, upload it to the grid, then return the
URI at which it was uploaded. If convergence is None then a random
encryption key will be used, else the plaintext will be hashed, then
that hash will be mixed together with the "convergence" string to form
the encryption key.
"""
return URI
def download_from_uri_to_file(uri=URI, filename=str):
"""Download a file from the grid, placing it on the node's local disk
at the given filename (which must be absolute[?]). Returns the
absolute filename where the file was written."""
return str
# debug stuff
def get_memory_usage():
"""Return a dict describes the amount of memory currently in use. The
keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
measuring memory consupmtion in bytes."""
return DictOf(str, int)
def speed_test(count=int, size=int, mutable=Any()):
"""Write 'count' tempfiles to disk, all of the given size. Measure
how long (in seconds) it takes to upload them all to the servers.
Then measure how long it takes to download all of them. If 'mutable'
is 'create', time creation of mutable files. If 'mutable' is
'upload', then time access to the same mutable file instead of
creating one.
Returns a tuple of (upload_time, download_time).
"""
return (float, float)
def measure_peer_response_time():
"""Send a short message to each connected peer, and measure the time
it takes for them to respond to it. This is a rough measure of the
application-level round trip time.
@return: a dictionary mapping peerid to a float (RTT time in seconds)
"""
return DictOf(Nodeid, float)
UploadResults = Any() #DictOf(str, str)
class RIEncryptedUploadable(RemoteInterface):
__remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
def get_size():
return Offset
def get_all_encoding_parameters():
return (int, int, int, long)
def read_encrypted(offset=Offset, length=ReadSize):
return ListOf(str)
def get_plaintext_hashtree_leaves(first=int, last=int, num_segments=int):
return ListOf(Hash)
def get_plaintext_hash():
return Hash
def close():
return None
class RICHKUploadHelper(RemoteInterface):
__remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
def get_version():
"""
Return a dictionary of version information.
"""
return DictOf(str, Any())
def upload(reader=RIEncryptedUploadable):
return UploadResults
class RIHelper(RemoteInterface):
__remote_name__ = "RIHelper.tahoe.allmydata.com"
def get_version():
"""
Return a dictionary of version information.
"""
return DictOf(str, Any())
def upload_chk(si=StorageIndex):
"""See if a file with a given storage index needs uploading. The
helper will ask the appropriate storage servers to see if the file
has already been uploaded. If so, the helper will return a set of
'upload results' that includes whatever hashes are needed to build
the read-cap, and perhaps a truncated sharemap.
If the file has not yet been uploaded (or if it was only partially
uploaded), the helper will return an empty upload-results dictionary
and also an RICHKUploadHelper object that will take care of the
upload process. The client should call upload() on this object and
pass it a reference to an RIEncryptedUploadable object that will
provide ciphertext. When the upload is finished, the upload() method
will finish and return the upload results.
"""
return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
class RIStatsProvider(RemoteInterface):
__remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
"""
Provides access to statistics and monitoring information.
"""
def get_stats():
"""
returns a dictionary containing 'counters' and 'stats', each a dictionary
with string counter/stat name keys, and numeric values. counters are
monotonically increasing measures of work done, and stats are instantaneous
measures (potentially time averaged internally)
"""
return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
class RIStatsGatherer(RemoteInterface):
__remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
"""
Provides a monitoring service for centralised collection of stats
"""
def provide(provider=RIStatsProvider, nickname=str):
"""
@param provider: a stats collector instance which should be polled
periodically by the gatherer to collect stats.
@param nickname: a name useful to identify the provided client
"""
return None
class IStatsProducer(Interface):
def get_stats():
"""
returns a dictionary, with str keys representing the names of stats
to be monitored, and numeric values.
"""
class RIKeyGenerator(RemoteInterface):
__remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
"""
Provides a service offering to make RSA key pairs.
"""
def get_rsa_key_pair(key_size=int):
"""
@param key_size: the size of the signature key.
@return: tuple(verifying_key, signing_key)
"""
return TupleOf(str, str)
class FileTooLargeError(Exception):
pass
class InsufficientVersionError(Exception):
def __init__(self, needed, got):
self.needed = needed
self.got = got
def __repr__(self):
return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
self.got)
Reference in New Issue View Git Blame Copy Permalink