Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

1101 lines (900 sloc) 46.16 kB
"""
Provides an abstraction of Cassandra's data model to allow for easy
manipulation of data inside Cassandra.
.. seealso:: :mod:`pycassa.columnfamilymap`
"""
import time
import struct
from UserDict import DictMixin
from pycassa.cassandra.ttypes import Column, ColumnOrSuperColumn,\
ColumnParent, ColumnPath, ConsistencyLevel, NotFoundException,\
SlicePredicate, SliceRange, SuperColumn, KeyRange,\
IndexExpression, IndexClause, CounterColumn, Mutation
import pycassa.marshal as marshal
import pycassa.types as types
from pycassa.batch import CfMutator
try:
from collections import OrderedDict
except ImportError:
from pycassa.util import OrderedDict
__all__ = ['gm_timestamp', 'ColumnFamily', 'PooledColumnFamily']
class ColumnValidatorDict(DictMixin):
def __init__(self, other_dict={}, name_packer=None, name_unpacker=None):
self.name_packer = name_packer or (lambda x: x)
self.name_unpacker = name_unpacker or (lambda x: x)
self.type_map = {}
self.packers = {}
self.unpackers = {}
for item, value in other_dict.items():
packed_item = self.name_packer(item)
self[packed_item] = value
def __getitem__(self, item):
packed_item = self.name_packer(item)
return self.type_map[packed_item]
def __setitem__(self, item, value):
packed_item = self.name_packer(item)
if isinstance(value, types.CassandraType):
self.type_map[packed_item] = value
self.packers[packed_item] = value.pack
self.unpackers[packed_item] = value.unpack
else:
self.type_map[packed_item] = marshal.extract_type_name(value)
self.packers[packed_item] = marshal.packer_for(value)
self.unpackers[packed_item] = marshal.unpacker_for(value)
def __delitem__(self, item):
packed_item = self.name_packer(item)
del self.type_map[packed_item]
del self.packers[packed_item]
del self.unpackers[packed_item]
def keys(self):
return map(self.name_unpacker, self.type_map.keys())
def gm_timestamp():
""" Gets the current GMT timestamp in microseconds. """
return int(time.time() * 1e6)
class ColumnFamily(object):
"""
An abstraction of a Cassandra column family or super column family.
Operations on this, such as :meth:`get` or :meth:`insert` will get data from or
insert data into the corresponding Cassandra column family.
"""
buffer_size = 1024
""" When calling :meth:`get_range()` or :meth:`get_indexed_slices()`,
the intermediate results need to be buffered if we are fetching many
rows, otherwise performance may suffer and the Cassandra server may
overallocate memory and fail. This is the size of that buffer in number
of rows. The default is 1024. """
column_buffer_size = 1024
""" The number of columns fetched at once for :meth:`xget()` """
read_consistency_level = ConsistencyLevel.ONE
""" The default consistency level for every read operation, such as
:meth:`get` or :meth:`get_range`. This may be overridden per-operation. This should be
an instance of :class:`~pycassa.cassandra.ttypes.ConsistencyLevel`.
The default level is ``ONE``. """
write_consistency_level = ConsistencyLevel.ONE
""" The default consistency level for every write operation, such as
:meth:`insert` or :meth:`remove`. This may be overridden per-operation. This should be
an instance of :class:`.~pycassa.cassandra.ttypes.ConsistencyLevel`.
The default level is ``ONE``. """
timestamp = gm_timestamp
""" Each :meth:`insert()` or :meth:`remove` sends a timestamp with every
column. This attribute is a function that is used to get
this timestamp when needed. The default function is :meth:`gm_timestamp()`."""
dict_class = OrderedDict
""" Results are returned as dictionaries. By default, python 2.7's
:class:`collections.OrderedDict` is used if available, otherwise
:class:`~pycassa.util.OrderedDict` is used so that order is maintained.
A different class, such as :class:`dict`, may be instead by used setting
this. """
autopack_names = True
""" Controls whether column names are automatically converted to or from
their natural type to the binary string format that Cassandra uses.
The data type used is controlled by :attr:`column_name_class` for
column names and :attr:`super_column_name_class` for super column names.
By default, this is :const:`True`. """
autopack_values = True
""" Whether column values are automatically converted to or from
their natural type to the binary string format that Cassandra uses.
The data type used is controlled by :attr:`default_validation_class`
and :attr:`column_validators`.
By default, this is :const:`True`. """
autopack_keys = True
""" Whether row keys are automatically converted to or from
their natural type to the binary string format that Cassandra uses.
The data type used is controlled by :attr:`key_validation_class`.
By default, this is :const:`True`.
"""
retry_counter_mutations = False
""" Whether to retry failed counter mutations. Counter mutations are
not idempotent so retrying could result in double counting.
By default, this is :const:`False`.
.. versionadded:: 1.5.0
"""
def _set_column_name_class(self, t):
if isinstance(t, types.CassandraType):
self._column_name_class = t
self._name_packer = t.pack
self._name_unpacker = t.unpack
else:
self._column_name_class = marshal.extract_type_name(t)
self._name_packer = marshal.packer_for(t)
self._name_unpacker = marshal.unpacker_for(t)
def _get_column_name_class(self):
return self._column_name_class
column_name_class = property(_get_column_name_class, _set_column_name_class)
""" The data type of column names, which pycassa will use
to determine how to pack and unpack them.
This is set automatically by inspecting the column family's
``comparator_type``, but it may also be set manually if you want
autopacking behavior without setting a ``comparator_type``. Options
include an instance of any class in :mod:`pycassa.types`, such as ``LongType()``.
"""
def _set_super_column_name_class(self, t):
if isinstance(t, types.CassandraType):
self._super_column_name_class = t
self._super_name_packer = t.pack
self._super_name_unpacker = t.unpack
else:
self._super_column_name_class = marshal.extract_type_name(t)
self._super_name_packer = marshal.packer_for(t)
self._super_name_unpacker = marshal.unpacker_for(t)
def _get_super_column_name_class(self):
return self._super_column_name_class
super_column_name_class = property(_get_super_column_name_class,
_set_super_column_name_class)
""" Like :attr:`column_name_class`, but for
super column names. """
def _set_default_validation_class(self, t):
if isinstance(t, types.CassandraType):
self._default_validation_class = t
self._default_value_packer = t.pack
self._default_value_unpacker = t.unpack
self._have_counters = isinstance(t, types.CounterColumnType)
else:
self._default_validation_class = marshal.extract_type_name(t)
self._default_value_packer = marshal.packer_for(t)
self._default_value_unpacker = marshal.unpacker_for(t)
self._have_counters = self._default_validation_class == "CounterColumnType"
if not self.super:
if self._have_counters:
def _make_counter_cosc(name, value, timestamp, ttl):
return ColumnOrSuperColumn(counter_column=CounterColumn(name, value))
self._make_cosc = _make_counter_cosc
else:
def _make_normal_cosc(name, value, timestamp, ttl):
return ColumnOrSuperColumn(Column(name, value, timestamp, ttl))
self._make_cosc = _make_normal_cosc
else:
if self._have_counters:
def _make_column(name, value, timestamp, ttl):
return CounterColumn(name, value)
self._make_column = _make_column
def _make_counter_super_cosc(scol_name, subcols):
return ColumnOrSuperColumn(counter_super_column=(SuperColumn(scol_name, subcols)))
self._make_cosc = _make_counter_super_cosc
else:
self._make_column = Column
def _make_super_cosc(scol_name, subcols):
return ColumnOrSuperColumn(super_column=(SuperColumn(scol_name, subcols)))
self._make_cosc = _make_super_cosc
def _get_default_validation_class(self):
return self._default_validation_class
default_validation_class = property(_get_default_validation_class,
_set_default_validation_class)
""" The default data type of column values, which pycassa
will use to determine how to pack and unpack them.
This is set automatically by inspecting the column family's
``default_validation_class``, but it may also be set manually if you want
autopacking behavior without setting a ``default_validation_class``. Options
include an instance of any class in :mod:`pycassa.types`, such as ``LongType()``.
"""
@property
def _allow_retries(self):
return not self._have_counters or self.retry_counter_mutations
def _set_column_validators(self, other_dict):
self._column_validators = ColumnValidatorDict(other_dict, self._pack_name, self._unpack_name)
def _get_column_validators(self):
return self._column_validators
column_validators = property(_get_column_validators, _set_column_validators)
""" Like :attr:`default_validation_class`, but is a
:class:`dict` mapping individual columns to types. """
def _set_key_validation_class(self, t):
if isinstance(t, types.CassandraType):
self._key_validation_class = t
self._key_packer = t.pack
self._key_unpacker = t.unpack
else:
self._key_validation_class = marshal.extract_type_name(t)
self._key_packer = marshal.packer_for(t)
self._key_unpacker = marshal.unpacker_for(t)
def _get_key_validation_class(self):
return self._key_validation_class
key_validation_class = property(_get_key_validation_class,
_set_key_validation_class)
""" The data type of row keys, which pycassa will use
to determine how to pack and unpack them.
This is set automatically by inspecting the column family's
``key_validation_class`` (which only exists in Cassandra 0.8 or greater),
but may be set manually if you want the autopacking behavior without
setting a ``key_validation_class`` or if you are using Cassandra 0.7.
Options include an instance of any class in :mod:`pycassa.types`,
such as ``LongType()``.
"""
def __init__(self, pool, column_family, **kwargs):
"""
`pool` is a :class:`~pycassa.pool.ConnectionPool` that the column
family will use for all operations. A connection is drawn from the
pool before each operations and is returned afterwards.
`column_family` should be the name of the column family that you
want to use in Cassandra. Note that the keyspace to be used is
determined by the pool.
"""
self.pool = pool
self.column_family = column_family
self.timestamp = gm_timestamp
self.load_schema()
recognized_kwargs = ("buffer_size", "read_consistency_level",
"write_consistency_level", "timestamp",
"dict_class", "buffer_size", "autopack_names",
"autopack_values", "autopack_keys",
"retry_counter_mutations")
for k, v in kwargs.iteritems():
if k in recognized_kwargs:
setattr(self, k, v)
else:
raise TypeError(
"ColumnFamily.__init__() got an unexpected keyword "
"argument '%s'" % (k,))
def load_schema(self):
"""
Loads the schema definition for this column family from
Cassandra and updates comparator and validation classes if
neccessary.
"""
ksdef = self.pool.execute('get_keyspace_description',
use_dict_for_col_metadata=True)
try:
self._cfdef = ksdef[self.column_family]
except KeyError:
nfe = NotFoundException()
nfe.why = 'Column family %s not found.' % self.column_family
raise nfe
self.super = self._cfdef.column_type == 'Super'
self._load_comparator_classes()
self._load_validation_classes()
self._load_key_class()
def _load_comparator_classes(self):
if not self.super:
self.column_name_class = self._cfdef.comparator_type
self.super_column_name_class = None
else:
self.column_name_class = self._cfdef.subcomparator_type
self.super_column_name_class = self._cfdef.comparator_type
def _load_validation_classes(self):
self.default_validation_class = self._cfdef.default_validation_class
self.column_validators = {}
for name, coldef in self._cfdef.column_metadata.items():
unpacked_name = self._unpack_name(name)
self.column_validators[unpacked_name] = coldef.validation_class
def _load_key_class(self):
if hasattr(self._cfdef, "key_validation_class"):
self.key_validation_class = self._cfdef.key_validation_class
else:
self.key_validation_class = 'BytesType'
def _col_to_dict(self, column, include_timestamp):
value = self._unpack_value(column.value, column.name)
if include_timestamp:
return (value, column.timestamp)
return value
def _scol_to_dict(self, super_column, include_timestamp):
ret = self.dict_class()
for column in super_column.columns:
ret[self._unpack_name(column.name)] = self._col_to_dict(column, include_timestamp)
return ret
def _scounter_to_dict(self, counter_super_column):
ret = self.dict_class()
for counter in counter_super_column.columns:
ret[self._unpack_name(counter.name)] = counter.value
return ret
def _cosc_to_dict(self, list_col_or_super, include_timestamp):
ret = self.dict_class()
for cosc in list_col_or_super:
if cosc.column:
col = cosc.column
ret[self._unpack_name(col.name)] = self._col_to_dict(col, include_timestamp)
elif cosc.counter_column:
counter = cosc.counter_column
ret[self._unpack_name(counter.name)] = counter.value
elif cosc.super_column:
scol = cosc.super_column
ret[self._unpack_name(scol.name, True)] = self._scol_to_dict(scol, include_timestamp)
else:
scounter = cosc.counter_super_column
ret[self._unpack_name(scounter.name, True)] = self._scounter_to_dict(scounter)
return ret
def _column_path(self, super_column=None, column=None):
return ColumnPath(self.column_family,
self._pack_name(super_column, is_supercol_name=True),
self._pack_name(column, False))
def _column_parent(self, super_column=None):
return ColumnParent(column_family=self.column_family,
super_column=self._pack_name(super_column, is_supercol_name=True))
def _slice_predicate(self, columns, column_start, column_finish,
column_reversed, column_count, super_column=None, pack=True):
is_supercol_name = self.super and super_column is None
if columns is not None:
packed_cols = []
for col in columns:
packed_cols.append(self._pack_name(col, is_supercol_name=is_supercol_name))
return SlicePredicate(column_names=packed_cols)
else:
if column_start != '' and pack:
column_start = self._pack_name(column_start,
is_supercol_name=is_supercol_name,
slice_start=(not column_reversed))
if column_finish != '' and pack:
column_finish = self._pack_name(column_finish,
is_supercol_name=is_supercol_name,
slice_start=column_reversed)
sr = SliceRange(start=column_start, finish=column_finish,
reversed=column_reversed, count=column_count)
return SlicePredicate(slice_range=sr)
def _pack_name(self, value, is_supercol_name=False, slice_start=None):
if value is None:
return
if not self.autopack_names:
if not isinstance(value, basestring):
raise TypeError("A str or unicode column name was expected, " +
"but %s was received instead (%s)"
% (value.__class__.__name__, str(value)))
return value
try:
if is_supercol_name:
return self._super_name_packer(value, slice_start)
else:
return self._name_packer(value, slice_start)
except struct.error:
if is_supercol_name:
d_type = self.super_column_name_class
else:
d_type = self.column_name_class
raise TypeError("%s is not a compatible type for %s" %
(value.__class__.__name__, d_type))
def _unpack_name(self, b, is_supercol_name=False):
if not self.autopack_names:
return b
try:
if is_supercol_name:
return self._super_name_unpacker(b)
else:
return self._name_unpacker(b)
except struct.error:
if is_supercol_name:
d_type = self.super_column_name_class
else:
d_type = self.column_name_class
raise TypeError("%s cannot be converted to a type matching %s" %
(b, d_type))
def _pack_value(self, value, col_name):
if value is None:
return
if not self.autopack_values:
if not isinstance(value, basestring):
raise TypeError("A str or unicode column value was expected for " +
"column '%s', but %s was received instead (%s)"
% (str(col_name), value.__class__.__name__, str(value)))
return value
packed_col_name = self._pack_name(col_name, False)
packer = self._column_validators.packers.get(packed_col_name, self._default_value_packer)
try:
return packer(value)
except struct.error:
d_type = self.column_validators.get(col_name, self._default_validation_class)
raise TypeError("%s is not a compatible type for %s" %
(value.__class__.__name__, d_type))
def _unpack_value(self, value, col_name):
if not self.autopack_values:
return value
unpacker = self._column_validators.unpackers.get(col_name, self._default_value_unpacker)
try:
return unpacker(value)
except struct.error:
d_type = self.column_validators.get(col_name, self.default_validation_class)
raise TypeError("%s cannot be converted to a type matching %s" %
(value, d_type))
def _pack_key(self, key):
if not self.autopack_keys or key == '':
return key
try:
return self._key_packer(key)
except struct.error:
d_type = self.key_validation_class
raise TypeError("%s is not a compatible type for %s" %
(key.__class__.__name__, d_type))
def _unpack_key(self, b):
if not self.autopack_keys:
return b
try:
return self._key_unpacker(b)
except struct.error:
d_type = self.key_validation_class
raise TypeError("%s cannot be converted to a type matching %s" %
(b, d_type))
def _make_mutation_list(self, columns, timestamp, ttl):
_pack_name = self._pack_name
_pack_value = self._pack_value
if not self.super:
return map(lambda (c, v): Mutation(self._make_cosc(_pack_name(c), _pack_value(v, c), timestamp, ttl)),
columns.iteritems())
else:
mut_list = []
for super_col, subcs in columns.items():
subcols = map(lambda (c, v): self._make_column(_pack_name(c), _pack_value(v, c), timestamp, ttl),
subcs.iteritems())
mut_list.append(Mutation(self._make_cosc(_pack_name(super_col, True), subcols)))
return mut_list
def xget(self, key, column_start="", column_finish="", column_reversed=False,
column_count=None, include_timestamp=False, read_consistency_level=None,
buffer_size=None):
"""
Like :meth:`get()`, but creates a generator that pages over the columns
automatically.
The number of columns fetched at once can be controlled with the
`buffer_size` parameter. The default is :attr:`column_buffer_size`.
The generator returns `(name, value)` tuples.
"""
packed_key = self._pack_key(key)
cp = self._column_parent(None)
rcl = read_consistency_level or self.read_consistency_level
if buffer_size is None:
buffer_size = self.column_buffer_size
count = i = 0
last_name = finish = ""
if column_start != "":
last_name = self._pack_name(column_start,
is_supercol_name=self.super,
slice_start=(not column_reversed))
if column_finish != "":
finish = self._pack_name(column_finish,
is_supercol_name=self.super,
slice_start=column_reversed)
while True:
if column_count is not None:
if i == 0 and column_count <= buffer_size:
buffer_size = column_count
else:
buffer_size = min(column_count - count + 1, buffer_size)
sp = self._slice_predicate(None, last_name, finish,
column_reversed, buffer_size, None, pack=False)
list_cosc = self.pool.execute('get_slice', packed_key, cp, sp, rcl)
if not list_cosc:
return
for j, cosc in enumerate(list_cosc):
if j == 0 and i != 0:
continue
if self.super:
if self._have_counters:
scol = cosc.counter_super_column
else:
scol = cosc.super_column
yield (self._unpack_name(scol.name, True), self._scol_to_dict(scol, include_timestamp))
else:
if self._have_counters:
col = cosc.counter_column
else:
col = cosc.column
yield (self._unpack_name(col.name, False), self._col_to_dict(col, include_timestamp))
count += 1
if column_count is not None and count >= column_count:
return
if len(list_cosc) != buffer_size:
return
if self.super:
if self._have_counters:
last_name = list_cosc[-1].counter_super_column.name
else:
last_name = list_cosc[-1].super_column.name
else:
if self._have_counters:
last_name = list_cosc[-1].counter_column.name
else:
last_name = list_cosc[-1].column.name
i += 1
def get(self, key, columns=None, column_start="", column_finish="",
column_reversed=False, column_count=100, include_timestamp=False,
super_column=None, read_consistency_level=None):
"""
Fetches all or part of the row with key `key`.
The columns fetched may be limited to a specified list of column names
using `columns`.
Alternatively, you may fetch a slice of columns or super columns from a row
using `column_start`, `column_finish`, and `column_count`.
Setting these will cause columns or super columns to be fetched starting with
`column_start`, continuing until `column_count` columns or super columns have
been fetched or `column_finish` is reached. If `column_start` is left as the
empty string, the slice will begin with the start of the row; leaving
`column_finish` blank will cause the slice to extend to the end of the row.
Note that `column_count` defaults to 100, so rows over this size will not be
completely fetched by default.
If `column_reversed` is ``True``, columns are fetched in reverse sorted order,
beginning with `column_start`. In this case, if `column_start` is the empty
string, the slice will begin with the end of the row.
You may fetch all or part of only a single super column by setting `super_column`.
If this is set, `column_start`, `column_finish`, `column_count`, and `column_reversed`
will apply to the subcolumns of `super_column`.
To include every column's timestamp in the result set, set `include_timestamp` to
``True``. Results will include a ``(value, timestamp)`` tuple for each column.
If this is a standard column family, the return type is of the form
``{column_name: column_value}``. If this is a super column family and `super_column`
is not specified, the results are of the form
``{super_column_name: {column_name, column_value}}``. If `super_column` is set,
the super column name will be excluded and the results are of the form
``{column_name: column_value}``.
"""
packed_key = self._pack_key(key)
single_column = columns is not None and len(columns) == 1
if (not self.super and single_column) or \
(self.super and super_column is not None and single_column):
column = None
if self.super and super_column is None:
super_column = columns[0]
else:
column = columns[0]
cp = self._column_path(super_column, column)
col_or_super = self.pool.execute('get', packed_key, cp,
read_consistency_level or self.read_consistency_level)
return self._cosc_to_dict([col_or_super], include_timestamp)
else:
cp = self._column_parent(super_column)
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, column_count, super_column)
list_col_or_super = self.pool.execute('get_slice', packed_key, cp, sp,
read_consistency_level or self.read_consistency_level)
if len(list_col_or_super) == 0:
raise NotFoundException()
return self._cosc_to_dict(list_col_or_super, include_timestamp)
def get_indexed_slices(self, index_clause, columns=None, column_start="", column_finish="",
column_reversed=False, column_count=100, include_timestamp=False,
read_consistency_level=None, buffer_size=None):
"""
Similar to :meth:`get_range()`, but an :class:`~pycassa.cassandra.ttypes.IndexClause`
is used instead of a key range.
`index_clause` limits the keys that are returned based on expressions
that compare the value of a column to a given value. At least one of the
expressions in the :class:`.IndexClause` must be on an indexed column.
Note that Cassandra does not support secondary indexes or get_indexed_slices()
for super column families.
.. seealso:: :meth:`~pycassa.index.create_index_clause()` and
:meth:`~pycassa.index.create_index_expression()`
"""
assert not self.super, "get_indexed_slices() is not " \
"supported by super column families"
cl = read_consistency_level or self.read_consistency_level
cp = self._column_parent()
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, column_count)
new_exprs = []
# Pack the values in the index clause expressions
for expr in index_clause.expressions:
value = self._pack_value(expr.value, expr.column_name)
name = self._pack_name(expr.column_name)
new_exprs.append(IndexExpression(name, expr.op, value))
packed_start_key = self._pack_key(index_clause.start_key)
clause = IndexClause(new_exprs, packed_start_key, index_clause.count)
# Figure out how we will chunk the request
if buffer_size is None:
buffer_size = self.buffer_size
row_count = clause.count
count = 0
i = 0
last_key = clause.start_key
while True:
if row_count is not None:
if i == 0 and row_count <= buffer_size:
# We don't need to chunk, grab exactly the number of rows
buffer_size = row_count
else:
buffer_size = min(row_count - count + 1, buffer_size)
clause.count = buffer_size
clause.start_key = last_key
key_slices = self.pool.execute('get_indexed_slices', cp, clause, sp, cl)
if key_slices is None:
return
for j, key_slice in enumerate(key_slices):
# Ignore the first element after the first iteration
# because it will be a duplicate.
if j == 0 and i != 0:
continue
unpacked_key = self._unpack_key(key_slice.key)
yield (unpacked_key,
self._cosc_to_dict(key_slice.columns, include_timestamp))
count += 1
if row_count is not None and count >= row_count:
return
if len(key_slices) != buffer_size:
return
last_key = key_slices[-1].key
i += 1
def multiget(self, keys, columns=None, column_start="", column_finish="",
column_reversed=False, column_count=100, include_timestamp=False,
super_column=None, read_consistency_level=None, buffer_size=None):
"""
Fetch multiple rows from a Cassandra server.
`keys` should be a list of keys to fetch.
`buffer_size` is the number of rows from the total list to fetch at a time.
If left as ``None``, the ColumnFamily's :attr:`buffer_size` will be used.
All other parameters are the same as :meth:`get()`, except that a list of keys may
be passed in.
Results will be returned in the form: ``{key: {column_name: column_value}}``. If
an OrderedDict is used, the rows will have the same order as `keys`.
"""
packed_keys = map(self._pack_key, keys)
cp = self._column_parent(super_column)
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, column_count, super_column)
consistency = read_consistency_level or self.read_consistency_level
buffer_size = buffer_size or self.buffer_size
offset = 0
keymap = {}
while offset < len(packed_keys):
new_keymap = self.pool.execute('multiget_slice',
packed_keys[offset:offset + buffer_size], cp, sp, consistency)
keymap.update(new_keymap)
offset += buffer_size
ret = self.dict_class()
# Keep the order of keys
for key in keys:
ret[key] = None
empty_keys = []
for packed_key, columns in keymap.iteritems():
unpacked_key = self._unpack_key(packed_key)
if len(columns) > 0:
ret[unpacked_key] = self._cosc_to_dict(columns, include_timestamp)
else:
empty_keys.append(unpacked_key)
for key in empty_keys:
try:
del ret[key]
except KeyError:
pass
return ret
MAX_COUNT = 2 ** 31 - 1
def get_count(self, key, super_column=None, read_consistency_level=None,
columns=None, column_start="", column_finish="",
column_reversed=False, max_count=None):
"""
Count the number of columns in the row with key `key`.
You may limit the columns or super columns counted to those in `columns`.
Additionally, you may limit the columns or super columns counted to
only those between `column_start` and `column_finish`.
You may also count only the number of subcolumns in a single super column
using `super_column`. If this is set, `columns`, `column_start`, and
`column_finish` only apply to the subcolumns of `super_column`.
To put an upper bound on the number of columns that are counted,
set `max_count`.
"""
if max_count is None:
max_count = self.MAX_COUNT
packed_key = self._pack_key(key)
cp = self._column_parent(super_column)
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, max_count, super_column)
return self.pool.execute('get_count', packed_key, cp, sp,
read_consistency_level or self.read_consistency_level)
def multiget_count(self, keys, super_column=None,
read_consistency_level=None,
columns=None, column_start="",
column_finish="", buffer_size=None,
column_reversed=False, max_count=None):
"""
Perform a column count in parallel on a set of rows.
The parameters are the same as for :meth:`multiget()`, except that a list
of keys may be used. A dictionary of the form ``{key: int}`` is
returned.
`buffer_size` is the number of rows from the total list to count at a time.
If left as ``None``, the ColumnFamily's :attr:`buffer_size` will be used.
To put an upper bound on the number of columns that are counted,
set `max_count`.
"""
if max_count is None:
max_count = self.MAX_COUNT
packed_keys = map(self._pack_key, keys)
cp = self._column_parent(super_column)
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, max_count, super_column)
consistency = read_consistency_level or self.read_consistency_level
buffer_size = buffer_size or self.buffer_size
offset = 0
keymap = {}
while offset < len(packed_keys):
new_keymap = self.pool.execute('multiget_count',
packed_keys[offset:offset + buffer_size], cp, sp, consistency)
keymap.update(new_keymap)
offset += buffer_size
ret = self.dict_class()
# Keep the order of keys
for key in keys:
ret[key] = None
for packed_key, count in keymap.iteritems():
ret[self._unpack_key(packed_key)] = count
return ret
def get_range(self, start="", finish="", columns=None, column_start="",
column_finish="", column_reversed=False, column_count=100,
row_count=None, include_timestamp=False,
super_column=None, read_consistency_level=None,
buffer_size=None, filter_empty=True):
"""
Get an iterator over rows in a specified key range.
The key range begins with `start` and ends with `finish`. If left
as empty strings, these extend to the beginning and end, respectively.
Note that if RandomPartitioner is used, rows are stored in the
order of the MD5 hash of their keys, so getting a lexicographical range
of keys is not feasible.
The `row_count` parameter limits the total number of rows that may be
returned. If left as ``None``, the number of rows that may be returned
is unlimted (this is the default).
When calling `get_range()`, the intermediate results need to be
buffered if we are fetching many rows, otherwise the Cassandra
server will overallocate memory and fail. `buffer_size` is the
size of that buffer in number of rows. If left as ``None``, the
ColumnFamily's :attr:`buffer_size` attribute will be used.
When `filter_empty` is left as ``True``, empty rows (including
`range ghosts <http://wiki.apache.org/cassandra/FAQ#range_ghosts>`_)
will be skipped and will not count towards `row_count`.
All other parameters are the same as those of :meth:`get()`.
A generator over ``(key, {column_name: column_value})`` is returned.
To convert this to a list, use ``list()`` on the result.
"""
cl = read_consistency_level or self.read_consistency_level
cp = self._column_parent(super_column)
sp = self._slice_predicate(columns, column_start, column_finish,
column_reversed, column_count, super_column)
count = 0
i = 0
last_key = self._pack_key(start)
finish = self._pack_key(finish)
if buffer_size is None:
buffer_size = self.buffer_size
while True:
if row_count is not None:
if i == 0 and row_count <= buffer_size:
# We don't need to chunk, grab exactly the number of rows
buffer_size = row_count
else:
buffer_size = min(row_count - count + 1, buffer_size)
key_range = KeyRange(start_key=last_key, end_key=finish, count=buffer_size)
key_slices = self.pool.execute('get_range_slices', cp, sp, key_range, cl)
# This may happen if nothing was ever inserted
if key_slices is None:
return
for j, key_slice in enumerate(key_slices):
# Ignore the first element after the first iteration
# because it will be a duplicate.
if j == 0 and i != 0:
continue
if filter_empty and not key_slice.columns:
continue
yield (self._unpack_key(key_slice.key),
self._cosc_to_dict(key_slice.columns, include_timestamp))
count += 1
if row_count is not None and count >= row_count:
return
if len(key_slices) != buffer_size:
return
last_key = key_slices[-1].key
i += 1
def insert(self, key, columns, timestamp=None, ttl=None,
write_consistency_level=None):
"""
Insert or update columns in the row with key `key`.
`columns` should be a dictionary of columns or super columns to insert
or update. If this is a standard column family, `columns` should
look like ``{column_name: column_value}``. If this is a super
column family, `columns` should look like
``{super_column_name: {sub_column_name: value}}``. If this is a
counter column family, you may use integers as values and those will
be used as counter adjustments.
A timestamp may be supplied for all inserted columns with `timestamp`.
`ttl` sets the "time to live" in number of seconds for the inserted
columns. After this many seconds, Cassandra will mark the columns as
deleted.
The timestamp Cassandra reports as being used for insert is returned.
"""
if timestamp is None:
timestamp = self.timestamp()
packed_key = self._pack_key(key)
mut_list = self._make_mutation_list(columns, timestamp, ttl)
mutations = {packed_key: {self.column_family: mut_list}}
self.pool.execute('batch_mutate', mutations,
write_consistency_level or self.write_consistency_level,
allow_retries=self._allow_retries)
return timestamp
def batch_insert(self, rows, timestamp=None, ttl=None, write_consistency_level=None):
"""
Like :meth:`insert()`, but multiple rows may be inserted at once.
The `rows` parameter should be of the form ``{key: {column_name: column_value}}``
if this is a standard column family or
``{key: {super_column_name: {column_name: column_value}}}`` if this is a super
column family.
"""
if timestamp == None:
timestamp = self.timestamp()
cf = self.column_family
mutations = {}
for key, columns in rows.iteritems():
packed_key = self._pack_key(key)
mut_list = self._make_mutation_list(columns, timestamp, ttl)
mutations[packed_key] = {cf: mut_list}
if mutations:
self.pool.execute('batch_mutate', mutations,
write_consistency_level or self.write_consistency_level,
allow_retries=self._allow_retries)
return timestamp
def add(self, key, column, value=1, super_column=None, write_consistency_level=None):
"""
Increment or decrement a counter.
`value` should be an integer, either positive or negative, to be added
to a counter column. By default, `value` is 1.
.. versionadded:: 1.1.0
Available in Cassandra 0.8.0 and later.
"""
packed_key = self._pack_key(key)
cp = self._column_parent(super_column)
column = self._pack_name(column)
self.pool.execute('add', packed_key, cp, CounterColumn(column, value),
write_consistency_level or self.write_consistency_level,
allow_retries=self._allow_retries)
def remove(self, key, columns=None, super_column=None,
write_consistency_level=None, timestamp=None, counter=None):
"""
Remove a specified row or a set of columns within the row with key `key`.
A set of columns or super columns to delete may be specified using
`columns`.
A single super column may be deleted by setting `super_column`. If
`super_column` is specified, `columns` will apply to the subcolumns
of `super_column`.
If `columns` and `super_column` are both ``None``, the entire row is
removed.
The timestamp used for the mutation is returned.
"""
if timestamp is None:
timestamp = self.timestamp()
batch = self.batch(write_consistency_level=write_consistency_level)
batch.remove(key, columns, super_column, timestamp)
batch.send()
return timestamp
def remove_counter(self, key, column, super_column=None, write_consistency_level=None):
"""
Remove a counter at the specified location.
Note that counters have limited support for deletes: if you remove a
counter, you must wait to issue any following update until the delete
has reached all the nodes and all of them have been fully compacted.
.. versionadded:: 1.1.0
Available in Cassandra 0.8.0 and later.
"""
packed_key = self._pack_key(key)
cp = self._column_path(super_column, column)
self.pool.execute('remove_counter', packed_key, cp,
write_consistency_level or self.write_consistency_level)
def batch(self, queue_size=100, write_consistency_level=None):
"""
Create batch mutator for doing multiple insert, update, and remove
operations using as few roundtrips as possible.
The `queue_size` parameter sets the max number of mutations per request.
A :class:`~pycassa.batch.CfMutator` is returned.
"""
return CfMutator(self, queue_size,
write_consistency_level or self.write_consistency_level,
allow_retries=self._allow_retries)
def truncate(self):
"""
Marks the entire ColumnFamily as deleted.
From the user's perspective, a successful call to ``truncate`` will
result complete data deletion from this column family. Internally,
however, disk space will not be immediatily released, as with all
deletes in Cassandra, this one only marks the data as deleted.
The operation succeeds only if all hosts in the cluster at available
and will throw an :exc:`.UnavailableException` if some hosts are
down.
"""
self.pool.execute('truncate', self.column_family)
PooledColumnFamily = ColumnFamily
Jump to Line
Something went wrong with that request. Please try again.