Skip to content

Instantly share code, notes, and snippets.

@rafacv
Forked from mmalone/fakecassandra.py
Created April 23, 2010 19:17
Show Gist options
  • Save rafacv/377023 to your computer and use it in GitHub Desktop.
Save rafacv/377023 to your computer and use it in GitHub Desktop.
In-memory Cassandra-ish thingy
# In-memory Cassandra-ish thingy... useful for unit tests. Maybe useful for other
# stuff too? No support for SuperColumns, but that should be easy enough to add.
import bisect
import copy
from cassandra.ttypes import NotFoundException, Column, ColumnPath, ColumnOrSuperColumn
class SSTable(object):
def __init__(self, data=None):
if data is None:
data = {}
self.data = data
self.sorted_keys = sorted(self.data.iterkeys())
def __setitem__(self, key, value):
if key not in self.data:
bisect.insort(self.sorted_keys, key)
self.data[key] = value
def __delitem__(self, key):
del self.data[key]
del self.sorted_keys[bisect.bisect_left(self.sorted_keys, key)]
def __getitem__(self, key):
return self.data[key]
def __contains__(self, key):
return self.data.__contains__(key)
def __len__(self):
return len(self.sorted_keys)
def get(self, *args, **kwargs):
return self.data.get(*args, **kwargs)
def clear(self):
self.data.clear()
del self.sorted_keys[:]
def copy(self):
return self.__class__(self.data.copy())
def key_slice(self, start, finish, count=None, reversed=False):
start = bisect.bisect_left(self.sorted_keys, start) if start else 0
finish = bisect.bisect_left(self.sorted_keys, finish) if finish else len(self.sorted_keys)
keys = self.sorted_keys[start:finish]
if reversed:
keys.reverse()
if count is not None:
keys = keys[:count]
return keys
def slice(self, start, finish, count=None, reversed=False):
return (self[key] for key in self.key_slice(start, finish, count, reversed))
class Cassandra(object):
def __init__(self, keyspaces=None):
self.database = {}
self.keyspaces = {}
if keyspaces is not None:
for keyspace, column_families in keyspaces.iteritems():
self.add_keyspace(keyspace, column_families)
def add_keyspace(self, keyspace, column_families):
self.database[keyspace] = SSTable()
self.keyspaces[keyspace] = column_families
def _slice_range(self, row, predicate):
if predicate.slice_range:
slice_range = predicate.slice_range
columns = list(row.slice(
start=slice_range.start,
finish=slice_range.finish,
count=slice_range.count,
reversed=slice_range.reversed
))
else:
columns = [row[column_name] for column_name in predicate.column_names if column_name in row]
return [ColumnOrSuperColumn(column=copy.deepcopy(column)) for column in columns]
def _get(self, keyspace, key, column_path, consistency_level):
key = self.database[keyspace].get(key)
if key is not None:
column_family = key[column_path.column_family] # Should raise misconfiguration exception here if KeyError.
try:
return column_family[column_path.column]
except KeyError:
pass
raise NotFoundException()
def get(self, keyspace, key, column_path, consistency_level):
return copy.deepcopy(self._get(keyspace, key, column_path, consistency_level))
def get_slice(self, keyspace, key, column_parent, predicate, consistency_level):
key = self.database[keyspace].get(key)
return self._slice_range(key[column_parent.column_family], predicate) if key is not None else []
def multiget_slice(self, keyspace, keys, column_parent, predicate, consistency_level):
return dict((key, self.get_slice(keyspace, key, column_parent, predicate, consistency_level)) for key in keys)
def get_range_slice(self, keyspace, column_parent, predicate, start_key, finish_key, row_count, consistency_level):
keys = self.database[keyspace].key_slice(start_key, finish_key, count=row_count)
return self.multiget_slice(keyspace, keys, column_parent, predicate, consistency_level)
def get_range_slices(self, keyspace, column_parent, predicate, key_range, consistency_level):
return self.get_range_slice(keyspace, column_parent, predicate, key_range.start_key, key_range.end_key, key_range.count, consistency_level)
def multiget_key_range(self, keyspace, column_family, key_ranges, count, consistency_level):
keys = []
for key_range in key_ranges:
key_slice = self.database[keyspace].key_slice(key_range.start_key, key_range.end_key)
keys.extend(key for key in key_slice if len(self.database[keyspace][key][column_family]))
if len(keys) >= count:
break
return keys[:count]
def insert(self, keyspace, key, column_path, value, timestamp, consistency_level):
try:
column = self._get(keyspace, key, column_path, consistency_level)
if timestamp > column.timestamp:
column.value = value
except NotFoundException:
if key not in self.database[keyspace]:
self.database[keyspace][key] = dict((column_family, SSTable()) for column_family in self.keyspaces[keyspace])
self.database[keyspace][key][column_path.column_family][column_path.column] = Column(column_path.column, value, timestamp)
def remove(self, keyspace, key, column_path, timestamp, consistency_level):
key = self.database[keyspace].get(key)
if key is not None:
column_family = key[column_path.column_family] # Should raise misconfiguration exception here if KeyError.
if column_path.column in column_family and timestamp >= column_family[column_path.column].timestamp:
del column_family[column_path.column]
def batch_insert(self, keyspace, key, cfmap, consistency_level):
for column_family, columns in cfmap.iteritems():
for column in columns:
self.insert(
keyspace,
key,
ColumnPath(column_family=column_family, column=column.column.name),
column.column.value,
column.column.timestamp,
consistency_level
)
def batch_mutate(self, keyspace, mutation_map, consistency_level):
for key, cfmap in mutation_map.iteritems():
for column_family, mutations in cfmap.iteritems():
for mutation in mutations:
if mutation.deletion:
if mutation.deletion.predicate.slice_range:
raise Exception('Slice range on batch mutations not supported yet.')
for column in mutation.deletion.predicate.column_names:
self.remove(
keyspace,
key,
ColumnPath(column_family=column_family, column=column),
mutation.deletion.timestamp,
consistency_level
)
else:
column = mutation.column_or_supercolumn.column
self.insert(
keyspace,
key,
ColumnPath(column_family=column_family, column=column.name),
column.value,
column.timestamp,
consistency_level
)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment