You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 

995 lines
36 KiB

#!/usr/bin/env python
# -*- coding: utf-8 -*-
#
# Copyright 2011-2013 Codernity (http://codernity.com)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from CodernityDB3.index import (Index,
IndexException,
DocIdNotFound,
ElemNotFound,
TryReindexException,
IndexPreconditionsException)
import os
import marshal
import io
import struct
import shutil
from CodernityDB3.storage import IU_Storage, DummyStorage
from CodernityDB3.env import cdb_environment
if cdb_environment.get('rlock_obj'):
from CodernityDB3 import patch
patch.patch_cache_rr(cdb_environment['rlock_obj'])
from CodernityDB3.rr_cache import cache1lvl
from CodernityDB3.misc import random_hex_32
try:
from CodernityDB3 import __version__
except ImportError:
from .__init__ import __version__
class IU_HashIndex(Index):
"""
That class is for Internal Use only, if you want to use HashIndex just subclass the :py:class:`HashIndex` instead this one.
That design is because main index logic should be always in database not in custom user indexes.
"""
def __init__(self, db_path, name, entry_line_format='<32s{key}IIcI', hash_lim=0xfffff, storage_class=None, key_format='c'):
"""
The index is capable to solve conflicts by `Separate chaining`
:param db_path: database path
:type db_path: string
:param name: index name
:type name: ascii string
:param line_format: line format, `key_format` parameter value will replace `{key}` if present.
:type line_format: string (32s{key}IIcI by default) {doc_id}{hash_key}{start}{size}{status}{next}
:param hash_lim: maximum hash functon results (remember about birthday problem) count from 0
:type hash_lim: integer
:param storage_class: Storage class by default it will open standard :py:class:`CodernityDB3.storage.Storage` (if string has to be accesible by globals()[storage_class])
:type storage_class: class name which will be instance of CodernityDB3.storage.Storage instance or None
:param key_format: a index key format
"""
# Fix types
if isinstance(db_path, str):
db_path = db_path.encode()
if isinstance(name, str):
name = name.encode()
if key_format and '{key}' in entry_line_format:
entry_line_format = entry_line_format.replace('{key}', key_format)
super(IU_HashIndex, self).__init__(db_path, name)
self.hash_lim = hash_lim
if not storage_class:
storage_class = IU_Storage
if storage_class and not isinstance(storage_class, str):
storage_class = storage_class.__name__
self.storage_class = storage_class
self.storage = None
self.bucket_line_format = "<I"
self.bucket_line_size = struct.calcsize(self.bucket_line_format)
self.entry_line_format = entry_line_format
self.entry_line_size = struct.calcsize(self.entry_line_format)
cache = cache1lvl(100)
self._find_key = cache(self._find_key)
self._locate_doc_id = cache(self._locate_doc_id)
self.bucket_struct = struct.Struct(self.bucket_line_format)
self.entry_struct = struct.Struct(self.entry_line_format)
self.data_start = (
self.hash_lim + 1) * self.bucket_line_size + self._start_ind + 2
def _fix_params(self):
super(IU_HashIndex, self)._fix_params()
self.bucket_line_size = struct.calcsize(self.bucket_line_format)
self.entry_line_size = struct.calcsize(self.entry_line_format)
self.bucket_struct = struct.Struct(self.bucket_line_format)
self.entry_struct = struct.Struct(self.entry_line_format)
self.data_start = (
self.hash_lim + 1) * self.bucket_line_size + self._start_ind + 2
def open_index(self):
if not os.path.isfile(os.path.join(self.db_path, self.name + '_buck')):
raise IndexException("Doesn't exists")
self.buckets = io.open(
os.path.join(self.db_path, self.name + "_buck"), 'r+b', buffering=0)
self._fix_params()
self._open_storage()
def create_index(self):
if os.path.isfile(os.path.join(self.db_path, self.name + '_buck')):
raise IndexException('Already exists')
with io.open(os.path.join(self.db_path, self.name + "_buck"), 'w+b') as f:
props = dict(name=self.name,
bucket_line_format=self.bucket_line_format,
entry_line_format=self.entry_line_format,
hash_lim=self.hash_lim,
version=self.__version__,
storage_class=self.storage_class)
f.write(marshal.dumps(props))
self.buckets = io.open(
os.path.join(self.db_path, self.name + "_buck"), 'r+b', buffering=0)
self._create_storage()
def destroy(self):
super(IU_HashIndex, self).destroy()
self._clear_cache()
def _open_storage(self):
s = globals()[self.storage_class]
if not self.storage:
self.storage = s(self.db_path, self.name)
self.storage.open()
def _create_storage(self):
s = globals()[self.storage_class]
if not self.storage:
self.storage = s(self.db_path, self.name)
self.storage.create()
# def close_index(self):
# self.buckets.flush()
# self.buckets.close()
# self.storage.close()
# @lfu_cache(100)
def _find_key(self, key):
"""
Find the key position
:param key: the key to find
"""
# Fix types
if isinstance(key, str):
key = key.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
if not location:
return None, None, 0, 0, 'u'
found_at, doc_id, l_key, start, size, status, _next = self._locate_key(
key, location)
if status == 'd': # when first record from many is deleted
while True:
found_at, doc_id, l_key, start, size, status, _next = self._locate_key(
key, _next)
if status != 'd':
break
return doc_id, l_key, start, size, status
else:
return None, None, 0, 0, 'u'
def _find_key_many(self, key, limit=1, offset=0):
# Fix types
if isinstance(key, str):
key = key.encode()
location = None
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
while offset:
if not location:
break
try:
found_at, doc_id, l_key, start, size, status, _next = self._locate_key(
key, location)
except IndexException:
break
else:
if status != 'd':
if l_key == key: # in case of hash function conflicts
offset -= 1
location = _next
while limit:
if not location:
break
try:
found_at, doc_id, l_key, start, size, status, _next = self._locate_key(
key, location)
except IndexException:
break
else:
if status != 'd':
if l_key == key: # in case of hash function conflicts
yield doc_id, start, size, status
limit -= 1
location = _next
def _calculate_position(self, key):
# Fix types
if isinstance(key, str):
key = key.encode()
return abs(hash(key) & self.hash_lim) * self.bucket_line_size + self._start_ind
# TODO add cache!
def _locate_key(self, key, start):
"""
Locate position of the key, it will iterate using `next` field in record
until required key will be find.
:param key: the key to locate
:param start: position to start from
"""
# Fix types
if isinstance(key, str):
key = key.encode()
location = start
while True:
self.buckets.seek(location)
data = self.buckets.read(self.entry_line_size)
# todo, maybe partial read there...
try:
doc_id, l_key, start, size, status, _next = self.entry_struct.unpack(data)
except struct.error:
raise ElemNotFound(
"Not found") # not found but might be also broken
if l_key == key:
break
else:
if not _next:
# not found
raise ElemNotFound("Not found")
else:
location = _next # go to next record
return location, doc_id, l_key, start, size, status, _next
# @lfu_cache(100)
def _locate_doc_id(self, doc_id, key, start):
"""
Locate position of the doc_id, it will iterate using `next` field in record
until required key will be find.
:param doc_id: the doc_id to locate
:param key: key value
:param start: position to start from
"""
# Fix types
if isinstance(doc_id, str):
doc_id = doc_id.encode()
if isinstance(key, str):
key = key.encode()
location = start
while True:
self.buckets.seek(location)
data = self.buckets.read(self.entry_line_size)
try:
l_doc_id, l_key, start, size, status, _next = self.entry_struct.unpack(data)
except:
raise DocIdNotFound(
"Doc_id '%s' for '%s' not found" % (doc_id, key))
if l_doc_id == doc_id and l_key == key: # added for consistency
break
else:
if not _next:
# not found
raise DocIdNotFound(
"Doc_id '%s' for '%s' not found" % (doc_id, key))
else:
location = _next # go to next record
return location, doc_id, l_key, start, size, status, _next
def _find_place(self, start):
"""
Find a place to where put the key. It will iterate using `next` field in record, until
empty `next` found
:param start: position to start from
"""
location = start
while True:
self.buckets.seek(location)
data = self.buckets.read(self.entry_line_size)
# todo, maybe partial read there...
doc_id, l_key, start, size, status, _next = self.entry_struct.unpack(data)
if not _next or status == 'd':
return self.buckets.tell() - self.entry_line_size, doc_id, l_key, start, size, status, _next
else:
location = _next # go to next record
def update(self, doc_id, key, u_start=0, u_size=0, u_status='o'):
# Fix types
if isinstance(doc_id, str):
doc_id = doc_id.encode()
if isinstance(key, str):
key = key.encode()
if isinstance(u_status, str):
u_status = u_status.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
# test if it's unique or not really unique hash
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
else:
raise ElemNotFound("Location '%s' not found" % doc_id)
found_at, _doc_id, _key, start, size, status, _next = self._locate_doc_id(doc_id, key, location)
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(doc_id,
key,
u_start,
u_size,
u_status,
_next))
self.flush()
self._find_key.delete(key)
self._locate_doc_id.delete(doc_id)
return True
def insert(self, doc_id, key, start, size, status='o'):
# Fix types
if isinstance(doc_id, str):
doc_id = doc_id.encode()
if isinstance(key, str):
key = key.encode()
if isinstance(status, str):
status = status.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
# conflict occurs?
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
else:
location = 0
if location:
# last key with that hash
try:
found_at, _doc_id, _key, _start, _size, _status, _next = self._locate_doc_id(doc_id, key, location)
except DocIdNotFound:
found_at, _doc_id, _key, _start, _size, _status, _next = self._find_place(location)
self.buckets.seek(0, 2)
wrote_at = self.buckets.tell()
self.buckets.write(self.entry_struct.pack(doc_id,
key,
start,
size,
status,
_next))
# self.flush()
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(_doc_id,
_key,
_start,
_size,
_status,
wrote_at))
else:
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(doc_id,
key,
start,
size,
status,
_next))
self.flush()
self._locate_doc_id.delete(doc_id)
self._find_key.delete(_key)
# self._find_key.delete(key)
# self._locate_key.delete(_key)
return True
# raise NotImplementedError
else:
self.buckets.seek(0, 2)
wrote_at = self.buckets.tell()
# check if position is bigger than all hash entries...
if wrote_at < self.data_start:
self.buckets.seek(self.data_start)
wrote_at = self.buckets.tell()
self.buckets.write(self.entry_struct.pack(doc_id,
key,
start,
size,
status,
0))
# self.flush()
self._find_key.delete(key)
self.buckets.seek(start_position)
self.buckets.write(self.bucket_struct.pack(wrote_at))
self.flush()
return True
def get(self, key):
# Fix types
if isinstance(key, str):
key = key.encode()
return self._find_key(self.make_key(key))
def get_many(self, key, limit=1, offset=0):
return self._find_key_many(self.make_key(key), limit, offset)
def all(self, limit=-1, offset=0):
self.buckets.seek(self.data_start)
while offset:
curr_data = self.buckets.read(self.entry_line_size)
if not curr_data:
break
try:
doc_id, key, start, size, status, _next = self.entry_struct.unpack(curr_data)
except IndexException:
break
else:
if status != 'd':
offset -= 1
while limit:
curr_data = self.buckets.read(self.entry_line_size)
if not curr_data:
break
try:
doc_id, key, start, size, status, _next = self.entry_struct.unpack(curr_data)
except IndexException:
break
else:
if status != 'd':
yield doc_id, key, start, size, status
limit -= 1
def _fix_link(self, key, pos_prev, pos_next):
# Fix types
if isinstance(key, str):
key = key.encode()
# CHECKIT why I need that hack
if pos_prev >= self.data_start:
self.buckets.seek(pos_prev)
data = self.buckets.read(self.entry_line_size)
if data:
doc_id, l_key, start, size, status, _next = self.entry_struct.unpack(data)
self.buckets.seek(pos_prev)
self.buckets.write(self.entry_struct.pack(doc_id,
l_key,
start,
size,
status,
pos_next))
self.flush()
if pos_next:
self.buckets.seek(pos_next)
data = self.buckets.read(self.entry_line_size)
if data:
doc_id, l_key, start, size, status, _next = self.entry_struct.unpack(data)
self.buckets.seek(pos_next)
self.buckets.write(self.entry_struct.pack(doc_id,
l_key,
start,
size,
status,
_next))
self.flush()
return
def delete(self, doc_id, key, start=0, size=0):
# Fix types
if isinstance(doc_id, str):
doc_id = doc_id.encode()
if isinstance(key, str):
key = key.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
else:
# case happens when trying to delete element with new index key in data
# after adding new index to database without reindex
raise TryReindexException()
found_at, _doc_id, _key, start, size, status, _next = self._locate_doc_id(doc_id, key, location)
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(doc_id,
key,
start,
size,
'd',
_next))
self.flush()
# self._fix_link(_key, _prev, _next)
self._find_key.delete(key)
self._locate_doc_id.delete(doc_id)
return True
def compact(self, hash_lim=None):
if not hash_lim:
hash_lim = self.hash_lim
compact_ind = self.__class__(
self.db_path, self.name + '_compact', hash_lim=hash_lim)
compact_ind.create_index()
gen = self.all()
while True:
try:
doc_id, key, start, size, status = next(gen)
except StopIteration:
break
self.storage._f.seek(start)
value = self.storage._f.read(size)
start_ = compact_ind.storage._f.tell()
compact_ind.storage._f.write(value)
compact_ind.insert(doc_id, key, start_, size, status)
compact_ind.close_index()
original_name = self.name
# os.unlink(os.path.join(self.db_path, self.name + "_buck"))
self.close_index()
shutil.move(os.path.join(compact_ind.db_path, compact_ind.
name + "_buck"), os.path.join(self.db_path, self.name + "_buck"))
shutil.move(os.path.join(compact_ind.db_path, compact_ind.
name + "_stor"), os.path.join(self.db_path, self.name + "_stor"))
# self.name = original_name
self.open_index() # reload...
self.name = original_name.decode()
self._save_params(dict(name=original_name))
self._fix_params()
self._clear_cache()
return True
def make_key(self, key):
return key
def make_key_value(self, data):
return '1', data
def _clear_cache(self):
self._find_key.clear()
self._locate_doc_id.clear()
def close_index(self):
super(IU_HashIndex, self).close_index()
self._clear_cache()
class IU_UniqueHashIndex(IU_HashIndex):
"""
Index for *unique* keys! Designed to be a **id** index.
That class is for Internal Use only, if you want to use UniqueHashIndex just subclass the :py:class:`UniqueHashIndex` instead this one.
That design is because main index logic should be always in database not in custom user indexes.
"""
def __init__(self, db_path, name, entry_line_format="<32s8sIIcI", *args, **kwargs):
# Fix types
if isinstance(db_path, str):
db_path = db_path.encode()
if isinstance(name, str):
name = name.encode()
if 'key' in kwargs:
raise IndexPreconditionsException(
"UniqueHashIndex doesn't accept key parameter'")
super(IU_UniqueHashIndex, self).__init__(db_path, name,
entry_line_format, *args, **kwargs)
self.create_key = random_hex_32 # : set the function to create random key when no _id given
# self.entry_struct=struct.Struct(entry_line_format)
# @lfu_cache(100)
def _find_key(self, key):
"""
Find the key position
:param key: the key to find
"""
# Fix types
if isinstance(key, str):
key = key.encode()
start_position = self._calculate_position(key)
self.seek = self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
found_at, l_key, rev, start, size, status, _next = self._locate_key(
key, location)
# Fix types
if isinstance(l_key, bytes):
l_key = l_key.decode()
if isinstance(rev, bytes):
rev = rev.decode()
if isinstance(status, bytes):
status = status.decode()
return l_key, rev, start, size, status
else:
return None, None, 0, 0, 'u'
def _find_key_many(self, *args, **kwargs):
raise NotImplementedError()
def _find_place(self, start, key):
"""
Find a place to where put the key. It will iterate using `next` field in record, until
empty `next` found
:param start: position to start from
"""
# Fix types
if isinstance(key, str):
key = key.encode()
location = start
while True:
self.buckets.seek(location)
data = self.buckets.read(self.entry_line_size)
# todo, maybe partial read there...
l_key, rev, start, size, status, _next = self.entry_struct.unpack(
data)
if l_key == key:
raise IndexException("The '%s' key already exists" % key)
if not _next or status == 'd':
return self.buckets.tell() - self.entry_line_size, l_key, rev, start, size, status, _next
else:
location = _next # go to next record
# @lfu_cache(100)
def _locate_key(self, key, start):
"""
Locate position of the key, it will iterate using `next` field in record
until required key will be find.
:param key: the key to locate
:param start: position to start from
"""
# Fix types
if isinstance(key, str):
key = key.encode()
location = start
while True:
self.buckets.seek(location)
data = self.buckets.read(self.entry_line_size)
# todo, maybe partial read there...
try:
l_key, rev, start, size, status, _next = self.entry_struct.unpack(data)
except struct.error:
raise ElemNotFound("Location '%s' not found" % key)
if l_key == key:
break
else:
if not _next:
# not found
raise ElemNotFound("Location '%s' not found" % key)
else:
location = _next # go to next record
return self.buckets.tell() - self.entry_line_size, l_key, rev, start, size, status, _next
def update(self, key, rev, u_start=0, u_size=0, u_status='o'):
# Fix types
if isinstance(key, str):
key = key.encode()
if isinstance(rev, str):
rev = rev.encode()
if isinstance(u_status, str):
u_status = u_status.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
# test if it's unique or not really unique hash
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
else:
raise ElemNotFound("Location '%s' not found" % key)
found_at, _key, _rev, start, size, status, _next = self._locate_key(
key, location)
if u_start == 0:
u_start = start
if u_size == 0:
u_size = size
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(key,
rev,
u_start,
u_size,
u_status,
_next))
self.flush()
self._find_key.delete(key)
return True
def insert(self, key, rev, start, size, status='o'):
# Fix types
if isinstance(key, str):
key = key.encode()
if isinstance(rev, str):
rev = rev.encode()
if isinstance(status, str):
status = status.encode()
start_position = self._calculate_position(key)
self.buckets.seek(start_position)
curr_data = self.buckets.read(self.bucket_line_size)
# conflict occurs?
if curr_data:
location = self.bucket_struct.unpack(curr_data)[0]
else:
location = 0
if location:
# last key with that hash
found_at, _key, _rev, _start, _size, _status, _next = self._find_place(
location, key)
self.buckets.seek(0, 2)
wrote_at = self.buckets.tell()
# check if position is bigger than all hash entries...
if wrote_at < self.data_start:
self.buckets.seek(self.data_start)
wrote_at = self.buckets.tell()
self.buckets.write(self.entry_struct.pack(key,
rev,
start,
size,
status,
_next))
# self.flush()
self.buckets.seek(found_at)
self.buckets.write(self.entry_struct.pack(_key,
_rev,
_start,
_size,
_status,
wrote_at))
self.flush()
self._find_key.delete(_key)
# self._locate_key.delete(_key)
return True
# raise NotImplementedError
else:
self.buckets.seek(0, 2)
wrote_at = self.buckets.tell()
# check if position is bigger than all hash entries...
if wrote_at < self.data_start:
self.buckets.seek(self.data_start)
wrote_at = self.buckets.tell()
self.buckets.write(self.entry_struct.pack(key,
rev,
start,
size,
status,
0))
self.buckets.seek(start_position)
self.buckets.write(self.bucket_struct.pack(wrote_at))
self.flush()
self._find_key.delete(key)
return True
def all(self, limit=-1, offset=0):
self.buckets.seek(self.data_start)
while offset:
curr_data = self.buckets.read(self.entry_line_size)
if not curr_data:
break
try:
doc_id, rev, start, size, status, next = self.entry_struct.unpack(curr_data)
except IndexException:
break
else:
if status != 'd':
offset -= 1
while limit:
curr_data = self.buckets.read(self.entry_line_size)
if not curr_data:
break
try:
doc_id, rev, start, size, status, next = self.entry_struct.unpack(curr_data)
except IndexException:
break
else:
if status != 'd':
yield doc_id, rev, start, size, status
limit -= 1
def get_many(self, *args, **kwargs):
raise NotImplementedError()
def delete(self, key, start=0, size=0):
# Fix types
if isinstance(key, str):
key = key.encode()
self.update(key, '00000000', start, size, 'd')
def make_key_value(self, data):
_id = data['_id']
try:
_id = data['_id'].encode()
except:
raise IndexPreconditionsException(
"_id must be valid string/bytes object")
if len(_id) != 32:
raise IndexPreconditionsException("Invalid _id lenght")
del data['_id']
del data['_rev']
return _id, data
def destroy(self):
Index.destroy(self)
self._clear_cache()
def _clear_cache(self):
self._find_key.clear()
def insert_with_storage(self, _id, _rev, value):
# Fix types
if isinstance(_id, str):
_id = _id.encode()
if isinstance(_rev, str):
_rev = _rev.encode()
if value:
start, size = self.storage.insert(value)
else:
start = 1
size = 0
return self.insert(_id, _rev, start, size)
def update_with_storage(self, _id, _rev, value):
# Fix types
if isinstance(_id, str):
_id = _id.encode()
if isinstance(_rev, str):
_rev = _rev.encode()
if value:
start, size = self.storage.insert(value)
else:
start = 1
size = 0
return self.update(_id, _rev, start, size)
class DummyHashIndex(IU_HashIndex):
def __init__(self, db_path, name, entry_line_format="<32s4sIIcI", *args, **kwargs):
super(DummyHashIndex, self).__init__(db_path, name,
entry_line_format, *args, **kwargs)
self.create_key = random_hex_32 # : set the function to create random key when no _id given
# self.entry_struct=struct.Struct(entry_line_format)
def update(self, *args, **kwargs):
return True
def insert(self, *args, **kwargs):
return True
def all(self, *args, **kwargs):
raise StopIteration
def get(self, *args, **kwargs):
raise ElemNotFound
def get_many(self, *args, **kwargs):
raise StopIteration
def delete(self, *args, **kwargs):
pass
def make_key_value(self, data):
return '1', {'_': 1}
def destroy(self):
pass
def _clear_cache(self):
pass
def _open_storage(self):
if not self.storage:
self.storage = DummyStorage()
self.storage.open()
def _create_storage(self):
if not self.storage:
self.storage = DummyStorage()
self.storage.create()
class IU_MultiHashIndex(IU_HashIndex):
"""
Class that allows to index more than one key per database record.
It operates very well on GET/INSERT. It's not optimized for
UPDATE operations (will always readd everything)
"""
def __init__(self, *args, **kwargs):
super(IU_MultiHashIndex, self).__init__(*args, **kwargs)
def insert(self, doc_id, key, start, size, status='o'):
if isinstance(key, (list, tuple)):
key = set(key)
elif not isinstance(key, set):
key = set([key])
ins = super(IU_MultiHashIndex, self).insert
for curr_key in key:
ins(doc_id, curr_key, start, size, status)
return True
def update(self, doc_id, key, u_start, u_size, u_status='o'):
if isinstance(key, (list, tuple)):
key = set(key)
elif not isinstance(key, set):
key = set([key])
upd = super(IU_MultiHashIndex, self).update
for curr_key in key:
upd(doc_id, curr_key, u_start, u_size, u_status)
def delete(self, doc_id, key, start=0, size=0):
if isinstance(key, (list, tuple)):
key = set(key)
elif not isinstance(key, set):
key = set([key])
delete = super(IU_MultiHashIndex, self).delete
for curr_key in key:
delete(doc_id, curr_key, start, size)
def get(self, key):
return super(IU_MultiHashIndex, self).get(key)
def make_key_value(self, data):
raise NotImplementedError()
# classes for public use, done in this way because of
# generation static files with indexes (_index directory)
class HashIndex(IU_HashIndex):
"""
That class is designed to be used in custom indexes.
"""
pass
class UniqueHashIndex(IU_UniqueHashIndex):
"""
That class is designed to be used in custom indexes. It's designed to be **id** index.
"""
pass
class MultiHashIndex(IU_MultiHashIndex):
"""
That class is designed to be used in custom indexes.
"""