blob: 169d4ebcf547e5a09ba44856c9376d387b8001b1 [file] [log] [blame]
#!/usr/bin/env python
#
# Copyright 2007 Google Inc.
#
# 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.
#
"""Key range representation and splitting."""
import os
try:
import json as simplejson
except ImportError:
try:
from mapreduce.lib import simplejson
except ImportError:
simplejson = None
from google.appengine.api import datastore
from google.appengine.api import namespace_manager
from google.appengine.datastore import datastore_pb
from google.appengine.ext import db
class Error(Exception):
"""Base class for exceptions in this module."""
class KeyRangeError(Error):
"""Error while trying to generate a KeyRange."""
class SimplejsonUnavailableError(Error):
"""Error using json functionality with unavailable json and simplejson."""
class KeyRange(object):
"""Represents a range of keys in the datastore.
A KeyRange object represents a key range
(key_start, include_start, key_end, include_end)
and a scan direction (KeyRange.DESC or KeyRange.ASC).
"""
DESC = "DESC"
ASC = "ASC"
def __init__(self,
key_start=None,
key_end=None,
direction=None,
include_start=True,
include_end=True,
namespace=None,
_app=None):
"""Initialize a KeyRange object.
Args:
key_start: The starting key for this range.
key_end: The ending key for this range.
direction: The direction of the query for this range.
include_start: Whether the start key should be included in the range.
include_end: Whether the end key should be included in the range.
namespace: The namespace for this range. If None then the current
namespace is used.
"""
if direction is None:
direction = KeyRange.ASC
assert direction in (KeyRange.ASC, KeyRange.DESC)
self.direction = direction
self.key_start = key_start
self.key_end = key_end
self.include_start = include_start
self.include_end = include_end
if namespace is not None:
self.namespace = namespace
else:
self.namespace = namespace_manager.get_namespace()
self._app = _app
def __str__(self):
if self.include_start:
left_side = "["
else:
left_side = "("
if self.include_end:
right_side = "]"
else:
right_side = ")"
return "%s%s%r to %r%s" % (self.direction, left_side, self.key_start,
self.key_end, right_side)
def __repr__(self):
return ("key_range.KeyRange(key_start=%r,key_end=%r,direction=%r,"
"include_start=%r,include_end=%r, namespace=%r)") % (
self.key_start,
self.key_end,
self.direction,
self.include_start,
self.include_end,
self.namespace)
def advance(self, key):
"""Updates the start of the range immediately past the specified key.
Args:
key: A db.Key.
"""
self.include_start = False
self.key_start = key
def filter_query(self, query):
"""Add query filter to restrict to this key range.
Args:
query: A db.Query instance.
Returns:
The input query restricted to this key range.
"""
assert isinstance(query, db.Query)
if self.include_start:
start_comparator = ">="
else:
start_comparator = ">"
if self.include_end:
end_comparator = "<="
else:
end_comparator = "<"
if self.key_start:
query.filter("__key__ %s" % start_comparator, self.key_start)
if self.key_end:
query.filter("__key__ %s" % end_comparator, self.key_end)
return query
def filter_datastore_query(self, query):
"""Add query filter to restrict to this key range.
Args:
query: A datastore.Query instance.
Returns:
The input query restricted to this key range.
"""
assert isinstance(query, datastore.Query)
if self.include_start:
start_comparator = ">="
else:
start_comparator = ">"
if self.include_end:
end_comparator = "<="
else:
end_comparator = "<"
if self.key_start:
query.update({"__key__ %s" % start_comparator: self.key_start})
if self.key_end:
query.update({"__key__ %s" % end_comparator: self.key_end})
return query
def __get_direction(self, asc, desc):
"""Check that self.direction is in (KeyRange.ASC, KeyRange.DESC).
Args:
asc: Argument to return if self.direction is KeyRange.ASC
desc: Argument to return if self.direction is KeyRange.DESC
Returns:
asc or desc appropriately
Raises:
KeyRangeError: if self.direction is not in (KeyRange.ASC, KeyRange.DESC).
"""
if self.direction == KeyRange.ASC:
return asc
elif self.direction == KeyRange.DESC:
return desc
else:
raise KeyRangeError("KeyRange direction unexpected: %s", self.direction)
def make_directed_query(self, kind_class, keys_only=False):
"""Construct a query for this key range, including the scan direction.
Args:
kind_class: A kind implementation class.
keys_only: bool, default False, use keys_only on Query?
Returns:
A db.Query instance.
Raises:
KeyRangeError: if self.direction is not in (KeyRange.ASC, KeyRange.DESC).
"""
assert self._app is None, '_app is not supported for db.Query'
direction = self.__get_direction("", "-")
query = db.Query(kind_class, namespace=self.namespace, keys_only=keys_only)
query.order("%s__key__" % direction)
query = self.filter_query(query)
return query
def make_directed_datastore_query(self, kind, keys_only=False):
"""Construct a query for this key range, including the scan direction.
Args:
kind: A string.
keys_only: bool, default False, use keys_only on Query?
Returns:
A datastore.Query instance.
Raises:
KeyRangeError: if self.direction is not in (KeyRange.ASC, KeyRange.DESC).
"""
direction = self.__get_direction(datastore.Query.ASCENDING,
datastore.Query.DESCENDING)
query = datastore.Query(kind, _app=self._app, keys_only=keys_only)
query.Order(("__key__", direction))
query = self.filter_datastore_query(query)
return query
def make_ascending_query(self, kind_class, keys_only=False):
"""Construct a query for this key range without setting the scan direction.
Args:
kind_class: A kind implementation class.
keys_only: bool, default False, query only for keys.
Returns:
A db.Query instance.
"""
assert self._app is None, '_app is not supported for db.Query'
query = db.Query(kind_class, namespace=self.namespace, keys_only=keys_only)
query.order("__key__")
query = self.filter_query(query)
return query
def make_ascending_datastore_query(self, kind, keys_only=False):
"""Construct a query for this key range without setting the scan direction.
Args:
kind: A string.
keys_only: bool, default False, use keys_only on Query?
Returns:
A datastore.Query instance.
"""
query = datastore.Query(kind,
namespace=self.namespace,
_app=self._app,
keys_only=keys_only)
query.Order(("__key__", datastore.Query.ASCENDING))
query = self.filter_datastore_query(query)
return query
def split_range(self, batch_size=0):
"""Split this key range into a list of at most two ranges.
This method attempts to split the key range approximately in half.
Numeric ranges are split in the middle into two equal ranges and
string ranges are split lexicographically in the middle. If the
key range is smaller than batch_size it is left unsplit.
Note that splitting is done without knowledge of the distribution
of actual entities in the key range, so there is no guarantee (nor
any particular reason to believe) that the entities of the range
are evenly split.
Args:
batch_size: The maximum size of a key range that should not be split.
Returns:
A list of one or two key ranges covering the same space as this range.
"""
key_start = self.key_start
key_end = self.key_end
include_start = self.include_start
include_end = self.include_end
key_pairs = []
if not key_start:
key_pairs.append((key_start, include_start, key_end, include_end,
KeyRange.ASC))
elif not key_end:
key_pairs.append((key_start, include_start, key_end, include_end,
KeyRange.DESC))
else:
key_split = KeyRange.split_keys(key_start, key_end, batch_size)
first_include_end = True
if key_split == key_start:
first_include_end = first_include_end and include_start
key_pairs.append((key_start, include_start,
key_split, first_include_end,
KeyRange.DESC))
second_include_end = include_end
if key_split == key_end:
second_include_end = False
key_pairs.append((key_split, False,
key_end, second_include_end,
KeyRange.ASC))
ranges = [KeyRange(key_start=start,
include_start=include_start,
key_end=end,
include_end=include_end,
direction=direction,
namespace=self.namespace,
_app=self._app)
for (start, include_start, end, include_end, direction)
in key_pairs]
return ranges
def __hash__(self):
return hash([self.key_start,
self.key_end,
self.direction,
self._app,
self.namespace])
def __cmp__(self, other):
"""Compare two key ranges.
Key ranges with a value of None for key_start or key_end, are always
considered to have include_start=False or include_end=False, respectively,
when comparing. Since None indicates an unbounded side of the range,
the include specifier is meaningless. The ordering generated is total
but somewhat arbitrary.
Args:
other: An object to compare to this one.
Returns:
-1: if this key range is less than other.
0: if this key range is equal to other.
1: if this key range is greater than other.
"""
if not isinstance(other, KeyRange):
return 1
self_list = [self.key_start, self.key_end, self.direction,
self.include_start, self.include_end, self._app,
self.namespace]
if not self.key_start:
self_list[3] = False
if not self.key_end:
self_list[4] = False
other_list = [other.key_start,
other.key_end,
other.direction,
other.include_start,
other.include_end,
other._app,
other.namespace]
if not other.key_start:
other_list[3] = False
if not other.key_end:
other_list[4] = False
return cmp(self_list, other_list)
@staticmethod
def bisect_string_range(start, end):
"""Returns a string that is approximately in the middle of the range.
(start, end) is treated as a string range, and it is assumed
start <= end in the usual lexicographic string ordering. The output key
mid is guaranteed to satisfy start <= mid <= end.
The method proceeds by comparing initial characters of start and
end. When the characters are equal, they are appended to the mid
string. In the first place that the characters differ, the
difference characters are averaged and this average is appended to
the mid string. If averaging resulted in rounding down, and
additional character is added to the mid string to make up for the
rounding down. This extra step is necessary for correctness in
the case that the average of the two characters is equal to the
character in the start string.
This method makes the assumption that most keys are ascii and it
attempts to perform splitting within the ascii range when that
results in a valid split.
Args:
start: A string.
end: A string such that start <= end.
Returns:
A string mid such that start <= mid <= end.
"""
if start == end:
return start
start += "\0"
end += "\0"
midpoint = []
expected_max = 127
for i in xrange(min(len(start), len(end))):
if start[i] == end[i]:
midpoint.append(start[i])
else:
ord_sum = ord(start[i]) + ord(end[i])
midpoint.append(unichr(ord_sum / 2))
if ord_sum % 2:
if len(start) > i + 1:
ord_start = ord(start[i+1])
else:
ord_start = 0
if ord_start < expected_max:
ord_split = (expected_max + ord_start) / 2
else:
ord_split = (0xFFFF + ord_start) / 2
midpoint.append(unichr(ord_split))
break
return "".join(midpoint)
@staticmethod
def split_keys(key_start, key_end, batch_size):
"""Return a key that is between key_start and key_end inclusive.
This method compares components of the ancestor paths of key_start
and key_end. The first place in the path that differs is
approximately split in half. If the kind components differ, a new
non-existent kind halfway between the two is used to split the
space. If the id_or_name components differ, then a new id_or_name
that is halfway between the two is selected. If the lower
id_or_name is numeric and the upper id_or_name is a string, then
the minumum string key u'\0' is used as the split id_or_name. The
key that is returned is the shared portion of the ancestor path
followed by the generated split component.
Args:
key_start: A db.Key instance for the lower end of a range.
key_end: A db.Key instance for the upper end of a range.
batch_size: The maximum size of a range that should not be split.
Returns:
A db.Key instance, k, such that key_start <= k <= key_end.
"""
assert key_start.app() == key_end.app()
assert key_start.namespace() == key_end.namespace()
path1 = key_start.to_path()
path2 = key_end.to_path()
len1 = len(path1)
len2 = len(path2)
assert len1 % 2 == 0
assert len2 % 2 == 0
out_path = []
min_path_len = min(len1, len2) / 2
for i in xrange(min_path_len):
kind1 = path1[2*i]
kind2 = path2[2*i]
if kind1 != kind2:
split_kind = KeyRange.bisect_string_range(kind1, kind2)
out_path.append(split_kind)
out_path.append(unichr(0))
break
last = (len1 == len2 == 2*(i + 1))
id_or_name1 = path1[2*i + 1]
id_or_name2 = path2[2*i + 1]
id_or_name_split = KeyRange._split_id_or_name(
id_or_name1, id_or_name2, batch_size, last)
if id_or_name1 == id_or_name_split:
out_path.append(kind1)
out_path.append(id_or_name1)
else:
out_path.append(kind1)
out_path.append(id_or_name_split)
break
return db.Key.from_path(
*out_path,
**{"_app": key_start.app(), "namespace": key_start.namespace()})
@staticmethod
def _split_id_or_name(id_or_name1, id_or_name2, batch_size, maintain_batches):
"""Return an id_or_name that is between id_or_name1 an id_or_name2.
Attempts to split the range [id_or_name1, id_or_name2] in half,
unless maintain_batches is true and the size of the range
[id_or_name1, id_or_name2] is less than or equal to batch_size.
Args:
id_or_name1: A number or string or the id_or_name component of a key
id_or_name2: A number or string or the id_or_name component of a key
batch_size: The range size that will not be split if maintain_batches
is true.
maintain_batches: A boolean for whether to keep small ranges intact.
Returns:
An id_or_name such that id_or_name1 <= id_or_name <= id_or_name2.
"""
if (isinstance(id_or_name1, (int, long)) and
isinstance(id_or_name2, (int, long))):
if not maintain_batches or id_or_name2 - id_or_name1 > batch_size:
return (id_or_name1 + id_or_name2) / 2
else:
return id_or_name1
elif (isinstance(id_or_name1, basestring) and
isinstance(id_or_name2, basestring)):
return KeyRange.bisect_string_range(id_or_name1, id_or_name2)
else:
if (not isinstance(id_or_name1, (int, long)) or
not isinstance(id_or_name2, basestring)):
raise KeyRangeError("Wrong key order: %r, %r" %
(id_or_name1, id_or_name2))
zero_ch = unichr(0)
if id_or_name2 == zero_ch:
return (id_or_name1 + 2**63 - 1) / 2
return zero_ch
@staticmethod
def guess_end_key(kind,
key_start,
probe_count=30,
split_rate=5):
"""Guess the end of a key range with a binary search of probe queries.
When the 'key_start' parameter has a key hierarchy, this function will
only determine the key range for keys in a similar hierarchy. That means
if the keys are in the form:
kind=Foo, name=bar/kind=Stuff, name=meep
only this range will be probed:
kind=Foo, name=*/kind=Stuff, name=*
That means other entities of kind 'Stuff' that are children of another
parent entity kind will be skipped:
kind=Other, name=cookie/kind=Stuff, name=meep
Args:
key_start: The starting key of the search range. In most cases this
should be id = 0 or name = '\0'.
kind: String name of the entity kind.
probe_count: Optional, how many probe queries to run.
split_rate: Exponential rate to use for splitting the range on the
way down from the full key space. For smaller ranges this should
be higher so more of the keyspace is skipped on initial descent.
Returns:
datastore.Key that is guaranteed to be as high or higher than the
highest key existing for this Kind. Doing a query between 'key_start' and
this returned Key (inclusive) will contain all entities of this Kind.
"""
app = key_start.app()
namespace = key_start.namespace()
full_path = key_start.to_path()
for index, piece in enumerate(full_path):
if index % 2 == 0:
continue
elif isinstance(piece, basestring):
full_path[index] = u"\xffff"
else:
full_path[index] = 2**63 - 1
key_end = datastore.Key.from_path(*full_path,
**{"_app": app, "namespace": namespace})
split_key = key_end
for i in xrange(probe_count):
for j in xrange(split_rate):
split_key = KeyRange.split_keys(key_start, split_key, 1)
results = datastore.Query(
kind,
{"__key__ >": split_key},
namespace=namespace,
_app=app,
keys_only=True).Get(1)
if results:
if results[0].name() and not key_start.name():
return KeyRange.guess_end_key(
kind, results[0], probe_count - 1, split_rate)
else:
split_rate = 1
key_start = results[0]
split_key = key_end
else:
key_end = split_key
return key_end
def to_json(self):
"""Serialize KeyRange to json.
Returns:
string with KeyRange json representation.
"""
if simplejson is None:
raise SimplejsonUnavailableError(
"JSON functionality requires json or simplejson to be available")
def key_to_str(key):
if key:
return str(key)
else:
return None
obj_dict = {
"direction": self.direction,
"key_start": key_to_str(self.key_start),
"key_end": key_to_str(self.key_end),
"include_start": self.include_start,
"include_end": self.include_end,
"namespace": self.namespace,
}
if self._app:
obj_dict["_app"] = self._app
return simplejson.dumps(obj_dict, sort_keys=True)
@staticmethod
def from_json(json_str):
"""Deserialize KeyRange from its json representation.
Args:
json_str: string with json representation created by key_range_to_json.
Returns:
deserialized KeyRange instance.
"""
if simplejson is None:
raise SimplejsonUnavailableError(
"JSON functionality requires json or simplejson to be available")
def key_from_str(key_str):
if key_str:
return db.Key(key_str)
else:
return None
json = simplejson.loads(json_str)
return KeyRange(key_from_str(json["key_start"]),
key_from_str(json["key_end"]),
json["direction"],
json["include_start"],
json["include_end"],
json.get("namespace"),
_app=json.get("_app"))