Download a globally available singleton when launching an application on Google App Engine using Python

Question

Using the Google application engine, is it possible to initialize a globally available singleton when the application starts? I have a large static tree structure that I must use for each request and want to initialize it in advance. The tree structure is too large (20 + MB) for placement in Memcache, and I'm trying to figure out what other alternatives I have.

EDIT: just add some clarity based on the answers I've received so far. I am loading a word dictionary into a trie / prefix tree structure. Trie is immutable because the vocabulary of words is fixed. I create anagrams based on the input character string, so a single request can access a fair number of three on a single request, possibly more than 1 MB, however I'm not sure.

Here is the python structure into which I load the word dictionary.

class Node(object):

    def __init__(self, letter='', final=False):
        self.letter = letter
        self.final = final
        self.children = {}

    def add(self, letters):
        node = self
        for index, letter in enumerate(letters):
            if letter not in node.children:
                node.children[letter] = Node(letter, index==len(letters)-1)
            node = node.children[letter]

Answer 1

How much of this whole tree do you need to get per query? How do you request it? Has this ever changed?

, "singleton", - . , , , .

Answer 2

, (, , ). , " " , ( memcache , : , ! -).

, " " , , "".

: , " ", , - . , - "" , "" , . - , :

import array
import random

def _doinsert(tree, payload):
  if tree is None: return HeavyTree(payload)
  tree.insert(payload)
  return tree

class HeavyTree(object):
  def __init__(self, payload):
    self.payload = payload
    self.left = self.right = None
  def insert(self, other):
    if other <= self.payload:
      self.left = _doinsert(self.left, other)
    else:
      self.right = _doinsert(self.right, other)
  def walk(self):
    if self.left:
      for x in self.left.walk(): yield x
    yield self.payload
    if self.right:
      for x in self.right.walk(): yield x
  def walknodes(self):
    yield self
    if self.left:
      for x in self.left.walknodes(): yield x
    if self.right:
      for x in self.right.walknodes(): yield x

data = [random.randint(0, 99) for _ in range(9)]
print 'data: ',
for x in data: print x,
print
theiter = iter(data)
thetree = HeavyTree(next(theiter))
for x in theiter: thetree.insert(x)

print
print 'Heavy tree:'
print 'nodes:',
for x in thetree.walknodes(): print x.payload,
print
print 'inord:',
for x in thetree.walk(): print x,
print

class LightTree(HeavyTree):
  def __init__(self, base, offset):
    self.base = base
    self.offset = offset
  @property
  def payload(self):
    return self.base[self.offset]
  @property
  def left(self):
    return self._astree(self.offset+1)
  @property
  def right(self):
    return self._astree(self.offset+2)
  def _astree(self, i):
    offset = self.base[i]
    if offset < 0: return None
    return LightTree(self.base, offset)

def heavy_to_light(heavy):
  for i, node in enumerate(heavy.walknodes()):
    node.id = i * 3
  base = array.array('l', (i+1) * 3 * [-1])
  for node in heavy.walknodes():
    base[node.id] = node.payload
    if node.left: base[node.id+1] = node.left.id
    if node.right: base[node.id+2] = node.right.id
  return LightTree(base, 0)

print
print 'Light tree:'
light = heavy_to_light(thetree)
print 'nodes:',
for x in light.walknodes(): print x.payload,
print
print 'base :',
for x in light.base: print x,
print
print 'inord:',
for x in light.walk(): print x,
print

:

data:  27 79 90 60 82 80 3 94 76

Heavy tree:
nodes: 27 3 79 60 76 90 82 80 94
inord: 3 27 60 76 79 80 82 90 94

Light tree:
nodes: 27 3 79 60 76 90 82 80 94
base : 27 3 6 3 -1 -1 79 9 15 60 -1 12 76 -1 -1 90 18 24 82 21 -1 80 -1 -1 94 -1 -1
inord: 3 27 60 76 79 80 82 90 94

, , .

, , (, , ), EE school ;-). ( unpickling ..)...! -)

Answer 3

, Google 300 , . , , , - .

, Google , , , . , 30 , , Google .

Answer 4

Memcache - . . trie- , memcache? - memcache, , trie memcache, . trie , . , .