1
# Copyright (C) 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
"""Tests for btree indices."""
31
from bzrlib.tests import (
32
TestCaseWithTransport,
35
split_suite_by_condition,
37
from bzrlib.transport import get_transport
40
def load_tests(standard_tests, module, loader):
41
# parameterise the TestBTreeNodes tests
42
node_tests, others = split_suite_by_condition(standard_tests,
43
condition_isinstance(TestBTreeNodes))
44
import bzrlib._btree_serializer_py as py_module
45
scenarios = [('python', {'parse_btree': py_module})]
46
if compiled_btreeparser.available():
47
scenarios.append(('C', {'parse_btree': compiled_btreeparser.module}))
48
return multiply_tests(node_tests, scenarios, others)
51
compiled_btreeparser = tests.ModuleAvailableFeature(
52
'bzrlib._btree_serializer_pyx')
55
class BTreeTestCase(TestCaseWithTransport):
56
# test names here are suffixed by the key length and reference list count
60
TestCaseWithTransport.setUp(self)
61
self._original_header = btree_index._RESERVED_HEADER_BYTES
63
btree_index._RESERVED_HEADER_BYTES = self._original_header
64
self.addCleanup(restore)
65
btree_index._RESERVED_HEADER_BYTES = 100
67
def make_nodes(self, count, key_elements, reference_lists):
68
"""Generate count*key_elements sample nodes."""
70
for prefix_pos in range(key_elements):
72
prefix = (str(prefix_pos) * 40,)
75
for pos in xrange(count):
76
# TODO: This creates odd keys. When count == 100,000, it
77
# creates a 240 byte key
78
key = prefix + (str(pos) * 40,)
79
value = "value:%s" % pos
81
# generate some references
83
for list_pos in range(reference_lists):
84
# as many keys in each list as its index + the key depth
85
# mod 2 - this generates both 0 length lists and
86
# ones slightly longer than the number of lists.
87
# It also ensures we have non homogeneous lists.
89
for ref_pos in range(list_pos + pos % 2):
91
# refer to a nearby key
92
refs[-1].append(prefix + ("ref" + str(pos - 1) * 40,))
94
# serial of this ref in the ref list
95
refs[-1].append(prefix + ("ref" + str(ref_pos) * 40,))
96
refs[-1] = tuple(refs[-1])
100
keys.append((key, value, refs))
103
def shrink_page_size(self):
104
"""Shrink the default page size so that less fits in a page."""
105
old_page_size = btree_index._PAGE_SIZE
107
btree_index._PAGE_SIZE = old_page_size
108
self.addCleanup(cleanup)
109
btree_index._PAGE_SIZE = 2048
112
class TestBTreeBuilder(BTreeTestCase):
114
def test_clear_cache(self):
115
builder = btree_index.BTreeBuilder(reference_lists=0, key_elements=1)
116
# This is a no-op, but we need the api to be consistent with other
117
# BTreeGraphIndex apis.
118
builder.clear_cache()
120
def test_empty_1_0(self):
121
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
122
# NamedTemporaryFile dies on builder.finish().read(). weird.
123
temp_file = builder.finish()
124
content = temp_file.read()
127
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=0\n"
131
def test_empty_2_1(self):
132
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=1)
133
# NamedTemporaryFile dies on builder.finish().read(). weird.
134
temp_file = builder.finish()
135
content = temp_file.read()
138
"B+Tree Graph Index 2\nnode_ref_lists=1\nkey_elements=2\nlen=0\n"
142
def test_root_leaf_1_0(self):
143
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
144
nodes = self.make_nodes(5, 1, 0)
146
builder.add_node(*node)
147
# NamedTemporaryFile dies on builder.finish().read(). weird.
148
temp_file = builder.finish()
149
content = temp_file.read()
151
self.assertEqual(131, len(content))
153
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=5\n"
156
node_content = content[73:]
157
node_bytes = zlib.decompress(node_content)
158
expected_node = ("type=leaf\n"
159
"0000000000000000000000000000000000000000\x00\x00value:0\n"
160
"1111111111111111111111111111111111111111\x00\x00value:1\n"
161
"2222222222222222222222222222222222222222\x00\x00value:2\n"
162
"3333333333333333333333333333333333333333\x00\x00value:3\n"
163
"4444444444444444444444444444444444444444\x00\x00value:4\n")
164
self.assertEqual(expected_node, node_bytes)
166
def test_root_leaf_2_2(self):
167
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
168
nodes = self.make_nodes(5, 2, 2)
170
builder.add_node(*node)
171
# NamedTemporaryFile dies on builder.finish().read(). weird.
172
temp_file = builder.finish()
173
content = temp_file.read()
175
self.assertEqual(238, len(content))
177
"B+Tree Graph Index 2\nnode_ref_lists=2\nkey_elements=2\nlen=10\n"
180
node_content = content[74:]
181
node_bytes = zlib.decompress(node_content)
184
"0000000000000000000000000000000000000000\x000000000000000000000000000000000000000000\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:0\n"
185
"0000000000000000000000000000000000000000\x001111111111111111111111111111111111111111\x000000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\r0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:1\n"
186
"0000000000000000000000000000000000000000\x002222222222222222222222222222222222222222\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:2\n"
187
"0000000000000000000000000000000000000000\x003333333333333333333333333333333333333333\x000000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\t0000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\r0000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\x00value:3\n"
188
"0000000000000000000000000000000000000000\x004444444444444444444444444444444444444444\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:4\n"
189
"1111111111111111111111111111111111111111\x000000000000000000000000000000000000000000\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:0\n"
190
"1111111111111111111111111111111111111111\x001111111111111111111111111111111111111111\x001111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\r1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:1\n"
191
"1111111111111111111111111111111111111111\x002222222222222222222222222222222222222222\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:2\n"
192
"1111111111111111111111111111111111111111\x003333333333333333333333333333333333333333\x001111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\t1111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\r1111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\x00value:3\n"
193
"1111111111111111111111111111111111111111\x004444444444444444444444444444444444444444\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:4\n"
196
self.assertEqual(expected_node, node_bytes)
198
def test_2_leaves_1_0(self):
199
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
200
nodes = self.make_nodes(400, 1, 0)
202
builder.add_node(*node)
203
# NamedTemporaryFile dies on builder.finish().read(). weird.
204
temp_file = builder.finish()
205
content = temp_file.read()
207
self.assertEqual(9283, len(content))
209
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=400\n"
212
root = content[77:4096]
213
leaf1 = content[4096:8192]
214
leaf2 = content[8192:]
215
root_bytes = zlib.decompress(root)
219
) + ("307" * 40) + "\n"
220
self.assertEqual(expected_root, root_bytes)
221
# We already know serialisation works for leaves, check key selection:
222
leaf1_bytes = zlib.decompress(leaf1)
223
sorted_node_keys = sorted(node[0] for node in nodes)
224
node = btree_index._LeafNode(leaf1_bytes, 1, 0)
225
self.assertEqual(231, len(node.keys))
226
self.assertEqual(sorted_node_keys[:231], sorted(node.keys))
227
leaf2_bytes = zlib.decompress(leaf2)
228
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
229
self.assertEqual(400 - 231, len(node.keys))
230
self.assertEqual(sorted_node_keys[231:], sorted(node.keys))
232
def test_last_page_rounded_1_layer(self):
233
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
234
nodes = self.make_nodes(10, 1, 0)
236
builder.add_node(*node)
237
# NamedTemporaryFile dies on builder.finish().read(). weird.
238
temp_file = builder.finish()
239
content = temp_file.read()
241
self.assertEqual(155, len(content))
243
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=10\n"
246
# Check thelast page is well formed
248
leaf2_bytes = zlib.decompress(leaf2)
249
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
250
self.assertEqual(10, len(node.keys))
251
sorted_node_keys = sorted(node[0] for node in nodes)
252
self.assertEqual(sorted_node_keys, sorted(node.keys))
254
def test_last_page_not_rounded_2_layer(self):
255
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
256
nodes = self.make_nodes(400, 1, 0)
258
builder.add_node(*node)
259
# NamedTemporaryFile dies on builder.finish().read(). weird.
260
temp_file = builder.finish()
261
content = temp_file.read()
263
self.assertEqual(9283, len(content))
265
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=400\n"
268
# Check the last page is well formed
269
leaf2 = content[8192:]
270
leaf2_bytes = zlib.decompress(leaf2)
271
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
272
self.assertEqual(400 - 231, len(node.keys))
273
sorted_node_keys = sorted(node[0] for node in nodes)
274
self.assertEqual(sorted_node_keys[231:], sorted(node.keys))
276
def test_three_level_tree_details(self):
277
# The left most pointer in the second internal node in a row should
278
# pointer to the second node that the internal node is for, _not_
279
# the first, otherwise the first node overlaps with the last node of
280
# the prior internal node on that row.
281
self.shrink_page_size()
282
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
283
# 40K nodes is enough to create a two internal nodes on the second
284
# level, with a 2K page size
285
nodes = self.make_nodes(20000, 2, 2)
288
builder.add_node(*node)
289
transport = get_transport('trace+' + self.get_url(''))
290
size = transport.put_file('index', self.time(builder.finish))
292
index = btree_index.BTreeGraphIndex(transport, 'index', size)
293
# Seed the metadata, we're using internal calls now.
295
self.assertEqual(3, len(index._row_lengths),
296
"Not enough rows: %r" % index._row_lengths)
297
self.assertEqual(4, len(index._row_offsets))
298
self.assertEqual(sum(index._row_lengths), index._row_offsets[-1])
299
internal_nodes = index._get_internal_nodes([0, 1, 2])
300
root_node = internal_nodes[0]
301
internal_node1 = internal_nodes[1]
302
internal_node2 = internal_nodes[2]
303
# The left most node node2 points at should be one after the right most
304
# node pointed at by node1.
305
self.assertEqual(internal_node2.offset, 1 + len(internal_node1.keys))
306
# The left most key of the second node pointed at by internal_node2
307
# should be its first key. We can check this by looking for its first key
308
# in the second node it points at
309
pos = index._row_offsets[2] + internal_node2.offset + 1
310
leaf = index._get_leaf_nodes([pos])[pos]
311
self.assertTrue(internal_node2.keys[0] in leaf.keys)
313
def test_2_leaves_2_2(self):
314
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
315
nodes = self.make_nodes(100, 2, 2)
317
builder.add_node(*node)
318
# NamedTemporaryFile dies on builder.finish().read(). weird.
319
temp_file = builder.finish()
320
content = temp_file.read()
322
self.assertEqual(12643, len(content))
324
"B+Tree Graph Index 2\nnode_ref_lists=2\nkey_elements=2\nlen=200\n"
327
root = content[77:4096]
328
leaf1 = content[4096:8192]
329
leaf2 = content[8192:12288]
330
leaf3 = content[12288:]
331
root_bytes = zlib.decompress(root)
335
+ ("0" * 40) + "\x00" + ("91" * 40) + "\n"
336
+ ("1" * 40) + "\x00" + ("81" * 40) + "\n"
338
self.assertEqual(expected_root, root_bytes)
339
# We assume the other leaf nodes have been written correctly - layering
342
def test_spill_index_stress_1_1(self):
343
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
344
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
345
builder.add_node(*nodes[0])
346
# Test the parts of the index that take up memory are doing so
348
self.assertEqual(1, len(builder._nodes))
349
self.assertIs(None, builder._nodes_by_key)
350
builder.add_node(*nodes[1])
351
self.assertEqual(0, len(builder._nodes))
352
self.assertIs(None, builder._nodes_by_key)
353
self.assertEqual(1, len(builder._backing_indices))
354
self.assertEqual(2, builder._backing_indices[0].key_count())
356
builder.add_node(*nodes[2])
357
self.assertEqual(1, len(builder._nodes))
358
self.assertIs(None, builder._nodes_by_key)
359
# And spills to a second backing index combing all
360
builder.add_node(*nodes[3])
361
self.assertEqual(0, len(builder._nodes))
362
self.assertIs(None, builder._nodes_by_key)
363
self.assertEqual(2, len(builder._backing_indices))
364
self.assertEqual(None, builder._backing_indices[0])
365
self.assertEqual(4, builder._backing_indices[1].key_count())
366
# The next spills to the 2-len slot
367
builder.add_node(*nodes[4])
368
builder.add_node(*nodes[5])
369
self.assertEqual(0, len(builder._nodes))
370
self.assertIs(None, builder._nodes_by_key)
371
self.assertEqual(2, len(builder._backing_indices))
372
self.assertEqual(2, builder._backing_indices[0].key_count())
373
self.assertEqual(4, builder._backing_indices[1].key_count())
374
# Next spill combines
375
builder.add_node(*nodes[6])
376
builder.add_node(*nodes[7])
377
self.assertEqual(3, len(builder._backing_indices))
378
self.assertEqual(None, builder._backing_indices[0])
379
self.assertEqual(None, builder._backing_indices[1])
380
self.assertEqual(8, builder._backing_indices[2].key_count())
381
# And so forth - counting up in binary.
382
builder.add_node(*nodes[8])
383
builder.add_node(*nodes[9])
384
self.assertEqual(3, len(builder._backing_indices))
385
self.assertEqual(2, builder._backing_indices[0].key_count())
386
self.assertEqual(None, builder._backing_indices[1])
387
self.assertEqual(8, builder._backing_indices[2].key_count())
388
builder.add_node(*nodes[10])
389
builder.add_node(*nodes[11])
390
self.assertEqual(3, len(builder._backing_indices))
391
self.assertEqual(None, builder._backing_indices[0])
392
self.assertEqual(4, builder._backing_indices[1].key_count())
393
self.assertEqual(8, builder._backing_indices[2].key_count())
394
builder.add_node(*nodes[12])
395
# Test that memory and disk are both used for query methods; and that
396
# None is skipped over happily.
397
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
398
list(builder.iter_all_entries()))
399
# Two nodes - one memory one disk
400
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
401
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
402
self.assertEqual(13, builder.key_count())
403
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
404
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
405
builder.add_node(*nodes[13])
406
self.assertEqual(3, len(builder._backing_indices))
407
self.assertEqual(2, builder._backing_indices[0].key_count())
408
self.assertEqual(4, builder._backing_indices[1].key_count())
409
self.assertEqual(8, builder._backing_indices[2].key_count())
410
builder.add_node(*nodes[14])
411
builder.add_node(*nodes[15])
412
self.assertEqual(4, len(builder._backing_indices))
413
self.assertEqual(None, builder._backing_indices[0])
414
self.assertEqual(None, builder._backing_indices[1])
415
self.assertEqual(None, builder._backing_indices[2])
416
self.assertEqual(16, builder._backing_indices[3].key_count())
417
# Now finish, and check we got a correctly ordered tree
418
transport = self.get_transport('')
419
size = transport.put_file('index', builder.finish())
420
index = btree_index.BTreeGraphIndex(transport, 'index', size)
421
nodes = list(index.iter_all_entries())
422
self.assertEqual(sorted(nodes), nodes)
423
self.assertEqual(16, len(nodes))
425
def test_spill_index_stress_1_1_no_combine(self):
426
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
427
builder.set_optimize(for_size=False, combine_backing_indices=False)
428
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
429
builder.add_node(*nodes[0])
430
# Test the parts of the index that take up memory are doing so
432
self.assertEqual(1, len(builder._nodes))
433
self.assertIs(None, builder._nodes_by_key)
434
builder.add_node(*nodes[1])
435
self.assertEqual(0, len(builder._nodes))
436
self.assertIs(None, builder._nodes_by_key)
437
self.assertEqual(1, len(builder._backing_indices))
438
self.assertEqual(2, builder._backing_indices[0].key_count())
440
builder.add_node(*nodes[2])
441
self.assertEqual(1, len(builder._nodes))
442
self.assertIs(None, builder._nodes_by_key)
443
# And spills to a second backing index but doesn't combine
444
builder.add_node(*nodes[3])
445
self.assertEqual(0, len(builder._nodes))
446
self.assertIs(None, builder._nodes_by_key)
447
self.assertEqual(2, len(builder._backing_indices))
448
for backing_index in builder._backing_indices:
449
self.assertEqual(2, backing_index.key_count())
450
# The next spills to the 3rd slot
451
builder.add_node(*nodes[4])
452
builder.add_node(*nodes[5])
453
self.assertEqual(0, len(builder._nodes))
454
self.assertIs(None, builder._nodes_by_key)
455
self.assertEqual(3, len(builder._backing_indices))
456
for backing_index in builder._backing_indices:
457
self.assertEqual(2, backing_index.key_count())
458
# Now spill a few more, and check that we don't combine
459
builder.add_node(*nodes[6])
460
builder.add_node(*nodes[7])
461
builder.add_node(*nodes[8])
462
builder.add_node(*nodes[9])
463
builder.add_node(*nodes[10])
464
builder.add_node(*nodes[11])
465
builder.add_node(*nodes[12])
466
self.assertEqual(6, len(builder._backing_indices))
467
for backing_index in builder._backing_indices:
468
self.assertEqual(2, backing_index.key_count())
469
# Test that memory and disk are both used for query methods; and that
470
# None is skipped over happily.
471
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
472
list(builder.iter_all_entries()))
473
# Two nodes - one memory one disk
474
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
475
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
476
self.assertEqual(13, builder.key_count())
477
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
478
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
479
builder.add_node(*nodes[13])
480
builder.add_node(*nodes[14])
481
builder.add_node(*nodes[15])
482
self.assertEqual(8, len(builder._backing_indices))
483
for backing_index in builder._backing_indices:
484
self.assertEqual(2, backing_index.key_count())
485
# Now finish, and check we got a correctly ordered tree
486
transport = self.get_transport('')
487
size = transport.put_file('index', builder.finish())
488
index = btree_index.BTreeGraphIndex(transport, 'index', size)
489
nodes = list(index.iter_all_entries())
490
self.assertEqual(sorted(nodes), nodes)
491
self.assertEqual(16, len(nodes))
493
def test_set_optimize(self):
494
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
495
builder.set_optimize(for_size=True)
496
self.assertTrue(builder._optimize_for_size)
497
builder.set_optimize(for_size=False)
498
self.assertFalse(builder._optimize_for_size)
499
# test that we can set combine_backing_indices without effecting
502
builder._optimize_for_size = obj
503
builder.set_optimize(combine_backing_indices=False)
504
self.assertFalse(builder._combine_backing_indices)
505
self.assertIs(obj, builder._optimize_for_size)
506
builder.set_optimize(combine_backing_indices=True)
507
self.assertTrue(builder._combine_backing_indices)
508
self.assertIs(obj, builder._optimize_for_size)
510
def test_spill_index_stress_2_2(self):
511
# test that references and longer keys don't confuse things.
512
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2,
514
nodes = self.make_nodes(16, 2, 2)
515
builder.add_node(*nodes[0])
516
# Test the parts of the index that take up memory are doing so
518
self.assertEqual(1, len(builder._nodes))
519
self.assertIs(None, builder._nodes_by_key)
520
builder.add_node(*nodes[1])
521
self.assertEqual(0, len(builder._nodes))
522
self.assertIs(None, builder._nodes_by_key)
523
self.assertEqual(1, len(builder._backing_indices))
524
self.assertEqual(2, builder._backing_indices[0].key_count())
526
old = dict(builder._get_nodes_by_key()) #Build up the nodes by key dict
527
builder.add_node(*nodes[2])
528
self.assertEqual(1, len(builder._nodes))
529
self.assertIsNot(None, builder._nodes_by_key)
530
self.assertNotEqual({}, builder._nodes_by_key)
531
# We should have a new entry
532
self.assertNotEqual(old, builder._nodes_by_key)
533
# And spills to a second backing index combing all
534
builder.add_node(*nodes[3])
535
self.assertEqual(0, len(builder._nodes))
536
self.assertIs(None, builder._nodes_by_key)
537
self.assertEqual(2, len(builder._backing_indices))
538
self.assertEqual(None, builder._backing_indices[0])
539
self.assertEqual(4, builder._backing_indices[1].key_count())
540
# The next spills to the 2-len slot
541
builder.add_node(*nodes[4])
542
builder.add_node(*nodes[5])
543
self.assertEqual(0, len(builder._nodes))
544
self.assertIs(None, builder._nodes_by_key)
545
self.assertEqual(2, len(builder._backing_indices))
546
self.assertEqual(2, builder._backing_indices[0].key_count())
547
self.assertEqual(4, builder._backing_indices[1].key_count())
548
# Next spill combines
549
builder.add_node(*nodes[6])
550
builder.add_node(*nodes[7])
551
self.assertEqual(3, len(builder._backing_indices))
552
self.assertEqual(None, builder._backing_indices[0])
553
self.assertEqual(None, builder._backing_indices[1])
554
self.assertEqual(8, builder._backing_indices[2].key_count())
555
# And so forth - counting up in binary.
556
builder.add_node(*nodes[8])
557
builder.add_node(*nodes[9])
558
self.assertEqual(3, len(builder._backing_indices))
559
self.assertEqual(2, builder._backing_indices[0].key_count())
560
self.assertEqual(None, builder._backing_indices[1])
561
self.assertEqual(8, builder._backing_indices[2].key_count())
562
builder.add_node(*nodes[10])
563
builder.add_node(*nodes[11])
564
self.assertEqual(3, len(builder._backing_indices))
565
self.assertEqual(None, builder._backing_indices[0])
566
self.assertEqual(4, builder._backing_indices[1].key_count())
567
self.assertEqual(8, builder._backing_indices[2].key_count())
568
builder.add_node(*nodes[12])
569
# Test that memory and disk are both used for query methods; and that
570
# None is skipped over happily.
571
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
572
list(builder.iter_all_entries()))
573
# Two nodes - one memory one disk
574
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
575
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
576
self.assertEqual(13, builder.key_count())
577
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
578
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
579
builder.add_node(*nodes[13])
580
self.assertEqual(3, len(builder._backing_indices))
581
self.assertEqual(2, builder._backing_indices[0].key_count())
582
self.assertEqual(4, builder._backing_indices[1].key_count())
583
self.assertEqual(8, builder._backing_indices[2].key_count())
584
builder.add_node(*nodes[14])
585
builder.add_node(*nodes[15])
586
self.assertEqual(4, len(builder._backing_indices))
587
self.assertEqual(None, builder._backing_indices[0])
588
self.assertEqual(None, builder._backing_indices[1])
589
self.assertEqual(None, builder._backing_indices[2])
590
self.assertEqual(16, builder._backing_indices[3].key_count())
591
# Now finish, and check we got a correctly ordered tree
592
transport = self.get_transport('')
593
size = transport.put_file('index', builder.finish())
594
index = btree_index.BTreeGraphIndex(transport, 'index', size)
595
nodes = list(index.iter_all_entries())
596
self.assertEqual(sorted(nodes), nodes)
597
self.assertEqual(16, len(nodes))
599
def test_spill_index_duplicate_key_caught_on_finish(self):
600
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
601
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
602
builder.add_node(*nodes[0])
603
builder.add_node(*nodes[1])
604
builder.add_node(*nodes[0])
605
self.assertRaises(errors.BadIndexDuplicateKey, builder.finish)
608
class TestBTreeIndex(BTreeTestCase):
610
def make_index(self, ref_lists=0, key_elements=1, nodes=[]):
611
builder = btree_index.BTreeBuilder(reference_lists=ref_lists,
612
key_elements=key_elements)
613
for key, value, references in nodes:
614
builder.add_node(key, value, references)
615
stream = builder.finish()
616
trans = get_transport('trace+' + self.get_url())
617
size = trans.put_file('index', stream)
618
return btree_index.BTreeGraphIndex(trans, 'index', size)
620
def test_clear_cache(self):
621
nodes = self.make_nodes(160, 2, 2)
622
index = self.make_index(ref_lists=2, key_elements=2, nodes=nodes)
623
self.assertEqual(1, len(list(index.iter_entries([nodes[30][0]]))))
624
self.assertEqual([1, 4], index._row_lengths)
625
self.assertIsNot(None, index._root_node)
626
internal_node_pre_clear = index._internal_node_cache.keys()
627
self.assertTrue(len(index._leaf_node_cache) > 0)
629
# We don't touch _root_node or _internal_node_cache, both should be
630
# small, and can save a round trip or two
631
self.assertIsNot(None, index._root_node)
632
# NOTE: We don't want to affect the _internal_node_cache, as we expect
633
# it will be small, and if we ever do touch this index again, it
634
# will save round-trips. This assertion isn't very strong,
635
# becuase without a 3-level index, we don't have any internal
637
self.assertEqual(internal_node_pre_clear,
638
index._internal_node_cache.keys())
639
self.assertEqual(0, len(index._leaf_node_cache))
641
def test_trivial_constructor(self):
642
transport = get_transport('trace+' + self.get_url(''))
643
index = btree_index.BTreeGraphIndex(transport, 'index', None)
644
# Checks the page size at load, but that isn't logged yet.
645
self.assertEqual([], transport._activity)
647
def test_with_size_constructor(self):
648
transport = get_transport('trace+' + self.get_url(''))
649
index = btree_index.BTreeGraphIndex(transport, 'index', 1)
650
# Checks the page size at load, but that isn't logged yet.
651
self.assertEqual([], transport._activity)
653
def test_empty_key_count_no_size(self):
654
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
655
transport = get_transport('trace+' + self.get_url(''))
656
transport.put_file('index', builder.finish())
657
index = btree_index.BTreeGraphIndex(transport, 'index', None)
658
del transport._activity[:]
659
self.assertEqual([], transport._activity)
660
self.assertEqual(0, index.key_count())
661
# The entire index should have been requested (as we generally have the
662
# size available, and doing many small readvs is inappropriate).
663
# We can't tell how much was actually read here, but - check the code.
664
self.assertEqual([('get', 'index')], transport._activity)
666
def test_empty_key_count(self):
667
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
668
transport = get_transport('trace+' + self.get_url(''))
669
size = transport.put_file('index', builder.finish())
670
self.assertEqual(72, size)
671
index = btree_index.BTreeGraphIndex(transport, 'index', size)
672
del transport._activity[:]
673
self.assertEqual([], transport._activity)
674
self.assertEqual(0, index.key_count())
675
# The entire index should have been read, as 4K > size
676
self.assertEqual([('readv', 'index', [(0, 72)], False, None)],
679
def test_non_empty_key_count_2_2(self):
680
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
681
nodes = self.make_nodes(35, 2, 2)
683
builder.add_node(*node)
684
transport = get_transport('trace+' + self.get_url(''))
685
size = transport.put_file('index', builder.finish())
686
index = btree_index.BTreeGraphIndex(transport, 'index', size)
687
del transport._activity[:]
688
self.assertEqual([], transport._activity)
689
self.assertEqual(70, index.key_count())
690
# The entire index should have been read, as it is one page long.
691
self.assertEqual([('readv', 'index', [(0, size)], False, None)],
693
self.assertEqual(1173, size)
695
def test__read_nodes_no_size_one_page_reads_once(self):
696
self.make_index(nodes=[(('key',), 'value', ())])
697
trans = get_transport('trace+' + self.get_url())
698
index = btree_index.BTreeGraphIndex(trans, 'index', None)
699
del trans._activity[:]
700
nodes = dict(index._read_nodes([0]))
701
self.assertEqual([0], nodes.keys())
703
self.assertEqual([('key',)], node.keys.keys())
704
self.assertEqual([('get', 'index')], trans._activity)
706
def test__read_nodes_no_size_multiple_pages(self):
707
index = self.make_index(2, 2, nodes=self.make_nodes(160, 2, 2))
709
num_pages = index._row_offsets[-1]
710
# Reopen with a traced transport and no size
711
trans = get_transport('trace+' + self.get_url())
712
index = btree_index.BTreeGraphIndex(trans, 'index', None)
713
del trans._activity[:]
714
nodes = dict(index._read_nodes([0]))
715
self.assertEqual(range(num_pages), nodes.keys())
717
def test_2_levels_key_count_2_2(self):
718
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
719
nodes = self.make_nodes(160, 2, 2)
721
builder.add_node(*node)
722
transport = get_transport('trace+' + self.get_url(''))
723
size = transport.put_file('index', builder.finish())
724
self.assertEqual(17692, size)
725
index = btree_index.BTreeGraphIndex(transport, 'index', size)
726
del transport._activity[:]
727
self.assertEqual([], transport._activity)
728
self.assertEqual(320, index.key_count())
729
# The entire index should not have been read.
730
self.assertEqual([('readv', 'index', [(0, 4096)], False, None)],
733
def test_validate_one_page(self):
734
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
735
nodes = self.make_nodes(45, 2, 2)
737
builder.add_node(*node)
738
transport = get_transport('trace+' + self.get_url(''))
739
size = transport.put_file('index', builder.finish())
740
index = btree_index.BTreeGraphIndex(transport, 'index', size)
741
del transport._activity[:]
742
self.assertEqual([], transport._activity)
744
# The entire index should have been read linearly.
745
self.assertEqual([('readv', 'index', [(0, size)], False, None)],
747
self.assertEqual(1488, size)
749
def test_validate_two_pages(self):
750
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
751
nodes = self.make_nodes(80, 2, 2)
753
builder.add_node(*node)
754
transport = get_transport('trace+' + self.get_url(''))
755
size = transport.put_file('index', builder.finish())
756
# Root page, 2 leaf pages
757
self.assertEqual(9339, size)
758
index = btree_index.BTreeGraphIndex(transport, 'index', size)
759
del transport._activity[:]
760
self.assertEqual([], transport._activity)
762
# The entire index should have been read linearly.
763
self.assertEqual([('readv', 'index', [(0, 4096)], False, None),
764
('readv', 'index', [(4096, 4096), (8192, 1147)], False, None)],
766
# XXX: TODO: write some badly-ordered nodes, and some pointers-to-wrong
767
# node and make validate find them.
769
def test_eq_ne(self):
770
# two indices are equal when constructed with the same parameters:
771
transport1 = get_transport('trace+' + self.get_url(''))
772
transport2 = get_transport(self.get_url(''))
774
btree_index.BTreeGraphIndex(transport1, 'index', None) ==
775
btree_index.BTreeGraphIndex(transport1, 'index', None))
777
btree_index.BTreeGraphIndex(transport1, 'index', 20) ==
778
btree_index.BTreeGraphIndex(transport1, 'index', 20))
780
btree_index.BTreeGraphIndex(transport1, 'index', 20) ==
781
btree_index.BTreeGraphIndex(transport2, 'index', 20))
783
btree_index.BTreeGraphIndex(transport1, 'inde1', 20) ==
784
btree_index.BTreeGraphIndex(transport1, 'inde2', 20))
786
btree_index.BTreeGraphIndex(transport1, 'index', 10) ==
787
btree_index.BTreeGraphIndex(transport1, 'index', 20))
789
btree_index.BTreeGraphIndex(transport1, 'index', None) !=
790
btree_index.BTreeGraphIndex(transport1, 'index', None))
792
btree_index.BTreeGraphIndex(transport1, 'index', 20) !=
793
btree_index.BTreeGraphIndex(transport1, 'index', 20))
795
btree_index.BTreeGraphIndex(transport1, 'index', 20) !=
796
btree_index.BTreeGraphIndex(transport2, 'index', 20))
798
btree_index.BTreeGraphIndex(transport1, 'inde1', 20) !=
799
btree_index.BTreeGraphIndex(transport1, 'inde2', 20))
801
btree_index.BTreeGraphIndex(transport1, 'index', 10) !=
802
btree_index.BTreeGraphIndex(transport1, 'index', 20))
804
def test_iter_all_only_root_no_size(self):
805
self.make_index(nodes=[(('key',), 'value', ())])
806
trans = get_transport('trace+' + self.get_url(''))
807
index = btree_index.BTreeGraphIndex(trans, 'index', None)
808
del trans._activity[:]
809
self.assertEqual([(('key',), 'value')],
810
[x[1:] for x in index.iter_all_entries()])
811
self.assertEqual([('get', 'index')], trans._activity)
813
def test_iter_all_entries_reads(self):
814
# iterating all entries reads the header, then does a linear
816
self.shrink_page_size()
817
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
818
# 20k nodes is enough to create a two internal nodes on the second
819
# level, with a 2K page size
820
nodes = self.make_nodes(10000, 2, 2)
822
builder.add_node(*node)
823
transport = get_transport('trace+' + self.get_url(''))
824
size = transport.put_file('index', builder.finish())
825
self.assertEqual(1303220, size, 'number of expected bytes in the'
827
page_size = btree_index._PAGE_SIZE
829
index = btree_index.BTreeGraphIndex(transport, 'index', size)
830
del transport._activity[:]
831
self.assertEqual([], transport._activity)
832
found_nodes = self.time(list, index.iter_all_entries())
834
for node in found_nodes:
835
self.assertTrue(node[0] is index)
836
bare_nodes.append(node[1:])
837
self.assertEqual(3, len(index._row_lengths),
838
"Not enough rows: %r" % index._row_lengths)
839
# Should be as long as the nodes we supplied
840
self.assertEqual(20000, len(found_nodes))
841
# Should have the same content
842
self.assertEqual(set(nodes), set(bare_nodes))
843
# Should have done linear scan IO up the index, ignoring
844
# the internal nodes:
845
# The entire index should have been read
846
total_pages = sum(index._row_lengths)
847
self.assertEqual(total_pages, index._row_offsets[-1])
848
self.assertEqual(1303220, size)
849
# The start of the leaves
850
first_byte = index._row_offsets[-2] * page_size
852
for offset in range(first_byte, size, page_size):
853
readv_request.append((offset, page_size))
854
# The last page is truncated
855
readv_request[-1] = (readv_request[-1][0], 1303220 % page_size)
856
expected = [('readv', 'index', [(0, page_size)], False, None),
857
('readv', 'index', readv_request, False, None)]
858
if expected != transport._activity:
859
self.assertEqualDiff(pprint.pformat(expected),
860
pprint.pformat(transport._activity))
862
def _test_iter_entries_references_resolved(self):
863
index = self.make_index(1, nodes=[
864
(('name', ), 'data', ([('ref', ), ('ref', )], )),
865
(('ref', ), 'refdata', ([], ))])
866
self.assertEqual(set([(index, ('name', ), 'data', ((('ref',),('ref',)),)),
867
(index, ('ref', ), 'refdata', ((), ))]),
868
set(index.iter_entries([('name',), ('ref',)])))
870
def test_iter_entries_references_2_refs_resolved(self):
871
# iterating some entries reads just the pages needed. For now, to
872
# get it working and start measuring, only 4K pages are read.
873
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
874
# 80 nodes is enough to create a two-level index.
875
nodes = self.make_nodes(160, 2, 2)
877
builder.add_node(*node)
878
transport = get_transport('trace+' + self.get_url(''))
879
size = transport.put_file('index', builder.finish())
881
index = btree_index.BTreeGraphIndex(transport, 'index', size)
882
del transport._activity[:]
883
self.assertEqual([], transport._activity)
885
found_nodes = list(index.iter_entries([nodes[30][0]]))
887
for node in found_nodes:
888
self.assertTrue(node[0] is index)
889
bare_nodes.append(node[1:])
890
# Should be as long as the nodes we supplied
891
self.assertEqual(1, len(found_nodes))
892
# Should have the same content
893
self.assertEqual(nodes[30], bare_nodes[0])
894
# Should have read the root node, then one leaf page:
895
self.assertEqual([('readv', 'index', [(0, 4096)], False, None),
896
('readv', 'index', [(8192, 4096), ], False, None)],
899
def test_iter_key_prefix_1_element_key_None(self):
900
index = self.make_index()
901
self.assertRaises(errors.BadIndexKey, list,
902
index.iter_entries_prefix([(None, )]))
904
def test_iter_key_prefix_wrong_length(self):
905
index = self.make_index()
906
self.assertRaises(errors.BadIndexKey, list,
907
index.iter_entries_prefix([('foo', None)]))
908
index = self.make_index(key_elements=2)
909
self.assertRaises(errors.BadIndexKey, list,
910
index.iter_entries_prefix([('foo', )]))
911
self.assertRaises(errors.BadIndexKey, list,
912
index.iter_entries_prefix([('foo', None, None)]))
914
def test_iter_key_prefix_1_key_element_no_refs(self):
915
index = self.make_index( nodes=[
916
(('name', ), 'data', ()),
917
(('ref', ), 'refdata', ())])
918
self.assertEqual(set([(index, ('name', ), 'data'),
919
(index, ('ref', ), 'refdata')]),
920
set(index.iter_entries_prefix([('name', ), ('ref', )])))
922
def test_iter_key_prefix_1_key_element_refs(self):
923
index = self.make_index(1, nodes=[
924
(('name', ), 'data', ([('ref', )], )),
925
(('ref', ), 'refdata', ([], ))])
926
self.assertEqual(set([(index, ('name', ), 'data', ((('ref',),),)),
927
(index, ('ref', ), 'refdata', ((), ))]),
928
set(index.iter_entries_prefix([('name', ), ('ref', )])))
930
def test_iter_key_prefix_2_key_element_no_refs(self):
931
index = self.make_index(key_elements=2, nodes=[
932
(('name', 'fin1'), 'data', ()),
933
(('name', 'fin2'), 'beta', ()),
934
(('ref', 'erence'), 'refdata', ())])
935
self.assertEqual(set([(index, ('name', 'fin1'), 'data'),
936
(index, ('ref', 'erence'), 'refdata')]),
937
set(index.iter_entries_prefix([('name', 'fin1'), ('ref', 'erence')])))
938
self.assertEqual(set([(index, ('name', 'fin1'), 'data'),
939
(index, ('name', 'fin2'), 'beta')]),
940
set(index.iter_entries_prefix([('name', None)])))
942
def test_iter_key_prefix_2_key_element_refs(self):
943
index = self.make_index(1, key_elements=2, nodes=[
944
(('name', 'fin1'), 'data', ([('ref', 'erence')], )),
945
(('name', 'fin2'), 'beta', ([], )),
946
(('ref', 'erence'), 'refdata', ([], ))])
947
self.assertEqual(set([(index, ('name', 'fin1'), 'data', ((('ref', 'erence'),),)),
948
(index, ('ref', 'erence'), 'refdata', ((), ))]),
949
set(index.iter_entries_prefix([('name', 'fin1'), ('ref', 'erence')])))
950
self.assertEqual(set([(index, ('name', 'fin1'), 'data', ((('ref', 'erence'),),)),
951
(index, ('name', 'fin2'), 'beta', ((), ))]),
952
set(index.iter_entries_prefix([('name', None)])))
954
# XXX: external_references tests are duplicated in test_index. We
955
# probably should have per_graph_index tests...
956
def test_external_references_no_refs(self):
957
index = self.make_index(ref_lists=0, nodes=[])
958
self.assertRaises(ValueError, index.external_references, 0)
960
def test_external_references_no_results(self):
961
index = self.make_index(ref_lists=1, nodes=[
962
(('key',), 'value', ([],))])
963
self.assertEqual(set(), index.external_references(0))
965
def test_external_references_missing_ref(self):
966
missing_key = ('missing',)
967
index = self.make_index(ref_lists=1, nodes=[
968
(('key',), 'value', ([missing_key],))])
969
self.assertEqual(set([missing_key]), index.external_references(0))
971
def test_external_references_multiple_ref_lists(self):
972
missing_key = ('missing',)
973
index = self.make_index(ref_lists=2, nodes=[
974
(('key',), 'value', ([], [missing_key]))])
975
self.assertEqual(set([]), index.external_references(0))
976
self.assertEqual(set([missing_key]), index.external_references(1))
978
def test_external_references_two_records(self):
979
index = self.make_index(ref_lists=1, nodes=[
980
(('key-1',), 'value', ([('key-2',)],)),
981
(('key-2',), 'value', ([],)),
983
self.assertEqual(set([]), index.external_references(0))
985
def test__find_ancestors_one_page(self):
988
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
989
(key1, 'value', ([key2],)),
990
(key2, 'value', ([],)),
994
search_keys = index._find_ancestors([key1], 0, parent_map, missing_keys)
995
self.assertEqual({key1: (key2,), key2: ()}, parent_map)
996
self.assertEqual(set(), missing_keys)
997
self.assertEqual(set(), search_keys)
999
def test__find_ancestors_one_page_w_missing(self):
1003
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1004
(key1, 'value', ([key2],)),
1005
(key2, 'value', ([],)),
1008
missing_keys = set()
1009
search_keys = index._find_ancestors([key2, key3], 0, parent_map,
1011
self.assertEqual({key2: ()}, parent_map)
1012
# we know that key3 is missing because we read the page that it would
1014
self.assertEqual(set([key3]), missing_keys)
1015
self.assertEqual(set(), search_keys)
1017
def test__find_ancestors_one_parent_missing(self):
1021
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1022
(key1, 'value', ([key2],)),
1023
(key2, 'value', ([key3],)),
1026
missing_keys = set()
1027
search_keys = index._find_ancestors([key1], 0, parent_map,
1029
self.assertEqual({key1: (key2,), key2: (key3,)}, parent_map)
1030
self.assertEqual(set(), missing_keys)
1031
# all we know is that key3 wasn't present on the page we were reading
1032
# but if you look, the last key is key2 which comes before key3, so we
1033
# don't know whether key3 would land on this page or not.
1034
self.assertEqual(set([key3]), search_keys)
1035
search_keys = index._find_ancestors(search_keys, 0, parent_map,
1037
# passing it back in, we are sure it is 'missing'
1038
self.assertEqual({key1: (key2,), key2: (key3,)}, parent_map)
1039
self.assertEqual(set([key3]), missing_keys)
1040
self.assertEqual(set([]), search_keys)
1042
def test__find_ancestors_dont_search_known(self):
1046
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1047
(key1, 'value', ([key2],)),
1048
(key2, 'value', ([key3],)),
1049
(key3, 'value', ([],)),
1051
# We already know about key2, so we won't try to search for key3
1052
parent_map = {key2: (key3,)}
1053
missing_keys = set()
1054
search_keys = index._find_ancestors([key1], 0, parent_map,
1056
self.assertEqual({key1: (key2,), key2: (key3,)}, parent_map)
1057
self.assertEqual(set(), missing_keys)
1058
self.assertEqual(set(), search_keys)
1060
def test__find_ancestors_multiple_pages(self):
1061
# We need to use enough keys that we actually cause a split
1062
start_time = 1249671539
1063
email = "joebob@example.com"
1067
for i in xrange(400):
1068
rev_id = '%s-%s-%s' % (email,
1069
osutils.compact_date(start_time + i),
1070
osutils.rand_chars(16))
1072
nodes.append((rev_key, 'value', ref_lists))
1073
# We have a ref 'list' of length 1, with a list of parents, with 1
1074
# parent which is a key
1075
ref_lists = ((rev_key,),)
1076
rev_keys.append(rev_key)
1077
index = self.make_index(ref_lists=1, key_elements=1, nodes=nodes)
1078
self.assertEqual(400, index.key_count())
1079
self.assertEqual(3, len(index._row_offsets))
1080
nodes = dict(index._read_nodes([1, 2]))
1083
min_l2_key = l2.min_key
1084
max_l1_key = l1.max_key
1085
self.assertTrue(max_l1_key < min_l2_key)
1086
parents_min_l2_key = l2.keys[min_l2_key][1][0]
1087
self.assertEqual((l1.max_key,), parents_min_l2_key)
1088
# Now, whatever key we select that would fall on the second page,
1089
# should give us all the parents until the page break
1090
key_idx = rev_keys.index(min_l2_key)
1091
next_key = rev_keys[key_idx+1]
1092
# So now when we get the parent map, we should get the key we are
1093
# looking for, min_l2_key, and then a reference to go look for the
1094
# parent of that key
1096
missing_keys = set()
1097
search_keys = index._find_ancestors([next_key], 0, parent_map,
1099
self.assertEqual([min_l2_key, next_key], sorted(parent_map))
1100
self.assertEqual(set(), missing_keys)
1101
self.assertEqual(set([max_l1_key]), search_keys)
1103
search_keys = index._find_ancestors([max_l1_key], 0, parent_map,
1105
self.assertEqual(sorted(l1.keys), sorted(parent_map))
1106
self.assertEqual(set(), missing_keys)
1107
self.assertEqual(set(), search_keys)
1109
def test__find_ancestors_empty_index(self):
1110
index = self.make_index(ref_lists=1, key_elements=1, nodes=[])
1112
missing_keys = set()
1113
search_keys = index._find_ancestors([('one',), ('two',)], 0, parent_map,
1115
self.assertEqual(set(), search_keys)
1116
self.assertEqual({}, parent_map)
1117
self.assertEqual(set([('one',), ('two',)]), missing_keys)
1119
def test_supports_unlimited_cache(self):
1120
builder = btree_index.BTreeBuilder(reference_lists=0, key_elements=1)
1121
# We need enough nodes to cause a page split (so we have both an
1122
# internal node and a couple leaf nodes. 500 seems to be enough.)
1123
nodes = self.make_nodes(500, 1, 0)
1125
builder.add_node(*node)
1126
stream = builder.finish()
1127
trans = get_transport(self.get_url())
1128
size = trans.put_file('index', stream)
1129
index = btree_index.BTreeGraphIndex(trans, 'index', size)
1130
self.assertEqual(500, index.key_count())
1131
# We have an internal node
1132
self.assertEqual(2, len(index._row_lengths))
1133
# We have at least 2 leaf nodes
1134
self.assertTrue(index._row_lengths[-1] >= 2)
1135
self.assertIsInstance(index._leaf_node_cache, lru_cache.LRUCache)
1136
self.assertEqual(btree_index._NODE_CACHE_SIZE,
1137
index._leaf_node_cache._max_cache)
1138
self.assertIsInstance(index._internal_node_cache, fifo_cache.FIFOCache)
1139
self.assertEqual(100, index._internal_node_cache._max_cache)
1140
# No change if unlimited_cache=False is passed
1141
index = btree_index.BTreeGraphIndex(trans, 'index', size,
1142
unlimited_cache=False)
1143
self.assertIsInstance(index._leaf_node_cache, lru_cache.LRUCache)
1144
self.assertEqual(btree_index._NODE_CACHE_SIZE,
1145
index._leaf_node_cache._max_cache)
1146
self.assertIsInstance(index._internal_node_cache, fifo_cache.FIFOCache)
1147
self.assertEqual(100, index._internal_node_cache._max_cache)
1148
index = btree_index.BTreeGraphIndex(trans, 'index', size,
1149
unlimited_cache=True)
1150
self.assertIsInstance(index._leaf_node_cache, dict)
1151
self.assertIs(type(index._internal_node_cache), dict)
1152
# Exercise the lookup code
1153
entries = set(index.iter_entries([n[0] for n in nodes]))
1154
self.assertEqual(500, len(entries))
1157
class TestBTreeNodes(BTreeTestCase):
1159
def restore_parser(self):
1160
btree_index._btree_serializer = self.saved_parser
1163
BTreeTestCase.setUp(self)
1164
self.saved_parser = btree_index._btree_serializer
1165
self.addCleanup(self.restore_parser)
1166
btree_index._btree_serializer = self.parse_btree
1168
def test_LeafNode_1_0(self):
1169
node_bytes = ("type=leaf\n"
1170
"0000000000000000000000000000000000000000\x00\x00value:0\n"
1171
"1111111111111111111111111111111111111111\x00\x00value:1\n"
1172
"2222222222222222222222222222222222222222\x00\x00value:2\n"
1173
"3333333333333333333333333333333333333333\x00\x00value:3\n"
1174
"4444444444444444444444444444444444444444\x00\x00value:4\n")
1175
node = btree_index._LeafNode(node_bytes, 1, 0)
1176
# We do direct access, or don't care about order, to leaf nodes most of
1177
# the time, so a dict is useful:
1179
("0000000000000000000000000000000000000000",): ("value:0", ()),
1180
("1111111111111111111111111111111111111111",): ("value:1", ()),
1181
("2222222222222222222222222222222222222222",): ("value:2", ()),
1182
("3333333333333333333333333333333333333333",): ("value:3", ()),
1183
("4444444444444444444444444444444444444444",): ("value:4", ()),
1186
def test_LeafNode_2_2(self):
1187
node_bytes = ("type=leaf\n"
1188
"00\x0000\x00\t00\x00ref00\x00value:0\n"
1189
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n"
1190
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n"
1191
"11\x0044\x00\t11\x00ref00\x00value:4\n"
1194
node = btree_index._LeafNode(node_bytes, 2, 2)
1195
# We do direct access, or don't care about order, to leaf nodes most of
1196
# the time, so a dict is useful:
1198
('00', '00'): ('value:0', ((), (('00', 'ref00'),))),
1199
('00', '11'): ('value:1',
1200
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01')))),
1201
('11', '33'): ('value:3',
1202
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22')))),
1203
('11', '44'): ('value:4', ((), (('11', 'ref00'),)))
1206
def test_InternalNode_1(self):
1207
node_bytes = ("type=internal\n"
1209
"0000000000000000000000000000000000000000\n"
1210
"1111111111111111111111111111111111111111\n"
1211
"2222222222222222222222222222222222222222\n"
1212
"3333333333333333333333333333333333333333\n"
1213
"4444444444444444444444444444444444444444\n"
1215
node = btree_index._InternalNode(node_bytes)
1216
# We want to bisect to find the right children from this node, so a
1217
# vector is most useful.
1219
("0000000000000000000000000000000000000000",),
1220
("1111111111111111111111111111111111111111",),
1221
("2222222222222222222222222222222222222222",),
1222
("3333333333333333333333333333333333333333",),
1223
("4444444444444444444444444444444444444444",),
1225
self.assertEqual(1, node.offset)
1227
def test_LeafNode_2_2(self):
1228
node_bytes = ("type=leaf\n"
1229
"00\x0000\x00\t00\x00ref00\x00value:0\n"
1230
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n"
1231
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n"
1232
"11\x0044\x00\t11\x00ref00\x00value:4\n"
1235
node = btree_index._LeafNode(node_bytes, 2, 2)
1236
# We do direct access, or don't care about order, to leaf nodes most of
1237
# the time, so a dict is useful:
1239
('00', '00'): ('value:0', ((), (('00', 'ref00'),))),
1240
('00', '11'): ('value:1',
1241
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01')))),
1242
('11', '33'): ('value:3',
1243
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22')))),
1244
('11', '44'): ('value:4', ((), (('11', 'ref00'),)))
1247
def assertFlattened(self, expected, key, value, refs):
1248
flat_key, flat_line = self.parse_btree._flatten_node(
1249
(None, key, value, refs), bool(refs))
1250
self.assertEqual('\x00'.join(key), flat_key)
1251
self.assertEqual(expected, flat_line)
1253
def test__flatten_node(self):
1254
self.assertFlattened('key\0\0value\n', ('key',), 'value', [])
1255
self.assertFlattened('key\0tuple\0\0value str\n',
1256
('key', 'tuple'), 'value str', [])
1257
self.assertFlattened('key\0tuple\0triple\0\0value str\n',
1258
('key', 'tuple', 'triple'), 'value str', [])
1259
self.assertFlattened('k\0t\0s\0ref\0value str\n',
1260
('k', 't', 's'), 'value str', [[('ref',)]])
1261
self.assertFlattened('key\0tuple\0ref\0key\0value str\n',
1262
('key', 'tuple'), 'value str', [[('ref', 'key')]])
1263
self.assertFlattened("00\x0000\x00\t00\x00ref00\x00value:0\n",
1264
('00', '00'), 'value:0', ((), (('00', 'ref00'),)))
1265
self.assertFlattened(
1266
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n",
1267
('00', '11'), 'value:1',
1268
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01'))))
1269
self.assertFlattened(
1270
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n",
1271
('11', '33'), 'value:3',
1272
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22'))))
1273
self.assertFlattened(
1274
"11\x0044\x00\t11\x00ref00\x00value:4\n",
1275
('11', '44'), 'value:4', ((), (('11', 'ref00'),)))
1278
class TestCompiledBtree(tests.TestCase):
1280
def test_exists(self):
1281
# This is just to let the user know if they don't have the feature
1283
self.requireFeature(compiled_btreeparser)
1286
class TestMultiBisectRight(tests.TestCase):
1288
def assertMultiBisectRight(self, offsets, search_keys, fixed_keys):
1289
self.assertEqual(offsets,
1290
btree_index.BTreeGraphIndex._multi_bisect_right(
1291
search_keys, fixed_keys))
1293
def test_after(self):
1294
self.assertMultiBisectRight([(1, ['b'])], ['b'], ['a'])
1295
self.assertMultiBisectRight([(3, ['e', 'f', 'g'])],
1296
['e', 'f', 'g'], ['a', 'b', 'c'])
1298
def test_before(self):
1299
self.assertMultiBisectRight([(0, ['a'])], ['a'], ['b'])
1300
self.assertMultiBisectRight([(0, ['a', 'b', 'c', 'd'])],
1301
['a', 'b', 'c', 'd'], ['e', 'f', 'g'])
1303
def test_exact(self):
1304
self.assertMultiBisectRight([(1, ['a'])], ['a'], ['a'])
1305
self.assertMultiBisectRight([(1, ['a']), (2, ['b'])], ['a', 'b'], ['a', 'b'])
1306
self.assertMultiBisectRight([(1, ['a']), (3, ['c'])],
1307
['a', 'c'], ['a', 'b', 'c'])
1309
def test_inbetween(self):
1310
self.assertMultiBisectRight([(1, ['b'])], ['b'], ['a', 'c'])
1311
self.assertMultiBisectRight([(1, ['b', 'c', 'd']), (2, ['f', 'g'])],
1312
['b', 'c', 'd', 'f', 'g'], ['a', 'e', 'h'])
1314
def test_mixed(self):
1315
self.assertMultiBisectRight([(0, ['a', 'b']), (2, ['d', 'e']),
1317
['a', 'b', 'd', 'e', 'g', 'h'],
1318
['c', 'd', 'f', 'g'])
1321
class TestExpandOffsets(tests.TestCase):
1323
def make_index(self, size, recommended_pages=None):
1324
"""Make an index with a generic size.
1326
This doesn't actually create anything on disk, it just primes a
1327
BTreeGraphIndex with the recommended information.
1329
index = btree_index.BTreeGraphIndex(get_transport('memory:///'),
1330
'test-index', size=size)
1331
if recommended_pages is not None:
1332
index._recommended_pages = recommended_pages
1335
def set_cached_offsets(self, index, cached_offsets):
1336
"""Monkeypatch to give a canned answer for _get_offsets_for...()."""
1337
def _get_offsets_to_cached_pages():
1338
cached = set(cached_offsets)
1340
index._get_offsets_to_cached_pages = _get_offsets_to_cached_pages
1342
def prepare_index(self, index, node_ref_lists, key_length, key_count,
1343
row_lengths, cached_offsets):
1344
"""Setup the BTreeGraphIndex with some pre-canned information."""
1345
index.node_ref_lists = node_ref_lists
1346
index._key_length = key_length
1347
index._key_count = key_count
1348
index._row_lengths = row_lengths
1349
index._compute_row_offsets()
1350
index._root_node = btree_index._InternalNode('internal\noffset=0\n')
1351
self.set_cached_offsets(index, cached_offsets)
1353
def make_100_node_index(self):
1354
index = self.make_index(4096*100, 6)
1355
# Consider we've already made a single request at the middle
1356
self.prepare_index(index, node_ref_lists=0, key_length=1,
1357
key_count=1000, row_lengths=[1, 99],
1358
cached_offsets=[0, 50])
1361
def make_1000_node_index(self):
1362
index = self.make_index(4096*1000, 6)
1363
# Pretend we've already made a single request in the middle
1364
self.prepare_index(index, node_ref_lists=0, key_length=1,
1365
key_count=90000, row_lengths=[1, 9, 990],
1366
cached_offsets=[0, 5, 500])
1369
def assertNumPages(self, expected_pages, index, size):
1371
self.assertEqual(expected_pages, index._compute_total_pages_in_index())
1373
def assertExpandOffsets(self, expected, index, offsets):
1374
self.assertEqual(expected, index._expand_offsets(offsets),
1375
'We did not get the expected value after expanding'
1378
def test_default_recommended_pages(self):
1379
index = self.make_index(None)
1380
# local transport recommends 4096 byte reads, which is 1 page
1381
self.assertEqual(1, index._recommended_pages)
1383
def test__compute_total_pages_in_index(self):
1384
index = self.make_index(None)
1385
self.assertNumPages(1, index, 1024)
1386
self.assertNumPages(1, index, 4095)
1387
self.assertNumPages(1, index, 4096)
1388
self.assertNumPages(2, index, 4097)
1389
self.assertNumPages(2, index, 8192)
1390
self.assertNumPages(76, index, 4096*75 + 10)
1392
def test__find_layer_start_and_stop(self):
1393
index = self.make_1000_node_index()
1394
self.assertEqual((0, 1), index._find_layer_first_and_end(0))
1395
self.assertEqual((1, 10), index._find_layer_first_and_end(1))
1396
self.assertEqual((1, 10), index._find_layer_first_and_end(9))
1397
self.assertEqual((10, 1000), index._find_layer_first_and_end(10))
1398
self.assertEqual((10, 1000), index._find_layer_first_and_end(99))
1399
self.assertEqual((10, 1000), index._find_layer_first_and_end(999))
1401
def test_unknown_size(self):
1402
# We should not expand if we don't know the file size
1403
index = self.make_index(None, 10)
1404
self.assertExpandOffsets([0], index, [0])
1405
self.assertExpandOffsets([1, 4, 9], index, [1, 4, 9])
1407
def test_more_than_recommended(self):
1408
index = self.make_index(4096*100, 2)
1409
self.assertExpandOffsets([1, 10], index, [1, 10])
1410
self.assertExpandOffsets([1, 10, 20], index, [1, 10, 20])
1412
def test_read_all_from_root(self):
1413
index = self.make_index(4096*10, 20)
1414
self.assertExpandOffsets(range(10), index, [0])
1416
def test_read_all_when_cached(self):
1417
# We've read enough that we can grab all the rest in a single request
1418
index = self.make_index(4096*10, 5)
1419
self.prepare_index(index, node_ref_lists=0, key_length=1,
1420
key_count=1000, row_lengths=[1, 9],
1421
cached_offsets=[0, 1, 2, 5, 6])
1422
# It should fill the remaining nodes, regardless of the one requested
1423
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [3])
1424
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [8])
1425
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [9])
1427
def test_no_root_node(self):
1428
index = self.make_index(4096*10, 5)
1429
self.assertExpandOffsets([0], index, [0])
1431
def test_include_neighbors(self):
1432
index = self.make_100_node_index()
1433
# We expand in both directions, until we have at least 'recommended'
1435
self.assertExpandOffsets([9, 10, 11, 12, 13, 14, 15], index, [12])
1436
self.assertExpandOffsets([88, 89, 90, 91, 92, 93, 94], index, [91])
1437
# If we hit an 'edge' we continue in the other direction
1438
self.assertExpandOffsets([1, 2, 3, 4, 5, 6], index, [2])
1439
self.assertExpandOffsets([94, 95, 96, 97, 98, 99], index, [98])
1441
# Requesting many nodes will expand all locations equally
1442
self.assertExpandOffsets([1, 2, 3, 80, 81, 82], index, [2, 81])
1443
self.assertExpandOffsets([1, 2, 3, 9, 10, 11, 80, 81, 82], index,
1446
def test_stop_at_cached(self):
1447
index = self.make_100_node_index()
1448
self.set_cached_offsets(index, [0, 10, 19])
1449
self.assertExpandOffsets([11, 12, 13, 14, 15, 16], index, [11])
1450
self.assertExpandOffsets([11, 12, 13, 14, 15, 16], index, [12])
1451
self.assertExpandOffsets([12, 13, 14, 15, 16, 17, 18], index, [15])
1452
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [16])
1453
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [17])
1454
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [18])
1456
def test_cannot_fully_expand(self):
1457
index = self.make_100_node_index()
1458
self.set_cached_offsets(index, [0, 10, 12])
1459
# We don't go into an endless loop if we are bound by cached nodes
1460
self.assertExpandOffsets([11], index, [11])
1462
def test_overlap(self):
1463
index = self.make_100_node_index()
1464
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [12, 13])
1465
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [11, 14])
1467
def test_stay_within_layer(self):
1468
index = self.make_1000_node_index()
1469
# When expanding a request, we won't read nodes from the next layer
1470
self.assertExpandOffsets([1, 2, 3, 4], index, [2])
1471
self.assertExpandOffsets([6, 7, 8, 9], index, [6])
1472
self.assertExpandOffsets([6, 7, 8, 9], index, [9])
1473
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [10])
1474
self.assertExpandOffsets([10, 11, 12, 13, 14, 15, 16], index, [13])
1476
self.set_cached_offsets(index, [0, 4, 12])
1477
self.assertExpandOffsets([5, 6, 7, 8, 9], index, [7])
1478
self.assertExpandOffsets([10, 11], index, [11])
1480
def test_small_requests_unexpanded(self):
1481
index = self.make_100_node_index()
1482
self.set_cached_offsets(index, [0])
1483
self.assertExpandOffsets([1], index, [1])
1484
self.assertExpandOffsets([50], index, [50])
1485
# If we request more than one node, then we'll expand
1486
self.assertExpandOffsets([49, 50, 51, 59, 60, 61], index, [50, 60])
1488
# The first pass does not expand
1489
index = self.make_1000_node_index()
1490
self.set_cached_offsets(index, [0])
1491
self.assertExpandOffsets([1], index, [1])
1492
self.set_cached_offsets(index, [0, 1])
1493
self.assertExpandOffsets([100], index, [100])
1494
self.set_cached_offsets(index, [0, 1, 100])
1495
# But after the first depth, we will expand
1496
self.assertExpandOffsets([2, 3, 4, 5, 6, 7], index, [2])
1497
self.assertExpandOffsets([2, 3, 4, 5, 6, 7], index, [4])
1498
self.set_cached_offsets(index, [0, 1, 2, 3, 4, 5, 6, 7, 100])
1499
self.assertExpandOffsets([102, 103, 104, 105, 106, 107, 108], index,
added
removed
bzrlib/tests/test_btree_index.py
