1
# Copyright (C) 2008-2011 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."""
32
from bzrlib.tests import (
33
TestCaseWithTransport,
38
load_tests = scenarios.load_tests_apply_scenarios
41
def btreeparser_scenarios():
42
import bzrlib._btree_serializer_py as py_module
43
scenarios = [('python', {'parse_btree': py_module})]
44
if compiled_btreeparser_feature.available():
45
scenarios.append(('C',
46
{'parse_btree': compiled_btreeparser_feature.module}))
50
compiled_btreeparser_feature = tests.ModuleAvailableFeature(
51
'bzrlib._btree_serializer_pyx')
54
class BTreeTestCase(TestCaseWithTransport):
55
# test names here are suffixed by the key length and reference list count
59
TestCaseWithTransport.setUp(self)
60
self.overrideAttr(btree_index, '_RESERVED_HEADER_BYTES', 100)
62
def make_nodes(self, count, key_elements, reference_lists):
63
"""Generate count*key_elements sample nodes."""
65
for prefix_pos in range(key_elements):
67
prefix = (str(prefix_pos) * 40,)
70
for pos in xrange(count):
71
# TODO: This creates odd keys. When count == 100,000, it
72
# creates a 240 byte key
73
key = prefix + (str(pos) * 40,)
74
value = "value:%s" % pos
76
# generate some references
78
for list_pos in range(reference_lists):
79
# as many keys in each list as its index + the key depth
80
# mod 2 - this generates both 0 length lists and
81
# ones slightly longer than the number of lists.
82
# It also ensures we have non homogeneous lists.
84
for ref_pos in range(list_pos + pos % 2):
86
# refer to a nearby key
87
refs[-1].append(prefix + ("ref" + str(pos - 1) * 40,))
89
# serial of this ref in the ref list
90
refs[-1].append(prefix + ("ref" + str(ref_pos) * 40,))
91
refs[-1] = tuple(refs[-1])
95
keys.append((key, value, refs))
98
def shrink_page_size(self):
99
"""Shrink the default page size so that less fits in a page."""
100
self.overrideAttr(btree_index, '_PAGE_SIZE')
101
btree_index._PAGE_SIZE = 2048
104
class TestBTreeBuilder(BTreeTestCase):
106
def test_clear_cache(self):
107
builder = btree_index.BTreeBuilder(reference_lists=0, key_elements=1)
108
# This is a no-op, but we need the api to be consistent with other
109
# BTreeGraphIndex apis.
110
builder.clear_cache()
112
def test_empty_1_0(self):
113
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
114
# NamedTemporaryFile dies on builder.finish().read(). weird.
115
temp_file = builder.finish()
116
content = temp_file.read()
119
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=0\n"
123
def test_empty_2_1(self):
124
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=1)
125
# NamedTemporaryFile dies on builder.finish().read(). weird.
126
temp_file = builder.finish()
127
content = temp_file.read()
130
"B+Tree Graph Index 2\nnode_ref_lists=1\nkey_elements=2\nlen=0\n"
134
def test_root_leaf_1_0(self):
135
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
136
nodes = self.make_nodes(5, 1, 0)
138
builder.add_node(*node)
139
# NamedTemporaryFile dies on builder.finish().read(). weird.
140
temp_file = builder.finish()
141
content = temp_file.read()
143
self.assertEqual(131, len(content))
145
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=5\n"
148
node_content = content[73:]
149
node_bytes = zlib.decompress(node_content)
150
expected_node = ("type=leaf\n"
151
"0000000000000000000000000000000000000000\x00\x00value:0\n"
152
"1111111111111111111111111111111111111111\x00\x00value:1\n"
153
"2222222222222222222222222222222222222222\x00\x00value:2\n"
154
"3333333333333333333333333333333333333333\x00\x00value:3\n"
155
"4444444444444444444444444444444444444444\x00\x00value:4\n")
156
self.assertEqual(expected_node, node_bytes)
158
def test_root_leaf_2_2(self):
159
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
160
nodes = self.make_nodes(5, 2, 2)
162
builder.add_node(*node)
163
# NamedTemporaryFile dies on builder.finish().read(). weird.
164
temp_file = builder.finish()
165
content = temp_file.read()
167
self.assertEqual(238, len(content))
169
"B+Tree Graph Index 2\nnode_ref_lists=2\nkey_elements=2\nlen=10\n"
172
node_content = content[74:]
173
node_bytes = zlib.decompress(node_content)
176
"0000000000000000000000000000000000000000\x000000000000000000000000000000000000000000\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:0\n"
177
"0000000000000000000000000000000000000000\x001111111111111111111111111111111111111111\x000000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\r0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:1\n"
178
"0000000000000000000000000000000000000000\x002222222222222222222222222222222222222222\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:2\n"
179
"0000000000000000000000000000000000000000\x003333333333333333333333333333333333333333\x000000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\t0000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\r0000000000000000000000000000000000000000\x00ref2222222222222222222222222222222222222222\x00value:3\n"
180
"0000000000000000000000000000000000000000\x004444444444444444444444444444444444444444\x00\t0000000000000000000000000000000000000000\x00ref0000000000000000000000000000000000000000\x00value:4\n"
181
"1111111111111111111111111111111111111111\x000000000000000000000000000000000000000000\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:0\n"
182
"1111111111111111111111111111111111111111\x001111111111111111111111111111111111111111\x001111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\r1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:1\n"
183
"1111111111111111111111111111111111111111\x002222222222222222222222222222222222222222\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:2\n"
184
"1111111111111111111111111111111111111111\x003333333333333333333333333333333333333333\x001111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\t1111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\r1111111111111111111111111111111111111111\x00ref2222222222222222222222222222222222222222\x00value:3\n"
185
"1111111111111111111111111111111111111111\x004444444444444444444444444444444444444444\x00\t1111111111111111111111111111111111111111\x00ref0000000000000000000000000000000000000000\x00value:4\n"
188
self.assertEqual(expected_node, node_bytes)
190
def test_2_leaves_1_0(self):
191
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
192
nodes = self.make_nodes(400, 1, 0)
194
builder.add_node(*node)
195
# NamedTemporaryFile dies on builder.finish().read(). weird.
196
temp_file = builder.finish()
197
content = temp_file.read()
199
self.assertEqual(9283, len(content))
201
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=400\n"
204
root = content[77:4096]
205
leaf1 = content[4096:8192]
206
leaf2 = content[8192:]
207
root_bytes = zlib.decompress(root)
211
) + ("307" * 40) + "\n"
212
self.assertEqual(expected_root, root_bytes)
213
# We already know serialisation works for leaves, check key selection:
214
leaf1_bytes = zlib.decompress(leaf1)
215
sorted_node_keys = sorted(node[0] for node in nodes)
216
node = btree_index._LeafNode(leaf1_bytes, 1, 0)
217
self.assertEqual(231, len(node))
218
self.assertEqual(sorted_node_keys[:231], node.all_keys())
219
leaf2_bytes = zlib.decompress(leaf2)
220
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
221
self.assertEqual(400 - 231, len(node))
222
self.assertEqual(sorted_node_keys[231:], node.all_keys())
224
def test_last_page_rounded_1_layer(self):
225
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
226
nodes = self.make_nodes(10, 1, 0)
228
builder.add_node(*node)
229
# NamedTemporaryFile dies on builder.finish().read(). weird.
230
temp_file = builder.finish()
231
content = temp_file.read()
233
self.assertEqual(155, len(content))
235
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=10\n"
238
# Check thelast page is well formed
240
leaf2_bytes = zlib.decompress(leaf2)
241
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
242
self.assertEqual(10, len(node))
243
sorted_node_keys = sorted(node[0] for node in nodes)
244
self.assertEqual(sorted_node_keys, node.all_keys())
246
def test_last_page_not_rounded_2_layer(self):
247
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
248
nodes = self.make_nodes(400, 1, 0)
250
builder.add_node(*node)
251
# NamedTemporaryFile dies on builder.finish().read(). weird.
252
temp_file = builder.finish()
253
content = temp_file.read()
255
self.assertEqual(9283, len(content))
257
"B+Tree Graph Index 2\nnode_ref_lists=0\nkey_elements=1\nlen=400\n"
260
# Check the last page is well formed
261
leaf2 = content[8192:]
262
leaf2_bytes = zlib.decompress(leaf2)
263
node = btree_index._LeafNode(leaf2_bytes, 1, 0)
264
self.assertEqual(400 - 231, len(node))
265
sorted_node_keys = sorted(node[0] for node in nodes)
266
self.assertEqual(sorted_node_keys[231:], node.all_keys())
268
def test_three_level_tree_details(self):
269
# The left most pointer in the second internal node in a row should
270
# pointer to the second node that the internal node is for, _not_
271
# the first, otherwise the first node overlaps with the last node of
272
# the prior internal node on that row.
273
self.shrink_page_size()
274
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
275
# 40K nodes is enough to create a two internal nodes on the second
276
# level, with a 2K page size
277
nodes = self.make_nodes(20000, 2, 2)
280
builder.add_node(*node)
281
t = transport.get_transport('trace+' + self.get_url(''))
282
size = t.put_file('index', self.time(builder.finish))
284
index = btree_index.BTreeGraphIndex(t, 'index', size)
285
# Seed the metadata, we're using internal calls now.
287
self.assertEqual(3, len(index._row_lengths),
288
"Not enough rows: %r" % index._row_lengths)
289
self.assertEqual(4, len(index._row_offsets))
290
self.assertEqual(sum(index._row_lengths), index._row_offsets[-1])
291
internal_nodes = index._get_internal_nodes([0, 1, 2])
292
root_node = internal_nodes[0]
293
internal_node1 = internal_nodes[1]
294
internal_node2 = internal_nodes[2]
295
# The left most node node2 points at should be one after the right most
296
# node pointed at by node1.
297
self.assertEqual(internal_node2.offset, 1 + len(internal_node1.keys))
298
# The left most key of the second node pointed at by internal_node2
299
# should be its first key. We can check this by looking for its first key
300
# in the second node it points at
301
pos = index._row_offsets[2] + internal_node2.offset + 1
302
leaf = index._get_leaf_nodes([pos])[pos]
303
self.assertTrue(internal_node2.keys[0] in leaf)
305
def test_2_leaves_2_2(self):
306
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
307
nodes = self.make_nodes(100, 2, 2)
309
builder.add_node(*node)
310
# NamedTemporaryFile dies on builder.finish().read(). weird.
311
temp_file = builder.finish()
312
content = temp_file.read()
314
self.assertEqual(12643, len(content))
316
"B+Tree Graph Index 2\nnode_ref_lists=2\nkey_elements=2\nlen=200\n"
319
root = content[77:4096]
320
leaf1 = content[4096:8192]
321
leaf2 = content[8192:12288]
322
leaf3 = content[12288:]
323
root_bytes = zlib.decompress(root)
327
+ ("0" * 40) + "\x00" + ("91" * 40) + "\n"
328
+ ("1" * 40) + "\x00" + ("81" * 40) + "\n"
330
self.assertEqual(expected_root, root_bytes)
331
# We assume the other leaf nodes have been written correctly - layering
334
def test_spill_index_stress_1_1(self):
335
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
336
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
337
builder.add_node(*nodes[0])
338
# Test the parts of the index that take up memory are doing so
340
self.assertEqual(1, len(builder._nodes))
341
self.assertIs(None, builder._nodes_by_key)
342
builder.add_node(*nodes[1])
343
self.assertEqual(0, len(builder._nodes))
344
self.assertIs(None, builder._nodes_by_key)
345
self.assertEqual(1, len(builder._backing_indices))
346
self.assertEqual(2, builder._backing_indices[0].key_count())
348
builder.add_node(*nodes[2])
349
self.assertEqual(1, len(builder._nodes))
350
self.assertIs(None, builder._nodes_by_key)
351
# And spills to a second backing index combing all
352
builder.add_node(*nodes[3])
353
self.assertEqual(0, len(builder._nodes))
354
self.assertIs(None, builder._nodes_by_key)
355
self.assertEqual(2, len(builder._backing_indices))
356
self.assertEqual(None, builder._backing_indices[0])
357
self.assertEqual(4, builder._backing_indices[1].key_count())
358
# The next spills to the 2-len slot
359
builder.add_node(*nodes[4])
360
builder.add_node(*nodes[5])
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(2, builder._backing_indices[0].key_count())
365
self.assertEqual(4, builder._backing_indices[1].key_count())
366
# Next spill combines
367
builder.add_node(*nodes[6])
368
builder.add_node(*nodes[7])
369
self.assertEqual(3, len(builder._backing_indices))
370
self.assertEqual(None, builder._backing_indices[0])
371
self.assertEqual(None, builder._backing_indices[1])
372
self.assertEqual(8, builder._backing_indices[2].key_count())
373
# And so forth - counting up in binary.
374
builder.add_node(*nodes[8])
375
builder.add_node(*nodes[9])
376
self.assertEqual(3, len(builder._backing_indices))
377
self.assertEqual(2, builder._backing_indices[0].key_count())
378
self.assertEqual(None, builder._backing_indices[1])
379
self.assertEqual(8, builder._backing_indices[2].key_count())
380
builder.add_node(*nodes[10])
381
builder.add_node(*nodes[11])
382
self.assertEqual(3, len(builder._backing_indices))
383
self.assertEqual(None, builder._backing_indices[0])
384
self.assertEqual(4, builder._backing_indices[1].key_count())
385
self.assertEqual(8, builder._backing_indices[2].key_count())
386
builder.add_node(*nodes[12])
387
# Test that memory and disk are both used for query methods; and that
388
# None is skipped over happily.
389
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
390
list(builder.iter_all_entries()))
391
# Two nodes - one memory one disk
392
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
393
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
394
self.assertEqual(13, builder.key_count())
395
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
396
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
397
builder.add_node(*nodes[13])
398
self.assertEqual(3, len(builder._backing_indices))
399
self.assertEqual(2, builder._backing_indices[0].key_count())
400
self.assertEqual(4, builder._backing_indices[1].key_count())
401
self.assertEqual(8, builder._backing_indices[2].key_count())
402
builder.add_node(*nodes[14])
403
builder.add_node(*nodes[15])
404
self.assertEqual(4, len(builder._backing_indices))
405
self.assertEqual(None, builder._backing_indices[0])
406
self.assertEqual(None, builder._backing_indices[1])
407
self.assertEqual(None, builder._backing_indices[2])
408
self.assertEqual(16, builder._backing_indices[3].key_count())
409
# Now finish, and check we got a correctly ordered tree
410
t = self.get_transport('')
411
size = t.put_file('index', builder.finish())
412
index = btree_index.BTreeGraphIndex(t, 'index', size)
413
nodes = list(index.iter_all_entries())
414
self.assertEqual(sorted(nodes), nodes)
415
self.assertEqual(16, len(nodes))
417
def test_spill_index_stress_1_1_no_combine(self):
418
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
419
builder.set_optimize(for_size=False, combine_backing_indices=False)
420
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
421
builder.add_node(*nodes[0])
422
# Test the parts of the index that take up memory are doing so
424
self.assertEqual(1, len(builder._nodes))
425
self.assertIs(None, builder._nodes_by_key)
426
builder.add_node(*nodes[1])
427
self.assertEqual(0, len(builder._nodes))
428
self.assertIs(None, builder._nodes_by_key)
429
self.assertEqual(1, len(builder._backing_indices))
430
self.assertEqual(2, builder._backing_indices[0].key_count())
432
builder.add_node(*nodes[2])
433
self.assertEqual(1, len(builder._nodes))
434
self.assertIs(None, builder._nodes_by_key)
435
# And spills to a second backing index but doesn't combine
436
builder.add_node(*nodes[3])
437
self.assertEqual(0, len(builder._nodes))
438
self.assertIs(None, builder._nodes_by_key)
439
self.assertEqual(2, len(builder._backing_indices))
440
for backing_index in builder._backing_indices:
441
self.assertEqual(2, backing_index.key_count())
442
# The next spills to the 3rd slot
443
builder.add_node(*nodes[4])
444
builder.add_node(*nodes[5])
445
self.assertEqual(0, len(builder._nodes))
446
self.assertIs(None, builder._nodes_by_key)
447
self.assertEqual(3, len(builder._backing_indices))
448
for backing_index in builder._backing_indices:
449
self.assertEqual(2, backing_index.key_count())
450
# Now spill a few more, and check that we don't combine
451
builder.add_node(*nodes[6])
452
builder.add_node(*nodes[7])
453
builder.add_node(*nodes[8])
454
builder.add_node(*nodes[9])
455
builder.add_node(*nodes[10])
456
builder.add_node(*nodes[11])
457
builder.add_node(*nodes[12])
458
self.assertEqual(6, len(builder._backing_indices))
459
for backing_index in builder._backing_indices:
460
self.assertEqual(2, backing_index.key_count())
461
# Test that memory and disk are both used for query methods; and that
462
# None is skipped over happily.
463
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
464
list(builder.iter_all_entries()))
465
# Two nodes - one memory one disk
466
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
467
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
468
self.assertEqual(13, builder.key_count())
469
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
470
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
471
builder.add_node(*nodes[13])
472
builder.add_node(*nodes[14])
473
builder.add_node(*nodes[15])
474
self.assertEqual(8, len(builder._backing_indices))
475
for backing_index in builder._backing_indices:
476
self.assertEqual(2, backing_index.key_count())
477
# Now finish, and check we got a correctly ordered tree
478
transport = self.get_transport('')
479
size = transport.put_file('index', builder.finish())
480
index = btree_index.BTreeGraphIndex(transport, 'index', size)
481
nodes = list(index.iter_all_entries())
482
self.assertEqual(sorted(nodes), nodes)
483
self.assertEqual(16, len(nodes))
485
def test_set_optimize(self):
486
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
487
builder.set_optimize(for_size=True)
488
self.assertTrue(builder._optimize_for_size)
489
builder.set_optimize(for_size=False)
490
self.assertFalse(builder._optimize_for_size)
491
# test that we can set combine_backing_indices without effecting
494
builder._optimize_for_size = obj
495
builder.set_optimize(combine_backing_indices=False)
496
self.assertFalse(builder._combine_backing_indices)
497
self.assertIs(obj, builder._optimize_for_size)
498
builder.set_optimize(combine_backing_indices=True)
499
self.assertTrue(builder._combine_backing_indices)
500
self.assertIs(obj, builder._optimize_for_size)
502
def test_spill_index_stress_2_2(self):
503
# test that references and longer keys don't confuse things.
504
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2,
506
nodes = self.make_nodes(16, 2, 2)
507
builder.add_node(*nodes[0])
508
# Test the parts of the index that take up memory are doing so
510
self.assertEqual(1, len(builder._nodes))
511
self.assertIs(None, builder._nodes_by_key)
512
builder.add_node(*nodes[1])
513
self.assertEqual(0, len(builder._nodes))
514
self.assertIs(None, builder._nodes_by_key)
515
self.assertEqual(1, len(builder._backing_indices))
516
self.assertEqual(2, builder._backing_indices[0].key_count())
518
old = dict(builder._get_nodes_by_key()) #Build up the nodes by key dict
519
builder.add_node(*nodes[2])
520
self.assertEqual(1, len(builder._nodes))
521
self.assertIsNot(None, builder._nodes_by_key)
522
self.assertNotEqual({}, builder._nodes_by_key)
523
# We should have a new entry
524
self.assertNotEqual(old, builder._nodes_by_key)
525
# And spills to a second backing index combing all
526
builder.add_node(*nodes[3])
527
self.assertEqual(0, len(builder._nodes))
528
self.assertIs(None, builder._nodes_by_key)
529
self.assertEqual(2, len(builder._backing_indices))
530
self.assertEqual(None, builder._backing_indices[0])
531
self.assertEqual(4, builder._backing_indices[1].key_count())
532
# The next spills to the 2-len slot
533
builder.add_node(*nodes[4])
534
builder.add_node(*nodes[5])
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(2, builder._backing_indices[0].key_count())
539
self.assertEqual(4, builder._backing_indices[1].key_count())
540
# Next spill combines
541
builder.add_node(*nodes[6])
542
builder.add_node(*nodes[7])
543
self.assertEqual(3, len(builder._backing_indices))
544
self.assertEqual(None, builder._backing_indices[0])
545
self.assertEqual(None, builder._backing_indices[1])
546
self.assertEqual(8, builder._backing_indices[2].key_count())
547
# And so forth - counting up in binary.
548
builder.add_node(*nodes[8])
549
builder.add_node(*nodes[9])
550
self.assertEqual(3, len(builder._backing_indices))
551
self.assertEqual(2, builder._backing_indices[0].key_count())
552
self.assertEqual(None, builder._backing_indices[1])
553
self.assertEqual(8, builder._backing_indices[2].key_count())
554
builder.add_node(*nodes[10])
555
builder.add_node(*nodes[11])
556
self.assertEqual(3, len(builder._backing_indices))
557
self.assertEqual(None, builder._backing_indices[0])
558
self.assertEqual(4, builder._backing_indices[1].key_count())
559
self.assertEqual(8, builder._backing_indices[2].key_count())
560
builder.add_node(*nodes[12])
561
# Test that memory and disk are both used for query methods; and that
562
# None is skipped over happily.
563
self.assertEqual([(builder,) + node for node in sorted(nodes[:13])],
564
list(builder.iter_all_entries()))
565
# Two nodes - one memory one disk
566
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
567
set(builder.iter_entries([nodes[12][0], nodes[11][0]])))
568
self.assertEqual(13, builder.key_count())
569
self.assertEqual(set([(builder,) + node for node in nodes[11:13]]),
570
set(builder.iter_entries_prefix([nodes[12][0], nodes[11][0]])))
571
builder.add_node(*nodes[13])
572
self.assertEqual(3, len(builder._backing_indices))
573
self.assertEqual(2, builder._backing_indices[0].key_count())
574
self.assertEqual(4, builder._backing_indices[1].key_count())
575
self.assertEqual(8, builder._backing_indices[2].key_count())
576
builder.add_node(*nodes[14])
577
builder.add_node(*nodes[15])
578
self.assertEqual(4, len(builder._backing_indices))
579
self.assertEqual(None, builder._backing_indices[0])
580
self.assertEqual(None, builder._backing_indices[1])
581
self.assertEqual(None, builder._backing_indices[2])
582
self.assertEqual(16, builder._backing_indices[3].key_count())
583
# Now finish, and check we got a correctly ordered tree
584
transport = self.get_transport('')
585
size = transport.put_file('index', builder.finish())
586
index = btree_index.BTreeGraphIndex(transport, 'index', size)
587
nodes = list(index.iter_all_entries())
588
self.assertEqual(sorted(nodes), nodes)
589
self.assertEqual(16, len(nodes))
591
def test_spill_index_duplicate_key_caught_on_finish(self):
592
builder = btree_index.BTreeBuilder(key_elements=1, spill_at=2)
593
nodes = [node[0:2] for node in self.make_nodes(16, 1, 0)]
594
builder.add_node(*nodes[0])
595
builder.add_node(*nodes[1])
596
builder.add_node(*nodes[0])
597
self.assertRaises(errors.BadIndexDuplicateKey, builder.finish)
600
class TestBTreeIndex(BTreeTestCase):
602
def make_index(self, ref_lists=0, key_elements=1, nodes=[]):
603
builder = btree_index.BTreeBuilder(reference_lists=ref_lists,
604
key_elements=key_elements)
605
for key, value, references in nodes:
606
builder.add_node(key, value, references)
607
stream = builder.finish()
608
trans = transport.get_transport('trace+' + self.get_url())
609
size = trans.put_file('index', stream)
610
return btree_index.BTreeGraphIndex(trans, 'index', size)
612
def make_index_with_offset(self, ref_lists=1, key_elements=1, nodes=[],
614
builder = btree_index.BTreeBuilder(key_elements=key_elements,
615
reference_lists=ref_lists)
616
builder.add_nodes(nodes)
617
transport = self.get_transport('')
618
# NamedTemporaryFile dies on builder.finish().read(). weird.
619
temp_file = builder.finish()
620
content = temp_file.read()
623
transport.put_bytes('index', (' '*offset)+content)
624
return btree_index.BTreeGraphIndex(transport, 'index', size=size,
627
def test_clear_cache(self):
628
nodes = self.make_nodes(160, 2, 2)
629
index = self.make_index(ref_lists=2, key_elements=2, nodes=nodes)
630
self.assertEqual(1, len(list(index.iter_entries([nodes[30][0]]))))
631
self.assertEqual([1, 4], index._row_lengths)
632
self.assertIsNot(None, index._root_node)
633
internal_node_pre_clear = index._internal_node_cache.keys()
634
self.assertTrue(len(index._leaf_node_cache) > 0)
636
# We don't touch _root_node or _internal_node_cache, both should be
637
# small, and can save a round trip or two
638
self.assertIsNot(None, index._root_node)
639
# NOTE: We don't want to affect the _internal_node_cache, as we expect
640
# it will be small, and if we ever do touch this index again, it
641
# will save round-trips. This assertion isn't very strong,
642
# becuase without a 3-level index, we don't have any internal
644
self.assertEqual(internal_node_pre_clear,
645
index._internal_node_cache.keys())
646
self.assertEqual(0, len(index._leaf_node_cache))
648
def test_trivial_constructor(self):
649
t = transport.get_transport('trace+' + self.get_url(''))
650
index = btree_index.BTreeGraphIndex(t, 'index', None)
651
# Checks the page size at load, but that isn't logged yet.
652
self.assertEqual([], t._activity)
654
def test_with_size_constructor(self):
655
t = transport.get_transport('trace+' + self.get_url(''))
656
index = btree_index.BTreeGraphIndex(t, 'index', 1)
657
# Checks the page size at load, but that isn't logged yet.
658
self.assertEqual([], t._activity)
660
def test_empty_key_count_no_size(self):
661
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
662
t = transport.get_transport('trace+' + self.get_url(''))
663
t.put_file('index', builder.finish())
664
index = btree_index.BTreeGraphIndex(t, 'index', None)
666
self.assertEqual([], t._activity)
667
self.assertEqual(0, index.key_count())
668
# The entire index should have been requested (as we generally have the
669
# size available, and doing many small readvs is inappropriate).
670
# We can't tell how much was actually read here, but - check the code.
671
self.assertEqual([('get', 'index')], t._activity)
673
def test_empty_key_count(self):
674
builder = btree_index.BTreeBuilder(key_elements=1, reference_lists=0)
675
t = transport.get_transport('trace+' + self.get_url(''))
676
size = t.put_file('index', builder.finish())
677
self.assertEqual(72, size)
678
index = btree_index.BTreeGraphIndex(t, 'index', size)
680
self.assertEqual([], t._activity)
681
self.assertEqual(0, index.key_count())
682
# The entire index should have been read, as 4K > size
683
self.assertEqual([('readv', 'index', [(0, 72)], False, None)],
686
def test_non_empty_key_count_2_2(self):
687
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
688
nodes = self.make_nodes(35, 2, 2)
690
builder.add_node(*node)
691
t = transport.get_transport('trace+' + self.get_url(''))
692
size = t.put_file('index', builder.finish())
693
index = btree_index.BTreeGraphIndex(t, 'index', size)
695
self.assertEqual([], t._activity)
696
self.assertEqual(70, index.key_count())
697
# The entire index should have been read, as it is one page long.
698
self.assertEqual([('readv', 'index', [(0, size)], False, None)],
700
self.assertEqual(1173, size)
702
def test_with_offset_no_size(self):
703
index = self.make_index_with_offset(key_elements=1, ref_lists=1,
705
nodes=self.make_nodes(200, 1, 1))
706
index._size = None # throw away the size info
707
self.assertEqual(200, index.key_count())
709
def test_with_small_offset(self):
710
index = self.make_index_with_offset(key_elements=1, ref_lists=1,
712
nodes=self.make_nodes(200, 1, 1))
713
self.assertEqual(200, index.key_count())
715
def test_with_large_offset(self):
716
index = self.make_index_with_offset(key_elements=1, ref_lists=1,
718
nodes=self.make_nodes(200, 1, 1))
719
self.assertEqual(200, index.key_count())
721
def test__read_nodes_no_size_one_page_reads_once(self):
722
self.make_index(nodes=[(('key',), 'value', ())])
723
trans = transport.get_transport('trace+' + self.get_url())
724
index = btree_index.BTreeGraphIndex(trans, 'index', None)
725
del trans._activity[:]
726
nodes = dict(index._read_nodes([0]))
727
self.assertEqual([0], nodes.keys())
729
self.assertEqual([('key',)], node.all_keys())
730
self.assertEqual([('get', 'index')], trans._activity)
732
def test__read_nodes_no_size_multiple_pages(self):
733
index = self.make_index(2, 2, nodes=self.make_nodes(160, 2, 2))
735
num_pages = index._row_offsets[-1]
736
# Reopen with a traced transport and no size
737
trans = transport.get_transport('trace+' + self.get_url())
738
index = btree_index.BTreeGraphIndex(trans, 'index', None)
739
del trans._activity[:]
740
nodes = dict(index._read_nodes([0]))
741
self.assertEqual(range(num_pages), nodes.keys())
743
def test_2_levels_key_count_2_2(self):
744
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
745
nodes = self.make_nodes(160, 2, 2)
747
builder.add_node(*node)
748
t = transport.get_transport('trace+' + self.get_url(''))
749
size = t.put_file('index', builder.finish())
750
self.assertEqual(17692, size)
751
index = btree_index.BTreeGraphIndex(t, 'index', size)
753
self.assertEqual([], t._activity)
754
self.assertEqual(320, index.key_count())
755
# The entire index should not have been read.
756
self.assertEqual([('readv', 'index', [(0, 4096)], False, None)],
759
def test_validate_one_page(self):
760
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
761
nodes = self.make_nodes(45, 2, 2)
763
builder.add_node(*node)
764
t = transport.get_transport('trace+' + self.get_url(''))
765
size = t.put_file('index', builder.finish())
766
index = btree_index.BTreeGraphIndex(t, 'index', size)
768
self.assertEqual([], t._activity)
770
# The entire index should have been read linearly.
771
self.assertEqual([('readv', 'index', [(0, size)], False, None)],
773
self.assertEqual(1488, size)
775
def test_validate_two_pages(self):
776
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
777
nodes = self.make_nodes(80, 2, 2)
779
builder.add_node(*node)
780
t = transport.get_transport('trace+' + self.get_url(''))
781
size = t.put_file('index', builder.finish())
782
# Root page, 2 leaf pages
783
self.assertEqual(9339, size)
784
index = btree_index.BTreeGraphIndex(t, 'index', size)
786
self.assertEqual([], t._activity)
788
# The entire index should have been read linearly.
790
[('readv', 'index', [(0, 4096)], False, None),
791
('readv', 'index', [(4096, 4096), (8192, 1147)], False, None)],
793
# XXX: TODO: write some badly-ordered nodes, and some pointers-to-wrong
794
# node and make validate find them.
796
def test_eq_ne(self):
797
# two indices are equal when constructed with the same parameters:
798
t1 = transport.get_transport('trace+' + self.get_url(''))
799
t2 = self.get_transport()
801
btree_index.BTreeGraphIndex(t1, 'index', None) ==
802
btree_index.BTreeGraphIndex(t1, 'index', None))
804
btree_index.BTreeGraphIndex(t1, 'index', 20) ==
805
btree_index.BTreeGraphIndex(t1, 'index', 20))
807
btree_index.BTreeGraphIndex(t1, 'index', 20) ==
808
btree_index.BTreeGraphIndex(t2, 'index', 20))
810
btree_index.BTreeGraphIndex(t1, 'inde1', 20) ==
811
btree_index.BTreeGraphIndex(t1, 'inde2', 20))
813
btree_index.BTreeGraphIndex(t1, 'index', 10) ==
814
btree_index.BTreeGraphIndex(t1, 'index', 20))
816
btree_index.BTreeGraphIndex(t1, 'index', None) !=
817
btree_index.BTreeGraphIndex(t1, 'index', None))
819
btree_index.BTreeGraphIndex(t1, 'index', 20) !=
820
btree_index.BTreeGraphIndex(t1, 'index', 20))
822
btree_index.BTreeGraphIndex(t1, 'index', 20) !=
823
btree_index.BTreeGraphIndex(t2, 'index', 20))
825
btree_index.BTreeGraphIndex(t1, 'inde1', 20) !=
826
btree_index.BTreeGraphIndex(t1, 'inde2', 20))
828
btree_index.BTreeGraphIndex(t1, 'index', 10) !=
829
btree_index.BTreeGraphIndex(t1, 'index', 20))
831
def test_iter_all_only_root_no_size(self):
832
self.make_index(nodes=[(('key',), 'value', ())])
833
t = transport.get_transport('trace+' + self.get_url(''))
834
index = btree_index.BTreeGraphIndex(t, 'index', None)
836
self.assertEqual([(('key',), 'value')],
837
[x[1:] for x in index.iter_all_entries()])
838
self.assertEqual([('get', 'index')], t._activity)
840
def test_iter_all_entries_reads(self):
841
# iterating all entries reads the header, then does a linear
843
self.shrink_page_size()
844
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
845
# 20k nodes is enough to create a two internal nodes on the second
846
# level, with a 2K page size
847
nodes = self.make_nodes(10000, 2, 2)
849
builder.add_node(*node)
850
t = transport.get_transport('trace+' + self.get_url(''))
851
size = t.put_file('index', builder.finish())
852
self.assertEqual(1303220, size, 'number of expected bytes in the'
854
page_size = btree_index._PAGE_SIZE
856
index = btree_index.BTreeGraphIndex(t, 'index', size)
858
self.assertEqual([], t._activity)
859
found_nodes = self.time(list, index.iter_all_entries())
861
for node in found_nodes:
862
self.assertTrue(node[0] is index)
863
bare_nodes.append(node[1:])
864
self.assertEqual(3, len(index._row_lengths),
865
"Not enough rows: %r" % index._row_lengths)
866
# Should be as long as the nodes we supplied
867
self.assertEqual(20000, len(found_nodes))
868
# Should have the same content
869
self.assertEqual(set(nodes), set(bare_nodes))
870
# Should have done linear scan IO up the index, ignoring
871
# the internal nodes:
872
# The entire index should have been read
873
total_pages = sum(index._row_lengths)
874
self.assertEqual(total_pages, index._row_offsets[-1])
875
self.assertEqual(1303220, size)
876
# The start of the leaves
877
first_byte = index._row_offsets[-2] * page_size
879
for offset in range(first_byte, size, page_size):
880
readv_request.append((offset, page_size))
881
# The last page is truncated
882
readv_request[-1] = (readv_request[-1][0], 1303220 % page_size)
883
expected = [('readv', 'index', [(0, page_size)], False, None),
884
('readv', 'index', readv_request, False, None)]
885
if expected != t._activity:
886
self.assertEqualDiff(pprint.pformat(expected),
887
pprint.pformat(transport._activity))
889
def _test_iter_entries_references_resolved(self):
890
index = self.make_index(1, nodes=[
891
(('name', ), 'data', ([('ref', ), ('ref', )], )),
892
(('ref', ), 'refdata', ([], ))])
893
self.assertEqual(set([(index, ('name', ), 'data', ((('ref',),('ref',)),)),
894
(index, ('ref', ), 'refdata', ((), ))]),
895
set(index.iter_entries([('name',), ('ref',)])))
897
def test_iter_entries_references_2_refs_resolved(self):
898
# iterating some entries reads just the pages needed. For now, to
899
# get it working and start measuring, only 4K pages are read.
900
builder = btree_index.BTreeBuilder(key_elements=2, reference_lists=2)
901
# 80 nodes is enough to create a two-level index.
902
nodes = self.make_nodes(160, 2, 2)
904
builder.add_node(*node)
905
t = transport.get_transport('trace+' + self.get_url(''))
906
size = t.put_file('index', builder.finish())
908
index = btree_index.BTreeGraphIndex(t, 'index', size)
910
self.assertEqual([], t._activity)
912
found_nodes = list(index.iter_entries([nodes[30][0]]))
914
for node in found_nodes:
915
self.assertTrue(node[0] is index)
916
bare_nodes.append(node[1:])
917
# Should be as long as the nodes we supplied
918
self.assertEqual(1, len(found_nodes))
919
# Should have the same content
920
self.assertEqual(nodes[30], bare_nodes[0])
921
# Should have read the root node, then one leaf page:
922
self.assertEqual([('readv', 'index', [(0, 4096)], False, None),
923
('readv', 'index', [(8192, 4096), ], False, None)],
926
def test_iter_key_prefix_1_element_key_None(self):
927
index = self.make_index()
928
self.assertRaises(errors.BadIndexKey, list,
929
index.iter_entries_prefix([(None, )]))
931
def test_iter_key_prefix_wrong_length(self):
932
index = self.make_index()
933
self.assertRaises(errors.BadIndexKey, list,
934
index.iter_entries_prefix([('foo', None)]))
935
index = self.make_index(key_elements=2)
936
self.assertRaises(errors.BadIndexKey, list,
937
index.iter_entries_prefix([('foo', )]))
938
self.assertRaises(errors.BadIndexKey, list,
939
index.iter_entries_prefix([('foo', None, None)]))
941
def test_iter_key_prefix_1_key_element_no_refs(self):
942
index = self.make_index( nodes=[
943
(('name', ), 'data', ()),
944
(('ref', ), 'refdata', ())])
945
self.assertEqual(set([(index, ('name', ), 'data'),
946
(index, ('ref', ), 'refdata')]),
947
set(index.iter_entries_prefix([('name', ), ('ref', )])))
949
def test_iter_key_prefix_1_key_element_refs(self):
950
index = self.make_index(1, nodes=[
951
(('name', ), 'data', ([('ref', )], )),
952
(('ref', ), 'refdata', ([], ))])
953
self.assertEqual(set([(index, ('name', ), 'data', ((('ref',),),)),
954
(index, ('ref', ), 'refdata', ((), ))]),
955
set(index.iter_entries_prefix([('name', ), ('ref', )])))
957
def test_iter_key_prefix_2_key_element_no_refs(self):
958
index = self.make_index(key_elements=2, nodes=[
959
(('name', 'fin1'), 'data', ()),
960
(('name', 'fin2'), 'beta', ()),
961
(('ref', 'erence'), 'refdata', ())])
962
self.assertEqual(set([(index, ('name', 'fin1'), 'data'),
963
(index, ('ref', 'erence'), 'refdata')]),
964
set(index.iter_entries_prefix([('name', 'fin1'), ('ref', 'erence')])))
965
self.assertEqual(set([(index, ('name', 'fin1'), 'data'),
966
(index, ('name', 'fin2'), 'beta')]),
967
set(index.iter_entries_prefix([('name', None)])))
969
def test_iter_key_prefix_2_key_element_refs(self):
970
index = self.make_index(1, key_elements=2, nodes=[
971
(('name', 'fin1'), 'data', ([('ref', 'erence')], )),
972
(('name', 'fin2'), 'beta', ([], )),
973
(('ref', 'erence'), 'refdata', ([], ))])
974
self.assertEqual(set([(index, ('name', 'fin1'), 'data', ((('ref', 'erence'),),)),
975
(index, ('ref', 'erence'), 'refdata', ((), ))]),
976
set(index.iter_entries_prefix([('name', 'fin1'), ('ref', 'erence')])))
977
self.assertEqual(set([(index, ('name', 'fin1'), 'data', ((('ref', 'erence'),),)),
978
(index, ('name', 'fin2'), 'beta', ((), ))]),
979
set(index.iter_entries_prefix([('name', None)])))
981
# XXX: external_references tests are duplicated in test_index. We
982
# probably should have per_graph_index tests...
983
def test_external_references_no_refs(self):
984
index = self.make_index(ref_lists=0, nodes=[])
985
self.assertRaises(ValueError, index.external_references, 0)
987
def test_external_references_no_results(self):
988
index = self.make_index(ref_lists=1, nodes=[
989
(('key',), 'value', ([],))])
990
self.assertEqual(set(), index.external_references(0))
992
def test_external_references_missing_ref(self):
993
missing_key = ('missing',)
994
index = self.make_index(ref_lists=1, nodes=[
995
(('key',), 'value', ([missing_key],))])
996
self.assertEqual(set([missing_key]), index.external_references(0))
998
def test_external_references_multiple_ref_lists(self):
999
missing_key = ('missing',)
1000
index = self.make_index(ref_lists=2, nodes=[
1001
(('key',), 'value', ([], [missing_key]))])
1002
self.assertEqual(set([]), index.external_references(0))
1003
self.assertEqual(set([missing_key]), index.external_references(1))
1005
def test_external_references_two_records(self):
1006
index = self.make_index(ref_lists=1, nodes=[
1007
(('key-1',), 'value', ([('key-2',)],)),
1008
(('key-2',), 'value', ([],)),
1010
self.assertEqual(set([]), index.external_references(0))
1012
def test__find_ancestors_one_page(self):
1015
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1016
(key1, 'value', ([key2],)),
1017
(key2, 'value', ([],)),
1020
missing_keys = set()
1021
search_keys = index._find_ancestors([key1], 0, parent_map, missing_keys)
1022
self.assertEqual({key1: (key2,), key2: ()}, parent_map)
1023
self.assertEqual(set(), missing_keys)
1024
self.assertEqual(set(), search_keys)
1026
def test__find_ancestors_one_page_w_missing(self):
1030
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1031
(key1, 'value', ([key2],)),
1032
(key2, 'value', ([],)),
1035
missing_keys = set()
1036
search_keys = index._find_ancestors([key2, key3], 0, parent_map,
1038
self.assertEqual({key2: ()}, parent_map)
1039
# we know that key3 is missing because we read the page that it would
1041
self.assertEqual(set([key3]), missing_keys)
1042
self.assertEqual(set(), search_keys)
1044
def test__find_ancestors_one_parent_missing(self):
1048
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1049
(key1, 'value', ([key2],)),
1050
(key2, 'value', ([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
# all we know is that key3 wasn't present on the page we were reading
1059
# but if you look, the last key is key2 which comes before key3, so we
1060
# don't know whether key3 would land on this page or not.
1061
self.assertEqual(set([key3]), search_keys)
1062
search_keys = index._find_ancestors(search_keys, 0, parent_map,
1064
# passing it back in, we are sure it is 'missing'
1065
self.assertEqual({key1: (key2,), key2: (key3,)}, parent_map)
1066
self.assertEqual(set([key3]), missing_keys)
1067
self.assertEqual(set([]), search_keys)
1069
def test__find_ancestors_dont_search_known(self):
1073
index = self.make_index(ref_lists=1, key_elements=1, nodes=[
1074
(key1, 'value', ([key2],)),
1075
(key2, 'value', ([key3],)),
1076
(key3, 'value', ([],)),
1078
# We already know about key2, so we won't try to search for key3
1079
parent_map = {key2: (key3,)}
1080
missing_keys = set()
1081
search_keys = index._find_ancestors([key1], 0, parent_map,
1083
self.assertEqual({key1: (key2,), key2: (key3,)}, parent_map)
1084
self.assertEqual(set(), missing_keys)
1085
self.assertEqual(set(), search_keys)
1087
def test__find_ancestors_multiple_pages(self):
1088
# We need to use enough keys that we actually cause a split
1089
start_time = 1249671539
1090
email = "joebob@example.com"
1094
for i in xrange(400):
1095
rev_id = '%s-%s-%s' % (email,
1096
osutils.compact_date(start_time + i),
1097
osutils.rand_chars(16))
1099
nodes.append((rev_key, 'value', ref_lists))
1100
# We have a ref 'list' of length 1, with a list of parents, with 1
1101
# parent which is a key
1102
ref_lists = ((rev_key,),)
1103
rev_keys.append(rev_key)
1104
index = self.make_index(ref_lists=1, key_elements=1, nodes=nodes)
1105
self.assertEqual(400, index.key_count())
1106
self.assertEqual(3, len(index._row_offsets))
1107
nodes = dict(index._read_nodes([1, 2]))
1110
min_l2_key = l2.min_key
1111
max_l1_key = l1.max_key
1112
self.assertTrue(max_l1_key < min_l2_key)
1113
parents_min_l2_key = l2[min_l2_key][1][0]
1114
self.assertEqual((l1.max_key,), parents_min_l2_key)
1115
# Now, whatever key we select that would fall on the second page,
1116
# should give us all the parents until the page break
1117
key_idx = rev_keys.index(min_l2_key)
1118
next_key = rev_keys[key_idx+1]
1119
# So now when we get the parent map, we should get the key we are
1120
# looking for, min_l2_key, and then a reference to go look for the
1121
# parent of that key
1123
missing_keys = set()
1124
search_keys = index._find_ancestors([next_key], 0, parent_map,
1126
self.assertEqual([min_l2_key, next_key], sorted(parent_map))
1127
self.assertEqual(set(), missing_keys)
1128
self.assertEqual(set([max_l1_key]), search_keys)
1130
search_keys = index._find_ancestors([max_l1_key], 0, parent_map,
1132
self.assertEqual(l1.all_keys(), sorted(parent_map))
1133
self.assertEqual(set(), missing_keys)
1134
self.assertEqual(set(), search_keys)
1136
def test__find_ancestors_empty_index(self):
1137
index = self.make_index(ref_lists=1, key_elements=1, nodes=[])
1139
missing_keys = set()
1140
search_keys = index._find_ancestors([('one',), ('two',)], 0, parent_map,
1142
self.assertEqual(set(), search_keys)
1143
self.assertEqual({}, parent_map)
1144
self.assertEqual(set([('one',), ('two',)]), missing_keys)
1146
def test_supports_unlimited_cache(self):
1147
builder = btree_index.BTreeBuilder(reference_lists=0, key_elements=1)
1148
# We need enough nodes to cause a page split (so we have both an
1149
# internal node and a couple leaf nodes. 500 seems to be enough.)
1150
nodes = self.make_nodes(500, 1, 0)
1152
builder.add_node(*node)
1153
stream = builder.finish()
1154
trans = self.get_transport()
1155
size = trans.put_file('index', stream)
1156
index = btree_index.BTreeGraphIndex(trans, 'index', size)
1157
self.assertEqual(500, index.key_count())
1158
# We have an internal node
1159
self.assertEqual(2, len(index._row_lengths))
1160
# We have at least 2 leaf nodes
1161
self.assertTrue(index._row_lengths[-1] >= 2)
1162
self.assertIsInstance(index._leaf_node_cache, lru_cache.LRUCache)
1163
self.assertEqual(btree_index._NODE_CACHE_SIZE,
1164
index._leaf_node_cache._max_cache)
1165
self.assertIsInstance(index._internal_node_cache, fifo_cache.FIFOCache)
1166
self.assertEqual(100, index._internal_node_cache._max_cache)
1167
# No change if unlimited_cache=False is passed
1168
index = btree_index.BTreeGraphIndex(trans, 'index', size,
1169
unlimited_cache=False)
1170
self.assertIsInstance(index._leaf_node_cache, lru_cache.LRUCache)
1171
self.assertEqual(btree_index._NODE_CACHE_SIZE,
1172
index._leaf_node_cache._max_cache)
1173
self.assertIsInstance(index._internal_node_cache, fifo_cache.FIFOCache)
1174
self.assertEqual(100, index._internal_node_cache._max_cache)
1175
index = btree_index.BTreeGraphIndex(trans, 'index', size,
1176
unlimited_cache=True)
1177
self.assertIsInstance(index._leaf_node_cache, dict)
1178
self.assertIs(type(index._internal_node_cache), dict)
1179
# Exercise the lookup code
1180
entries = set(index.iter_entries([n[0] for n in nodes]))
1181
self.assertEqual(500, len(entries))
1184
class TestBTreeNodes(BTreeTestCase):
1186
scenarios = btreeparser_scenarios()
1189
BTreeTestCase.setUp(self)
1190
self.overrideAttr(btree_index, '_btree_serializer', self.parse_btree)
1192
def test_LeafNode_1_0(self):
1193
node_bytes = ("type=leaf\n"
1194
"0000000000000000000000000000000000000000\x00\x00value:0\n"
1195
"1111111111111111111111111111111111111111\x00\x00value:1\n"
1196
"2222222222222222222222222222222222222222\x00\x00value:2\n"
1197
"3333333333333333333333333333333333333333\x00\x00value:3\n"
1198
"4444444444444444444444444444444444444444\x00\x00value:4\n")
1199
node = btree_index._LeafNode(node_bytes, 1, 0)
1200
# We do direct access, or don't care about order, to leaf nodes most of
1201
# the time, so a dict is useful:
1203
("0000000000000000000000000000000000000000",): ("value:0", ()),
1204
("1111111111111111111111111111111111111111",): ("value:1", ()),
1205
("2222222222222222222222222222222222222222",): ("value:2", ()),
1206
("3333333333333333333333333333333333333333",): ("value:3", ()),
1207
("4444444444444444444444444444444444444444",): ("value:4", ()),
1208
}, dict(node.all_items()))
1210
def test_LeafNode_2_2(self):
1211
node_bytes = ("type=leaf\n"
1212
"00\x0000\x00\t00\x00ref00\x00value:0\n"
1213
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n"
1214
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n"
1215
"11\x0044\x00\t11\x00ref00\x00value:4\n"
1218
node = btree_index._LeafNode(node_bytes, 2, 2)
1219
# We do direct access, or don't care about order, to leaf nodes most of
1220
# the time, so a dict is useful:
1222
('00', '00'): ('value:0', ((), (('00', 'ref00'),))),
1223
('00', '11'): ('value:1',
1224
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01')))),
1225
('11', '33'): ('value:3',
1226
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22')))),
1227
('11', '44'): ('value:4', ((), (('11', 'ref00'),)))
1228
}, dict(node.all_items()))
1230
def test_InternalNode_1(self):
1231
node_bytes = ("type=internal\n"
1233
"0000000000000000000000000000000000000000\n"
1234
"1111111111111111111111111111111111111111\n"
1235
"2222222222222222222222222222222222222222\n"
1236
"3333333333333333333333333333333333333333\n"
1237
"4444444444444444444444444444444444444444\n"
1239
node = btree_index._InternalNode(node_bytes)
1240
# We want to bisect to find the right children from this node, so a
1241
# vector is most useful.
1243
("0000000000000000000000000000000000000000",),
1244
("1111111111111111111111111111111111111111",),
1245
("2222222222222222222222222222222222222222",),
1246
("3333333333333333333333333333333333333333",),
1247
("4444444444444444444444444444444444444444",),
1249
self.assertEqual(1, node.offset)
1251
def test_LeafNode_2_2(self):
1252
node_bytes = ("type=leaf\n"
1253
"00\x0000\x00\t00\x00ref00\x00value:0\n"
1254
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n"
1255
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n"
1256
"11\x0044\x00\t11\x00ref00\x00value:4\n"
1259
node = btree_index._LeafNode(node_bytes, 2, 2)
1260
# We do direct access, or don't care about order, to leaf nodes most of
1261
# the time, so a dict is useful:
1263
('00', '00'): ('value:0', ((), (('00', 'ref00'),))),
1264
('00', '11'): ('value:1',
1265
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01')))),
1266
('11', '33'): ('value:3',
1267
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22')))),
1268
('11', '44'): ('value:4', ((), (('11', 'ref00'),)))
1269
}, dict(node.all_items()))
1271
def assertFlattened(self, expected, key, value, refs):
1272
flat_key, flat_line = self.parse_btree._flatten_node(
1273
(None, key, value, refs), bool(refs))
1274
self.assertEqual('\x00'.join(key), flat_key)
1275
self.assertEqual(expected, flat_line)
1277
def test__flatten_node(self):
1278
self.assertFlattened('key\0\0value\n', ('key',), 'value', [])
1279
self.assertFlattened('key\0tuple\0\0value str\n',
1280
('key', 'tuple'), 'value str', [])
1281
self.assertFlattened('key\0tuple\0triple\0\0value str\n',
1282
('key', 'tuple', 'triple'), 'value str', [])
1283
self.assertFlattened('k\0t\0s\0ref\0value str\n',
1284
('k', 't', 's'), 'value str', [[('ref',)]])
1285
self.assertFlattened('key\0tuple\0ref\0key\0value str\n',
1286
('key', 'tuple'), 'value str', [[('ref', 'key')]])
1287
self.assertFlattened("00\x0000\x00\t00\x00ref00\x00value:0\n",
1288
('00', '00'), 'value:0', ((), (('00', 'ref00'),)))
1289
self.assertFlattened(
1290
"00\x0011\x0000\x00ref00\t00\x00ref00\r01\x00ref01\x00value:1\n",
1291
('00', '11'), 'value:1',
1292
((('00', 'ref00'),), (('00', 'ref00'), ('01', 'ref01'))))
1293
self.assertFlattened(
1294
"11\x0033\x0011\x00ref22\t11\x00ref22\r11\x00ref22\x00value:3\n",
1295
('11', '33'), 'value:3',
1296
((('11', 'ref22'),), (('11', 'ref22'), ('11', 'ref22'))))
1297
self.assertFlattened(
1298
"11\x0044\x00\t11\x00ref00\x00value:4\n",
1299
('11', '44'), 'value:4', ((), (('11', 'ref00'),)))
1302
class TestCompiledBtree(tests.TestCase):
1304
def test_exists(self):
1305
# This is just to let the user know if they don't have the feature
1307
self.requireFeature(compiled_btreeparser_feature)
1310
class TestMultiBisectRight(tests.TestCase):
1312
def assertMultiBisectRight(self, offsets, search_keys, fixed_keys):
1313
self.assertEqual(offsets,
1314
btree_index.BTreeGraphIndex._multi_bisect_right(
1315
search_keys, fixed_keys))
1317
def test_after(self):
1318
self.assertMultiBisectRight([(1, ['b'])], ['b'], ['a'])
1319
self.assertMultiBisectRight([(3, ['e', 'f', 'g'])],
1320
['e', 'f', 'g'], ['a', 'b', 'c'])
1322
def test_before(self):
1323
self.assertMultiBisectRight([(0, ['a'])], ['a'], ['b'])
1324
self.assertMultiBisectRight([(0, ['a', 'b', 'c', 'd'])],
1325
['a', 'b', 'c', 'd'], ['e', 'f', 'g'])
1327
def test_exact(self):
1328
self.assertMultiBisectRight([(1, ['a'])], ['a'], ['a'])
1329
self.assertMultiBisectRight([(1, ['a']), (2, ['b'])], ['a', 'b'], ['a', 'b'])
1330
self.assertMultiBisectRight([(1, ['a']), (3, ['c'])],
1331
['a', 'c'], ['a', 'b', 'c'])
1333
def test_inbetween(self):
1334
self.assertMultiBisectRight([(1, ['b'])], ['b'], ['a', 'c'])
1335
self.assertMultiBisectRight([(1, ['b', 'c', 'd']), (2, ['f', 'g'])],
1336
['b', 'c', 'd', 'f', 'g'], ['a', 'e', 'h'])
1338
def test_mixed(self):
1339
self.assertMultiBisectRight([(0, ['a', 'b']), (2, ['d', 'e']),
1341
['a', 'b', 'd', 'e', 'g', 'h'],
1342
['c', 'd', 'f', 'g'])
1345
class TestExpandOffsets(tests.TestCase):
1347
def make_index(self, size, recommended_pages=None):
1348
"""Make an index with a generic size.
1350
This doesn't actually create anything on disk, it just primes a
1351
BTreeGraphIndex with the recommended information.
1353
index = btree_index.BTreeGraphIndex(
1354
transport.get_transport('memory:///'), 'test-index', size=size)
1355
if recommended_pages is not None:
1356
index._recommended_pages = recommended_pages
1359
def set_cached_offsets(self, index, cached_offsets):
1360
"""Monkeypatch to give a canned answer for _get_offsets_for...()."""
1361
def _get_offsets_to_cached_pages():
1362
cached = set(cached_offsets)
1364
index._get_offsets_to_cached_pages = _get_offsets_to_cached_pages
1366
def prepare_index(self, index, node_ref_lists, key_length, key_count,
1367
row_lengths, cached_offsets):
1368
"""Setup the BTreeGraphIndex with some pre-canned information."""
1369
index.node_ref_lists = node_ref_lists
1370
index._key_length = key_length
1371
index._key_count = key_count
1372
index._row_lengths = row_lengths
1373
index._compute_row_offsets()
1374
index._root_node = btree_index._InternalNode('internal\noffset=0\n')
1375
self.set_cached_offsets(index, cached_offsets)
1377
def make_100_node_index(self):
1378
index = self.make_index(4096*100, 6)
1379
# Consider we've already made a single request at the middle
1380
self.prepare_index(index, node_ref_lists=0, key_length=1,
1381
key_count=1000, row_lengths=[1, 99],
1382
cached_offsets=[0, 50])
1385
def make_1000_node_index(self):
1386
index = self.make_index(4096*1000, 6)
1387
# Pretend we've already made a single request in the middle
1388
self.prepare_index(index, node_ref_lists=0, key_length=1,
1389
key_count=90000, row_lengths=[1, 9, 990],
1390
cached_offsets=[0, 5, 500])
1393
def assertNumPages(self, expected_pages, index, size):
1395
self.assertEqual(expected_pages, index._compute_total_pages_in_index())
1397
def assertExpandOffsets(self, expected, index, offsets):
1398
self.assertEqual(expected, index._expand_offsets(offsets),
1399
'We did not get the expected value after expanding'
1402
def test_default_recommended_pages(self):
1403
index = self.make_index(None)
1404
# local transport recommends 4096 byte reads, which is 1 page
1405
self.assertEqual(1, index._recommended_pages)
1407
def test__compute_total_pages_in_index(self):
1408
index = self.make_index(None)
1409
self.assertNumPages(1, index, 1024)
1410
self.assertNumPages(1, index, 4095)
1411
self.assertNumPages(1, index, 4096)
1412
self.assertNumPages(2, index, 4097)
1413
self.assertNumPages(2, index, 8192)
1414
self.assertNumPages(76, index, 4096*75 + 10)
1416
def test__find_layer_start_and_stop(self):
1417
index = self.make_1000_node_index()
1418
self.assertEqual((0, 1), index._find_layer_first_and_end(0))
1419
self.assertEqual((1, 10), index._find_layer_first_and_end(1))
1420
self.assertEqual((1, 10), index._find_layer_first_and_end(9))
1421
self.assertEqual((10, 1000), index._find_layer_first_and_end(10))
1422
self.assertEqual((10, 1000), index._find_layer_first_and_end(99))
1423
self.assertEqual((10, 1000), index._find_layer_first_and_end(999))
1425
def test_unknown_size(self):
1426
# We should not expand if we don't know the file size
1427
index = self.make_index(None, 10)
1428
self.assertExpandOffsets([0], index, [0])
1429
self.assertExpandOffsets([1, 4, 9], index, [1, 4, 9])
1431
def test_more_than_recommended(self):
1432
index = self.make_index(4096*100, 2)
1433
self.assertExpandOffsets([1, 10], index, [1, 10])
1434
self.assertExpandOffsets([1, 10, 20], index, [1, 10, 20])
1436
def test_read_all_from_root(self):
1437
index = self.make_index(4096*10, 20)
1438
self.assertExpandOffsets(range(10), index, [0])
1440
def test_read_all_when_cached(self):
1441
# We've read enough that we can grab all the rest in a single request
1442
index = self.make_index(4096*10, 5)
1443
self.prepare_index(index, node_ref_lists=0, key_length=1,
1444
key_count=1000, row_lengths=[1, 9],
1445
cached_offsets=[0, 1, 2, 5, 6])
1446
# It should fill the remaining nodes, regardless of the one requested
1447
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [3])
1448
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [8])
1449
self.assertExpandOffsets([3, 4, 7, 8, 9], index, [9])
1451
def test_no_root_node(self):
1452
index = self.make_index(4096*10, 5)
1453
self.assertExpandOffsets([0], index, [0])
1455
def test_include_neighbors(self):
1456
index = self.make_100_node_index()
1457
# We expand in both directions, until we have at least 'recommended'
1459
self.assertExpandOffsets([9, 10, 11, 12, 13, 14, 15], index, [12])
1460
self.assertExpandOffsets([88, 89, 90, 91, 92, 93, 94], index, [91])
1461
# If we hit an 'edge' we continue in the other direction
1462
self.assertExpandOffsets([1, 2, 3, 4, 5, 6], index, [2])
1463
self.assertExpandOffsets([94, 95, 96, 97, 98, 99], index, [98])
1465
# Requesting many nodes will expand all locations equally
1466
self.assertExpandOffsets([1, 2, 3, 80, 81, 82], index, [2, 81])
1467
self.assertExpandOffsets([1, 2, 3, 9, 10, 11, 80, 81, 82], index,
1470
def test_stop_at_cached(self):
1471
index = self.make_100_node_index()
1472
self.set_cached_offsets(index, [0, 10, 19])
1473
self.assertExpandOffsets([11, 12, 13, 14, 15, 16], index, [11])
1474
self.assertExpandOffsets([11, 12, 13, 14, 15, 16], index, [12])
1475
self.assertExpandOffsets([12, 13, 14, 15, 16, 17, 18], index, [15])
1476
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [16])
1477
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [17])
1478
self.assertExpandOffsets([13, 14, 15, 16, 17, 18], index, [18])
1480
def test_cannot_fully_expand(self):
1481
index = self.make_100_node_index()
1482
self.set_cached_offsets(index, [0, 10, 12])
1483
# We don't go into an endless loop if we are bound by cached nodes
1484
self.assertExpandOffsets([11], index, [11])
1486
def test_overlap(self):
1487
index = self.make_100_node_index()
1488
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [12, 13])
1489
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [11, 14])
1491
def test_stay_within_layer(self):
1492
index = self.make_1000_node_index()
1493
# When expanding a request, we won't read nodes from the next layer
1494
self.assertExpandOffsets([1, 2, 3, 4], index, [2])
1495
self.assertExpandOffsets([6, 7, 8, 9], index, [6])
1496
self.assertExpandOffsets([6, 7, 8, 9], index, [9])
1497
self.assertExpandOffsets([10, 11, 12, 13, 14, 15], index, [10])
1498
self.assertExpandOffsets([10, 11, 12, 13, 14, 15, 16], index, [13])
1500
self.set_cached_offsets(index, [0, 4, 12])
1501
self.assertExpandOffsets([5, 6, 7, 8, 9], index, [7])
1502
self.assertExpandOffsets([10, 11], index, [11])
1504
def test_small_requests_unexpanded(self):
1505
index = self.make_100_node_index()
1506
self.set_cached_offsets(index, [0])
1507
self.assertExpandOffsets([1], index, [1])
1508
self.assertExpandOffsets([50], index, [50])
1509
# If we request more than one node, then we'll expand
1510
self.assertExpandOffsets([49, 50, 51, 59, 60, 61], index, [50, 60])
1512
# The first pass does not expand
1513
index = self.make_1000_node_index()
1514
self.set_cached_offsets(index, [0])
1515
self.assertExpandOffsets([1], index, [1])
1516
self.set_cached_offsets(index, [0, 1])
1517
self.assertExpandOffsets([100], index, [100])
1518
self.set_cached_offsets(index, [0, 1, 100])
1519
# But after the first depth, we will expand
1520
self.assertExpandOffsets([2, 3, 4, 5, 6, 7], index, [2])
1521
self.assertExpandOffsets([2, 3, 4, 5, 6, 7], index, [4])
1522
self.set_cached_offsets(index, [0, 1, 2, 3, 4, 5, 6, 7, 100])
1523
self.assertExpandOffsets([102, 103, 104, 105, 106, 107, 108], index,
added
removed
bzrlib/tests/test_btree_index.py
