1
# Copyright (C) 2008 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
20
from bisect import bisect_right
35
from bzrlib.index import _OPTION_NODE_REFS, _OPTION_KEY_ELEMENTS, _OPTION_LEN
36
from bzrlib.transport import get_transport
39
_BTSIGNATURE = "B+Tree Graph Index 2\n"
40
_OPTION_ROW_LENGTHS = "row_lengths="
41
_LEAF_FLAG = "type=leaf\n"
42
_INTERNAL_FLAG = "type=internal\n"
43
_INTERNAL_OFFSET = "offset="
45
_RESERVED_HEADER_BYTES = 120
48
# 4K per page: 4MB - 1000 entries
49
_NODE_CACHE_SIZE = 1000
52
class _BuilderRow(object):
53
"""The stored state accumulated while writing out a row in the index.
55
:ivar spool: A temporary file used to accumulate nodes for this row
57
:ivar nodes: The count of nodes emitted so far.
61
"""Create a _BuilderRow."""
63
self.spool = tempfile.TemporaryFile()
66
def finish_node(self, pad=True):
67
byte_lines, _, padding = self.writer.finish()
70
self.spool.write("\x00" * _RESERVED_HEADER_BYTES)
72
if not pad and padding:
74
skipped_bytes = padding
75
self.spool.writelines(byte_lines)
76
remainder = (self.spool.tell() + skipped_bytes) % _PAGE_SIZE
78
raise AssertionError("incorrect node length: %d, %d"
79
% (self.spool.tell(), remainder))
84
class _InternalBuilderRow(_BuilderRow):
85
"""The stored state accumulated while writing out internal rows."""
87
def finish_node(self, pad=True):
89
raise AssertionError("Must pad internal nodes only.")
90
_BuilderRow.finish_node(self)
93
class _LeafBuilderRow(_BuilderRow):
94
"""The stored state accumulated while writing out a leaf rows."""
97
class BTreeBuilder(index.GraphIndexBuilder):
98
"""A Builder for B+Tree based Graph indices.
100
The resulting graph has the structure:
102
_SIGNATURE OPTIONS NODES
103
_SIGNATURE := 'B+Tree Graph Index 1' NEWLINE
104
OPTIONS := REF_LISTS KEY_ELEMENTS LENGTH
105
REF_LISTS := 'node_ref_lists=' DIGITS NEWLINE
106
KEY_ELEMENTS := 'key_elements=' DIGITS NEWLINE
107
LENGTH := 'len=' DIGITS NEWLINE
108
ROW_LENGTHS := 'row_lengths' DIGITS (COMMA DIGITS)*
109
NODES := NODE_COMPRESSED*
110
NODE_COMPRESSED:= COMPRESSED_BYTES{4096}
111
NODE_RAW := INTERNAL | LEAF
112
INTERNAL := INTERNAL_FLAG POINTERS
113
LEAF := LEAF_FLAG ROWS
114
KEY_ELEMENT := Not-whitespace-utf8
115
KEY := KEY_ELEMENT (NULL KEY_ELEMENT)*
117
ROW := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
119
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
120
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
122
VALUE := no-newline-no-null-bytes
125
def __init__(self, reference_lists=0, key_elements=1, spill_at=100000):
126
"""See GraphIndexBuilder.__init__.
128
:param spill_at: Optional parameter controlling the maximum number
129
of nodes that BTreeBuilder will hold in memory.
131
index.GraphIndexBuilder.__init__(self, reference_lists=reference_lists,
132
key_elements=key_elements)
133
self._spill_at = spill_at
134
self._backing_indices = []
135
# A map of {key: (node_refs, value)}
137
# Indicate it hasn't been built yet
138
self._nodes_by_key = None
139
self._optimize_for_size = False
141
def add_node(self, key, value, references=()):
142
"""Add a node to the index.
144
If adding the node causes the builder to reach its spill_at threshold,
145
disk spilling will be triggered.
147
:param key: The key. keys are non-empty tuples containing
148
as many whitespace-free utf8 bytestrings as the key length
149
defined for this index.
150
:param references: An iterable of iterables of keys. Each is a
151
reference to another key.
152
:param value: The value to associate with the key. It may be any
153
bytes as long as it does not contain \0 or \n.
155
# we don't care about absent_references
156
node_refs, _ = self._check_key_ref_value(key, references, value)
157
if key in self._nodes:
158
raise errors.BadIndexDuplicateKey(key, self)
159
self._nodes[key] = (node_refs, value)
161
if self._nodes_by_key is not None and self._key_length > 1:
162
self._update_nodes_by_key(key, value, node_refs)
163
if len(self._keys) < self._spill_at:
165
self._spill_mem_keys_to_disk()
167
def _spill_mem_keys_to_disk(self):
168
"""Write the in memory keys down to disk to cap memory consumption.
170
If we already have some keys written to disk, we will combine them so
171
as to preserve the sorted order. The algorithm for combining uses
172
powers of two. So on the first spill, write all mem nodes into a
173
single index. On the second spill, combine the mem nodes with the nodes
174
on disk to create a 2x sized disk index and get rid of the first index.
175
On the third spill, create a single new disk index, which will contain
176
the mem nodes, and preserve the existing 2x sized index. On the fourth,
177
combine mem with the first and second indexes, creating a new one of
178
size 4x. On the fifth create a single new one, etc.
180
if self._combine_backing_indices:
181
(new_backing_file, size,
182
backing_pos) = self._spill_mem_keys_and_combine()
184
new_backing_file, size = self._spill_mem_keys_without_combining()
185
dir_path, base_name = osutils.split(new_backing_file.name)
186
# Note: The transport here isn't strictly needed, because we will use
187
# direct access to the new_backing._file object
188
new_backing = BTreeGraphIndex(get_transport(dir_path),
190
# GC will clean up the file
191
new_backing._file = new_backing_file
192
if self._combine_backing_indices:
193
if len(self._backing_indices) == backing_pos:
194
self._backing_indices.append(None)
195
self._backing_indices[backing_pos] = new_backing
196
for backing_pos in range(backing_pos):
197
self._backing_indices[backing_pos] = None
199
self._backing_indices.append(new_backing)
202
self._nodes_by_key = None
204
def _spill_mem_keys_without_combining(self):
205
return self._write_nodes(self._iter_mem_nodes(), allow_optimize=False)
207
def _spill_mem_keys_and_combine(self):
208
iterators_to_combine = [self._iter_mem_nodes()]
210
for pos, backing in enumerate(self._backing_indices):
214
iterators_to_combine.append(backing.iter_all_entries())
215
backing_pos = pos + 1
216
new_backing_file, size = \
217
self._write_nodes(self._iter_smallest(iterators_to_combine),
218
allow_optimize=False)
219
return new_backing_file, size, backing_pos
221
def add_nodes(self, nodes):
222
"""Add nodes to the index.
224
:param nodes: An iterable of (key, node_refs, value) entries to add.
226
if self.reference_lists:
227
for (key, value, node_refs) in nodes:
228
self.add_node(key, value, node_refs)
230
for (key, value) in nodes:
231
self.add_node(key, value)
233
def _iter_mem_nodes(self):
234
"""Iterate over the nodes held in memory."""
236
if self.reference_lists:
237
for key in sorted(nodes):
238
references, value = nodes[key]
239
yield self, key, value, references
241
for key in sorted(nodes):
242
references, value = nodes[key]
243
yield self, key, value
245
def _iter_smallest(self, iterators_to_combine):
246
if len(iterators_to_combine) == 1:
247
for value in iterators_to_combine[0]:
251
for iterator in iterators_to_combine:
253
current_values.append(iterator.next())
254
except StopIteration:
255
current_values.append(None)
258
# Decorate candidates with the value to allow 2.4's min to be used.
259
candidates = [(item[1][1], item) for item
260
in enumerate(current_values) if item[1] is not None]
261
if not len(candidates):
263
selected = min(candidates)
264
# undecorate back to (pos, node)
265
selected = selected[1]
266
if last == selected[1][1]:
267
raise errors.BadIndexDuplicateKey(last, self)
268
last = selected[1][1]
269
# Yield, with self as the index
270
yield (self,) + selected[1][1:]
273
current_values[pos] = iterators_to_combine[pos].next()
274
except StopIteration:
275
current_values[pos] = None
277
def _add_key(self, string_key, line, rows, allow_optimize=True):
278
"""Add a key to the current chunk.
280
:param string_key: The key to add.
281
:param line: The fully serialised key and value.
282
:param allow_optimize: If set to False, prevent setting the optimize
283
flag when writing out. This is used by the _spill_mem_keys_to_disk
286
if rows[-1].writer is None:
287
# opening a new leaf chunk;
288
for pos, internal_row in enumerate(rows[:-1]):
289
# flesh out any internal nodes that are needed to
290
# preserve the height of the tree
291
if internal_row.writer is None:
293
if internal_row.nodes == 0:
294
length -= _RESERVED_HEADER_BYTES # padded
296
optimize_for_size = self._optimize_for_size
298
optimize_for_size = False
299
internal_row.writer = chunk_writer.ChunkWriter(length, 0,
300
optimize_for_size=optimize_for_size)
301
internal_row.writer.write(_INTERNAL_FLAG)
302
internal_row.writer.write(_INTERNAL_OFFSET +
303
str(rows[pos + 1].nodes) + "\n")
306
if rows[-1].nodes == 0:
307
length -= _RESERVED_HEADER_BYTES # padded
308
rows[-1].writer = chunk_writer.ChunkWriter(length,
309
optimize_for_size=self._optimize_for_size)
310
rows[-1].writer.write(_LEAF_FLAG)
311
if rows[-1].writer.write(line):
312
# this key did not fit in the node:
313
rows[-1].finish_node()
314
key_line = string_key + "\n"
316
for row in reversed(rows[:-1]):
317
# Mark the start of the next node in the node above. If it
318
# doesn't fit then propagate upwards until we find one that
320
if row.writer.write(key_line):
323
# We've found a node that can handle the pointer.
326
# If we reached the current root without being able to mark the
327
# division point, then we need a new root:
330
if 'index' in debug.debug_flags:
331
trace.mutter('Inserting new global row.')
332
new_row = _InternalBuilderRow()
334
rows.insert(0, new_row)
335
# This will be padded, hence the -100
336
new_row.writer = chunk_writer.ChunkWriter(
337
_PAGE_SIZE - _RESERVED_HEADER_BYTES,
339
optimize_for_size=self._optimize_for_size)
340
new_row.writer.write(_INTERNAL_FLAG)
341
new_row.writer.write(_INTERNAL_OFFSET +
342
str(rows[1].nodes - 1) + "\n")
343
new_row.writer.write(key_line)
344
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
346
def _write_nodes(self, node_iterator, allow_optimize=True):
347
"""Write node_iterator out as a B+Tree.
349
:param node_iterator: An iterator of sorted nodes. Each node should
350
match the output given by iter_all_entries.
351
:param allow_optimize: If set to False, prevent setting the optimize
352
flag when writing out. This is used by the _spill_mem_keys_to_disk
354
:return: A file handle for a temporary file containing a B+Tree for
357
# The index rows - rows[0] is the root, rows[1] is the layer under it
360
# forward sorted by key. In future we may consider topological sorting,
361
# at the cost of table scans for direct lookup, or a second index for
364
# A stack with the number of nodes of each size. 0 is the root node
365
# and must always be 1 (if there are any nodes in the tree).
366
self.row_lengths = []
367
# Loop over all nodes adding them to the bottom row
368
# (rows[-1]). When we finish a chunk in a row,
369
# propagate the key that didn't fit (comes after the chunk) to the
370
# row above, transitively.
371
for node in node_iterator:
373
# First key triggers the first row
374
rows.append(_LeafBuilderRow())
376
string_key, line = _btree_serializer._flatten_node(node,
377
self.reference_lists)
378
self._add_key(string_key, line, rows, allow_optimize=allow_optimize)
379
for row in reversed(rows):
380
pad = (type(row) != _LeafBuilderRow)
381
row.finish_node(pad=pad)
382
result = tempfile.NamedTemporaryFile(prefix='bzr-index-')
383
lines = [_BTSIGNATURE]
384
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
385
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
386
lines.append(_OPTION_LEN + str(key_count) + '\n')
387
row_lengths = [row.nodes for row in rows]
388
lines.append(_OPTION_ROW_LENGTHS + ','.join(map(str, row_lengths)) + '\n')
389
result.writelines(lines)
390
position = sum(map(len, lines))
392
if position > _RESERVED_HEADER_BYTES:
393
raise AssertionError("Could not fit the header in the"
394
" reserved space: %d > %d"
395
% (position, _RESERVED_HEADER_BYTES))
396
# write the rows out:
398
reserved = _RESERVED_HEADER_BYTES # reserved space for first node
401
# copy nodes to the finalised file.
402
# Special case the first node as it may be prefixed
403
node = row.spool.read(_PAGE_SIZE)
404
result.write(node[reserved:])
405
result.write("\x00" * (reserved - position))
406
position = 0 # Only the root row actually has an offset
407
copied_len = osutils.pumpfile(row.spool, result)
408
if copied_len != (row.nodes - 1) * _PAGE_SIZE:
409
if type(row) != _LeafBuilderRow:
410
raise AssertionError("Incorrect amount of data copied"
411
" expected: %d, got: %d"
412
% ((row.nodes - 1) * _PAGE_SIZE,
420
"""Finalise the index.
422
:return: A file handle for a temporary file containing the nodes added
425
return self._write_nodes(self.iter_all_entries())[0]
427
def iter_all_entries(self):
428
"""Iterate over all keys within the index
430
:return: An iterable of (index, key, value, reference_lists). There is
431
no defined order for the result iteration - it will be in the most
432
efficient order for the index (in this case dictionary hash order).
434
if 'evil' in debug.debug_flags:
435
trace.mutter_callsite(3,
436
"iter_all_entries scales with size of history.")
437
# Doing serial rather than ordered would be faster; but this shouldn't
438
# be getting called routinely anyway.
439
iterators = [self._iter_mem_nodes()]
440
for backing in self._backing_indices:
441
if backing is not None:
442
iterators.append(backing.iter_all_entries())
443
if len(iterators) == 1:
445
return self._iter_smallest(iterators)
447
def iter_entries(self, keys):
448
"""Iterate over keys within the index.
450
:param keys: An iterable providing the keys to be retrieved.
451
:return: An iterable of (index, key, value, reference_lists). There is no
452
defined order for the result iteration - it will be in the most
453
efficient order for the index (keys iteration order in this case).
456
local_keys = keys.intersection(self._keys)
457
if self.reference_lists:
458
for key in local_keys:
459
node = self._nodes[key]
460
yield self, key, node[1], node[0]
462
for key in local_keys:
463
node = self._nodes[key]
464
yield self, key, node[1]
465
# Find things that are in backing indices that have not been handled
467
if not self._backing_indices:
468
return # We won't find anything there either
469
# Remove all of the keys that we found locally
470
keys.difference_update(local_keys)
471
for backing in self._backing_indices:
476
for node in backing.iter_entries(keys):
478
yield (self,) + node[1:]
480
def iter_entries_prefix(self, keys):
481
"""Iterate over keys within the index using prefix matching.
483
Prefix matching is applied within the tuple of a key, not to within
484
the bytestring of each key element. e.g. if you have the keys ('foo',
485
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
486
only the former key is returned.
488
:param keys: An iterable providing the key prefixes to be retrieved.
489
Each key prefix takes the form of a tuple the length of a key, but
490
with the last N elements 'None' rather than a regular bytestring.
491
The first element cannot be 'None'.
492
:return: An iterable as per iter_all_entries, but restricted to the
493
keys with a matching prefix to those supplied. No additional keys
494
will be returned, and every match that is in the index will be
497
# XXX: To much duplication with the GraphIndex class; consider finding
498
# a good place to pull out the actual common logic.
502
for backing in self._backing_indices:
505
for node in backing.iter_entries_prefix(keys):
506
yield (self,) + node[1:]
507
if self._key_length == 1:
511
raise errors.BadIndexKey(key)
512
if len(key) != self._key_length:
513
raise errors.BadIndexKey(key)
515
node = self._nodes[key]
518
if self.reference_lists:
519
yield self, key, node[1], node[0]
521
yield self, key, node[1]
526
raise errors.BadIndexKey(key)
527
if len(key) != self._key_length:
528
raise errors.BadIndexKey(key)
529
# find what it refers to:
530
key_dict = self._get_nodes_by_key()
532
# find the subdict to return
534
while len(elements) and elements[0] is not None:
535
key_dict = key_dict[elements[0]]
538
# a non-existant lookup.
543
key_dict = dicts.pop(-1)
544
# can't be empty or would not exist
545
item, value = key_dict.iteritems().next()
546
if type(value) == dict:
548
dicts.extend(key_dict.itervalues())
551
for value in key_dict.itervalues():
552
yield (self, ) + value
554
yield (self, ) + key_dict
556
def _get_nodes_by_key(self):
557
if self._nodes_by_key is None:
559
if self.reference_lists:
560
for key, (references, value) in self._nodes.iteritems():
561
key_dict = nodes_by_key
562
for subkey in key[:-1]:
563
key_dict = key_dict.setdefault(subkey, {})
564
key_dict[key[-1]] = key, value, references
566
for key, (references, value) in self._nodes.iteritems():
567
key_dict = nodes_by_key
568
for subkey in key[:-1]:
569
key_dict = key_dict.setdefault(subkey, {})
570
key_dict[key[-1]] = key, value
571
self._nodes_by_key = nodes_by_key
572
return self._nodes_by_key
575
"""Return an estimate of the number of keys in this index.
577
For InMemoryGraphIndex the estimate is exact.
579
return len(self._keys) + sum(backing.key_count() for backing in
580
self._backing_indices if backing is not None)
583
"""In memory index's have no known corruption at the moment."""
586
class _LeafNode(object):
587
"""A leaf node for a serialised B+Tree index."""
589
__slots__ = ('keys', 'min_key', 'max_key')
591
def __init__(self, bytes, key_length, ref_list_length):
592
"""Parse bytes to create a leaf node object."""
593
# splitlines mangles the \r delimiters.. don't use it.
594
key_list = _btree_serializer._parse_leaf_lines(bytes,
595
key_length, ref_list_length)
597
self.min_key = key_list[0][0]
598
self.max_key = key_list[-1][0]
600
self.min_key = self.max_key = None
601
self.keys = dict(key_list)
604
class _InternalNode(object):
605
"""An internal node for a serialised B+Tree index."""
607
__slots__ = ('keys', 'offset')
609
def __init__(self, bytes):
610
"""Parse bytes to create an internal node object."""
611
# splitlines mangles the \r delimiters.. don't use it.
612
self.keys = self._parse_lines(bytes.split('\n'))
614
def _parse_lines(self, lines):
616
self.offset = int(lines[1][7:])
617
for line in lines[2:]:
620
nodes.append(tuple(map(intern, line.split('\0'))))
624
class BTreeGraphIndex(object):
625
"""Access to nodes via the standard GraphIndex interface for B+Tree's.
627
Individual nodes are held in a LRU cache. This holds the root node in
628
memory except when very large walks are done.
631
def __init__(self, transport, name, size):
632
"""Create a B+Tree index object on the index name.
634
:param transport: The transport to read data for the index from.
635
:param name: The file name of the index on transport.
636
:param size: Optional size of the index in bytes. This allows
637
compatibility with the GraphIndex API, as well as ensuring that
638
the initial read (to read the root node header) can be done
639
without over-reading even on empty indices, and on small indices
640
allows single-IO to read the entire index.
642
self._transport = transport
646
self._recommended_pages = self._compute_recommended_pages()
647
self._root_node = None
648
# Default max size is 100,000 leave values
649
self._leaf_value_cache = None # lru_cache.LRUCache(100*1000)
650
self._leaf_node_cache = lru_cache.LRUCache(_NODE_CACHE_SIZE)
651
# We could limit this, but even a 300k record btree has only 3k leaf
652
# nodes, and only 20 internal nodes. So the default of 100 nodes in an
653
# LRU would mean we always cache everything anyway, no need to pay the
655
self._internal_node_cache = fifo_cache.FIFOCache(100)
656
self._key_count = None
657
self._row_lengths = None
658
self._row_offsets = None # Start of each row, [-1] is the end
660
def __eq__(self, other):
661
"""Equal when self and other were created with the same parameters."""
663
type(self) == type(other) and
664
self._transport == other._transport and
665
self._name == other._name and
666
self._size == other._size)
668
def __ne__(self, other):
669
return not self.__eq__(other)
671
def _get_and_cache_nodes(self, nodes):
672
"""Read nodes and cache them in the lru.
674
The nodes list supplied is sorted and then read from disk, each node
675
being inserted it into the _node_cache.
677
Note: Asking for more nodes than the _node_cache can contain will
678
result in some of the results being immediately discarded, to prevent
679
this an assertion is raised if more nodes are asked for than are
682
:return: A dict of {node_pos: node}
685
start_of_leaves = None
686
for node_pos, node in self._read_nodes(sorted(nodes)):
687
if node_pos == 0: # Special case
688
self._root_node = node
690
if start_of_leaves is None:
691
start_of_leaves = self._row_offsets[-2]
692
if node_pos < start_of_leaves:
693
self._internal_node_cache.add(node_pos, node)
695
self._leaf_node_cache.add(node_pos, node)
696
found[node_pos] = node
699
def _compute_recommended_pages(self):
700
"""Convert transport's recommended_page_size into btree pages.
702
recommended_page_size is in bytes, we want to know how many _PAGE_SIZE
703
pages fit in that length.
705
recommended_read = self._transport.recommended_page_size()
706
recommended_pages = int(math.ceil(recommended_read /
708
return recommended_pages
710
def _compute_total_pages_in_index(self):
711
"""How many pages are in the index.
713
If we have read the header we will use the value stored there.
714
Otherwise it will be computed based on the length of the index.
716
if self._size is None:
717
raise AssertionError('_compute_total_pages_in_index should not be'
718
' called when self._size is None')
719
if self._root_node is not None:
720
# This is the number of pages as defined by the header
721
return self._row_offsets[-1]
722
# This is the number of pages as defined by the size of the index. They
723
# should be indentical.
724
total_pages = int(math.ceil(self._size / float(_PAGE_SIZE)))
727
def _expand_offsets(self, offsets):
728
"""Find extra pages to download.
730
The idea is that we always want to make big-enough requests (like 64kB
731
for http), so that we don't waste round trips. So given the entries
732
that we already have cached and the new pages being downloaded figure
733
out what other pages we might want to read.
735
See also doc/developers/btree_index_prefetch.txt for more details.
737
:param offsets: The offsets to be read
738
:return: A list of offsets to download
740
if 'index' in debug.debug_flags:
741
trace.mutter('expanding: %s\toffsets: %s', self._name, offsets)
743
if len(offsets) >= self._recommended_pages:
744
# Don't add more, we are already requesting more than enough
745
if 'index' in debug.debug_flags:
746
trace.mutter(' not expanding large request (%s >= %s)',
747
len(offsets), self._recommended_pages)
749
if self._size is None:
750
# Don't try anything, because we don't know where the file ends
751
if 'index' in debug.debug_flags:
752
trace.mutter(' not expanding without knowing index size')
754
total_pages = self._compute_total_pages_in_index()
755
cached_offsets = self._get_offsets_to_cached_pages()
756
# If reading recommended_pages would read the rest of the index, just
758
if total_pages - len(cached_offsets) <= self._recommended_pages:
759
# Read whatever is left
761
expanded = [x for x in xrange(total_pages)
762
if x not in cached_offsets]
764
expanded = range(total_pages)
765
if 'index' in debug.debug_flags:
766
trace.mutter(' reading all unread pages: %s', expanded)
769
if self._root_node is None:
770
# ATM on the first read of the root node of a large index, we don't
771
# bother pre-reading any other pages. This is because the
772
# likelyhood of actually reading interesting pages is very low.
773
# See doc/developers/btree_index_prefetch.txt for a discussion, and
774
# a possible implementation when we are guessing that the second
775
# layer index is small
776
final_offsets = offsets
778
tree_depth = len(self._row_lengths)
779
if len(cached_offsets) < tree_depth and len(offsets) == 1:
780
# We haven't read enough to justify expansion
781
# If we are only going to read the root node, and 1 leaf node,
782
# then it isn't worth expanding our request. Once we've read at
783
# least 2 nodes, then we are probably doing a search, and we
784
# start expanding our requests.
785
if 'index' in debug.debug_flags:
786
trace.mutter(' not expanding on first reads')
788
final_offsets = self._expand_to_neighbors(offsets, cached_offsets,
791
final_offsets = sorted(final_offsets)
792
if 'index' in debug.debug_flags:
793
trace.mutter('expanded: %s', final_offsets)
796
def _expand_to_neighbors(self, offsets, cached_offsets, total_pages):
797
"""Expand requests to neighbors until we have enough pages.
799
This is called from _expand_offsets after policy has determined that we
801
We only want to expand requests within a given layer. We cheat a little
802
bit and assume all requests will be in the same layer. This is true
803
given the current design, but if it changes this algorithm may perform
806
:param offsets: requested offsets
807
:param cached_offsets: offsets for pages we currently have cached
808
:return: A set() of offsets after expansion
810
final_offsets = set(offsets)
812
new_tips = set(final_offsets)
813
while len(final_offsets) < self._recommended_pages and new_tips:
817
first, end = self._find_layer_first_and_end(pos)
820
and previous not in cached_offsets
821
and previous not in final_offsets
822
and previous >= first):
823
next_tips.add(previous)
825
if (after < total_pages
826
and after not in cached_offsets
827
and after not in final_offsets
830
# This would keep us from going bigger than
831
# recommended_pages by only expanding the first offsets.
832
# However, if we are making a 'wide' request, it is
833
# reasonable to expand all points equally.
834
# if len(final_offsets) > recommended_pages:
836
final_offsets.update(next_tips)
840
def external_references(self, ref_list_num):
841
if self._root_node is None:
842
self._get_root_node()
843
if ref_list_num + 1 > self.node_ref_lists:
844
raise ValueError('No ref list %d, index has %d ref lists'
845
% (ref_list_num, self.node_ref_lists))
848
for node in self.iter_all_entries():
850
refs.update(node[3][ref_list_num])
853
def _find_layer_first_and_end(self, offset):
854
"""Find the start/stop nodes for the layer corresponding to offset.
856
:return: (first, end)
857
first is the first node in this layer
858
end is the first node of the next layer
861
for roffset in self._row_offsets:
868
def _get_offsets_to_cached_pages(self):
869
"""Determine what nodes we already have cached."""
870
cached_offsets = set(self._internal_node_cache.keys())
871
cached_offsets.update(self._leaf_node_cache.keys())
872
if self._root_node is not None:
873
cached_offsets.add(0)
874
return cached_offsets
876
def _get_root_node(self):
877
if self._root_node is None:
878
# We may not have a root node yet
879
self._get_internal_nodes([0])
880
return self._root_node
882
def _get_nodes(self, cache, node_indexes):
885
for idx in node_indexes:
886
if idx == 0 and self._root_node is not None:
887
found[0] = self._root_node
890
found[idx] = cache[idx]
895
needed = self._expand_offsets(needed)
896
found.update(self._get_and_cache_nodes(needed))
899
def _get_internal_nodes(self, node_indexes):
900
"""Get a node, from cache or disk.
902
After getting it, the node will be cached.
904
return self._get_nodes(self._internal_node_cache, node_indexes)
906
def _cache_leaf_values(self, nodes):
907
"""Cache directly from key => value, skipping the btree."""
908
if self._leaf_value_cache is not None:
909
for node in nodes.itervalues():
910
for key, value in node.keys.iteritems():
911
if key in self._leaf_value_cache:
912
# Don't add the rest of the keys, we've seen this node
915
self._leaf_value_cache[key] = value
917
def _get_leaf_nodes(self, node_indexes):
918
"""Get a bunch of nodes, from cache or disk."""
919
found = self._get_nodes(self._leaf_node_cache, node_indexes)
920
self._cache_leaf_values(found)
923
def iter_all_entries(self):
924
"""Iterate over all keys within the index.
926
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
927
The former tuple is used when there are no reference lists in the
928
index, making the API compatible with simple key:value index types.
929
There is no defined order for the result iteration - it will be in
930
the most efficient order for the index.
932
if 'evil' in debug.debug_flags:
933
trace.mutter_callsite(3,
934
"iter_all_entries scales with size of history.")
935
if not self.key_count():
937
if self._row_offsets[-1] == 1:
938
# There is only the root node, and we read that via key_count()
939
if self.node_ref_lists:
940
for key, (value, refs) in sorted(self._root_node.keys.items()):
941
yield (self, key, value, refs)
943
for key, (value, refs) in sorted(self._root_node.keys.items()):
944
yield (self, key, value)
946
start_of_leaves = self._row_offsets[-2]
947
end_of_leaves = self._row_offsets[-1]
948
needed_offsets = range(start_of_leaves, end_of_leaves)
949
if needed_offsets == [0]:
950
# Special case when we only have a root node, as we have already
952
nodes = [(0, self._root_node)]
954
nodes = self._read_nodes(needed_offsets)
955
# We iterate strictly in-order so that we can use this function
956
# for spilling index builds to disk.
957
if self.node_ref_lists:
958
for _, node in nodes:
959
for key, (value, refs) in sorted(node.keys.items()):
960
yield (self, key, value, refs)
962
for _, node in nodes:
963
for key, (value, refs) in sorted(node.keys.items()):
964
yield (self, key, value)
967
def _multi_bisect_right(in_keys, fixed_keys):
968
"""Find the positions where each 'in_key' would fit in fixed_keys.
970
This is equivalent to doing "bisect_right" on each in_key into
973
:param in_keys: A sorted list of keys to match with fixed_keys
974
:param fixed_keys: A sorted list of keys to match against
975
:return: A list of (integer position, [key list]) tuples.
980
# no pointers in the fixed_keys list, which means everything must
982
return [(0, in_keys)]
984
# TODO: Iterating both lists will generally take M + N steps
985
# Bisecting each key will generally take M * log2 N steps.
986
# If we had an efficient way to compare, we could pick the method
987
# based on which has the fewer number of steps.
988
# There is also the argument that bisect_right is a compiled
989
# function, so there is even more to be gained.
990
# iter_steps = len(in_keys) + len(fixed_keys)
991
# bisect_steps = len(in_keys) * math.log(len(fixed_keys), 2)
992
if len(in_keys) == 1: # Bisect will always be faster for M = 1
993
return [(bisect_right(fixed_keys, in_keys[0]), in_keys)]
994
# elif bisect_steps < iter_steps:
996
# for key in in_keys:
997
# offsets.setdefault(bisect_right(fixed_keys, key),
999
# return [(o, offsets[o]) for o in sorted(offsets)]
1000
in_keys_iter = iter(in_keys)
1001
fixed_keys_iter = enumerate(fixed_keys)
1002
cur_in_key = in_keys_iter.next()
1003
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1005
class InputDone(Exception): pass
1006
class FixedDone(Exception): pass
1011
# TODO: Another possibility is that rather than iterating on each side,
1012
# we could use a combination of bisecting and iterating. For
1013
# example, while cur_in_key < fixed_key, bisect to find its
1014
# point, then iterate all matching keys, then bisect (restricted
1015
# to only the remainder) for the next one, etc.
1018
if cur_in_key < cur_fixed_key:
1020
cur_out = (cur_fixed_offset, cur_keys)
1021
output.append(cur_out)
1022
while cur_in_key < cur_fixed_key:
1023
cur_keys.append(cur_in_key)
1025
cur_in_key = in_keys_iter.next()
1026
except StopIteration:
1028
# At this point cur_in_key must be >= cur_fixed_key
1029
# step the cur_fixed_key until we pass the cur key, or walk off
1031
while cur_in_key >= cur_fixed_key:
1033
cur_fixed_offset, cur_fixed_key = fixed_keys_iter.next()
1034
except StopIteration:
1037
# We consumed all of the input, nothing more to do
1040
# There was some input left, but we consumed all of fixed, so we
1041
# have to add one more for the tail
1042
cur_keys = [cur_in_key]
1043
cur_keys.extend(in_keys_iter)
1044
cur_out = (len(fixed_keys), cur_keys)
1045
output.append(cur_out)
1048
def _walk_through_internal_nodes(self, keys):
1049
"""Take the given set of keys, and find the corresponding LeafNodes.
1051
:param keys: An unsorted iterable of keys to search for
1052
:return: (nodes, index_and_keys)
1053
nodes is a dict mapping {index: LeafNode}
1054
keys_at_index is a list of tuples of [(index, [keys for Leaf])]
1056
# 6 seconds spent in miss_torture using the sorted() line.
1057
# Even with out of order disk IO it seems faster not to sort it when
1058
# large queries are being made.
1059
keys_at_index = [(0, sorted(keys))]
1061
for row_pos, next_row_start in enumerate(self._row_offsets[1:-1]):
1062
node_indexes = [idx for idx, s_keys in keys_at_index]
1063
nodes = self._get_internal_nodes(node_indexes)
1065
next_nodes_and_keys = []
1066
for node_index, sub_keys in keys_at_index:
1067
node = nodes[node_index]
1068
positions = self._multi_bisect_right(sub_keys, node.keys)
1069
node_offset = next_row_start + node.offset
1070
next_nodes_and_keys.extend([(node_offset + pos, s_keys)
1071
for pos, s_keys in positions])
1072
keys_at_index = next_nodes_and_keys
1073
# We should now be at the _LeafNodes
1074
node_indexes = [idx for idx, s_keys in keys_at_index]
1076
# TODO: We may *not* want to always read all the nodes in one
1077
# big go. Consider setting a max size on this.
1078
nodes = self._get_leaf_nodes(node_indexes)
1079
return nodes, keys_at_index
1081
def iter_entries(self, keys):
1082
"""Iterate over keys within the index.
1084
:param keys: An iterable providing the keys to be retrieved.
1085
:return: An iterable as per iter_all_entries, but restricted to the
1086
keys supplied. No additional keys will be returned, and every
1087
key supplied that is in the index will be returned.
1089
# 6 seconds spent in miss_torture using the sorted() line.
1090
# Even with out of order disk IO it seems faster not to sort it when
1091
# large queries are being made.
1092
# However, now that we are doing multi-way bisecting, we need the keys
1093
# in sorted order anyway. We could change the multi-way code to not
1094
# require sorted order. (For example, it bisects for the first node,
1095
# does an in-order search until a key comes before the current point,
1096
# which it then bisects for, etc.)
1097
keys = frozenset(keys)
1101
if not self.key_count():
1105
if self._leaf_value_cache is None:
1109
value = self._leaf_value_cache.get(key, None)
1110
if value is not None:
1111
# This key is known not to be here, skip it
1113
if self.node_ref_lists:
1114
yield (self, key, value, refs)
1116
yield (self, key, value)
1118
needed_keys.append(key)
1124
nodes, nodes_and_keys = self._walk_through_internal_nodes(needed_keys)
1125
for node_index, sub_keys in nodes_and_keys:
1128
node = nodes[node_index]
1129
for next_sub_key in sub_keys:
1130
if next_sub_key in node.keys:
1131
value, refs = node.keys[next_sub_key]
1132
if self.node_ref_lists:
1133
yield (self, next_sub_key, value, refs)
1135
yield (self, next_sub_key, value)
1137
def _find_ancestors(self, keys, ref_list_num, parent_map, missing_keys):
1138
"""Find the parent_map information for the set of keys.
1140
This populates the parent_map dict and missing_keys set based on the
1141
queried keys. It also can fill out an arbitrary number of parents that
1142
it finds while searching for the supplied keys.
1144
It is unlikely that you want to call this directly. See
1145
"CombinedGraphIndex.find_ancestry()" for a more appropriate API.
1147
:param keys: A keys whose ancestry we want to return
1148
Every key will either end up in 'parent_map' or 'missing_keys'.
1149
:param ref_list_num: This index in the ref_lists is the parents we
1151
:param parent_map: {key: parent_keys} for keys that are present in this
1152
index. This may contain more entries than were in 'keys', that are
1153
reachable ancestors of the keys requested.
1154
:param missing_keys: keys which are known to be missing in this index.
1155
This may include parents that were not directly requested, but we
1156
were able to determine that they are not present in this index.
1157
:return: search_keys parents that were found but not queried to know
1158
if they are missing or present. Callers can re-query this index for
1159
those keys, and they will be placed into parent_map or missing_keys
1161
if not self.key_count():
1162
# We use key_count() to trigger reading the root node and
1163
# determining info about this BTreeGraphIndex
1164
# If we don't have any keys, then everything is missing
1165
missing_keys.update(keys)
1167
if ref_list_num >= self.node_ref_lists:
1168
raise ValueError('No ref list %d, index has %d ref lists'
1169
% (ref_list_num, self.node_ref_lists))
1171
# The main trick we are trying to accomplish is that when we find a
1172
# key listing its parents, we expect that the parent key is also likely
1173
# to sit on the same page. Allowing us to expand parents quickly
1174
# without suffering the full stack of bisecting, etc.
1175
nodes, nodes_and_keys = self._walk_through_internal_nodes(keys)
1177
# These are parent keys which could not be immediately resolved on the
1178
# page where the child was present. Note that we may already be
1179
# searching for that key, and it may actually be present [or known
1180
# missing] on one of the other pages we are reading.
1182
# We could try searching for them in the immediate previous or next
1183
# page. If they occur "later" we could put them in a pending lookup
1184
# set, and then for each node we read thereafter we could check to
1185
# see if they are present.
1186
# However, we don't know the impact of keeping this list of things
1187
# that I'm going to search for every node I come across from here on
1189
# It doesn't handle the case when the parent key is missing on a
1190
# page that we *don't* read. So we already have to handle being
1191
# re-entrant for that.
1192
# Since most keys contain a date string, they are more likely to be
1193
# found earlier in the file than later, but we would know that right
1194
# away (key < min_key), and wouldn't keep searching it on every other
1195
# page that we read.
1196
# Mostly, it is an idea, one which should be benchmarked.
1197
parents_not_on_page = set()
1199
for node_index, sub_keys in nodes_and_keys:
1202
# sub_keys is all of the keys we are looking for that should exist
1203
# on this page, if they aren't here, then they won't be found
1204
node = nodes[node_index]
1205
node_keys = node.keys
1206
parents_to_check = set()
1207
for next_sub_key in sub_keys:
1208
if next_sub_key not in node_keys:
1209
# This one is just not present in the index at all
1210
missing_keys.add(next_sub_key)
1212
value, refs = node_keys[next_sub_key]
1213
parent_keys = refs[ref_list_num]
1214
parent_map[next_sub_key] = parent_keys
1215
parents_to_check.update(parent_keys)
1216
# Don't look for things we've already found
1217
parents_to_check = parents_to_check.difference(parent_map)
1218
# this can be used to test the benefit of having the check loop
1220
# parents_not_on_page.update(parents_to_check)
1222
while parents_to_check:
1223
next_parents_to_check = set()
1224
for key in parents_to_check:
1225
if key in node_keys:
1226
value, refs = node_keys[key]
1227
parent_keys = refs[ref_list_num]
1228
parent_map[key] = parent_keys
1229
next_parents_to_check.update(parent_keys)
1231
# This parent either is genuinely missing, or should be
1232
# found on another page. Perf test whether it is better
1233
# to check if this node should fit on this page or not.
1234
# in the 'everything-in-one-pack' scenario, this *not*
1235
# doing the check is 237ms vs 243ms.
1236
# So slightly better, but I assume the standard 'lots
1237
# of packs' is going to show a reasonable improvement
1238
# from the check, because it avoids 'going around
1239
# again' for everything that is in another index
1240
# parents_not_on_page.add(key)
1241
# Missing for some reason
1242
if key < node.min_key:
1243
# in the case of bzr.dev, 3.4k/5.3k misses are
1244
# 'earlier' misses (65%)
1245
parents_not_on_page.add(key)
1246
elif key > node.max_key:
1247
# This parent key would be present on a different
1249
parents_not_on_page.add(key)
1251
# assert key != node.min_key and key != node.max_key
1252
# If it was going to be present, it would be on
1253
# *this* page, so mark it missing.
1254
missing_keys.add(key)
1255
parents_to_check = next_parents_to_check.difference(parent_map)
1256
# Might want to do another .difference() from missing_keys
1257
# parents_not_on_page could have been found on a different page, or be
1258
# known to be missing. So cull out everything that has already been
1260
search_keys = parents_not_on_page.difference(
1261
parent_map).difference(missing_keys)
1264
def iter_entries_prefix(self, keys):
1265
"""Iterate over keys within the index using prefix matching.
1267
Prefix matching is applied within the tuple of a key, not to within
1268
the bytestring of each key element. e.g. if you have the keys ('foo',
1269
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1270
only the former key is returned.
1272
WARNING: Note that this method currently causes a full index parse
1273
unconditionally (which is reasonably appropriate as it is a means for
1274
thunking many small indices into one larger one and still supplies
1275
iter_all_entries at the thunk layer).
1277
:param keys: An iterable providing the key prefixes to be retrieved.
1278
Each key prefix takes the form of a tuple the length of a key, but
1279
with the last N elements 'None' rather than a regular bytestring.
1280
The first element cannot be 'None'.
1281
:return: An iterable as per iter_all_entries, but restricted to the
1282
keys with a matching prefix to those supplied. No additional keys
1283
will be returned, and every match that is in the index will be
1286
keys = sorted(set(keys))
1289
# Load if needed to check key lengths
1290
if self._key_count is None:
1291
self._get_root_node()
1292
# TODO: only access nodes that can satisfy the prefixes we are looking
1293
# for. For now, to meet API usage (as this function is not used by
1294
# current bzrlib) just suck the entire index and iterate in memory.
1296
if self.node_ref_lists:
1297
if self._key_length == 1:
1298
for _1, key, value, refs in self.iter_all_entries():
1299
nodes[key] = value, refs
1302
for _1, key, value, refs in self.iter_all_entries():
1303
key_value = key, value, refs
1304
# For a key of (foo, bar, baz) create
1305
# _nodes_by_key[foo][bar][baz] = key_value
1306
key_dict = nodes_by_key
1307
for subkey in key[:-1]:
1308
key_dict = key_dict.setdefault(subkey, {})
1309
key_dict[key[-1]] = key_value
1311
if self._key_length == 1:
1312
for _1, key, value in self.iter_all_entries():
1316
for _1, key, value in self.iter_all_entries():
1317
key_value = key, value
1318
# For a key of (foo, bar, baz) create
1319
# _nodes_by_key[foo][bar][baz] = key_value
1320
key_dict = nodes_by_key
1321
for subkey in key[:-1]:
1322
key_dict = key_dict.setdefault(subkey, {})
1323
key_dict[key[-1]] = key_value
1324
if self._key_length == 1:
1328
raise errors.BadIndexKey(key)
1329
if len(key) != self._key_length:
1330
raise errors.BadIndexKey(key)
1332
if self.node_ref_lists:
1333
value, node_refs = nodes[key]
1334
yield self, key, value, node_refs
1336
yield self, key, nodes[key]
1343
raise errors.BadIndexKey(key)
1344
if len(key) != self._key_length:
1345
raise errors.BadIndexKey(key)
1346
# find what it refers to:
1347
key_dict = nodes_by_key
1348
elements = list(key)
1349
# find the subdict whose contents should be returned.
1351
while len(elements) and elements[0] is not None:
1352
key_dict = key_dict[elements[0]]
1355
# a non-existant lookup.
1360
key_dict = dicts.pop(-1)
1361
# can't be empty or would not exist
1362
item, value = key_dict.iteritems().next()
1363
if type(value) == dict:
1365
dicts.extend(key_dict.itervalues())
1368
for value in key_dict.itervalues():
1369
# each value is the key:value:node refs tuple
1371
yield (self, ) + value
1373
# the last thing looked up was a terminal element
1374
yield (self, ) + key_dict
1376
def key_count(self):
1377
"""Return an estimate of the number of keys in this index.
1379
For BTreeGraphIndex the estimate is exact as it is contained in the
1382
if self._key_count is None:
1383
self._get_root_node()
1384
return self._key_count
1386
def _compute_row_offsets(self):
1387
"""Fill out the _row_offsets attribute based on _row_lengths."""
1390
for row in self._row_lengths:
1391
offsets.append(row_offset)
1393
offsets.append(row_offset)
1394
self._row_offsets = offsets
1396
def _parse_header_from_bytes(self, bytes):
1397
"""Parse the header from a region of bytes.
1399
:param bytes: The data to parse.
1400
:return: An offset, data tuple such as readv yields, for the unparsed
1401
data. (which may be of length 0).
1403
signature = bytes[0:len(self._signature())]
1404
if not signature == self._signature():
1405
raise errors.BadIndexFormatSignature(self._name, BTreeGraphIndex)
1406
lines = bytes[len(self._signature()):].splitlines()
1407
options_line = lines[0]
1408
if not options_line.startswith(_OPTION_NODE_REFS):
1409
raise errors.BadIndexOptions(self)
1411
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
1413
raise errors.BadIndexOptions(self)
1414
options_line = lines[1]
1415
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
1416
raise errors.BadIndexOptions(self)
1418
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
1420
raise errors.BadIndexOptions(self)
1421
options_line = lines[2]
1422
if not options_line.startswith(_OPTION_LEN):
1423
raise errors.BadIndexOptions(self)
1425
self._key_count = int(options_line[len(_OPTION_LEN):])
1427
raise errors.BadIndexOptions(self)
1428
options_line = lines[3]
1429
if not options_line.startswith(_OPTION_ROW_LENGTHS):
1430
raise errors.BadIndexOptions(self)
1432
self._row_lengths = map(int, [length for length in
1433
options_line[len(_OPTION_ROW_LENGTHS):].split(',')
1436
raise errors.BadIndexOptions(self)
1437
self._compute_row_offsets()
1439
# calculate the bytes we have processed
1440
header_end = (len(signature) + sum(map(len, lines[0:4])) + 4)
1441
return header_end, bytes[header_end:]
1443
def _read_nodes(self, nodes):
1444
"""Read some nodes from disk into the LRU cache.
1446
This performs a readv to get the node data into memory, and parses each
1447
node, then yields it to the caller. The nodes are requested in the
1448
supplied order. If possible doing sort() on the list before requesting
1449
a read may improve performance.
1451
:param nodes: The nodes to read. 0 - first node, 1 - second node etc.
1454
# may be the byte string of the whole file
1456
# list of (offset, length) regions of the file that should, evenually
1457
# be read in to data_ranges, either from 'bytes' or from the transport
1460
offset = index * _PAGE_SIZE
1463
# Root node - special case
1465
size = min(_PAGE_SIZE, self._size)
1467
# The only case where we don't know the size, is for very
1468
# small indexes. So we read the whole thing
1469
bytes = self._transport.get_bytes(self._name)
1470
self._size = len(bytes)
1471
# the whole thing should be parsed out of 'bytes'
1472
ranges.append((0, len(bytes)))
1475
if offset > self._size:
1476
raise AssertionError('tried to read past the end'
1477
' of the file %s > %s'
1478
% (offset, self._size))
1479
size = min(size, self._size - offset)
1480
ranges.append((offset, size))
1483
elif bytes is not None:
1484
# already have the whole file
1485
data_ranges = [(start, bytes[start:start+_PAGE_SIZE])
1486
for start in xrange(0, len(bytes), _PAGE_SIZE)]
1487
elif self._file is None:
1488
data_ranges = self._transport.readv(self._name, ranges)
1491
for offset, size in ranges:
1492
self._file.seek(offset)
1493
data_ranges.append((offset, self._file.read(size)))
1494
for offset, data in data_ranges:
1496
# extract the header
1497
offset, data = self._parse_header_from_bytes(data)
1500
bytes = zlib.decompress(data)
1501
if bytes.startswith(_LEAF_FLAG):
1502
node = _LeafNode(bytes, self._key_length, self.node_ref_lists)
1503
elif bytes.startswith(_INTERNAL_FLAG):
1504
node = _InternalNode(bytes)
1506
raise AssertionError("Unknown node type for %r" % bytes)
1507
yield offset / _PAGE_SIZE, node
1509
def _signature(self):
1510
"""The file signature for this index type."""
1514
"""Validate that everything in the index can be accessed."""
1515
# just read and parse every node.
1516
self._get_root_node()
1517
if len(self._row_lengths) > 1:
1518
start_node = self._row_offsets[1]
1520
# We shouldn't be reading anything anyway
1522
node_end = self._row_offsets[-1]
1523
for node in self._read_nodes(range(start_node, node_end)):
1528
from bzrlib import _btree_serializer_pyx as _btree_serializer
1530
from bzrlib import _btree_serializer_py as _btree_serializer
added
removed
bzrlib/btree_index.py
