51
from ..static_tuple import StaticTuple
59
from brzlib.static_tuple import StaticTuple
54
62
# If each line is 50 bytes, and you have 255 internal pages, with 255-way fan
55
63
# out, it takes 3.1MB to cache the layer.
56
_PAGE_CACHE_SIZE = 4 * 1024 * 1024
64
_PAGE_CACHE_SIZE = 4*1024*1024
57
65
# Per thread caches for 2 reasons:
58
66
# - in the server we may be serving very different content, so we get less
90
97
def _search_key_plain(key):
91
98
"""Map the key tuple into a search string that just uses the key bytes."""
92
return b'\x00'.join(key)
99
return '\x00'.join(key)
95
102
search_key_registry = registry.Registry()
96
search_key_registry.register(b'plain', _search_key_plain)
103
search_key_registry.register('plain', _search_key_plain)
99
106
class CHKMap(object):
131
138
has_deletes = False
132
139
# Check preconditions first.
133
140
as_st = StaticTuple.from_sequence
134
new_items = {as_st(key) for (old, key, value) in delta
135
if key is not None and old is None}
141
new_items = set([as_st(key) for (old, key, value) in delta
142
if key is not None and old is None])
136
143
existing_new = list(self.iteritems(key_filter=new_items))
138
145
raise errors.InconsistentDeltaDelta(delta,
139
"New items are already in the map %r." % existing_new)
146
"New items are already in the map %r." % existing_new)
140
147
# Now apply changes.
141
148
for old, new, value in delta:
142
149
if old is not None and old != new:
165
172
:param node: A tuple key or node object.
166
173
:return: A node object.
168
if isinstance(node, StaticTuple):
175
if type(node) is StaticTuple:
169
176
bytes = self._read_bytes(node)
170
177
return _deserialise(bytes, node,
171
search_key_func=self._search_key_func)
178
search_key_func=self._search_key_func)
177
184
return _get_cache()[key]
179
186
stream = self._store.get_record_stream([key], 'unordered', True)
180
bytes = next(stream).get_bytes_as('fulltext')
187
bytes = stream.next().get_bytes_as('fulltext')
181
188
_get_cache()[key] = bytes
184
def _dump_tree(self, include_keys=False, encoding='utf-8'):
191
def _dump_tree(self, include_keys=False):
185
192
"""Return the tree in a string representation."""
186
193
self._ensure_root()
187
def decode(x): return x.decode(encoding)
188
res = self._dump_tree_node(self._root_node, prefix=b'', indent='',
189
decode=decode, include_keys=include_keys)
190
res.append('') # Give a trailing '\n'
194
res = self._dump_tree_node(self._root_node, prefix='', indent='',
195
include_keys=include_keys)
196
res.append('') # Give a trailing '\n'
191
197
return '\n'.join(res)
193
def _dump_tree_node(self, node, prefix, indent, decode, include_keys=True):
199
def _dump_tree_node(self, node, prefix, indent, include_keys=True):
194
200
"""For this node and all children, generate a string representation."""
196
202
if not include_keys:
199
205
node_key = node.key()
200
206
if node_key is not None:
201
key_str = ' %s' % (decode(node_key[0]),)
207
key_str = ' %s' % (node_key[0],)
203
209
key_str = ' None'
204
result.append('%s%r %s%s' % (indent, decode(prefix), node.__class__.__name__,
210
result.append('%s%r %s%s' % (indent, prefix, node.__class__.__name__,
206
if isinstance(node, InternalNode):
212
if type(node) is InternalNode:
207
213
# Trigger all child nodes to get loaded
208
214
list(node._iter_nodes(self._store))
209
for prefix, sub in sorted(node._items.items()):
215
for prefix, sub in sorted(node._items.iteritems()):
210
216
result.extend(self._dump_tree_node(sub, prefix, indent + ' ',
211
decode=decode, include_keys=include_keys))
217
include_keys=include_keys))
213
for key, value in sorted(node._items.items()):
219
for key, value in sorted(node._items.iteritems()):
214
220
# Don't use prefix nor indent here to line up when used in
215
221
# tests in conjunction with assertEqualDiff
216
result.append(' %r %r' % (
217
tuple([decode(ke) for ke in key]), decode(value)))
222
result.append(' %r %r' % (tuple(key), value))
221
226
def from_dict(klass, store, initial_value, maximum_size=0, key_width=1,
222
search_key_func=None):
227
search_key_func=None):
223
228
"""Create a CHKMap in store with initial_value as the content.
225
230
:param store: The store to record initial_value in, a VersionedFiles
236
241
:return: The root chk of the resulting CHKMap.
238
243
root_key = klass._create_directly(store, initial_value,
239
maximum_size=maximum_size, key_width=key_width,
240
search_key_func=search_key_func)
241
if not isinstance(root_key, StaticTuple):
244
maximum_size=maximum_size, key_width=key_width,
245
search_key_func=search_key_func)
246
if type(root_key) is not StaticTuple:
242
247
raise AssertionError('we got a %s instead of a StaticTuple'
243
248
% (type(root_key),))
262
267
node.set_maximum_size(maximum_size)
263
268
node._key_width = key_width
264
269
as_st = StaticTuple.from_sequence
265
node._items = dict((as_st(key), val)
266
for key, val in initial_value.items())
267
node._raw_size = sum(node._key_value_len(key, value)
268
for key, value in node._items.items())
270
node._items = dict([(as_st(key), val) for key, val
271
in initial_value.iteritems()])
272
node._raw_size = sum([node._key_value_len(key, value)
273
for key,value in node._items.iteritems()])
269
274
node._len = len(node._items)
270
275
node._compute_search_prefix()
271
276
node._compute_serialised_prefix()
272
if (node._len > 1 and
274
node._current_size() > maximum_size):
279
and node._current_size() > maximum_size):
275
280
prefix, node_details = node._split(store)
276
281
if len(node_details) == 1:
277
282
raise AssertionError('Failed to split using node._split')
323
328
# key_path (a list of tuples, tail-sharing down the tree.)
324
329
self_pending = []
325
330
basis_pending = []
327
331
def process_node(node, path, a_map, pending):
328
332
# take a node and expand it
329
333
node = a_map._get_node(node)
330
if isinstance(node, LeafNode):
334
if type(node) == LeafNode:
331
335
path = (node._key, path)
332
336
for key, value in node._items.items():
333
337
# For a LeafNode, the key is a serialized_key, rather than
339
343
path = (node._key, path)
340
344
for prefix, child in node._items.items():
341
345
heapq.heappush(pending, (prefix, None, child, path))
343
346
def process_common_internal_nodes(self_node, basis_node):
344
347
self_items = set(self_node._items.items())
345
348
basis_items = set(basis_node._items.items())
349
352
path = (basis_node._key, None)
350
353
for prefix, child in basis_items - self_items:
351
354
heapq.heappush(basis_pending, (prefix, None, child, path))
353
355
def process_common_leaf_nodes(self_node, basis_node):
354
356
self_items = set(self_node._items.items())
355
357
basis_items = set(basis_node._items.items())
361
363
for key, value in basis_items - self_items:
362
364
prefix = basis._search_key_func(key)
363
365
heapq.heappush(basis_pending, (prefix, key, value, path))
365
366
def process_common_prefix_nodes(self_node, self_path,
366
367
basis_node, basis_path):
367
368
# Would it be more efficient if we could request both at the same
369
370
self_node = self._get_node(self_node)
370
371
basis_node = basis._get_node(basis_node)
371
if (isinstance(self_node, InternalNode) and
372
isinstance(basis_node, InternalNode)):
372
if (type(self_node) == InternalNode
373
and type(basis_node) == InternalNode):
373
374
# Matching internal nodes
374
375
process_common_internal_nodes(self_node, basis_node)
375
elif (isinstance(self_node, LeafNode) and
376
isinstance(basis_node, LeafNode)):
376
elif (type(self_node) == LeafNode
377
and type(basis_node) == LeafNode):
377
378
process_common_leaf_nodes(self_node, basis_node)
379
380
process_node(self_node, self_path, self, self_pending)
382
383
self_seen = set()
383
384
basis_seen = set()
384
385
excluded_keys = set()
386
386
def check_excluded(key_path):
387
387
# Note that this is N^2, it depends on us trimming trees
388
388
# aggressively to not become slow.
389
389
# A better implementation would probably have a reverse map
390
390
# back to the children of a node, and jump straight to it when
391
391
# a common node is detected, the proceed to remove the already
392
# pending children. breezy.graph has a searcher module with a
392
# pending children. brzlib.graph has a searcher module with a
393
393
# similar problem.
394
394
while key_path is not None:
395
395
key, key_path = key_path
472
472
basis_details = heapq.heappop(basis_pending)
473
473
if self_details[2] != basis_details[2]:
474
474
yield (self_details[1],
475
basis_details[2], self_details[2])
475
basis_details[2], self_details[2])
477
477
# At least one side wasn't a simple value
478
if (self._node_key(self_pending[0][2])
479
== self._node_key(basis_pending[0][2])):
478
if (self._node_key(self_pending[0][2]) ==
479
self._node_key(basis_pending[0][2])):
480
480
# Identical pointers, skip (and don't bother adding to
481
481
# excluded, it won't turn up again.
482
482
heapq.heappop(self_pending)
542
542
self._root_node = node_details[0][1]
544
544
self._root_node = InternalNode(prefix,
545
search_key_func=self._search_key_func)
545
search_key_func=self._search_key_func)
546
546
self._root_node.set_maximum_size(node_details[0][1].maximum_size)
547
547
self._root_node._key_width = node_details[0][1]._key_width
548
548
for split, node in node_details:
551
551
def _node_key(self, node):
552
552
"""Get the key for a node whether it's a tuple or node."""
553
if isinstance(node, tuple):
553
if type(node) is tuple:
554
554
node = StaticTuple.from_sequence(node)
555
if isinstance(node, StaticTuple):
555
if type(node) is StaticTuple:
561
561
"""remove key from the map."""
562
562
key = StaticTuple.from_sequence(key)
563
563
self._ensure_root()
564
if isinstance(self._root_node, InternalNode):
564
if type(self._root_node) is InternalNode:
565
565
unmapped = self._root_node.unmap(self._store, key,
566
check_remap=check_remap)
566
check_remap=check_remap)
568
568
unmapped = self._root_node.unmap(self._store, key)
569
569
self._root_node = unmapped
596
596
__slots__ = ('_key', '_len', '_maximum_size', '_key_width',
597
597
'_raw_size', '_items', '_search_prefix', '_search_key_func'
600
600
def __init__(self, key_width=1):
601
601
"""Create a node.
711
710
'%s(key:%s len:%s size:%s max:%s prefix:%s keywidth:%s items:%s)' \
712
711
% (self.__class__.__name__, self._key, self._len, self._raw_size,
713
self._maximum_size, self._search_prefix, self._key_width, items_str)
712
self._maximum_size, self._search_prefix, self._key_width, items_str)
715
714
def _current_size(self):
716
715
"""Answer the current serialised size of this node.
728
727
prefix_len = len(self._common_serialised_prefix)
729
728
bytes_for_items = (self._raw_size - (prefix_len * self._len))
730
return (9 + # 'chkleaf:\n' +
731
len(str(self._maximum_size)) + 1 +
732
len(str(self._key_width)) + 1 +
733
len(str(self._len)) + 1 +
729
return (9 # 'chkleaf:\n'
730
+ len(str(self._maximum_size)) + 1
731
+ len(str(self._key_width)) + 1
732
+ len(str(self._len)) + 1
738
737
def deserialise(klass, bytes, key, search_key_func=None):
769
768
# Short items, we need to match based on a prefix
770
filters.setdefault(len(key), set()).add(key)
769
length_filter = filters.setdefault(len(key), set())
770
length_filter.add(key)
772
filters_itemview = filters.items()
773
for item in self._items.items():
774
for length, length_filter in filters_itemview:
772
filters = filters.items()
773
for item in self._items.iteritems():
774
for length, length_filter in filters:
775
775
if item[0][:length] in length_filter:
779
yield from self._items.items()
779
for item in self._items.iteritems():
781
782
def _key_value_len(self, key, value):
782
783
# TODO: Should probably be done without actually joining the key, but
783
784
# then that can be done via the C extension
784
return (len(self._serialise_key(key)) + 1 +
785
len(b'%d' % value.count(b'\n')) + 1 +
785
return (len(self._serialise_key(key)) + 1
786
+ len(str(value.count('\n'))) + 1
788
789
def _search_key(self, key):
789
790
return self._search_key_func(key)
814
815
self._search_prefix = self.common_prefix(
815
816
self._search_prefix, search_key)
816
if (self._len > 1 and
817
self._maximum_size and
818
self._current_size() > self._maximum_size):
818
and self._maximum_size
819
and self._current_size() > self._maximum_size):
819
820
# Check to see if all of the search_keys for this node are
820
821
# identical. We allow the node to grow under that circumstance
821
822
# (we could track this as common state, but it is infrequent)
822
if (search_key != self._search_prefix or
823
not self._are_search_keys_identical()):
823
if (search_key != self._search_prefix
824
or not self._are_search_keys_identical()):
837
838
common_prefix = self._search_prefix
838
839
split_at = len(common_prefix) + 1
840
for key, value in self._items.items():
841
for key, value in self._items.iteritems():
841
842
search_key = self._search_key(key)
842
843
prefix = search_key[:split_at]
843
844
# TODO: Generally only 1 key can be exactly the right length,
849
850
# may get a '\00' node anywhere, but won't have keys of
850
851
# different lengths.
851
852
if len(prefix) < split_at:
852
prefix += b'\x00' * (split_at - len(prefix))
853
prefix += '\x00'*(split_at - len(prefix))
853
854
if prefix not in result:
854
855
node = LeafNode(search_key_func=self._search_key_func)
855
856
node.set_maximum_size(self._maximum_size)
865
866
result.pop(prefix)
866
867
new_node = InternalNode(sub_prefix,
867
search_key_func=self._search_key_func)
868
search_key_func=self._search_key_func)
868
869
new_node.set_maximum_size(self._maximum_size)
869
870
new_node._key_width = self._key_width
870
871
for split, node in node_details:
871
872
new_node.add_node(split, node)
872
873
result[prefix] = new_node
873
return common_prefix, list(result.items())
874
return common_prefix, result.items()
875
876
def map(self, store, key, value):
876
877
"""Map key to value."""
884
885
if self._search_prefix is _unknown:
885
886
raise AssertionError('%r must be known' % self._search_prefix)
886
return self._search_prefix, [(b"", self)]
887
return self._search_prefix, [("", self)]
888
_serialise_key = b'\x00'.join
889
_serialise_key = '\x00'.join
890
891
def serialise(self, store):
891
892
"""Serialise the LeafNode to store.
893
894
:param store: A VersionedFiles honouring the CHK extensions.
894
895
:return: An iterable of the keys inserted by this operation.
896
lines = [b"chkleaf:\n"]
897
lines.append(b"%d\n" % self._maximum_size)
898
lines.append(b"%d\n" % self._key_width)
899
lines.append(b"%d\n" % self._len)
897
lines = ["chkleaf:\n"]
898
lines.append("%d\n" % self._maximum_size)
899
lines.append("%d\n" % self._key_width)
900
lines.append("%d\n" % self._len)
900
901
if self._common_serialised_prefix is None:
902
903
if len(self._items) != 0:
903
904
raise AssertionError('If _common_serialised_prefix is None'
904
' we should have no items')
905
' we should have no items')
906
lines.append(b'%s\n' % (self._common_serialised_prefix,))
907
lines.append('%s\n' % (self._common_serialised_prefix,))
907
908
prefix_len = len(self._common_serialised_prefix)
908
909
for key, value in sorted(self._items.items()):
909
910
# Always add a final newline
910
value_lines = osutils.chunks_to_lines([value + b'\n'])
911
serialized = b"%s\x00%d\n" % (self._serialise_key(key),
911
value_lines = osutils.chunks_to_lines([value + '\n'])
912
serialized = "%s\x00%s\n" % (self._serialise_key(key),
913
914
if not serialized.startswith(self._common_serialised_prefix):
914
915
raise AssertionError('We thought the common prefix was %r'
915
' but entry %r does not have it in common'
916
% (self._common_serialised_prefix, serialized))
916
' but entry %r does not have it in common'
917
% (self._common_serialised_prefix, serialized))
917
918
lines.append(serialized[prefix_len:])
918
919
lines.extend(value_lines)
919
920
sha1, _, _ = store.add_lines((None,), (), lines)
920
self._key = StaticTuple(b"sha1:" + sha1,).intern()
921
data = b''.join(lines)
922
if len(data) != self._current_size():
921
self._key = StaticTuple("sha1:" + sha1,).intern()
922
bytes = ''.join(lines)
923
if len(bytes) != self._current_size():
923
924
raise AssertionError('Invalid _current_size')
924
_get_cache()[self._key] = data
925
_get_cache()[self._key] = bytes
925
926
return [self._key]
994
995
__slots__ = ('_node_width',)
996
def __init__(self, prefix=b'', search_key_func=None):
997
def __init__(self, prefix='', search_key_func=None):
997
998
Node.__init__(self)
998
999
# The size of an internalnode with default values and no children.
999
1000
# How many octets key prefixes within this node are.
1014
1015
raise AssertionError("_search_prefix should not be None")
1015
1016
if not prefix.startswith(self._search_prefix):
1016
1017
raise AssertionError("prefixes mismatch: %s must start with %s"
1017
% (prefix, self._search_prefix))
1018
% (prefix,self._search_prefix))
1018
1019
if len(prefix) != len(self._search_prefix) + 1:
1019
1020
raise AssertionError("prefix wrong length: len(%s) is not %d" %
1020
(prefix, len(self._search_prefix) + 1))
1021
(prefix, len(self._search_prefix) + 1))
1021
1022
self._len += len(node)
1022
1023
if not len(self._items):
1023
1024
self._node_width = len(prefix)
1024
1025
if self._node_width != len(self._search_prefix) + 1:
1025
1026
raise AssertionError("node width mismatch: %d is not %d" %
1026
(self._node_width, len(self._search_prefix) + 1))
1027
(self._node_width, len(self._search_prefix) + 1))
1027
1028
self._items[prefix] = node
1028
1029
self._key = None
1030
1031
def _current_size(self):
1031
1032
"""Answer the current serialised size of this node."""
1032
return (self._raw_size + len(str(self._len)) + len(str(self._key_width))
1033
+ len(str(self._maximum_size)))
1033
return (self._raw_size + len(str(self._len)) + len(str(self._key_width)) +
1034
len(str(self._maximum_size)))
1036
1037
def deserialise(klass, bytes, key, search_key_func=None):
1070
1071
# yielding all nodes, yield whatever we have, and queue up a read
1071
1072
# for whatever we are missing
1072
1073
shortcut = True
1073
for prefix, node in self._items.items():
1074
for prefix, node in self._items.iteritems():
1074
1075
if node.__class__ is StaticTuple:
1075
1076
keys[node] = (prefix, None)
1121
1122
for key in key_filter:
1122
1123
search_prefix = self._search_prefix_filter(key)
1123
1124
length_filter = length_filters.setdefault(
1124
len(search_prefix), set())
1125
len(search_prefix), set())
1125
1126
length_filter.add(search_prefix)
1126
1127
prefix_to_keys.setdefault(search_prefix, []).append(key)
1128
if (self._node_width in length_filters and
1129
len(length_filters) == 1):
1129
if (self._node_width in length_filters
1130
and len(length_filters) == 1):
1130
1131
# all of the search prefixes match exactly _node_width. This
1131
1132
# means that everything is an exact match, and we can do a
1132
1133
# lookup into self._items, rather than iterating over the items
1147
1148
# The slow way. We walk every item in self._items, and check to
1148
1149
# see if there are any matches
1149
length_filters_itemview = length_filters.items()
1150
for prefix, node in self._items.items():
1150
length_filters = length_filters.items()
1151
for prefix, node in self._items.iteritems():
1151
1152
node_key_filter = []
1152
for length, length_filter in length_filters_itemview:
1153
for length, length_filter in length_filters:
1153
1154
sub_prefix = prefix[:length]
1154
1155
if sub_prefix in length_filter:
1155
1156
node_key_filter.extend(prefix_to_keys[sub_prefix])
1156
if node_key_filter: # this key matched something, yield it
1157
if node_key_filter: # this key matched something, yield it
1157
1158
if node.__class__ is StaticTuple:
1158
1159
keys[node] = (prefix, node_key_filter)
1170
1171
node = _deserialise(bytes, key,
1171
search_key_func=self._search_key_func)
1172
search_key_func=self._search_key_func)
1172
1173
prefix, node_key_filter = keys[key]
1173
1174
self._items[prefix] = node
1174
1175
found_keys.add(key)
1190
1191
for record in stream:
1191
1192
bytes = record.get_bytes_as('fulltext')
1192
1193
node = _deserialise(bytes, record.key,
1193
search_key_func=self._search_key_func)
1194
search_key_func=self._search_key_func)
1194
1195
prefix, node_key_filter = keys[record.key]
1195
1196
node_and_filters.append((node, node_key_filter))
1196
1197
self._items[prefix] = node
1205
1206
search_key = self._search_key(key)
1206
1207
if self._node_width != len(self._search_prefix) + 1:
1207
1208
raise AssertionError("node width mismatch: %d is not %d" %
1208
(self._node_width, len(self._search_prefix) + 1))
1209
(self._node_width, len(self._search_prefix) + 1))
1209
1210
if not search_key.startswith(self._search_prefix):
1210
1211
# This key doesn't fit in this index, so we need to split at the
1211
1212
# point where it would fit, insert self into that internal node,
1213
1214
new_prefix = self.common_prefix(self._search_prefix,
1215
1216
new_parent = InternalNode(new_prefix,
1216
search_key_func=self._search_key_func)
1217
search_key_func=self._search_key_func)
1217
1218
new_parent.set_maximum_size(self._maximum_size)
1218
1219
new_parent._key_width = self._key_width
1219
new_parent.add_node(self._search_prefix[:len(new_prefix) + 1],
1220
new_parent.add_node(self._search_prefix[:len(new_prefix)+1],
1221
1222
return new_parent.map(store, key, value)
1222
children = [node for node, _ in self._iter_nodes(
1223
store, key_filter=[key])]
1223
children = [node for node, _
1224
in self._iter_nodes(store, key_filter=[key])]
1225
1226
child = children[0]
1227
1228
# new child needed:
1228
1229
child = self._new_child(search_key, LeafNode)
1229
1230
old_len = len(child)
1230
if isinstance(child, LeafNode):
1231
if type(child) is LeafNode:
1231
1232
old_size = child._current_size()
1233
1234
old_size = None
1239
1240
self._items[search_key] = child
1240
1241
self._key = None
1241
1242
new_node = self
1242
if isinstance(child, LeafNode):
1243
if type(child) is LeafNode:
1243
1244
if old_size is None:
1244
1245
# The old node was an InternalNode which means it has now
1245
1246
# collapsed, so we need to check if it will chain to a
1255
1256
# amount is over a configurable limit.
1256
1257
new_size = child._current_size()
1257
1258
shrinkage = old_size - new_size
1258
if (shrinkage > 0 and new_size < _INTERESTING_NEW_SIZE or
1259
shrinkage > _INTERESTING_SHRINKAGE_LIMIT):
1259
if (shrinkage > 0 and new_size < _INTERESTING_NEW_SIZE
1260
or shrinkage > _INTERESTING_SHRINKAGE_LIMIT):
1261
1262
"checking remap as size shrunk by %d to be %d",
1262
1263
shrinkage, new_size)
1263
1264
new_node = self._check_remap(store)
1264
1265
if new_node._search_prefix is None:
1265
1266
raise AssertionError("_search_prefix should not be None")
1266
return new_node._search_prefix, [(b'', new_node)]
1267
return new_node._search_prefix, [('', new_node)]
1267
1268
# child has overflown - create a new intermediate node.
1268
1269
# XXX: This is where we might want to try and expand our depth
1269
1270
# to refer to more bytes of every child (which would give us
1274
1275
child.add_node(split, node)
1275
1276
self._len = self._len - old_len + len(child)
1276
1277
self._key = None
1277
return self._search_prefix, [(b"", self)]
1278
return self._search_prefix, [("", self)]
1279
1280
def _new_child(self, search_key, klass):
1280
1281
"""Create a new child node of type klass."""
1291
1292
:param store: A VersionedFiles honouring the CHK extensions.
1292
1293
:return: An iterable of the keys inserted by this operation.
1294
for node in self._items.values():
1295
if isinstance(node, StaticTuple):
1295
for node in self._items.itervalues():
1296
if type(node) is StaticTuple:
1296
1297
# Never deserialised.
1298
1299
if node._key is not None:
1301
1302
for key in node.serialise(store):
1303
lines = [b"chknode:\n"]
1304
lines.append(b"%d\n" % self._maximum_size)
1305
lines.append(b"%d\n" % self._key_width)
1306
lines.append(b"%d\n" % self._len)
1304
lines = ["chknode:\n"]
1305
lines.append("%d\n" % self._maximum_size)
1306
lines.append("%d\n" % self._key_width)
1307
lines.append("%d\n" % self._len)
1307
1308
if self._search_prefix is None:
1308
1309
raise AssertionError("_search_prefix should not be None")
1309
lines.append(b'%s\n' % (self._search_prefix,))
1310
lines.append('%s\n' % (self._search_prefix,))
1310
1311
prefix_len = len(self._search_prefix)
1311
1312
for prefix, node in sorted(self._items.items()):
1312
if isinstance(node, StaticTuple):
1313
if type(node) is StaticTuple:
1315
1316
key = node._key[0]
1316
serialised = b"%s\x00%s\n" % (prefix, key)
1317
serialised = "%s\x00%s\n" % (prefix, key)
1317
1318
if not serialised.startswith(self._search_prefix):
1318
1319
raise AssertionError("prefixes mismatch: %s must start with %s"
1319
% (serialised, self._search_prefix))
1320
% (serialised, self._search_prefix))
1320
1321
lines.append(serialised[prefix_len:])
1321
1322
sha1, _, _ = store.add_lines((None,), (), lines)
1322
self._key = StaticTuple(b"sha1:" + sha1,).intern()
1323
_get_cache()[self._key] = b''.join(lines)
1323
self._key = StaticTuple("sha1:" + sha1,).intern()
1324
_get_cache()[self._key] = ''.join(lines)
1324
1325
yield self._key
1326
1327
def _search_key(self, key):
1327
1328
"""Return the serialised key for key in this node."""
1328
1329
# search keys are fixed width. All will be self._node_width wide, so we
1329
1330
# pad as necessary.
1330
return (self._search_key_func(key) + b'\x00' * self._node_width)[:self._node_width]
1331
return (self._search_key_func(key) + '\x00'*self._node_width)[:self._node_width]
1332
1333
def _search_prefix_filter(self, key):
1333
1334
"""Serialise key for use as a prefix filter in iteritems."""
1341
1342
prefix for reaching node.
1343
1344
if offset >= self._node_width:
1344
for node in valueview(self._items):
1345
for node in self._items.values():
1345
1346
for result in node._split(offset):
1349
for key, node in self._items.items():
1348
1352
def refs(self):
1349
1353
"""Return the references to other CHK's held by this node."""
1350
1354
if self._key is None:
1351
1355
raise AssertionError("unserialised nodes have no refs.")
1353
for value in self._items.values():
1354
if isinstance(value, StaticTuple):
1357
for value in self._items.itervalues():
1358
if type(value) is StaticTuple:
1355
1359
refs.append(value)
1357
1361
refs.append(value.key())
1389
1393
self._items[search_key] = unmapped
1390
1394
if len(self._items) == 1:
1391
1395
# this node is no longer needed:
1392
return list(self._items.values())[0]
1393
if isinstance(unmapped, InternalNode):
1396
return self._items.values()[0]
1397
if type(unmapped) is InternalNode:
1395
1399
if check_remap:
1396
1400
return self._check_remap(store)
1436
1440
# c) With 255-way fan out, we don't want to read all 255 and destroy
1437
1441
# the page cache, just to determine that we really don't need it.
1438
1442
for node, _ in self._iter_nodes(store, batch_size=16):
1439
if isinstance(node, InternalNode):
1443
if type(node) is InternalNode:
1440
1444
# Without looking at any leaf nodes, we are sure
1442
for key, value in node._items.items():
1446
for key, value in node._items.iteritems():
1443
1447
if new_leaf._map_no_split(key, value):
1445
1449
trace.mutter("remap generated a new LeafNode")
1446
1450
return new_leaf
1449
def _deserialise(data, key, search_key_func):
1453
def _deserialise(bytes, key, search_key_func):
1450
1454
"""Helper for repositorydetails - convert bytes to a node."""
1451
if data.startswith(b"chkleaf:\n"):
1452
node = LeafNode.deserialise(data, key, search_key_func=search_key_func)
1453
elif data.startswith(b"chknode:\n"):
1454
node = InternalNode.deserialise(data, key,
1455
search_key_func=search_key_func)
1455
if bytes.startswith("chkleaf:\n"):
1456
node = LeafNode.deserialise(bytes, key, search_key_func=search_key_func)
1457
elif bytes.startswith("chknode:\n"):
1458
node = InternalNode.deserialise(bytes, key,
1459
search_key_func=search_key_func)
1457
1461
raise AssertionError("Unknown node type.")
1521
1525
bytes = record.get_bytes_as('fulltext')
1522
1526
node = _deserialise(bytes, record.key,
1523
1527
search_key_func=self._search_key_func)
1524
if isinstance(node, InternalNode):
1528
if type(node) is InternalNode:
1525
1529
# Note we don't have to do node.refs() because we know that
1526
1530
# there are no children that have been pushed into this node
1527
1531
# Note: Using as_st() here seemed to save 1.2MB, which would
1528
1532
# indicate that we keep 100k prefix_refs around while
1529
1533
# processing. They *should* be shorter lived than that...
1530
1534
# It does cost us ~10s of processing time
1531
prefix_refs = list(node._items.items())
1535
#prefix_refs = [as_st(item) for item in node._items.iteritems()]
1536
prefix_refs = node._items.items()
1534
1539
prefix_refs = []
1535
1540
# Note: We don't use a StaticTuple here. Profiling showed a
1536
1541
# minor memory improvement (0.8MB out of 335MB peak 0.2%)
1537
1542
# But a significant slowdown (15s / 145s, or 10%)
1538
items = list(node._items.items())
1543
items = node._items.items()
1539
1544
yield record, node, prefix_refs, items
1541
1546
def _read_old_roots(self):
1545
1550
self._read_nodes_from_store(self._old_root_keys):
1546
1551
# Uninteresting node
1547
1552
prefix_refs = [p_r for p_r in prefix_refs
1548
if p_r[1] not in all_old_chks]
1553
if p_r[1] not in all_old_chks]
1549
1554
new_refs = [p_r[1] for p_r in prefix_refs]
1550
1555
all_old_chks.update(new_refs)
1551
1556
# TODO: This might be a good time to turn items into StaticTuple
1565
1570
# handled the interesting ones
1566
1571
for prefix, ref in old_chks_to_enqueue:
1567
1572
not_interesting = True
1568
for i in range(len(prefix), 0, -1):
1573
for i in xrange(len(prefix), 0, -1):
1569
1574
if prefix[:i] in new_prefixes:
1570
1575
not_interesting = False
1601
1606
# At this level, we now know all the uninteresting references
1602
1607
# So we filter and queue up whatever is remaining
1603
1608
prefix_refs = [p_r for p_r in prefix_refs
1604
if p_r[1] not in self._all_old_chks and
1605
p_r[1] not in processed_new_refs]
1609
if p_r[1] not in self._all_old_chks
1610
and p_r[1] not in processed_new_refs]
1606
1611
refs = [p_r[1] for p_r in prefix_refs]
1607
1612
new_prefixes.update([p_r[0] for p_r in prefix_refs])
1608
1613
self._new_queue.extend(refs)
1614
1619
# self._new_item_queue will hold the contents of multiple
1615
1620
# records for an extended lifetime
1616
1621
new_items = [item for item in items
1617
if item not in self._all_old_items]
1622
if item not in self._all_old_items]
1618
1623
self._new_item_queue.extend(new_items)
1619
1624
new_prefixes.update([self._search_key_func(item[0])
1620
1625
for item in new_items])
1625
1630
# 'ab', then we also need to include 'a'.) So expand the
1626
1631
# new_prefixes to include all shorter prefixes
1627
1632
for prefix in list(new_prefixes):
1628
new_prefixes.update([prefix[:i] for i in range(1, len(prefix))])
1633
new_prefixes.update([prefix[:i] for i in xrange(1, len(prefix))])
1629
1634
self._enqueue_old(new_prefixes, old_chks_to_enqueue)
1631
1636
def _flush_new_queue(self):
1638
1643
processed_new_refs = self._processed_new_refs
1639
1644
all_old_items = self._all_old_items
1640
1645
new_items = [item for item in self._new_item_queue
1641
if item not in all_old_items]
1646
if item not in all_old_items]
1642
1647
self._new_item_queue = []
1644
1649
yield None, new_items
1682
1687
for record, _, prefix_refs, items in self._read_nodes_from_store(refs):
1683
1688
# TODO: Use StaticTuple here?
1684
1689
self._all_old_items.update(items)
1685
refs = [r for _, r in prefix_refs if r not in all_old_chks]
1690
refs = [r for _,r in prefix_refs if r not in all_old_chks]
1686
1691
self._old_queue.extend(refs)
1687
1692
all_old_chks.update(refs)
1723
from ._chk_map_pyx import (
1728
from brzlib._chk_map_pyx import (
1724
1729
_bytes_to_text_key,
1725
1730
_search_key_16,
1726
1731
_search_key_255,
1727
1732
_deserialise_leaf_node,
1728
1733
_deserialise_internal_node,
1730
except ImportError as e:
1735
except ImportError, e:
1731
1736
osutils.failed_to_load_extension(e)
1732
from ._chk_map_py import (
1737
from brzlib._chk_map_py import (
1733
1738
_bytes_to_text_key,
1734
1739
_search_key_16,
1735
1740
_search_key_255,
1736
1741
_deserialise_leaf_node,
1737
1742
_deserialise_internal_node,
1739
search_key_registry.register(b'hash-16-way', _search_key_16)
1740
search_key_registry.register(b'hash-255-way', _search_key_255)
1744
search_key_registry.register('hash-16-way', _search_key_16)
1745
search_key_registry.register('hash-255-way', _search_key_255)
1743
1748
def _check_key(key):
1746
1751
This generally shouldn't be used in production code, but it can be helpful
1747
1752
to debug problems.
1749
if not isinstance(key, StaticTuple):
1754
if type(key) is not StaticTuple:
1750
1755
raise TypeError('key %r is not StaticTuple but %s' % (key, type(key)))
1751
1756
if len(key) != 1:
1752
raise ValueError('key %r should have length 1, not %d' %
1754
if not isinstance(key[0], str):
1757
raise ValueError('key %r should have length 1, not %d' % (key, len(key),))
1758
if type(key[0]) is not str:
1755
1759
raise TypeError('key %r should hold a str, not %r'
1756
1760
% (key, type(key[0])))
1757
1761
if not key[0].startswith('sha1:'):
1758
1762
raise ValueError('key %r should point to a sha1:' % (key,))
added
removed
brzlib/chk_map.py
