40
from __future__ import absolute_import
43
from bzrlib import lazy_import
45
from . import lazy_import
44
46
lazy_import.lazy_import(globals(), """
57
from bzrlib.static_tuple import StaticTuple
63
from .static_tuple import StaticTuple
60
66
# If each line is 50 bytes, and you have 255 internal pages, with 255-way fan
90
96
_INTERESTING_NEW_SIZE = 50
91
97
# If a ChildNode shrinks by more than this amount, we check for a remap
92
98
_INTERESTING_SHRINKAGE_LIMIT = 20
93
# If we delete more than this many nodes applying a delta, we check for a remap
94
_INTERESTING_DELETES_LIMIT = 5
97
101
def _search_key_plain(key):
98
102
"""Map the key tuple into a search string that just uses the key bytes."""
99
return '\x00'.join(key)
103
return b'\x00'.join(key)
102
106
search_key_registry = registry.Registry()
135
139
into the map; if old_key is not None, then the old mapping
136
140
of old_key is removed.
139
143
# Check preconditions first.
140
144
as_st = StaticTuple.from_sequence
141
new_items = set([as_st(key) for (old, key, value) in delta
142
if key is not None and old is None])
145
new_items = {as_st(key) for (old, key, value) in delta
146
if key is not None and old is None}
143
147
existing_new = list(self.iteritems(key_filter=new_items))
145
149
raise errors.InconsistentDeltaDelta(delta,
148
152
for old, new, value in delta:
149
153
if old is not None and old != new:
150
154
self.unmap(old, check_remap=False)
152
156
for old, new, value in delta:
153
157
if new is not None:
154
158
self.map(new, value)
155
if delete_count > _INTERESTING_DELETES_LIMIT:
156
trace.mutter("checking remap as %d deletions", delete_count)
157
160
self._check_remap()
158
161
return self._save()
160
163
def _ensure_root(self):
161
164
"""Ensure that the root node is an object not a key."""
162
if type(self._root_node) is StaticTuple:
165
if isinstance(self._root_node, StaticTuple):
163
166
# Demand-load the root
164
167
self._root_node = self._get_node(self._root_node)
173
176
:param node: A tuple key or node object.
174
177
:return: A node object.
176
if type(node) is StaticTuple:
179
if isinstance(node, StaticTuple):
177
180
bytes = self._read_bytes(node)
178
181
return _deserialise(bytes, node,
179
182
search_key_func=self._search_key_func)
194
197
self._ensure_root()
195
198
res = self._dump_tree_node(self._root_node, prefix='', indent='',
196
199
include_keys=include_keys)
197
res.append('') # Give a trailing '\n'
198
return '\n'.join(res)
200
res.append(b'') # Give a trailing '\n'
201
return b'\n'.join(res)
200
203
def _dump_tree_node(self, node, prefix, indent, include_keys=True):
201
204
"""For this node and all children, generate a string representation."""
206
209
node_key = node.key()
207
210
if node_key is not None:
208
key_str = ' %s' % (node_key[0],)
211
key_str = b' %s' % (node_key[0],)
211
result.append('%s%r %s%s' % (indent, prefix, node.__class__.__name__,
213
if type(node) is InternalNode:
214
result.append(b'%s%r %s%s' % (indent, prefix, node.__class__.__name__,
216
if isinstance(node, InternalNode):
214
217
# Trigger all child nodes to get loaded
215
218
list(node._iter_nodes(self._store))
216
for prefix, sub in sorted(node._items.iteritems()):
219
for prefix, sub in sorted(viewitems(node._items)):
217
220
result.extend(self._dump_tree_node(sub, prefix, indent + ' ',
218
221
include_keys=include_keys))
220
for key, value in sorted(node._items.iteritems()):
223
for key, value in sorted(viewitems(node._items)):
221
224
# Don't use prefix nor indent here to line up when used in
222
225
# tests in conjunction with assertEqualDiff
223
result.append(' %r %r' % (tuple(key), value))
226
result.append(b' %r %r' % (tuple(key), value))
244
247
root_key = klass._create_directly(store, initial_value,
245
248
maximum_size=maximum_size, key_width=key_width,
246
249
search_key_func=search_key_func)
247
if type(root_key) is not StaticTuple:
250
if not isinstance(root_key, StaticTuple):
248
251
raise AssertionError('we got a %s instead of a StaticTuple'
249
252
% (type(root_key),))
256
259
result._root_node.set_maximum_size(maximum_size)
257
260
result._root_node._key_width = key_width
259
for key, value in initial_value.items():
262
for key, value in viewitems(initial_value):
260
263
delta.append((None, key, value))
261
264
root_key = result.apply_delta(delta)
268
271
node.set_maximum_size(maximum_size)
269
272
node._key_width = key_width
270
273
as_st = StaticTuple.from_sequence
271
node._items = dict([(as_st(key), val) for key, val
272
in initial_value.iteritems()])
273
node._raw_size = sum([node._key_value_len(key, value)
274
for key,value in node._items.iteritems()])
274
node._items = dict((as_st(key), val)
275
for key, val in viewitems(initial_value))
276
node._raw_size = sum(node._key_value_len(key, value)
277
for key, value in viewitems(node._items))
275
278
node._len = len(node._items)
276
279
node._compute_search_prefix()
277
280
node._compute_serialised_prefix()
332
335
def process_node(node, path, a_map, pending):
333
336
# take a node and expand it
334
337
node = a_map._get_node(node)
335
if type(node) == LeafNode:
338
if isinstance(node, LeafNode):
336
339
path = (node._key, path)
337
for key, value in node._items.items():
340
for key, value in viewitems(node._items):
338
341
# For a LeafNode, the key is a serialized_key, rather than
339
342
# a search_key, but the heap is using search_keys
340
343
search_key = node._search_key_func(key)
343
346
# type(node) == InternalNode
344
347
path = (node._key, path)
345
for prefix, child in node._items.items():
348
for prefix, child in viewitems(node._items):
346
349
heapq.heappush(pending, (prefix, None, child, path))
347
350
def process_common_internal_nodes(self_node, basis_node):
348
self_items = set(self_node._items.items())
349
basis_items = set(basis_node._items.items())
351
self_items = set(viewitems(self_node._items))
352
basis_items = set(viewitems(basis_node._items))
350
353
path = (self_node._key, None)
351
354
for prefix, child in self_items - basis_items:
352
355
heapq.heappush(self_pending, (prefix, None, child, path))
354
357
for prefix, child in basis_items - self_items:
355
358
heapq.heappush(basis_pending, (prefix, None, child, path))
356
359
def process_common_leaf_nodes(self_node, basis_node):
357
self_items = set(self_node._items.items())
358
basis_items = set(basis_node._items.items())
360
self_items = set(viewitems(self_node._items))
361
basis_items = set(viewitems(basis_node._items))
359
362
path = (self_node._key, None)
360
363
for key, value in self_items - basis_items:
361
364
prefix = self._search_key_func(key)
371
374
self_node = self._get_node(self_node)
372
375
basis_node = basis._get_node(basis_node)
373
if (type(self_node) == InternalNode
374
and type(basis_node) == InternalNode):
376
if (isinstance(self_node, InternalNode)
377
and isinstance(basis_node, InternalNode)):
375
378
# Matching internal nodes
376
379
process_common_internal_nodes(self_node, basis_node)
377
elif (type(self_node) == LeafNode
378
and type(basis_node) == LeafNode):
380
elif (isinstance(self_node, LeafNode)
381
and isinstance(basis_node, LeafNode)):
379
382
process_common_leaf_nodes(self_node, basis_node)
381
384
process_node(self_node, self_path, self, self_pending)
390
393
# A better implementation would probably have a reverse map
391
394
# back to the children of a node, and jump straight to it when
392
395
# a common node is detected, the proceed to remove the already
393
# pending children. bzrlib.graph has a searcher module with a
396
# pending children. breezy.graph has a searcher module with a
394
397
# similar problem.
395
398
while key_path is not None:
396
399
key, key_path = key_path
552
555
def _node_key(self, node):
553
556
"""Get the key for a node whether it's a tuple or node."""
554
if type(node) is tuple:
557
if isinstance(node, tuple):
555
558
node = StaticTuple.from_sequence(node)
556
if type(node) is StaticTuple:
559
if isinstance(node, StaticTuple):
562
565
"""remove key from the map."""
563
566
key = StaticTuple.from_sequence(key)
564
567
self._ensure_root()
565
if type(self._root_node) is InternalNode:
568
if isinstance(self._root_node, InternalNode):
566
569
unmapped = self._root_node.unmap(self._store, key,
567
570
check_remap=check_remap)
572
575
def _check_remap(self):
573
576
"""Check if nodes can be collapsed."""
574
577
self._ensure_root()
575
if type(self._root_node) is InternalNode:
576
self._root_node._check_remap(self._store)
578
if isinstance(self._root_node, InternalNode):
579
self._root_node = self._root_node._check_remap(self._store)
579
582
"""Save the map completely.
581
584
:return: The key of the root node.
583
if type(self._root_node) is StaticTuple:
586
if isinstance(self._root_node, StaticTuple):
585
588
return self._root_node
586
589
keys = list(self._root_node.serialise(self._store))
769
772
# Short items, we need to match based on a prefix
770
length_filter = filters.setdefault(len(key), set())
771
length_filter.add(key)
773
filters.setdefault(len(key), set()).add(key)
773
filters = filters.items()
774
for item in self._items.iteritems():
775
for length, length_filter in filters:
775
filters_itemview = viewitems(filters)
776
for item in viewitems(self._items):
777
for length, length_filter in filters_itemview:
776
778
if item[0][:length] in length_filter:
780
for item in self._items.iteritems():
782
for item in viewitems(self._items):
783
785
def _key_value_len(self, key, value):
784
786
# TODO: Should probably be done without actually joining the key, but
785
787
# then that can be done via the C extension
786
788
return (len(self._serialise_key(key)) + 1
787
+ len(str(value.count('\n'))) + 1
789
+ len(str(value.count(b'\n'))) + 1
788
790
+ len(value) + 1)
790
792
def _search_key(self, key):
839
841
common_prefix = self._search_prefix
840
842
split_at = len(common_prefix) + 1
842
for key, value in self._items.iteritems():
844
for key, value in viewitems(self._items):
843
845
search_key = self._search_key(key)
844
846
prefix = search_key[:split_at]
845
847
# TODO: Generally only 1 key can be exactly the right length,
851
853
# may get a '\00' node anywhere, but won't have keys of
852
854
# different lengths.
853
855
if len(prefix) < split_at:
854
prefix += '\x00'*(split_at - len(prefix))
856
prefix += b'\x00'*(split_at - len(prefix))
855
857
if prefix not in result:
856
858
node = LeafNode(search_key_func=self._search_key_func)
857
859
node.set_maximum_size(self._maximum_size)
872
874
for split, node in node_details:
873
875
new_node.add_node(split, node)
874
876
result[prefix] = new_node
875
return common_prefix, result.items()
877
return common_prefix, list(viewitems(result))
877
879
def map(self, store, key, value):
878
880
"""Map key to value."""
887
889
raise AssertionError('%r must be known' % self._search_prefix)
888
890
return self._search_prefix, [("", self)]
890
_serialise_key = '\x00'.join
892
_serialise_key = b'\x00'.join
892
894
def serialise(self, store):
893
895
"""Serialise the LeafNode to store.
895
897
:param store: A VersionedFiles honouring the CHK extensions.
896
898
:return: An iterable of the keys inserted by this operation.
898
lines = ["chkleaf:\n"]
899
lines.append("%d\n" % self._maximum_size)
900
lines.append("%d\n" % self._key_width)
901
lines.append("%d\n" % self._len)
900
lines = [b"chkleaf:\n"]
901
lines.append(b"%d\n" % self._maximum_size)
902
lines.append(b"%d\n" % self._key_width)
903
lines.append(b"%d\n" % self._len)
902
904
if self._common_serialised_prefix is None:
904
906
if len(self._items) != 0:
905
907
raise AssertionError('If _common_serialised_prefix is None'
906
908
' we should have no items')
908
lines.append('%s\n' % (self._common_serialised_prefix,))
910
lines.append(b'%s\n' % (self._common_serialised_prefix,))
909
911
prefix_len = len(self._common_serialised_prefix)
910
for key, value in sorted(self._items.items()):
912
for key, value in sorted(viewitems(self._items)):
911
913
# Always add a final newline
912
value_lines = osutils.chunks_to_lines([value + '\n'])
913
serialized = "%s\x00%s\n" % (self._serialise_key(key),
914
value_lines = osutils.chunks_to_lines([value + b'\n'])
915
serialized = b"%s\x00%d\n" % (self._serialise_key(key),
914
916
len(value_lines))
915
917
if not serialized.startswith(self._common_serialised_prefix):
916
918
raise AssertionError('We thought the common prefix was %r'
919
921
lines.append(serialized[prefix_len:])
920
922
lines.extend(value_lines)
921
923
sha1, _, _ = store.add_lines((None,), (), lines)
922
self._key = StaticTuple("sha1:" + sha1,).intern()
923
bytes = ''.join(lines)
924
if len(bytes) != self._current_size():
924
self._key = StaticTuple(b"sha1:" + sha1,).intern()
925
data = b''.join(lines)
926
if len(data) != self._current_size():
925
927
raise AssertionError('Invalid _current_size')
926
_get_cache().add(self._key, bytes)
928
_get_cache()[self._key] = data
927
929
return [self._key]
1072
1074
# yielding all nodes, yield whatever we have, and queue up a read
1073
1075
# for whatever we are missing
1074
1076
shortcut = True
1075
for prefix, node in self._items.iteritems():
1077
for prefix, node in viewitems(self._items):
1076
1078
if node.__class__ is StaticTuple:
1077
1079
keys[node] = (prefix, None)
1149
1151
# The slow way. We walk every item in self._items, and check to
1150
1152
# see if there are any matches
1151
length_filters = length_filters.items()
1152
for prefix, node in self._items.iteritems():
1153
length_filters_itemview = viewitems(length_filters)
1154
for prefix, node in viewitems(self._items):
1153
1155
node_key_filter = []
1154
for length, length_filter in length_filters:
1156
for length, length_filter in length_filters_itemview:
1155
1157
sub_prefix = prefix[:length]
1156
1158
if sub_prefix in length_filter:
1157
1159
node_key_filter.extend(prefix_to_keys[sub_prefix])
1196
1198
prefix, node_key_filter = keys[record.key]
1197
1199
node_and_filters.append((node, node_key_filter))
1198
1200
self._items[prefix] = node
1199
_get_cache().add(record.key, bytes)
1201
_get_cache()[record.key] = bytes
1200
1202
for info in node_and_filters:
1241
1243
self._items[search_key] = child
1242
1244
self._key = None
1243
1245
new_node = self
1244
if type(child) is LeafNode:
1246
if isinstance(child, LeafNode):
1245
1247
if old_size is None:
1246
1248
# The old node was an InternalNode which means it has now
1247
1249
# collapsed, so we need to check if it will chain to a
1293
1295
:param store: A VersionedFiles honouring the CHK extensions.
1294
1296
:return: An iterable of the keys inserted by this operation.
1296
for node in self._items.itervalues():
1297
if type(node) is StaticTuple:
1298
for node in viewvalues(self._items):
1299
if isinstance(node, StaticTuple):
1298
1300
# Never deserialised.
1300
1302
if node._key is not None:
1303
1305
for key in node.serialise(store):
1305
lines = ["chknode:\n"]
1306
lines.append("%d\n" % self._maximum_size)
1307
lines.append("%d\n" % self._key_width)
1308
lines.append("%d\n" % self._len)
1307
lines = [b"chknode:\n"]
1308
lines.append(b"%d\n" % self._maximum_size)
1309
lines.append(b"%d\n" % self._key_width)
1310
lines.append(b"%d\n" % self._len)
1309
1311
if self._search_prefix is None:
1310
1312
raise AssertionError("_search_prefix should not be None")
1311
lines.append('%s\n' % (self._search_prefix,))
1313
lines.append(b'%s\n' % (self._search_prefix,))
1312
1314
prefix_len = len(self._search_prefix)
1313
for prefix, node in sorted(self._items.items()):
1314
if type(node) is StaticTuple:
1315
for prefix, node in sorted(viewitems(self._items)):
1316
if isinstance(node, StaticTuple):
1317
1319
key = node._key[0]
1318
serialised = "%s\x00%s\n" % (prefix, key)
1320
serialised = b"%s\x00%s\n" % (prefix, key)
1319
1321
if not serialised.startswith(self._search_prefix):
1320
1322
raise AssertionError("prefixes mismatch: %s must start with %s"
1321
1323
% (serialised, self._search_prefix))
1322
1324
lines.append(serialised[prefix_len:])
1323
1325
sha1, _, _ = store.add_lines((None,), (), lines)
1324
self._key = StaticTuple("sha1:" + sha1,).intern()
1325
_get_cache().add(self._key, ''.join(lines))
1326
self._key = StaticTuple(b"sha1:" + sha1,).intern()
1327
_get_cache()[self._key] = b''.join(lines)
1326
1328
yield self._key
1328
1330
def _search_key(self, key):
1329
1331
"""Return the serialised key for key in this node."""
1330
1332
# search keys are fixed width. All will be self._node_width wide, so we
1331
1333
# pad as necessary.
1332
return (self._search_key_func(key) + '\x00'*self._node_width)[:self._node_width]
1334
return (self._search_key_func(key) + b'\x00'*self._node_width)[:self._node_width]
1334
1336
def _search_prefix_filter(self, key):
1335
1337
"""Serialise key for use as a prefix filter in iteritems."""
1343
1345
prefix for reaching node.
1345
1347
if offset >= self._node_width:
1346
for node in self._items.values():
1348
for node in valueview(self._items):
1347
1349
for result in node._split(offset):
1350
for key, node in self._items.items():
1353
1352
def refs(self):
1354
1353
"""Return the references to other CHK's held by this node."""
1355
1354
if self._key is None:
1356
1355
raise AssertionError("unserialised nodes have no refs.")
1358
for value in self._items.itervalues():
1359
if type(value) is StaticTuple:
1357
for value in viewvalues(self._items):
1358
if isinstance(value, StaticTuple):
1360
1359
refs.append(value)
1362
1361
refs.append(value.key())
1372
1371
return self._search_prefix
1374
1373
def unmap(self, store, key, check_remap=True):
1375
"""Remove key from this node and it's children."""
1374
"""Remove key from this node and its children."""
1376
1375
if not len(self._items):
1377
1376
raise AssertionError("can't unmap in an empty InternalNode.")
1378
1377
children = [node for node, _
1394
1393
self._items[search_key] = unmapped
1395
1394
if len(self._items) == 1:
1396
1395
# this node is no longer needed:
1397
return self._items.values()[0]
1398
if type(unmapped) is InternalNode:
1396
return list(viewvalues(self._items))[0]
1397
if isinstance(unmapped, InternalNode):
1400
1399
if check_remap:
1401
1400
return self._check_remap(store)
1441
1440
# c) With 255-way fan out, we don't want to read all 255 and destroy
1442
1441
# the page cache, just to determine that we really don't need it.
1443
1442
for node, _ in self._iter_nodes(store, batch_size=16):
1444
if type(node) is InternalNode:
1443
if isinstance(node, InternalNode):
1445
1444
# Without looking at any leaf nodes, we are sure
1447
for key, value in node._items.iteritems():
1446
for key, value in viewitems(node._items):
1448
1447
if new_leaf._map_no_split(key, value):
1450
1449
trace.mutter("remap generated a new LeafNode")
1451
1450
return new_leaf
1454
def _deserialise(bytes, key, search_key_func):
1453
def _deserialise(data, key, search_key_func):
1455
1454
"""Helper for repositorydetails - convert bytes to a node."""
1456
if bytes.startswith("chkleaf:\n"):
1457
node = LeafNode.deserialise(bytes, key, search_key_func=search_key_func)
1458
elif bytes.startswith("chknode:\n"):
1459
node = InternalNode.deserialise(bytes, key,
1455
if data.startswith(b"chkleaf:\n"):
1456
node = LeafNode.deserialise(data, key, search_key_func=search_key_func)
1457
elif data.startswith(b"chknode:\n"):
1458
node = InternalNode.deserialise(data, key,
1460
1459
search_key_func=search_key_func)
1462
1461
raise AssertionError("Unknown node type.")
1526
1525
bytes = record.get_bytes_as('fulltext')
1527
1526
node = _deserialise(bytes, record.key,
1528
1527
search_key_func=self._search_key_func)
1529
if type(node) is InternalNode:
1528
if isinstance(node, InternalNode):
1530
1529
# Note we don't have to do node.refs() because we know that
1531
1530
# there are no children that have been pushed into this node
1532
1531
# Note: Using as_st() here seemed to save 1.2MB, which would
1533
1532
# indicate that we keep 100k prefix_refs around while
1534
1533
# processing. They *should* be shorter lived than that...
1535
1534
# It does cost us ~10s of processing time
1536
#prefix_refs = [as_st(item) for item in node._items.iteritems()]
1537
prefix_refs = node._items.items()
1535
prefix_refs = list(viewitems(node._items))
1540
1538
prefix_refs = []
1541
1539
# Note: We don't use a StaticTuple here. Profiling showed a
1542
1540
# minor memory improvement (0.8MB out of 335MB peak 0.2%)
1543
1541
# But a significant slowdown (15s / 145s, or 10%)
1544
items = node._items.items()
1542
items = list(viewitems(node._items))
1545
1543
yield record, node, prefix_refs, items
1547
1545
def _read_old_roots(self):
1571
1569
# handled the interesting ones
1572
1570
for prefix, ref in old_chks_to_enqueue:
1573
1571
not_interesting = True
1574
for i in xrange(len(prefix), 0, -1):
1572
for i in range(len(prefix), 0, -1):
1575
1573
if prefix[:i] in new_prefixes:
1576
1574
not_interesting = False
1631
1629
# 'ab', then we also need to include 'a'.) So expand the
1632
1630
# new_prefixes to include all shorter prefixes
1633
1631
for prefix in list(new_prefixes):
1634
new_prefixes.update([prefix[:i] for i in xrange(1, len(prefix))])
1632
new_prefixes.update([prefix[:i] for i in range(1, len(prefix))])
1635
1633
self._enqueue_old(new_prefixes, old_chks_to_enqueue)
1637
1635
def _flush_new_queue(self):
1729
from bzrlib._chk_map_pyx import (
1727
from breezy._chk_map_pyx import (
1730
1729
_search_key_16,
1731
1730
_search_key_255,
1732
1731
_deserialise_leaf_node,
1733
1732
_deserialise_internal_node,
1735
except ImportError, e:
1734
except ImportError as e:
1736
1735
osutils.failed_to_load_extension(e)
1737
from bzrlib._chk_map_py import (
1736
from breezy._chk_map_py import (
1738
1738
_search_key_16,
1739
1739
_search_key_255,
1740
1740
_deserialise_leaf_node,
1750
1750
This generally shouldn't be used in production code, but it can be helpful
1751
1751
to debug problems.
1753
if type(key) is not StaticTuple:
1753
if not isinstance(key, StaticTuple):
1754
1754
raise TypeError('key %r is not StaticTuple but %s' % (key, type(key)))
1755
1755
if len(key) != 1:
1756
1756
raise ValueError('key %r should have length 1, not %d' % (key, len(key),))
1757
if type(key[0]) is not str:
1757
if not isinstance(key[0], str):
1758
1758
raise TypeError('key %r should hold a str, not %r'
1759
1759
% (key, type(key[0])))
1760
1760
if not key[0].startswith('sha1:'):
added
removed
breezy/chk_map.py
