139
152
texts/deltas (via (fileid, revisionid) tuples).
140
153
:param signature_index: A GraphIndex for determining what signatures are
141
154
present in the Pack and accessing the locations of their texts.
155
:param chk_index: A GraphIndex for accessing content by CHK, if the
143
158
self.revision_index = revision_index
144
159
self.inventory_index = inventory_index
145
160
self.text_index = text_index
146
161
self.signature_index = signature_index
162
self.chk_index = chk_index
148
164
def access_tuple(self):
149
165
"""Return a tuple (transport, name) for the pack content."""
150
166
return self.pack_transport, self.file_name()
168
def _check_references(self):
169
"""Make sure our external references are present.
171
Packs are allowed to have deltas whose base is not in the pack, but it
172
must be present somewhere in this collection. It is not allowed to
173
have deltas based on a fallback repository.
174
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
177
for (index_name, external_refs, index) in [
179
self._get_external_refs(self.text_index),
180
self._pack_collection.text_index.combined_index),
182
self._get_external_refs(self.inventory_index),
183
self._pack_collection.inventory_index.combined_index),
185
missing = external_refs.difference(
186
k for (idx, k, v, r) in
187
index.iter_entries(external_refs))
189
missing_items[index_name] = sorted(list(missing))
191
from pprint import pformat
192
raise errors.BzrCheckError(
193
"Newly created pack file %r has delta references to "
194
"items not in its repository:\n%s"
195
% (self, pformat(missing_items)))
152
197
def file_name(self):
153
198
"""Get the file name for the pack on disk."""
154
199
return self.name + '.pack'
172
225
"""The text index is the name + .tix."""
173
226
return self.index_name('text', name)
175
def _external_compression_parents_of_texts(self):
178
for node in self.text_index.iter_all_entries():
180
refs.update(node[3][1])
228
def _replace_index_with_readonly(self, index_type):
229
unlimited_cache = False
230
if index_type == 'chk':
231
unlimited_cache = True
232
setattr(self, index_type + '_index',
233
self.index_class(self.index_transport,
234
self.index_name(index_type, self.name),
235
self.index_sizes[self.index_offset(index_type)],
236
unlimited_cache=unlimited_cache))
184
239
class ExistingPack(Pack):
185
240
"""An in memory proxy for an existing .pack and its disk indices."""
187
242
def __init__(self, pack_transport, name, revision_index, inventory_index,
188
text_index, signature_index):
243
text_index, signature_index, chk_index=None):
189
244
"""Create an ExistingPack object.
191
246
:param pack_transport: The transport where the pack file resides.
192
247
:param name: The name of the pack on disk in the pack_transport.
194
249
Pack.__init__(self, revision_index, inventory_index, text_index,
250
signature_index, chk_index)
197
252
self.pack_transport = pack_transport
198
253
if None in (revision_index, inventory_index, text_index,
206
261
return not self.__eq__(other)
208
263
def __repr__(self):
209
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
210
id(self), self.pack_transport, self.name)
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
return index.external_references(1)
213
338
class NewPack(Pack):
214
339
"""An in memory proxy for a pack which is being created."""
216
# A map of index 'type' to the file extension and position in the
218
index_definitions = {
219
'revision': ('.rix', 0),
220
'inventory': ('.iix', 1),
222
'signature': ('.six', 3),
225
def __init__(self, upload_transport, index_transport, pack_transport,
226
upload_suffix='', file_mode=None, index_builder_class=None,
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
228
342
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
344
:param pack_collection: A PackCollection into which this is being inserted.
237
345
:param upload_suffix: An optional suffix to be given to any temporary
238
346
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
240
:param index_builder_class: Required keyword parameter - the class of
241
index builder to use.
242
:param index_class: Required keyword parameter - the class of index
347
:param file_mode: Unix permissions for newly created file.
245
349
# The relative locations of the packs are constrained, but all are
246
350
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
247
356
Pack.__init__(self,
248
357
# Revisions: parents list, no text compression.
249
358
index_builder_class(reference_lists=1),
258
367
# Signatures: Just blobs to store, no compression, no parents
260
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
373
self._pack_collection = pack_collection
262
374
# When we make readonly indices, we need this.
263
self.index_class = index_class
375
self.index_class = pack_collection._index_class
264
376
# where should the new pack be opened
265
self.upload_transport = upload_transport
377
self.upload_transport = pack_collection._upload_transport
266
378
# where are indices written out to
267
self.index_transport = index_transport
379
self.index_transport = pack_collection._index_transport
268
380
# where is the pack renamed to when it is finished?
269
self.pack_transport = pack_transport
381
self.pack_transport = pack_collection._pack_transport
270
382
# What file mode to upload the pack and indices with.
271
383
self._file_mode = file_mode
272
384
# tracks the content written to the .pack file.
273
385
self._hash = osutils.md5()
274
# a four-tuple with the length in bytes of the indices, once the pack
275
# is finalised. (rev, inv, text, sigs)
386
# a tuple with the length in bytes of the indices, once the pack
387
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
276
388
self.index_sizes = None
277
389
# How much data to cache when writing packs. Note that this is not
278
390
# synchronised with reads, because it's not in the transport layer, so
352
475
- stores the index size tuple for the pack in the index_sizes
357
self._write_data('', flush=True)
358
self.name = self._hash.hexdigest()
478
self.finish_content()
480
self._check_references()
360
482
# XXX: It'd be better to write them all to temporary names, then
361
483
# rename them all into place, so that the window when only some are
362
484
# visible is smaller. On the other hand none will be seen until
363
485
# they're in the names list.
364
486
self.index_sizes = [None, None, None, None]
365
self._write_index('revision', self.revision_index, 'revision')
366
self._write_index('inventory', self.inventory_index, 'inventory')
367
self._write_index('text', self.text_index, 'file texts')
487
self._write_index('revision', self.revision_index, 'revision', suspend)
488
self._write_index('inventory', self.inventory_index, 'inventory',
490
self._write_index('text', self.text_index, 'file texts', suspend)
368
491
self._write_index('signature', self.signature_index,
369
'revision signatures')
492
'revision signatures', suspend)
493
if self.chk_index is not None:
494
self.index_sizes.append(None)
495
self._write_index('chk', self.chk_index,
496
'content hash bytes', suspend)
370
497
self.write_stream.close()
371
498
# Note that this will clobber an existing pack with the same name,
372
499
# without checking for hash collisions. While this is undesirable this
397
525
self._hash.update(bytes)
398
526
self._buffer[:] = [[], 0]
400
def index_name(self, index_type, name):
401
"""Get the disk name of an index type for pack name 'name'."""
402
return name + NewPack.index_definitions[index_type][0]
404
def index_offset(self, index_type):
405
"""Get the position in a index_size array for a given index type."""
406
return NewPack.index_definitions[index_type][1]
408
def _replace_index_with_readonly(self, index_type):
409
setattr(self, index_type + '_index',
410
self.index_class(self.index_transport,
411
self.index_name(index_type, self.name),
412
self.index_sizes[self.index_offset(index_type)]))
528
def _get_external_refs(self, index):
529
return index._external_references()
414
531
def set_write_cache_size(self, size):
415
532
self._cache_limit = size
417
def _write_index(self, index_type, index, label):
534
def _write_index(self, index_type, index, label, suspend=False):
418
535
"""Write out an index.
420
537
:param index_type: The type of index to write - e.g. 'revision'.
422
539
:param label: What label to give the index e.g. 'revision'.
424
541
index_name = self.index_name(index_type, self.name)
425
self.index_sizes[self.index_offset(index_type)] = \
426
self.index_transport.put_file(index_name, index.finish(),
427
mode=self._file_mode)
543
transport = self.upload_transport
545
transport = self.index_transport
546
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
547
index_name, index.finish(), mode=self._file_mode)
428
548
if 'pack' in debug.debug_flags:
429
549
# XXX: size might be interesting?
430
550
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
431
551
time.ctime(), label, self.upload_transport.base,
432
552
self.random_name, time.time() - self.start_time)
433
# Replace the writable index on this object with a readonly,
553
# Replace the writable index on this object with a readonly,
434
554
# presently unloaded index. We should alter
435
555
# the index layer to make its finish() error if add_node is
436
556
# subsequently used. RBC
445
565
such as 'revision index'.
447
567
A CombinedIndex provides an index on a single key space built up
448
from several on-disk indices. The AggregateIndex builds on this
568
from several on-disk indices. The AggregateIndex builds on this
449
569
to provide a knit access layer, and allows having up to one writable
450
570
index within the collection.
452
572
# XXX: Probably 'can be written to' could/should be separated from 'acts
453
573
# like a knit index' -- mbp 20071024
456
"""Create an AggregateIndex."""
575
def __init__(self, reload_func=None, flush_func=None):
576
"""Create an AggregateIndex.
578
:param reload_func: A function to call if we find we are missing an
579
index. Should have the form reload_func() => True if the list of
580
active pack files has changed.
582
self._reload_func = reload_func
457
583
self.index_to_pack = {}
458
self.combined_index = CombinedGraphIndex([])
459
self.data_access = _DirectPackAccess(self.index_to_pack)
460
self.add_callback = None
462
def replace_indices(self, index_to_pack, indices):
463
"""Replace the current mappings with fresh ones.
465
This should probably not be used eventually, rather incremental add and
466
removal of indices. It has been added during refactoring of existing
469
:param index_to_pack: A mapping from index objects to
470
(transport, name) tuples for the pack file data.
471
:param indices: A list of indices.
473
# refresh the revision pack map dict without replacing the instance.
474
self.index_to_pack.clear()
475
self.index_to_pack.update(index_to_pack)
476
# XXX: API break - clearly a 'replace' method would be good?
477
self.combined_index._indices[:] = indices
478
# the current add nodes callback for the current writable index if
584
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
585
self.data_access = _DirectPackAccess(self.index_to_pack,
586
reload_func=reload_func,
587
flush_func=flush_func)
480
588
self.add_callback = None
482
590
def add_index(self, index, pack):
562
677
def _extra_init(self):
563
678
"""A template hook to allow extending the constructor trivially."""
680
def _pack_map_and_index_list(self, index_attribute):
681
"""Convert a list of packs to an index pack map and index list.
683
:param index_attribute: The attribute that the desired index is found
685
:return: A tuple (map, list) where map contains the dict from
686
index:pack_tuple, and list contains the indices in the preferred
691
for pack_obj in self.packs:
692
index = getattr(pack_obj, index_attribute)
693
indices.append(index)
694
pack_map[index] = pack_obj
695
return pack_map, indices
697
def _index_contents(self, indices, key_filter=None):
698
"""Get an iterable of the index contents from a pack_map.
700
:param indices: The list of indices to query
701
:param key_filter: An optional filter to limit the keys returned.
703
all_index = CombinedGraphIndex(indices)
704
if key_filter is None:
705
return all_index.iter_all_entries()
707
return all_index.iter_entries(key_filter)
565
709
def pack(self, pb=None):
566
710
"""Create a new pack by reading data from other packs.
568
712
This does little more than a bulk copy of data. One key difference
569
713
is that data with the same item key across multiple packs is elided
570
714
from the output. The new pack is written into the current pack store
571
along with its indices, and the name added to the pack names. The
715
along with its indices, and the name added to the pack names. The
572
716
source packs are not altered and are not required to be in the current
604
749
def open_pack(self):
605
750
"""Open a pack for the pack we are creating."""
606
return NewPack(self._pack_collection._upload_transport,
607
self._pack_collection._index_transport,
608
self._pack_collection._pack_transport, upload_suffix=self.suffix,
609
file_mode=self._pack_collection.repo.bzrdir._get_file_mode(),
610
index_builder_class=self._pack_collection._index_builder_class,
611
index_class=self._pack_collection._index_class)
751
new_pack = self._pack_collection.pack_factory(self._pack_collection,
752
upload_suffix=self.suffix,
753
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
754
# We know that we will process all nodes in order, and don't need to
755
# query, so don't combine any indices spilled to disk until we are done
756
new_pack.revision_index.set_optimize(combine_backing_indices=False)
757
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
758
new_pack.text_index.set_optimize(combine_backing_indices=False)
759
new_pack.signature_index.set_optimize(combine_backing_indices=False)
762
def _update_pack_order(self, entries, index_to_pack_map):
763
"""Determine how we want our packs to be ordered.
765
This changes the sort order of the self.packs list so that packs unused
766
by 'entries' will be at the end of the list, so that future requests
767
can avoid probing them. Used packs will be at the front of the
768
self.packs list, in the order of their first use in 'entries'.
770
:param entries: A list of (index, ...) tuples
771
:param index_to_pack_map: A mapping from index objects to pack objects.
775
for entry in entries:
777
if index not in seen_indexes:
778
packs.append(index_to_pack_map[index])
779
seen_indexes.add(index)
780
if len(packs) == len(self.packs):
781
if 'pack' in debug.debug_flags:
782
mutter('Not changing pack list, all packs used.')
784
seen_packs = set(packs)
785
for pack in self.packs:
786
if pack not in seen_packs:
789
if 'pack' in debug.debug_flags:
790
old_names = [p.access_tuple()[1] for p in self.packs]
791
new_names = [p.access_tuple()[1] for p in packs]
792
mutter('Reordering packs\nfrom: %s\n to: %s',
793
old_names, new_names)
613
796
def _copy_revision_texts(self):
614
797
"""Copy revision data to the new pack."""
756
940
self._pack_collection.allocate(new_pack)
759
def _copy_nodes(self, nodes, index_map, writer, write_index):
760
"""Copy knit nodes between packs with no graph references."""
943
def _copy_chks(self, refs=None):
944
# XXX: Todo, recursive follow-pointers facility when fetching some
946
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
chk_nodes = self._index_contents(chk_indices, refs)
950
# TODO: This isn't strictly tasteful as we are accessing some private
951
# variables (_serializer). Perhaps a better way would be to have
952
# Repository._deserialise_chk_node()
953
search_key_func = chk_map.search_key_registry.get(
954
self._pack_collection.repo._serializer.search_key_name)
955
def accumlate_refs(lines):
956
# XXX: move to a generic location
958
bytes = ''.join(lines)
959
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
960
new_refs.update(node.refs())
961
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
962
self.new_pack.chk_index, output_lines=accumlate_refs)
965
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
"""Copy knit nodes between packs with no graph references.
969
:param output_lines: Output full texts of copied items.
761
971
pb = ui.ui_factory.nested_progress_bar()
763
973
return self._do_copy_nodes(nodes, index_map, writer,
974
write_index, pb, output_lines=output_lines)
768
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
978
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
769
980
# for record verification
770
981
knit = KnitVersionedFiles(None, None)
771
982
# plan a readv on each source pack:
792
1003
# linear scan up the pack
793
1004
pack_readv_requests.sort()
795
transport, path = index_map[index]
796
reader = pack.make_readv_reader(transport, path,
797
[offset[0:2] for offset in pack_readv_requests])
1006
pack_obj = index_map[index]
1007
transport, path = pack_obj.access_tuple()
1009
reader = pack.make_readv_reader(transport, path,
1010
[offset[0:2] for offset in pack_readv_requests])
1011
except errors.NoSuchFile:
1012
if self._reload_func is not None:
798
1015
for (names, read_func), (_1, _2, (key, eol_flag)) in \
799
1016
izip(reader.iter_records(), pack_readv_requests):
800
1017
raw_data = read_func(None)
801
1018
# check the header only
802
df, _ = knit._parse_record_header(key, raw_data)
1019
if output_lines is not None:
1020
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
df, _ = knit._parse_record_header(key, raw_data)
804
1024
pos, size = writer.add_bytes_record(raw_data, names)
805
1025
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
806
1026
pb.update("Copied record", record_index)
860
1086
record_index += 1
862
1088
def _get_text_nodes(self):
863
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
864
self.packs, 'text_index')[0]
865
return text_index_map, self._pack_collection._index_contents(text_index_map,
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
866
1092
self._text_filter)
868
1094
def _least_readv_node_readv(self, nodes):
869
1095
"""Generate request groups for nodes using the least readv's.
871
1097
:param nodes: An iterable of graph index nodes.
872
1098
:return: Total node count and an iterator of the data needed to perform
873
1099
readvs to obtain the data for nodes. Each item yielded by the
874
1100
iterator is a tuple with:
875
1101
index, readv_vector, node_vector. readv_vector is a list ready to
876
1102
hand to the transport readv method, and node_vector is a list of
877
(key, eol_flag, references) for the the node retrieved by the
1103
(key, eol_flag, references) for the node retrieved by the
878
1104
matching readv_vector.
880
1106
# group by pack so we do one readv per pack
1140
1382
self._pack_transport = pack_transport
1141
1383
self._index_builder_class = index_builder_class
1142
1384
self._index_class = index_class
1143
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1385
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1144
1387
self.packs = []
1145
1388
# name:Pack mapping
1146
1390
self._packs_by_name = {}
1147
1391
# the previous pack-names content
1148
1392
self._packs_at_load = None
1149
1393
# when a pack is being created by this object, the state of that pack.
1150
1394
self._new_pack = None
1151
1395
# aggregated revision index data
1152
self.revision_index = AggregateIndex()
1153
self.inventory_index = AggregateIndex()
1154
self.text_index = AggregateIndex()
1155
self.signature_index = AggregateIndex()
1396
flush = self._flush_new_pack
1397
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1398
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1401
all_indices = [self.revision_index, self.inventory_index,
1402
self.text_index, self.signature_index]
1404
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1405
all_indices.append(self.chk_index)
1407
# used to determine if we're using a chk_index elsewhere.
1408
self.chk_index = None
1409
# Tell all the CombinedGraphIndex objects about each other, so they can
1410
# share hints about which pack names to search first.
1411
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1412
for combined_idx in all_combined:
1413
combined_idx.set_sibling_indices(
1414
set(all_combined).difference([combined_idx]))
1416
self._resumed_packs = []
1419
return '%s(%r)' % (self.__class__.__name__, self.repo)
1157
1421
def add_pack_to_memory(self, pack):
1158
1422
"""Make a Pack object available to the repository to satisfy queries.
1160
1424
:param pack: A Pack object.
1162
1426
if pack.name in self._packs_by_name:
1163
raise AssertionError()
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
1164
1429
self.packs.append(pack)
1165
1430
self._packs_by_name[pack.name] = pack
1166
1431
self.revision_index.add_index(pack.revision_index, pack)
1167
1432
self.inventory_index.add_index(pack.inventory_index, pack)
1168
1433
self.text_index.add_index(pack.text_index, pack)
1169
1434
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
1171
1438
def all_packs(self):
1172
1439
"""Return a list of all the Pack objects this repository has.
1194
1461
in synchronisation with certain steps. Otherwise the names collection
1195
1462
is not flushed.
1197
:return: True if packing took place.
1464
:return: Something evaluating true if packing took place.
1468
return self._do_autopack()
1469
except errors.RetryAutopack:
1470
# If we get a RetryAutopack exception, we should abort the
1471
# current action, and retry.
1474
def _do_autopack(self):
1199
1475
# XXX: Should not be needed when the management of indices is sane.
1200
1476
total_revisions = self.revision_index.combined_index.key_count()
1201
1477
total_packs = len(self._names)
1202
1478
if self._max_pack_count(total_revisions) >= total_packs:
1204
# XXX: the following may want to be a class, to pack with a given
1206
mutter('Auto-packing repository %s, which has %d pack files, '
1207
'containing %d revisions into %d packs.', self, total_packs,
1208
total_revisions, self._max_pack_count(total_revisions))
1209
1480
# determine which packs need changing
1210
1481
pack_distribution = self.pack_distribution(total_revisions)
1211
1482
existing_packs = []
1219
1490
# group their data with the relevant commit, and that may
1220
1491
# involve rewriting ancient history - which autopack tries to
1221
1492
# avoid. Alternatively we could not group the data but treat
1222
# each of these as having a single revision, and thus add
1493
# each of these as having a single revision, and thus add
1223
1494
# one revision for each to the total revision count, to get
1224
1495
# a matching distribution.
1226
1497
existing_packs.append((revision_count, pack))
1227
1498
pack_operations = self.plan_autopack_combinations(
1228
1499
existing_packs, pack_distribution)
1229
self._execute_pack_operations(pack_operations)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
1232
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1233
1514
"""Execute a series of pack operations.
1235
1516
:param pack_operations: A list of [revision_count, packs_to_combine].
1236
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1239
1520
for revision_count, packs in pack_operations:
1240
1521
# we may have no-ops from the setup logic
1241
1522
if len(packs) == 0:
1243
_packer_class(self, packs, '.autopack').pack()
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
1244
1536
for pack in packs:
1245
1537
self._remove_pack_from_memory(pack)
1246
1538
# record the newly available packs and stop advertising the old
1248
self._save_pack_names(clear_obsolete_packs=True)
1249
# Move the old packs out of the way now they are no longer referenced.
1250
for revision_count, packs in pack_operations:
1251
self._obsolete_packs(packs)
1540
to_be_obsoleted = []
1541
for _, packs in pack_operations:
1542
to_be_obsoleted.extend(packs)
1543
result = self._save_pack_names(clear_obsolete_packs=True,
1544
obsolete_packs=to_be_obsoleted)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
1253
1551
def lock_names(self):
1254
1552
"""Acquire the mutex around the pack-names index.
1256
1554
This cannot be used in the middle of a read-only transaction on the
1259
1557
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None, clean_obsolete_packs=False):
1262
1564
"""Pack the pack collection totally."""
1263
1565
self.ensure_loaded()
1264
1566
total_packs = len(self._names)
1266
# This is arguably wrong because we might not be optimal, but for
1267
# now lets leave it in. (e.g. reconcile -> one pack. But not
1567
if self._already_packed():
1270
1569
total_revisions = self.revision_index.combined_index.key_count()
1271
1570
# XXX: the following may want to be a class, to pack with a given
1273
1572
mutter('Packing repository %s, which has %d pack files, '
1274
'containing %d revisions into 1 packs.', self, total_packs,
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
1276
1575
# determine which packs need changing
1277
pack_distribution = [1]
1278
1576
pack_operations = [[0, []]]
1279
1577
for pack in self.all_packs():
1280
pack_operations[-1][0] += pack.get_revision_count()
1281
pack_operations[-1][1].append(pack)
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1282
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
if clean_obsolete_packs:
1586
self._clear_obsolete_packs()
1284
1588
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1285
1589
"""Plan a pack operation.
1365
1677
inv_index = self._make_index(name, '.iix')
1366
1678
txt_index = self._make_index(name, '.tix')
1367
1679
sig_index = self._make_index(name, '.six')
1680
if self.chk_index is not None:
1681
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1368
1684
result = ExistingPack(self._pack_transport, name, rev_index,
1369
inv_index, txt_index, sig_index)
1685
inv_index, txt_index, sig_index, chk_index)
1370
1686
self.add_pack_to_memory(result)
1689
def _resume_pack(self, name):
1690
"""Get a suspended Pack object by name.
1692
:param name: The name of the pack - e.g. '123456'
1693
:return: A Pack object.
1695
if not re.match('[a-f0-9]{32}', name):
1696
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1698
raise errors.UnresumableWriteGroup(
1699
self.repo, [name], 'Malformed write group token')
1701
rev_index = self._make_index(name, '.rix', resume=True)
1702
inv_index = self._make_index(name, '.iix', resume=True)
1703
txt_index = self._make_index(name, '.tix', resume=True)
1704
sig_index = self._make_index(name, '.six', resume=True)
1705
if self.chk_index is not None:
1706
chk_index = self._make_index(name, '.cix', resume=True,
1707
unlimited_cache=True)
1710
result = self.resumed_pack_factory(name, rev_index, inv_index,
1711
txt_index, sig_index, self._upload_transport,
1712
self._pack_transport, self._index_transport, self,
1713
chk_index=chk_index)
1714
except errors.NoSuchFile, e:
1715
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1716
self.add_pack_to_memory(result)
1717
self._resumed_packs.append(result)
1373
1720
def allocate(self, a_new_pack):
1374
1721
"""Allocate name in the list of packs.
1393
1740
return self._index_class(self.transport, 'pack-names', None
1394
1741
).iter_all_entries()
1396
def _make_index(self, name, suffix):
1743
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1397
1744
size_offset = self._suffix_offsets[suffix]
1398
1745
index_name = name + suffix
1399
index_size = self._names[name][size_offset]
1400
return self._index_class(
1401
self._index_transport, index_name, index_size)
1747
transport = self._upload_transport
1748
index_size = transport.stat(index_name).st_size
1750
transport = self._index_transport
1751
index_size = self._names[name][size_offset]
1752
return self._index_class(transport, index_name, index_size,
1753
unlimited_cache=unlimited_cache)
1403
1755
def _max_pack_count(self, total_revisions):
1404
1756
"""Return the maximum number of packs to use for total revisions.
1406
1758
:param total_revisions: The total number of revisions in the
1432
1784
:param return: None.
1434
1786
for pack in packs:
1435
pack.pack_transport.rename(pack.file_name(),
1436
'../obsolete_packs/' + pack.file_name())
1788
pack.pack_transport.rename(pack.file_name(),
1789
'../obsolete_packs/' + pack.file_name())
1790
except (errors.PathError, errors.TransportError), e:
1791
# TODO: Should these be warnings or mutters?
1792
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1437
1794
# TODO: Probably needs to know all possible indices for this pack
1438
1795
# - or maybe list the directory and move all indices matching this
1439
1796
# name whether we recognize it or not?
1440
for suffix in ('.iix', '.six', '.tix', '.rix'):
1441
self._index_transport.rename(pack.name + suffix,
1442
'../obsolete_packs/' + pack.name + suffix)
1797
suffixes = ['.iix', '.six', '.tix', '.rix']
1798
if self.chk_index is not None:
1799
suffixes.append('.cix')
1800
for suffix in suffixes:
1802
self._index_transport.rename(pack.name + suffix,
1803
'../obsolete_packs/' + pack.name + suffix)
1804
except (errors.PathError, errors.TransportError), e:
1805
mutter("couldn't rename obsolete index, skipping it:\n%s"
1444
1808
def pack_distribution(self, total_revisions):
1445
1809
"""Generate a list of the number of revisions to put in each pack.
1464
1828
def _remove_pack_from_memory(self, pack):
1465
1829
"""Remove pack from the packs accessed by this repository.
1467
1831
Only affects memory state, until self._save_pack_names() is invoked.
1469
1833
self._names.pop(pack.name)
1470
1834
self._packs_by_name.pop(pack.name)
1471
1835
self._remove_pack_indices(pack)
1836
self.packs.remove(pack)
1473
def _remove_pack_indices(self, pack):
1474
"""Remove the indices for pack from the aggregated indices."""
1475
self.revision_index.remove_index(pack.revision_index, pack)
1476
self.inventory_index.remove_index(pack.inventory_index, pack)
1477
self.text_index.remove_index(pack.text_index, pack)
1478
self.signature_index.remove_index(pack.signature_index, pack)
1838
def _remove_pack_indices(self, pack, ignore_missing=False):
1839
"""Remove the indices for pack from the aggregated indices.
1841
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1843
for index_type in Pack.index_definitions.keys():
1844
attr_name = index_type + '_index'
1845
aggregate_index = getattr(self, attr_name)
1846
if aggregate_index is not None:
1847
pack_index = getattr(pack, attr_name)
1849
aggregate_index.remove_index(pack_index)
1480
1855
def reset(self):
1481
1856
"""Clear all cached data."""
1482
1857
# cached revision data
1483
self.repo._revision_knit = None
1484
1858
self.revision_index.clear()
1485
1859
# cached signature data
1486
self.repo._signature_knit = None
1487
1860
self.signature_index.clear()
1488
1861
# cached file text data
1489
1862
self.text_index.clear()
1490
self.repo._text_knit = None
1491
1863
# cached inventory data
1492
1864
self.inventory_index.clear()
1866
if self.chk_index is not None:
1867
self.chk_index.clear()
1493
1868
# remove the open pack
1494
1869
self._new_pack = None
1495
1870
# information about packs.
1498
1873
self._packs_by_name = {}
1499
1874
self._packs_at_load = None
1501
def _make_index_map(self, index_suffix):
1502
"""Return information on existing indices.
1504
:param suffix: Index suffix added to pack name.
1506
:returns: (pack_map, indices) where indices is a list of GraphIndex
1507
objects, and pack_map is a mapping from those objects to the
1508
pack tuple they describe.
1510
# TODO: stop using this; it creates new indices unnecessarily.
1511
self.ensure_loaded()
1512
suffix_map = {'.rix': 'revision_index',
1513
'.six': 'signature_index',
1514
'.iix': 'inventory_index',
1515
'.tix': 'text_index',
1517
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1518
suffix_map[index_suffix])
1520
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1521
"""Convert a list of packs to an index pack map and index list.
1523
:param packs: The packs list to process.
1524
:param index_attribute: The attribute that the desired index is found
1526
:return: A tuple (map, list) where map contains the dict from
1527
index:pack_tuple, and lsit contains the indices in the same order
1533
index = getattr(pack, index_attribute)
1534
indices.append(index)
1535
pack_map[index] = (pack.pack_transport, pack.file_name())
1536
return pack_map, indices
1538
def _index_contents(self, pack_map, key_filter=None):
1539
"""Get an iterable of the index contents from a pack_map.
1541
:param pack_map: A map from indices to pack details.
1542
:param key_filter: An optional filter to limit the
1545
indices = [index for index in pack_map.iterkeys()]
1546
all_index = CombinedGraphIndex(indices)
1547
if key_filter is None:
1548
return all_index.iter_all_entries()
1550
return all_index.iter_entries(key_filter)
1552
1876
def _unlock_names(self):
1553
1877
"""Release the mutex around the pack-names index."""
1554
1878
self.repo.control_files.unlock()
1556
def _save_pack_names(self, clear_obsolete_packs=False):
1557
"""Save the list of packs.
1559
This will take out the mutex around the pack names list for the
1560
duration of the method call. If concurrent updates have been made, a
1561
three-way merge between the current list and the current in memory list
1564
:param clear_obsolete_packs: If True, clear out the contents of the
1565
obsolete_packs directory.
1569
builder = self._index_builder_class()
1570
# load the disk nodes across
1572
for index, key, value in self._iter_disk_pack_index():
1573
disk_nodes.add((key, value))
1574
# do a two-way diff against our original content
1575
current_nodes = set()
1576
for name, sizes in self._names.iteritems():
1578
((name, ), ' '.join(str(size) for size in sizes)))
1579
deleted_nodes = self._packs_at_load - current_nodes
1580
new_nodes = current_nodes - self._packs_at_load
1581
disk_nodes.difference_update(deleted_nodes)
1582
disk_nodes.update(new_nodes)
1583
# TODO: handle same-name, index-size-changes here -
1584
# e.g. use the value from disk, not ours, *unless* we're the one
1586
for key, value in disk_nodes:
1587
builder.add_node(key, value)
1588
self.transport.put_file('pack-names', builder.finish(),
1589
mode=self.repo.bzrdir._get_file_mode())
1590
# move the baseline forward
1591
self._packs_at_load = disk_nodes
1592
if clear_obsolete_packs:
1593
self._clear_obsolete_packs()
1595
self._unlock_names()
1596
# synchronise the memory packs list with what we just wrote:
1880
def _diff_pack_names(self):
1881
"""Read the pack names from disk, and compare it to the one in memory.
1883
:return: (disk_nodes, deleted_nodes, new_nodes)
1884
disk_nodes The final set of nodes that should be referenced
1885
deleted_nodes Nodes which have been removed from when we started
1886
new_nodes Nodes that are newly introduced
1888
# load the disk nodes across
1890
for index, key, value in self._iter_disk_pack_index():
1891
disk_nodes.add((key, value))
1892
orig_disk_nodes = set(disk_nodes)
1894
# do a two-way diff against our original content
1895
current_nodes = set()
1896
for name, sizes in self._names.iteritems():
1898
((name, ), ' '.join(str(size) for size in sizes)))
1900
# Packs no longer present in the repository, which were present when we
1901
# locked the repository
1902
deleted_nodes = self._packs_at_load - current_nodes
1903
# Packs which this process is adding
1904
new_nodes = current_nodes - self._packs_at_load
1906
# Update the disk_nodes set to include the ones we are adding, and
1907
# remove the ones which were removed by someone else
1908
disk_nodes.difference_update(deleted_nodes)
1909
disk_nodes.update(new_nodes)
1911
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1913
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1914
"""Given the correct set of pack files, update our saved info.
1916
:return: (removed, added, modified)
1917
removed pack names removed from self._names
1918
added pack names added to self._names
1919
modified pack names that had changed value
1924
## self._packs_at_load = disk_nodes
1597
1925
new_names = dict(disk_nodes)
1598
1926
# drop no longer present nodes
1599
1927
for pack in self.all_packs():
1600
1928
if (pack.name,) not in new_names:
1929
removed.append(pack.name)
1601
1930
self._remove_pack_from_memory(pack)
1602
1931
# add new nodes/refresh existing ones
1603
1932
for key, value in disk_nodes:
1617
1946
self._remove_pack_from_memory(self.get_pack_by_name(name))
1618
1947
self._names[name] = sizes
1619
1948
self.get_pack_by_name(name)
1949
modified.append(name)
1622
1952
self._names[name] = sizes
1623
1953
self.get_pack_by_name(name)
1625
def _clear_obsolete_packs(self):
1955
return removed, added, modified
1957
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1958
"""Save the list of packs.
1960
This will take out the mutex around the pack names list for the
1961
duration of the method call. If concurrent updates have been made, a
1962
three-way merge between the current list and the current in memory list
1965
:param clear_obsolete_packs: If True, clear out the contents of the
1966
obsolete_packs directory.
1967
:param obsolete_packs: Packs that are obsolete once the new pack-names
1968
file has been written.
1969
:return: A list of the names saved that were not previously on disk.
1971
already_obsolete = []
1974
builder = self._index_builder_class()
1975
(disk_nodes, deleted_nodes, new_nodes,
1976
orig_disk_nodes) = self._diff_pack_names()
1977
# TODO: handle same-name, index-size-changes here -
1978
# e.g. use the value from disk, not ours, *unless* we're the one
1980
for key, value in disk_nodes:
1981
builder.add_node(key, value)
1982
self.transport.put_file('pack-names', builder.finish(),
1983
mode=self.repo.bzrdir._get_file_mode())
1984
self._packs_at_load = disk_nodes
1985
if clear_obsolete_packs:
1988
to_preserve = set([o.name for o in obsolete_packs])
1989
already_obsolete = self._clear_obsolete_packs(to_preserve)
1991
self._unlock_names()
1992
# synchronise the memory packs list with what we just wrote:
1993
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1995
# TODO: We could add one more condition here. "if o.name not in
1996
# orig_disk_nodes and o != the new_pack we haven't written to
1997
# disk yet. However, the new pack object is not easily
1998
# accessible here (it would have to be passed through the
1999
# autopacking code, etc.)
2000
obsolete_packs = [o for o in obsolete_packs
2001
if o.name not in already_obsolete]
2002
self._obsolete_packs(obsolete_packs)
2003
return [new_node[0][0] for new_node in new_nodes]
2005
def reload_pack_names(self):
2006
"""Sync our pack listing with what is present in the repository.
2008
This should be called when we find out that something we thought was
2009
present is now missing. This happens when another process re-packs the
2012
:return: True if the in-memory list of packs has been altered at all.
2014
# The ensure_loaded call is to handle the case where the first call
2015
# made involving the collection was to reload_pack_names, where we
2016
# don't have a view of disk contents. Its a bit of a bandaid, and
2017
# causes two reads of pack-names, but its a rare corner case not struck
2018
# with regular push/pull etc.
2019
first_read = self.ensure_loaded()
2022
# out the new value.
2023
(disk_nodes, deleted_nodes, new_nodes,
2024
orig_disk_nodes) = self._diff_pack_names()
2025
# _packs_at_load is meant to be the explicit list of names in
2026
# 'pack-names' at then start. As such, it should not contain any
2027
# pending names that haven't been written out yet.
2028
self._packs_at_load = orig_disk_nodes
2030
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2031
if removed or added or modified:
2035
def _restart_autopack(self):
2036
"""Reload the pack names list, and restart the autopack code."""
2037
if not self.reload_pack_names():
2038
# Re-raise the original exception, because something went missing
2039
# and a restart didn't find it
2041
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2043
def _clear_obsolete_packs(self, preserve=None):
1626
2044
"""Delete everything from the obsolete-packs directory.
2046
:return: A list of pack identifiers (the filename without '.pack') that
2047
were found in obsolete_packs.
1628
2050
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2051
if preserve is None:
1629
2053
for filename in obsolete_pack_transport.list_dir('.'):
2054
name, ext = osutils.splitext(filename)
2057
if name in preserve:
1631
2060
obsolete_pack_transport.delete(filename)
1632
2061
except (errors.PathError, errors.TransportError), e:
1633
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2062
warning("couldn't delete obsolete pack, skipping it:\n%s"
1635
2066
def _start_write_group(self):
1636
2067
# Do not permit preparation for writing if we're not in a 'write lock'.
1637
2068
if not self.repo.is_write_locked():
1638
2069
raise errors.NotWriteLocked(self)
1639
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1640
self._pack_transport, upload_suffix='.pack',
1641
file_mode=self.repo.bzrdir._get_file_mode(),
1642
index_builder_class=self._index_builder_class,
1643
index_class=self._index_class)
2070
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2071
file_mode=self.repo.bzrdir._get_file_mode())
1644
2072
# allow writing: queue writes to a new index
1645
2073
self.revision_index.add_writable_index(self._new_pack.revision_index,
1646
2074
self._new_pack)
1660
2094
# FIXME: just drop the transient index.
1661
2095
# forget what names there are
1662
2096
if self._new_pack is not None:
1663
self._new_pack.abort()
1664
self._remove_pack_indices(self._new_pack)
1665
self._new_pack = None
1666
self.repo._text_knit = None
2097
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2098
operation.add_cleanup(setattr, self, '_new_pack', None)
2099
# If we aborted while in the middle of finishing the write
2100
# group, _remove_pack_indices could fail because the indexes are
2101
# already gone. But they're not there we shouldn't fail in this
2102
# case, so we pass ignore_missing=True.
2103
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2104
ignore_missing=True)
2105
operation.run_simple()
2106
for resumed_pack in self._resumed_packs:
2107
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2108
# See comment in previous finally block.
2109
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2110
ignore_missing=True)
2111
operation.run_simple()
2112
del self._resumed_packs[:]
2114
def _remove_resumed_pack_indices(self):
2115
for resumed_pack in self._resumed_packs:
2116
self._remove_pack_indices(resumed_pack)
2117
del self._resumed_packs[:]
2119
def _check_new_inventories(self):
2120
"""Detect missing inventories in this write group.
2122
:returns: list of strs, summarising any problems found. If the list is
2123
empty no problems were found.
2125
# The base implementation does no checks. GCRepositoryPackCollection
1668
2129
def _commit_write_group(self):
2131
for prefix, versioned_file in (
2132
('revisions', self.repo.revisions),
2133
('inventories', self.repo.inventories),
2134
('texts', self.repo.texts),
2135
('signatures', self.repo.signatures),
2137
missing = versioned_file.get_missing_compression_parent_keys()
2138
all_missing.update([(prefix,) + key for key in missing])
2140
raise errors.BzrCheckError(
2141
"Repository %s has missing compression parent(s) %r "
2142
% (self.repo, sorted(all_missing)))
2143
problems = self._check_new_inventories()
2145
problems_summary = '\n'.join(problems)
2146
raise errors.BzrCheckError(
2147
"Cannot add revision(s) to repository: " + problems_summary)
1669
2148
self._remove_pack_indices(self._new_pack)
2149
any_new_content = False
1670
2150
if self._new_pack.data_inserted():
1671
2151
# get all the data to disk and read to use
1672
2152
self._new_pack.finish()
1673
2153
self.allocate(self._new_pack)
1674
2154
self._new_pack = None
1675
if not self.autopack():
2155
any_new_content = True
2157
self._new_pack.abort()
2158
self._new_pack = None
2159
for resumed_pack in self._resumed_packs:
2160
# XXX: this is a pretty ugly way to turn the resumed pack into a
2161
# properly committed pack.
2162
self._names[resumed_pack.name] = None
2163
self._remove_pack_from_memory(resumed_pack)
2164
resumed_pack.finish()
2165
self.allocate(resumed_pack)
2166
any_new_content = True
2167
del self._resumed_packs[:]
2169
result = self.autopack()
1676
2171
# when autopack takes no steps, the names list is still
1678
self._save_pack_names()
2173
return self._save_pack_names()
2177
def _suspend_write_group(self):
2178
tokens = [pack.name for pack in self._resumed_packs]
2179
self._remove_pack_indices(self._new_pack)
2180
if self._new_pack.data_inserted():
2181
# get all the data to disk and read to use
2182
self._new_pack.finish(suspend=True)
2183
tokens.append(self._new_pack.name)
2184
self._new_pack = None
1680
2186
self._new_pack.abort()
1681
2187
self._new_pack = None
1682
self.repo._text_knit = None
2188
self._remove_resumed_pack_indices()
2191
def _resume_write_group(self, tokens):
2192
for token in tokens:
2193
self._resume_pack(token)
1685
2196
class KnitPackRepository(KnitRepository):
1686
2197
"""Repository with knit objects stored inside pack containers.
1688
2199
The layering for a KnitPackRepository is:
1690
2201
Graph | HPSS | Repository public layer |
1742
2256
deltas=True, parents=True, is_locked=self.is_locked),
1743
2257
data_access=self._pack_collection.text_index.data_access,
1744
2258
max_delta_chain=200)
2259
if _format.supports_chks:
2260
# No graph, no compression:- references from chks are between
2261
# different objects not temporal versions of the same; and without
2262
# some sort of temporal structure knit compression will just fail.
2263
self.chk_bytes = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2265
add_callback=self._pack_collection.chk_index.add_callback,
2266
deltas=False, parents=False, is_locked=self.is_locked),
2267
data_access=self._pack_collection.chk_index.data_access,
2270
self.chk_bytes = None
1745
2271
# True when the repository object is 'write locked' (as opposed to the
1746
# physical lock only taken out around changes to the pack-names list.)
2272
# physical lock only taken out around changes to the pack-names list.)
1747
2273
# Another way to represent this would be a decorator around the control
1748
2274
# files object that presents logical locks as physical ones - if this
1749
2275
# gets ugly consider that alternative design. RBC 20071011
1753
2279
self._reconcile_does_inventory_gc = True
1754
2280
self._reconcile_fixes_text_parents = True
1755
2281
self._reconcile_backsup_inventory = False
1756
self._fetch_order = 'unordered'
1758
def _warn_if_deprecated(self):
2283
def _warn_if_deprecated(self, branch=None):
1759
2284
# This class isn't deprecated, but one sub-format is
1760
2285
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
1761
from bzrlib import repository
1762
if repository._deprecation_warning_done:
1764
repository._deprecation_warning_done = True
1765
warning("Format %s for %s is deprecated - please use"
1766
" 'bzr upgrade --1.6.1-rich-root'"
1767
% (self._format, self.bzrdir.transport.base))
2286
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1769
2288
def _abort_write_group(self):
2289
self.revisions._index._key_dependencies.clear()
1770
2290
self._pack_collection._abort_write_group()
1772
def _find_inconsistent_revision_parents(self):
1773
"""Find revisions with incorrectly cached parents.
1775
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1776
parents-in-revision).
1778
if not self.is_locked():
1779
raise errors.ObjectNotLocked(self)
1780
pb = ui.ui_factory.nested_progress_bar()
1783
revision_nodes = self._pack_collection.revision_index \
1784
.combined_index.iter_all_entries()
1785
index_positions = []
1786
# Get the cached index values for all revisions, and also the location
1787
# in each index of the revision text so we can perform linear IO.
1788
for index, key, value, refs in revision_nodes:
1789
pos, length = value[1:].split(' ')
1790
index_positions.append((index, int(pos), key[0],
1791
tuple(parent[0] for parent in refs[0])))
1792
pb.update("Reading revision index.", 0, 0)
1793
index_positions.sort()
1794
batch_count = len(index_positions) / 1000 + 1
1795
pb.update("Checking cached revision graph.", 0, batch_count)
1796
for offset in xrange(batch_count):
1797
pb.update("Checking cached revision graph.", offset)
1798
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1801
rev_ids = [item[2] for item in to_query]
1802
revs = self.get_revisions(rev_ids)
1803
for revision, item in zip(revs, to_query):
1804
index_parents = item[3]
1805
rev_parents = tuple(revision.parent_ids)
1806
if index_parents != rev_parents:
1807
result.append((revision.revision_id, index_parents, rev_parents))
1812
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1813
def get_parents(self, revision_ids):
1814
"""See graph._StackedParentsProvider.get_parents."""
1815
parent_map = self.get_parent_map(revision_ids)
1816
return [parent_map.get(r, None) for r in revision_ids]
2292
def _get_source(self, to_format):
2293
if to_format.network_name() == self._format.network_name():
2294
return KnitPackStreamSource(self, to_format)
2295
return super(KnitPackRepository, self)._get_source(to_format)
1818
2297
def _make_parents_provider(self):
1819
2298
return graph.CachingParentsProvider(self)
1821
2300
def _refresh_data(self):
1822
if self._write_lock_count == 1 or (
1823
self.control_files._lock_count == 1 and
1824
self.control_files._lock_mode == 'r'):
1825
# forget what names there are
1826
self._pack_collection.reset()
1827
# XXX: Better to do an in-memory merge when acquiring a new lock -
1828
# factor out code from _save_pack_names.
1829
self._pack_collection.ensure_loaded()
2301
if not self.is_locked():
2303
self._pack_collection.reload_pack_names()
1831
2305
def _start_write_group(self):
1832
2306
self._pack_collection._start_write_group()
1834
2308
def _commit_write_group(self):
1835
return self._pack_collection._commit_write_group()
2309
hint = self._pack_collection._commit_write_group()
2310
self.revisions._index._key_dependencies.clear()
2313
def suspend_write_group(self):
2314
# XXX check self._write_group is self.get_transaction()?
2315
tokens = self._pack_collection._suspend_write_group()
2316
self.revisions._index._key_dependencies.clear()
2317
self._write_group = None
2320
def _resume_write_group(self, tokens):
2321
self._start_write_group()
2323
self._pack_collection._resume_write_group(tokens)
2324
except errors.UnresumableWriteGroup:
2325
self._abort_write_group()
2327
for pack in self._pack_collection._resumed_packs:
2328
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1837
2330
def get_transaction(self):
1838
2331
if self._write_lock_count:
1847
2340
return self._write_lock_count
1849
2342
def lock_write(self, token=None):
1850
if not self._write_lock_count and self.is_locked():
2343
locked = self.is_locked()
2344
if not self._write_lock_count and locked:
1851
2345
raise errors.ReadOnlyError(self)
1852
2346
self._write_lock_count += 1
1853
2347
if self._write_lock_count == 1:
1854
2348
self._transaction = transactions.WriteTransaction()
2350
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2351
note('%r was write locked again', self)
2352
self._prev_lock = 'w'
1855
2353
for repo in self._fallback_repositories:
1856
2354
# Writes don't affect fallback repos
1857
2355
repo.lock_read()
1858
self._refresh_data()
2356
self._refresh_data()
1860
2358
def lock_read(self):
2359
locked = self.is_locked()
1861
2360
if self._write_lock_count:
1862
2361
self._write_lock_count += 1
1864
2363
self.control_files.lock_read()
2365
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2366
note('%r was read locked again', self)
2367
self._prev_lock = 'r'
1865
2368
for repo in self._fallback_repositories:
1866
# Writes don't affect fallback repos
1867
2369
repo.lock_read()
1868
self._refresh_data()
2370
self._refresh_data()
1870
2372
def leave_lock_in_place(self):
1871
2373
# not supported - raise an error
1906
2413
transaction = self._transaction
1907
2414
self._transaction = None
1908
2415
transaction.finish()
1909
for repo in self._fallback_repositories:
1912
2417
self.control_files.unlock()
2419
if not self.is_locked():
1913
2420
for repo in self._fallback_repositories:
2424
class KnitPackStreamSource(StreamSource):
2425
"""A StreamSource used to transfer data between same-format KnitPack repos.
2427
This source assumes:
2428
1) Same serialization format for all objects
2429
2) Same root information
2430
3) XML format inventories
2431
4) Atomic inserts (so we can stream inventory texts before text
2436
def __init__(self, from_repository, to_format):
2437
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2438
self._text_keys = None
2439
self._text_fetch_order = 'unordered'
2441
def _get_filtered_inv_stream(self, revision_ids):
2442
from_repo = self.from_repository
2443
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2444
parent_keys = [(p,) for p in parent_ids]
2445
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2446
parent_text_keys = set(find_text_keys(
2447
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2448
content_text_keys = set()
2449
knit = KnitVersionedFiles(None, None)
2450
factory = KnitPlainFactory()
2451
def find_text_keys_from_content(record):
2452
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2453
raise ValueError("Unknown content storage kind for"
2454
" inventory text: %s" % (record.storage_kind,))
2455
# It's a knit record, it has a _raw_record field (even if it was
2456
# reconstituted from a network stream).
2457
raw_data = record._raw_record
2458
# read the entire thing
2459
revision_id = record.key[-1]
2460
content, _ = knit._parse_record(revision_id, raw_data)
2461
if record.storage_kind == 'knit-delta-gz':
2462
line_iterator = factory.get_linedelta_content(content)
2463
elif record.storage_kind == 'knit-ft-gz':
2464
line_iterator = factory.get_fulltext_content(content)
2465
content_text_keys.update(find_text_keys(
2466
[(line, revision_id) for line in line_iterator]))
2467
revision_keys = [(r,) for r in revision_ids]
2468
def _filtered_inv_stream():
2469
source_vf = from_repo.inventories
2470
stream = source_vf.get_record_stream(revision_keys,
2472
for record in stream:
2473
if record.storage_kind == 'absent':
2474
raise errors.NoSuchRevision(from_repo, record.key)
2475
find_text_keys_from_content(record)
2477
self._text_keys = content_text_keys - parent_text_keys
2478
return ('inventories', _filtered_inv_stream())
2480
def _get_text_stream(self):
2481
# Note: We know we don't have to handle adding root keys, because both
2482
# the source and target are the identical network name.
2483
text_stream = self.from_repository.texts.get_record_stream(
2484
self._text_keys, self._text_fetch_order, False)
2485
return ('texts', text_stream)
2487
def get_stream(self, search):
2488
revision_ids = search.get_keys()
2489
for stream_info in self._fetch_revision_texts(revision_ids):
2491
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2492
yield self._get_filtered_inv_stream(revision_ids)
2493
yield self._get_text_stream()
1917
2497
class RepositoryFormatPack(MetaDirRepositoryFormat):
1918
2498
"""Format logic for pack structured repositories.
2199
2774
_commit_builder_class = PackRootCommitBuilder
2200
2775
rich_root_data = True
2201
2776
supports_tree_reference = False # no subtrees
2202
_serializer = xml7.serializer_v7
2204
2778
supports_external_lookups = True
2205
2779
# What index classes to use
2206
2780
index_builder_class = InMemoryGraphIndex
2207
2781
index_class = GraphIndex
2784
def _serializer(self):
2785
return xml7.serializer_v7
2209
2787
def _get_matching_bzrdir(self):
2210
return bzrdir.format_registry.make_bzrdir(
2788
matching = bzrdir.format_registry.make_bzrdir(
2211
2789
'1.6.1-rich-root')
2790
matching.repository_format = self
2213
2793
def _ignore_setting_bzrdir(self, format):
2216
2796
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2218
def check_conversion_target(self, target_format):
2219
if not target_format.rich_root_data:
2220
raise errors.BadConversionTarget(
2221
'Does not support rich root data.', target_format)
2223
2798
def get_format_string(self):
2224
2799
"""See RepositoryFormat.get_format_string()."""
2225
2800
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2229
2804
" (deprecated)")
2232
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2233
"""A no-subtrees development repository.
2235
This format should be retained until the second release after bzr 1.7.
2237
This is pack-1.6.1 with B+Tree indices.
2807
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2808
"""A repository with stacking and btree indexes,
2809
without rich roots or subtrees.
2811
This is equivalent to pack-1.6 with B+Tree indices.
2240
2814
repository_class = KnitPackRepository
2241
2815
_commit_builder_class = PackCommitBuilder
2242
_serializer = xml5.serializer_v5
2243
2816
supports_external_lookups = True
2244
2817
# What index classes to use
2245
2818
index_builder_class = BTreeBuilder
2246
2819
index_class = BTreeGraphIndex
2822
def _serializer(self):
2823
return xml5.serializer_v5
2248
2825
def _get_matching_bzrdir(self):
2249
return bzrdir.format_registry.make_bzrdir('development2')
2826
return bzrdir.format_registry.make_bzrdir('1.9')
2251
2828
def _ignore_setting_bzrdir(self, format):
2256
2833
def get_format_string(self):
2257
2834
"""See RepositoryFormat.get_format_string()."""
2258
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2835
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2260
2837
def get_format_description(self):
2261
2838
"""See RepositoryFormat.get_format_description()."""
2262
return ("Development repository format, currently the same as "
2263
"1.6.1 with B+Trees.\n")
2265
def check_conversion_target(self, target_format):
2839
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2842
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2843
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2845
1.6-rich-root with B+Tree indices.
2848
repository_class = KnitPackRepository
2849
_commit_builder_class = PackRootCommitBuilder
2850
rich_root_data = True
2851
supports_tree_reference = False # no subtrees
2852
supports_external_lookups = True
2853
# What index classes to use
2854
index_builder_class = BTreeBuilder
2855
index_class = BTreeGraphIndex
2858
def _serializer(self):
2859
return xml6.serializer_v6
2861
def _get_matching_bzrdir(self):
2862
return bzrdir.format_registry.make_bzrdir(
2865
def _ignore_setting_bzrdir(self, format):
2868
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2870
def get_format_string(self):
2871
"""See RepositoryFormat.get_format_string()."""
2872
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2874
def get_format_description(self):
2875
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2269
2878
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2270
2879
"""A subtrees development repository.
2272
2881
This format should be retained until the second release after bzr 1.7.
2274
2883
1.6.1-subtree[as it might have been] with B+Tree indices.
2885
This is [now] retained until we have a CHK based subtree format in
2277
2889
repository_class = KnitPackRepository
2278
2890
_commit_builder_class = PackRootCommitBuilder
2279
2891
rich_root_data = True
2280
2893
supports_tree_reference = True
2281
_serializer = xml7.serializer_v7
2282
2894
supports_external_lookups = True
2283
2895
# What index classes to use
2284
2896
index_builder_class = BTreeBuilder
2285
2897
index_class = BTreeGraphIndex
2900
def _serializer(self):
2901
return xml7.serializer_v7
2287
2903
def _get_matching_bzrdir(self):
2288
2904
return bzrdir.format_registry.make_bzrdir(
2289
'development2-subtree')
2905
'development-subtree')
2291
2907
def _ignore_setting_bzrdir(self, format):
2294
2910
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2296
def check_conversion_target(self, target_format):
2297
if not target_format.rich_root_data:
2298
raise errors.BadConversionTarget(
2299
'Does not support rich root data.', target_format)
2300
if not getattr(target_format, 'supports_tree_reference', False):
2301
raise errors.BadConversionTarget(
2302
'Does not support nested trees', target_format)
2304
2912
def get_format_string(self):
2305
2913
"""See RepositoryFormat.get_format_string()."""
2306
2914
return ("Bazaar development format 2 with subtree support "
added
removed
bzrlib/repofmt/pack_repo.py
