1
# Copyright (C) 2007-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
37
from bzrlib.index import (
39
GraphIndexPrefixAdapter,
49
from bzrlib.decorators import (
54
from bzrlib.lock import LogicalLockResult
55
from bzrlib.repository import (
59
RepositoryWriteLockResult,
61
from bzrlib.vf_repository import (
62
MetaDirVersionedFileRepository,
63
MetaDirVersionedFileRepositoryFormat,
64
VersionedFileCommitBuilder,
65
VersionedFileRootCommitBuilder,
67
from bzrlib.trace import (
74
class PackCommitBuilder(VersionedFileCommitBuilder):
75
"""Subclass of VersionedFileCommitBuilder to add texts with pack semantics.
77
Specifically this uses one knit object rather than one knit object per
78
added text, reducing memory and object pressure.
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
83
revision_id=None, lossy=False):
84
VersionedFileCommitBuilder.__init__(self, repository, parents, config,
85
timestamp=timestamp, timezone=timezone, committer=committer,
86
revprops=revprops, revision_id=revision_id, lossy=lossy)
87
self._file_graph = graph.Graph(
88
repository._pack_collection.text_index.combined_index)
90
def _heads(self, file_id, revision_ids):
91
keys = [(file_id, revision_id) for revision_id in revision_ids]
92
return set([key[1] for key in self._file_graph.heads(keys)])
95
class PackRootCommitBuilder(VersionedFileRootCommitBuilder):
96
"""A subclass of RootCommitBuilder to add texts with pack semantics.
98
Specifically this uses one knit object rather than one knit object per
99
added text, reducing memory and object pressure.
102
def __init__(self, repository, parents, config, timestamp=None,
103
timezone=None, committer=None, revprops=None,
104
revision_id=None, lossy=False):
105
super(PackRootCommitBuilder, self).__init__(repository, parents,
106
config, timestamp=timestamp, timezone=timezone,
107
committer=committer, revprops=revprops, revision_id=revision_id,
109
self._file_graph = graph.Graph(
110
repository._pack_collection.text_index.combined_index)
112
def _heads(self, file_id, revision_ids):
113
keys = [(file_id, revision_id) for revision_id in revision_ids]
114
return set([key[1] for key in self._file_graph.heads(keys)])
118
"""An in memory proxy for a pack and its indices.
120
This is a base class that is not directly used, instead the classes
121
ExistingPack and NewPack are used.
124
# A map of index 'type' to the file extension and position in the
126
index_definitions = {
128
'revision': ('.rix', 0),
129
'inventory': ('.iix', 1),
131
'signature': ('.six', 3),
134
def __init__(self, revision_index, inventory_index, text_index,
135
signature_index, chk_index=None):
136
"""Create a pack instance.
138
:param revision_index: A GraphIndex for determining what revisions are
139
present in the Pack and accessing the locations of their texts.
140
:param inventory_index: A GraphIndex for determining what inventories are
141
present in the Pack and accessing the locations of their
143
:param text_index: A GraphIndex for determining what file texts
144
are present in the pack and accessing the locations of their
145
texts/deltas (via (fileid, revisionid) tuples).
146
:param signature_index: A GraphIndex for determining what signatures are
147
present in the Pack and accessing the locations of their texts.
148
:param chk_index: A GraphIndex for accessing content by CHK, if the
151
self.revision_index = revision_index
152
self.inventory_index = inventory_index
153
self.text_index = text_index
154
self.signature_index = signature_index
155
self.chk_index = chk_index
157
def access_tuple(self):
158
"""Return a tuple (transport, name) for the pack content."""
159
return self.pack_transport, self.file_name()
161
def _check_references(self):
162
"""Make sure our external references are present.
164
Packs are allowed to have deltas whose base is not in the pack, but it
165
must be present somewhere in this collection. It is not allowed to
166
have deltas based on a fallback repository.
167
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
170
for (index_name, external_refs, index) in [
172
self._get_external_refs(self.text_index),
173
self._pack_collection.text_index.combined_index),
175
self._get_external_refs(self.inventory_index),
176
self._pack_collection.inventory_index.combined_index),
178
missing = external_refs.difference(
179
k for (idx, k, v, r) in
180
index.iter_entries(external_refs))
182
missing_items[index_name] = sorted(list(missing))
184
from pprint import pformat
185
raise errors.BzrCheckError(
186
"Newly created pack file %r has delta references to "
187
"items not in its repository:\n%s"
188
% (self, pformat(missing_items)))
191
"""Get the file name for the pack on disk."""
192
return self.name + '.pack'
194
def get_revision_count(self):
195
return self.revision_index.key_count()
197
def index_name(self, index_type, name):
198
"""Get the disk name of an index type for pack name 'name'."""
199
return name + Pack.index_definitions[index_type][0]
201
def index_offset(self, index_type):
202
"""Get the position in a index_size array for a given index type."""
203
return Pack.index_definitions[index_type][1]
205
def inventory_index_name(self, name):
206
"""The inv index is the name + .iix."""
207
return self.index_name('inventory', name)
209
def revision_index_name(self, name):
210
"""The revision index is the name + .rix."""
211
return self.index_name('revision', name)
213
def signature_index_name(self, name):
214
"""The signature index is the name + .six."""
215
return self.index_name('signature', name)
217
def text_index_name(self, name):
218
"""The text index is the name + .tix."""
219
return self.index_name('text', name)
221
def _replace_index_with_readonly(self, index_type):
222
unlimited_cache = False
223
if index_type == 'chk':
224
unlimited_cache = True
225
index = self.index_class(self.index_transport,
226
self.index_name(index_type, self.name),
227
self.index_sizes[self.index_offset(index_type)],
228
unlimited_cache=unlimited_cache)
229
if index_type == 'chk':
230
index._leaf_factory = btree_index._gcchk_factory
231
setattr(self, index_type + '_index', index)
234
class ExistingPack(Pack):
235
"""An in memory proxy for an existing .pack and its disk indices."""
237
def __init__(self, pack_transport, name, revision_index, inventory_index,
238
text_index, signature_index, chk_index=None):
239
"""Create an ExistingPack object.
241
:param pack_transport: The transport where the pack file resides.
242
:param name: The name of the pack on disk in the pack_transport.
244
Pack.__init__(self, revision_index, inventory_index, text_index,
245
signature_index, chk_index)
247
self.pack_transport = pack_transport
248
if None in (revision_index, inventory_index, text_index,
249
signature_index, name, pack_transport):
250
raise AssertionError()
252
def __eq__(self, other):
253
return self.__dict__ == other.__dict__
255
def __ne__(self, other):
256
return not self.__eq__(other)
259
return "<%s.%s object at 0x%x, %s, %s" % (
260
self.__class__.__module__, self.__class__.__name__, id(self),
261
self.pack_transport, self.name)
264
class ResumedPack(ExistingPack):
266
def __init__(self, name, revision_index, inventory_index, text_index,
267
signature_index, upload_transport, pack_transport, index_transport,
268
pack_collection, chk_index=None):
269
"""Create a ResumedPack object."""
270
ExistingPack.__init__(self, pack_transport, name, revision_index,
271
inventory_index, text_index, signature_index,
273
self.upload_transport = upload_transport
274
self.index_transport = index_transport
275
self.index_sizes = [None, None, None, None]
277
('revision', revision_index),
278
('inventory', inventory_index),
279
('text', text_index),
280
('signature', signature_index),
282
if chk_index is not None:
283
indices.append(('chk', chk_index))
284
self.index_sizes.append(None)
285
for index_type, index in indices:
286
offset = self.index_offset(index_type)
287
self.index_sizes[offset] = index._size
288
self.index_class = pack_collection._index_class
289
self._pack_collection = pack_collection
290
self._state = 'resumed'
291
# XXX: perhaps check that the .pack file exists?
293
def access_tuple(self):
294
if self._state == 'finished':
295
return Pack.access_tuple(self)
296
elif self._state == 'resumed':
297
return self.upload_transport, self.file_name()
299
raise AssertionError(self._state)
302
self.upload_transport.delete(self.file_name())
303
indices = [self.revision_index, self.inventory_index, self.text_index,
304
self.signature_index]
305
if self.chk_index is not None:
306
indices.append(self.chk_index)
307
for index in indices:
308
index._transport.delete(index._name)
311
self._check_references()
312
index_types = ['revision', 'inventory', 'text', 'signature']
313
if self.chk_index is not None:
314
index_types.append('chk')
315
for index_type in index_types:
316
old_name = self.index_name(index_type, self.name)
317
new_name = '../indices/' + old_name
318
self.upload_transport.move(old_name, new_name)
319
self._replace_index_with_readonly(index_type)
320
new_name = '../packs/' + self.file_name()
321
self.upload_transport.move(self.file_name(), new_name)
322
self._state = 'finished'
324
def _get_external_refs(self, index):
325
"""Return compression parents for this index that are not present.
327
This returns any compression parents that are referenced by this index,
328
which are not contained *in* this index. They may be present elsewhere.
330
return index.external_references(1)
334
"""An in memory proxy for a pack which is being created."""
336
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
337
"""Create a NewPack instance.
339
:param pack_collection: A PackCollection into which this is being inserted.
340
:param upload_suffix: An optional suffix to be given to any temporary
341
files created during the pack creation. e.g '.autopack'
342
:param file_mode: Unix permissions for newly created file.
344
# The relative locations of the packs are constrained, but all are
345
# passed in because the caller has them, so as to avoid object churn.
346
index_builder_class = pack_collection._index_builder_class
347
if pack_collection.chk_index is not None:
348
chk_index = index_builder_class(reference_lists=0)
352
# Revisions: parents list, no text compression.
353
index_builder_class(reference_lists=1),
354
# Inventory: We want to map compression only, but currently the
355
# knit code hasn't been updated enough to understand that, so we
356
# have a regular 2-list index giving parents and compression
358
index_builder_class(reference_lists=2),
359
# Texts: compression and per file graph, for all fileids - so two
360
# reference lists and two elements in the key tuple.
361
index_builder_class(reference_lists=2, key_elements=2),
362
# Signatures: Just blobs to store, no compression, no parents
364
index_builder_class(reference_lists=0),
365
# CHK based storage - just blobs, no compression or parents.
368
self._pack_collection = pack_collection
369
# When we make readonly indices, we need this.
370
self.index_class = pack_collection._index_class
371
# where should the new pack be opened
372
self.upload_transport = pack_collection._upload_transport
373
# where are indices written out to
374
self.index_transport = pack_collection._index_transport
375
# where is the pack renamed to when it is finished?
376
self.pack_transport = pack_collection._pack_transport
377
# What file mode to upload the pack and indices with.
378
self._file_mode = file_mode
379
# tracks the content written to the .pack file.
380
self._hash = osutils.md5()
381
# a tuple with the length in bytes of the indices, once the pack
382
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
383
self.index_sizes = None
384
# How much data to cache when writing packs. Note that this is not
385
# synchronised with reads, because it's not in the transport layer, so
386
# is not safe unless the client knows it won't be reading from the pack
388
self._cache_limit = 0
389
# the temporary pack file name.
390
self.random_name = osutils.rand_chars(20) + upload_suffix
391
# when was this pack started ?
392
self.start_time = time.time()
393
# open an output stream for the data added to the pack.
394
self.write_stream = self.upload_transport.open_write_stream(
395
self.random_name, mode=self._file_mode)
396
if 'pack' in debug.debug_flags:
397
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
398
time.ctime(), self.upload_transport.base, self.random_name,
399
time.time() - self.start_time)
400
# A list of byte sequences to be written to the new pack, and the
401
# aggregate size of them. Stored as a list rather than separate
402
# variables so that the _write_data closure below can update them.
403
self._buffer = [[], 0]
404
# create a callable for adding data
406
# robertc says- this is a closure rather than a method on the object
407
# so that the variables are locals, and faster than accessing object
409
def _write_data(bytes, flush=False, _buffer=self._buffer,
410
_write=self.write_stream.write, _update=self._hash.update):
411
_buffer[0].append(bytes)
412
_buffer[1] += len(bytes)
414
if _buffer[1] > self._cache_limit or flush:
415
bytes = ''.join(_buffer[0])
419
# expose this on self, for the occasion when clients want to add data.
420
self._write_data = _write_data
421
# a pack writer object to serialise pack records.
422
self._writer = pack.ContainerWriter(self._write_data)
424
# what state is the pack in? (open, finished, aborted)
426
# no name until we finish writing the content
430
"""Cancel creating this pack."""
431
self._state = 'aborted'
432
self.write_stream.close()
433
# Remove the temporary pack file.
434
self.upload_transport.delete(self.random_name)
435
# The indices have no state on disk.
437
def access_tuple(self):
438
"""Return a tuple (transport, name) for the pack content."""
439
if self._state == 'finished':
440
return Pack.access_tuple(self)
441
elif self._state == 'open':
442
return self.upload_transport, self.random_name
444
raise AssertionError(self._state)
446
def data_inserted(self):
447
"""True if data has been added to this pack."""
448
return bool(self.get_revision_count() or
449
self.inventory_index.key_count() or
450
self.text_index.key_count() or
451
self.signature_index.key_count() or
452
(self.chk_index is not None and self.chk_index.key_count()))
454
def finish_content(self):
455
if self.name is not None:
459
self._write_data('', flush=True)
460
self.name = self._hash.hexdigest()
462
def finish(self, suspend=False):
463
"""Finish the new pack.
466
- finalises the content
467
- assigns a name (the md5 of the content, currently)
468
- writes out the associated indices
469
- renames the pack into place.
470
- stores the index size tuple for the pack in the index_sizes
473
self.finish_content()
475
self._check_references()
477
# XXX: It'd be better to write them all to temporary names, then
478
# rename them all into place, so that the window when only some are
479
# visible is smaller. On the other hand none will be seen until
480
# they're in the names list.
481
self.index_sizes = [None, None, None, None]
482
self._write_index('revision', self.revision_index, 'revision',
484
self._write_index('inventory', self.inventory_index, 'inventory',
486
self._write_index('text', self.text_index, 'file texts', suspend)
487
self._write_index('signature', self.signature_index,
488
'revision signatures', suspend)
489
if self.chk_index is not None:
490
self.index_sizes.append(None)
491
self._write_index('chk', self.chk_index,
492
'content hash bytes', suspend)
493
self.write_stream.close(
494
want_fdatasync=self._pack_collection.config_stack.get('repository.fdatasync'))
495
# Note that this will clobber an existing pack with the same name,
496
# without checking for hash collisions. While this is undesirable this
497
# is something that can be rectified in a subsequent release. One way
498
# to rectify it may be to leave the pack at the original name, writing
499
# its pack-names entry as something like 'HASH: index-sizes
500
# temporary-name'. Allocate that and check for collisions, if it is
501
# collision free then rename it into place. If clients know this scheme
502
# they can handle missing-file errors by:
503
# - try for HASH.pack
504
# - try for temporary-name
505
# - refresh the pack-list to see if the pack is now absent
506
new_name = self.name + '.pack'
508
new_name = '../packs/' + new_name
509
self.upload_transport.move(self.random_name, new_name)
510
self._state = 'finished'
511
if 'pack' in debug.debug_flags:
512
# XXX: size might be interesting?
513
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
514
time.ctime(), self.upload_transport.base, self.random_name,
515
new_name, time.time() - self.start_time)
518
"""Flush any current data."""
520
bytes = ''.join(self._buffer[0])
521
self.write_stream.write(bytes)
522
self._hash.update(bytes)
523
self._buffer[:] = [[], 0]
525
def _get_external_refs(self, index):
526
return index._external_references()
528
def set_write_cache_size(self, size):
529
self._cache_limit = size
531
def _write_index(self, index_type, index, label, suspend=False):
532
"""Write out an index.
534
:param index_type: The type of index to write - e.g. 'revision'.
535
:param index: The index object to serialise.
536
:param label: What label to give the index e.g. 'revision'.
538
index_name = self.index_name(index_type, self.name)
540
transport = self.upload_transport
542
transport = self.index_transport
543
index_tempfile = index.finish()
544
index_bytes = index_tempfile.read()
545
write_stream = transport.open_write_stream(index_name,
546
mode=self._file_mode)
547
write_stream.write(index_bytes)
549
want_fdatasync=self._pack_collection.config_stack.get('repository.fdatasync'))
550
self.index_sizes[self.index_offset(index_type)] = len(index_bytes)
551
if 'pack' in debug.debug_flags:
552
# XXX: size might be interesting?
553
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
554
time.ctime(), label, self.upload_transport.base,
555
self.random_name, time.time() - self.start_time)
556
# Replace the writable index on this object with a readonly,
557
# presently unloaded index. We should alter
558
# the index layer to make its finish() error if add_node is
559
# subsequently used. RBC
560
self._replace_index_with_readonly(index_type)
563
class AggregateIndex(object):
564
"""An aggregated index for the RepositoryPackCollection.
566
AggregateIndex is reponsible for managing the PackAccess object,
567
Index-To-Pack mapping, and all indices list for a specific type of index
568
such as 'revision index'.
570
A CombinedIndex provides an index on a single key space built up
571
from several on-disk indices. The AggregateIndex builds on this
572
to provide a knit access layer, and allows having up to one writable
573
index within the collection.
575
# XXX: Probably 'can be written to' could/should be separated from 'acts
576
# like a knit index' -- mbp 20071024
578
def __init__(self, reload_func=None, flush_func=None):
579
"""Create an AggregateIndex.
581
:param reload_func: A function to call if we find we are missing an
582
index. Should have the form reload_func() => True if the list of
583
active pack files has changed.
585
self._reload_func = reload_func
586
self.index_to_pack = {}
587
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
588
self.data_access = _DirectPackAccess(self.index_to_pack,
589
reload_func=reload_func,
590
flush_func=flush_func)
591
self.add_callback = None
593
def add_index(self, index, pack):
594
"""Add index to the aggregate, which is an index for Pack pack.
596
Future searches on the aggregate index will seach this new index
597
before all previously inserted indices.
599
:param index: An Index for the pack.
600
:param pack: A Pack instance.
602
# expose it to the index map
603
self.index_to_pack[index] = pack.access_tuple()
604
# put it at the front of the linear index list
605
self.combined_index.insert_index(0, index, pack.name)
607
def add_writable_index(self, index, pack):
608
"""Add an index which is able to have data added to it.
610
There can be at most one writable index at any time. Any
611
modifications made to the knit are put into this index.
613
:param index: An index from the pack parameter.
614
:param pack: A Pack instance.
616
if self.add_callback is not None:
617
raise AssertionError(
618
"%s already has a writable index through %s" % \
619
(self, self.add_callback))
620
# allow writing: queue writes to a new index
621
self.add_index(index, pack)
622
# Updates the index to packs mapping as a side effect,
623
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
624
self.add_callback = index.add_nodes
627
"""Reset all the aggregate data to nothing."""
628
self.data_access.set_writer(None, None, (None, None))
629
self.index_to_pack.clear()
630
del self.combined_index._indices[:]
631
del self.combined_index._index_names[:]
632
self.add_callback = None
634
def remove_index(self, index):
635
"""Remove index from the indices used to answer queries.
637
:param index: An index from the pack parameter.
639
del self.index_to_pack[index]
640
pos = self.combined_index._indices.index(index)
641
del self.combined_index._indices[pos]
642
del self.combined_index._index_names[pos]
643
if (self.add_callback is not None and
644
getattr(index, 'add_nodes', None) == self.add_callback):
645
self.add_callback = None
646
self.data_access.set_writer(None, None, (None, None))
649
class Packer(object):
650
"""Create a pack from packs."""
652
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
656
:param pack_collection: A RepositoryPackCollection object where the
657
new pack is being written to.
658
:param packs: The packs to combine.
659
:param suffix: The suffix to use on the temporary files for the pack.
660
:param revision_ids: Revision ids to limit the pack to.
661
:param reload_func: A function to call if a pack file/index goes
662
missing. The side effect of calling this function should be to
663
update self.packs. See also AggregateIndex
667
self.revision_ids = revision_ids
668
# The pack object we are creating.
670
self._pack_collection = pack_collection
671
self._reload_func = reload_func
672
# The index layer keys for the revisions being copied. None for 'all
674
self._revision_keys = None
675
# What text keys to copy. None for 'all texts'. This is set by
676
# _copy_inventory_texts
677
self._text_filter = None
679
def pack(self, pb=None):
680
"""Create a new pack by reading data from other packs.
682
This does little more than a bulk copy of data. One key difference
683
is that data with the same item key across multiple packs is elided
684
from the output. The new pack is written into the current pack store
685
along with its indices, and the name added to the pack names. The
686
source packs are not altered and are not required to be in the current
689
:param pb: An optional progress bar to use. A nested bar is created if
691
:return: A Pack object, or None if nothing was copied.
693
# open a pack - using the same name as the last temporary file
694
# - which has already been flushed, so it's safe.
695
# XXX: - duplicate code warning with start_write_group; fix before
696
# considering 'done'.
697
if self._pack_collection._new_pack is not None:
698
raise errors.BzrError('call to %s.pack() while another pack is'
700
% (self.__class__.__name__,))
701
if self.revision_ids is not None:
702
if len(self.revision_ids) == 0:
703
# silly fetch request.
706
self.revision_ids = frozenset(self.revision_ids)
707
self.revision_keys = frozenset((revid,) for revid in
710
self.pb = ui.ui_factory.nested_progress_bar()
714
return self._create_pack_from_packs()
720
"""Open a pack for the pack we are creating."""
721
new_pack = self._pack_collection.pack_factory(self._pack_collection,
722
upload_suffix=self.suffix,
723
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
724
# We know that we will process all nodes in order, and don't need to
725
# query, so don't combine any indices spilled to disk until we are done
726
new_pack.revision_index.set_optimize(combine_backing_indices=False)
727
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
728
new_pack.text_index.set_optimize(combine_backing_indices=False)
729
new_pack.signature_index.set_optimize(combine_backing_indices=False)
732
def _copy_revision_texts(self):
733
"""Copy revision data to the new pack."""
734
raise NotImplementedError(self._copy_revision_texts)
736
def _copy_inventory_texts(self):
737
"""Copy the inventory texts to the new pack.
739
self._revision_keys is used to determine what inventories to copy.
741
Sets self._text_filter appropriately.
743
raise NotImplementedError(self._copy_inventory_texts)
745
def _copy_text_texts(self):
746
raise NotImplementedError(self._copy_text_texts)
748
def _create_pack_from_packs(self):
749
raise NotImplementedError(self._create_pack_from_packs)
751
def _log_copied_texts(self):
752
if 'pack' in debug.debug_flags:
753
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
754
time.ctime(), self._pack_collection._upload_transport.base,
755
self.new_pack.random_name,
756
self.new_pack.text_index.key_count(),
757
time.time() - self.new_pack.start_time)
759
def _use_pack(self, new_pack):
760
"""Return True if new_pack should be used.
762
:param new_pack: The pack that has just been created.
763
:return: True if the pack should be used.
765
return new_pack.data_inserted()
768
class RepositoryPackCollection(object):
769
"""Management of packs within a repository.
771
:ivar _names: map of {pack_name: (index_size,)}
775
resumed_pack_factory = None
776
normal_packer_class = None
777
optimising_packer_class = None
779
def __init__(self, repo, transport, index_transport, upload_transport,
780
pack_transport, index_builder_class, index_class,
782
"""Create a new RepositoryPackCollection.
784
:param transport: Addresses the repository base directory
785
(typically .bzr/repository/).
786
:param index_transport: Addresses the directory containing indices.
787
:param upload_transport: Addresses the directory into which packs are written
788
while they're being created.
789
:param pack_transport: Addresses the directory of existing complete packs.
790
:param index_builder_class: The index builder class to use.
791
:param index_class: The index class to use.
792
:param use_chk_index: Whether to setup and manage a CHK index.
794
# XXX: This should call self.reset()
796
self.transport = transport
797
self._index_transport = index_transport
798
self._upload_transport = upload_transport
799
self._pack_transport = pack_transport
800
self._index_builder_class = index_builder_class
801
self._index_class = index_class
802
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
807
self._packs_by_name = {}
808
# the previous pack-names content
809
self._packs_at_load = None
810
# when a pack is being created by this object, the state of that pack.
811
self._new_pack = None
812
# aggregated revision index data
813
flush = self._flush_new_pack
814
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
815
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
816
self.text_index = AggregateIndex(self.reload_pack_names, flush)
817
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
818
all_indices = [self.revision_index, self.inventory_index,
819
self.text_index, self.signature_index]
821
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
822
all_indices.append(self.chk_index)
824
# used to determine if we're using a chk_index elsewhere.
825
self.chk_index = None
826
# Tell all the CombinedGraphIndex objects about each other, so they can
827
# share hints about which pack names to search first.
828
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
829
for combined_idx in all_combined:
830
combined_idx.set_sibling_indices(
831
set(all_combined).difference([combined_idx]))
833
self._resumed_packs = []
834
self.config_stack = config.LocationStack(self.transport.base)
837
return '%s(%r)' % (self.__class__.__name__, self.repo)
839
def add_pack_to_memory(self, pack):
840
"""Make a Pack object available to the repository to satisfy queries.
842
:param pack: A Pack object.
844
if pack.name in self._packs_by_name:
845
raise AssertionError(
846
'pack %s already in _packs_by_name' % (pack.name,))
847
self.packs.append(pack)
848
self._packs_by_name[pack.name] = pack
849
self.revision_index.add_index(pack.revision_index, pack)
850
self.inventory_index.add_index(pack.inventory_index, pack)
851
self.text_index.add_index(pack.text_index, pack)
852
self.signature_index.add_index(pack.signature_index, pack)
853
if self.chk_index is not None:
854
self.chk_index.add_index(pack.chk_index, pack)
857
"""Return a list of all the Pack objects this repository has.
859
Note that an in-progress pack being created is not returned.
861
:return: A list of Pack objects for all the packs in the repository.
864
for name in self.names():
865
result.append(self.get_pack_by_name(name))
869
"""Pack the pack collection incrementally.
871
This will not attempt global reorganisation or recompression,
872
rather it will just ensure that the total number of packs does
873
not grow without bound. It uses the _max_pack_count method to
874
determine if autopacking is needed, and the pack_distribution
875
method to determine the number of revisions in each pack.
877
If autopacking takes place then the packs name collection will have
878
been flushed to disk - packing requires updating the name collection
879
in synchronisation with certain steps. Otherwise the names collection
882
:return: Something evaluating true if packing took place.
886
return self._do_autopack()
887
except errors.RetryAutopack:
888
# If we get a RetryAutopack exception, we should abort the
889
# current action, and retry.
892
def _do_autopack(self):
893
# XXX: Should not be needed when the management of indices is sane.
894
total_revisions = self.revision_index.combined_index.key_count()
895
total_packs = len(self._names)
896
if self._max_pack_count(total_revisions) >= total_packs:
898
# determine which packs need changing
899
pack_distribution = self.pack_distribution(total_revisions)
901
for pack in self.all_packs():
902
revision_count = pack.get_revision_count()
903
if revision_count == 0:
904
# revision less packs are not generated by normal operation,
905
# only by operations like sign-my-commits, and thus will not
906
# tend to grow rapdily or without bound like commit containing
907
# packs do - leave them alone as packing them really should
908
# group their data with the relevant commit, and that may
909
# involve rewriting ancient history - which autopack tries to
910
# avoid. Alternatively we could not group the data but treat
911
# each of these as having a single revision, and thus add
912
# one revision for each to the total revision count, to get
913
# a matching distribution.
915
existing_packs.append((revision_count, pack))
916
pack_operations = self.plan_autopack_combinations(
917
existing_packs, pack_distribution)
918
num_new_packs = len(pack_operations)
919
num_old_packs = sum([len(po[1]) for po in pack_operations])
920
num_revs_affected = sum([po[0] for po in pack_operations])
921
mutter('Auto-packing repository %s, which has %d pack files, '
922
'containing %d revisions. Packing %d files into %d affecting %d'
923
' revisions', self, total_packs, total_revisions, num_old_packs,
924
num_new_packs, num_revs_affected)
925
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
926
reload_func=self._restart_autopack)
927
mutter('Auto-packing repository %s completed', self)
930
def _execute_pack_operations(self, pack_operations, packer_class,
932
"""Execute a series of pack operations.
934
:param pack_operations: A list of [revision_count, packs_to_combine].
935
:param packer_class: The class of packer to use
936
:return: The new pack names.
938
for revision_count, packs in pack_operations:
939
# we may have no-ops from the setup logic
942
packer = packer_class(self, packs, '.autopack',
943
reload_func=reload_func)
945
result = packer.pack()
946
except errors.RetryWithNewPacks:
947
# An exception is propagating out of this context, make sure
948
# this packer has cleaned up. Packer() doesn't set its new_pack
949
# state into the RepositoryPackCollection object, so we only
950
# have access to it directly here.
951
if packer.new_pack is not None:
952
packer.new_pack.abort()
957
self._remove_pack_from_memory(pack)
958
# record the newly available packs and stop advertising the old
961
for _, packs in pack_operations:
962
to_be_obsoleted.extend(packs)
963
result = self._save_pack_names(clear_obsolete_packs=True,
964
obsolete_packs=to_be_obsoleted)
967
def _flush_new_pack(self):
968
if self._new_pack is not None:
969
self._new_pack.flush()
971
def lock_names(self):
972
"""Acquire the mutex around the pack-names index.
974
This cannot be used in the middle of a read-only transaction on the
977
self.repo.control_files.lock_write()
979
def _already_packed(self):
980
"""Is the collection already packed?"""
981
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
983
def pack(self, hint=None, clean_obsolete_packs=False):
984
"""Pack the pack collection totally."""
986
total_packs = len(self._names)
987
if self._already_packed():
989
total_revisions = self.revision_index.combined_index.key_count()
990
# XXX: the following may want to be a class, to pack with a given
992
mutter('Packing repository %s, which has %d pack files, '
993
'containing %d revisions with hint %r.', self, total_packs,
994
total_revisions, hint)
997
self._try_pack_operations(hint)
998
except RetryPackOperations:
1002
if clean_obsolete_packs:
1003
self._clear_obsolete_packs()
1005
def _try_pack_operations(self, hint):
1006
"""Calculate the pack operations based on the hint (if any), and
1009
# determine which packs need changing
1010
pack_operations = [[0, []]]
1011
for pack in self.all_packs():
1012
if hint is None or pack.name in hint:
1013
# Either no hint was provided (so we are packing everything),
1014
# or this pack was included in the hint.
1015
pack_operations[-1][0] += pack.get_revision_count()
1016
pack_operations[-1][1].append(pack)
1017
self._execute_pack_operations(pack_operations,
1018
packer_class=self.optimising_packer_class,
1019
reload_func=self._restart_pack_operations)
1021
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1022
"""Plan a pack operation.
1024
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1026
:param pack_distribution: A list with the number of revisions desired
1029
if len(existing_packs) <= len(pack_distribution):
1031
existing_packs.sort(reverse=True)
1032
pack_operations = [[0, []]]
1033
# plan out what packs to keep, and what to reorganise
1034
while len(existing_packs):
1035
# take the largest pack, and if it's less than the head of the
1036
# distribution chart we will include its contents in the new pack
1037
# for that position. If it's larger, we remove its size from the
1038
# distribution chart
1039
next_pack_rev_count, next_pack = existing_packs.pop(0)
1040
if next_pack_rev_count >= pack_distribution[0]:
1041
# this is already packed 'better' than this, so we can
1042
# not waste time packing it.
1043
while next_pack_rev_count > 0:
1044
next_pack_rev_count -= pack_distribution[0]
1045
if next_pack_rev_count >= 0:
1047
del pack_distribution[0]
1049
# didn't use that entire bucket up
1050
pack_distribution[0] = -next_pack_rev_count
1052
# add the revisions we're going to add to the next output pack
1053
pack_operations[-1][0] += next_pack_rev_count
1054
# allocate this pack to the next pack sub operation
1055
pack_operations[-1][1].append(next_pack)
1056
if pack_operations[-1][0] >= pack_distribution[0]:
1057
# this pack is used up, shift left.
1058
del pack_distribution[0]
1059
pack_operations.append([0, []])
1060
# Now that we know which pack files we want to move, shove them all
1061
# into a single pack file.
1063
final_pack_list = []
1064
for num_revs, pack_files in pack_operations:
1065
final_rev_count += num_revs
1066
final_pack_list.extend(pack_files)
1067
if len(final_pack_list) == 1:
1068
raise AssertionError('We somehow generated an autopack with a'
1069
' single pack file being moved.')
1071
return [[final_rev_count, final_pack_list]]
1073
def ensure_loaded(self):
1074
"""Ensure we have read names from disk.
1076
:return: True if the disk names had not been previously read.
1078
# NB: if you see an assertion error here, it's probably access against
1079
# an unlocked repo. Naughty.
1080
if not self.repo.is_locked():
1081
raise errors.ObjectNotLocked(self.repo)
1082
if self._names is None:
1084
self._packs_at_load = set()
1085
for index, key, value in self._iter_disk_pack_index():
1087
self._names[name] = self._parse_index_sizes(value)
1088
self._packs_at_load.add((key, value))
1092
# populate all the metadata.
1096
def _parse_index_sizes(self, value):
1097
"""Parse a string of index sizes."""
1098
return tuple([int(digits) for digits in value.split(' ')])
1100
def get_pack_by_name(self, name):
1101
"""Get a Pack object by name.
1103
:param name: The name of the pack - e.g. '123456'
1104
:return: A Pack object.
1107
return self._packs_by_name[name]
1109
rev_index = self._make_index(name, '.rix')
1110
inv_index = self._make_index(name, '.iix')
1111
txt_index = self._make_index(name, '.tix')
1112
sig_index = self._make_index(name, '.six')
1113
if self.chk_index is not None:
1114
chk_index = self._make_index(name, '.cix', is_chk=True)
1117
result = ExistingPack(self._pack_transport, name, rev_index,
1118
inv_index, txt_index, sig_index, chk_index)
1119
self.add_pack_to_memory(result)
1122
def _resume_pack(self, name):
1123
"""Get a suspended Pack object by name.
1125
:param name: The name of the pack - e.g. '123456'
1126
:return: A Pack object.
1128
if not re.match('[a-f0-9]{32}', name):
1129
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1131
raise errors.UnresumableWriteGroup(
1132
self.repo, [name], 'Malformed write group token')
1134
rev_index = self._make_index(name, '.rix', resume=True)
1135
inv_index = self._make_index(name, '.iix', resume=True)
1136
txt_index = self._make_index(name, '.tix', resume=True)
1137
sig_index = self._make_index(name, '.six', resume=True)
1138
if self.chk_index is not None:
1139
chk_index = self._make_index(name, '.cix', resume=True,
1143
result = self.resumed_pack_factory(name, rev_index, inv_index,
1144
txt_index, sig_index, self._upload_transport,
1145
self._pack_transport, self._index_transport, self,
1146
chk_index=chk_index)
1147
except errors.NoSuchFile, e:
1148
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1149
self.add_pack_to_memory(result)
1150
self._resumed_packs.append(result)
1153
def allocate(self, a_new_pack):
1154
"""Allocate name in the list of packs.
1156
:param a_new_pack: A NewPack instance to be added to the collection of
1157
packs for this repository.
1159
self.ensure_loaded()
1160
if a_new_pack.name in self._names:
1161
raise errors.BzrError(
1162
'Pack %r already exists in %s' % (a_new_pack.name, self))
1163
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1164
self.add_pack_to_memory(a_new_pack)
1166
def _iter_disk_pack_index(self):
1167
"""Iterate over the contents of the pack-names index.
1169
This is used when loading the list from disk, and before writing to
1170
detect updates from others during our write operation.
1171
:return: An iterator of the index contents.
1173
return self._index_class(self.transport, 'pack-names', None
1174
).iter_all_entries()
1176
def _make_index(self, name, suffix, resume=False, is_chk=False):
1177
size_offset = self._suffix_offsets[suffix]
1178
index_name = name + suffix
1180
transport = self._upload_transport
1181
index_size = transport.stat(index_name).st_size
1183
transport = self._index_transport
1184
index_size = self._names[name][size_offset]
1185
index = self._index_class(transport, index_name, index_size,
1186
unlimited_cache=is_chk)
1187
if is_chk and self._index_class is btree_index.BTreeGraphIndex:
1188
index._leaf_factory = btree_index._gcchk_factory
1191
def _max_pack_count(self, total_revisions):
1192
"""Return the maximum number of packs to use for total revisions.
1194
:param total_revisions: The total number of revisions in the
1197
if not total_revisions:
1199
digits = str(total_revisions)
1201
for digit in digits:
1202
result += int(digit)
1206
"""Provide an order to the underlying names."""
1207
return sorted(self._names.keys())
1209
def _obsolete_packs(self, packs):
1210
"""Move a number of packs which have been obsoleted out of the way.
1212
Each pack and its associated indices are moved out of the way.
1214
Note: for correctness this function should only be called after a new
1215
pack names index has been written without these pack names, and with
1216
the names of packs that contain the data previously available via these
1219
:param packs: The packs to obsolete.
1220
:param return: None.
1224
pack.pack_transport.move(pack.file_name(),
1225
'../obsolete_packs/' + pack.file_name())
1226
except (errors.PathError, errors.TransportError), e:
1227
# TODO: Should these be warnings or mutters?
1228
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1230
# TODO: Probably needs to know all possible indices for this pack
1231
# - or maybe list the directory and move all indices matching this
1232
# name whether we recognize it or not?
1233
suffixes = ['.iix', '.six', '.tix', '.rix']
1234
if self.chk_index is not None:
1235
suffixes.append('.cix')
1236
for suffix in suffixes:
1238
self._index_transport.move(pack.name + suffix,
1239
'../obsolete_packs/' + pack.name + suffix)
1240
except (errors.PathError, errors.TransportError), e:
1241
mutter("couldn't rename obsolete index, skipping it:\n%s"
1244
def pack_distribution(self, total_revisions):
1245
"""Generate a list of the number of revisions to put in each pack.
1247
:param total_revisions: The total number of revisions in the
1250
if total_revisions == 0:
1252
digits = reversed(str(total_revisions))
1254
for exponent, count in enumerate(digits):
1255
size = 10 ** exponent
1256
for pos in range(int(count)):
1258
return list(reversed(result))
1260
def _pack_tuple(self, name):
1261
"""Return a tuple with the transport and file name for a pack name."""
1262
return self._pack_transport, name + '.pack'
1264
def _remove_pack_from_memory(self, pack):
1265
"""Remove pack from the packs accessed by this repository.
1267
Only affects memory state, until self._save_pack_names() is invoked.
1269
self._names.pop(pack.name)
1270
self._packs_by_name.pop(pack.name)
1271
self._remove_pack_indices(pack)
1272
self.packs.remove(pack)
1274
def _remove_pack_indices(self, pack, ignore_missing=False):
1275
"""Remove the indices for pack from the aggregated indices.
1277
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1279
for index_type in Pack.index_definitions.keys():
1280
attr_name = index_type + '_index'
1281
aggregate_index = getattr(self, attr_name)
1282
if aggregate_index is not None:
1283
pack_index = getattr(pack, attr_name)
1285
aggregate_index.remove_index(pack_index)
1292
"""Clear all cached data."""
1293
# cached revision data
1294
self.revision_index.clear()
1295
# cached signature data
1296
self.signature_index.clear()
1297
# cached file text data
1298
self.text_index.clear()
1299
# cached inventory data
1300
self.inventory_index.clear()
1302
if self.chk_index is not None:
1303
self.chk_index.clear()
1304
# remove the open pack
1305
self._new_pack = None
1306
# information about packs.
1309
self._packs_by_name = {}
1310
self._packs_at_load = None
1312
def _unlock_names(self):
1313
"""Release the mutex around the pack-names index."""
1314
self.repo.control_files.unlock()
1316
def _diff_pack_names(self):
1317
"""Read the pack names from disk, and compare it to the one in memory.
1319
:return: (disk_nodes, deleted_nodes, new_nodes)
1320
disk_nodes The final set of nodes that should be referenced
1321
deleted_nodes Nodes which have been removed from when we started
1322
new_nodes Nodes that are newly introduced
1324
# load the disk nodes across
1326
for index, key, value in self._iter_disk_pack_index():
1327
disk_nodes.add((key, value))
1328
orig_disk_nodes = set(disk_nodes)
1330
# do a two-way diff against our original content
1331
current_nodes = set()
1332
for name, sizes in self._names.iteritems():
1334
((name, ), ' '.join(str(size) for size in sizes)))
1336
# Packs no longer present in the repository, which were present when we
1337
# locked the repository
1338
deleted_nodes = self._packs_at_load - current_nodes
1339
# Packs which this process is adding
1340
new_nodes = current_nodes - self._packs_at_load
1342
# Update the disk_nodes set to include the ones we are adding, and
1343
# remove the ones which were removed by someone else
1344
disk_nodes.difference_update(deleted_nodes)
1345
disk_nodes.update(new_nodes)
1347
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1349
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1350
"""Given the correct set of pack files, update our saved info.
1352
:return: (removed, added, modified)
1353
removed pack names removed from self._names
1354
added pack names added to self._names
1355
modified pack names that had changed value
1360
## self._packs_at_load = disk_nodes
1361
new_names = dict(disk_nodes)
1362
# drop no longer present nodes
1363
for pack in self.all_packs():
1364
if (pack.name,) not in new_names:
1365
removed.append(pack.name)
1366
self._remove_pack_from_memory(pack)
1367
# add new nodes/refresh existing ones
1368
for key, value in disk_nodes:
1370
sizes = self._parse_index_sizes(value)
1371
if name in self._names:
1373
if sizes != self._names[name]:
1374
# the pack for name has had its indices replaced - rare but
1375
# important to handle. XXX: probably can never happen today
1376
# because the three-way merge code above does not handle it
1377
# - you may end up adding the same key twice to the new
1378
# disk index because the set values are the same, unless
1379
# the only index shows up as deleted by the set difference
1380
# - which it may. Until there is a specific test for this,
1381
# assume it's broken. RBC 20071017.
1382
self._remove_pack_from_memory(self.get_pack_by_name(name))
1383
self._names[name] = sizes
1384
self.get_pack_by_name(name)
1385
modified.append(name)
1388
self._names[name] = sizes
1389
self.get_pack_by_name(name)
1391
return removed, added, modified
1393
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1394
"""Save the list of packs.
1396
This will take out the mutex around the pack names list for the
1397
duration of the method call. If concurrent updates have been made, a
1398
three-way merge between the current list and the current in memory list
1401
:param clear_obsolete_packs: If True, clear out the contents of the
1402
obsolete_packs directory.
1403
:param obsolete_packs: Packs that are obsolete once the new pack-names
1404
file has been written.
1405
:return: A list of the names saved that were not previously on disk.
1407
already_obsolete = []
1410
builder = self._index_builder_class()
1411
(disk_nodes, deleted_nodes, new_nodes,
1412
orig_disk_nodes) = self._diff_pack_names()
1413
# TODO: handle same-name, index-size-changes here -
1414
# e.g. use the value from disk, not ours, *unless* we're the one
1416
for key, value in disk_nodes:
1417
builder.add_node(key, value)
1418
self.transport.put_file('pack-names', builder.finish(),
1419
mode=self.repo.bzrdir._get_file_mode())
1420
self._packs_at_load = disk_nodes
1421
if clear_obsolete_packs:
1424
to_preserve = set([o.name for o in obsolete_packs])
1425
already_obsolete = self._clear_obsolete_packs(to_preserve)
1427
self._unlock_names()
1428
# synchronise the memory packs list with what we just wrote:
1429
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1431
# TODO: We could add one more condition here. "if o.name not in
1432
# orig_disk_nodes and o != the new_pack we haven't written to
1433
# disk yet. However, the new pack object is not easily
1434
# accessible here (it would have to be passed through the
1435
# autopacking code, etc.)
1436
obsolete_packs = [o for o in obsolete_packs
1437
if o.name not in already_obsolete]
1438
self._obsolete_packs(obsolete_packs)
1439
return [new_node[0][0] for new_node in new_nodes]
1441
def reload_pack_names(self):
1442
"""Sync our pack listing with what is present in the repository.
1444
This should be called when we find out that something we thought was
1445
present is now missing. This happens when another process re-packs the
1448
:return: True if the in-memory list of packs has been altered at all.
1450
# The ensure_loaded call is to handle the case where the first call
1451
# made involving the collection was to reload_pack_names, where we
1452
# don't have a view of disk contents. It's a bit of a bandaid, and
1453
# causes two reads of pack-names, but it's a rare corner case not
1454
# struck with regular push/pull etc.
1455
first_read = self.ensure_loaded()
1458
# out the new value.
1459
(disk_nodes, deleted_nodes, new_nodes,
1460
orig_disk_nodes) = self._diff_pack_names()
1461
# _packs_at_load is meant to be the explicit list of names in
1462
# 'pack-names' at then start. As such, it should not contain any
1463
# pending names that haven't been written out yet.
1464
self._packs_at_load = orig_disk_nodes
1466
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1467
if removed or added or modified:
1471
def _restart_autopack(self):
1472
"""Reload the pack names list, and restart the autopack code."""
1473
if not self.reload_pack_names():
1474
# Re-raise the original exception, because something went missing
1475
# and a restart didn't find it
1477
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1479
def _restart_pack_operations(self):
1480
"""Reload the pack names list, and restart the autopack code."""
1481
if not self.reload_pack_names():
1482
# Re-raise the original exception, because something went missing
1483
# and a restart didn't find it
1485
raise RetryPackOperations(self.repo, False, sys.exc_info())
1487
def _clear_obsolete_packs(self, preserve=None):
1488
"""Delete everything from the obsolete-packs directory.
1490
:return: A list of pack identifiers (the filename without '.pack') that
1491
were found in obsolete_packs.
1494
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1495
if preserve is None:
1497
for filename in obsolete_pack_transport.list_dir('.'):
1498
name, ext = osutils.splitext(filename)
1501
if name in preserve:
1504
obsolete_pack_transport.delete(filename)
1505
except (errors.PathError, errors.TransportError), e:
1506
warning("couldn't delete obsolete pack, skipping it:\n%s"
1510
def _start_write_group(self):
1511
# Do not permit preparation for writing if we're not in a 'write lock'.
1512
if not self.repo.is_write_locked():
1513
raise errors.NotWriteLocked(self)
1514
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1515
file_mode=self.repo.bzrdir._get_file_mode())
1516
# allow writing: queue writes to a new index
1517
self.revision_index.add_writable_index(self._new_pack.revision_index,
1519
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1521
self.text_index.add_writable_index(self._new_pack.text_index,
1523
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
1524
self.signature_index.add_writable_index(self._new_pack.signature_index,
1526
if self.chk_index is not None:
1527
self.chk_index.add_writable_index(self._new_pack.chk_index,
1529
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
1530
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
1532
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1533
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1534
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1535
self.repo.texts._index._add_callback = self.text_index.add_callback
1537
def _abort_write_group(self):
1538
# FIXME: just drop the transient index.
1539
# forget what names there are
1540
if self._new_pack is not None:
1541
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
1542
operation.add_cleanup(setattr, self, '_new_pack', None)
1543
# If we aborted while in the middle of finishing the write
1544
# group, _remove_pack_indices could fail because the indexes are
1545
# already gone. But they're not there we shouldn't fail in this
1546
# case, so we pass ignore_missing=True.
1547
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
1548
ignore_missing=True)
1549
operation.run_simple()
1550
for resumed_pack in self._resumed_packs:
1551
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
1552
# See comment in previous finally block.
1553
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
1554
ignore_missing=True)
1555
operation.run_simple()
1556
del self._resumed_packs[:]
1558
def _remove_resumed_pack_indices(self):
1559
for resumed_pack in self._resumed_packs:
1560
self._remove_pack_indices(resumed_pack)
1561
del self._resumed_packs[:]
1563
def _check_new_inventories(self):
1564
"""Detect missing inventories in this write group.
1566
:returns: list of strs, summarising any problems found. If the list is
1567
empty no problems were found.
1569
# The base implementation does no checks. GCRepositoryPackCollection
1573
def _commit_write_group(self):
1575
for prefix, versioned_file in (
1576
('revisions', self.repo.revisions),
1577
('inventories', self.repo.inventories),
1578
('texts', self.repo.texts),
1579
('signatures', self.repo.signatures),
1581
missing = versioned_file.get_missing_compression_parent_keys()
1582
all_missing.update([(prefix,) + key for key in missing])
1584
raise errors.BzrCheckError(
1585
"Repository %s has missing compression parent(s) %r "
1586
% (self.repo, sorted(all_missing)))
1587
problems = self._check_new_inventories()
1589
problems_summary = '\n'.join(problems)
1590
raise errors.BzrCheckError(
1591
"Cannot add revision(s) to repository: " + problems_summary)
1592
self._remove_pack_indices(self._new_pack)
1593
any_new_content = False
1594
if self._new_pack.data_inserted():
1595
# get all the data to disk and read to use
1596
self._new_pack.finish()
1597
self.allocate(self._new_pack)
1598
self._new_pack = None
1599
any_new_content = True
1601
self._new_pack.abort()
1602
self._new_pack = None
1603
for resumed_pack in self._resumed_packs:
1604
# XXX: this is a pretty ugly way to turn the resumed pack into a
1605
# properly committed pack.
1606
self._names[resumed_pack.name] = None
1607
self._remove_pack_from_memory(resumed_pack)
1608
resumed_pack.finish()
1609
self.allocate(resumed_pack)
1610
any_new_content = True
1611
del self._resumed_packs[:]
1613
result = self.autopack()
1615
# when autopack takes no steps, the names list is still
1617
return self._save_pack_names()
1621
def _suspend_write_group(self):
1622
tokens = [pack.name for pack in self._resumed_packs]
1623
self._remove_pack_indices(self._new_pack)
1624
if self._new_pack.data_inserted():
1625
# get all the data to disk and read to use
1626
self._new_pack.finish(suspend=True)
1627
tokens.append(self._new_pack.name)
1628
self._new_pack = None
1630
self._new_pack.abort()
1631
self._new_pack = None
1632
self._remove_resumed_pack_indices()
1635
def _resume_write_group(self, tokens):
1636
for token in tokens:
1637
self._resume_pack(token)
1640
class PackRepository(MetaDirVersionedFileRepository):
1641
"""Repository with knit objects stored inside pack containers.
1643
The layering for a KnitPackRepository is:
1645
Graph | HPSS | Repository public layer |
1646
===================================================
1647
Tuple based apis below, string based, and key based apis above
1648
---------------------------------------------------
1650
Provides .texts, .revisions etc
1651
This adapts the N-tuple keys to physical knit records which only have a
1652
single string identifier (for historical reasons), which in older formats
1653
was always the revision_id, and in the mapped code for packs is always
1654
the last element of key tuples.
1655
---------------------------------------------------
1657
A separate GraphIndex is used for each of the
1658
texts/inventories/revisions/signatures contained within each individual
1659
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1661
===================================================
1665
# These attributes are inherited from the Repository base class. Setting
1666
# them to None ensures that if the constructor is changed to not initialize
1667
# them, or a subclass fails to call the constructor, that an error will
1668
# occur rather than the system working but generating incorrect data.
1669
_commit_builder_class = None
1672
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1674
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1675
self._commit_builder_class = _commit_builder_class
1676
self._serializer = _serializer
1677
self._reconcile_fixes_text_parents = True
1678
if self._format.supports_external_lookups:
1679
self._unstacked_provider = graph.CachingParentsProvider(
1680
self._make_parents_provider_unstacked())
1682
self._unstacked_provider = graph.CachingParentsProvider(self)
1683
self._unstacked_provider.disable_cache()
1686
def _all_revision_ids(self):
1687
"""See Repository.all_revision_ids()."""
1688
return [key[0] for key in self.revisions.keys()]
1690
def _abort_write_group(self):
1691
self.revisions._index._key_dependencies.clear()
1692
self._pack_collection._abort_write_group()
1694
def _make_parents_provider(self):
1695
if not self._format.supports_external_lookups:
1696
return self._unstacked_provider
1697
return graph.StackedParentsProvider(_LazyListJoin(
1698
[self._unstacked_provider], self._fallback_repositories))
1700
def _refresh_data(self):
1701
if not self.is_locked():
1703
self._pack_collection.reload_pack_names()
1704
self._unstacked_provider.disable_cache()
1705
self._unstacked_provider.enable_cache()
1707
def _start_write_group(self):
1708
self._pack_collection._start_write_group()
1710
def _commit_write_group(self):
1711
hint = self._pack_collection._commit_write_group()
1712
self.revisions._index._key_dependencies.clear()
1713
# The commit may have added keys that were previously cached as
1714
# missing, so reset the cache.
1715
self._unstacked_provider.disable_cache()
1716
self._unstacked_provider.enable_cache()
1719
def suspend_write_group(self):
1720
# XXX check self._write_group is self.get_transaction()?
1721
tokens = self._pack_collection._suspend_write_group()
1722
self.revisions._index._key_dependencies.clear()
1723
self._write_group = None
1726
def _resume_write_group(self, tokens):
1727
self._start_write_group()
1729
self._pack_collection._resume_write_group(tokens)
1730
except errors.UnresumableWriteGroup:
1731
self._abort_write_group()
1733
for pack in self._pack_collection._resumed_packs:
1734
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1736
def get_transaction(self):
1737
if self._write_lock_count:
1738
return self._transaction
1740
return self.control_files.get_transaction()
1742
def is_locked(self):
1743
return self._write_lock_count or self.control_files.is_locked()
1745
def is_write_locked(self):
1746
return self._write_lock_count
1748
def lock_write(self, token=None):
1749
"""Lock the repository for writes.
1751
:return: A bzrlib.repository.RepositoryWriteLockResult.
1753
locked = self.is_locked()
1754
if not self._write_lock_count and locked:
1755
raise errors.ReadOnlyError(self)
1756
self._write_lock_count += 1
1757
if self._write_lock_count == 1:
1758
self._transaction = transactions.WriteTransaction()
1760
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
1761
note('%r was write locked again', self)
1762
self._prev_lock = 'w'
1763
self._unstacked_provider.enable_cache()
1764
for repo in self._fallback_repositories:
1765
# Writes don't affect fallback repos
1767
self._refresh_data()
1768
return RepositoryWriteLockResult(self.unlock, None)
1770
def lock_read(self):
1771
"""Lock the repository for reads.
1773
:return: A bzrlib.lock.LogicalLockResult.
1775
locked = self.is_locked()
1776
if self._write_lock_count:
1777
self._write_lock_count += 1
1779
self.control_files.lock_read()
1781
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
1782
note('%r was read locked again', self)
1783
self._prev_lock = 'r'
1784
self._unstacked_provider.enable_cache()
1785
for repo in self._fallback_repositories:
1787
self._refresh_data()
1788
return LogicalLockResult(self.unlock)
1790
def leave_lock_in_place(self):
1791
# not supported - raise an error
1792
raise NotImplementedError(self.leave_lock_in_place)
1794
def dont_leave_lock_in_place(self):
1795
# not supported - raise an error
1796
raise NotImplementedError(self.dont_leave_lock_in_place)
1799
def pack(self, hint=None, clean_obsolete_packs=False):
1800
"""Compress the data within the repository.
1802
This will pack all the data to a single pack. In future it may
1803
recompress deltas or do other such expensive operations.
1805
self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
1808
def reconcile(self, other=None, thorough=False):
1809
"""Reconcile this repository."""
1810
from bzrlib.reconcile import PackReconciler
1811
reconciler = PackReconciler(self, thorough=thorough)
1812
reconciler.reconcile()
1815
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1816
raise NotImplementedError(self._reconcile_pack)
1818
@only_raises(errors.LockNotHeld, errors.LockBroken)
1820
if self._write_lock_count == 1 and self._write_group is not None:
1821
self.abort_write_group()
1822
self._unstacked_provider.disable_cache()
1823
self._transaction = None
1824
self._write_lock_count = 0
1825
raise errors.BzrError(
1826
'Must end write group before releasing write lock on %s'
1828
if self._write_lock_count:
1829
self._write_lock_count -= 1
1830
if not self._write_lock_count:
1831
transaction = self._transaction
1832
self._transaction = None
1833
transaction.finish()
1835
self.control_files.unlock()
1837
if not self.is_locked():
1838
self._unstacked_provider.disable_cache()
1839
for repo in self._fallback_repositories:
1843
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
1844
"""Format logic for pack structured repositories.
1846
This repository format has:
1847
- a list of packs in pack-names
1848
- packs in packs/NAME.pack
1849
- indices in indices/NAME.{iix,six,tix,rix}
1850
- knit deltas in the packs, knit indices mapped to the indices.
1851
- thunk objects to support the knits programming API.
1852
- a format marker of its own
1853
- an optional 'shared-storage' flag
1854
- an optional 'no-working-trees' flag
1858
# Set this attribute in derived classes to control the repository class
1859
# created by open and initialize.
1860
repository_class = None
1861
# Set this attribute in derived classes to control the
1862
# _commit_builder_class that the repository objects will have passed to
1863
# their constructor.
1864
_commit_builder_class = None
1865
# Set this attribute in derived clases to control the _serializer that the
1866
# repository objects will have passed to their constructor.
1868
# Packs are not confused by ghosts.
1869
supports_ghosts = True
1870
# External references are not supported in pack repositories yet.
1871
supports_external_lookups = False
1872
# Most pack formats do not use chk lookups.
1873
supports_chks = False
1874
# What index classes to use
1875
index_builder_class = None
1877
_fetch_uses_deltas = True
1879
supports_funky_characters = True
1880
revision_graph_can_have_wrong_parents = True
1882
def initialize(self, a_bzrdir, shared=False):
1883
"""Create a pack based repository.
1885
:param a_bzrdir: bzrdir to contain the new repository; must already
1887
:param shared: If true the repository will be initialized as a shared
1890
mutter('creating repository in %s.', a_bzrdir.transport.base)
1891
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1892
builder = self.index_builder_class()
1893
files = [('pack-names', builder.finish())]
1894
utf8_files = [('format', self.get_format_string())]
1896
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1897
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
1898
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
1901
def open(self, a_bzrdir, _found=False, _override_transport=None):
1902
"""See RepositoryFormat.open().
1904
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1905
repository at a slightly different url
1906
than normal. I.e. during 'upgrade'.
1909
format = RepositoryFormat.find_format(a_bzrdir)
1910
if _override_transport is not None:
1911
repo_transport = _override_transport
1913
repo_transport = a_bzrdir.get_repository_transport(None)
1914
control_files = lockable_files.LockableFiles(repo_transport,
1915
'lock', lockdir.LockDir)
1916
return self.repository_class(_format=self,
1918
control_files=control_files,
1919
_commit_builder_class=self._commit_builder_class,
1920
_serializer=self._serializer)
1923
class RetryPackOperations(errors.RetryWithNewPacks):
1924
"""Raised when we are packing and we find a missing file.
1926
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1927
code it should try again.
1930
internal_error = True
1932
_fmt = ("Pack files have changed, reload and try pack again."
1933
" context: %(context)s %(orig_error)s")
1936
class _DirectPackAccess(object):
1937
"""Access to data in one or more packs with less translation."""
1939
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1940
"""Create a _DirectPackAccess object.
1942
:param index_to_packs: A dict mapping index objects to the transport
1943
and file names for obtaining data.
1944
:param reload_func: A function to call if we determine that the pack
1945
files have moved and we need to reload our caches. See
1946
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1948
self._container_writer = None
1949
self._write_index = None
1950
self._indices = index_to_packs
1951
self._reload_func = reload_func
1952
self._flush_func = flush_func
1954
def add_raw_records(self, key_sizes, raw_data):
1955
"""Add raw knit bytes to a storage area.
1957
The data is spooled to the container writer in one bytes-record per
1960
:param sizes: An iterable of tuples containing the key and size of each
1962
:param raw_data: A bytestring containing the data.
1963
:return: A list of memos to retrieve the record later. Each memo is an
1964
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1965
length), where the index field is the write_index object supplied
1966
to the PackAccess object.
1968
if type(raw_data) is not str:
1969
raise AssertionError(
1970
'data must be plain bytes was %s' % type(raw_data))
1973
for key, size in key_sizes:
1974
p_offset, p_length = self._container_writer.add_bytes_record(
1975
raw_data[offset:offset+size], [])
1977
result.append((self._write_index, p_offset, p_length))
1981
"""Flush pending writes on this access object.
1983
This will flush any buffered writes to a NewPack.
1985
if self._flush_func is not None:
1988
def get_raw_records(self, memos_for_retrieval):
1989
"""Get the raw bytes for a records.
1991
:param memos_for_retrieval: An iterable containing the (index, pos,
1992
length) memo for retrieving the bytes. The Pack access method
1993
looks up the pack to use for a given record in its index_to_pack
1995
:return: An iterator over the bytes of the records.
1997
# first pass, group into same-index requests
1999
current_index = None
2000
for (index, offset, length) in memos_for_retrieval:
2001
if current_index == index:
2002
current_list.append((offset, length))
2004
if current_index is not None:
2005
request_lists.append((current_index, current_list))
2006
current_index = index
2007
current_list = [(offset, length)]
2008
# handle the last entry
2009
if current_index is not None:
2010
request_lists.append((current_index, current_list))
2011
for index, offsets in request_lists:
2013
transport, path = self._indices[index]
2015
# A KeyError here indicates that someone has triggered an index
2016
# reload, and this index has gone missing, we need to start
2018
if self._reload_func is None:
2019
# If we don't have a _reload_func there is nothing that can
2022
raise errors.RetryWithNewPacks(index,
2023
reload_occurred=True,
2024
exc_info=sys.exc_info())
2026
reader = pack.make_readv_reader(transport, path, offsets)
2027
for names, read_func in reader.iter_records():
2028
yield read_func(None)
2029
except errors.NoSuchFile:
2030
# A NoSuchFile error indicates that a pack file has gone
2031
# missing on disk, we need to trigger a reload, and start over.
2032
if self._reload_func is None:
2034
raise errors.RetryWithNewPacks(transport.abspath(path),
2035
reload_occurred=False,
2036
exc_info=sys.exc_info())
2038
def set_writer(self, writer, index, transport_packname):
2039
"""Set a writer to use for adding data."""
2040
if index is not None:
2041
self._indices[index] = transport_packname
2042
self._container_writer = writer
2043
self._write_index = index
2045
def reload_or_raise(self, retry_exc):
2046
"""Try calling the reload function, or re-raise the original exception.
2048
This should be called after _DirectPackAccess raises a
2049
RetryWithNewPacks exception. This function will handle the common logic
2050
of determining when the error is fatal versus being temporary.
2051
It will also make sure that the original exception is raised, rather
2052
than the RetryWithNewPacks exception.
2054
If this function returns, then the calling function should retry
2055
whatever operation was being performed. Otherwise an exception will
2058
:param retry_exc: A RetryWithNewPacks exception.
2061
if self._reload_func is None:
2063
elif not self._reload_func():
2064
# The reload claimed that nothing changed
2065
if not retry_exc.reload_occurred:
2066
# If there wasn't an earlier reload, then we really were
2067
# expecting to find changes. We didn't find them, so this is a
2071
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2072
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
