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
36
from bzrlib.index import (
38
GraphIndexPrefixAdapter,
48
from bzrlib.decorators import (
53
from bzrlib.lock import LogicalLockResult
54
from bzrlib.repository import (
58
RepositoryWriteLockResult,
60
from bzrlib.vf_repository import (
61
MetaDirVersionedFileRepository,
62
MetaDirVersionedFileRepositoryFormat,
63
VersionedFileCommitBuilder,
64
VersionedFileRootCommitBuilder,
66
from bzrlib.trace import (
73
class PackCommitBuilder(VersionedFileCommitBuilder):
74
"""Subclass of VersionedFileCommitBuilder to add texts with pack semantics.
76
Specifically this uses one knit object rather than one knit object per
77
added text, reducing memory and object pressure.
80
def __init__(self, repository, parents, config, timestamp=None,
81
timezone=None, committer=None, revprops=None,
82
revision_id=None, lossy=False):
83
VersionedFileCommitBuilder.__init__(self, repository, parents, config,
84
timestamp=timestamp, timezone=timezone, committer=committer,
85
revprops=revprops, revision_id=revision_id, lossy=lossy)
86
self._file_graph = graph.Graph(
87
repository._pack_collection.text_index.combined_index)
89
def _heads(self, file_id, revision_ids):
90
keys = [(file_id, revision_id) for revision_id in revision_ids]
91
return set([key[1] for key in self._file_graph.heads(keys)])
94
class PackRootCommitBuilder(VersionedFileRootCommitBuilder):
95
"""A subclass of RootCommitBuilder to add texts with pack semantics.
97
Specifically this uses one knit object rather than one knit object per
98
added text, reducing memory and object pressure.
101
def __init__(self, repository, parents, config, timestamp=None,
102
timezone=None, committer=None, revprops=None,
103
revision_id=None, lossy=False):
104
super(PackRootCommitBuilder, self).__init__(repository, parents,
105
config, timestamp=timestamp, timezone=timezone,
106
committer=committer, revprops=revprops, revision_id=revision_id,
108
self._file_graph = graph.Graph(
109
repository._pack_collection.text_index.combined_index)
111
def _heads(self, file_id, revision_ids):
112
keys = [(file_id, revision_id) for revision_id in revision_ids]
113
return set([key[1] for key in self._file_graph.heads(keys)])
117
"""An in memory proxy for a pack and its indices.
119
This is a base class that is not directly used, instead the classes
120
ExistingPack and NewPack are used.
123
# A map of index 'type' to the file extension and position in the
125
index_definitions = {
127
'revision': ('.rix', 0),
128
'inventory': ('.iix', 1),
130
'signature': ('.six', 3),
133
def __init__(self, revision_index, inventory_index, text_index,
134
signature_index, chk_index=None):
135
"""Create a pack instance.
137
:param revision_index: A GraphIndex for determining what revisions are
138
present in the Pack and accessing the locations of their texts.
139
:param inventory_index: A GraphIndex for determining what inventories are
140
present in the Pack and accessing the locations of their
142
:param text_index: A GraphIndex for determining what file texts
143
are present in the pack and accessing the locations of their
144
texts/deltas (via (fileid, revisionid) tuples).
145
:param signature_index: A GraphIndex for determining what signatures are
146
present in the Pack and accessing the locations of their texts.
147
:param chk_index: A GraphIndex for accessing content by CHK, if the
150
self.revision_index = revision_index
151
self.inventory_index = inventory_index
152
self.text_index = text_index
153
self.signature_index = signature_index
154
self.chk_index = chk_index
156
def access_tuple(self):
157
"""Return a tuple (transport, name) for the pack content."""
158
return self.pack_transport, self.file_name()
160
def _check_references(self):
161
"""Make sure our external references are present.
163
Packs are allowed to have deltas whose base is not in the pack, but it
164
must be present somewhere in this collection. It is not allowed to
165
have deltas based on a fallback repository.
166
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
169
for (index_name, external_refs, index) in [
171
self._get_external_refs(self.text_index),
172
self._pack_collection.text_index.combined_index),
174
self._get_external_refs(self.inventory_index),
175
self._pack_collection.inventory_index.combined_index),
177
missing = external_refs.difference(
178
k for (idx, k, v, r) in
179
index.iter_entries(external_refs))
181
missing_items[index_name] = sorted(list(missing))
183
from pprint import pformat
184
raise errors.BzrCheckError(
185
"Newly created pack file %r has delta references to "
186
"items not in its repository:\n%s"
187
% (self, pformat(missing_items)))
190
"""Get the file name for the pack on disk."""
191
return self.name + '.pack'
193
def get_revision_count(self):
194
return self.revision_index.key_count()
196
def index_name(self, index_type, name):
197
"""Get the disk name of an index type for pack name 'name'."""
198
return name + Pack.index_definitions[index_type][0]
200
def index_offset(self, index_type):
201
"""Get the position in a index_size array for a given index type."""
202
return Pack.index_definitions[index_type][1]
204
def inventory_index_name(self, name):
205
"""The inv index is the name + .iix."""
206
return self.index_name('inventory', name)
208
def revision_index_name(self, name):
209
"""The revision index is the name + .rix."""
210
return self.index_name('revision', name)
212
def signature_index_name(self, name):
213
"""The signature index is the name + .six."""
214
return self.index_name('signature', name)
216
def text_index_name(self, name):
217
"""The text index is the name + .tix."""
218
return self.index_name('text', name)
220
def _replace_index_with_readonly(self, index_type):
221
unlimited_cache = False
222
if index_type == 'chk':
223
unlimited_cache = True
224
index = self.index_class(self.index_transport,
225
self.index_name(index_type, self.name),
226
self.index_sizes[self.index_offset(index_type)],
227
unlimited_cache=unlimited_cache)
228
if index_type == 'chk':
229
index._leaf_factory = btree_index._gcchk_factory
230
setattr(self, index_type + '_index', index)
233
class ExistingPack(Pack):
234
"""An in memory proxy for an existing .pack and its disk indices."""
236
def __init__(self, pack_transport, name, revision_index, inventory_index,
237
text_index, signature_index, chk_index=None):
238
"""Create an ExistingPack object.
240
:param pack_transport: The transport where the pack file resides.
241
:param name: The name of the pack on disk in the pack_transport.
243
Pack.__init__(self, revision_index, inventory_index, text_index,
244
signature_index, chk_index)
246
self.pack_transport = pack_transport
247
if None in (revision_index, inventory_index, text_index,
248
signature_index, name, pack_transport):
249
raise AssertionError()
251
def __eq__(self, other):
252
return self.__dict__ == other.__dict__
254
def __ne__(self, other):
255
return not self.__eq__(other)
258
return "<%s.%s object at 0x%x, %s, %s" % (
259
self.__class__.__module__, self.__class__.__name__, id(self),
260
self.pack_transport, self.name)
263
class ResumedPack(ExistingPack):
265
def __init__(self, name, revision_index, inventory_index, text_index,
266
signature_index, upload_transport, pack_transport, index_transport,
267
pack_collection, chk_index=None):
268
"""Create a ResumedPack object."""
269
ExistingPack.__init__(self, pack_transport, name, revision_index,
270
inventory_index, text_index, signature_index,
272
self.upload_transport = upload_transport
273
self.index_transport = index_transport
274
self.index_sizes = [None, None, None, None]
276
('revision', revision_index),
277
('inventory', inventory_index),
278
('text', text_index),
279
('signature', signature_index),
281
if chk_index is not None:
282
indices.append(('chk', chk_index))
283
self.index_sizes.append(None)
284
for index_type, index in indices:
285
offset = self.index_offset(index_type)
286
self.index_sizes[offset] = index._size
287
self.index_class = pack_collection._index_class
288
self._pack_collection = pack_collection
289
self._state = 'resumed'
290
# XXX: perhaps check that the .pack file exists?
292
def access_tuple(self):
293
if self._state == 'finished':
294
return Pack.access_tuple(self)
295
elif self._state == 'resumed':
296
return self.upload_transport, self.file_name()
298
raise AssertionError(self._state)
301
self.upload_transport.delete(self.file_name())
302
indices = [self.revision_index, self.inventory_index, self.text_index,
303
self.signature_index]
304
if self.chk_index is not None:
305
indices.append(self.chk_index)
306
for index in indices:
307
index._transport.delete(index._name)
310
self._check_references()
311
index_types = ['revision', 'inventory', 'text', 'signature']
312
if self.chk_index is not None:
313
index_types.append('chk')
314
for index_type in index_types:
315
old_name = self.index_name(index_type, self.name)
316
new_name = '../indices/' + old_name
317
self.upload_transport.move(old_name, new_name)
318
self._replace_index_with_readonly(index_type)
319
new_name = '../packs/' + self.file_name()
320
self.upload_transport.move(self.file_name(), new_name)
321
self._state = 'finished'
323
def _get_external_refs(self, index):
324
"""Return compression parents for this index that are not present.
326
This returns any compression parents that are referenced by this index,
327
which are not contained *in* this index. They may be present elsewhere.
329
return index.external_references(1)
333
"""An in memory proxy for a pack which is being created."""
335
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
336
"""Create a NewPack instance.
338
:param pack_collection: A PackCollection into which this is being inserted.
339
:param upload_suffix: An optional suffix to be given to any temporary
340
files created during the pack creation. e.g '.autopack'
341
:param file_mode: Unix permissions for newly created file.
343
# The relative locations of the packs are constrained, but all are
344
# passed in because the caller has them, so as to avoid object churn.
345
index_builder_class = pack_collection._index_builder_class
346
if pack_collection.chk_index is not None:
347
chk_index = index_builder_class(reference_lists=0)
351
# Revisions: parents list, no text compression.
352
index_builder_class(reference_lists=1),
353
# Inventory: We want to map compression only, but currently the
354
# knit code hasn't been updated enough to understand that, so we
355
# have a regular 2-list index giving parents and compression
357
index_builder_class(reference_lists=2),
358
# Texts: compression and per file graph, for all fileids - so two
359
# reference lists and two elements in the key tuple.
360
index_builder_class(reference_lists=2, key_elements=2),
361
# Signatures: Just blobs to store, no compression, no parents
363
index_builder_class(reference_lists=0),
364
# CHK based storage - just blobs, no compression or parents.
367
self._pack_collection = pack_collection
368
# When we make readonly indices, we need this.
369
self.index_class = pack_collection._index_class
370
# where should the new pack be opened
371
self.upload_transport = pack_collection._upload_transport
372
# where are indices written out to
373
self.index_transport = pack_collection._index_transport
374
# where is the pack renamed to when it is finished?
375
self.pack_transport = pack_collection._pack_transport
376
# What file mode to upload the pack and indices with.
377
self._file_mode = file_mode
378
# tracks the content written to the .pack file.
379
self._hash = osutils.md5()
380
# a tuple with the length in bytes of the indices, once the pack
381
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
382
self.index_sizes = None
383
# How much data to cache when writing packs. Note that this is not
384
# synchronised with reads, because it's not in the transport layer, so
385
# is not safe unless the client knows it won't be reading from the pack
387
self._cache_limit = 0
388
# the temporary pack file name.
389
self.random_name = osutils.rand_chars(20) + upload_suffix
390
# when was this pack started ?
391
self.start_time = time.time()
392
# open an output stream for the data added to the pack.
393
self.write_stream = self.upload_transport.open_write_stream(
394
self.random_name, mode=self._file_mode)
395
if 'pack' in debug.debug_flags:
396
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
397
time.ctime(), self.upload_transport.base, self.random_name,
398
time.time() - self.start_time)
399
# A list of byte sequences to be written to the new pack, and the
400
# aggregate size of them. Stored as a list rather than separate
401
# variables so that the _write_data closure below can update them.
402
self._buffer = [[], 0]
403
# create a callable for adding data
405
# robertc says- this is a closure rather than a method on the object
406
# so that the variables are locals, and faster than accessing object
408
def _write_data(bytes, flush=False, _buffer=self._buffer,
409
_write=self.write_stream.write, _update=self._hash.update):
410
_buffer[0].append(bytes)
411
_buffer[1] += len(bytes)
413
if _buffer[1] > self._cache_limit or flush:
414
bytes = ''.join(_buffer[0])
418
# expose this on self, for the occasion when clients want to add data.
419
self._write_data = _write_data
420
# a pack writer object to serialise pack records.
421
self._writer = pack.ContainerWriter(self._write_data)
423
# what state is the pack in? (open, finished, aborted)
425
# no name until we finish writing the content
429
"""Cancel creating this pack."""
430
self._state = 'aborted'
431
self.write_stream.close()
432
# Remove the temporary pack file.
433
self.upload_transport.delete(self.random_name)
434
# The indices have no state on disk.
436
def access_tuple(self):
437
"""Return a tuple (transport, name) for the pack content."""
438
if self._state == 'finished':
439
return Pack.access_tuple(self)
440
elif self._state == 'open':
441
return self.upload_transport, self.random_name
443
raise AssertionError(self._state)
445
def data_inserted(self):
446
"""True if data has been added to this pack."""
447
return bool(self.get_revision_count() or
448
self.inventory_index.key_count() or
449
self.text_index.key_count() or
450
self.signature_index.key_count() or
451
(self.chk_index is not None and self.chk_index.key_count()))
453
def finish_content(self):
454
if self.name is not None:
458
self._write_data('', flush=True)
459
self.name = self._hash.hexdigest()
461
def finish(self, suspend=False):
462
"""Finish the new pack.
465
- finalises the content
466
- assigns a name (the md5 of the content, currently)
467
- writes out the associated indices
468
- renames the pack into place.
469
- stores the index size tuple for the pack in the index_sizes
472
self.finish_content()
474
self._check_references()
476
# XXX: It'd be better to write them all to temporary names, then
477
# rename them all into place, so that the window when only some are
478
# visible is smaller. On the other hand none will be seen until
479
# they're in the names list.
480
self.index_sizes = [None, None, None, None]
481
self._write_index('revision', self.revision_index, 'revision', suspend)
482
self._write_index('inventory', self.inventory_index, 'inventory',
484
self._write_index('text', self.text_index, 'file texts', suspend)
485
self._write_index('signature', self.signature_index,
486
'revision signatures', suspend)
487
if self.chk_index is not None:
488
self.index_sizes.append(None)
489
self._write_index('chk', self.chk_index,
490
'content hash bytes', suspend)
491
self.write_stream.close()
492
# Note that this will clobber an existing pack with the same name,
493
# without checking for hash collisions. While this is undesirable this
494
# is something that can be rectified in a subsequent release. One way
495
# to rectify it may be to leave the pack at the original name, writing
496
# its pack-names entry as something like 'HASH: index-sizes
497
# temporary-name'. Allocate that and check for collisions, if it is
498
# collision free then rename it into place. If clients know this scheme
499
# they can handle missing-file errors by:
500
# - try for HASH.pack
501
# - try for temporary-name
502
# - refresh the pack-list to see if the pack is now absent
503
new_name = self.name + '.pack'
505
new_name = '../packs/' + new_name
506
self.upload_transport.move(self.random_name, new_name)
507
self._state = 'finished'
508
if 'pack' in debug.debug_flags:
509
# XXX: size might be interesting?
510
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
511
time.ctime(), self.upload_transport.base, self.random_name,
512
new_name, time.time() - self.start_time)
515
"""Flush any current data."""
517
bytes = ''.join(self._buffer[0])
518
self.write_stream.write(bytes)
519
self._hash.update(bytes)
520
self._buffer[:] = [[], 0]
522
def _get_external_refs(self, index):
523
return index._external_references()
525
def set_write_cache_size(self, size):
526
self._cache_limit = size
528
def _write_index(self, index_type, index, label, suspend=False):
529
"""Write out an index.
531
:param index_type: The type of index to write - e.g. 'revision'.
532
:param index: The index object to serialise.
533
:param label: What label to give the index e.g. 'revision'.
535
index_name = self.index_name(index_type, self.name)
537
transport = self.upload_transport
539
transport = self.index_transport
540
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
541
index_name, index.finish(), mode=self._file_mode)
542
if 'pack' in debug.debug_flags:
543
# XXX: size might be interesting?
544
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
545
time.ctime(), label, self.upload_transport.base,
546
self.random_name, time.time() - self.start_time)
547
# Replace the writable index on this object with a readonly,
548
# presently unloaded index. We should alter
549
# the index layer to make its finish() error if add_node is
550
# subsequently used. RBC
551
self._replace_index_with_readonly(index_type)
554
class AggregateIndex(object):
555
"""An aggregated index for the RepositoryPackCollection.
557
AggregateIndex is reponsible for managing the PackAccess object,
558
Index-To-Pack mapping, and all indices list for a specific type of index
559
such as 'revision index'.
561
A CombinedIndex provides an index on a single key space built up
562
from several on-disk indices. The AggregateIndex builds on this
563
to provide a knit access layer, and allows having up to one writable
564
index within the collection.
566
# XXX: Probably 'can be written to' could/should be separated from 'acts
567
# like a knit index' -- mbp 20071024
569
def __init__(self, reload_func=None, flush_func=None):
570
"""Create an AggregateIndex.
572
:param reload_func: A function to call if we find we are missing an
573
index. Should have the form reload_func() => True if the list of
574
active pack files has changed.
576
self._reload_func = reload_func
577
self.index_to_pack = {}
578
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
579
self.data_access = _DirectPackAccess(self.index_to_pack,
580
reload_func=reload_func,
581
flush_func=flush_func)
582
self.add_callback = None
584
def add_index(self, index, pack):
585
"""Add index to the aggregate, which is an index for Pack pack.
587
Future searches on the aggregate index will seach this new index
588
before all previously inserted indices.
590
:param index: An Index for the pack.
591
:param pack: A Pack instance.
593
# expose it to the index map
594
self.index_to_pack[index] = pack.access_tuple()
595
# put it at the front of the linear index list
596
self.combined_index.insert_index(0, index, pack.name)
598
def add_writable_index(self, index, pack):
599
"""Add an index which is able to have data added to it.
601
There can be at most one writable index at any time. Any
602
modifications made to the knit are put into this index.
604
:param index: An index from the pack parameter.
605
:param pack: A Pack instance.
607
if self.add_callback is not None:
608
raise AssertionError(
609
"%s already has a writable index through %s" % \
610
(self, self.add_callback))
611
# allow writing: queue writes to a new index
612
self.add_index(index, pack)
613
# Updates the index to packs mapping as a side effect,
614
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
615
self.add_callback = index.add_nodes
618
"""Reset all the aggregate data to nothing."""
619
self.data_access.set_writer(None, None, (None, None))
620
self.index_to_pack.clear()
621
del self.combined_index._indices[:]
622
del self.combined_index._index_names[:]
623
self.add_callback = None
625
def remove_index(self, index):
626
"""Remove index from the indices used to answer queries.
628
:param index: An index from the pack parameter.
630
del self.index_to_pack[index]
631
pos = self.combined_index._indices.index(index)
632
del self.combined_index._indices[pos]
633
del self.combined_index._index_names[pos]
634
if (self.add_callback is not None and
635
getattr(index, 'add_nodes', None) == self.add_callback):
636
self.add_callback = None
637
self.data_access.set_writer(None, None, (None, None))
640
class Packer(object):
641
"""Create a pack from packs."""
643
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
647
:param pack_collection: A RepositoryPackCollection object where the
648
new pack is being written to.
649
:param packs: The packs to combine.
650
:param suffix: The suffix to use on the temporary files for the pack.
651
:param revision_ids: Revision ids to limit the pack to.
652
:param reload_func: A function to call if a pack file/index goes
653
missing. The side effect of calling this function should be to
654
update self.packs. See also AggregateIndex
658
self.revision_ids = revision_ids
659
# The pack object we are creating.
661
self._pack_collection = pack_collection
662
self._reload_func = reload_func
663
# The index layer keys for the revisions being copied. None for 'all
665
self._revision_keys = None
666
# What text keys to copy. None for 'all texts'. This is set by
667
# _copy_inventory_texts
668
self._text_filter = None
670
def pack(self, pb=None):
671
"""Create a new pack by reading data from other packs.
673
This does little more than a bulk copy of data. One key difference
674
is that data with the same item key across multiple packs is elided
675
from the output. The new pack is written into the current pack store
676
along with its indices, and the name added to the pack names. The
677
source packs are not altered and are not required to be in the current
680
:param pb: An optional progress bar to use. A nested bar is created if
682
:return: A Pack object, or None if nothing was copied.
684
# open a pack - using the same name as the last temporary file
685
# - which has already been flushed, so it's safe.
686
# XXX: - duplicate code warning with start_write_group; fix before
687
# considering 'done'.
688
if self._pack_collection._new_pack is not None:
689
raise errors.BzrError('call to %s.pack() while another pack is'
691
% (self.__class__.__name__,))
692
if self.revision_ids is not None:
693
if len(self.revision_ids) == 0:
694
# silly fetch request.
697
self.revision_ids = frozenset(self.revision_ids)
698
self.revision_keys = frozenset((revid,) for revid in
701
self.pb = ui.ui_factory.nested_progress_bar()
705
return self._create_pack_from_packs()
711
"""Open a pack for the pack we are creating."""
712
new_pack = self._pack_collection.pack_factory(self._pack_collection,
713
upload_suffix=self.suffix,
714
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
715
# We know that we will process all nodes in order, and don't need to
716
# query, so don't combine any indices spilled to disk until we are done
717
new_pack.revision_index.set_optimize(combine_backing_indices=False)
718
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
719
new_pack.text_index.set_optimize(combine_backing_indices=False)
720
new_pack.signature_index.set_optimize(combine_backing_indices=False)
723
def _copy_revision_texts(self):
724
"""Copy revision data to the new pack."""
725
raise NotImplementedError(self._copy_revision_texts)
727
def _copy_inventory_texts(self):
728
"""Copy the inventory texts to the new pack.
730
self._revision_keys is used to determine what inventories to copy.
732
Sets self._text_filter appropriately.
734
raise NotImplementedError(self._copy_inventory_texts)
736
def _copy_text_texts(self):
737
raise NotImplementedError(self._copy_text_texts)
739
def _create_pack_from_packs(self):
740
raise NotImplementedError(self._create_pack_from_packs)
742
def _log_copied_texts(self):
743
if 'pack' in debug.debug_flags:
744
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
745
time.ctime(), self._pack_collection._upload_transport.base,
746
self.new_pack.random_name,
747
self.new_pack.text_index.key_count(),
748
time.time() - self.new_pack.start_time)
750
def _use_pack(self, new_pack):
751
"""Return True if new_pack should be used.
753
:param new_pack: The pack that has just been created.
754
:return: True if the pack should be used.
756
return new_pack.data_inserted()
759
class RepositoryPackCollection(object):
760
"""Management of packs within a repository.
762
:ivar _names: map of {pack_name: (index_size,)}
766
resumed_pack_factory = None
767
normal_packer_class = None
768
optimising_packer_class = None
770
def __init__(self, repo, transport, index_transport, upload_transport,
771
pack_transport, index_builder_class, index_class,
773
"""Create a new RepositoryPackCollection.
775
:param transport: Addresses the repository base directory
776
(typically .bzr/repository/).
777
:param index_transport: Addresses the directory containing indices.
778
:param upload_transport: Addresses the directory into which packs are written
779
while they're being created.
780
:param pack_transport: Addresses the directory of existing complete packs.
781
:param index_builder_class: The index builder class to use.
782
:param index_class: The index class to use.
783
:param use_chk_index: Whether to setup and manage a CHK index.
785
# XXX: This should call self.reset()
787
self.transport = transport
788
self._index_transport = index_transport
789
self._upload_transport = upload_transport
790
self._pack_transport = pack_transport
791
self._index_builder_class = index_builder_class
792
self._index_class = index_class
793
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
798
self._packs_by_name = {}
799
# the previous pack-names content
800
self._packs_at_load = None
801
# when a pack is being created by this object, the state of that pack.
802
self._new_pack = None
803
# aggregated revision index data
804
flush = self._flush_new_pack
805
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
806
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
807
self.text_index = AggregateIndex(self.reload_pack_names, flush)
808
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
809
all_indices = [self.revision_index, self.inventory_index,
810
self.text_index, self.signature_index]
812
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
813
all_indices.append(self.chk_index)
815
# used to determine if we're using a chk_index elsewhere.
816
self.chk_index = None
817
# Tell all the CombinedGraphIndex objects about each other, so they can
818
# share hints about which pack names to search first.
819
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
820
for combined_idx in all_combined:
821
combined_idx.set_sibling_indices(
822
set(all_combined).difference([combined_idx]))
824
self._resumed_packs = []
827
return '%s(%r)' % (self.__class__.__name__, self.repo)
829
def add_pack_to_memory(self, pack):
830
"""Make a Pack object available to the repository to satisfy queries.
832
:param pack: A Pack object.
834
if pack.name in self._packs_by_name:
835
raise AssertionError(
836
'pack %s already in _packs_by_name' % (pack.name,))
837
self.packs.append(pack)
838
self._packs_by_name[pack.name] = pack
839
self.revision_index.add_index(pack.revision_index, pack)
840
self.inventory_index.add_index(pack.inventory_index, pack)
841
self.text_index.add_index(pack.text_index, pack)
842
self.signature_index.add_index(pack.signature_index, pack)
843
if self.chk_index is not None:
844
self.chk_index.add_index(pack.chk_index, pack)
847
"""Return a list of all the Pack objects this repository has.
849
Note that an in-progress pack being created is not returned.
851
:return: A list of Pack objects for all the packs in the repository.
854
for name in self.names():
855
result.append(self.get_pack_by_name(name))
859
"""Pack the pack collection incrementally.
861
This will not attempt global reorganisation or recompression,
862
rather it will just ensure that the total number of packs does
863
not grow without bound. It uses the _max_pack_count method to
864
determine if autopacking is needed, and the pack_distribution
865
method to determine the number of revisions in each pack.
867
If autopacking takes place then the packs name collection will have
868
been flushed to disk - packing requires updating the name collection
869
in synchronisation with certain steps. Otherwise the names collection
872
:return: Something evaluating true if packing took place.
876
return self._do_autopack()
877
except errors.RetryAutopack:
878
# If we get a RetryAutopack exception, we should abort the
879
# current action, and retry.
882
def _do_autopack(self):
883
# XXX: Should not be needed when the management of indices is sane.
884
total_revisions = self.revision_index.combined_index.key_count()
885
total_packs = len(self._names)
886
if self._max_pack_count(total_revisions) >= total_packs:
888
# determine which packs need changing
889
pack_distribution = self.pack_distribution(total_revisions)
891
for pack in self.all_packs():
892
revision_count = pack.get_revision_count()
893
if revision_count == 0:
894
# revision less packs are not generated by normal operation,
895
# only by operations like sign-my-commits, and thus will not
896
# tend to grow rapdily or without bound like commit containing
897
# packs do - leave them alone as packing them really should
898
# group their data with the relevant commit, and that may
899
# involve rewriting ancient history - which autopack tries to
900
# avoid. Alternatively we could not group the data but treat
901
# each of these as having a single revision, and thus add
902
# one revision for each to the total revision count, to get
903
# a matching distribution.
905
existing_packs.append((revision_count, pack))
906
pack_operations = self.plan_autopack_combinations(
907
existing_packs, pack_distribution)
908
num_new_packs = len(pack_operations)
909
num_old_packs = sum([len(po[1]) for po in pack_operations])
910
num_revs_affected = sum([po[0] for po in pack_operations])
911
mutter('Auto-packing repository %s, which has %d pack files, '
912
'containing %d revisions. Packing %d files into %d affecting %d'
913
' revisions', self, total_packs, total_revisions, num_old_packs,
914
num_new_packs, num_revs_affected)
915
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
916
reload_func=self._restart_autopack)
917
mutter('Auto-packing repository %s completed', self)
920
def _execute_pack_operations(self, pack_operations, packer_class,
922
"""Execute a series of pack operations.
924
:param pack_operations: A list of [revision_count, packs_to_combine].
925
:param packer_class: The class of packer to use
926
:return: The new pack names.
928
for revision_count, packs in pack_operations:
929
# we may have no-ops from the setup logic
932
packer = packer_class(self, packs, '.autopack',
933
reload_func=reload_func)
935
result = packer.pack()
936
except errors.RetryWithNewPacks:
937
# An exception is propagating out of this context, make sure
938
# this packer has cleaned up. Packer() doesn't set its new_pack
939
# state into the RepositoryPackCollection object, so we only
940
# have access to it directly here.
941
if packer.new_pack is not None:
942
packer.new_pack.abort()
947
self._remove_pack_from_memory(pack)
948
# record the newly available packs and stop advertising the old
951
for _, packs in pack_operations:
952
to_be_obsoleted.extend(packs)
953
result = self._save_pack_names(clear_obsolete_packs=True,
954
obsolete_packs=to_be_obsoleted)
957
def _flush_new_pack(self):
958
if self._new_pack is not None:
959
self._new_pack.flush()
961
def lock_names(self):
962
"""Acquire the mutex around the pack-names index.
964
This cannot be used in the middle of a read-only transaction on the
967
self.repo.control_files.lock_write()
969
def _already_packed(self):
970
"""Is the collection already packed?"""
971
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
973
def pack(self, hint=None, clean_obsolete_packs=False):
974
"""Pack the pack collection totally."""
976
total_packs = len(self._names)
977
if self._already_packed():
979
total_revisions = self.revision_index.combined_index.key_count()
980
# XXX: the following may want to be a class, to pack with a given
982
mutter('Packing repository %s, which has %d pack files, '
983
'containing %d revisions with hint %r.', self, total_packs,
984
total_revisions, hint)
987
self._try_pack_operations(hint)
988
except RetryPackOperations:
992
if clean_obsolete_packs:
993
self._clear_obsolete_packs()
995
def _try_pack_operations(self, hint):
996
"""Calculate the pack operations based on the hint (if any), and
999
# determine which packs need changing
1000
pack_operations = [[0, []]]
1001
for pack in self.all_packs():
1002
if hint is None or pack.name in hint:
1003
# Either no hint was provided (so we are packing everything),
1004
# or this pack was included in the hint.
1005
pack_operations[-1][0] += pack.get_revision_count()
1006
pack_operations[-1][1].append(pack)
1007
self._execute_pack_operations(pack_operations,
1008
packer_class=self.optimising_packer_class,
1009
reload_func=self._restart_pack_operations)
1011
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1012
"""Plan a pack operation.
1014
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1016
:param pack_distribution: A list with the number of revisions desired
1019
if len(existing_packs) <= len(pack_distribution):
1021
existing_packs.sort(reverse=True)
1022
pack_operations = [[0, []]]
1023
# plan out what packs to keep, and what to reorganise
1024
while len(existing_packs):
1025
# take the largest pack, and if it's less than the head of the
1026
# distribution chart we will include its contents in the new pack
1027
# for that position. If it's larger, we remove its size from the
1028
# distribution chart
1029
next_pack_rev_count, next_pack = existing_packs.pop(0)
1030
if next_pack_rev_count >= pack_distribution[0]:
1031
# this is already packed 'better' than this, so we can
1032
# not waste time packing it.
1033
while next_pack_rev_count > 0:
1034
next_pack_rev_count -= pack_distribution[0]
1035
if next_pack_rev_count >= 0:
1037
del pack_distribution[0]
1039
# didn't use that entire bucket up
1040
pack_distribution[0] = -next_pack_rev_count
1042
# add the revisions we're going to add to the next output pack
1043
pack_operations[-1][0] += next_pack_rev_count
1044
# allocate this pack to the next pack sub operation
1045
pack_operations[-1][1].append(next_pack)
1046
if pack_operations[-1][0] >= pack_distribution[0]:
1047
# this pack is used up, shift left.
1048
del pack_distribution[0]
1049
pack_operations.append([0, []])
1050
# Now that we know which pack files we want to move, shove them all
1051
# into a single pack file.
1053
final_pack_list = []
1054
for num_revs, pack_files in pack_operations:
1055
final_rev_count += num_revs
1056
final_pack_list.extend(pack_files)
1057
if len(final_pack_list) == 1:
1058
raise AssertionError('We somehow generated an autopack with a'
1059
' single pack file being moved.')
1061
return [[final_rev_count, final_pack_list]]
1063
def ensure_loaded(self):
1064
"""Ensure we have read names from disk.
1066
:return: True if the disk names had not been previously read.
1068
# NB: if you see an assertion error here, it's probably access against
1069
# an unlocked repo. Naughty.
1070
if not self.repo.is_locked():
1071
raise errors.ObjectNotLocked(self.repo)
1072
if self._names is None:
1074
self._packs_at_load = set()
1075
for index, key, value in self._iter_disk_pack_index():
1077
self._names[name] = self._parse_index_sizes(value)
1078
self._packs_at_load.add((key, value))
1082
# populate all the metadata.
1086
def _parse_index_sizes(self, value):
1087
"""Parse a string of index sizes."""
1088
return tuple([int(digits) for digits in value.split(' ')])
1090
def get_pack_by_name(self, name):
1091
"""Get a Pack object by name.
1093
:param name: The name of the pack - e.g. '123456'
1094
:return: A Pack object.
1097
return self._packs_by_name[name]
1099
rev_index = self._make_index(name, '.rix')
1100
inv_index = self._make_index(name, '.iix')
1101
txt_index = self._make_index(name, '.tix')
1102
sig_index = self._make_index(name, '.six')
1103
if self.chk_index is not None:
1104
chk_index = self._make_index(name, '.cix', is_chk=True)
1107
result = ExistingPack(self._pack_transport, name, rev_index,
1108
inv_index, txt_index, sig_index, chk_index)
1109
self.add_pack_to_memory(result)
1112
def _resume_pack(self, name):
1113
"""Get a suspended Pack object by name.
1115
:param name: The name of the pack - e.g. '123456'
1116
:return: A Pack object.
1118
if not re.match('[a-f0-9]{32}', name):
1119
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1121
raise errors.UnresumableWriteGroup(
1122
self.repo, [name], 'Malformed write group token')
1124
rev_index = self._make_index(name, '.rix', resume=True)
1125
inv_index = self._make_index(name, '.iix', resume=True)
1126
txt_index = self._make_index(name, '.tix', resume=True)
1127
sig_index = self._make_index(name, '.six', resume=True)
1128
if self.chk_index is not None:
1129
chk_index = self._make_index(name, '.cix', resume=True,
1133
result = self.resumed_pack_factory(name, rev_index, inv_index,
1134
txt_index, sig_index, self._upload_transport,
1135
self._pack_transport, self._index_transport, self,
1136
chk_index=chk_index)
1137
except errors.NoSuchFile, e:
1138
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1139
self.add_pack_to_memory(result)
1140
self._resumed_packs.append(result)
1143
def allocate(self, a_new_pack):
1144
"""Allocate name in the list of packs.
1146
:param a_new_pack: A NewPack instance to be added to the collection of
1147
packs for this repository.
1149
self.ensure_loaded()
1150
if a_new_pack.name in self._names:
1151
raise errors.BzrError(
1152
'Pack %r already exists in %s' % (a_new_pack.name, self))
1153
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1154
self.add_pack_to_memory(a_new_pack)
1156
def _iter_disk_pack_index(self):
1157
"""Iterate over the contents of the pack-names index.
1159
This is used when loading the list from disk, and before writing to
1160
detect updates from others during our write operation.
1161
:return: An iterator of the index contents.
1163
return self._index_class(self.transport, 'pack-names', None
1164
).iter_all_entries()
1166
def _make_index(self, name, suffix, resume=False, is_chk=False):
1167
size_offset = self._suffix_offsets[suffix]
1168
index_name = name + suffix
1170
transport = self._upload_transport
1171
index_size = transport.stat(index_name).st_size
1173
transport = self._index_transport
1174
index_size = self._names[name][size_offset]
1175
index = self._index_class(transport, index_name, index_size,
1176
unlimited_cache=is_chk)
1177
if is_chk and self._index_class is btree_index.BTreeGraphIndex:
1178
index._leaf_factory = btree_index._gcchk_factory
1181
def _max_pack_count(self, total_revisions):
1182
"""Return the maximum number of packs to use for total revisions.
1184
:param total_revisions: The total number of revisions in the
1187
if not total_revisions:
1189
digits = str(total_revisions)
1191
for digit in digits:
1192
result += int(digit)
1196
"""Provide an order to the underlying names."""
1197
return sorted(self._names.keys())
1199
def _obsolete_packs(self, packs):
1200
"""Move a number of packs which have been obsoleted out of the way.
1202
Each pack and its associated indices are moved out of the way.
1204
Note: for correctness this function should only be called after a new
1205
pack names index has been written without these pack names, and with
1206
the names of packs that contain the data previously available via these
1209
:param packs: The packs to obsolete.
1210
:param return: None.
1214
pack.pack_transport.move(pack.file_name(),
1215
'../obsolete_packs/' + pack.file_name())
1216
except (errors.PathError, errors.TransportError), e:
1217
# TODO: Should these be warnings or mutters?
1218
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1220
# TODO: Probably needs to know all possible indices for this pack
1221
# - or maybe list the directory and move all indices matching this
1222
# name whether we recognize it or not?
1223
suffixes = ['.iix', '.six', '.tix', '.rix']
1224
if self.chk_index is not None:
1225
suffixes.append('.cix')
1226
for suffix in suffixes:
1228
self._index_transport.move(pack.name + suffix,
1229
'../obsolete_packs/' + pack.name + suffix)
1230
except (errors.PathError, errors.TransportError), e:
1231
mutter("couldn't rename obsolete index, skipping it:\n%s"
1234
def pack_distribution(self, total_revisions):
1235
"""Generate a list of the number of revisions to put in each pack.
1237
:param total_revisions: The total number of revisions in the
1240
if total_revisions == 0:
1242
digits = reversed(str(total_revisions))
1244
for exponent, count in enumerate(digits):
1245
size = 10 ** exponent
1246
for pos in range(int(count)):
1248
return list(reversed(result))
1250
def _pack_tuple(self, name):
1251
"""Return a tuple with the transport and file name for a pack name."""
1252
return self._pack_transport, name + '.pack'
1254
def _remove_pack_from_memory(self, pack):
1255
"""Remove pack from the packs accessed by this repository.
1257
Only affects memory state, until self._save_pack_names() is invoked.
1259
self._names.pop(pack.name)
1260
self._packs_by_name.pop(pack.name)
1261
self._remove_pack_indices(pack)
1262
self.packs.remove(pack)
1264
def _remove_pack_indices(self, pack, ignore_missing=False):
1265
"""Remove the indices for pack from the aggregated indices.
1267
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1269
for index_type in Pack.index_definitions.keys():
1270
attr_name = index_type + '_index'
1271
aggregate_index = getattr(self, attr_name)
1272
if aggregate_index is not None:
1273
pack_index = getattr(pack, attr_name)
1275
aggregate_index.remove_index(pack_index)
1282
"""Clear all cached data."""
1283
# cached revision data
1284
self.revision_index.clear()
1285
# cached signature data
1286
self.signature_index.clear()
1287
# cached file text data
1288
self.text_index.clear()
1289
# cached inventory data
1290
self.inventory_index.clear()
1292
if self.chk_index is not None:
1293
self.chk_index.clear()
1294
# remove the open pack
1295
self._new_pack = None
1296
# information about packs.
1299
self._packs_by_name = {}
1300
self._packs_at_load = None
1302
def _unlock_names(self):
1303
"""Release the mutex around the pack-names index."""
1304
self.repo.control_files.unlock()
1306
def _diff_pack_names(self):
1307
"""Read the pack names from disk, and compare it to the one in memory.
1309
:return: (disk_nodes, deleted_nodes, new_nodes)
1310
disk_nodes The final set of nodes that should be referenced
1311
deleted_nodes Nodes which have been removed from when we started
1312
new_nodes Nodes that are newly introduced
1314
# load the disk nodes across
1316
for index, key, value in self._iter_disk_pack_index():
1317
disk_nodes.add((key, value))
1318
orig_disk_nodes = set(disk_nodes)
1320
# do a two-way diff against our original content
1321
current_nodes = set()
1322
for name, sizes in self._names.iteritems():
1324
((name, ), ' '.join(str(size) for size in sizes)))
1326
# Packs no longer present in the repository, which were present when we
1327
# locked the repository
1328
deleted_nodes = self._packs_at_load - current_nodes
1329
# Packs which this process is adding
1330
new_nodes = current_nodes - self._packs_at_load
1332
# Update the disk_nodes set to include the ones we are adding, and
1333
# remove the ones which were removed by someone else
1334
disk_nodes.difference_update(deleted_nodes)
1335
disk_nodes.update(new_nodes)
1337
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1339
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1340
"""Given the correct set of pack files, update our saved info.
1342
:return: (removed, added, modified)
1343
removed pack names removed from self._names
1344
added pack names added to self._names
1345
modified pack names that had changed value
1350
## self._packs_at_load = disk_nodes
1351
new_names = dict(disk_nodes)
1352
# drop no longer present nodes
1353
for pack in self.all_packs():
1354
if (pack.name,) not in new_names:
1355
removed.append(pack.name)
1356
self._remove_pack_from_memory(pack)
1357
# add new nodes/refresh existing ones
1358
for key, value in disk_nodes:
1360
sizes = self._parse_index_sizes(value)
1361
if name in self._names:
1363
if sizes != self._names[name]:
1364
# the pack for name has had its indices replaced - rare but
1365
# important to handle. XXX: probably can never happen today
1366
# because the three-way merge code above does not handle it
1367
# - you may end up adding the same key twice to the new
1368
# disk index because the set values are the same, unless
1369
# the only index shows up as deleted by the set difference
1370
# - which it may. Until there is a specific test for this,
1371
# assume it's broken. RBC 20071017.
1372
self._remove_pack_from_memory(self.get_pack_by_name(name))
1373
self._names[name] = sizes
1374
self.get_pack_by_name(name)
1375
modified.append(name)
1378
self._names[name] = sizes
1379
self.get_pack_by_name(name)
1381
return removed, added, modified
1383
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1384
"""Save the list of packs.
1386
This will take out the mutex around the pack names list for the
1387
duration of the method call. If concurrent updates have been made, a
1388
three-way merge between the current list and the current in memory list
1391
:param clear_obsolete_packs: If True, clear out the contents of the
1392
obsolete_packs directory.
1393
:param obsolete_packs: Packs that are obsolete once the new pack-names
1394
file has been written.
1395
:return: A list of the names saved that were not previously on disk.
1397
already_obsolete = []
1400
builder = self._index_builder_class()
1401
(disk_nodes, deleted_nodes, new_nodes,
1402
orig_disk_nodes) = self._diff_pack_names()
1403
# TODO: handle same-name, index-size-changes here -
1404
# e.g. use the value from disk, not ours, *unless* we're the one
1406
for key, value in disk_nodes:
1407
builder.add_node(key, value)
1408
self.transport.put_file('pack-names', builder.finish(),
1409
mode=self.repo.bzrdir._get_file_mode())
1410
self._packs_at_load = disk_nodes
1411
if clear_obsolete_packs:
1414
to_preserve = set([o.name for o in obsolete_packs])
1415
already_obsolete = self._clear_obsolete_packs(to_preserve)
1417
self._unlock_names()
1418
# synchronise the memory packs list with what we just wrote:
1419
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1421
# TODO: We could add one more condition here. "if o.name not in
1422
# orig_disk_nodes and o != the new_pack we haven't written to
1423
# disk yet. However, the new pack object is not easily
1424
# accessible here (it would have to be passed through the
1425
# autopacking code, etc.)
1426
obsolete_packs = [o for o in obsolete_packs
1427
if o.name not in already_obsolete]
1428
self._obsolete_packs(obsolete_packs)
1429
return [new_node[0][0] for new_node in new_nodes]
1431
def reload_pack_names(self):
1432
"""Sync our pack listing with what is present in the repository.
1434
This should be called when we find out that something we thought was
1435
present is now missing. This happens when another process re-packs the
1438
:return: True if the in-memory list of packs has been altered at all.
1440
# The ensure_loaded call is to handle the case where the first call
1441
# made involving the collection was to reload_pack_names, where we
1442
# don't have a view of disk contents. It's a bit of a bandaid, and
1443
# causes two reads of pack-names, but it's a rare corner case not
1444
# struck with regular push/pull etc.
1445
first_read = self.ensure_loaded()
1448
# out the new value.
1449
(disk_nodes, deleted_nodes, new_nodes,
1450
orig_disk_nodes) = self._diff_pack_names()
1451
# _packs_at_load is meant to be the explicit list of names in
1452
# 'pack-names' at then start. As such, it should not contain any
1453
# pending names that haven't been written out yet.
1454
self._packs_at_load = orig_disk_nodes
1456
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1457
if removed or added or modified:
1461
def _restart_autopack(self):
1462
"""Reload the pack names list, and restart the autopack code."""
1463
if not self.reload_pack_names():
1464
# Re-raise the original exception, because something went missing
1465
# and a restart didn't find it
1467
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1469
def _restart_pack_operations(self):
1470
"""Reload the pack names list, and restart the autopack code."""
1471
if not self.reload_pack_names():
1472
# Re-raise the original exception, because something went missing
1473
# and a restart didn't find it
1475
raise RetryPackOperations(self.repo, False, sys.exc_info())
1477
def _clear_obsolete_packs(self, preserve=None):
1478
"""Delete everything from the obsolete-packs directory.
1480
:return: A list of pack identifiers (the filename without '.pack') that
1481
were found in obsolete_packs.
1484
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1485
if preserve is None:
1487
for filename in obsolete_pack_transport.list_dir('.'):
1488
name, ext = osutils.splitext(filename)
1491
if name in preserve:
1494
obsolete_pack_transport.delete(filename)
1495
except (errors.PathError, errors.TransportError), e:
1496
warning("couldn't delete obsolete pack, skipping it:\n%s"
1500
def _start_write_group(self):
1501
# Do not permit preparation for writing if we're not in a 'write lock'.
1502
if not self.repo.is_write_locked():
1503
raise errors.NotWriteLocked(self)
1504
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1505
file_mode=self.repo.bzrdir._get_file_mode())
1506
# allow writing: queue writes to a new index
1507
self.revision_index.add_writable_index(self._new_pack.revision_index,
1509
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1511
self.text_index.add_writable_index(self._new_pack.text_index,
1513
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
1514
self.signature_index.add_writable_index(self._new_pack.signature_index,
1516
if self.chk_index is not None:
1517
self.chk_index.add_writable_index(self._new_pack.chk_index,
1519
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
1520
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
1522
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1523
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1524
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1525
self.repo.texts._index._add_callback = self.text_index.add_callback
1527
def _abort_write_group(self):
1528
# FIXME: just drop the transient index.
1529
# forget what names there are
1530
if self._new_pack is not None:
1531
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
1532
operation.add_cleanup(setattr, self, '_new_pack', None)
1533
# If we aborted while in the middle of finishing the write
1534
# group, _remove_pack_indices could fail because the indexes are
1535
# already gone. But they're not there we shouldn't fail in this
1536
# case, so we pass ignore_missing=True.
1537
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
1538
ignore_missing=True)
1539
operation.run_simple()
1540
for resumed_pack in self._resumed_packs:
1541
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
1542
# See comment in previous finally block.
1543
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
1544
ignore_missing=True)
1545
operation.run_simple()
1546
del self._resumed_packs[:]
1548
def _remove_resumed_pack_indices(self):
1549
for resumed_pack in self._resumed_packs:
1550
self._remove_pack_indices(resumed_pack)
1551
del self._resumed_packs[:]
1553
def _check_new_inventories(self):
1554
"""Detect missing inventories in this write group.
1556
:returns: list of strs, summarising any problems found. If the list is
1557
empty no problems were found.
1559
# The base implementation does no checks. GCRepositoryPackCollection
1563
def _commit_write_group(self):
1565
for prefix, versioned_file in (
1566
('revisions', self.repo.revisions),
1567
('inventories', self.repo.inventories),
1568
('texts', self.repo.texts),
1569
('signatures', self.repo.signatures),
1571
missing = versioned_file.get_missing_compression_parent_keys()
1572
all_missing.update([(prefix,) + key for key in missing])
1574
raise errors.BzrCheckError(
1575
"Repository %s has missing compression parent(s) %r "
1576
% (self.repo, sorted(all_missing)))
1577
problems = self._check_new_inventories()
1579
problems_summary = '\n'.join(problems)
1580
raise errors.BzrCheckError(
1581
"Cannot add revision(s) to repository: " + problems_summary)
1582
self._remove_pack_indices(self._new_pack)
1583
any_new_content = False
1584
if self._new_pack.data_inserted():
1585
# get all the data to disk and read to use
1586
self._new_pack.finish()
1587
self.allocate(self._new_pack)
1588
self._new_pack = None
1589
any_new_content = True
1591
self._new_pack.abort()
1592
self._new_pack = None
1593
for resumed_pack in self._resumed_packs:
1594
# XXX: this is a pretty ugly way to turn the resumed pack into a
1595
# properly committed pack.
1596
self._names[resumed_pack.name] = None
1597
self._remove_pack_from_memory(resumed_pack)
1598
resumed_pack.finish()
1599
self.allocate(resumed_pack)
1600
any_new_content = True
1601
del self._resumed_packs[:]
1603
result = self.autopack()
1605
# when autopack takes no steps, the names list is still
1607
return self._save_pack_names()
1611
def _suspend_write_group(self):
1612
tokens = [pack.name for pack in self._resumed_packs]
1613
self._remove_pack_indices(self._new_pack)
1614
if self._new_pack.data_inserted():
1615
# get all the data to disk and read to use
1616
self._new_pack.finish(suspend=True)
1617
tokens.append(self._new_pack.name)
1618
self._new_pack = None
1620
self._new_pack.abort()
1621
self._new_pack = None
1622
self._remove_resumed_pack_indices()
1625
def _resume_write_group(self, tokens):
1626
for token in tokens:
1627
self._resume_pack(token)
1630
class PackRepository(MetaDirVersionedFileRepository):
1631
"""Repository with knit objects stored inside pack containers.
1633
The layering for a KnitPackRepository is:
1635
Graph | HPSS | Repository public layer |
1636
===================================================
1637
Tuple based apis below, string based, and key based apis above
1638
---------------------------------------------------
1640
Provides .texts, .revisions etc
1641
This adapts the N-tuple keys to physical knit records which only have a
1642
single string identifier (for historical reasons), which in older formats
1643
was always the revision_id, and in the mapped code for packs is always
1644
the last element of key tuples.
1645
---------------------------------------------------
1647
A separate GraphIndex is used for each of the
1648
texts/inventories/revisions/signatures contained within each individual
1649
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1651
===================================================
1655
# These attributes are inherited from the Repository base class. Setting
1656
# them to None ensures that if the constructor is changed to not initialize
1657
# them, or a subclass fails to call the constructor, that an error will
1658
# occur rather than the system working but generating incorrect data.
1659
_commit_builder_class = None
1662
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1664
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1665
self._commit_builder_class = _commit_builder_class
1666
self._serializer = _serializer
1667
self._reconcile_fixes_text_parents = True
1668
if self._format.supports_external_lookups:
1669
self._unstacked_provider = graph.CachingParentsProvider(
1670
self._make_parents_provider_unstacked())
1672
self._unstacked_provider = graph.CachingParentsProvider(self)
1673
self._unstacked_provider.disable_cache()
1676
def _all_revision_ids(self):
1677
"""See Repository.all_revision_ids()."""
1678
return [key[0] for key in self.revisions.keys()]
1680
def _abort_write_group(self):
1681
self.revisions._index._key_dependencies.clear()
1682
self._pack_collection._abort_write_group()
1684
def _make_parents_provider(self):
1685
if not self._format.supports_external_lookups:
1686
return self._unstacked_provider
1687
return graph.StackedParentsProvider(_LazyListJoin(
1688
[self._unstacked_provider], self._fallback_repositories))
1690
def _refresh_data(self):
1691
if not self.is_locked():
1693
self._pack_collection.reload_pack_names()
1694
self._unstacked_provider.disable_cache()
1695
self._unstacked_provider.enable_cache()
1697
def _start_write_group(self):
1698
self._pack_collection._start_write_group()
1700
def _commit_write_group(self):
1701
hint = self._pack_collection._commit_write_group()
1702
self.revisions._index._key_dependencies.clear()
1703
# The commit may have added keys that were previously cached as
1704
# missing, so reset the cache.
1705
self._unstacked_provider.disable_cache()
1706
self._unstacked_provider.enable_cache()
1709
def suspend_write_group(self):
1710
# XXX check self._write_group is self.get_transaction()?
1711
tokens = self._pack_collection._suspend_write_group()
1712
self.revisions._index._key_dependencies.clear()
1713
self._write_group = None
1716
def _resume_write_group(self, tokens):
1717
self._start_write_group()
1719
self._pack_collection._resume_write_group(tokens)
1720
except errors.UnresumableWriteGroup:
1721
self._abort_write_group()
1723
for pack in self._pack_collection._resumed_packs:
1724
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1726
def get_transaction(self):
1727
if self._write_lock_count:
1728
return self._transaction
1730
return self.control_files.get_transaction()
1732
def is_locked(self):
1733
return self._write_lock_count or self.control_files.is_locked()
1735
def is_write_locked(self):
1736
return self._write_lock_count
1738
def lock_write(self, token=None):
1739
"""Lock the repository for writes.
1741
:return: A bzrlib.repository.RepositoryWriteLockResult.
1743
locked = self.is_locked()
1744
if not self._write_lock_count and locked:
1745
raise errors.ReadOnlyError(self)
1746
self._write_lock_count += 1
1747
if self._write_lock_count == 1:
1748
self._transaction = transactions.WriteTransaction()
1750
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
1751
note('%r was write locked again', self)
1752
self._prev_lock = 'w'
1753
self._unstacked_provider.enable_cache()
1754
for repo in self._fallback_repositories:
1755
# Writes don't affect fallback repos
1757
self._refresh_data()
1758
return RepositoryWriteLockResult(self.unlock, None)
1760
def lock_read(self):
1761
"""Lock the repository for reads.
1763
:return: A bzrlib.lock.LogicalLockResult.
1765
locked = self.is_locked()
1766
if self._write_lock_count:
1767
self._write_lock_count += 1
1769
self.control_files.lock_read()
1771
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
1772
note('%r was read locked again', self)
1773
self._prev_lock = 'r'
1774
self._unstacked_provider.enable_cache()
1775
for repo in self._fallback_repositories:
1777
self._refresh_data()
1778
return LogicalLockResult(self.unlock)
1780
def leave_lock_in_place(self):
1781
# not supported - raise an error
1782
raise NotImplementedError(self.leave_lock_in_place)
1784
def dont_leave_lock_in_place(self):
1785
# not supported - raise an error
1786
raise NotImplementedError(self.dont_leave_lock_in_place)
1789
def pack(self, hint=None, clean_obsolete_packs=False):
1790
"""Compress the data within the repository.
1792
This will pack all the data to a single pack. In future it may
1793
recompress deltas or do other such expensive operations.
1795
self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
1798
def reconcile(self, other=None, thorough=False):
1799
"""Reconcile this repository."""
1800
from bzrlib.reconcile import PackReconciler
1801
reconciler = PackReconciler(self, thorough=thorough)
1802
reconciler.reconcile()
1805
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1806
raise NotImplementedError(self._reconcile_pack)
1808
@only_raises(errors.LockNotHeld, errors.LockBroken)
1810
if self._write_lock_count == 1 and self._write_group is not None:
1811
self.abort_write_group()
1812
self._unstacked_provider.disable_cache()
1813
self._transaction = None
1814
self._write_lock_count = 0
1815
raise errors.BzrError(
1816
'Must end write group before releasing write lock on %s'
1818
if self._write_lock_count:
1819
self._write_lock_count -= 1
1820
if not self._write_lock_count:
1821
transaction = self._transaction
1822
self._transaction = None
1823
transaction.finish()
1825
self.control_files.unlock()
1827
if not self.is_locked():
1828
self._unstacked_provider.disable_cache()
1829
for repo in self._fallback_repositories:
1833
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
1834
"""Format logic for pack structured repositories.
1836
This repository format has:
1837
- a list of packs in pack-names
1838
- packs in packs/NAME.pack
1839
- indices in indices/NAME.{iix,six,tix,rix}
1840
- knit deltas in the packs, knit indices mapped to the indices.
1841
- thunk objects to support the knits programming API.
1842
- a format marker of its own
1843
- an optional 'shared-storage' flag
1844
- an optional 'no-working-trees' flag
1848
# Set this attribute in derived classes to control the repository class
1849
# created by open and initialize.
1850
repository_class = None
1851
# Set this attribute in derived classes to control the
1852
# _commit_builder_class that the repository objects will have passed to
1853
# their constructor.
1854
_commit_builder_class = None
1855
# Set this attribute in derived clases to control the _serializer that the
1856
# repository objects will have passed to their constructor.
1858
# Packs are not confused by ghosts.
1859
supports_ghosts = True
1860
# External references are not supported in pack repositories yet.
1861
supports_external_lookups = False
1862
# Most pack formats do not use chk lookups.
1863
supports_chks = False
1864
# What index classes to use
1865
index_builder_class = None
1867
_fetch_uses_deltas = True
1869
supports_funky_characters = True
1870
revision_graph_can_have_wrong_parents = True
1872
def initialize(self, a_bzrdir, shared=False):
1873
"""Create a pack based repository.
1875
:param a_bzrdir: bzrdir to contain the new repository; must already
1877
:param shared: If true the repository will be initialized as a shared
1880
mutter('creating repository in %s.', a_bzrdir.transport.base)
1881
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1882
builder = self.index_builder_class()
1883
files = [('pack-names', builder.finish())]
1884
utf8_files = [('format', self.get_format_string())]
1886
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1887
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
1888
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
1891
def open(self, a_bzrdir, _found=False, _override_transport=None):
1892
"""See RepositoryFormat.open().
1894
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1895
repository at a slightly different url
1896
than normal. I.e. during 'upgrade'.
1899
format = RepositoryFormat.find_format(a_bzrdir)
1900
if _override_transport is not None:
1901
repo_transport = _override_transport
1903
repo_transport = a_bzrdir.get_repository_transport(None)
1904
control_files = lockable_files.LockableFiles(repo_transport,
1905
'lock', lockdir.LockDir)
1906
return self.repository_class(_format=self,
1908
control_files=control_files,
1909
_commit_builder_class=self._commit_builder_class,
1910
_serializer=self._serializer)
1913
class RetryPackOperations(errors.RetryWithNewPacks):
1914
"""Raised when we are packing and we find a missing file.
1916
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1917
code it should try again.
1920
internal_error = True
1922
_fmt = ("Pack files have changed, reload and try pack again."
1923
" context: %(context)s %(orig_error)s")
1926
class _DirectPackAccess(object):
1927
"""Access to data in one or more packs with less translation."""
1929
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1930
"""Create a _DirectPackAccess object.
1932
:param index_to_packs: A dict mapping index objects to the transport
1933
and file names for obtaining data.
1934
:param reload_func: A function to call if we determine that the pack
1935
files have moved and we need to reload our caches. See
1936
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1938
self._container_writer = None
1939
self._write_index = None
1940
self._indices = index_to_packs
1941
self._reload_func = reload_func
1942
self._flush_func = flush_func
1944
def add_raw_records(self, key_sizes, raw_data):
1945
"""Add raw knit bytes to a storage area.
1947
The data is spooled to the container writer in one bytes-record per
1950
:param sizes: An iterable of tuples containing the key and size of each
1952
:param raw_data: A bytestring containing the data.
1953
:return: A list of memos to retrieve the record later. Each memo is an
1954
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1955
length), where the index field is the write_index object supplied
1956
to the PackAccess object.
1958
if type(raw_data) is not str:
1959
raise AssertionError(
1960
'data must be plain bytes was %s' % type(raw_data))
1963
for key, size in key_sizes:
1964
p_offset, p_length = self._container_writer.add_bytes_record(
1965
raw_data[offset:offset+size], [])
1967
result.append((self._write_index, p_offset, p_length))
1971
"""Flush pending writes on this access object.
1973
This will flush any buffered writes to a NewPack.
1975
if self._flush_func is not None:
1978
def get_raw_records(self, memos_for_retrieval):
1979
"""Get the raw bytes for a records.
1981
:param memos_for_retrieval: An iterable containing the (index, pos,
1982
length) memo for retrieving the bytes. The Pack access method
1983
looks up the pack to use for a given record in its index_to_pack
1985
:return: An iterator over the bytes of the records.
1987
# first pass, group into same-index requests
1989
current_index = None
1990
for (index, offset, length) in memos_for_retrieval:
1991
if current_index == index:
1992
current_list.append((offset, length))
1994
if current_index is not None:
1995
request_lists.append((current_index, current_list))
1996
current_index = index
1997
current_list = [(offset, length)]
1998
# handle the last entry
1999
if current_index is not None:
2000
request_lists.append((current_index, current_list))
2001
for index, offsets in request_lists:
2003
transport, path = self._indices[index]
2005
# A KeyError here indicates that someone has triggered an index
2006
# reload, and this index has gone missing, we need to start
2008
if self._reload_func is None:
2009
# If we don't have a _reload_func there is nothing that can
2012
raise errors.RetryWithNewPacks(index,
2013
reload_occurred=True,
2014
exc_info=sys.exc_info())
2016
reader = pack.make_readv_reader(transport, path, offsets)
2017
for names, read_func in reader.iter_records():
2018
yield read_func(None)
2019
except errors.NoSuchFile:
2020
# A NoSuchFile error indicates that a pack file has gone
2021
# missing on disk, we need to trigger a reload, and start over.
2022
if self._reload_func is None:
2024
raise errors.RetryWithNewPacks(transport.abspath(path),
2025
reload_occurred=False,
2026
exc_info=sys.exc_info())
2028
def set_writer(self, writer, index, transport_packname):
2029
"""Set a writer to use for adding data."""
2030
if index is not None:
2031
self._indices[index] = transport_packname
2032
self._container_writer = writer
2033
self._write_index = index
2035
def reload_or_raise(self, retry_exc):
2036
"""Try calling the reload function, or re-raise the original exception.
2038
This should be called after _DirectPackAccess raises a
2039
RetryWithNewPacks exception. This function will handle the common logic
2040
of determining when the error is fatal versus being temporary.
2041
It will also make sure that the original exception is raised, rather
2042
than the RetryWithNewPacks exception.
2044
If this function returns, then the calling function should retry
2045
whatever operation was being performed. Otherwise an exception will
2048
:param retry_exc: A RetryWithNewPacks exception.
2051
if self._reload_func is None:
2053
elif not self._reload_func():
2054
# The reload claimed that nothing changed
2055
if not retry_exc.reload_occurred:
2056
# If there wasn't an earlier reload, then we really were
2057
# expecting to find changes. We didn't find them, so this is a
2061
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2062
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
