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 ..lazy_import import lazy_import
21
lazy_import(globals(), """
33
from breezy.bzr import (
36
from breezy.bzr.index import (
49
from ..decorators import (
52
from ..lock import LogicalLockResult
53
from ..repository import (
55
RepositoryWriteLockResult,
57
from ..bzr.repository import (
59
RepositoryFormatMetaDir,
61
from ..bzr.vf_repository import (
62
MetaDirVersionedFileRepository,
63
MetaDirVersionedFileRepositoryFormat,
64
VersionedFileCommitBuilder,
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 {key[1] for key in self._file_graph.heads(keys)}
95
"""An in memory proxy for a pack and its indices.
97
This is a base class that is not directly used, instead the classes
98
ExistingPack and NewPack are used.
101
# A map of index 'type' to the file extension and position in the
103
index_definitions = {
105
'revision': ('.rix', 0),
106
'inventory': ('.iix', 1),
108
'signature': ('.six', 3),
111
def __init__(self, revision_index, inventory_index, text_index,
112
signature_index, chk_index=None):
113
"""Create a pack instance.
115
:param revision_index: A GraphIndex for determining what revisions are
116
present in the Pack and accessing the locations of their texts.
117
:param inventory_index: A GraphIndex for determining what inventories are
118
present in the Pack and accessing the locations of their
120
:param text_index: A GraphIndex for determining what file texts
121
are present in the pack and accessing the locations of their
122
texts/deltas (via (fileid, revisionid) tuples).
123
:param signature_index: A GraphIndex for determining what signatures are
124
present in the Pack and accessing the locations of their texts.
125
:param chk_index: A GraphIndex for accessing content by CHK, if the
128
self.revision_index = revision_index
129
self.inventory_index = inventory_index
130
self.text_index = text_index
131
self.signature_index = signature_index
132
self.chk_index = chk_index
134
def access_tuple(self):
135
"""Return a tuple (transport, name) for the pack content."""
136
return self.pack_transport, self.file_name()
138
def _check_references(self):
139
"""Make sure our external references are present.
141
Packs are allowed to have deltas whose base is not in the pack, but it
142
must be present somewhere in this collection. It is not allowed to
143
have deltas based on a fallback repository.
144
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
147
for (index_name, external_refs, index) in [
149
self._get_external_refs(self.text_index),
150
self._pack_collection.text_index.combined_index),
152
self._get_external_refs(self.inventory_index),
153
self._pack_collection.inventory_index.combined_index),
155
missing = external_refs.difference(
156
k for (idx, k, v, r) in
157
index.iter_entries(external_refs))
159
missing_items[index_name] = sorted(list(missing))
161
from pprint import pformat
162
raise errors.BzrCheckError(
163
"Newly created pack file %r has delta references to "
164
"items not in its repository:\n%s"
165
% (self, pformat(missing_items)))
168
"""Get the file name for the pack on disk."""
169
return self.name + '.pack'
171
def get_revision_count(self):
172
return self.revision_index.key_count()
174
def index_name(self, index_type, name):
175
"""Get the disk name of an index type for pack name 'name'."""
176
return name + Pack.index_definitions[index_type][0]
178
def index_offset(self, index_type):
179
"""Get the position in a index_size array for a given index type."""
180
return Pack.index_definitions[index_type][1]
182
def inventory_index_name(self, name):
183
"""The inv index is the name + .iix."""
184
return self.index_name('inventory', name)
186
def revision_index_name(self, name):
187
"""The revision index is the name + .rix."""
188
return self.index_name('revision', name)
190
def signature_index_name(self, name):
191
"""The signature index is the name + .six."""
192
return self.index_name('signature', name)
194
def text_index_name(self, name):
195
"""The text index is the name + .tix."""
196
return self.index_name('text', name)
198
def _replace_index_with_readonly(self, index_type):
199
unlimited_cache = False
200
if index_type == 'chk':
201
unlimited_cache = True
202
index = self.index_class(self.index_transport,
203
self.index_name(index_type, self.name),
204
self.index_sizes[self.index_offset(
206
unlimited_cache=unlimited_cache)
207
if index_type == 'chk':
208
index._leaf_factory = btree_index._gcchk_factory
209
setattr(self, index_type + '_index', index)
211
def __lt__(self, other):
212
if not isinstance(other, Pack):
213
raise TypeError(other)
214
return (id(self) < id(other))
217
return hash((type(self), self.revision_index, self.inventory_index,
218
self.text_index, self.signature_index, self.chk_index))
221
class ExistingPack(Pack):
222
"""An in memory proxy for an existing .pack and its disk indices."""
224
def __init__(self, pack_transport, name, revision_index, inventory_index,
225
text_index, signature_index, chk_index=None):
226
"""Create an ExistingPack object.
228
:param pack_transport: The transport where the pack file resides.
229
:param name: The name of the pack on disk in the pack_transport.
231
Pack.__init__(self, revision_index, inventory_index, text_index,
232
signature_index, chk_index)
234
self.pack_transport = pack_transport
235
if None in (revision_index, inventory_index, text_index,
236
signature_index, name, pack_transport):
237
raise AssertionError()
239
def __eq__(self, other):
240
return self.__dict__ == other.__dict__
242
def __ne__(self, other):
243
return not self.__eq__(other)
246
return "<%s.%s object at 0x%x, %s, %s" % (
247
self.__class__.__module__, self.__class__.__name__, id(self),
248
self.pack_transport, self.name)
251
return hash((type(self), self.name))
254
class ResumedPack(ExistingPack):
256
def __init__(self, name, revision_index, inventory_index, text_index,
257
signature_index, upload_transport, pack_transport, index_transport,
258
pack_collection, chk_index=None):
259
"""Create a ResumedPack object."""
260
ExistingPack.__init__(self, pack_transport, name, revision_index,
261
inventory_index, text_index, signature_index,
263
self.upload_transport = upload_transport
264
self.index_transport = index_transport
265
self.index_sizes = [None, None, None, None]
267
('revision', revision_index),
268
('inventory', inventory_index),
269
('text', text_index),
270
('signature', signature_index),
272
if chk_index is not None:
273
indices.append(('chk', chk_index))
274
self.index_sizes.append(None)
275
for index_type, index in indices:
276
offset = self.index_offset(index_type)
277
self.index_sizes[offset] = index._size
278
self.index_class = pack_collection._index_class
279
self._pack_collection = pack_collection
280
self._state = 'resumed'
281
# XXX: perhaps check that the .pack file exists?
283
def access_tuple(self):
284
if self._state == 'finished':
285
return Pack.access_tuple(self)
286
elif self._state == 'resumed':
287
return self.upload_transport, self.file_name()
289
raise AssertionError(self._state)
292
self.upload_transport.delete(self.file_name())
293
indices = [self.revision_index, self.inventory_index, self.text_index,
294
self.signature_index]
295
if self.chk_index is not None:
296
indices.append(self.chk_index)
297
for index in indices:
298
index._transport.delete(index._name)
301
self._check_references()
302
index_types = ['revision', 'inventory', 'text', 'signature']
303
if self.chk_index is not None:
304
index_types.append('chk')
305
for index_type in index_types:
306
old_name = self.index_name(index_type, self.name)
307
new_name = '../indices/' + old_name
308
self.upload_transport.move(old_name, new_name)
309
self._replace_index_with_readonly(index_type)
310
new_name = '../packs/' + self.file_name()
311
self.upload_transport.move(self.file_name(), new_name)
312
self._state = 'finished'
314
def _get_external_refs(self, index):
315
"""Return compression parents for this index that are not present.
317
This returns any compression parents that are referenced by this index,
318
which are not contained *in* this index. They may be present elsewhere.
320
return index.external_references(1)
324
"""An in memory proxy for a pack which is being created."""
326
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
327
"""Create a NewPack instance.
329
:param pack_collection: A PackCollection into which this is being inserted.
330
:param upload_suffix: An optional suffix to be given to any temporary
331
files created during the pack creation. e.g '.autopack'
332
:param file_mode: Unix permissions for newly created file.
334
# The relative locations of the packs are constrained, but all are
335
# passed in because the caller has them, so as to avoid object churn.
336
index_builder_class = pack_collection._index_builder_class
337
if pack_collection.chk_index is not None:
338
chk_index = index_builder_class(reference_lists=0)
342
# Revisions: parents list, no text compression.
343
index_builder_class(reference_lists=1),
344
# Inventory: We want to map compression only, but currently the
345
# knit code hasn't been updated enough to understand that, so we
346
# have a regular 2-list index giving parents and compression
348
index_builder_class(reference_lists=2),
349
# Texts: compression and per file graph, for all fileids - so two
350
# reference lists and two elements in the key tuple.
351
index_builder_class(reference_lists=2, key_elements=2),
352
# Signatures: Just blobs to store, no compression, no parents
354
index_builder_class(reference_lists=0),
355
# CHK based storage - just blobs, no compression or parents.
358
self._pack_collection = pack_collection
359
# When we make readonly indices, we need this.
360
self.index_class = pack_collection._index_class
361
# where should the new pack be opened
362
self.upload_transport = pack_collection._upload_transport
363
# where are indices written out to
364
self.index_transport = pack_collection._index_transport
365
# where is the pack renamed to when it is finished?
366
self.pack_transport = pack_collection._pack_transport
367
# What file mode to upload the pack and indices with.
368
self._file_mode = file_mode
369
# tracks the content written to the .pack file.
370
self._hash = osutils.md5()
371
# a tuple with the length in bytes of the indices, once the pack
372
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
373
self.index_sizes = None
374
# How much data to cache when writing packs. Note that this is not
375
# synchronised with reads, because it's not in the transport layer, so
376
# is not safe unless the client knows it won't be reading from the pack
378
self._cache_limit = 0
379
# the temporary pack file name.
380
self.random_name = osutils.rand_chars(20) + upload_suffix
381
# when was this pack started ?
382
self.start_time = time.time()
383
# open an output stream for the data added to the pack.
384
self.write_stream = self.upload_transport.open_write_stream(
385
self.random_name, mode=self._file_mode)
386
if 'pack' in debug.debug_flags:
387
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
388
time.ctime(), self.upload_transport.base, self.random_name,
389
time.time() - self.start_time)
390
# A list of byte sequences to be written to the new pack, and the
391
# aggregate size of them. Stored as a list rather than separate
392
# variables so that the _write_data closure below can update them.
393
self._buffer = [[], 0]
394
# create a callable for adding data
396
# robertc says- this is a closure rather than a method on the object
397
# so that the variables are locals, and faster than accessing object
400
def _write_data(bytes, flush=False, _buffer=self._buffer,
401
_write=self.write_stream.write, _update=self._hash.update):
402
_buffer[0].append(bytes)
403
_buffer[1] += len(bytes)
405
if _buffer[1] > self._cache_limit or flush:
406
bytes = b''.join(_buffer[0])
410
# expose this on self, for the occasion when clients want to add data.
411
self._write_data = _write_data
412
# a pack writer object to serialise pack records.
413
self._writer = pack.ContainerWriter(self._write_data)
415
# what state is the pack in? (open, finished, aborted)
417
# no name until we finish writing the content
421
"""Cancel creating this pack."""
422
self._state = 'aborted'
423
self.write_stream.close()
424
# Remove the temporary pack file.
425
self.upload_transport.delete(self.random_name)
426
# The indices have no state on disk.
428
def access_tuple(self):
429
"""Return a tuple (transport, name) for the pack content."""
430
if self._state == 'finished':
431
return Pack.access_tuple(self)
432
elif self._state == 'open':
433
return self.upload_transport, self.random_name
435
raise AssertionError(self._state)
437
def data_inserted(self):
438
"""True if data has been added to this pack."""
439
return bool(self.get_revision_count() or
440
self.inventory_index.key_count() or
441
self.text_index.key_count() or
442
self.signature_index.key_count() or
443
(self.chk_index is not None and self.chk_index.key_count()))
445
def finish_content(self):
446
if self.name is not None:
450
self._write_data(b'', flush=True)
451
self.name = self._hash.hexdigest()
453
def finish(self, suspend=False):
454
"""Finish the new pack.
457
- finalises the content
458
- assigns a name (the md5 of the content, currently)
459
- writes out the associated indices
460
- renames the pack into place.
461
- stores the index size tuple for the pack in the index_sizes
464
self.finish_content()
466
self._check_references()
468
# XXX: It'd be better to write them all to temporary names, then
469
# rename them all into place, so that the window when only some are
470
# visible is smaller. On the other hand none will be seen until
471
# they're in the names list.
472
self.index_sizes = [None, None, None, None]
473
self._write_index('revision', self.revision_index, 'revision',
475
self._write_index('inventory', self.inventory_index, 'inventory',
477
self._write_index('text', self.text_index, 'file texts', suspend)
478
self._write_index('signature', self.signature_index,
479
'revision signatures', suspend)
480
if self.chk_index is not None:
481
self.index_sizes.append(None)
482
self._write_index('chk', self.chk_index,
483
'content hash bytes', suspend)
484
self.write_stream.close(
485
want_fdatasync=self._pack_collection.config_stack.get('repository.fdatasync'))
486
# Note that this will clobber an existing pack with the same name,
487
# without checking for hash collisions. While this is undesirable this
488
# is something that can be rectified in a subsequent release. One way
489
# to rectify it may be to leave the pack at the original name, writing
490
# its pack-names entry as something like 'HASH: index-sizes
491
# temporary-name'. Allocate that and check for collisions, if it is
492
# collision free then rename it into place. If clients know this scheme
493
# they can handle missing-file errors by:
494
# - try for HASH.pack
495
# - try for temporary-name
496
# - refresh the pack-list to see if the pack is now absent
497
new_name = self.name + '.pack'
499
new_name = '../packs/' + new_name
500
self.upload_transport.move(self.random_name, new_name)
501
self._state = 'finished'
502
if 'pack' in debug.debug_flags:
503
# XXX: size might be interesting?
504
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
505
time.ctime(), self.upload_transport.base, self.random_name,
506
new_name, time.time() - self.start_time)
509
"""Flush any current data."""
511
bytes = b''.join(self._buffer[0])
512
self.write_stream.write(bytes)
513
self._hash.update(bytes)
514
self._buffer[:] = [[], 0]
516
def _get_external_refs(self, index):
517
return index._external_references()
519
def set_write_cache_size(self, size):
520
self._cache_limit = size
522
def _write_index(self, index_type, index, label, suspend=False):
523
"""Write out an index.
525
:param index_type: The type of index to write - e.g. 'revision'.
526
:param index: The index object to serialise.
527
:param label: What label to give the index e.g. 'revision'.
529
index_name = self.index_name(index_type, self.name)
531
transport = self.upload_transport
533
transport = self.index_transport
534
index_tempfile = index.finish()
535
index_bytes = index_tempfile.read()
536
write_stream = transport.open_write_stream(index_name,
537
mode=self._file_mode)
538
write_stream.write(index_bytes)
540
want_fdatasync=self._pack_collection.config_stack.get('repository.fdatasync'))
541
self.index_sizes[self.index_offset(index_type)] = len(index_bytes)
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.controldir._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 = []
825
self.config_stack = config.LocationStack(self.transport.base)
828
return '%s(%r)' % (self.__class__.__name__, self.repo)
830
def add_pack_to_memory(self, pack):
831
"""Make a Pack object available to the repository to satisfy queries.
833
:param pack: A Pack object.
835
if pack.name in self._packs_by_name:
836
raise AssertionError(
837
'pack %s already in _packs_by_name' % (pack.name,))
838
self.packs.append(pack)
839
self._packs_by_name[pack.name] = pack
840
self.revision_index.add_index(pack.revision_index, pack)
841
self.inventory_index.add_index(pack.inventory_index, pack)
842
self.text_index.add_index(pack.text_index, pack)
843
self.signature_index.add_index(pack.signature_index, pack)
844
if self.chk_index is not None:
845
self.chk_index.add_index(pack.chk_index, pack)
848
"""Return a list of all the Pack objects this repository has.
850
Note that an in-progress pack being created is not returned.
852
:return: A list of Pack objects for all the packs in the repository.
855
for name in self.names():
856
result.append(self.get_pack_by_name(name))
860
"""Pack the pack collection incrementally.
862
This will not attempt global reorganisation or recompression,
863
rather it will just ensure that the total number of packs does
864
not grow without bound. It uses the _max_pack_count method to
865
determine if autopacking is needed, and the pack_distribution
866
method to determine the number of revisions in each pack.
868
If autopacking takes place then the packs name collection will have
869
been flushed to disk - packing requires updating the name collection
870
in synchronisation with certain steps. Otherwise the names collection
873
:return: Something evaluating true if packing took place.
877
return self._do_autopack()
878
except errors.RetryAutopack:
879
# If we get a RetryAutopack exception, we should abort the
880
# current action, and retry.
883
def _do_autopack(self):
884
# XXX: Should not be needed when the management of indices is sane.
885
total_revisions = self.revision_index.combined_index.key_count()
886
total_packs = len(self._names)
887
if self._max_pack_count(total_revisions) >= total_packs:
889
# determine which packs need changing
890
pack_distribution = self.pack_distribution(total_revisions)
892
for pack in self.all_packs():
893
revision_count = pack.get_revision_count()
894
if revision_count == 0:
895
# revision less packs are not generated by normal operation,
896
# only by operations like sign-my-commits, and thus will not
897
# tend to grow rapdily or without bound like commit containing
898
# packs do - leave them alone as packing them really should
899
# group their data with the relevant commit, and that may
900
# involve rewriting ancient history - which autopack tries to
901
# avoid. Alternatively we could not group the data but treat
902
# each of these as having a single revision, and thus add
903
# one revision for each to the total revision count, to get
904
# a matching distribution.
906
existing_packs.append((revision_count, pack))
907
pack_operations = self.plan_autopack_combinations(
908
existing_packs, pack_distribution)
909
num_new_packs = len(pack_operations)
910
num_old_packs = sum([len(po[1]) for po in pack_operations])
911
num_revs_affected = sum([po[0] for po in pack_operations])
912
mutter('Auto-packing repository %s, which has %d pack files, '
913
'containing %d revisions. Packing %d files into %d affecting %d'
915
self), total_packs, total_revisions, num_old_packs,
916
num_new_packs, num_revs_affected)
917
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
918
reload_func=self._restart_autopack)
919
mutter('Auto-packing repository %s completed', str(self))
922
def _execute_pack_operations(self, pack_operations, packer_class,
924
"""Execute a series of pack operations.
926
:param pack_operations: A list of [revision_count, packs_to_combine].
927
:param packer_class: The class of packer to use
928
:return: The new pack names.
930
for revision_count, packs in pack_operations:
931
# we may have no-ops from the setup logic
934
packer = packer_class(self, packs, '.autopack',
935
reload_func=reload_func)
937
result = packer.pack()
938
except errors.RetryWithNewPacks:
939
# An exception is propagating out of this context, make sure
940
# this packer has cleaned up. Packer() doesn't set its new_pack
941
# state into the RepositoryPackCollection object, so we only
942
# have access to it directly here.
943
if packer.new_pack is not None:
944
packer.new_pack.abort()
949
self._remove_pack_from_memory(pack)
950
# record the newly available packs and stop advertising the old
953
for _, packs in pack_operations:
954
to_be_obsoleted.extend(packs)
955
result = self._save_pack_names(clear_obsolete_packs=True,
956
obsolete_packs=to_be_obsoleted)
959
def _flush_new_pack(self):
960
if self._new_pack is not None:
961
self._new_pack.flush()
963
def lock_names(self):
964
"""Acquire the mutex around the pack-names index.
966
This cannot be used in the middle of a read-only transaction on the
969
self.repo.control_files.lock_write()
971
def _already_packed(self):
972
"""Is the collection already packed?"""
973
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
975
def pack(self, hint=None, clean_obsolete_packs=False):
976
"""Pack the pack collection totally."""
978
total_packs = len(self._names)
979
if self._already_packed():
981
total_revisions = self.revision_index.combined_index.key_count()
982
# XXX: the following may want to be a class, to pack with a given
984
mutter('Packing repository %s, which has %d pack files, '
985
'containing %d revisions with hint %r.', str(self), total_packs,
986
total_revisions, hint)
989
self._try_pack_operations(hint)
990
except RetryPackOperations:
994
if clean_obsolete_packs:
995
self._clear_obsolete_packs()
997
def _try_pack_operations(self, hint):
998
"""Calculate the pack operations based on the hint (if any), and
1001
# determine which packs need changing
1002
pack_operations = [[0, []]]
1003
for pack in self.all_packs():
1004
if hint is None or pack.name in hint:
1005
# Either no hint was provided (so we are packing everything),
1006
# or this pack was included in the hint.
1007
pack_operations[-1][0] += pack.get_revision_count()
1008
pack_operations[-1][1].append(pack)
1009
self._execute_pack_operations(pack_operations,
1010
packer_class=self.optimising_packer_class,
1011
reload_func=self._restart_pack_operations)
1013
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1014
"""Plan a pack operation.
1016
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1018
:param pack_distribution: A list with the number of revisions desired
1021
if len(existing_packs) <= len(pack_distribution):
1023
existing_packs.sort(reverse=True)
1024
pack_operations = [[0, []]]
1025
# plan out what packs to keep, and what to reorganise
1026
while len(existing_packs):
1027
# take the largest pack, and if it's less than the head of the
1028
# distribution chart we will include its contents in the new pack
1029
# for that position. If it's larger, we remove its size from the
1030
# distribution chart
1031
next_pack_rev_count, next_pack = existing_packs.pop(0)
1032
if next_pack_rev_count >= pack_distribution[0]:
1033
# this is already packed 'better' than this, so we can
1034
# not waste time packing it.
1035
while next_pack_rev_count > 0:
1036
next_pack_rev_count -= pack_distribution[0]
1037
if next_pack_rev_count >= 0:
1039
del pack_distribution[0]
1041
# didn't use that entire bucket up
1042
pack_distribution[0] = -next_pack_rev_count
1044
# add the revisions we're going to add to the next output pack
1045
pack_operations[-1][0] += next_pack_rev_count
1046
# allocate this pack to the next pack sub operation
1047
pack_operations[-1][1].append(next_pack)
1048
if pack_operations[-1][0] >= pack_distribution[0]:
1049
# this pack is used up, shift left.
1050
del pack_distribution[0]
1051
pack_operations.append([0, []])
1052
# Now that we know which pack files we want to move, shove them all
1053
# into a single pack file.
1055
final_pack_list = []
1056
for num_revs, pack_files in pack_operations:
1057
final_rev_count += num_revs
1058
final_pack_list.extend(pack_files)
1059
if len(final_pack_list) == 1:
1060
raise AssertionError('We somehow generated an autopack with a'
1061
' single pack file being moved.')
1063
return [[final_rev_count, final_pack_list]]
1065
def ensure_loaded(self):
1066
"""Ensure we have read names from disk.
1068
:return: True if the disk names had not been previously read.
1070
# NB: if you see an assertion error here, it's probably access against
1071
# an unlocked repo. Naughty.
1072
if not self.repo.is_locked():
1073
raise errors.ObjectNotLocked(self.repo)
1074
if self._names is None:
1076
self._packs_at_load = set()
1077
for index, key, value in self._iter_disk_pack_index():
1078
name = key[0].decode('ascii')
1079
self._names[name] = self._parse_index_sizes(value)
1080
self._packs_at_load.add((name, value))
1084
# populate all the metadata.
1088
def _parse_index_sizes(self, value):
1089
"""Parse a string of index sizes."""
1090
return tuple(int(digits) for digits in value.split(b' '))
1092
def get_pack_by_name(self, name):
1093
"""Get a Pack object by name.
1095
:param name: The name of the pack - e.g. '123456'
1096
:return: A Pack object.
1099
return self._packs_by_name[name]
1101
rev_index = self._make_index(name, '.rix')
1102
inv_index = self._make_index(name, '.iix')
1103
txt_index = self._make_index(name, '.tix')
1104
sig_index = self._make_index(name, '.six')
1105
if self.chk_index is not None:
1106
chk_index = self._make_index(name, '.cix', is_chk=True)
1109
result = ExistingPack(self._pack_transport, name, rev_index,
1110
inv_index, txt_index, sig_index, chk_index)
1111
self.add_pack_to_memory(result)
1114
def _resume_pack(self, name):
1115
"""Get a suspended Pack object by name.
1117
:param name: The name of the pack - e.g. '123456'
1118
:return: A Pack object.
1120
if not re.match('[a-f0-9]{32}', name):
1121
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1123
raise errors.UnresumableWriteGroup(
1124
self.repo, [name], 'Malformed write group token')
1126
rev_index = self._make_index(name, '.rix', resume=True)
1127
inv_index = self._make_index(name, '.iix', resume=True)
1128
txt_index = self._make_index(name, '.tix', resume=True)
1129
sig_index = self._make_index(name, '.six', resume=True)
1130
if self.chk_index is not None:
1131
chk_index = self._make_index(name, '.cix', resume=True,
1135
result = self.resumed_pack_factory(name, rev_index, inv_index,
1136
txt_index, sig_index, self._upload_transport,
1137
self._pack_transport, self._index_transport, self,
1138
chk_index=chk_index)
1139
except errors.NoSuchFile as e:
1140
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1141
self.add_pack_to_memory(result)
1142
self._resumed_packs.append(result)
1145
def allocate(self, a_new_pack):
1146
"""Allocate name in the list of packs.
1148
:param a_new_pack: A NewPack instance to be added to the collection of
1149
packs for this repository.
1151
self.ensure_loaded()
1152
if a_new_pack.name in self._names:
1153
raise errors.BzrError(
1154
'Pack %r already exists in %s' % (a_new_pack.name, self))
1155
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1156
self.add_pack_to_memory(a_new_pack)
1158
def _iter_disk_pack_index(self):
1159
"""Iterate over the contents of the pack-names index.
1161
This is used when loading the list from disk, and before writing to
1162
detect updates from others during our write operation.
1163
:return: An iterator of the index contents.
1165
return self._index_class(self.transport, 'pack-names', None
1166
).iter_all_entries()
1168
def _make_index(self, name, suffix, resume=False, is_chk=False):
1169
size_offset = self._suffix_offsets[suffix]
1170
index_name = name + suffix
1172
transport = self._upload_transport
1173
index_size = transport.stat(index_name).st_size
1175
transport = self._index_transport
1176
index_size = self._names[name][size_offset]
1177
index = self._index_class(transport, index_name, index_size,
1178
unlimited_cache=is_chk)
1179
if is_chk and self._index_class is btree_index.BTreeGraphIndex:
1180
index._leaf_factory = btree_index._gcchk_factory
1183
def _max_pack_count(self, total_revisions):
1184
"""Return the maximum number of packs to use for total revisions.
1186
:param total_revisions: The total number of revisions in the
1189
if not total_revisions:
1191
digits = str(total_revisions)
1193
for digit in digits:
1194
result += int(digit)
1198
"""Provide an order to the underlying names."""
1199
return sorted(self._names.keys())
1201
def _obsolete_packs(self, packs):
1202
"""Move a number of packs which have been obsoleted out of the way.
1204
Each pack and its associated indices are moved out of the way.
1206
Note: for correctness this function should only be called after a new
1207
pack names index has been written without these pack names, and with
1208
the names of packs that contain the data previously available via these
1211
:param packs: The packs to obsolete.
1212
:param return: None.
1217
pack.pack_transport.move(pack.file_name(),
1218
'../obsolete_packs/' + pack.file_name())
1219
except errors.NoSuchFile:
1220
# perhaps obsolete_packs was removed? Let's create it and
1223
pack.pack_transport.mkdir('../obsolete_packs/')
1224
except errors.FileExists:
1226
pack.pack_transport.move(pack.file_name(),
1227
'../obsolete_packs/' + pack.file_name())
1228
except (errors.PathError, errors.TransportError) as e:
1229
# TODO: Should these be warnings or mutters?
1230
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1232
# TODO: Probably needs to know all possible indices for this pack
1233
# - or maybe list the directory and move all indices matching this
1234
# name whether we recognize it or not?
1235
suffixes = ['.iix', '.six', '.tix', '.rix']
1236
if self.chk_index is not None:
1237
suffixes.append('.cix')
1238
for suffix in suffixes:
1240
self._index_transport.move(pack.name + suffix,
1241
'../obsolete_packs/' + pack.name + suffix)
1242
except (errors.PathError, errors.TransportError) as e:
1243
mutter("couldn't rename obsolete index, skipping it:\n%s"
1246
def pack_distribution(self, total_revisions):
1247
"""Generate a list of the number of revisions to put in each pack.
1249
:param total_revisions: The total number of revisions in the
1252
if total_revisions == 0:
1254
digits = reversed(str(total_revisions))
1256
for exponent, count in enumerate(digits):
1257
size = 10 ** exponent
1258
for pos in range(int(count)):
1260
return list(reversed(result))
1262
def _pack_tuple(self, name):
1263
"""Return a tuple with the transport and file name for a pack name."""
1264
return self._pack_transport, name + '.pack'
1266
def _remove_pack_from_memory(self, pack):
1267
"""Remove pack from the packs accessed by this repository.
1269
Only affects memory state, until self._save_pack_names() is invoked.
1271
self._names.pop(pack.name)
1272
self._packs_by_name.pop(pack.name)
1273
self._remove_pack_indices(pack)
1274
self.packs.remove(pack)
1276
def _remove_pack_indices(self, pack, ignore_missing=False):
1277
"""Remove the indices for pack from the aggregated indices.
1279
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1281
for index_type in Pack.index_definitions:
1282
attr_name = index_type + '_index'
1283
aggregate_index = getattr(self, attr_name)
1284
if aggregate_index is not None:
1285
pack_index = getattr(pack, attr_name)
1287
aggregate_index.remove_index(pack_index)
1294
"""Clear all cached data."""
1295
# cached revision data
1296
self.revision_index.clear()
1297
# cached signature data
1298
self.signature_index.clear()
1299
# cached file text data
1300
self.text_index.clear()
1301
# cached inventory data
1302
self.inventory_index.clear()
1304
if self.chk_index is not None:
1305
self.chk_index.clear()
1306
# remove the open pack
1307
self._new_pack = None
1308
# information about packs.
1311
self._packs_by_name = {}
1312
self._packs_at_load = None
1314
def _unlock_names(self):
1315
"""Release the mutex around the pack-names index."""
1316
self.repo.control_files.unlock()
1318
def _diff_pack_names(self):
1319
"""Read the pack names from disk, and compare it to the one in memory.
1321
:return: (disk_nodes, deleted_nodes, new_nodes)
1322
disk_nodes The final set of nodes that should be referenced
1323
deleted_nodes Nodes which have been removed from when we started
1324
new_nodes Nodes that are newly introduced
1326
# load the disk nodes across
1328
for index, key, value in self._iter_disk_pack_index():
1329
disk_nodes.add((key[0].decode('ascii'), value))
1330
orig_disk_nodes = set(disk_nodes)
1332
# do a two-way diff against our original content
1333
current_nodes = set()
1334
for name, sizes in self._names.items():
1336
(name, b' '.join(b'%d' % size for size in sizes)))
1338
# Packs no longer present in the repository, which were present when we
1339
# locked the repository
1340
deleted_nodes = self._packs_at_load - current_nodes
1341
# Packs which this process is adding
1342
new_nodes = current_nodes - self._packs_at_load
1344
# Update the disk_nodes set to include the ones we are adding, and
1345
# remove the ones which were removed by someone else
1346
disk_nodes.difference_update(deleted_nodes)
1347
disk_nodes.update(new_nodes)
1349
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1351
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1352
"""Given the correct set of pack files, update our saved info.
1354
:return: (removed, added, modified)
1355
removed pack names removed from self._names
1356
added pack names added to self._names
1357
modified pack names that had changed value
1362
## self._packs_at_load = disk_nodes
1363
new_names = dict(disk_nodes)
1364
# drop no longer present nodes
1365
for pack in self.all_packs():
1366
if pack.name not in new_names:
1367
removed.append(pack.name)
1368
self._remove_pack_from_memory(pack)
1369
# add new nodes/refresh existing ones
1370
for name, value in disk_nodes:
1371
sizes = self._parse_index_sizes(value)
1372
if name in self._names:
1374
if sizes != self._names[name]:
1375
# the pack for name has had its indices replaced - rare but
1376
# important to handle. XXX: probably can never happen today
1377
# because the three-way merge code above does not handle it
1378
# - you may end up adding the same key twice to the new
1379
# disk index because the set values are the same, unless
1380
# the only index shows up as deleted by the set difference
1381
# - which it may. Until there is a specific test for this,
1382
# assume it's broken. RBC 20071017.
1383
self._remove_pack_from_memory(self.get_pack_by_name(name))
1384
self._names[name] = sizes
1385
self.get_pack_by_name(name)
1386
modified.append(name)
1389
self._names[name] = sizes
1390
self.get_pack_by_name(name)
1392
return removed, added, modified
1394
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1395
"""Save the list of packs.
1397
This will take out the mutex around the pack names list for the
1398
duration of the method call. If concurrent updates have been made, a
1399
three-way merge between the current list and the current in memory list
1402
:param clear_obsolete_packs: If True, clear out the contents of the
1403
obsolete_packs directory.
1404
:param obsolete_packs: Packs that are obsolete once the new pack-names
1405
file has been written.
1406
:return: A list of the names saved that were not previously on disk.
1408
already_obsolete = []
1411
builder = self._index_builder_class()
1412
(disk_nodes, deleted_nodes, new_nodes,
1413
orig_disk_nodes) = self._diff_pack_names()
1414
# TODO: handle same-name, index-size-changes here -
1415
# e.g. use the value from disk, not ours, *unless* we're the one
1417
for name, value in disk_nodes:
1418
builder.add_node((name.encode('ascii'), ), value)
1419
self.transport.put_file('pack-names', builder.finish(),
1420
mode=self.repo.controldir._get_file_mode())
1421
self._packs_at_load = disk_nodes
1422
if clear_obsolete_packs:
1425
to_preserve = {o.name for o in obsolete_packs}
1426
already_obsolete = self._clear_obsolete_packs(to_preserve)
1428
self._unlock_names()
1429
# synchronise the memory packs list with what we just wrote:
1430
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1432
# TODO: We could add one more condition here. "if o.name not in
1433
# orig_disk_nodes and o != the new_pack we haven't written to
1434
# disk yet. However, the new pack object is not easily
1435
# accessible here (it would have to be passed through the
1436
# autopacking code, etc.)
1437
obsolete_packs = [o for o in obsolete_packs
1438
if o.name not in already_obsolete]
1439
self._obsolete_packs(obsolete_packs)
1440
return [new_node[0] for new_node in new_nodes]
1442
def reload_pack_names(self):
1443
"""Sync our pack listing with what is present in the repository.
1445
This should be called when we find out that something we thought was
1446
present is now missing. This happens when another process re-packs the
1449
:return: True if the in-memory list of packs has been altered at all.
1451
# The ensure_loaded call is to handle the case where the first call
1452
# made involving the collection was to reload_pack_names, where we
1453
# don't have a view of disk contents. It's a bit of a bandaid, and
1454
# causes two reads of pack-names, but it's a rare corner case not
1455
# struck with regular push/pull etc.
1456
first_read = self.ensure_loaded()
1459
# out the new value.
1460
(disk_nodes, deleted_nodes, new_nodes,
1461
orig_disk_nodes) = self._diff_pack_names()
1462
# _packs_at_load is meant to be the explicit list of names in
1463
# 'pack-names' at then start. As such, it should not contain any
1464
# pending names that haven't been written out yet.
1465
self._packs_at_load = orig_disk_nodes
1467
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1468
if removed or added or modified:
1472
def _restart_autopack(self):
1473
"""Reload the pack names list, and restart the autopack code."""
1474
if not self.reload_pack_names():
1475
# Re-raise the original exception, because something went missing
1476
# and a restart didn't find it
1478
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1480
def _restart_pack_operations(self):
1481
"""Reload the pack names list, and restart the autopack code."""
1482
if not self.reload_pack_names():
1483
# Re-raise the original exception, because something went missing
1484
# and a restart didn't find it
1486
raise RetryPackOperations(self.repo, False, sys.exc_info())
1488
def _clear_obsolete_packs(self, preserve=None):
1489
"""Delete everything from the obsolete-packs directory.
1491
:return: A list of pack identifiers (the filename without '.pack') that
1492
were found in obsolete_packs.
1495
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1496
if preserve is None:
1499
obsolete_pack_files = obsolete_pack_transport.list_dir('.')
1500
except errors.NoSuchFile:
1502
for filename in obsolete_pack_files:
1503
name, ext = osutils.splitext(filename)
1506
if name in preserve:
1509
obsolete_pack_transport.delete(filename)
1510
except (errors.PathError, errors.TransportError) as e:
1511
warning("couldn't delete obsolete pack, skipping it:\n%s"
1515
def _start_write_group(self):
1516
# Do not permit preparation for writing if we're not in a 'write lock'.
1517
if not self.repo.is_write_locked():
1518
raise errors.NotWriteLocked(self)
1519
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1520
file_mode=self.repo.controldir._get_file_mode())
1521
# allow writing: queue writes to a new index
1522
self.revision_index.add_writable_index(self._new_pack.revision_index,
1524
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1526
self.text_index.add_writable_index(self._new_pack.text_index,
1528
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
1529
self.signature_index.add_writable_index(self._new_pack.signature_index,
1531
if self.chk_index is not None:
1532
self.chk_index.add_writable_index(self._new_pack.chk_index,
1534
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
1535
self._new_pack.chk_index.set_optimize(
1536
combine_backing_indices=False)
1538
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1539
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1540
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1541
self.repo.texts._index._add_callback = self.text_index.add_callback
1543
def _abort_write_group(self):
1544
# FIXME: just drop the transient index.
1545
# forget what names there are
1546
if self._new_pack is not None:
1547
with contextlib.ExitStack() as stack:
1548
stack.callback(setattr, self, '_new_pack', None)
1549
# If we aborted while in the middle of finishing the write
1550
# group, _remove_pack_indices could fail because the indexes are
1551
# already gone. But they're not there we shouldn't fail in this
1552
# case, so we pass ignore_missing=True.
1553
stack.callback(self._remove_pack_indices, self._new_pack,
1554
ignore_missing=True)
1555
self._new_pack.abort()
1556
for resumed_pack in self._resumed_packs:
1557
with contextlib.ExitStack() as stack:
1558
# See comment in previous finally block.
1559
stack.callback(self._remove_pack_indices, resumed_pack,
1560
ignore_missing=True)
1561
resumed_pack.abort()
1562
del self._resumed_packs[:]
1564
def _remove_resumed_pack_indices(self):
1565
for resumed_pack in self._resumed_packs:
1566
self._remove_pack_indices(resumed_pack)
1567
del self._resumed_packs[:]
1569
def _check_new_inventories(self):
1570
"""Detect missing inventories in this write group.
1572
:returns: list of strs, summarising any problems found. If the list is
1573
empty no problems were found.
1575
# The base implementation does no checks. GCRepositoryPackCollection
1579
def _commit_write_group(self):
1581
for prefix, versioned_file in (
1582
('revisions', self.repo.revisions),
1583
('inventories', self.repo.inventories),
1584
('texts', self.repo.texts),
1585
('signatures', self.repo.signatures),
1587
missing = versioned_file.get_missing_compression_parent_keys()
1588
all_missing.update([(prefix,) + key for key in missing])
1590
raise errors.BzrCheckError(
1591
"Repository %s has missing compression parent(s) %r "
1592
% (self.repo, sorted(all_missing)))
1593
problems = self._check_new_inventories()
1595
problems_summary = '\n'.join(problems)
1596
raise errors.BzrCheckError(
1597
"Cannot add revision(s) to repository: " + problems_summary)
1598
self._remove_pack_indices(self._new_pack)
1599
any_new_content = False
1600
if self._new_pack.data_inserted():
1601
# get all the data to disk and read to use
1602
self._new_pack.finish()
1603
self.allocate(self._new_pack)
1604
self._new_pack = None
1605
any_new_content = True
1607
self._new_pack.abort()
1608
self._new_pack = None
1609
for resumed_pack in self._resumed_packs:
1610
# XXX: this is a pretty ugly way to turn the resumed pack into a
1611
# properly committed pack.
1612
self._names[resumed_pack.name] = None
1613
self._remove_pack_from_memory(resumed_pack)
1614
resumed_pack.finish()
1615
self.allocate(resumed_pack)
1616
any_new_content = True
1617
del self._resumed_packs[:]
1619
result = self.autopack()
1621
# when autopack takes no steps, the names list is still
1623
return self._save_pack_names()
1627
def _suspend_write_group(self):
1628
tokens = [pack.name for pack in self._resumed_packs]
1629
self._remove_pack_indices(self._new_pack)
1630
if self._new_pack.data_inserted():
1631
# get all the data to disk and read to use
1632
self._new_pack.finish(suspend=True)
1633
tokens.append(self._new_pack.name)
1634
self._new_pack = None
1636
self._new_pack.abort()
1637
self._new_pack = None
1638
self._remove_resumed_pack_indices()
1641
def _resume_write_group(self, tokens):
1642
for token in tokens:
1643
self._resume_pack(token)
1646
class PackRepository(MetaDirVersionedFileRepository):
1647
"""Repository with knit objects stored inside pack containers.
1649
The layering for a KnitPackRepository is:
1651
Graph | HPSS | Repository public layer |
1652
===================================================
1653
Tuple based apis below, string based, and key based apis above
1654
---------------------------------------------------
1656
Provides .texts, .revisions etc
1657
This adapts the N-tuple keys to physical knit records which only have a
1658
single string identifier (for historical reasons), which in older formats
1659
was always the revision_id, and in the mapped code for packs is always
1660
the last element of key tuples.
1661
---------------------------------------------------
1663
A separate GraphIndex is used for each of the
1664
texts/inventories/revisions/signatures contained within each individual
1665
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1667
===================================================
1671
# These attributes are inherited from the Repository base class. Setting
1672
# them to None ensures that if the constructor is changed to not initialize
1673
# them, or a subclass fails to call the constructor, that an error will
1674
# occur rather than the system working but generating incorrect data.
1675
_commit_builder_class = None
1678
def __init__(self, _format, a_controldir, control_files, _commit_builder_class,
1680
MetaDirRepository.__init__(self, _format, a_controldir, control_files)
1681
self._commit_builder_class = _commit_builder_class
1682
self._serializer = _serializer
1683
self._reconcile_fixes_text_parents = True
1684
if self._format.supports_external_lookups:
1685
self._unstacked_provider = graph.CachingParentsProvider(
1686
self._make_parents_provider_unstacked())
1688
self._unstacked_provider = graph.CachingParentsProvider(self)
1689
self._unstacked_provider.disable_cache()
1691
def _all_revision_ids(self):
1692
"""See Repository.all_revision_ids()."""
1693
with self.lock_read():
1694
return [key[0] for key in self.revisions.keys()]
1696
def _abort_write_group(self):
1697
self.revisions._index._key_dependencies.clear()
1698
self._pack_collection._abort_write_group()
1700
def _make_parents_provider(self):
1701
if not self._format.supports_external_lookups:
1702
return self._unstacked_provider
1703
return graph.StackedParentsProvider(_LazyListJoin(
1704
[self._unstacked_provider], self._fallback_repositories))
1706
def _refresh_data(self):
1707
if not self.is_locked():
1709
self._pack_collection.reload_pack_names()
1710
self._unstacked_provider.disable_cache()
1711
self._unstacked_provider.enable_cache()
1713
def _start_write_group(self):
1714
self._pack_collection._start_write_group()
1716
def _commit_write_group(self):
1717
hint = self._pack_collection._commit_write_group()
1718
self.revisions._index._key_dependencies.clear()
1719
# The commit may have added keys that were previously cached as
1720
# missing, so reset the cache.
1721
self._unstacked_provider.disable_cache()
1722
self._unstacked_provider.enable_cache()
1725
def suspend_write_group(self):
1726
# XXX check self._write_group is self.get_transaction()?
1727
tokens = self._pack_collection._suspend_write_group()
1728
self.revisions._index._key_dependencies.clear()
1729
self._write_group = None
1732
def _resume_write_group(self, tokens):
1733
self._start_write_group()
1735
self._pack_collection._resume_write_group(tokens)
1736
except errors.UnresumableWriteGroup:
1737
self._abort_write_group()
1739
for pack in self._pack_collection._resumed_packs:
1740
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1742
def get_transaction(self):
1743
if self._write_lock_count:
1744
return self._transaction
1746
return self.control_files.get_transaction()
1748
def is_locked(self):
1749
return self._write_lock_count or self.control_files.is_locked()
1751
def is_write_locked(self):
1752
return self._write_lock_count
1754
def lock_write(self, token=None):
1755
"""Lock the repository for writes.
1757
:return: A breezy.repository.RepositoryWriteLockResult.
1759
locked = self.is_locked()
1760
if not self._write_lock_count and locked:
1761
raise errors.ReadOnlyError(self)
1762
self._write_lock_count += 1
1763
if self._write_lock_count == 1:
1764
self._transaction = transactions.WriteTransaction()
1766
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
1767
note('%r was write locked again', self)
1768
self._prev_lock = 'w'
1769
self._unstacked_provider.enable_cache()
1770
for repo in self._fallback_repositories:
1771
# Writes don't affect fallback repos
1773
self._refresh_data()
1774
return RepositoryWriteLockResult(self.unlock, None)
1776
def lock_read(self):
1777
"""Lock the repository for reads.
1779
:return: A breezy.lock.LogicalLockResult.
1781
locked = self.is_locked()
1782
if self._write_lock_count:
1783
self._write_lock_count += 1
1785
self.control_files.lock_read()
1787
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
1788
note('%r was read locked again', self)
1789
self._prev_lock = 'r'
1790
self._unstacked_provider.enable_cache()
1791
for repo in self._fallback_repositories:
1793
self._refresh_data()
1794
return LogicalLockResult(self.unlock)
1796
def leave_lock_in_place(self):
1797
# not supported - raise an error
1798
raise NotImplementedError(self.leave_lock_in_place)
1800
def dont_leave_lock_in_place(self):
1801
# not supported - raise an error
1802
raise NotImplementedError(self.dont_leave_lock_in_place)
1804
def pack(self, hint=None, clean_obsolete_packs=False):
1805
"""Compress the data within the repository.
1807
This will pack all the data to a single pack. In future it may
1808
recompress deltas or do other such expensive operations.
1810
with self.lock_write():
1811
self._pack_collection.pack(
1812
hint=hint, clean_obsolete_packs=clean_obsolete_packs)
1814
def reconcile(self, other=None, thorough=False):
1815
"""Reconcile this repository."""
1816
from .reconcile import PackReconciler
1817
with self.lock_write():
1818
reconciler = PackReconciler(self, thorough=thorough)
1819
return reconciler.reconcile()
1821
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1822
raise NotImplementedError(self._reconcile_pack)
1824
@only_raises(errors.LockNotHeld, errors.LockBroken)
1826
if self._write_lock_count == 1 and self._write_group is not None:
1827
self.abort_write_group()
1828
self._unstacked_provider.disable_cache()
1829
self._transaction = None
1830
self._write_lock_count = 0
1831
raise errors.BzrError(
1832
'Must end write group before releasing write lock on %s'
1834
if self._write_lock_count:
1835
self._write_lock_count -= 1
1836
if not self._write_lock_count:
1837
transaction = self._transaction
1838
self._transaction = None
1839
transaction.finish()
1841
self.control_files.unlock()
1843
if not self.is_locked():
1844
self._unstacked_provider.disable_cache()
1845
for repo in self._fallback_repositories:
1849
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
1850
"""Format logic for pack structured repositories.
1852
This repository format has:
1853
- a list of packs in pack-names
1854
- packs in packs/NAME.pack
1855
- indices in indices/NAME.{iix,six,tix,rix}
1856
- knit deltas in the packs, knit indices mapped to the indices.
1857
- thunk objects to support the knits programming API.
1858
- a format marker of its own
1859
- an optional 'shared-storage' flag
1860
- an optional 'no-working-trees' flag
1864
# Set this attribute in derived classes to control the repository class
1865
# created by open and initialize.
1866
repository_class = None
1867
# Set this attribute in derived classes to control the
1868
# _commit_builder_class that the repository objects will have passed to
1869
# their constructor.
1870
_commit_builder_class = None
1871
# Set this attribute in derived clases to control the _serializer that the
1872
# repository objects will have passed to their constructor.
1874
# Packs are not confused by ghosts.
1875
supports_ghosts = True
1876
# External references are not supported in pack repositories yet.
1877
supports_external_lookups = False
1878
# Most pack formats do not use chk lookups.
1879
supports_chks = False
1880
# What index classes to use
1881
index_builder_class = None
1883
_fetch_uses_deltas = True
1885
supports_funky_characters = True
1886
revision_graph_can_have_wrong_parents = True
1888
def initialize(self, a_controldir, shared=False):
1889
"""Create a pack based repository.
1891
:param a_controldir: bzrdir to contain the new repository; must already
1893
:param shared: If true the repository will be initialized as a shared
1896
mutter('creating repository in %s.', a_controldir.transport.base)
1897
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1898
builder = self.index_builder_class()
1899
files = [('pack-names', builder.finish())]
1900
utf8_files = [('format', self.get_format_string())]
1902
self._upload_blank_content(
1903
a_controldir, dirs, files, utf8_files, shared)
1904
repository = self.open(a_controldir=a_controldir, _found=True)
1905
self._run_post_repo_init_hooks(repository, a_controldir, shared)
1908
def open(self, a_controldir, _found=False, _override_transport=None):
1909
"""See RepositoryFormat.open().
1911
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1912
repository at a slightly different url
1913
than normal. I.e. during 'upgrade'.
1916
format = RepositoryFormatMetaDir.find_format(a_controldir)
1917
if _override_transport is not None:
1918
repo_transport = _override_transport
1920
repo_transport = a_controldir.get_repository_transport(None)
1921
control_files = lockable_files.LockableFiles(repo_transport,
1922
'lock', lockdir.LockDir)
1923
return self.repository_class(_format=self,
1924
a_controldir=a_controldir,
1925
control_files=control_files,
1926
_commit_builder_class=self._commit_builder_class,
1927
_serializer=self._serializer)
1930
class RetryPackOperations(errors.RetryWithNewPacks):
1931
"""Raised when we are packing and we find a missing file.
1933
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1934
code it should try again.
1937
internal_error = True
1939
_fmt = ("Pack files have changed, reload and try pack again."
1940
" context: %(context)s %(orig_error)s")
1943
class _DirectPackAccess(object):
1944
"""Access to data in one or more packs with less translation."""
1946
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1947
"""Create a _DirectPackAccess object.
1949
:param index_to_packs: A dict mapping index objects to the transport
1950
and file names for obtaining data.
1951
:param reload_func: A function to call if we determine that the pack
1952
files have moved and we need to reload our caches. See
1953
breezy.repo_fmt.pack_repo.AggregateIndex for more details.
1955
self._container_writer = None
1956
self._write_index = None
1957
self._indices = index_to_packs
1958
self._reload_func = reload_func
1959
self._flush_func = flush_func
1961
def add_raw_record(self, key, size, raw_data):
1962
"""Add raw knit bytes to a storage area.
1964
The data is spooled to the container writer in one bytes-record per
1967
:param key: key of the data segment
1968
:param size: length of the data segment
1969
:param raw_data: A bytestring containing the data.
1970
:return: An opaque index memo For _DirectPackAccess the memo is
1971
(index, pos, length), where the index field is the write_index
1972
object supplied to the PackAccess object.
1974
p_offset, p_length = self._container_writer.add_bytes_record(
1976
return (self._write_index, p_offset, p_length)
1978
def add_raw_records(self, key_sizes, raw_data):
1979
"""Add raw knit bytes to a storage area.
1981
The data is spooled to the container writer in one bytes-record per
1984
:param sizes: An iterable of tuples containing the key and size of each
1986
:param raw_data: A bytestring containing the data.
1987
:return: A list of memos to retrieve the record later. Each memo is an
1988
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1989
length), where the index field is the write_index object supplied
1990
to the PackAccess object.
1992
raw_data = b''.join(raw_data)
1993
if not isinstance(raw_data, bytes):
1994
raise AssertionError(
1995
'data must be plain bytes was %s' % type(raw_data))
1998
for key, size in key_sizes:
2000
self.add_raw_record(key, size, [raw_data[offset:offset + size]]))
2005
"""Flush pending writes on this access object.
2007
This will flush any buffered writes to a NewPack.
2009
if self._flush_func is not None:
2012
def get_raw_records(self, memos_for_retrieval):
2013
"""Get the raw bytes for a records.
2015
:param memos_for_retrieval: An iterable containing the (index, pos,
2016
length) memo for retrieving the bytes. The Pack access method
2017
looks up the pack to use for a given record in its index_to_pack
2019
:return: An iterator over the bytes of the records.
2021
# first pass, group into same-index requests
2023
current_index = None
2024
for (index, offset, length) in memos_for_retrieval:
2025
if current_index == index:
2026
current_list.append((offset, length))
2028
if current_index is not None:
2029
request_lists.append((current_index, current_list))
2030
current_index = index
2031
current_list = [(offset, length)]
2032
# handle the last entry
2033
if current_index is not None:
2034
request_lists.append((current_index, current_list))
2035
for index, offsets in request_lists:
2037
transport, path = self._indices[index]
2039
# A KeyError here indicates that someone has triggered an index
2040
# reload, and this index has gone missing, we need to start
2042
if self._reload_func is None:
2043
# If we don't have a _reload_func there is nothing that can
2046
raise errors.RetryWithNewPacks(index,
2047
reload_occurred=True,
2048
exc_info=sys.exc_info())
2050
reader = pack.make_readv_reader(transport, path, offsets)
2051
for names, read_func in reader.iter_records():
2052
yield read_func(None)
2053
except errors.NoSuchFile:
2054
# A NoSuchFile error indicates that a pack file has gone
2055
# missing on disk, we need to trigger a reload, and start over.
2056
if self._reload_func is None:
2058
raise errors.RetryWithNewPacks(transport.abspath(path),
2059
reload_occurred=False,
2060
exc_info=sys.exc_info())
2062
def set_writer(self, writer, index, transport_packname):
2063
"""Set a writer to use for adding data."""
2064
if index is not None:
2065
self._indices[index] = transport_packname
2066
self._container_writer = writer
2067
self._write_index = index
2069
def reload_or_raise(self, retry_exc):
2070
"""Try calling the reload function, or re-raise the original exception.
2072
This should be called after _DirectPackAccess raises a
2073
RetryWithNewPacks exception. This function will handle the common logic
2074
of determining when the error is fatal versus being temporary.
2075
It will also make sure that the original exception is raised, rather
2076
than the RetryWithNewPacks exception.
2078
If this function returns, then the calling function should retry
2079
whatever operation was being performed. Otherwise an exception will
2082
:param retry_exc: A RetryWithNewPacks exception.
2085
if self._reload_func is None:
2087
elif not self._reload_func():
2088
# The reload claimed that nothing changed
2089
if not retry_exc.reload_occurred:
2090
# If there wasn't an earlier reload, then we really were
2091
# expecting to find changes. We didn't find them, so this is a
2095
# GZ 2017-03-27: No real reason this needs the original traceback.
2096
raise retry_exc.exc_info[1]
added
removed
breezy/bzr/pack_repo.py
