1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
37
from bzrlib.index import (
39
GraphIndexPrefixAdapter,
41
from bzrlib.knit import (
47
from bzrlib import tsort
54
revision as _mod_revision,
57
from bzrlib.decorators import needs_write_lock, only_raises
58
from bzrlib.btree_index import (
62
from bzrlib.index import (
66
from bzrlib.repofmt.knitrepo import KnitRepository
67
from bzrlib.repository import (
69
MetaDirRepositoryFormat,
74
from bzrlib.trace import (
81
class PackCommitBuilder(CommitBuilder):
82
"""A subclass of CommitBuilder to add texts with pack semantics.
84
Specifically this uses one knit object rather than one knit object per
85
added text, reducing memory and object pressure.
88
def __init__(self, repository, parents, config, timestamp=None,
89
timezone=None, committer=None, revprops=None,
91
CommitBuilder.__init__(self, repository, parents, config,
92
timestamp=timestamp, timezone=timezone, committer=committer,
93
revprops=revprops, revision_id=revision_id)
94
self._file_graph = graph.Graph(
95
repository._pack_collection.text_index.combined_index)
97
def _heads(self, file_id, revision_ids):
98
keys = [(file_id, revision_id) for revision_id in revision_ids]
99
return set([key[1] for key in self._file_graph.heads(keys)])
102
class PackRootCommitBuilder(RootCommitBuilder):
103
"""A subclass of RootCommitBuilder to add texts with pack semantics.
105
Specifically this uses one knit object rather than one knit object per
106
added text, reducing memory and object pressure.
109
def __init__(self, repository, parents, config, timestamp=None,
110
timezone=None, committer=None, revprops=None,
112
CommitBuilder.__init__(self, repository, parents, config,
113
timestamp=timestamp, timezone=timezone, committer=committer,
114
revprops=revprops, revision_id=revision_id)
115
self._file_graph = graph.Graph(
116
repository._pack_collection.text_index.combined_index)
118
def _heads(self, file_id, revision_ids):
119
keys = [(file_id, revision_id) for revision_id in revision_ids]
120
return set([key[1] for key in self._file_graph.heads(keys)])
124
"""An in memory proxy for a pack and its indices.
126
This is a base class that is not directly used, instead the classes
127
ExistingPack and NewPack are used.
130
# A map of index 'type' to the file extension and position in the
132
index_definitions = {
134
'revision': ('.rix', 0),
135
'inventory': ('.iix', 1),
137
'signature': ('.six', 3),
140
def __init__(self, revision_index, inventory_index, text_index,
141
signature_index, chk_index=None):
142
"""Create a pack instance.
144
:param revision_index: A GraphIndex for determining what revisions are
145
present in the Pack and accessing the locations of their texts.
146
:param inventory_index: A GraphIndex for determining what inventories are
147
present in the Pack and accessing the locations of their
149
:param text_index: A GraphIndex for determining what file texts
150
are present in the pack and accessing the locations of their
151
texts/deltas (via (fileid, revisionid) tuples).
152
:param signature_index: A GraphIndex for determining what signatures are
153
present in the Pack and accessing the locations of their texts.
154
:param chk_index: A GraphIndex for accessing content by CHK, if the
157
self.revision_index = revision_index
158
self.inventory_index = inventory_index
159
self.text_index = text_index
160
self.signature_index = signature_index
161
self.chk_index = chk_index
163
def access_tuple(self):
164
"""Return a tuple (transport, name) for the pack content."""
165
return self.pack_transport, self.file_name()
167
def _check_references(self):
168
"""Make sure our external references are present.
170
Packs are allowed to have deltas whose base is not in the pack, but it
171
must be present somewhere in this collection. It is not allowed to
172
have deltas based on a fallback repository.
173
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
176
for (index_name, external_refs, index) in [
178
self._get_external_refs(self.text_index),
179
self._pack_collection.text_index.combined_index),
181
self._get_external_refs(self.inventory_index),
182
self._pack_collection.inventory_index.combined_index),
184
missing = external_refs.difference(
185
k for (idx, k, v, r) in
186
index.iter_entries(external_refs))
188
missing_items[index_name] = sorted(list(missing))
190
from pprint import pformat
191
raise errors.BzrCheckError(
192
"Newly created pack file %r has delta references to "
193
"items not in its repository:\n%s"
194
% (self, pformat(missing_items)))
197
"""Get the file name for the pack on disk."""
198
return self.name + '.pack'
200
def get_revision_count(self):
201
return self.revision_index.key_count()
203
def index_name(self, index_type, name):
204
"""Get the disk name of an index type for pack name 'name'."""
205
return name + Pack.index_definitions[index_type][0]
207
def index_offset(self, index_type):
208
"""Get the position in a index_size array for a given index type."""
209
return Pack.index_definitions[index_type][1]
211
def inventory_index_name(self, name):
212
"""The inv index is the name + .iix."""
213
return self.index_name('inventory', name)
215
def revision_index_name(self, name):
216
"""The revision index is the name + .rix."""
217
return self.index_name('revision', name)
219
def signature_index_name(self, name):
220
"""The signature index is the name + .six."""
221
return self.index_name('signature', name)
223
def text_index_name(self, name):
224
"""The text index is the name + .tix."""
225
return self.index_name('text', name)
227
def _replace_index_with_readonly(self, index_type):
228
unlimited_cache = False
229
if index_type == 'chk':
230
unlimited_cache = True
231
setattr(self, index_type + '_index',
232
self.index_class(self.index_transport,
233
self.index_name(index_type, self.name),
234
self.index_sizes[self.index_offset(index_type)],
235
unlimited_cache=unlimited_cache))
238
class ExistingPack(Pack):
239
"""An in memory proxy for an existing .pack and its disk indices."""
241
def __init__(self, pack_transport, name, revision_index, inventory_index,
242
text_index, signature_index, chk_index=None):
243
"""Create an ExistingPack object.
245
:param pack_transport: The transport where the pack file resides.
246
:param name: The name of the pack on disk in the pack_transport.
248
Pack.__init__(self, revision_index, inventory_index, text_index,
249
signature_index, chk_index)
251
self.pack_transport = pack_transport
252
if None in (revision_index, inventory_index, text_index,
253
signature_index, name, pack_transport):
254
raise AssertionError()
256
def __eq__(self, other):
257
return self.__dict__ == other.__dict__
259
def __ne__(self, other):
260
return not self.__eq__(other)
263
return "<%s.%s object at 0x%x, %s, %s" % (
264
self.__class__.__module__, self.__class__.__name__, id(self),
265
self.pack_transport, self.name)
268
class ResumedPack(ExistingPack):
270
def __init__(self, name, revision_index, inventory_index, text_index,
271
signature_index, upload_transport, pack_transport, index_transport,
272
pack_collection, chk_index=None):
273
"""Create a ResumedPack object."""
274
ExistingPack.__init__(self, pack_transport, name, revision_index,
275
inventory_index, text_index, signature_index,
277
self.upload_transport = upload_transport
278
self.index_transport = index_transport
279
self.index_sizes = [None, None, None, None]
281
('revision', revision_index),
282
('inventory', inventory_index),
283
('text', text_index),
284
('signature', signature_index),
286
if chk_index is not None:
287
indices.append(('chk', chk_index))
288
self.index_sizes.append(None)
289
for index_type, index in indices:
290
offset = self.index_offset(index_type)
291
self.index_sizes[offset] = index._size
292
self.index_class = pack_collection._index_class
293
self._pack_collection = pack_collection
294
self._state = 'resumed'
295
# XXX: perhaps check that the .pack file exists?
297
def access_tuple(self):
298
if self._state == 'finished':
299
return Pack.access_tuple(self)
300
elif self._state == 'resumed':
301
return self.upload_transport, self.file_name()
303
raise AssertionError(self._state)
306
self.upload_transport.delete(self.file_name())
307
indices = [self.revision_index, self.inventory_index, self.text_index,
308
self.signature_index]
309
if self.chk_index is not None:
310
indices.append(self.chk_index)
311
for index in indices:
312
index._transport.delete(index._name)
315
self._check_references()
316
index_types = ['revision', 'inventory', 'text', 'signature']
317
if self.chk_index is not None:
318
index_types.append('chk')
319
for index_type in index_types:
320
old_name = self.index_name(index_type, self.name)
321
new_name = '../indices/' + old_name
322
self.upload_transport.rename(old_name, new_name)
323
self._replace_index_with_readonly(index_type)
324
new_name = '../packs/' + self.file_name()
325
self.upload_transport.rename(self.file_name(), new_name)
326
self._state = 'finished'
328
def _get_external_refs(self, index):
329
"""Return compression parents for this index that are not present.
331
This returns any compression parents that are referenced by this index,
332
which are not contained *in* this index. They may be present elsewhere.
334
return index.external_references(1)
338
"""An in memory proxy for a pack which is being created."""
340
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
341
"""Create a NewPack instance.
343
:param pack_collection: A PackCollection into which this is being inserted.
344
:param upload_suffix: An optional suffix to be given to any temporary
345
files created during the pack creation. e.g '.autopack'
346
:param file_mode: Unix permissions for newly created file.
348
# The relative locations of the packs are constrained, but all are
349
# passed in because the caller has them, so as to avoid object churn.
350
index_builder_class = pack_collection._index_builder_class
351
if pack_collection.chk_index is not None:
352
chk_index = index_builder_class(reference_lists=0)
356
# Revisions: parents list, no text compression.
357
index_builder_class(reference_lists=1),
358
# Inventory: We want to map compression only, but currently the
359
# knit code hasn't been updated enough to understand that, so we
360
# have a regular 2-list index giving parents and compression
362
index_builder_class(reference_lists=2),
363
# Texts: compression and per file graph, for all fileids - so two
364
# reference lists and two elements in the key tuple.
365
index_builder_class(reference_lists=2, key_elements=2),
366
# Signatures: Just blobs to store, no compression, no parents
368
index_builder_class(reference_lists=0),
369
# CHK based storage - just blobs, no compression or parents.
372
self._pack_collection = pack_collection
373
# When we make readonly indices, we need this.
374
self.index_class = pack_collection._index_class
375
# where should the new pack be opened
376
self.upload_transport = pack_collection._upload_transport
377
# where are indices written out to
378
self.index_transport = pack_collection._index_transport
379
# where is the pack renamed to when it is finished?
380
self.pack_transport = pack_collection._pack_transport
381
# What file mode to upload the pack and indices with.
382
self._file_mode = file_mode
383
# tracks the content written to the .pack file.
384
self._hash = osutils.md5()
385
# a tuple with the length in bytes of the indices, once the pack
386
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
387
self.index_sizes = None
388
# How much data to cache when writing packs. Note that this is not
389
# synchronised with reads, because it's not in the transport layer, so
390
# is not safe unless the client knows it won't be reading from the pack
392
self._cache_limit = 0
393
# the temporary pack file name.
394
self.random_name = osutils.rand_chars(20) + upload_suffix
395
# when was this pack started ?
396
self.start_time = time.time()
397
# open an output stream for the data added to the pack.
398
self.write_stream = self.upload_transport.open_write_stream(
399
self.random_name, mode=self._file_mode)
400
if 'pack' in debug.debug_flags:
401
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
402
time.ctime(), self.upload_transport.base, self.random_name,
403
time.time() - self.start_time)
404
# A list of byte sequences to be written to the new pack, and the
405
# aggregate size of them. Stored as a list rather than separate
406
# variables so that the _write_data closure below can update them.
407
self._buffer = [[], 0]
408
# create a callable for adding data
410
# robertc says- this is a closure rather than a method on the object
411
# so that the variables are locals, and faster than accessing object
413
def _write_data(bytes, flush=False, _buffer=self._buffer,
414
_write=self.write_stream.write, _update=self._hash.update):
415
_buffer[0].append(bytes)
416
_buffer[1] += len(bytes)
418
if _buffer[1] > self._cache_limit or flush:
419
bytes = ''.join(_buffer[0])
423
# expose this on self, for the occasion when clients want to add data.
424
self._write_data = _write_data
425
# a pack writer object to serialise pack records.
426
self._writer = pack.ContainerWriter(self._write_data)
428
# what state is the pack in? (open, finished, aborted)
430
# no name until we finish writing the content
434
"""Cancel creating this pack."""
435
self._state = 'aborted'
436
self.write_stream.close()
437
# Remove the temporary pack file.
438
self.upload_transport.delete(self.random_name)
439
# The indices have no state on disk.
441
def access_tuple(self):
442
"""Return a tuple (transport, name) for the pack content."""
443
if self._state == 'finished':
444
return Pack.access_tuple(self)
445
elif self._state == 'open':
446
return self.upload_transport, self.random_name
448
raise AssertionError(self._state)
450
def data_inserted(self):
451
"""True if data has been added to this pack."""
452
return bool(self.get_revision_count() or
453
self.inventory_index.key_count() or
454
self.text_index.key_count() or
455
self.signature_index.key_count() or
456
(self.chk_index is not None and self.chk_index.key_count()))
458
def finish_content(self):
459
if self.name is not None:
463
self._write_data('', flush=True)
464
self.name = self._hash.hexdigest()
466
def finish(self, suspend=False):
467
"""Finish the new pack.
470
- finalises the content
471
- assigns a name (the md5 of the content, currently)
472
- writes out the associated indices
473
- renames the pack into place.
474
- stores the index size tuple for the pack in the index_sizes
477
self.finish_content()
479
self._check_references()
481
# XXX: It'd be better to write them all to temporary names, then
482
# rename them all into place, so that the window when only some are
483
# visible is smaller. On the other hand none will be seen until
484
# they're in the names list.
485
self.index_sizes = [None, None, None, None]
486
self._write_index('revision', self.revision_index, 'revision', suspend)
487
self._write_index('inventory', self.inventory_index, 'inventory',
489
self._write_index('text', self.text_index, 'file texts', suspend)
490
self._write_index('signature', self.signature_index,
491
'revision signatures', suspend)
492
if self.chk_index is not None:
493
self.index_sizes.append(None)
494
self._write_index('chk', self.chk_index,
495
'content hash bytes', suspend)
496
self.write_stream.close()
497
# Note that this will clobber an existing pack with the same name,
498
# without checking for hash collisions. While this is undesirable this
499
# is something that can be rectified in a subsequent release. One way
500
# to rectify it may be to leave the pack at the original name, writing
501
# its pack-names entry as something like 'HASH: index-sizes
502
# temporary-name'. Allocate that and check for collisions, if it is
503
# collision free then rename it into place. If clients know this scheme
504
# they can handle missing-file errors by:
505
# - try for HASH.pack
506
# - try for temporary-name
507
# - refresh the pack-list to see if the pack is now absent
508
new_name = self.name + '.pack'
510
new_name = '../packs/' + new_name
511
self.upload_transport.rename(self.random_name, new_name)
512
self._state = 'finished'
513
if 'pack' in debug.debug_flags:
514
# XXX: size might be interesting?
515
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
516
time.ctime(), self.upload_transport.base, self.random_name,
517
new_name, time.time() - self.start_time)
520
"""Flush any current data."""
522
bytes = ''.join(self._buffer[0])
523
self.write_stream.write(bytes)
524
self._hash.update(bytes)
525
self._buffer[:] = [[], 0]
527
def _get_external_refs(self, index):
528
return index._external_references()
530
def set_write_cache_size(self, size):
531
self._cache_limit = size
533
def _write_index(self, index_type, index, label, suspend=False):
534
"""Write out an index.
536
:param index_type: The type of index to write - e.g. 'revision'.
537
:param index: The index object to serialise.
538
:param label: What label to give the index e.g. 'revision'.
540
index_name = self.index_name(index_type, self.name)
542
transport = self.upload_transport
544
transport = self.index_transport
545
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
546
index_name, index.finish(), mode=self._file_mode)
547
if 'pack' in debug.debug_flags:
548
# XXX: size might be interesting?
549
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
550
time.ctime(), label, self.upload_transport.base,
551
self.random_name, time.time() - self.start_time)
552
# Replace the writable index on this object with a readonly,
553
# presently unloaded index. We should alter
554
# the index layer to make its finish() error if add_node is
555
# subsequently used. RBC
556
self._replace_index_with_readonly(index_type)
559
class AggregateIndex(object):
560
"""An aggregated index for the RepositoryPackCollection.
562
AggregateIndex is reponsible for managing the PackAccess object,
563
Index-To-Pack mapping, and all indices list for a specific type of index
564
such as 'revision index'.
566
A CombinedIndex provides an index on a single key space built up
567
from several on-disk indices. The AggregateIndex builds on this
568
to provide a knit access layer, and allows having up to one writable
569
index within the collection.
571
# XXX: Probably 'can be written to' could/should be separated from 'acts
572
# like a knit index' -- mbp 20071024
574
def __init__(self, reload_func=None, flush_func=None):
575
"""Create an AggregateIndex.
577
:param reload_func: A function to call if we find we are missing an
578
index. Should have the form reload_func() => True if the list of
579
active pack files has changed.
581
self._reload_func = reload_func
582
self.index_to_pack = {}
583
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
584
self.data_access = _DirectPackAccess(self.index_to_pack,
585
reload_func=reload_func,
586
flush_func=flush_func)
587
self.add_callback = None
589
def replace_indices(self, index_to_pack, indices):
590
"""Replace the current mappings with fresh ones.
592
This should probably not be used eventually, rather incremental add and
593
removal of indices. It has been added during refactoring of existing
596
:param index_to_pack: A mapping from index objects to
597
(transport, name) tuples for the pack file data.
598
:param indices: A list of indices.
600
# refresh the revision pack map dict without replacing the instance.
601
self.index_to_pack.clear()
602
self.index_to_pack.update(index_to_pack)
603
# XXX: API break - clearly a 'replace' method would be good?
604
self.combined_index._indices[:] = indices
605
# the current add nodes callback for the current writable index if
607
self.add_callback = None
609
def add_index(self, index, pack):
610
"""Add index to the aggregate, which is an index for Pack pack.
612
Future searches on the aggregate index will seach this new index
613
before all previously inserted indices.
615
:param index: An Index for the pack.
616
:param pack: A Pack instance.
618
# expose it to the index map
619
self.index_to_pack[index] = pack.access_tuple()
620
# put it at the front of the linear index list
621
self.combined_index.insert_index(0, index)
623
def add_writable_index(self, index, pack):
624
"""Add an index which is able to have data added to it.
626
There can be at most one writable index at any time. Any
627
modifications made to the knit are put into this index.
629
:param index: An index from the pack parameter.
630
:param pack: A Pack instance.
632
if self.add_callback is not None:
633
raise AssertionError(
634
"%s already has a writable index through %s" % \
635
(self, self.add_callback))
636
# allow writing: queue writes to a new index
637
self.add_index(index, pack)
638
# Updates the index to packs mapping as a side effect,
639
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
640
self.add_callback = index.add_nodes
643
"""Reset all the aggregate data to nothing."""
644
self.data_access.set_writer(None, None, (None, None))
645
self.index_to_pack.clear()
646
del self.combined_index._indices[:]
647
self.add_callback = None
649
def remove_index(self, index, pack):
650
"""Remove index from the indices used to answer queries.
652
:param index: An index from the pack parameter.
653
:param pack: A Pack instance.
655
del self.index_to_pack[index]
656
self.combined_index._indices.remove(index)
657
if (self.add_callback is not None and
658
getattr(index, 'add_nodes', None) == self.add_callback):
659
self.add_callback = None
660
self.data_access.set_writer(None, None, (None, None))
663
class Packer(object):
664
"""Create a pack from packs."""
666
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
670
:param pack_collection: A RepositoryPackCollection object where the
671
new pack is being written to.
672
:param packs: The packs to combine.
673
:param suffix: The suffix to use on the temporary files for the pack.
674
:param revision_ids: Revision ids to limit the pack to.
675
:param reload_func: A function to call if a pack file/index goes
676
missing. The side effect of calling this function should be to
677
update self.packs. See also AggregateIndex
681
self.revision_ids = revision_ids
682
# The pack object we are creating.
684
self._pack_collection = pack_collection
685
self._reload_func = reload_func
686
# The index layer keys for the revisions being copied. None for 'all
688
self._revision_keys = None
689
# What text keys to copy. None for 'all texts'. This is set by
690
# _copy_inventory_texts
691
self._text_filter = None
694
def _extra_init(self):
695
"""A template hook to allow extending the constructor trivially."""
697
def _pack_map_and_index_list(self, index_attribute):
698
"""Convert a list of packs to an index pack map and index list.
700
:param index_attribute: The attribute that the desired index is found
702
:return: A tuple (map, list) where map contains the dict from
703
index:pack_tuple, and list contains the indices in the preferred
708
for pack_obj in self.packs:
709
index = getattr(pack_obj, index_attribute)
710
indices.append(index)
711
pack_map[index] = pack_obj
712
return pack_map, indices
714
def _index_contents(self, indices, key_filter=None):
715
"""Get an iterable of the index contents from a pack_map.
717
:param indices: The list of indices to query
718
:param key_filter: An optional filter to limit the keys returned.
720
all_index = CombinedGraphIndex(indices)
721
if key_filter is None:
722
return all_index.iter_all_entries()
724
return all_index.iter_entries(key_filter)
726
def pack(self, pb=None):
727
"""Create a new pack by reading data from other packs.
729
This does little more than a bulk copy of data. One key difference
730
is that data with the same item key across multiple packs is elided
731
from the output. The new pack is written into the current pack store
732
along with its indices, and the name added to the pack names. The
733
source packs are not altered and are not required to be in the current
736
:param pb: An optional progress bar to use. A nested bar is created if
738
:return: A Pack object, or None if nothing was copied.
740
# open a pack - using the same name as the last temporary file
741
# - which has already been flushed, so its safe.
742
# XXX: - duplicate code warning with start_write_group; fix before
743
# considering 'done'.
744
if self._pack_collection._new_pack is not None:
745
raise errors.BzrError('call to %s.pack() while another pack is'
747
% (self.__class__.__name__,))
748
if self.revision_ids is not None:
749
if len(self.revision_ids) == 0:
750
# silly fetch request.
753
self.revision_ids = frozenset(self.revision_ids)
754
self.revision_keys = frozenset((revid,) for revid in
757
self.pb = ui.ui_factory.nested_progress_bar()
761
return self._create_pack_from_packs()
767
"""Open a pack for the pack we are creating."""
768
new_pack = self._pack_collection.pack_factory(self._pack_collection,
769
upload_suffix=self.suffix,
770
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
771
# We know that we will process all nodes in order, and don't need to
772
# query, so don't combine any indices spilled to disk until we are done
773
new_pack.revision_index.set_optimize(combine_backing_indices=False)
774
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
775
new_pack.text_index.set_optimize(combine_backing_indices=False)
776
new_pack.signature_index.set_optimize(combine_backing_indices=False)
779
def _update_pack_order(self, entries, index_to_pack_map):
780
"""Determine how we want our packs to be ordered.
782
This changes the sort order of the self.packs list so that packs unused
783
by 'entries' will be at the end of the list, so that future requests
784
can avoid probing them. Used packs will be at the front of the
785
self.packs list, in the order of their first use in 'entries'.
787
:param entries: A list of (index, ...) tuples
788
:param index_to_pack_map: A mapping from index objects to pack objects.
792
for entry in entries:
794
if index not in seen_indexes:
795
packs.append(index_to_pack_map[index])
796
seen_indexes.add(index)
797
if len(packs) == len(self.packs):
798
if 'pack' in debug.debug_flags:
799
mutter('Not changing pack list, all packs used.')
801
seen_packs = set(packs)
802
for pack in self.packs:
803
if pack not in seen_packs:
806
if 'pack' in debug.debug_flags:
807
old_names = [p.access_tuple()[1] for p in self.packs]
808
new_names = [p.access_tuple()[1] for p in packs]
809
mutter('Reordering packs\nfrom: %s\n to: %s',
810
old_names, new_names)
813
def _copy_revision_texts(self):
814
"""Copy revision data to the new pack."""
816
if self.revision_ids:
817
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
820
# select revision keys
821
revision_index_map, revision_indices = self._pack_map_and_index_list(
823
revision_nodes = self._index_contents(revision_indices, revision_keys)
824
revision_nodes = list(revision_nodes)
825
self._update_pack_order(revision_nodes, revision_index_map)
826
# copy revision keys and adjust values
827
self.pb.update("Copying revision texts", 1)
828
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
829
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
830
self.new_pack.revision_index, readv_group_iter, total_items))
831
if 'pack' in debug.debug_flags:
832
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
833
time.ctime(), self._pack_collection._upload_transport.base,
834
self.new_pack.random_name,
835
self.new_pack.revision_index.key_count(),
836
time.time() - self.new_pack.start_time)
837
self._revision_keys = revision_keys
839
def _copy_inventory_texts(self):
840
"""Copy the inventory texts to the new pack.
842
self._revision_keys is used to determine what inventories to copy.
844
Sets self._text_filter appropriately.
846
# select inventory keys
847
inv_keys = self._revision_keys # currently the same keyspace, and note that
848
# querying for keys here could introduce a bug where an inventory item
849
# is missed, so do not change it to query separately without cross
850
# checking like the text key check below.
851
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
853
inv_nodes = self._index_contents(inventory_indices, inv_keys)
854
# copy inventory keys and adjust values
855
# XXX: Should be a helper function to allow different inv representation
857
self.pb.update("Copying inventory texts", 2)
858
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
859
# Only grab the output lines if we will be processing them
860
output_lines = bool(self.revision_ids)
861
inv_lines = self._copy_nodes_graph(inventory_index_map,
862
self.new_pack._writer, self.new_pack.inventory_index,
863
readv_group_iter, total_items, output_lines=output_lines)
864
if self.revision_ids:
865
self._process_inventory_lines(inv_lines)
867
# eat the iterator to cause it to execute.
869
self._text_filter = None
870
if 'pack' in debug.debug_flags:
871
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
872
time.ctime(), self._pack_collection._upload_transport.base,
873
self.new_pack.random_name,
874
self.new_pack.inventory_index.key_count(),
875
time.time() - self.new_pack.start_time)
877
def _copy_text_texts(self):
879
text_index_map, text_nodes = self._get_text_nodes()
880
if self._text_filter is not None:
881
# We could return the keys copied as part of the return value from
882
# _copy_nodes_graph but this doesn't work all that well with the
883
# need to get line output too, so we check separately, and as we're
884
# going to buffer everything anyway, we check beforehand, which
885
# saves reading knit data over the wire when we know there are
887
text_nodes = set(text_nodes)
888
present_text_keys = set(_node[1] for _node in text_nodes)
889
missing_text_keys = set(self._text_filter) - present_text_keys
890
if missing_text_keys:
891
# TODO: raise a specific error that can handle many missing
893
mutter("missing keys during fetch: %r", missing_text_keys)
894
a_missing_key = missing_text_keys.pop()
895
raise errors.RevisionNotPresent(a_missing_key[1],
897
# copy text keys and adjust values
898
self.pb.update("Copying content texts", 3)
899
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
900
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
901
self.new_pack.text_index, readv_group_iter, total_items))
902
self._log_copied_texts()
904
def _create_pack_from_packs(self):
905
self.pb.update("Opening pack", 0, 5)
906
self.new_pack = self.open_pack()
907
new_pack = self.new_pack
908
# buffer data - we won't be reading-back during the pack creation and
909
# this makes a significant difference on sftp pushes.
910
new_pack.set_write_cache_size(1024*1024)
911
if 'pack' in debug.debug_flags:
912
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
913
for a_pack in self.packs]
914
if self.revision_ids is not None:
915
rev_count = len(self.revision_ids)
918
mutter('%s: create_pack: creating pack from source packs: '
919
'%s%s %s revisions wanted %s t=0',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
plain_pack_list, rev_count)
922
self._copy_revision_texts()
923
self._copy_inventory_texts()
924
self._copy_text_texts()
925
# select signature keys
926
signature_filter = self._revision_keys # same keyspace
927
signature_index_map, signature_indices = self._pack_map_and_index_list(
929
signature_nodes = self._index_contents(signature_indices,
931
# copy signature keys and adjust values
932
self.pb.update("Copying signature texts", 4)
933
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
934
new_pack.signature_index)
935
if 'pack' in debug.debug_flags:
936
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
937
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
938
new_pack.signature_index.key_count(),
939
time.time() - new_pack.start_time)
941
# NB XXX: how to check CHK references are present? perhaps by yielding
942
# the items? How should that interact with stacked repos?
943
if new_pack.chk_index is not None:
945
if 'pack' in debug.debug_flags:
946
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
947
time.ctime(), self._pack_collection._upload_transport.base,
948
new_pack.random_name,
949
new_pack.chk_index.key_count(),
950
time.time() - new_pack.start_time)
951
new_pack._check_references()
952
if not self._use_pack(new_pack):
955
self.pb.update("Finishing pack", 5)
957
self._pack_collection.allocate(new_pack)
960
def _copy_chks(self, refs=None):
961
# XXX: Todo, recursive follow-pointers facility when fetching some
963
chk_index_map, chk_indices = self._pack_map_and_index_list(
965
chk_nodes = self._index_contents(chk_indices, refs)
967
# TODO: This isn't strictly tasteful as we are accessing some private
968
# variables (_serializer). Perhaps a better way would be to have
969
# Repository._deserialise_chk_node()
970
search_key_func = chk_map.search_key_registry.get(
971
self._pack_collection.repo._serializer.search_key_name)
972
def accumlate_refs(lines):
973
# XXX: move to a generic location
975
bytes = ''.join(lines)
976
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
977
new_refs.update(node.refs())
978
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
979
self.new_pack.chk_index, output_lines=accumlate_refs)
982
def _copy_nodes(self, nodes, index_map, writer, write_index,
984
"""Copy knit nodes between packs with no graph references.
986
:param output_lines: Output full texts of copied items.
988
pb = ui.ui_factory.nested_progress_bar()
990
return self._do_copy_nodes(nodes, index_map, writer,
991
write_index, pb, output_lines=output_lines)
995
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
997
# for record verification
998
knit = KnitVersionedFiles(None, None)
999
# plan a readv on each source pack:
1001
nodes = sorted(nodes)
1002
# how to map this into knit.py - or knit.py into this?
1003
# we don't want the typical knit logic, we want grouping by pack
1004
# at this point - perhaps a helper library for the following code
1005
# duplication points?
1007
for index, key, value in nodes:
1008
if index not in request_groups:
1009
request_groups[index] = []
1010
request_groups[index].append((key, value))
1012
pb.update("Copied record", record_index, len(nodes))
1013
for index, items in request_groups.iteritems():
1014
pack_readv_requests = []
1015
for key, value in items:
1016
# ---- KnitGraphIndex.get_position
1017
bits = value[1:].split(' ')
1018
offset, length = int(bits[0]), int(bits[1])
1019
pack_readv_requests.append((offset, length, (key, value[0])))
1020
# linear scan up the pack
1021
pack_readv_requests.sort()
1023
pack_obj = index_map[index]
1024
transport, path = pack_obj.access_tuple()
1026
reader = pack.make_readv_reader(transport, path,
1027
[offset[0:2] for offset in pack_readv_requests])
1028
except errors.NoSuchFile:
1029
if self._reload_func is not None:
1032
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1033
izip(reader.iter_records(), pack_readv_requests):
1034
raw_data = read_func(None)
1035
# check the header only
1036
if output_lines is not None:
1037
output_lines(knit._parse_record(key[-1], raw_data)[0])
1039
df, _ = knit._parse_record_header(key, raw_data)
1041
pos, size = writer.add_bytes_record(raw_data, names)
1042
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1043
pb.update("Copied record", record_index)
1046
def _copy_nodes_graph(self, index_map, writer, write_index,
1047
readv_group_iter, total_items, output_lines=False):
1048
"""Copy knit nodes between packs.
1050
:param output_lines: Return lines present in the copied data as
1051
an iterator of line,version_id.
1053
pb = ui.ui_factory.nested_progress_bar()
1055
for result in self._do_copy_nodes_graph(index_map, writer,
1056
write_index, output_lines, pb, readv_group_iter, total_items):
1059
# Python 2.4 does not permit try:finally: in a generator.
1065
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1066
output_lines, pb, readv_group_iter, total_items):
1067
# for record verification
1068
knit = KnitVersionedFiles(None, None)
1069
# for line extraction when requested (inventories only)
1071
factory = KnitPlainFactory()
1073
pb.update("Copied record", record_index, total_items)
1074
for index, readv_vector, node_vector in readv_group_iter:
1076
pack_obj = index_map[index]
1077
transport, path = pack_obj.access_tuple()
1079
reader = pack.make_readv_reader(transport, path, readv_vector)
1080
except errors.NoSuchFile:
1081
if self._reload_func is not None:
1084
for (names, read_func), (key, eol_flag, references) in \
1085
izip(reader.iter_records(), node_vector):
1086
raw_data = read_func(None)
1088
# read the entire thing
1089
content, _ = knit._parse_record(key[-1], raw_data)
1090
if len(references[-1]) == 0:
1091
line_iterator = factory.get_fulltext_content(content)
1093
line_iterator = factory.get_linedelta_content(content)
1094
for line in line_iterator:
1097
# check the header only
1098
df, _ = knit._parse_record_header(key, raw_data)
1100
pos, size = writer.add_bytes_record(raw_data, names)
1101
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1102
pb.update("Copied record", record_index)
1105
def _get_text_nodes(self):
1106
text_index_map, text_indices = self._pack_map_and_index_list(
1108
return text_index_map, self._index_contents(text_indices,
1111
def _least_readv_node_readv(self, nodes):
1112
"""Generate request groups for nodes using the least readv's.
1114
:param nodes: An iterable of graph index nodes.
1115
:return: Total node count and an iterator of the data needed to perform
1116
readvs to obtain the data for nodes. Each item yielded by the
1117
iterator is a tuple with:
1118
index, readv_vector, node_vector. readv_vector is a list ready to
1119
hand to the transport readv method, and node_vector is a list of
1120
(key, eol_flag, references) for the node retrieved by the
1121
matching readv_vector.
1123
# group by pack so we do one readv per pack
1124
nodes = sorted(nodes)
1127
for index, key, value, references in nodes:
1128
if index not in request_groups:
1129
request_groups[index] = []
1130
request_groups[index].append((key, value, references))
1132
for index, items in request_groups.iteritems():
1133
pack_readv_requests = []
1134
for key, value, references in items:
1135
# ---- KnitGraphIndex.get_position
1136
bits = value[1:].split(' ')
1137
offset, length = int(bits[0]), int(bits[1])
1138
pack_readv_requests.append(
1139
((offset, length), (key, value[0], references)))
1140
# linear scan up the pack to maximum range combining.
1141
pack_readv_requests.sort()
1142
# split out the readv and the node data.
1143
pack_readv = [readv for readv, node in pack_readv_requests]
1144
node_vector = [node for readv, node in pack_readv_requests]
1145
result.append((index, pack_readv, node_vector))
1146
return total, result
1148
def _log_copied_texts(self):
1149
if 'pack' in debug.debug_flags:
1150
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1151
time.ctime(), self._pack_collection._upload_transport.base,
1152
self.new_pack.random_name,
1153
self.new_pack.text_index.key_count(),
1154
time.time() - self.new_pack.start_time)
1156
def _process_inventory_lines(self, inv_lines):
1157
"""Use up the inv_lines generator and setup a text key filter."""
1158
repo = self._pack_collection.repo
1159
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1160
inv_lines, self.revision_keys)
1162
for fileid, file_revids in fileid_revisions.iteritems():
1163
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1164
self._text_filter = text_filter
1166
def _revision_node_readv(self, revision_nodes):
1167
"""Return the total revisions and the readv's to issue.
1169
:param revision_nodes: The revision index contents for the packs being
1170
incorporated into the new pack.
1171
:return: As per _least_readv_node_readv.
1173
return self._least_readv_node_readv(revision_nodes)
1175
def _use_pack(self, new_pack):
1176
"""Return True if new_pack should be used.
1178
:param new_pack: The pack that has just been created.
1179
:return: True if the pack should be used.
1181
return new_pack.data_inserted()
1184
class OptimisingPacker(Packer):
1185
"""A packer which spends more time to create better disk layouts."""
1187
def _revision_node_readv(self, revision_nodes):
1188
"""Return the total revisions and the readv's to issue.
1190
This sort places revisions in topological order with the ancestors
1193
:param revision_nodes: The revision index contents for the packs being
1194
incorporated into the new pack.
1195
:return: As per _least_readv_node_readv.
1197
# build an ancestors dict
1200
for index, key, value, references in revision_nodes:
1201
ancestors[key] = references[0]
1202
by_key[key] = (index, value, references)
1203
order = tsort.topo_sort(ancestors)
1205
# Single IO is pathological, but it will work as a starting point.
1207
for key in reversed(order):
1208
index, value, references = by_key[key]
1209
# ---- KnitGraphIndex.get_position
1210
bits = value[1:].split(' ')
1211
offset, length = int(bits[0]), int(bits[1])
1213
(index, [(offset, length)], [(key, value[0], references)]))
1214
# TODO: combine requests in the same index that are in ascending order.
1215
return total, requests
1217
def open_pack(self):
1218
"""Open a pack for the pack we are creating."""
1219
new_pack = super(OptimisingPacker, self).open_pack()
1220
# Turn on the optimization flags for all the index builders.
1221
new_pack.revision_index.set_optimize(for_size=True)
1222
new_pack.inventory_index.set_optimize(for_size=True)
1223
new_pack.text_index.set_optimize(for_size=True)
1224
new_pack.signature_index.set_optimize(for_size=True)
1228
class ReconcilePacker(Packer):
1229
"""A packer which regenerates indices etc as it copies.
1231
This is used by ``bzr reconcile`` to cause parent text pointers to be
1235
def _extra_init(self):
1236
self._data_changed = False
1238
def _process_inventory_lines(self, inv_lines):
1239
"""Generate a text key reference map rather for reconciling with."""
1240
repo = self._pack_collection.repo
1241
refs = repo._find_text_key_references_from_xml_inventory_lines(
1243
self._text_refs = refs
1244
# during reconcile we:
1245
# - convert unreferenced texts to full texts
1246
# - correct texts which reference a text not copied to be full texts
1247
# - copy all others as-is but with corrected parents.
1248
# - so at this point we don't know enough to decide what becomes a full
1250
self._text_filter = None
1252
def _copy_text_texts(self):
1253
"""generate what texts we should have and then copy."""
1254
self.pb.update("Copying content texts", 3)
1255
# we have three major tasks here:
1256
# 1) generate the ideal index
1257
repo = self._pack_collection.repo
1258
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1259
_1, key, _2, refs in
1260
self.new_pack.revision_index.iter_all_entries()])
1261
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1262
# 2) generate a text_nodes list that contains all the deltas that can
1263
# be used as-is, with corrected parents.
1266
discarded_nodes = []
1267
NULL_REVISION = _mod_revision.NULL_REVISION
1268
text_index_map, text_nodes = self._get_text_nodes()
1269
for node in text_nodes:
1275
ideal_parents = tuple(ideal_index[node[1]])
1277
discarded_nodes.append(node)
1278
self._data_changed = True
1280
if ideal_parents == (NULL_REVISION,):
1282
if ideal_parents == node[3][0]:
1284
ok_nodes.append(node)
1285
elif ideal_parents[0:1] == node[3][0][0:1]:
1286
# the left most parent is the same, or there are no parents
1287
# today. Either way, we can preserve the representation as
1288
# long as we change the refs to be inserted.
1289
self._data_changed = True
1290
ok_nodes.append((node[0], node[1], node[2],
1291
(ideal_parents, node[3][1])))
1292
self._data_changed = True
1294
# Reinsert this text completely
1295
bad_texts.append((node[1], ideal_parents))
1296
self._data_changed = True
1297
# we're finished with some data.
1300
# 3) bulk copy the ok data
1301
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1302
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1303
self.new_pack.text_index, readv_group_iter, total_items))
1304
# 4) adhoc copy all the other texts.
1305
# We have to topologically insert all texts otherwise we can fail to
1306
# reconcile when parts of a single delta chain are preserved intact,
1307
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1308
# reinserted, and if d3 has incorrect parents it will also be
1309
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1310
# copied), so we will try to delta, but d2 is not currently able to be
1311
# extracted because it's basis d1 is not present. Topologically sorting
1312
# addresses this. The following generates a sort for all the texts that
1313
# are being inserted without having to reference the entire text key
1314
# space (we only topo sort the revisions, which is smaller).
1315
topo_order = tsort.topo_sort(ancestors)
1316
rev_order = dict(zip(topo_order, range(len(topo_order))))
1317
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1318
transaction = repo.get_transaction()
1319
file_id_index = GraphIndexPrefixAdapter(
1320
self.new_pack.text_index,
1322
add_nodes_callback=self.new_pack.text_index.add_nodes)
1323
data_access = _DirectPackAccess(
1324
{self.new_pack.text_index:self.new_pack.access_tuple()})
1325
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1326
self.new_pack.access_tuple())
1327
output_texts = KnitVersionedFiles(
1328
_KnitGraphIndex(self.new_pack.text_index,
1329
add_callback=self.new_pack.text_index.add_nodes,
1330
deltas=True, parents=True, is_locked=repo.is_locked),
1331
data_access=data_access, max_delta_chain=200)
1332
for key, parent_keys in bad_texts:
1333
# We refer to the new pack to delta data being output.
1334
# A possible improvement would be to catch errors on short reads
1335
# and only flush then.
1336
self.new_pack.flush()
1338
for parent_key in parent_keys:
1339
if parent_key[0] != key[0]:
1340
# Graph parents must match the fileid
1341
raise errors.BzrError('Mismatched key parent %r:%r' %
1343
parents.append(parent_key[1])
1344
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1345
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1346
output_texts.add_lines(key, parent_keys, text_lines,
1347
random_id=True, check_content=False)
1348
# 5) check that nothing inserted has a reference outside the keyspace.
1349
missing_text_keys = self.new_pack.text_index._external_references()
1350
if missing_text_keys:
1351
raise errors.BzrCheckError('Reference to missing compression parents %r'
1352
% (missing_text_keys,))
1353
self._log_copied_texts()
1355
def _use_pack(self, new_pack):
1356
"""Override _use_pack to check for reconcile having changed content."""
1357
# XXX: we might be better checking this at the copy time.
1358
original_inventory_keys = set()
1359
inv_index = self._pack_collection.inventory_index.combined_index
1360
for entry in inv_index.iter_all_entries():
1361
original_inventory_keys.add(entry[1])
1362
new_inventory_keys = set()
1363
for entry in new_pack.inventory_index.iter_all_entries():
1364
new_inventory_keys.add(entry[1])
1365
if new_inventory_keys != original_inventory_keys:
1366
self._data_changed = True
1367
return new_pack.data_inserted() and self._data_changed
1370
class RepositoryPackCollection(object):
1371
"""Management of packs within a repository.
1373
:ivar _names: map of {pack_name: (index_size,)}
1376
pack_factory = NewPack
1377
resumed_pack_factory = ResumedPack
1379
def __init__(self, repo, transport, index_transport, upload_transport,
1380
pack_transport, index_builder_class, index_class,
1382
"""Create a new RepositoryPackCollection.
1384
:param transport: Addresses the repository base directory
1385
(typically .bzr/repository/).
1386
:param index_transport: Addresses the directory containing indices.
1387
:param upload_transport: Addresses the directory into which packs are written
1388
while they're being created.
1389
:param pack_transport: Addresses the directory of existing complete packs.
1390
:param index_builder_class: The index builder class to use.
1391
:param index_class: The index class to use.
1392
:param use_chk_index: Whether to setup and manage a CHK index.
1394
# XXX: This should call self.reset()
1396
self.transport = transport
1397
self._index_transport = index_transport
1398
self._upload_transport = upload_transport
1399
self._pack_transport = pack_transport
1400
self._index_builder_class = index_builder_class
1401
self._index_class = index_class
1402
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1407
self._packs_by_name = {}
1408
# the previous pack-names content
1409
self._packs_at_load = None
1410
# when a pack is being created by this object, the state of that pack.
1411
self._new_pack = None
1412
# aggregated revision index data
1413
flush = self._flush_new_pack
1414
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1415
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1416
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1417
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1419
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1421
# used to determine if we're using a chk_index elsewhere.
1422
self.chk_index = None
1424
self._resumed_packs = []
1427
return '%s(%r)' % (self.__class__.__name__, self.repo)
1429
def add_pack_to_memory(self, pack):
1430
"""Make a Pack object available to the repository to satisfy queries.
1432
:param pack: A Pack object.
1434
if pack.name in self._packs_by_name:
1435
raise AssertionError(
1436
'pack %s already in _packs_by_name' % (pack.name,))
1437
self.packs.append(pack)
1438
self._packs_by_name[pack.name] = pack
1439
self.revision_index.add_index(pack.revision_index, pack)
1440
self.inventory_index.add_index(pack.inventory_index, pack)
1441
self.text_index.add_index(pack.text_index, pack)
1442
self.signature_index.add_index(pack.signature_index, pack)
1443
if self.chk_index is not None:
1444
self.chk_index.add_index(pack.chk_index, pack)
1446
def all_packs(self):
1447
"""Return a list of all the Pack objects this repository has.
1449
Note that an in-progress pack being created is not returned.
1451
:return: A list of Pack objects for all the packs in the repository.
1454
for name in self.names():
1455
result.append(self.get_pack_by_name(name))
1459
"""Pack the pack collection incrementally.
1461
This will not attempt global reorganisation or recompression,
1462
rather it will just ensure that the total number of packs does
1463
not grow without bound. It uses the _max_pack_count method to
1464
determine if autopacking is needed, and the pack_distribution
1465
method to determine the number of revisions in each pack.
1467
If autopacking takes place then the packs name collection will have
1468
been flushed to disk - packing requires updating the name collection
1469
in synchronisation with certain steps. Otherwise the names collection
1472
:return: Something evaluating true if packing took place.
1476
return self._do_autopack()
1477
except errors.RetryAutopack:
1478
# If we get a RetryAutopack exception, we should abort the
1479
# current action, and retry.
1482
def _do_autopack(self):
1483
# XXX: Should not be needed when the management of indices is sane.
1484
total_revisions = self.revision_index.combined_index.key_count()
1485
total_packs = len(self._names)
1486
if self._max_pack_count(total_revisions) >= total_packs:
1488
# determine which packs need changing
1489
pack_distribution = self.pack_distribution(total_revisions)
1491
for pack in self.all_packs():
1492
revision_count = pack.get_revision_count()
1493
if revision_count == 0:
1494
# revision less packs are not generated by normal operation,
1495
# only by operations like sign-my-commits, and thus will not
1496
# tend to grow rapdily or without bound like commit containing
1497
# packs do - leave them alone as packing them really should
1498
# group their data with the relevant commit, and that may
1499
# involve rewriting ancient history - which autopack tries to
1500
# avoid. Alternatively we could not group the data but treat
1501
# each of these as having a single revision, and thus add
1502
# one revision for each to the total revision count, to get
1503
# a matching distribution.
1505
existing_packs.append((revision_count, pack))
1506
pack_operations = self.plan_autopack_combinations(
1507
existing_packs, pack_distribution)
1508
num_new_packs = len(pack_operations)
1509
num_old_packs = sum([len(po[1]) for po in pack_operations])
1510
num_revs_affected = sum([po[0] for po in pack_operations])
1511
mutter('Auto-packing repository %s, which has %d pack files, '
1512
'containing %d revisions. Packing %d files into %d affecting %d'
1513
' revisions', self, total_packs, total_revisions, num_old_packs,
1514
num_new_packs, num_revs_affected)
1515
result = self._execute_pack_operations(pack_operations,
1516
reload_func=self._restart_autopack)
1517
mutter('Auto-packing repository %s completed', self)
1520
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1522
"""Execute a series of pack operations.
1524
:param pack_operations: A list of [revision_count, packs_to_combine].
1525
:param _packer_class: The class of packer to use (default: Packer).
1526
:return: The new pack names.
1528
for revision_count, packs in pack_operations:
1529
# we may have no-ops from the setup logic
1532
packer = _packer_class(self, packs, '.autopack',
1533
reload_func=reload_func)
1536
except errors.RetryWithNewPacks:
1537
# An exception is propagating out of this context, make sure
1538
# this packer has cleaned up. Packer() doesn't set its new_pack
1539
# state into the RepositoryPackCollection object, so we only
1540
# have access to it directly here.
1541
if packer.new_pack is not None:
1542
packer.new_pack.abort()
1545
self._remove_pack_from_memory(pack)
1546
# record the newly available packs and stop advertising the old
1548
result = self._save_pack_names(clear_obsolete_packs=True)
1549
# Move the old packs out of the way now they are no longer referenced.
1550
for revision_count, packs in pack_operations:
1551
self._obsolete_packs(packs)
1554
def _flush_new_pack(self):
1555
if self._new_pack is not None:
1556
self._new_pack.flush()
1558
def lock_names(self):
1559
"""Acquire the mutex around the pack-names index.
1561
This cannot be used in the middle of a read-only transaction on the
1564
self.repo.control_files.lock_write()
1566
def _already_packed(self):
1567
"""Is the collection already packed?"""
1568
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1570
def pack(self, hint=None):
1571
"""Pack the pack collection totally."""
1572
self.ensure_loaded()
1573
total_packs = len(self._names)
1574
if self._already_packed():
1576
total_revisions = self.revision_index.combined_index.key_count()
1577
# XXX: the following may want to be a class, to pack with a given
1579
mutter('Packing repository %s, which has %d pack files, '
1580
'containing %d revisions with hint %r.', self, total_packs,
1581
total_revisions, hint)
1582
# determine which packs need changing
1583
pack_operations = [[0, []]]
1584
for pack in self.all_packs():
1585
if hint is None or pack.name in hint:
1586
# Either no hint was provided (so we are packing everything),
1587
# or this pack was included in the hint.
1588
pack_operations[-1][0] += pack.get_revision_count()
1589
pack_operations[-1][1].append(pack)
1590
self._execute_pack_operations(pack_operations, OptimisingPacker)
1592
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1593
"""Plan a pack operation.
1595
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1597
:param pack_distribution: A list with the number of revisions desired
1600
if len(existing_packs) <= len(pack_distribution):
1602
existing_packs.sort(reverse=True)
1603
pack_operations = [[0, []]]
1604
# plan out what packs to keep, and what to reorganise
1605
while len(existing_packs):
1606
# take the largest pack, and if its less than the head of the
1607
# distribution chart we will include its contents in the new pack
1608
# for that position. If its larger, we remove its size from the
1609
# distribution chart
1610
next_pack_rev_count, next_pack = existing_packs.pop(0)
1611
if next_pack_rev_count >= pack_distribution[0]:
1612
# this is already packed 'better' than this, so we can
1613
# not waste time packing it.
1614
while next_pack_rev_count > 0:
1615
next_pack_rev_count -= pack_distribution[0]
1616
if next_pack_rev_count >= 0:
1618
del pack_distribution[0]
1620
# didn't use that entire bucket up
1621
pack_distribution[0] = -next_pack_rev_count
1623
# add the revisions we're going to add to the next output pack
1624
pack_operations[-1][0] += next_pack_rev_count
1625
# allocate this pack to the next pack sub operation
1626
pack_operations[-1][1].append(next_pack)
1627
if pack_operations[-1][0] >= pack_distribution[0]:
1628
# this pack is used up, shift left.
1629
del pack_distribution[0]
1630
pack_operations.append([0, []])
1631
# Now that we know which pack files we want to move, shove them all
1632
# into a single pack file.
1634
final_pack_list = []
1635
for num_revs, pack_files in pack_operations:
1636
final_rev_count += num_revs
1637
final_pack_list.extend(pack_files)
1638
if len(final_pack_list) == 1:
1639
raise AssertionError('We somehow generated an autopack with a'
1640
' single pack file being moved.')
1642
return [[final_rev_count, final_pack_list]]
1644
def ensure_loaded(self):
1645
"""Ensure we have read names from disk.
1647
:return: True if the disk names had not been previously read.
1649
# NB: if you see an assertion error here, its probably access against
1650
# an unlocked repo. Naughty.
1651
if not self.repo.is_locked():
1652
raise errors.ObjectNotLocked(self.repo)
1653
if self._names is None:
1655
self._packs_at_load = set()
1656
for index, key, value in self._iter_disk_pack_index():
1658
self._names[name] = self._parse_index_sizes(value)
1659
self._packs_at_load.add((key, value))
1663
# populate all the metadata.
1667
def _parse_index_sizes(self, value):
1668
"""Parse a string of index sizes."""
1669
return tuple([int(digits) for digits in value.split(' ')])
1671
def get_pack_by_name(self, name):
1672
"""Get a Pack object by name.
1674
:param name: The name of the pack - e.g. '123456'
1675
:return: A Pack object.
1678
return self._packs_by_name[name]
1680
rev_index = self._make_index(name, '.rix')
1681
inv_index = self._make_index(name, '.iix')
1682
txt_index = self._make_index(name, '.tix')
1683
sig_index = self._make_index(name, '.six')
1684
if self.chk_index is not None:
1685
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1688
result = ExistingPack(self._pack_transport, name, rev_index,
1689
inv_index, txt_index, sig_index, chk_index)
1690
self.add_pack_to_memory(result)
1693
def _resume_pack(self, name):
1694
"""Get a suspended Pack object by name.
1696
:param name: The name of the pack - e.g. '123456'
1697
:return: A Pack object.
1699
if not re.match('[a-f0-9]{32}', name):
1700
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1702
raise errors.UnresumableWriteGroup(
1703
self.repo, [name], 'Malformed write group token')
1705
rev_index = self._make_index(name, '.rix', resume=True)
1706
inv_index = self._make_index(name, '.iix', resume=True)
1707
txt_index = self._make_index(name, '.tix', resume=True)
1708
sig_index = self._make_index(name, '.six', resume=True)
1709
if self.chk_index is not None:
1710
chk_index = self._make_index(name, '.cix', resume=True,
1711
unlimited_cache=True)
1714
result = self.resumed_pack_factory(name, rev_index, inv_index,
1715
txt_index, sig_index, self._upload_transport,
1716
self._pack_transport, self._index_transport, self,
1717
chk_index=chk_index)
1718
except errors.NoSuchFile, e:
1719
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1720
self.add_pack_to_memory(result)
1721
self._resumed_packs.append(result)
1724
def allocate(self, a_new_pack):
1725
"""Allocate name in the list of packs.
1727
:param a_new_pack: A NewPack instance to be added to the collection of
1728
packs for this repository.
1730
self.ensure_loaded()
1731
if a_new_pack.name in self._names:
1732
raise errors.BzrError(
1733
'Pack %r already exists in %s' % (a_new_pack.name, self))
1734
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1735
self.add_pack_to_memory(a_new_pack)
1737
def _iter_disk_pack_index(self):
1738
"""Iterate over the contents of the pack-names index.
1740
This is used when loading the list from disk, and before writing to
1741
detect updates from others during our write operation.
1742
:return: An iterator of the index contents.
1744
return self._index_class(self.transport, 'pack-names', None
1745
).iter_all_entries()
1747
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1748
size_offset = self._suffix_offsets[suffix]
1749
index_name = name + suffix
1751
transport = self._upload_transport
1752
index_size = transport.stat(index_name).st_size
1754
transport = self._index_transport
1755
index_size = self._names[name][size_offset]
1756
return self._index_class(transport, index_name, index_size,
1757
unlimited_cache=unlimited_cache)
1759
def _max_pack_count(self, total_revisions):
1760
"""Return the maximum number of packs to use for total revisions.
1762
:param total_revisions: The total number of revisions in the
1765
if not total_revisions:
1767
digits = str(total_revisions)
1769
for digit in digits:
1770
result += int(digit)
1774
"""Provide an order to the underlying names."""
1775
return sorted(self._names.keys())
1777
def _obsolete_packs(self, packs):
1778
"""Move a number of packs which have been obsoleted out of the way.
1780
Each pack and its associated indices are moved out of the way.
1782
Note: for correctness this function should only be called after a new
1783
pack names index has been written without these pack names, and with
1784
the names of packs that contain the data previously available via these
1787
:param packs: The packs to obsolete.
1788
:param return: None.
1791
pack.pack_transport.rename(pack.file_name(),
1792
'../obsolete_packs/' + pack.file_name())
1793
# TODO: Probably needs to know all possible indices for this pack
1794
# - or maybe list the directory and move all indices matching this
1795
# name whether we recognize it or not?
1796
suffixes = ['.iix', '.six', '.tix', '.rix']
1797
if self.chk_index is not None:
1798
suffixes.append('.cix')
1799
for suffix in suffixes:
1800
self._index_transport.rename(pack.name + suffix,
1801
'../obsolete_packs/' + pack.name + suffix)
1803
def pack_distribution(self, total_revisions):
1804
"""Generate a list of the number of revisions to put in each pack.
1806
:param total_revisions: The total number of revisions in the
1809
if total_revisions == 0:
1811
digits = reversed(str(total_revisions))
1813
for exponent, count in enumerate(digits):
1814
size = 10 ** exponent
1815
for pos in range(int(count)):
1817
return list(reversed(result))
1819
def _pack_tuple(self, name):
1820
"""Return a tuple with the transport and file name for a pack name."""
1821
return self._pack_transport, name + '.pack'
1823
def _remove_pack_from_memory(self, pack):
1824
"""Remove pack from the packs accessed by this repository.
1826
Only affects memory state, until self._save_pack_names() is invoked.
1828
self._names.pop(pack.name)
1829
self._packs_by_name.pop(pack.name)
1830
self._remove_pack_indices(pack)
1831
self.packs.remove(pack)
1833
def _remove_pack_indices(self, pack):
1834
"""Remove the indices for pack from the aggregated indices."""
1835
self.revision_index.remove_index(pack.revision_index, pack)
1836
self.inventory_index.remove_index(pack.inventory_index, pack)
1837
self.text_index.remove_index(pack.text_index, pack)
1838
self.signature_index.remove_index(pack.signature_index, pack)
1839
if self.chk_index is not None:
1840
self.chk_index.remove_index(pack.chk_index, pack)
1843
"""Clear all cached data."""
1844
# cached revision data
1845
self.revision_index.clear()
1846
# cached signature data
1847
self.signature_index.clear()
1848
# cached file text data
1849
self.text_index.clear()
1850
# cached inventory data
1851
self.inventory_index.clear()
1853
if self.chk_index is not None:
1854
self.chk_index.clear()
1855
# remove the open pack
1856
self._new_pack = None
1857
# information about packs.
1860
self._packs_by_name = {}
1861
self._packs_at_load = None
1863
def _unlock_names(self):
1864
"""Release the mutex around the pack-names index."""
1865
self.repo.control_files.unlock()
1867
def _diff_pack_names(self):
1868
"""Read the pack names from disk, and compare it to the one in memory.
1870
:return: (disk_nodes, deleted_nodes, new_nodes)
1871
disk_nodes The final set of nodes that should be referenced
1872
deleted_nodes Nodes which have been removed from when we started
1873
new_nodes Nodes that are newly introduced
1875
# load the disk nodes across
1877
for index, key, value in self._iter_disk_pack_index():
1878
disk_nodes.add((key, value))
1880
# do a two-way diff against our original content
1881
current_nodes = set()
1882
for name, sizes in self._names.iteritems():
1884
((name, ), ' '.join(str(size) for size in sizes)))
1886
# Packs no longer present in the repository, which were present when we
1887
# locked the repository
1888
deleted_nodes = self._packs_at_load - current_nodes
1889
# Packs which this process is adding
1890
new_nodes = current_nodes - self._packs_at_load
1892
# Update the disk_nodes set to include the ones we are adding, and
1893
# remove the ones which were removed by someone else
1894
disk_nodes.difference_update(deleted_nodes)
1895
disk_nodes.update(new_nodes)
1897
return disk_nodes, deleted_nodes, new_nodes
1899
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1900
"""Given the correct set of pack files, update our saved info.
1902
:return: (removed, added, modified)
1903
removed pack names removed from self._names
1904
added pack names added to self._names
1905
modified pack names that had changed value
1910
## self._packs_at_load = disk_nodes
1911
new_names = dict(disk_nodes)
1912
# drop no longer present nodes
1913
for pack in self.all_packs():
1914
if (pack.name,) not in new_names:
1915
removed.append(pack.name)
1916
self._remove_pack_from_memory(pack)
1917
# add new nodes/refresh existing ones
1918
for key, value in disk_nodes:
1920
sizes = self._parse_index_sizes(value)
1921
if name in self._names:
1923
if sizes != self._names[name]:
1924
# the pack for name has had its indices replaced - rare but
1925
# important to handle. XXX: probably can never happen today
1926
# because the three-way merge code above does not handle it
1927
# - you may end up adding the same key twice to the new
1928
# disk index because the set values are the same, unless
1929
# the only index shows up as deleted by the set difference
1930
# - which it may. Until there is a specific test for this,
1931
# assume its broken. RBC 20071017.
1932
self._remove_pack_from_memory(self.get_pack_by_name(name))
1933
self._names[name] = sizes
1934
self.get_pack_by_name(name)
1935
modified.append(name)
1938
self._names[name] = sizes
1939
self.get_pack_by_name(name)
1941
return removed, added, modified
1943
def _save_pack_names(self, clear_obsolete_packs=False):
1944
"""Save the list of packs.
1946
This will take out the mutex around the pack names list for the
1947
duration of the method call. If concurrent updates have been made, a
1948
three-way merge between the current list and the current in memory list
1951
:param clear_obsolete_packs: If True, clear out the contents of the
1952
obsolete_packs directory.
1953
:return: A list of the names saved that were not previously on disk.
1957
builder = self._index_builder_class()
1958
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1959
# TODO: handle same-name, index-size-changes here -
1960
# e.g. use the value from disk, not ours, *unless* we're the one
1962
for key, value in disk_nodes:
1963
builder.add_node(key, value)
1964
self.transport.put_file('pack-names', builder.finish(),
1965
mode=self.repo.bzrdir._get_file_mode())
1966
# move the baseline forward
1967
self._packs_at_load = disk_nodes
1968
if clear_obsolete_packs:
1969
self._clear_obsolete_packs()
1971
self._unlock_names()
1972
# synchronise the memory packs list with what we just wrote:
1973
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1974
return [new_node[0][0] for new_node in new_nodes]
1976
def reload_pack_names(self):
1977
"""Sync our pack listing with what is present in the repository.
1979
This should be called when we find out that something we thought was
1980
present is now missing. This happens when another process re-packs the
1983
:return: True if the in-memory list of packs has been altered at all.
1985
# The ensure_loaded call is to handle the case where the first call
1986
# made involving the collection was to reload_pack_names, where we
1987
# don't have a view of disk contents. Its a bit of a bandaid, and
1988
# causes two reads of pack-names, but its a rare corner case not struck
1989
# with regular push/pull etc.
1990
first_read = self.ensure_loaded()
1993
# out the new value.
1994
disk_nodes, _, _ = self._diff_pack_names()
1995
self._packs_at_load = disk_nodes
1997
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1998
if removed or added or modified:
2002
def _restart_autopack(self):
2003
"""Reload the pack names list, and restart the autopack code."""
2004
if not self.reload_pack_names():
2005
# Re-raise the original exception, because something went missing
2006
# and a restart didn't find it
2008
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2010
def _clear_obsolete_packs(self):
2011
"""Delete everything from the obsolete-packs directory.
2013
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2014
for filename in obsolete_pack_transport.list_dir('.'):
2016
obsolete_pack_transport.delete(filename)
2017
except (errors.PathError, errors.TransportError), e:
2018
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2020
def _start_write_group(self):
2021
# Do not permit preparation for writing if we're not in a 'write lock'.
2022
if not self.repo.is_write_locked():
2023
raise errors.NotWriteLocked(self)
2024
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2025
file_mode=self.repo.bzrdir._get_file_mode())
2026
# allow writing: queue writes to a new index
2027
self.revision_index.add_writable_index(self._new_pack.revision_index,
2029
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2031
self.text_index.add_writable_index(self._new_pack.text_index,
2033
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2034
self.signature_index.add_writable_index(self._new_pack.signature_index,
2036
if self.chk_index is not None:
2037
self.chk_index.add_writable_index(self._new_pack.chk_index,
2039
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2040
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2042
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2043
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2044
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2045
self.repo.texts._index._add_callback = self.text_index.add_callback
2047
def _abort_write_group(self):
2048
# FIXME: just drop the transient index.
2049
# forget what names there are
2050
if self._new_pack is not None:
2052
self._new_pack.abort()
2054
# XXX: If we aborted while in the middle of finishing the write
2055
# group, _remove_pack_indices can fail because the indexes are
2056
# already gone. If they're not there we shouldn't fail in this
2057
# case. -- mbp 20081113
2058
self._remove_pack_indices(self._new_pack)
2059
self._new_pack = None
2060
for resumed_pack in self._resumed_packs:
2062
resumed_pack.abort()
2064
# See comment in previous finally block.
2066
self._remove_pack_indices(resumed_pack)
2069
del self._resumed_packs[:]
2071
def _remove_resumed_pack_indices(self):
2072
for resumed_pack in self._resumed_packs:
2073
self._remove_pack_indices(resumed_pack)
2074
del self._resumed_packs[:]
2076
def _check_new_inventories(self):
2077
"""Detect missing inventories in this write group.
2079
:returns: list of strs, summarising any problems found. If the list is
2080
empty no problems were found.
2082
# The base implementation does no checks. GCRepositoryPackCollection
2086
def _commit_write_group(self):
2088
for prefix, versioned_file in (
2089
('revisions', self.repo.revisions),
2090
('inventories', self.repo.inventories),
2091
('texts', self.repo.texts),
2092
('signatures', self.repo.signatures),
2094
missing = versioned_file.get_missing_compression_parent_keys()
2095
all_missing.update([(prefix,) + key for key in missing])
2097
raise errors.BzrCheckError(
2098
"Repository %s has missing compression parent(s) %r "
2099
% (self.repo, sorted(all_missing)))
2100
problems = self._check_new_inventories()
2102
problems_summary = '\n'.join(problems)
2103
raise errors.BzrCheckError(
2104
"Cannot add revision(s) to repository: " + problems_summary)
2105
self._remove_pack_indices(self._new_pack)
2106
any_new_content = False
2107
if self._new_pack.data_inserted():
2108
# get all the data to disk and read to use
2109
self._new_pack.finish()
2110
self.allocate(self._new_pack)
2111
self._new_pack = None
2112
any_new_content = True
2114
self._new_pack.abort()
2115
self._new_pack = None
2116
for resumed_pack in self._resumed_packs:
2117
# XXX: this is a pretty ugly way to turn the resumed pack into a
2118
# properly committed pack.
2119
self._names[resumed_pack.name] = None
2120
self._remove_pack_from_memory(resumed_pack)
2121
resumed_pack.finish()
2122
self.allocate(resumed_pack)
2123
any_new_content = True
2124
del self._resumed_packs[:]
2126
result = self.autopack()
2128
# when autopack takes no steps, the names list is still
2130
return self._save_pack_names()
2134
def _suspend_write_group(self):
2135
tokens = [pack.name for pack in self._resumed_packs]
2136
self._remove_pack_indices(self._new_pack)
2137
if self._new_pack.data_inserted():
2138
# get all the data to disk and read to use
2139
self._new_pack.finish(suspend=True)
2140
tokens.append(self._new_pack.name)
2141
self._new_pack = None
2143
self._new_pack.abort()
2144
self._new_pack = None
2145
self._remove_resumed_pack_indices()
2148
def _resume_write_group(self, tokens):
2149
for token in tokens:
2150
self._resume_pack(token)
2153
class KnitPackRepository(KnitRepository):
2154
"""Repository with knit objects stored inside pack containers.
2156
The layering for a KnitPackRepository is:
2158
Graph | HPSS | Repository public layer |
2159
===================================================
2160
Tuple based apis below, string based, and key based apis above
2161
---------------------------------------------------
2163
Provides .texts, .revisions etc
2164
This adapts the N-tuple keys to physical knit records which only have a
2165
single string identifier (for historical reasons), which in older formats
2166
was always the revision_id, and in the mapped code for packs is always
2167
the last element of key tuples.
2168
---------------------------------------------------
2170
A separate GraphIndex is used for each of the
2171
texts/inventories/revisions/signatures contained within each individual
2172
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2174
===================================================
2178
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2180
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2181
_commit_builder_class, _serializer)
2182
index_transport = self._transport.clone('indices')
2183
self._pack_collection = RepositoryPackCollection(self, self._transport,
2185
self._transport.clone('upload'),
2186
self._transport.clone('packs'),
2187
_format.index_builder_class,
2188
_format.index_class,
2189
use_chk_index=self._format.supports_chks,
2191
self.inventories = KnitVersionedFiles(
2192
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2193
add_callback=self._pack_collection.inventory_index.add_callback,
2194
deltas=True, parents=True, is_locked=self.is_locked),
2195
data_access=self._pack_collection.inventory_index.data_access,
2196
max_delta_chain=200)
2197
self.revisions = KnitVersionedFiles(
2198
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2199
add_callback=self._pack_collection.revision_index.add_callback,
2200
deltas=False, parents=True, is_locked=self.is_locked,
2201
track_external_parent_refs=True),
2202
data_access=self._pack_collection.revision_index.data_access,
2204
self.signatures = KnitVersionedFiles(
2205
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2206
add_callback=self._pack_collection.signature_index.add_callback,
2207
deltas=False, parents=False, is_locked=self.is_locked),
2208
data_access=self._pack_collection.signature_index.data_access,
2210
self.texts = KnitVersionedFiles(
2211
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2212
add_callback=self._pack_collection.text_index.add_callback,
2213
deltas=True, parents=True, is_locked=self.is_locked),
2214
data_access=self._pack_collection.text_index.data_access,
2215
max_delta_chain=200)
2216
if _format.supports_chks:
2217
# No graph, no compression:- references from chks are between
2218
# different objects not temporal versions of the same; and without
2219
# some sort of temporal structure knit compression will just fail.
2220
self.chk_bytes = KnitVersionedFiles(
2221
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2222
add_callback=self._pack_collection.chk_index.add_callback,
2223
deltas=False, parents=False, is_locked=self.is_locked),
2224
data_access=self._pack_collection.chk_index.data_access,
2227
self.chk_bytes = None
2228
# True when the repository object is 'write locked' (as opposed to the
2229
# physical lock only taken out around changes to the pack-names list.)
2230
# Another way to represent this would be a decorator around the control
2231
# files object that presents logical locks as physical ones - if this
2232
# gets ugly consider that alternative design. RBC 20071011
2233
self._write_lock_count = 0
2234
self._transaction = None
2236
self._reconcile_does_inventory_gc = True
2237
self._reconcile_fixes_text_parents = True
2238
self._reconcile_backsup_inventory = False
2240
def _warn_if_deprecated(self, branch=None):
2241
# This class isn't deprecated, but one sub-format is
2242
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2243
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2245
def _abort_write_group(self):
2246
self.revisions._index._key_dependencies.clear()
2247
self._pack_collection._abort_write_group()
2249
def _get_source(self, to_format):
2250
if to_format.network_name() == self._format.network_name():
2251
return KnitPackStreamSource(self, to_format)
2252
return super(KnitPackRepository, self)._get_source(to_format)
2254
def _make_parents_provider(self):
2255
return graph.CachingParentsProvider(self)
2257
def _refresh_data(self):
2258
if not self.is_locked():
2260
self._pack_collection.reload_pack_names()
2262
def _start_write_group(self):
2263
self._pack_collection._start_write_group()
2265
def _commit_write_group(self):
2266
hint = self._pack_collection._commit_write_group()
2267
self.revisions._index._key_dependencies.clear()
2270
def suspend_write_group(self):
2271
# XXX check self._write_group is self.get_transaction()?
2272
tokens = self._pack_collection._suspend_write_group()
2273
self.revisions._index._key_dependencies.clear()
2274
self._write_group = None
2277
def _resume_write_group(self, tokens):
2278
self._start_write_group()
2280
self._pack_collection._resume_write_group(tokens)
2281
except errors.UnresumableWriteGroup:
2282
self._abort_write_group()
2284
for pack in self._pack_collection._resumed_packs:
2285
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2287
def get_transaction(self):
2288
if self._write_lock_count:
2289
return self._transaction
2291
return self.control_files.get_transaction()
2293
def is_locked(self):
2294
return self._write_lock_count or self.control_files.is_locked()
2296
def is_write_locked(self):
2297
return self._write_lock_count
2299
def lock_write(self, token=None):
2300
locked = self.is_locked()
2301
if not self._write_lock_count and locked:
2302
raise errors.ReadOnlyError(self)
2303
self._write_lock_count += 1
2304
if self._write_lock_count == 1:
2305
self._transaction = transactions.WriteTransaction()
2307
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2308
note('%r was write locked again', self)
2309
self._prev_lock = 'w'
2310
for repo in self._fallback_repositories:
2311
# Writes don't affect fallback repos
2313
self._refresh_data()
2315
def lock_read(self):
2316
locked = self.is_locked()
2317
if self._write_lock_count:
2318
self._write_lock_count += 1
2320
self.control_files.lock_read()
2322
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2323
note('%r was read locked again', self)
2324
self._prev_lock = 'r'
2325
for repo in self._fallback_repositories:
2327
self._refresh_data()
2329
def leave_lock_in_place(self):
2330
# not supported - raise an error
2331
raise NotImplementedError(self.leave_lock_in_place)
2333
def dont_leave_lock_in_place(self):
2334
# not supported - raise an error
2335
raise NotImplementedError(self.dont_leave_lock_in_place)
2338
def pack(self, hint=None):
2339
"""Compress the data within the repository.
2341
This will pack all the data to a single pack. In future it may
2342
recompress deltas or do other such expensive operations.
2344
self._pack_collection.pack(hint=hint)
2347
def reconcile(self, other=None, thorough=False):
2348
"""Reconcile this repository."""
2349
from bzrlib.reconcile import PackReconciler
2350
reconciler = PackReconciler(self, thorough=thorough)
2351
reconciler.reconcile()
2354
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2355
packer = ReconcilePacker(collection, packs, extension, revs)
2356
return packer.pack(pb)
2358
@only_raises(errors.LockNotHeld, errors.LockBroken)
2360
if self._write_lock_count == 1 and self._write_group is not None:
2361
self.abort_write_group()
2362
self._transaction = None
2363
self._write_lock_count = 0
2364
raise errors.BzrError(
2365
'Must end write group before releasing write lock on %s'
2367
if self._write_lock_count:
2368
self._write_lock_count -= 1
2369
if not self._write_lock_count:
2370
transaction = self._transaction
2371
self._transaction = None
2372
transaction.finish()
2374
self.control_files.unlock()
2376
if not self.is_locked():
2377
for repo in self._fallback_repositories:
2381
class KnitPackStreamSource(StreamSource):
2382
"""A StreamSource used to transfer data between same-format KnitPack repos.
2384
This source assumes:
2385
1) Same serialization format for all objects
2386
2) Same root information
2387
3) XML format inventories
2388
4) Atomic inserts (so we can stream inventory texts before text
2393
def __init__(self, from_repository, to_format):
2394
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2395
self._text_keys = None
2396
self._text_fetch_order = 'unordered'
2398
def _get_filtered_inv_stream(self, revision_ids):
2399
from_repo = self.from_repository
2400
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2401
parent_keys = [(p,) for p in parent_ids]
2402
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2403
parent_text_keys = set(find_text_keys(
2404
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2405
content_text_keys = set()
2406
knit = KnitVersionedFiles(None, None)
2407
factory = KnitPlainFactory()
2408
def find_text_keys_from_content(record):
2409
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2410
raise ValueError("Unknown content storage kind for"
2411
" inventory text: %s" % (record.storage_kind,))
2412
# It's a knit record, it has a _raw_record field (even if it was
2413
# reconstituted from a network stream).
2414
raw_data = record._raw_record
2415
# read the entire thing
2416
revision_id = record.key[-1]
2417
content, _ = knit._parse_record(revision_id, raw_data)
2418
if record.storage_kind == 'knit-delta-gz':
2419
line_iterator = factory.get_linedelta_content(content)
2420
elif record.storage_kind == 'knit-ft-gz':
2421
line_iterator = factory.get_fulltext_content(content)
2422
content_text_keys.update(find_text_keys(
2423
[(line, revision_id) for line in line_iterator]))
2424
revision_keys = [(r,) for r in revision_ids]
2425
def _filtered_inv_stream():
2426
source_vf = from_repo.inventories
2427
stream = source_vf.get_record_stream(revision_keys,
2429
for record in stream:
2430
if record.storage_kind == 'absent':
2431
raise errors.NoSuchRevision(from_repo, record.key)
2432
find_text_keys_from_content(record)
2434
self._text_keys = content_text_keys - parent_text_keys
2435
return ('inventories', _filtered_inv_stream())
2437
def _get_text_stream(self):
2438
# Note: We know we don't have to handle adding root keys, because both
2439
# the source and target are the identical network name.
2440
text_stream = self.from_repository.texts.get_record_stream(
2441
self._text_keys, self._text_fetch_order, False)
2442
return ('texts', text_stream)
2444
def get_stream(self, search):
2445
revision_ids = search.get_keys()
2446
for stream_info in self._fetch_revision_texts(revision_ids):
2448
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2449
yield self._get_filtered_inv_stream(revision_ids)
2450
yield self._get_text_stream()
2454
class RepositoryFormatPack(MetaDirRepositoryFormat):
2455
"""Format logic for pack structured repositories.
2457
This repository format has:
2458
- a list of packs in pack-names
2459
- packs in packs/NAME.pack
2460
- indices in indices/NAME.{iix,six,tix,rix}
2461
- knit deltas in the packs, knit indices mapped to the indices.
2462
- thunk objects to support the knits programming API.
2463
- a format marker of its own
2464
- an optional 'shared-storage' flag
2465
- an optional 'no-working-trees' flag
2469
# Set this attribute in derived classes to control the repository class
2470
# created by open and initialize.
2471
repository_class = None
2472
# Set this attribute in derived classes to control the
2473
# _commit_builder_class that the repository objects will have passed to
2474
# their constructor.
2475
_commit_builder_class = None
2476
# Set this attribute in derived clases to control the _serializer that the
2477
# repository objects will have passed to their constructor.
2479
# Packs are not confused by ghosts.
2480
supports_ghosts = True
2481
# External references are not supported in pack repositories yet.
2482
supports_external_lookups = False
2483
# Most pack formats do not use chk lookups.
2484
supports_chks = False
2485
# What index classes to use
2486
index_builder_class = None
2488
_fetch_uses_deltas = True
2491
def initialize(self, a_bzrdir, shared=False):
2492
"""Create a pack based repository.
2494
:param a_bzrdir: bzrdir to contain the new repository; must already
2496
:param shared: If true the repository will be initialized as a shared
2499
mutter('creating repository in %s.', a_bzrdir.transport.base)
2500
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2501
builder = self.index_builder_class()
2502
files = [('pack-names', builder.finish())]
2503
utf8_files = [('format', self.get_format_string())]
2505
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2506
return self.open(a_bzrdir=a_bzrdir, _found=True)
2508
def open(self, a_bzrdir, _found=False, _override_transport=None):
2509
"""See RepositoryFormat.open().
2511
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2512
repository at a slightly different url
2513
than normal. I.e. during 'upgrade'.
2516
format = RepositoryFormat.find_format(a_bzrdir)
2517
if _override_transport is not None:
2518
repo_transport = _override_transport
2520
repo_transport = a_bzrdir.get_repository_transport(None)
2521
control_files = lockable_files.LockableFiles(repo_transport,
2522
'lock', lockdir.LockDir)
2523
return self.repository_class(_format=self,
2525
control_files=control_files,
2526
_commit_builder_class=self._commit_builder_class,
2527
_serializer=self._serializer)
2530
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2531
"""A no-subtrees parameterized Pack repository.
2533
This format was introduced in 0.92.
2536
repository_class = KnitPackRepository
2537
_commit_builder_class = PackCommitBuilder
2539
def _serializer(self):
2540
return xml5.serializer_v5
2541
# What index classes to use
2542
index_builder_class = InMemoryGraphIndex
2543
index_class = GraphIndex
2545
def _get_matching_bzrdir(self):
2546
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2548
def _ignore_setting_bzrdir(self, format):
2551
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2553
def get_format_string(self):
2554
"""See RepositoryFormat.get_format_string()."""
2555
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2557
def get_format_description(self):
2558
"""See RepositoryFormat.get_format_description()."""
2559
return "Packs containing knits without subtree support"
2562
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2563
"""A subtrees parameterized Pack repository.
2565
This repository format uses the xml7 serializer to get:
2566
- support for recording full info about the tree root
2567
- support for recording tree-references
2569
This format was introduced in 0.92.
2572
repository_class = KnitPackRepository
2573
_commit_builder_class = PackRootCommitBuilder
2574
rich_root_data = True
2575
supports_tree_reference = True
2577
def _serializer(self):
2578
return xml7.serializer_v7
2579
# What index classes to use
2580
index_builder_class = InMemoryGraphIndex
2581
index_class = GraphIndex
2583
def _get_matching_bzrdir(self):
2584
return bzrdir.format_registry.make_bzrdir(
2585
'pack-0.92-subtree')
2587
def _ignore_setting_bzrdir(self, format):
2590
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2592
def get_format_string(self):
2593
"""See RepositoryFormat.get_format_string()."""
2594
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2596
def get_format_description(self):
2597
"""See RepositoryFormat.get_format_description()."""
2598
return "Packs containing knits with subtree support\n"
2601
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2602
"""A rich-root, no subtrees parameterized Pack repository.
2604
This repository format uses the xml6 serializer to get:
2605
- support for recording full info about the tree root
2607
This format was introduced in 1.0.
2610
repository_class = KnitPackRepository
2611
_commit_builder_class = PackRootCommitBuilder
2612
rich_root_data = True
2613
supports_tree_reference = False
2615
def _serializer(self):
2616
return xml6.serializer_v6
2617
# What index classes to use
2618
index_builder_class = InMemoryGraphIndex
2619
index_class = GraphIndex
2621
def _get_matching_bzrdir(self):
2622
return bzrdir.format_registry.make_bzrdir(
2625
def _ignore_setting_bzrdir(self, format):
2628
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2630
def get_format_string(self):
2631
"""See RepositoryFormat.get_format_string()."""
2632
return ("Bazaar pack repository format 1 with rich root"
2633
" (needs bzr 1.0)\n")
2635
def get_format_description(self):
2636
"""See RepositoryFormat.get_format_description()."""
2637
return "Packs containing knits with rich root support\n"
2640
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2641
"""Repository that supports external references to allow stacking.
2645
Supports external lookups, which results in non-truncated ghosts after
2646
reconcile compared to pack-0.92 formats.
2649
repository_class = KnitPackRepository
2650
_commit_builder_class = PackCommitBuilder
2651
supports_external_lookups = True
2652
# What index classes to use
2653
index_builder_class = InMemoryGraphIndex
2654
index_class = GraphIndex
2657
def _serializer(self):
2658
return xml5.serializer_v5
2660
def _get_matching_bzrdir(self):
2661
return bzrdir.format_registry.make_bzrdir('1.6')
2663
def _ignore_setting_bzrdir(self, format):
2666
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2668
def get_format_string(self):
2669
"""See RepositoryFormat.get_format_string()."""
2670
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2672
def get_format_description(self):
2673
"""See RepositoryFormat.get_format_description()."""
2674
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2677
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2678
"""A repository with rich roots and stacking.
2680
New in release 1.6.1.
2682
Supports stacking on other repositories, allowing data to be accessed
2683
without being stored locally.
2686
repository_class = KnitPackRepository
2687
_commit_builder_class = PackRootCommitBuilder
2688
rich_root_data = True
2689
supports_tree_reference = False # no subtrees
2690
supports_external_lookups = True
2691
# What index classes to use
2692
index_builder_class = InMemoryGraphIndex
2693
index_class = GraphIndex
2696
def _serializer(self):
2697
return xml6.serializer_v6
2699
def _get_matching_bzrdir(self):
2700
return bzrdir.format_registry.make_bzrdir(
2703
def _ignore_setting_bzrdir(self, format):
2706
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2708
def get_format_string(self):
2709
"""See RepositoryFormat.get_format_string()."""
2710
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2712
def get_format_description(self):
2713
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2716
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2717
"""A repository with rich roots and external references.
2721
Supports external lookups, which results in non-truncated ghosts after
2722
reconcile compared to pack-0.92 formats.
2724
This format was deprecated because the serializer it uses accidentally
2725
supported subtrees, when the format was not intended to. This meant that
2726
someone could accidentally fetch from an incorrect repository.
2729
repository_class = KnitPackRepository
2730
_commit_builder_class = PackRootCommitBuilder
2731
rich_root_data = True
2732
supports_tree_reference = False # no subtrees
2734
supports_external_lookups = True
2735
# What index classes to use
2736
index_builder_class = InMemoryGraphIndex
2737
index_class = GraphIndex
2740
def _serializer(self):
2741
return xml7.serializer_v7
2743
def _get_matching_bzrdir(self):
2744
matching = bzrdir.format_registry.make_bzrdir(
2746
matching.repository_format = self
2749
def _ignore_setting_bzrdir(self, format):
2752
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2754
def get_format_string(self):
2755
"""See RepositoryFormat.get_format_string()."""
2756
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2758
def get_format_description(self):
2759
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2763
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2764
"""A repository with stacking and btree indexes,
2765
without rich roots or subtrees.
2767
This is equivalent to pack-1.6 with B+Tree indices.
2770
repository_class = KnitPackRepository
2771
_commit_builder_class = PackCommitBuilder
2772
supports_external_lookups = True
2773
# What index classes to use
2774
index_builder_class = BTreeBuilder
2775
index_class = BTreeGraphIndex
2778
def _serializer(self):
2779
return xml5.serializer_v5
2781
def _get_matching_bzrdir(self):
2782
return bzrdir.format_registry.make_bzrdir('1.9')
2784
def _ignore_setting_bzrdir(self, format):
2787
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2789
def get_format_string(self):
2790
"""See RepositoryFormat.get_format_string()."""
2791
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2793
def get_format_description(self):
2794
"""See RepositoryFormat.get_format_description()."""
2795
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2798
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2799
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2801
1.6-rich-root with B+Tree indices.
2804
repository_class = KnitPackRepository
2805
_commit_builder_class = PackRootCommitBuilder
2806
rich_root_data = True
2807
supports_tree_reference = False # no subtrees
2808
supports_external_lookups = True
2809
# What index classes to use
2810
index_builder_class = BTreeBuilder
2811
index_class = BTreeGraphIndex
2814
def _serializer(self):
2815
return xml6.serializer_v6
2817
def _get_matching_bzrdir(self):
2818
return bzrdir.format_registry.make_bzrdir(
2821
def _ignore_setting_bzrdir(self, format):
2824
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2826
def get_format_string(self):
2827
"""See RepositoryFormat.get_format_string()."""
2828
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2830
def get_format_description(self):
2831
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2834
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2835
"""A subtrees development repository.
2837
This format should be retained until the second release after bzr 1.7.
2839
1.6.1-subtree[as it might have been] with B+Tree indices.
2841
This is [now] retained until we have a CHK based subtree format in
2845
repository_class = KnitPackRepository
2846
_commit_builder_class = PackRootCommitBuilder
2847
rich_root_data = True
2848
supports_tree_reference = True
2849
supports_external_lookups = True
2850
# What index classes to use
2851
index_builder_class = BTreeBuilder
2852
index_class = BTreeGraphIndex
2855
def _serializer(self):
2856
return xml7.serializer_v7
2858
def _get_matching_bzrdir(self):
2859
return bzrdir.format_registry.make_bzrdir(
2860
'development-subtree')
2862
def _ignore_setting_bzrdir(self, format):
2865
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2867
def get_format_string(self):
2868
"""See RepositoryFormat.get_format_string()."""
2869
return ("Bazaar development format 2 with subtree support "
2870
"(needs bzr.dev from before 1.8)\n")
2872
def get_format_description(self):
2873
"""See RepositoryFormat.get_format_description()."""
2874
return ("Development repository format, currently the same as "
2875
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
