594
557
return new_pack.data_inserted() and self._data_changed
597
class GCCHKCanonicalizingPacker(GCCHKPacker):
598
"""A packer that ensures inventories have canonical-form CHK maps.
600
Ideally this would be part of reconcile, but it's very slow and rarely
601
needed. (It repairs repositories affected by
602
https://bugs.launchpad.net/bzr/+bug/522637).
605
def __init__(self, *args, **kwargs):
606
super(GCCHKCanonicalizingPacker, self).__init__(*args, **kwargs)
607
self._data_changed = False
609
def _exhaust_stream(self, source_vf, keys, message, vf_to_stream, pb_offset):
610
"""Create and exhaust a stream, but don't insert it.
612
This is useful to get the side-effects of generating a stream.
614
self.pb.update('scanning %s' % (message,), pb_offset)
615
child_pb = ui.ui_factory.nested_progress_bar()
617
list(vf_to_stream(source_vf, keys, message, child_pb))
621
def _copy_inventory_texts(self):
622
source_vf, target_vf = self._build_vfs('inventory', True, True)
623
source_chk_vf, target_chk_vf = self._get_chk_vfs_for_copy()
624
inventory_keys = source_vf.keys()
625
# First, copy the existing CHKs on the assumption that most of them
626
# will be correct. This will save us from having to reinsert (and
627
# recompress) these records later at the cost of perhaps preserving a
629
# (Iterate but don't insert _get_filtered_inv_stream to populate the
630
# variables needed by GCCHKPacker._copy_chk_texts.)
631
self._exhaust_stream(source_vf, inventory_keys, 'inventories',
632
self._get_filtered_inv_stream, 2)
633
GCCHKPacker._copy_chk_texts(self)
634
# Now copy and fix the inventories, and any regenerated CHKs.
635
def chk_canonicalizing_inv_stream(source_vf, keys, message, pb=None):
636
return self._get_filtered_canonicalizing_inv_stream(
637
source_vf, keys, message, pb, source_chk_vf, target_chk_vf)
638
self._copy_stream(source_vf, target_vf, inventory_keys,
639
'inventories', chk_canonicalizing_inv_stream, 4)
641
def _copy_chk_texts(self):
642
# No-op; in this class this happens during _copy_inventory_texts.
645
def _get_filtered_canonicalizing_inv_stream(self, source_vf, keys, message,
646
pb=None, source_chk_vf=None, target_chk_vf=None):
647
"""Filter the texts of inventories, regenerating CHKs to make sure they
650
total_keys = len(keys)
651
target_chk_vf = versionedfile.NoDupeAddLinesDecorator(target_chk_vf)
652
def _filtered_inv_stream():
653
stream = source_vf.get_record_stream(keys, 'groupcompress', True)
654
search_key_name = None
655
for idx, record in enumerate(stream):
656
# Inventories should always be with revisions; assume success.
657
bytes = record.get_bytes_as('fulltext')
658
chk_inv = inventory.CHKInventory.deserialise(
659
source_chk_vf, bytes, record.key)
661
pb.update('inv', idx, total_keys)
662
chk_inv.id_to_entry._ensure_root()
663
if search_key_name is None:
664
# Find the name corresponding to the search_key_func
665
search_key_reg = chk_map.search_key_registry
666
for search_key_name, func in search_key_reg.iteritems():
667
if func == chk_inv.id_to_entry._search_key_func:
669
canonical_inv = inventory.CHKInventory.from_inventory(
670
target_chk_vf, chk_inv,
671
maximum_size=chk_inv.id_to_entry._root_node._maximum_size,
672
search_key_name=search_key_name)
673
if chk_inv.id_to_entry.key() != canonical_inv.id_to_entry.key():
675
'Non-canonical CHK map for id_to_entry of inv: %s '
676
'(root is %s, should be %s)' % (chk_inv.revision_id,
677
chk_inv.id_to_entry.key()[0],
678
canonical_inv.id_to_entry.key()[0]))
679
self._data_changed = True
680
p_id_map = chk_inv.parent_id_basename_to_file_id
681
p_id_map._ensure_root()
682
canon_p_id_map = canonical_inv.parent_id_basename_to_file_id
683
if p_id_map.key() != canon_p_id_map.key():
685
'Non-canonical CHK map for parent_id_to_basename of '
686
'inv: %s (root is %s, should be %s)'
687
% (chk_inv.revision_id, p_id_map.key()[0],
688
canon_p_id_map.key()[0]))
689
self._data_changed = True
690
yield versionedfile.ChunkedContentFactory(record.key,
691
record.parents, record.sha1,
692
canonical_inv.to_lines())
693
# We have finished processing all of the inventory records, we
694
# don't need these sets anymore
695
return _filtered_inv_stream()
697
def _use_pack(self, new_pack):
698
"""Override _use_pack to check for reconcile having changed content."""
699
return new_pack.data_inserted() and self._data_changed
702
560
class GCRepositoryPackCollection(RepositoryPackCollection):
704
562
pack_factory = GCPack
705
563
resumed_pack_factory = ResumedGCPack
706
normal_packer_class = GCCHKPacker
707
optimising_packer_class = GCCHKPacker
709
def _check_new_inventories(self):
710
"""Detect missing inventories or chk root entries for the new revisions
713
:returns: list of strs, summarising any problems found. If the list is
714
empty no problems were found.
565
def _execute_pack_operations(self, pack_operations,
566
_packer_class=GCCHKPacker,
568
"""Execute a series of pack operations.
570
:param pack_operations: A list of [revision_count, packs_to_combine].
571
:param _packer_class: The class of packer to use (default: Packer).
716
# Ensure that all revisions added in this write group have:
717
# - corresponding inventories,
718
# - chk root entries for those inventories,
719
# - and any present parent inventories have their chk root
721
# And all this should be independent of any fallback repository.
723
key_deps = self.repo.revisions._index._key_dependencies
724
new_revisions_keys = key_deps.get_new_keys()
725
no_fallback_inv_index = self.repo.inventories._index
726
no_fallback_chk_bytes_index = self.repo.chk_bytes._index
727
no_fallback_texts_index = self.repo.texts._index
728
inv_parent_map = no_fallback_inv_index.get_parent_map(
730
# Are any inventories for corresponding to the new revisions missing?
731
corresponding_invs = set(inv_parent_map)
732
missing_corresponding = set(new_revisions_keys)
733
missing_corresponding.difference_update(corresponding_invs)
734
if missing_corresponding:
735
problems.append("inventories missing for revisions %s" %
736
(sorted(missing_corresponding),))
738
# Are any chk root entries missing for any inventories? This includes
739
# any present parent inventories, which may be used when calculating
740
# deltas for streaming.
741
all_inv_keys = set(corresponding_invs)
742
for parent_inv_keys in inv_parent_map.itervalues():
743
all_inv_keys.update(parent_inv_keys)
744
# Filter out ghost parents.
745
all_inv_keys.intersection_update(
746
no_fallback_inv_index.get_parent_map(all_inv_keys))
747
parent_invs_only_keys = all_inv_keys.symmetric_difference(
750
inv_ids = [key[-1] for key in all_inv_keys]
751
parent_invs_only_ids = [key[-1] for key in parent_invs_only_keys]
752
root_key_info = _build_interesting_key_sets(
753
self.repo, inv_ids, parent_invs_only_ids)
754
expected_chk_roots = root_key_info.all_keys()
755
present_chk_roots = no_fallback_chk_bytes_index.get_parent_map(
757
missing_chk_roots = expected_chk_roots.difference(present_chk_roots)
758
if missing_chk_roots:
759
problems.append("missing referenced chk root keys: %s"
760
% (sorted(missing_chk_roots),))
761
# Don't bother checking any further.
763
# Find all interesting chk_bytes records, and make sure they are
764
# present, as well as the text keys they reference.
765
chk_bytes_no_fallbacks = self.repo.chk_bytes.without_fallbacks()
766
chk_bytes_no_fallbacks._search_key_func = \
767
self.repo.chk_bytes._search_key_func
768
chk_diff = chk_map.iter_interesting_nodes(
769
chk_bytes_no_fallbacks, root_key_info.interesting_root_keys,
770
root_key_info.uninteresting_root_keys)
773
for record in _filter_text_keys(chk_diff, text_keys,
774
chk_map._bytes_to_text_key):
776
except errors.NoSuchRevision, e:
777
# XXX: It would be nice if we could give a more precise error here.
778
problems.append("missing chk node(s) for id_to_entry maps")
779
chk_diff = chk_map.iter_interesting_nodes(
780
chk_bytes_no_fallbacks, root_key_info.interesting_pid_root_keys,
781
root_key_info.uninteresting_pid_root_keys)
783
for interesting_rec, interesting_map in chk_diff:
785
except errors.NoSuchRevision, e:
787
"missing chk node(s) for parent_id_basename_to_file_id maps")
788
present_text_keys = no_fallback_texts_index.get_parent_map(text_keys)
789
missing_text_keys = text_keys.difference(present_text_keys)
790
if missing_text_keys:
791
problems.append("missing text keys: %r"
792
% (sorted(missing_text_keys),))
796
class CHKInventoryRepository(PackRepository):
797
"""subclass of PackRepository that uses CHK based inventories."""
574
# XXX: Copied across from RepositoryPackCollection simply because we
575
# want to override the _packer_class ... :(
576
for revision_count, packs in pack_operations:
577
# we may have no-ops from the setup logic
580
packer = GCCHKPacker(self, packs, '.autopack',
581
reload_func=reload_func)
584
except errors.RetryWithNewPacks:
585
# An exception is propagating out of this context, make sure
586
# this packer has cleaned up. Packer() doesn't set its new_pack
587
# state into the RepositoryPackCollection object, so we only
588
# have access to it directly here.
589
if packer.new_pack is not None:
590
packer.new_pack.abort()
593
self._remove_pack_from_memory(pack)
594
# record the newly available packs and stop advertising the old
596
self._save_pack_names(clear_obsolete_packs=True)
597
# Move the old packs out of the way now they are no longer referenced.
598
for revision_count, packs in pack_operations:
599
self._obsolete_packs(packs)
602
class CHKInventoryRepository(KnitPackRepository):
603
"""subclass of KnitPackRepository that uses CHK based inventories."""
799
605
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
801
607
"""Overridden to change pack collection class."""
802
super(CHKInventoryRepository, self).__init__(_format, a_bzrdir,
803
control_files, _commit_builder_class, _serializer)
608
KnitPackRepository.__init__(self, _format, a_bzrdir, control_files,
609
_commit_builder_class, _serializer)
610
# and now replace everything it did :)
804
611
index_transport = self._transport.clone('indices')
805
612
self._pack_collection = GCRepositoryPackCollection(self,
806
613
self._transport, index_transport,
1085
def reconcile_canonicalize_chks(self):
1086
"""Reconcile this repository to make sure all CHKs are in canonical
1089
from bzrlib.reconcile import PackReconciler
1090
reconciler = PackReconciler(self, thorough=True, canonicalize_chks=True)
1091
reconciler.reconcile()
1094
876
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1095
877
packer = GCCHKReconcilePacker(collection, packs, extension)
1096
878
return packer.pack(pb)
1098
def _canonicalize_chks_pack(self, collection, packs, extension, revs, pb):
1099
packer = GCCHKCanonicalizingPacker(collection, packs, extension, revs)
1100
return packer.pack(pb)
1102
880
def _get_source(self, to_format):
1103
881
"""Return a source for streaming from this repository."""
1104
if self._format._serializer == to_format._serializer:
882
if isinstance(to_format, remote.RemoteRepositoryFormat):
883
# Can't just check attributes on to_format with the current code,
885
to_format._ensure_real()
886
to_format = to_format._custom_format
887
if to_format.__class__ is self._format.__class__:
1105
888
# We must be exactly the same format, otherwise stuff like the chk
1106
# page layout might be different.
1107
# Actually, this test is just slightly looser than exact so that
1108
# CHK2 <-> 2a transfers will work.
889
# page layout might be different
1109
890
return GroupCHKStreamSource(self, to_format)
1110
891
return super(CHKInventoryRepository, self)._get_source(to_format)
1112
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1113
"""Find revisions with different parent lists in the revision object
1114
and in the index graph.
1116
:param revisions_iterator: None, or an iterator of (revid,
1117
Revision-or-None). This iterator controls the revisions checked.
1118
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1119
parents-in-revision).
1121
if not self.is_locked():
1122
raise AssertionError()
1124
if revisions_iterator is None:
1125
revisions_iterator = self._iter_revisions(None)
1126
for revid, revision in revisions_iterator:
1127
if revision is None:
1129
parent_map = vf.get_parent_map([(revid,)])
1130
parents_according_to_index = tuple(parent[-1] for parent in
1131
parent_map[(revid,)])
1132
parents_according_to_revision = tuple(revision.parent_ids)
1133
if parents_according_to_index != parents_according_to_revision:
1134
yield (revid, parents_according_to_index,
1135
parents_according_to_revision)
1137
def _check_for_inconsistent_revision_parents(self):
1138
inconsistencies = list(self._find_inconsistent_revision_parents())
1140
raise errors.BzrCheckError(
1141
"Revision index has inconsistent parents.")
1144
class GroupCHKStreamSource(StreamSource):
894
class GroupCHKStreamSource(KnitPackStreamSource):
1145
895
"""Used when both the source and target repo are GroupCHK repos."""
1147
897
def __init__(self, from_repository, to_format):
1234
988
self._chk_p_id_roots = None
1235
989
yield 'chk_bytes', _get_parent_id_basename_to_file_id_pages()
1237
def _get_text_stream(self):
1238
# Note: We know we don't have to handle adding root keys, because both
1239
# the source and target are the identical network name.
1240
text_stream = self.from_repository.texts.get_record_stream(
1241
self._text_keys, self._text_fetch_order, False)
1242
return ('texts', text_stream)
1244
991
def get_stream(self, search):
1245
def wrap_and_count(pb, rc, stream):
1246
"""Yield records from stream while showing progress."""
1248
for record in stream:
1249
if count == rc.STEP:
1251
pb.update('Estimate', rc.current, rc.max)
1256
992
revision_ids = search.get_keys()
1257
pb = ui.ui_factory.nested_progress_bar()
1258
rc = self._record_counter
1259
self._record_counter.setup(len(revision_ids))
1260
993
for stream_info in self._fetch_revision_texts(revision_ids):
1261
yield (stream_info[0],
1262
wrap_and_count(pb, rc, stream_info[1]))
1263
995
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
996
yield self._get_inventory_stream(self._revision_keys)
1264
997
# TODO: The keys to exclude might be part of the search recipe
1265
998
# For now, exclude all parents that are at the edge of ancestry, for
1266
999
# which we have inventories
1267
1000
from_repo = self.from_repository
1268
1001
parent_keys = from_repo._find_parent_keys_of_revisions(
1269
1002
self._revision_keys)
1270
self.from_repository.revisions.clear_cache()
1271
self.from_repository.signatures.clear_cache()
1272
# Clear the repo's get_parent_map cache too.
1273
self.from_repository._unstacked_provider.disable_cache()
1274
self.from_repository._unstacked_provider.enable_cache()
1275
s = self._get_inventory_stream(self._revision_keys)
1276
yield (s[0], wrap_and_count(pb, rc, s[1]))
1277
self.from_repository.inventories.clear_cache()
1278
1003
for stream_info in self._get_filtered_chk_streams(parent_keys):
1279
yield (stream_info[0], wrap_and_count(pb, rc, stream_info[1]))
1280
self.from_repository.chk_bytes.clear_cache()
1281
s = self._get_text_stream()
1282
yield (s[0], wrap_and_count(pb, rc, s[1]))
1283
self.from_repository.texts.clear_cache()
1284
pb.update('Done', rc.max, rc.max)
1005
yield self._get_text_stream()
1287
1007
def get_stream_for_missing_keys(self, missing_keys):
1288
1008
# missing keys can only occur when we are byte copying and not
1310
1030
yield stream_info
1313
class _InterestingKeyInfo(object):
1315
self.interesting_root_keys = set()
1316
self.interesting_pid_root_keys = set()
1317
self.uninteresting_root_keys = set()
1318
self.uninteresting_pid_root_keys = set()
1320
def all_interesting(self):
1321
return self.interesting_root_keys.union(self.interesting_pid_root_keys)
1323
def all_uninteresting(self):
1324
return self.uninteresting_root_keys.union(
1325
self.uninteresting_pid_root_keys)
1328
return self.all_interesting().union(self.all_uninteresting())
1331
def _build_interesting_key_sets(repo, inventory_ids, parent_only_inv_ids):
1332
result = _InterestingKeyInfo()
1333
for inv in repo.iter_inventories(inventory_ids, 'unordered'):
1334
root_key = inv.id_to_entry.key()
1335
pid_root_key = inv.parent_id_basename_to_file_id.key()
1336
if inv.revision_id in parent_only_inv_ids:
1337
result.uninteresting_root_keys.add(root_key)
1338
result.uninteresting_pid_root_keys.add(pid_root_key)
1340
result.interesting_root_keys.add(root_key)
1341
result.interesting_pid_root_keys.add(pid_root_key)
1345
def _filter_text_keys(interesting_nodes_iterable, text_keys, bytes_to_text_key):
1346
"""Iterate the result of iter_interesting_nodes, yielding the records
1347
and adding to text_keys.
1349
text_keys_update = text_keys.update
1350
for record, items in interesting_nodes_iterable:
1351
text_keys_update([bytes_to_text_key(b) for n,b in items])
1355
class RepositoryFormat2a(RepositoryFormatPack):
1356
"""A CHK repository that uses the bencode revision serializer."""
1033
class RepositoryFormatCHK1(RepositoryFormatPack):
1034
"""A hashed CHK+group compress pack repository."""
1358
1036
repository_class = CHKInventoryRepository
1359
1037
supports_external_lookups = True
1360
1038
supports_chks = True
1039
# For right now, setting this to True gives us InterModel1And2 rather
1040
# than InterDifferingSerializer
1361
1041
_commit_builder_class = PackRootCommitBuilder
1362
1042
rich_root_data = True
1363
_serializer = chk_serializer.chk_bencode_serializer
1043
_serializer = chk_serializer.chk_serializer_255_bigpage
1364
1044
_commit_inv_deltas = True
1365
1045
# What index classes to use
1366
1046
index_builder_class = BTreeBuilder
added
removed
bzrlib/repofmt/groupcompress_repo.py
