273
275
bytes = apply_delta_to_source(self._content, content_start, end)
278
def set_chunked_content(self, content_chunks, length):
279
"""Set the content of this block to the given chunks."""
280
# If we have lots of short lines, it is may be more efficient to join
281
# the content ahead of time. If the content is <10MiB, we don't really
282
# care about the extra memory consumption, so we can just pack it and
283
# be done. However, timing showed 18s => 17.9s for repacking 1k revs of
284
# mysql, which is below the noise margin
285
self._content_length = length
286
self._content_chunks = content_chunks
288
self._z_content = None
276
290
def set_content(self, content):
277
291
"""Set the content of this block."""
278
292
self._content_length = len(content)
279
293
self._content = content
280
294
self._z_content = None
296
def _create_z_content_using_lzma(self):
297
if self._content_chunks is not None:
298
self._content = ''.join(self._content_chunks)
299
self._content_chunks = None
300
if self._content is None:
301
raise AssertionError('Nothing to compress')
302
self._z_content = pylzma.compress(self._content)
303
self._z_content_length = len(self._z_content)
305
def _create_z_content_from_chunks(self):
306
compressor = zlib.compressobj(zlib.Z_DEFAULT_COMPRESSION)
307
compressed_chunks = map(compressor.compress, self._content_chunks)
308
compressed_chunks.append(compressor.flush())
309
self._z_content = ''.join(compressed_chunks)
310
self._z_content_length = len(self._z_content)
312
def _create_z_content(self):
313
if self._z_content is not None:
316
self._create_z_content_using_lzma()
318
if self._content_chunks is not None:
319
self._create_z_content_from_chunks()
321
self._z_content = zlib.compress(self._content)
322
self._z_content_length = len(self._z_content)
282
324
def to_bytes(self):
283
325
"""Encode the information into a byte stream."""
284
compress = zlib.compress
286
compress = pylzma.compress
287
if self._z_content is None:
288
if self._content is None:
289
raise AssertionError('Nothing to compress')
290
self._z_content = compress(self._content)
291
self._z_content_length = len(self._z_content)
326
self._create_z_content()
293
328
header = self.GCB_LZ_HEADER
503
553
# time (self._block._content) is a little expensive.
504
554
self._block._ensure_content(self._last_byte)
506
def _check_rebuild_block(self):
556
def _check_rebuild_action(self):
507
557
"""Check to see if our block should be repacked."""
508
558
total_bytes_used = 0
509
559
last_byte_used = 0
510
560
for factory in self._factories:
511
561
total_bytes_used += factory._end - factory._start
512
last_byte_used = max(last_byte_used, factory._end)
513
# If we are using most of the bytes from the block, we have nothing
514
# else to check (currently more that 1/2)
562
if last_byte_used < factory._end:
563
last_byte_used = factory._end
564
# If we are using more than half of the bytes from the block, we have
565
# nothing else to check
515
566
if total_bytes_used * 2 >= self._block._content_length:
517
# Can we just strip off the trailing bytes? If we are going to be
518
# transmitting more than 50% of the front of the content, go ahead
567
return None, last_byte_used, total_bytes_used
568
# We are using less than 50% of the content. Is the content we are
569
# using at the beginning of the block? If so, we can just trim the
570
# tail, rather than rebuilding from scratch.
519
571
if total_bytes_used * 2 > last_byte_used:
520
self._trim_block(last_byte_used)
572
return 'trim', last_byte_used, total_bytes_used
523
574
# We are using a small amount of the data, and it isn't just packed
524
575
# nicely at the front, so rebuild the content.
531
582
# expanding many deltas into fulltexts, as well.
532
583
# If we build a cheap enough 'strip', then we could try a strip,
533
584
# if that expands the content, we then rebuild.
534
self._rebuild_block()
585
return 'rebuild', last_byte_used, total_bytes_used
587
def check_is_well_utilized(self):
588
"""Is the current block considered 'well utilized'?
590
This heuristic asks if the current block considers itself to be a fully
591
developed group, rather than just a loose collection of data.
593
if len(self._factories) == 1:
594
# A block of length 1 could be improved by combining with other
595
# groups - don't look deeper. Even larger than max size groups
596
# could compress well with adjacent versions of the same thing.
598
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
599
block_size = self._block._content_length
600
if total_bytes_used < block_size * self._max_cut_fraction:
601
# This block wants to trim itself small enough that we want to
602
# consider it under-utilized.
604
# TODO: This code is meant to be the twin of _insert_record_stream's
605
# 'start_new_block' logic. It would probably be better to factor
606
# out that logic into a shared location, so that it stays
608
# We currently assume a block is properly utilized whenever it is >75%
609
# of the size of a 'full' block. In normal operation, a block is
610
# considered full when it hits 4MB of same-file content. So any block
611
# >3MB is 'full enough'.
612
# The only time this isn't true is when a given block has large-object
613
# content. (a single file >4MB, etc.)
614
# Under these circumstances, we allow a block to grow to
615
# 2 x largest_content. Which means that if a given block had a large
616
# object, it may actually be under-utilized. However, given that this
617
# is 'pack-on-the-fly' it is probably reasonable to not repack large
618
# content blobs on-the-fly. Note that because we return False for all
619
# 1-item blobs, we will repack them; we may wish to reevaluate our
620
# treatment of large object blobs in the future.
621
if block_size >= self._full_enough_block_size:
623
# If a block is <3MB, it still may be considered 'full' if it contains
624
# mixed content. The current rule is 2MB of mixed content is considered
625
# full. So check to see if this block contains mixed content, and
626
# set the threshold appropriately.
628
for factory in self._factories:
629
prefix = factory.key[:-1]
630
if common_prefix is None:
631
common_prefix = prefix
632
elif prefix != common_prefix:
633
# Mixed content, check the size appropriately
634
if block_size >= self._full_enough_mixed_block_size:
637
# The content failed both the mixed check and the single-content check
638
# so obviously it is not fully utilized
639
# TODO: there is one other constraint that isn't being checked
640
# namely, that the entries in the block are in the appropriate
641
# order. For example, you could insert the entries in exactly
642
# reverse groupcompress order, and we would think that is ok.
643
# (all the right objects are in one group, and it is fully
644
# utilized, etc.) For now, we assume that case is rare,
645
# especially since we should always fetch in 'groupcompress'
649
def _check_rebuild_block(self):
650
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
654
self._trim_block(last_byte_used)
655
elif action == 'rebuild':
656
self._rebuild_block()
658
raise ValueError('unknown rebuild action: %r' % (action,))
536
660
def _wire_bytes(self):
537
661
"""Return a byte stream suitable for transmitting over the wire."""
925
1057
versioned_files.stream.close()
1060
class _BatchingBlockFetcher(object):
1061
"""Fetch group compress blocks in batches.
1063
:ivar total_bytes: int of expected number of bytes needed to fetch the
1064
currently pending batch.
1067
def __init__(self, gcvf, locations):
1069
self.locations = locations
1071
self.batch_memos = {}
1072
self.memos_to_get = []
1073
self.total_bytes = 0
1074
self.last_read_memo = None
1077
def add_key(self, key):
1078
"""Add another to key to fetch.
1080
:return: The estimated number of bytes needed to fetch the batch so
1083
self.keys.append(key)
1084
index_memo, _, _, _ = self.locations[key]
1085
read_memo = index_memo[0:3]
1086
# Three possibilities for this read_memo:
1087
# - it's already part of this batch; or
1088
# - it's not yet part of this batch, but is already cached; or
1089
# - it's not yet part of this batch and will need to be fetched.
1090
if read_memo in self.batch_memos:
1091
# This read memo is already in this batch.
1092
return self.total_bytes
1094
cached_block = self.gcvf._group_cache[read_memo]
1096
# This read memo is new to this batch, and the data isn't cached
1098
self.batch_memos[read_memo] = None
1099
self.memos_to_get.append(read_memo)
1100
byte_length = read_memo[2]
1101
self.total_bytes += byte_length
1103
# This read memo is new to this batch, but cached.
1104
# Keep a reference to the cached block in batch_memos because it's
1105
# certain that we'll use it when this batch is processed, but
1106
# there's a risk that it would fall out of _group_cache between now
1108
self.batch_memos[read_memo] = cached_block
1109
return self.total_bytes
1111
def _flush_manager(self):
1112
if self.manager is not None:
1113
for factory in self.manager.get_record_stream():
1116
self.last_read_memo = None
1118
def yield_factories(self, full_flush=False):
1119
"""Yield factories for keys added since the last yield. They will be
1120
returned in the order they were added via add_key.
1122
:param full_flush: by default, some results may not be returned in case
1123
they can be part of the next batch. If full_flush is True, then
1124
all results are returned.
1126
if self.manager is None and not self.keys:
1128
# Fetch all memos in this batch.
1129
blocks = self.gcvf._get_blocks(self.memos_to_get)
1130
# Turn blocks into factories and yield them.
1131
memos_to_get_stack = list(self.memos_to_get)
1132
memos_to_get_stack.reverse()
1133
for key in self.keys:
1134
index_memo, _, parents, _ = self.locations[key]
1135
read_memo = index_memo[:3]
1136
if self.last_read_memo != read_memo:
1137
# We are starting a new block. If we have a
1138
# manager, we have found everything that fits for
1139
# now, so yield records
1140
for factory in self._flush_manager():
1142
# Now start a new manager.
1143
if memos_to_get_stack and memos_to_get_stack[-1] == read_memo:
1144
# The next block from _get_blocks will be the block we
1146
block_read_memo, block = blocks.next()
1147
if block_read_memo != read_memo:
1148
raise AssertionError(
1149
"block_read_memo out of sync with read_memo"
1150
"(%r != %r)" % (block_read_memo, read_memo))
1151
self.batch_memos[read_memo] = block
1152
memos_to_get_stack.pop()
1154
block = self.batch_memos[read_memo]
1155
self.manager = _LazyGroupContentManager(block)
1156
self.last_read_memo = read_memo
1157
start, end = index_memo[3:5]
1158
self.manager.add_factory(key, parents, start, end)
1160
for factory in self._flush_manager():
1163
self.batch_memos.clear()
1164
del self.memos_to_get[:]
1165
self.total_bytes = 0
928
1168
class GroupCompressVersionedFiles(VersionedFiles):
929
1169
"""A group-compress based VersionedFiles implementation."""
931
def __init__(self, index, access, delta=True):
1171
def __init__(self, index, access, delta=True, _unadded_refs=None):
932
1172
"""Create a GroupCompressVersionedFiles object.
934
1174
:param index: The index object storing access and graph data.
935
1175
:param access: The access object storing raw data.
936
1176
:param delta: Whether to delta compress or just entropy compress.
1177
:param _unadded_refs: private parameter, don't use.
938
1179
self._index = index
939
1180
self._access = access
940
1181
self._delta = delta
941
self._unadded_refs = {}
1182
if _unadded_refs is None:
1184
self._unadded_refs = _unadded_refs
942
1185
self._group_cache = LRUSizeCache(max_size=50*1024*1024)
943
1186
self._fallback_vfs = []
1188
def without_fallbacks(self):
1189
"""Return a clone of this object without any fallbacks configured."""
1190
return GroupCompressVersionedFiles(self._index, self._access,
1191
self._delta, _unadded_refs=dict(self._unadded_refs))
945
1193
def add_lines(self, key, parents, lines, parent_texts=None,
946
1194
left_matching_blocks=None, nostore_sha=None, random_id=False,
947
1195
check_content=True):
992
1240
nostore_sha=nostore_sha))[0]
993
1241
return sha1, length, None
1243
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
1244
"""See VersionedFiles._add_text()."""
1245
self._index._check_write_ok()
1246
self._check_add(key, None, random_id, check_content=False)
1247
if text.__class__ is not str:
1248
raise errors.BzrBadParameterUnicode("text")
1250
# The caller might pass None if there is no graph data, but kndx
1251
# indexes can't directly store that, so we give them
1252
# an empty tuple instead.
1254
# double handling for now. Make it work until then.
1256
record = FulltextContentFactory(key, parents, None, text)
1257
sha1 = list(self._insert_record_stream([record], random_id=random_id,
1258
nostore_sha=nostore_sha))[0]
1259
return sha1, length, None
995
1261
def add_fallback_versioned_files(self, a_versioned_files):
996
1262
"""Add a source of texts for texts not present in this knit.
1002
1268
def annotate(self, key):
1003
1269
"""See VersionedFiles.annotate."""
1005
parent_map = self.get_parent_map([key])
1007
raise errors.RevisionNotPresent(key, self)
1008
if parent_map[key] is not None:
1009
search = graph._make_breadth_first_searcher([key])
1013
present, ghosts = search.next_with_ghosts()
1014
except StopIteration:
1016
keys.update(present)
1017
parent_map = self.get_parent_map(keys)
1020
parent_map = {key:()}
1021
head_cache = _mod_graph.FrozenHeadsCache(graph)
1023
reannotate = annotate.reannotate
1024
for record in self.get_record_stream(keys, 'topological', True):
1026
chunks = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1027
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
1028
parent_cache[key] = list(
1029
reannotate(parent_lines, chunks, key, None, head_cache))
1030
return parent_cache[key]
1032
def check(self, progress_bar=None):
1270
ann = annotate.Annotator(self)
1271
return ann.annotate_flat(key)
1273
def get_annotator(self):
1274
return annotate.Annotator(self)
1276
def check(self, progress_bar=None, keys=None):
1033
1277
"""See VersionedFiles.check()."""
1035
for record in self.get_record_stream(keys, 'unordered', True):
1036
record.get_bytes_as('fulltext')
1280
for record in self.get_record_stream(keys, 'unordered', True):
1281
record.get_bytes_as('fulltext')
1283
return self.get_record_stream(keys, 'unordered', True)
1038
1285
def _check_add(self, key, lines, random_id, check_content):
1039
1286
"""check that version_id and lines are safe to add."""
1082
1345
missing.difference_update(set(new_result))
1083
1346
return result, source_results
1085
def _get_block(self, index_memo):
1086
read_memo = index_memo[0:3]
1089
block = self._group_cache[read_memo]
1092
zdata = self._access.get_raw_records([read_memo]).next()
1093
# decompress - whole thing - this is not a bug, as it
1094
# permits caching. We might want to store the partially
1095
# decompresed group and decompress object, so that recent
1096
# texts are not penalised by big groups.
1097
block = GroupCompressBlock.from_bytes(zdata)
1098
self._group_cache[read_memo] = block
1100
# print len(zdata), len(plain)
1101
# parse - requires split_lines, better to have byte offsets
1102
# here (but not by much - we only split the region for the
1103
# recipe, and we often want to end up with lines anyway.
1348
def _get_blocks(self, read_memos):
1349
"""Get GroupCompressBlocks for the given read_memos.
1351
:returns: a series of (read_memo, block) pairs, in the order they were
1355
for read_memo in read_memos:
1357
block = self._group_cache[read_memo]
1361
cached[read_memo] = block
1363
not_cached_seen = set()
1364
for read_memo in read_memos:
1365
if read_memo in cached:
1366
# Don't fetch what we already have
1368
if read_memo in not_cached_seen:
1369
# Don't try to fetch the same data twice
1371
not_cached.append(read_memo)
1372
not_cached_seen.add(read_memo)
1373
raw_records = self._access.get_raw_records(not_cached)
1374
for read_memo in read_memos:
1376
yield read_memo, cached[read_memo]
1378
# Read the block, and cache it.
1379
zdata = raw_records.next()
1380
block = GroupCompressBlock.from_bytes(zdata)
1381
self._group_cache[read_memo] = block
1382
cached[read_memo] = block
1383
yield read_memo, block
1106
1385
def get_missing_compression_parent_keys(self):
1107
1386
"""Return the keys of missing compression parents.
1273
1552
unadded_keys, source_result)
1274
1553
for key in missing:
1275
1554
yield AbsentContentFactory(key)
1277
last_read_memo = None
1278
# TODO: This works fairly well at batching up existing groups into a
1279
# streamable format, and possibly allowing for taking one big
1280
# group and splitting it when it isn't fully utilized.
1281
# However, it doesn't allow us to find under-utilized groups and
1282
# combine them into a bigger group on the fly.
1283
# (Consider the issue with how chk_map inserts texts
1284
# one-at-a-time.) This could be done at insert_record_stream()
1285
# time, but it probably would decrease the number of
1286
# bytes-on-the-wire for fetch.
1555
# Batch up as many keys as we can until either:
1556
# - we encounter an unadded ref, or
1557
# - we run out of keys, or
1558
# - the total bytes to retrieve for this batch > BATCH_SIZE
1559
batcher = _BatchingBlockFetcher(self, locations)
1287
1560
for source, keys in source_keys:
1288
1561
if source is self:
1289
1562
for key in keys:
1290
1563
if key in self._unadded_refs:
1291
if manager is not None:
1292
for factory in manager.get_record_stream():
1294
last_read_memo = manager = None
1564
# Flush batch, then yield unadded ref from
1566
for factory in batcher.yield_factories(full_flush=True):
1295
1568
bytes, sha1 = self._compressor.extract(key)
1296
1569
parents = self._unadded_refs[key]
1297
1570
yield FulltextContentFactory(key, parents, sha1, bytes)
1299
index_memo, _, parents, (method, _) = locations[key]
1300
read_memo = index_memo[0:3]
1301
if last_read_memo != read_memo:
1302
# We are starting a new block. If we have a
1303
# manager, we have found everything that fits for
1304
# now, so yield records
1305
if manager is not None:
1306
for factory in manager.get_record_stream():
1308
# Now start a new manager
1309
block = self._get_block(index_memo)
1310
manager = _LazyGroupContentManager(block)
1311
last_read_memo = read_memo
1312
start, end = index_memo[3:5]
1313
manager.add_factory(key, parents, start, end)
1572
if batcher.add_key(key) > BATCH_SIZE:
1573
# Ok, this batch is big enough. Yield some results.
1574
for factory in batcher.yield_factories():
1315
if manager is not None:
1316
for factory in manager.get_record_stream():
1318
last_read_memo = manager = None
1577
for factory in batcher.yield_factories(full_flush=True):
1319
1579
for record in source.get_record_stream(keys, ordering,
1320
1580
include_delta_closure):
1322
if manager is not None:
1323
for factory in manager.get_record_stream():
1582
for factory in batcher.yield_factories(full_flush=True):
1326
1585
def get_sha1s(self, keys):
1327
1586
"""See VersionedFiles.get_sha1s()."""
1413
1673
if reuse_blocks:
1414
1674
# If the reuse_blocks flag is set, check to see if we can just
1415
1675
# copy a groupcompress block as-is.
1676
# We only check on the first record (groupcompress-block) not
1677
# on all of the (groupcompress-block-ref) entries.
1678
# The reuse_this_block flag is then kept for as long as
1679
if record.storage_kind == 'groupcompress-block':
1680
# Check to see if we really want to re-use this block
1681
insert_manager = record._manager
1682
reuse_this_block = insert_manager.check_is_well_utilized()
1684
reuse_this_block = False
1685
if reuse_this_block:
1686
# We still want to reuse this block
1416
1687
if record.storage_kind == 'groupcompress-block':
1417
1688
# Insert the raw block into the target repo
1418
1689
insert_manager = record._manager
1419
insert_manager._check_rebuild_block()
1420
1690
bytes = record._manager._block.to_bytes()
1421
1691
_, start, length = self._access.add_raw_records(
1422
1692
[(None, len(bytes))], bytes)[0]
1554
1830
:param add_callback: If not None, allow additions to the index and call
1555
1831
this callback with a list of added GraphIndex nodes:
1556
1832
[(node, value, node_refs), ...]
1833
:param track_external_parent_refs: As keys are added, keep track of the
1834
keys they reference, so that we can query get_missing_parents(),
1836
:param inconsistency_fatal: When asked to add records that are already
1837
present, and the details are inconsistent with the existing
1838
record, raise an exception instead of warning (and skipping the
1558
1841
self._add_callback = add_callback
1559
1842
self._graph_index = graph_index
1560
1843
self._parents = parents
1561
1844
self.has_graph = parents
1562
1845
self._is_locked = is_locked
1846
self._inconsistency_fatal = inconsistency_fatal
1847
if track_external_parent_refs:
1848
self._key_dependencies = knit._KeyRefs(
1849
track_new_keys=track_new_keys)
1851
self._key_dependencies = None
1564
1853
def add_records(self, records, random_id=False):
1565
1854
"""Add multiple records to the index.
1715
2031
delta_end = int(bits[3])
1716
2032
return node[0], start, stop, basis_end, delta_end
2034
def scan_unvalidated_index(self, graph_index):
2035
"""Inform this _GCGraphIndex that there is an unvalidated index.
2037
This allows this _GCGraphIndex to keep track of any missing
2038
compression parents we may want to have filled in to make those
2039
indices valid. It also allows _GCGraphIndex to track any new keys.
2041
:param graph_index: A GraphIndex
2043
key_dependencies = self._key_dependencies
2044
if key_dependencies is None:
2046
for node in graph_index.iter_all_entries():
2047
# Add parent refs from graph_index (and discard parent refs
2048
# that the graph_index has).
2049
key_dependencies.add_references(node[1], node[3][0])
1719
2052
from bzrlib._groupcompress_py import (
added
removed
bzrlib/groupcompress.py
