20
a cache for chewed-up 'file change' data structures, which are basically just
21
a different way of storing a revision delta. the cache improves lookup times
22
10x over bazaar's xml revision structure, though, so currently still worth
20
a cache for chewed-up "change" data structures, which are basically just a
21
different way of storing a revision. the cache improves lookup times 10x
22
over bazaar's xml revision structure, though, so currently still worth doing.
25
24
once a revision is committed in bazaar, it never changes, so once we have
26
25
cached a change, it's good forever.
36
from sqlite3 import dbapi2
32
from loggerhead import util
33
from loggerhead.lockfile import LockFile
36
with_lock = util.with_lock('_lock', 'ChangeCache')
38
SQLITE_INTERFACE = os.environ.get('SQLITE_INTERFACE', 'sqlite')
40
if SQLITE_INTERFACE == 'pysqlite2':
38
41
from pysqlite2 import dbapi2
40
# We take an optimistic approach to concurrency here: we might do work twice
41
# in the case of races, but not crash or corrupt data.
43
def safe_init_db(filename, init_sql):
44
# To avoid races around creating the database, we create the db in
45
# a temporary file and rename it into the ultimate location.
46
fd, temp_path = tempfile.mkstemp(dir=os.path.dirname(filename))
48
con = dbapi2.connect(temp_path)
53
os.rename(temp_path, filename)
43
elif SQLITE_INTERFACE == 'sqlite':
44
import sqlite as dbapi2
47
raise AssertionError("bad sqlite interface %r!?"%SQLITE_INTERFACE)
49
_select_stmt = ("select data from revisiondata where revid = ?"
50
).replace('?', _param_marker)
51
_insert_stmt = ("insert into revisiondata (revid, data) "
52
"values (?, ?)").replace('?', _param_marker)
53
_update_stmt = ("update revisiondata set data = ? where revid = ?"
54
).replace('?', _param_marker)
55
59
class FakeShelf(object):
57
60
def __init__(self, filename):
58
61
create_table = not os.path.exists(filename)
61
filename, "create table RevisionData "
62
"(revid binary primary key, data binary)")
63
62
self.connection = dbapi2.connect(filename)
64
63
self.cursor = self.connection.cursor()
66
def _create_table(self, filename):
67
con = dbapi2.connect(filename)
66
def _create_table(self):
70
68
"create table RevisionData "
71
69
"(revid binary primary key, data binary)")
70
self.connection.commit()
75
71
def _serialize(self, obj):
76
return dbapi2.Binary(cPickle.dumps(obj, protocol=2))
72
r = dbapi2.Binary(cPickle.dumps(obj, protocol=2))
78
74
def _unserialize(self, data):
79
75
return cPickle.loads(str(data))
81
76
def get(self, revid):
83
"select data from revisiondata where revid = ?", (revid, ))
77
self.cursor.execute(_select_stmt, (revid,))
84
78
filechange = self.cursor.fetchone()
85
79
if filechange is None:
88
82
return self._unserialize(filechange[0])
90
def add(self, revid, object):
93
"insert into revisiondata (revid, data) values (?, ?)",
94
(revid, self._serialize(object)))
95
self.connection.commit()
96
except dbapi2.IntegrityError:
97
# If another thread or process attempted to set the same key, we
98
# assume it set it to the same value and carry on with our day.
83
def add(self, revid_obj_pairs, commit=True):
84
for (r, d) in revid_obj_pairs:
85
self.cursor.execute(_insert_stmt, (r, self._serialize(d)))
87
self.connection.commit()
88
def update(self, revid_obj_pairs, commit=True):
89
for (r, d) in revid_obj_pairs:
90
self.cursor.execute(_update_stmt, (self._serialize(d), r))
92
self.connection.commit()
95
"select count(*) from revisiondata")
96
return self.cursor.fetchone()[0]
97
def close(self, commit=False):
99
self.connection.commit()
100
self.connection.close()
102
class ChangeCache (object):
104
def __init__(self, history, cache_path):
105
self.history = history
106
self.log = history.log
108
if not os.path.exists(cache_path):
111
self._changes_filename = os.path.join(cache_path, 'changes.sql')
113
# use a lockfile since the cache folder could be shared across different processes.
114
self._lock = LockFile(os.path.join(cache_path, 'lock'))
117
## # this is fluff; don't slow down startup time with it.
118
## # but it is racy in tests :(
120
## self.log.info('Using change cache %s; %d entries.' % (cache_path, self.size()))
121
## threading.Thread(target=log_sizes).start()
124
return FakeShelf(self._changes_filename)
128
self.log.debug('Closing cache file.')
140
def get_changes(self, revid_list):
142
get a list of changes by their revision_ids. any changes missing
143
from the cache are fetched by calling L{History.get_change_uncached}
144
and inserted into the cache before returning.
148
missing_revid_indices = []
149
cache = self._cache()
150
for revid in revid_list:
151
entry = cache.get(revid)
152
if entry is not None:
155
missing_revids.append(revid)
156
missing_revid_indices.append(len(out))
159
missing_entries = self.history.get_changes_uncached(missing_revids)
160
missing_entry_dict = {}
161
for entry in missing_entries:
162
missing_entry_dict[entry.revid] = entry
163
revid_entry_pairs = []
164
for i, revid in zip(missing_revid_indices, missing_revids):
165
out[i] = entry = missing_entry_dict.get(revid)
166
if entry is not None:
167
revid_entry_pairs.append((revid, entry))
168
cache.add(revid_entry_pairs)
169
return filter(None, out)
173
cache = self._cache()
174
last_revid = util.to_utf8(self.history.last_revid)
175
revision_history = self.history.get_revision_history()
176
return (cache.count() >= len(revision_history)
177
and cache.get(last_revid) is not None)
181
return self._cache().count()
183
def check_rebuild(self, max_time=3600):
185
check if we need to fill in any missing pieces of the cache. pull in
186
any missing changes, but don't work any longer than C{max_time}
189
if self.closed() or self.full():
192
self.log.info('Building revision cache...')
193
start_time = time.time()
194
last_update = time.time()
197
work = list(self.history.get_revision_history())
199
for i in xrange(0, len(work), jump):
201
# must call into history so we grab the branch lock (otherwise, lock inversion)
202
self.history.get_changes(r)
208
if now - start_time > max_time:
209
self.log.info('Cache rebuilding will pause for now.')
212
if now - last_update > 60:
213
self.log.info('Revision cache rebuilding continues: %d/%d' % (min(count, len(work)), len(work)))
214
last_update = time.time()
216
# give someone else a chance at the lock
218
self.log.info('Revision cache rebuild completed.')
102
221
class FileChangeCache(object):
104
def __init__(self, cache_path):
222
def __init__(self, history, cache_path):
223
self.history = history
106
225
if not os.path.exists(cache_path):
107
226
os.mkdir(cache_path)
109
228
self._changes_filename = os.path.join(cache_path, 'filechanges.sql')
111
def get_file_changes(self, entry):
230
# use a lockfile since the cache folder could be shared across
231
# different processes.
232
self._lock = LockFile(os.path.join(cache_path, 'filechange-lock'))
235
def get_file_changes(self, entries):
238
missing_entry_indices = []
112
239
cache = FakeShelf(self._changes_filename)
113
changes = cache.get(entry.revid)
115
changes = self.history.get_file_changes_uncached(entry)
116
cache.add(entry.revid, changes)
120
class RevInfoDiskCache(object):
121
"""Like `RevInfoMemoryCache` but backed in a sqlite DB."""
123
def __init__(self, cache_path):
124
if not os.path.exists(cache_path):
126
filename = os.path.join(cache_path, 'revinfo.sql')
127
create_table = not os.path.exists(filename)
130
filename, "create table Data "
131
"(key binary primary key, revid binary, data binary)")
132
self.connection = dbapi2.connect(filename)
133
self.cursor = self.connection.cursor()
135
def get(self, key, revid):
137
"select revid, data from data where key = ?", (dbapi2.Binary(key),))
138
row = self.cursor.fetchone()
141
elif str(row[0]) != revid:
144
return marshal.loads(zlib.decompress(row[1]))
146
def set(self, key, revid, data):
149
'delete from data where key = ?', (dbapi2.Binary(key), ))
150
blob = zlib.compress(marshal.dumps(data))
152
"insert into data (key, revid, data) values (?, ?, ?)",
153
map(dbapi2.Binary, [key, revid, blob]))
154
self.connection.commit()
155
except dbapi2.IntegrityError:
156
# If another thread or process attempted to set the same key, we
157
# don't care too much -- it's only a cache after all!
240
for entry in entries:
241
changes = cache.get(entry.revid)
242
if changes is not None:
245
missing_entries.append(entry)
246
missing_entry_indices.append(len(out))
249
missing_changes = self.history.get_file_changes_uncached(missing_entries)
250
revid_changes_pairs = []
251
for i, entry, changes in zip(
252
missing_entry_indices, missing_entries, missing_changes):
253
revid_changes_pairs.append((entry.revid, changes))
255
cache.add(revid_changes_pairs)
added
removed
loggerhead/changecache.py
