25
25
cached a change, it's good forever.
34
31
from loggerhead import util
35
from loggerhead.util import decorator
36
32
from loggerhead.lockfile import LockFile
39
34
with_lock = util.with_lock('_lock', 'ChangeCache')
42
class ChangeCache (object):
36
SQLITE_INTERFACE = os.environ.get('SQLITE_INTERFACE', 'sqlite')
38
if SQLITE_INTERFACE == 'pysqlite2':
39
from pysqlite2 import dbapi2
41
elif SQLITE_INTERFACE == 'sqlite':
42
import sqlite as dbapi2
46
_select_stmt = ("select data from revisiondata where revid = ?"
47
).replace('?', _param_marker)
48
_insert_stmt = ("insert into revisiondata (revid, data) "
49
"values (?, ?)").replace('?', _param_marker)
54
class FakeShelf(object):
55
def __init__(self, filename):
56
create_table = not os.path.exists(filename)
57
self.connection = dbapi2.connect(filename)
58
self.cursor = self.connection.cursor()
61
def _create_table(self):
63
"create table RevisionData "
64
"(revid binary primary key, data binary)")
65
self.connection.commit()
66
def _serialize(self, obj):
67
r = dbapi2.Binary(cPickle.dumps(obj, protocol=2))
69
def _unserialize(self, data):
70
return cPickle.loads(str(data))
72
self.cursor.execute(_select_stmt, (revid,))
73
filechange = self.cursor.fetchone()
74
if filechange is None:
77
return self._unserialize(filechange[0])
78
def add(self, revid_obj_pairs):
79
for (r, d) in revid_obj_pairs:
80
self.cursor.execute(_insert_stmt, (r, self._serialize(d)))
81
self.connection.commit()
84
class FileChangeCache(object):
44
85
def __init__(self, history, cache_path):
45
86
self.history = history
46
self.log = history.log
48
88
if not os.path.exists(cache_path):
49
89
os.mkdir(cache_path)
51
# keep a separate cache for the diffs, because they're very time-consuming to fetch.
52
self._changes_filename = os.path.join(cache_path, 'changes')
53
self._changes_diffs_filename = os.path.join(cache_path, 'changes-diffs')
55
# use a lockfile since the cache folder could be shared across different processes.
56
self._lock = LockFile(os.path.join(cache_path, 'lock'))
59
# this is fluff; don't slow down startup time with it.
62
self.log.info('Using change cache %s; %d/%d entries.' % (cache_path, s1, s2))
63
threading.Thread(target=log_sizes).start()
67
self.log.debug('Closing cache file.')
79
def get_changes(self, revid_list, get_diffs=False):
81
get a list of changes by their revision_ids. any changes missing
82
from the cache are fetched by calling L{History.get_change_uncached}
83
and inserted into the cache before returning.
86
cache = shelve.open(self._changes_diffs_filename, 'c', protocol=2)
88
cache = shelve.open(self._changes_filename, 'c', protocol=2)
91
self._changes_filename = os.path.join(cache_path, 'filechanges.sql')
93
# use a lockfile since the cache folder could be shared across
94
# different processes.
95
self._lock = LockFile(os.path.join(cache_path, 'filechange-lock'))
98
def get_file_changes(self, entries):
93
for revid in revid_list:
94
# if the revid is in unicode, use the utf-8 encoding as the key
95
srevid = util.to_utf8(revid)
98
out.append(cache[srevid])
101
missing_entry_indices = []
102
cache = FakeShelf(self._changes_filename)
103
for entry in entries:
104
changes = cache.get(entry.revid)
105
if changes is not None:
100
#self.log.debug('Entry cache miss: %r' % (revid,))
108
missing_entries.append(entry)
109
missing_entry_indices.append(len(out))
102
fetch_list.append(revid)
103
sfetch_list.append(srevid)
105
if len(fetch_list) > 0:
106
# some revisions weren't in the cache; fetch them
107
changes = self.history.get_changes_uncached(fetch_list, get_diffs)
110
for i in xrange(len(revid_list)):
112
cache[sfetch_list.pop(0)] = out[i] = changes.pop(0)
112
missing_changes = self.history.get_file_changes_uncached(missing_entries)
113
revid_changes_pairs = []
114
for i, entry, changes in zip(
115
missing_entry_indices, missing_entries, missing_changes):
116
revid_changes_pairs.append((entry.revid, changes))
118
cache.add(revid_changes_pairs)
118
def full(self, get_diffs=False):
120
cache = shelve.open(self._changes_diffs_filename, 'c', protocol=2)
122
cache = shelve.open(self._changes_filename, 'c', protocol=2)
124
return (len(cache) >= len(self.history.get_revision_history())) and (util.to_utf8(self.history.last_revid) in cache)
130
cache = shelve.open(self._changes_filename, 'c', protocol=2)
133
cache = shelve.open(self._changes_diffs_filename, 'c', protocol=2)
138
def check_rebuild(self, max_time=3600):
140
check if we need to fill in any missing pieces of the cache. pull in
141
any missing changes, but don't work any longer than C{max_time}
144
if self.closed() or self.full():
147
self.log.info('Building revision cache...')
148
start_time = time.time()
149
last_update = time.time()
152
work = list(self.history.get_revision_history())
154
for i in xrange(0, len(work), jump):
156
# must call into history so we grab the branch lock (otherwise, lock inversion)
157
self.history.get_changes(r)
163
if now - start_time > max_time:
164
self.log.info('Cache rebuilding will pause for now.')
167
if now - last_update > 60:
168
self.log.info('Revision cache rebuilding continues: %d/%d' % (min(count, len(work)), len(work)))
169
last_update = time.time()
171
# give someone else a chance at the lock
173
self.log.info('Revision cache rebuild completed.')
added
removed
loggerhead/changecache.py
