bzr branch
http://gegoxaren.bato24.eu/bzr/brz/remove-bazaar
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
1 |
# Copyright (C) 2009 Canonical Ltd
|
2 |
#
|
|
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.
|
|
7 |
#
|
|
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.
|
|
12 |
#
|
|
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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
16 |
||
17 |
"""A manager of caches."""
|
|
18 |
||
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
19 |
import atexit |
20 |
import os |
|
21 |
import shutil |
|
22 |
import tempfile |
|
23 |
import weakref |
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
24 |
|
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
25 |
from bzrlib import lru_cache, trace |
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
26 |
from bzrlib.plugins.fastimport import ( |
27 |
branch_mapper, |
|
28 |
)
|
|
|
0.123.3
by Jelmer Vernooij
Fix some imports. |
29 |
from fastimport.helpers import ( |
30 |
single_plural, |
|
31 |
)
|
|
|
0.64.279
by Jelmer Vernooij
Merge split of python-fastimport into a separate package. |
32 |
from fastimport.reftracker import ( |
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
33 |
RefTracker, |
34 |
)
|
|
35 |
||
36 |
||
37 |
class _Cleanup(object): |
|
38 |
"""This class makes sure we clean up when CacheManager goes away. |
|
39 |
||
40 |
We use a helper class to ensure that we are never in a refcycle.
|
|
41 |
"""
|
|
42 |
||
43 |
def __init__(self, disk_blobs): |
|
44 |
self.disk_blobs = disk_blobs |
|
45 |
self.tempdir = None |
|
46 |
self.small_blobs = None |
|
47 |
||
48 |
def __del__(self): |
|
49 |
self.finalize() |
|
50 |
||
51 |
def finalize(self): |
|
52 |
if self.disk_blobs is not None: |
|
53 |
for info in self.disk_blobs.itervalues(): |
|
54 |
if info[-1] is not None: |
|
55 |
os.unlink(info[-1]) |
|
56 |
self.disk_blobs = None |
|
57 |
if self.small_blobs is not None: |
|
58 |
self.small_blobs.close() |
|
59 |
self.small_blobs = None |
|
60 |
if self.tempdir is not None: |
|
61 |
shutil.rmtree(self.tempdir) |
|
62 |
||
63 |
||
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
64 |
class CacheManager(object): |
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
65 |
|
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
66 |
_small_blob_threshold = 25*1024 |
67 |
_sticky_cache_size = 300*1024*1024 |
|
68 |
_sticky_flushed_size = 100*1024*1024 |
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
69 |
|
|
0.83.1
by Ian Clatworthy
head tracking tests and fix |
70 |
def __init__(self, info=None, verbose=False, inventory_cache_size=10): |
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
71 |
"""Create a manager of caches. |
72 |
||
73 |
:param info: a ConfigObj holding the output from
|
|
74 |
the --info processor, or None if no hints are available
|
|
75 |
"""
|
|
76 |
self.verbose = verbose |
|
77 |
||
78 |
# dataref -> data. datref is either :mark or the sha-1.
|
|
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
79 |
# Sticky blobs are referenced more than once, and are saved until their
|
80 |
# refcount goes to 0
|
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
81 |
self._blobs = {} |
82 |
self._sticky_blobs = {} |
|
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
83 |
self._sticky_memory_bytes = 0 |
84 |
# if we overflow our memory cache, then we will dump large blobs to
|
|
85 |
# disk in this directory
|
|
86 |
self._tempdir = None |
|
87 |
# id => (offset, n_bytes, fname)
|
|
88 |
# if fname is None, then the content is stored in the small file
|
|
89 |
self._disk_blobs = {} |
|
90 |
self._cleanup = _Cleanup(self._disk_blobs) |
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
91 |
|
92 |
# revision-id -> Inventory cache
|
|
93 |
# these are large and we probably don't need too many as
|
|
94 |
# most parents are recent in history
|
|
95 |
self.inventories = lru_cache.LRUCache(inventory_cache_size) |
|
96 |
||
97 |
# import commmit-ids -> revision-id lookup table
|
|
98 |
# we need to keep all of these but they are small
|
|
99 |
self.revision_ids = {} |
|
100 |
||
|
0.64.159
by Ian Clatworthy
make the file-id cache optional and branch-ref aware |
101 |
# (path, branch_ref) -> file-ids - as generated.
|
102 |
# (Use store_file_id/fetch_fileid methods rather than direct access.)
|
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
103 |
|
104 |
# Work out the blobs to make sticky - None means all
|
|
|
0.64.168
by Ian Clatworthy
blob reference counting, not just sticky vs otherwise |
105 |
self._blob_ref_counts = {} |
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
106 |
if info is not None: |
107 |
try: |
|
|
0.64.168
by Ian Clatworthy
blob reference counting, not just sticky vs otherwise |
108 |
blobs_by_counts = info['Blob reference counts'] |
109 |
# The parser hands values back as lists, already parsed
|
|
110 |
for count, blob_list in blobs_by_counts.items(): |
|
111 |
n = int(count) |
|
112 |
for b in blob_list: |
|
113 |
self._blob_ref_counts[b] = n |
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
114 |
except KeyError: |
115 |
# info not in file - possible when no blobs used
|
|
116 |
pass
|
|
117 |
||
|
0.112.4
by Max Bowsher
Store the BranchMapper in the CacheManager so it can be got from other places. |
118 |
# BranchMapper has no state (for now?), but we keep it around rather
|
119 |
# than reinstantiate on every usage
|
|
120 |
self.branch_mapper = branch_mapper.BranchMapper() |
|
121 |
||
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
122 |
self.reftracker = RefTracker() |
123 |
||
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
124 |
def dump_stats(self, note=trace.note): |
125 |
"""Dump some statistics about what we cached.""" |
|
126 |
# TODO: add in inventory stastistics
|
|
127 |
note("Cache statistics:") |
|
128 |
self._show_stats_for(self._sticky_blobs, "sticky blobs", note=note) |
|
129 |
self._show_stats_for(self.revision_ids, "revision-ids", note=note) |
|
130 |
# These aren't interesting so omit from the output, at least for now
|
|
131 |
#self._show_stats_for(self._blobs, "other blobs", note=note)
|
|
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
132 |
#self.reftracker.dump_stats(note=note)
|
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
133 |
|
|
0.64.159
by Ian Clatworthy
make the file-id cache optional and branch-ref aware |
134 |
def _show_stats_for(self, dict, label, note=trace.note, tuple_key=False): |
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
135 |
"""Dump statistics about a given dictionary. |
136 |
||
137 |
By the key and value need to support len().
|
|
138 |
"""
|
|
139 |
count = len(dict) |
|
|
0.64.159
by Ian Clatworthy
make the file-id cache optional and branch-ref aware |
140 |
if tuple_key: |
141 |
size = sum(map(len, (''.join(k) for k in dict.keys()))) |
|
142 |
else: |
|
143 |
size = sum(map(len, dict.keys())) |
|
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
144 |
size += sum(map(len, dict.values())) |
|
0.64.168
by Ian Clatworthy
blob reference counting, not just sticky vs otherwise |
145 |
size = size * 1.0 / 1024 |
146 |
unit = 'K' |
|
147 |
if size > 1024: |
|
148 |
size = size / 1024 |
|
149 |
unit = 'M' |
|
150 |
if size > 1024: |
|
151 |
size = size / 1024 |
|
152 |
unit = 'G' |
|
153 |
note(" %-12s: %8.1f %s (%d %s)" % (label, size, unit, count, |
|
|
0.123.3
by Jelmer Vernooij
Fix some imports. |
154 |
single_plural(count, "item", "items"))) |
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
155 |
|
156 |
def clear_all(self): |
|
157 |
"""Free up any memory used by the caches.""" |
|
158 |
self._blobs.clear() |
|
159 |
self._sticky_blobs.clear() |
|
160 |
self.revision_ids.clear() |
|
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
161 |
self.reftracker.clear() |
|
0.64.153
by Ian Clatworthy
clear caches before packing; show cache stats in verbose mode |
162 |
self.inventories.clear() |
163 |
||
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
164 |
def _flush_blobs_to_disk(self): |
165 |
blobs = self._sticky_blobs.keys() |
|
166 |
sticky_blobs = self._sticky_blobs |
|
167 |
total_blobs = len(sticky_blobs) |
|
168 |
blobs.sort(key=lambda k:len(sticky_blobs[k])) |
|
169 |
if self._tempdir is None: |
|
|
0.123.3
by Jelmer Vernooij
Fix some imports. |
170 |
tempdir = tempfile.mkdtemp(prefix='fastimport_blobs-') |
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
171 |
self._tempdir = tempdir |
172 |
self._cleanup.tempdir = self._tempdir |
|
173 |
self._cleanup.small_blobs = tempfile.TemporaryFile( |
|
174 |
prefix='small-blobs-', dir=self._tempdir) |
|
175 |
small_blob_ref = weakref.ref(self._cleanup.small_blobs) |
|
176 |
# Even though we add it to _Cleanup it seems that the object can be
|
|
177 |
# destroyed 'too late' for cleanup to actually occur. Probably a
|
|
178 |
# combination of bzr's "die directly, don't clean up" and how
|
|
179 |
# exceptions close the running stack.
|
|
180 |
def exit_cleanup(): |
|
181 |
small_blob = small_blob_ref() |
|
182 |
if small_blob is not None: |
|
183 |
small_blob.close() |
|
184 |
shutil.rmtree(tempdir, ignore_errors=True) |
|
185 |
atexit.register(exit_cleanup) |
|
186 |
count = 0 |
|
187 |
bytes = 0 |
|
188 |
n_small_bytes = 0 |
|
189 |
while self._sticky_memory_bytes > self._sticky_flushed_size: |
|
190 |
id = blobs.pop() |
|
191 |
blob = self._sticky_blobs.pop(id) |
|
192 |
n_bytes = len(blob) |
|
193 |
self._sticky_memory_bytes -= n_bytes |
|
194 |
if n_bytes < self._small_blob_threshold: |
|
195 |
f = self._cleanup.small_blobs |
|
196 |
f.seek(0, os.SEEK_END) |
|
197 |
self._disk_blobs[id] = (f.tell(), n_bytes, None) |
|
198 |
f.write(blob) |
|
199 |
n_small_bytes += n_bytes |
|
200 |
else: |
|
201 |
fd, name = tempfile.mkstemp(prefix='blob-', dir=self._tempdir) |
|
202 |
os.write(fd, blob) |
|
203 |
os.close(fd) |
|
204 |
self._disk_blobs[id] = (0, n_bytes, name) |
|
205 |
bytes += n_bytes |
|
206 |
del blob |
|
207 |
count += 1 |
|
208 |
trace.note('flushed %d/%d blobs w/ %.1fMB (%.1fMB small) to disk' |
|
209 |
% (count, total_blobs, bytes / 1024. / 1024, |
|
210 |
n_small_bytes / 1024. / 1024)) |
|
211 |
||
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
212 |
def store_blob(self, id, data): |
213 |
"""Store a blob of data.""" |
|
|
0.64.169
by Ian Clatworthy
fix blob tracking when -v not given |
214 |
# Note: If we're not reference counting, everything has to be sticky
|
215 |
if not self._blob_ref_counts or id in self._blob_ref_counts: |
|
216 |
self._sticky_blobs[id] = data |
|
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
217 |
self._sticky_memory_bytes += len(data) |
218 |
if self._sticky_memory_bytes > self._sticky_cache_size: |
|
219 |
self._flush_blobs_to_disk() |
|
|
0.64.169
by Ian Clatworthy
fix blob tracking when -v not given |
220 |
elif data == '': |
221 |
# Empty data is always sticky
|
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
222 |
self._sticky_blobs[id] = data |
223 |
else: |
|
224 |
self._blobs[id] = data |
|
225 |
||
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
226 |
def _decref(self, id, cache, fn): |
227 |
if not self._blob_ref_counts: |
|
228 |
return False |
|
229 |
count = self._blob_ref_counts.get(id, None) |
|
230 |
if count is not None: |
|
231 |
count -= 1 |
|
232 |
if count <= 0: |
|
233 |
del cache[id] |
|
234 |
if fn is not None: |
|
235 |
os.unlink(fn) |
|
236 |
del self._blob_ref_counts[id] |
|
237 |
return True |
|
238 |
else: |
|
239 |
self._blob_ref_counts[id] = count |
|
240 |
return False |
|
241 |
||
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
242 |
def fetch_blob(self, id): |
243 |
"""Fetch a blob of data.""" |
|
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
244 |
if id in self._blobs: |
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
245 |
return self._blobs.pop(id) |
|
0.64.264
by Ian Clatworthy
Merge John's smarter caching of blobs to improve memory footprint |
246 |
if id in self._disk_blobs: |
247 |
(offset, n_bytes, fn) = self._disk_blobs[id] |
|
248 |
if fn is None: |
|
249 |
f = self._cleanup.small_blobs |
|
250 |
f.seek(offset) |
|
251 |
content = f.read(n_bytes) |
|
252 |
else: |
|
253 |
fp = open(fn, 'rb') |
|
254 |
try: |
|
255 |
content = fp.read() |
|
256 |
finally: |
|
257 |
fp.close() |
|
258 |
self._decref(id, self._disk_blobs, fn) |
|
259 |
return content |
|
260 |
content = self._sticky_blobs[id] |
|
261 |
if self._decref(id, self._sticky_blobs, None): |
|
262 |
self._sticky_memory_bytes -= len(content) |
|
263 |
return content |
|
|
0.78.3
by Ian Clatworthy
move GenericCacheManager into its own module |
264 |
|
|
0.123.6
by Jelmer Vernooij
Split out reftracker. |
265 |