def deduplicate(log, *subpath, timeout=None):
log.info('locking the deduplication db...')
with lock.Lock(path.saviour('.duperemove.hashfile-lock'), timeout=timeout):
log.info('deduplicating...')
run = log.pipe_powered(subprocess.run,
stdout=logging.INFO, stderr=logging.WARNING)
r = run(['duperemove',
'--hashfile', path.saviour('.duperemove.hashfile'),
'-hdr', path.saviour('_', *subpath)])
assert r.returncode in (0, 22) # nothing to deduplicate
def _deduplicate(log, db_name, resource_name, timeout=None):
log.info(f'locking the deduplication db {db_name}...')
hashfilesdir = path.saviour('.duperemove', 'hashfiles')
if not os.path.exists(hashfilesdir):
os.makedirs(hashfilesdir)
os.system(f'chattr +C {hashfilesdir} || true')
db_file = path.saviour('.duperemove', 'hashfiles', db_name)
db_lock = path.saviour('.duperemove', 'locks', db_name, makedirs=True)
with lock.Lock(db_lock, timeout=timeout):
log.info(f'deduplicating {resource_name} ({db_name})...')
run = log.pipe_powered(subprocess.run,
stdout=logging.INFO,
stderr=logging.WARNING)
r = run([
'duperemove', '--dedupe-options=nofiemap', '--io-threads=2',
'--cpu-threads=2', '--hashfile', db_file, '-hdr',
path.saviour('_', resource_name, 'data')
])
assert r.returncode in (0, 22) # nothing to deduplicate
def is_fetcheable(source, url, timeout=2):
if source == 'local':
return os.path.exists(path.saviour(url))
elif source != 'direct':
url = source + '/' + url
url = 'http://' + url if '://' not in source else url
try:
r = requests.head(url, allow_redirects=False, timeout=timeout)
return r.status_code < 400
except (requests.exceptions.BaseHTTPError, urllib3.exceptions.HTTPError,
requests.exceptions.Timeout, OSError) as ex:
log.warning(f'{ex}')
return False
return False
def fetch(self, url, out_path):
sources = saviour_sources()
for i, (source, cache) in enumerate(sources):
if is_fetcheable(source, url) or i == len(sources) - 1:
if source == 'local':
reflink.auto(path.saviour(url), out_path)
return
sess = self._get_requests_session(direct=not cache)
if source == 'direct':
surl = url
else:
surl = source + '/' + url
surl = 'http://' + surl if '://' not in source else surl
log.debug(f'fetching{"/caching" if cache else ""} '
f'{os.path.basename(url)} from {surl}')
r = sess.get(surl) # not raw because that punctures cache
with open(out_path, 'wb') as f:
f.write(r.content)
return
def mirror(config, *what_to_mirror, deduplicate=None):
total_failures = []
failures = collections.defaultdict(list)
with open(config) as f:
config = ruamel.yaml.YAML(typ='safe').load(f)
if 'mirror' in config and not config['mirror']:
log.warning('mirroring is disabled in config')
return
hows, whats = config['how'], config['what']
if not what_to_mirror:
what_to_mirror = whats.keys()
else:
what_to_mirror = ([
k for k in whats.keys() if any(
fnmatch.fnmatch(k, req) for req in what_to_mirror)
] + [k for k in what_to_mirror if '=' in k])
if not what_to_mirror:
log.error('nothing to mirror')
return
for resource in what_to_mirror:
log.debug(f'processing {resource}...')
if '=' not in resource: # example: alpine-3.13=alpine/v3.13/main/x86
resource_name, tail = resource, ''
s = whats[resource_name]
else: # example: alpine-3.13=alpine/v3.13/main/x86
resource_name, s = resource.split('=', 1)
# FIXME UGLY: config overrides are stronger that = (more syntax?)
# TODO: whats shouldn't be a dict, I think, just a list of strings
if resource_name in whats:
s = whats[resource_name]
if s is None:
s = resource_name
if '/' in s:
how_name, suffix = s.split('/', 1)
suffix = '/' + suffix
else:
how_name, suffix = s, ''
try:
how = hows[how_name]
except KeyError:
log.error(f'missing how section on {how_name}')
raise SystemExit()
url = how['url'] + suffix
method = how['method']
sources = (how['sources'] if 'sources' in how else [how['url']])
sources = [s + suffix for s in sources]
extra_args = {
k: v
for k, v in how.items()
if k not in ('url', 'sources', 'method', 'validate', 'deduplicate')
}
if f'method_{method}' not in globals():
log.error(f'unsupported method {method}')
raise SystemExit()
meth = globals()[f'method_{method}']
symlink = path.saviour(url.rstrip('/'))
# usually symlink points to data, but while we're working on it,
# it temporarily points to a consistent snapshot of it named `snap`
data = os.path.realpath(path.saviour('_', resource_name, 'data'))
snap = os.path.realpath(path.saviour('_', resource_name, 'snap'))
temp = os.path.realpath(path.saviour('_', resource_name, 'temp'))
lockfile = path.saviour('_', resource_name) + '-lock'
assert data.startswith(os.path.realpath(path.SAVIOUR))
assert snap.startswith(os.path.realpath(path.SAVIOUR))
assert temp.startswith(os.path.realpath(path.SAVIOUR))
sublog = log.Sublogger(f'{method} {resource_name}')
sublog.info('locking...')
with lock.Lock(lockfile):
os.makedirs(os.path.dirname(snap), exist_ok=True)
if os.path.exists(temp):
sublog.info('removing stale temp...')
_remove(temp)
if os.path.exists(symlink): # it's already published
if os.path.exists(data) and not os.path.exists(snap):
# `data` is present and is the best we have to publish
sublog.info('snapshotting...')
reflink.always(data, temp, preserve=True)
os.rename(temp, snap)
if os.path.exists(snap):
# link to a consistent snapshot while we work on `data`
_symlink(snap, symlink)
for source in sources:
sublog.info(f'trying {source}...')
try:
meth(sublog, source, snap, data, **extra_args)
assert os.path.exists(data)
if 'validate' in how:
sublog.info(f'validating with {how["validate"]}...')
validator = globals()[f'validate_{how["validate"]}']
validator(sublog, source, data)
sublog.info('validated')
break
except Exception as _:
traceback.print_exc()
failures[resource_name].append(source)
fingertip.util.log.warning(f'failed to mirror {source}')
if len(failures[resource_name]) == len(sources):
sublog.error(f'failed to mirror '
f'from all {len(sources)} sources')
total_failures.append(resource_name)
continue
_symlink(data, symlink)
if os.path.exists(snap):
os.rename(snap, temp) # move it out the way asap
sublog.info('removing now obsolete snapshot...')
_remove(temp)
how_deduplicate = how.get('deduplicate', True)
db_name = how_deduplicate if how_deduplicate is not True else how_name
if how_deduplicate and deduplicate is not False:
try:
_deduplicate(sublog, db_name, resource_name, timeout=1)
except lock.LockTimeout:
log.warning(f'skipped deduplication of {resource_name}, '
f'db {db_name} was locked')
if total_failures:
fingertip.util.log.error(f'failed: {", ".join(total_failures)}')
raise FailureToMirrorError(", ".join(total_failures))
log.info('saviour has completed mirroring')
def __init__(self, url, *path_components, enough_to_have=None):
if not path_components:
path_components = [url.replace('/', '::')]
self.url = url
cache_path = path.downloads('git', *path_components, makedirs=True)
self.path = temp.disappearing_dir(os.path.dirname(cache_path),
path_components[-1])
lock_working_copy_path = self.path + '-lock'
lock_cache_path = cache_path + '-lock'
lock.Lock.__init__(self, lock_working_copy_path)
sources = saviour_sources()
self.self_destruct = False
with lock.Lock(lock_cache_path), lock.Lock(lock_working_copy_path):
cache_is_enough = False
if os.path.exists(cache_path):
try:
cr = git.Repo(cache_path)
cache_is_enough = (enough_to_have
and _has_rev(cr, enough_to_have))
except git.GitError as e:
log.error(f'something wrong with git cache {cache_path}')
log.error(str(e))
_remove(self.path)
for i, (source, cache) in enumerate(sources):
last_source = i == len(sources) - 1
if cache and cache_is_enough:
log.info(f'not re-fetching {url} from {source} '
f'because {enough_to_have} '
'is already present in cache')
git.Repo.clone_from(cache_path, self.path, mirror=True)
break
if source == 'local':
surl = path.saviour(url).replace('//', '/') # workaround
if not os.path.exists(surl) and not last_source:
continue
log.info(f'cloning {url} from local saviour mirror')
git.Repo.clone_from(surl, self.path, mirror=True)
break
elif source == 'direct':
surl = url
else:
surl = source + '/' + url
surl = 'http://' + surl if '://' not in source else surl
log.info(f'cloning {url} from {source} '
f'cache_exists={os.path.exists(cache_path)}...')
try:
# TODO: bare clone
# no harm in referencing cache, even w/o cached+
git.Repo.clone_from(surl,
self.path,
mirror=True,
dissociate=True,
reference_if_able=cache_path)
except git.GitError:
log.warning(f'could not clone {url} from {source}')
if last_source:
raise
continue
break
_remove(cache_path)
reflink.auto(self.path, cache_path)
git.Repo.__init__(self, self.path)
self.remotes[0].set_url(url)
self.self_destruct = True
def mirror(config, *what_to_mirror):
total_failures = []
failures = collections.defaultdict(list)
with open(config) as f:
config = ruamel.yaml.YAML(typ='safe').load(f)
hows, whats = config['how'], config['what']
if not what_to_mirror:
what_to_mirror = whats.keys()
else:
what_to_mirror = [k for k in whats.keys()
if any((fnmatch.fnmatch(k, req)
for req in what_to_mirror))]
for resource_name in what_to_mirror or whats.keys():
s = whats[resource_name]
log.debug(f'processing {resource_name}...')
if s is None:
how, suffix = resource_name, ''
elif '/' in s:
how, suffix = s.split('/', 1)
suffix = '/' + suffix
else:
how, suffix = s, ''
try:
how = hows[how]
except KeyError:
log.error(f'missing how section on {how}')
raise SystemExit()
url = how['url'] + suffix
method = how['method']
sources = (how['sources'] if 'sources' in how else [how['url']])
sources = [s + suffix for s in sources]
extra_args = {k: v for k, v in how.items()
if k not in ('url', 'sources', 'method')}
if f'method_{method}' not in globals():
log.error(f'unsupported method {method}')
raise SystemExit()
meth = globals()[f'method_{method}']
symlink = path.saviour(url.rstrip('/'))
# usually symlink points to data, but while we're working on it,
# it temporarily points to a consistent snapshot of it named `snap`
data = path.saviour('_', resource_name, 'data')
snap = path.saviour('_', resource_name, 'snap')
temp = path.saviour('_', resource_name, 'temp')
lockfile = path.saviour('_', resource_name) + '-lock'
assert data.startswith(path.SAVIOUR)
assert snap.startswith(path.SAVIOUR)
assert temp.startswith(path.SAVIOUR)
sublog = log.Sublogger(f'{method} {resource_name}')
sublog.info('locking...')
with lock.Lock(lockfile):
os.makedirs(os.path.dirname(snap), exist_ok=True)
if os.path.exists(temp):
sublog.info('removing stale temp...')
_remove(temp)
if os.path.exists(symlink): # it's already published
if os.path.exists(data) and not os.path.exists(snap):
# `data` is present and is the best we have to publish
sublog.info('snapshotting...')
reflink.always(data, temp, preserve=True)
os.rename(temp, snap)
if os.path.exists(snap):
# link to a consistent snapshot while we work on `data`
_symlink(snap, symlink)
for source in sources:
sublog.info(f'trying {source}...')
try:
meth(sublog, source, snap, data, **extra_args)
assert os.path.exists(data)
break
except Exception as _:
traceback.print_exc()
failures[resource_name].append(source)
fingertip.util.log.warning(f'failed to mirror {source}')
if len(failures[resource_name]) == len(sources):
sublog.error(f'failed to mirror '
f'from all {len(sources)} sources')
total_failures.append(resource_name)
continue
_symlink(data, symlink)
if os.path.exists(snap):
os.rename(snap, temp) # move it out the way asap
sublog.info('removing now obsolete snapshot...')
_remove(temp)
try:
deduplicate(sublog, resource_name, timeout=1)
except lock.LockTimeout:
log.warning('skipped deduplication, db was locked')
if total_failures:
fingertip.util.log.error(f'failed: {", ".join(total_failures)}')
raise SystemExit()
log.info('saviour has completed mirroring')
def __init__(self, url, *path_components, enough_to_have=None):
assert path_components
self.url = url
cache_path = path.downloads('git', *path_components, makedirs=True)
cache_exists = os.path.exists(cache_path)
self.path = temp.disappearing_dir(os.path.dirname(cache_path),
path_components[-1])
lock_working_copy_path = self.path + '-lock'
lock_cache_path = cache_path + '-lock'
lock.Lock.__init__(self, lock_working_copy_path)
update_not_needed = None
sources = saviour_sources()
self.self_destruct = False
with lock.Lock(lock_cache_path), lock.Lock(lock_working_copy_path):
_remove(self.path)
for i, (source, cache) in enumerate(sources):
last_source = i == len(sources) - 1
if cache and cache_exists and update_not_needed is None:
cr = git.Repo(cache_path)
update_not_needed = enough_to_have and (
enough_to_have in (t.name for t in cr.tags) or
enough_to_have in (h.name for h in cr.heads) or
enough_to_have in (c.hexsha for c in cr.iter_commits())
# that's not all revspecs, but best-effort is fine
)
if update_not_needed:
log.info(f'not re-fetching {url} from {source} '
f'because {enough_to_have} '
'is already present in cache')
git.Repo.clone_from(cache_path, self.path, mirror=True)
break
if source == 'local':
surl = path.saviour(url).replace('//', '/') # workaround
if not os.path.exists(surl) and not last_source:
continue
log.info(f'cloning {url} from local saviour mirror')
git.Repo.clone_from(surl, self.path, mirror=True)
break
elif source == 'direct':
surl = url
else:
surl = source + '/' + url
surl = 'http://' + surl if '://' not in source else surl
log.info(f'cloning {url} from {source} '
f'cache_exists={cache_exists}...')
try:
# TODO: bare clone
# no harm in referencing cache, even w/o cached+
git.Repo.clone_from(surl, self.path, mirror=True,
dissociate=True,
reference_if_able=cache_path)
except git.GitError:
log.warning(f'could not clone {url} from {source}')
if last_source:
raise
continue
break
_remove(cache_path)
reflink.auto(self.path, cache_path)
git.Repo.__init__(self, self.path)
self.remotes[0].set_url(url)
self.self_destruct = True