def _autocontainer_loop(self, account, marker=None, limit=None,
concurrency=1, **kwargs):
from functools import partial
container_marker = self.flatns_manager(marker) if marker else None
count = 0
kwargs['pool_manager'] = get_pool_manager(
pool_maxsize=concurrency * 2)
# Start to list contents at 'marker' inside the last visited container
if container_marker:
for element in depaginate(
self.app.client_manager.storage.object_list,
listing_key=lambda x: x['objects'],
marker_key=lambda x: x.get('next_marker'),
truncated_key=lambda x: x['truncated'],
account=account, container=container_marker,
marker=marker, **kwargs):
count += 1
yield element
if limit and count >= limit:
return
pool = GreenPool(concurrency)
for object_list in pool.imap(
partial(self._list_autocontainer_objects,
account=account, **kwargs),
depaginate(self.app.client_manager.storage.container_list,
item_key=lambda x: x[0],
marker_key=lambda x: x[-1][0],
account=account,
marker=container_marker)):
for element in object_list:
count += 1
yield element
if limit and count >= limit:
return
def __init__(self,
namespace,
concurrency=50,
error_file=None,
rebuild_file=None,
check_xattr=True,
limit_listings=0,
request_attempts=1,
logger=None,
verbose=False,
check_hash=False,
**_kwargs):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.check_xattr = bool(check_xattr)
self.check_hash = bool(check_hash)
self.logger = logger or get_logger(
{'namespace': namespace}, name='integrity', verbose=verbose)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
outfile = open(self.error_file, 'a')
self.error_writer = csv.writer(outfile, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
self.fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(self.fd, delimiter='|')
self.api = ObjectStorageApi(namespace,
logger=self.logger,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.rdir_client = RdirClient({"namespace": namespace},
logger=self.logger)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = {}
self.running = {}
self.running_lock = Semaphore(1)
self.result_queue = Queue(concurrency)
self.run_time = 0
def __init__(self,
namespace,
concurrency=50,
error_file=None,
rebuild_file=None,
full=True,
limit_listings=0,
request_attempts=1):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.full = bool(full)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
f = open(self.error_file, 'a')
self.error_writer = csv.writer(f, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(fd, delimiter='|')
conf = {'namespace': namespace}
self.account_client = AccountClient(conf,
max_retries=request_attempts - 1)
self.container_client = ContainerClient(
conf,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.blob_client = BlobClient(conf=conf)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = {}
self.running = {}
def run(self, *args, **kwargs):
try:
self.logger.info('conscience agent: starting')
pool = GreenPool(len(self.watchers))
for watcher in self.watchers:
pool.spawn(watcher.start)
self.running = True
while self.running:
sleep(1)
for w in self.watchers:
if w.failed:
self.watchers.remove(w)
self.logger.warn('restart watcher "%s"', w.name)
new_w = ServiceWatcher(self.conf, w.service)
self.watchers.append(new_w)
pool.spawn(new_w.start)
except Exception as e:
self.logger.error('ERROR in main loop %s', e)
raise e
finally:
self.logger.warn('conscience agent: stopping')
self.running = False
self.stop_watchers()
def mover_pass(self, **kwargs):
start_time = report_time = time.time()
total_errors = 0
mover_time = 0
pool = GreenPool(self.concurrency)
paths = paths_gen(self.volume)
for path in paths:
loop_time = time.time()
now = time.time()
if now - self.last_usage_check >= self.usage_check_interval:
free_ratio = statfs(self.volume)
usage = (1 - float(free_ratio)) * 100
if usage <= self.usage_target:
self.logger.info(
'current usage %.2f%%: target reached (%.2f%%)', usage,
self.usage_target)
break
self.last_usage_check = now
# Spawn a chunk move task.
# The call will block if no green thread is available.
pool.spawn_n(self.safe_chunk_move, path)
self.chunks_run_time = ratelimit(self.chunks_run_time,
self.max_chunks_per_second)
self.total_chunks_processed += 1
now = time.time()
if now - self.last_reported >= self.report_interval:
self.logger.info(
'%(start_time)s '
'%(passes)d '
'%(errors)d '
'%(c_rate).2f '
'%(b_rate).2f '
'%(total).2f '
'%(mover_time).2f'
'%(mover_rate).2f' % {
'start_time': time.ctime(report_time),
'passes': self.passes,
'errors': self.errors,
'c_rate': self.passes / (now - report_time),
'b_rate': self.bytes_processed / (now - report_time),
'total': (now - start_time),
'mover_time': mover_time,
'mover_rate': mover_time / (now - start_time)
})
report_time = now
total_errors += self.errors
self.passes = 0
self.bytes_processed = 0
self.last_reported = now
mover_time += (now - loop_time)
if self.limit != 0 and self.total_chunks_processed >= self.limit:
break
pool.waitall()
elapsed = (time.time() - start_time) or 0.000001
self.logger.info(
'%(elapsed).02f '
'%(errors)d '
'%(chunk_rate).2f '
'%(bytes_rate).2f '
'%(mover_time).2f '
'%(mover_rate).2f' % {
'elapsed': elapsed,
'errors': total_errors + self.errors,
'chunk_rate': self.total_chunks_processed / elapsed,
'bytes_rate': self.total_bytes_processed / elapsed,
'mover_time': mover_time,
'mover_rate': mover_time / elapsed
})
def __init__(self,
namespace,
concurrency=50,
error_file=None,
rebuild_file=None,
check_xattr=True,
limit_listings=0,
request_attempts=1,
logger=None,
verbose=False,
check_hash=False,
min_time_in_error=0.0,
required_confirmations=0,
beanstalkd_addr=None,
beanstalkd_tube=BlobRebuilder.DEFAULT_BEANSTALKD_WORKER_TUBE,
cache_size=2**24,
**_kwargs):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.error_sender = None
self.check_xattr = bool(check_xattr)
self.check_hash = bool(check_hash)
self.logger = logger or get_logger(
{'namespace': namespace}, name='integrity', verbose=verbose)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
outfile = open(self.error_file, 'a')
self.error_writer = csv.writer(outfile, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
self.fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(self.fd, delimiter='|')
if beanstalkd_addr:
self.error_sender = BeanstalkdSender(beanstalkd_addr,
beanstalkd_tube, self.logger)
self.api = ObjectStorageApi(namespace,
logger=self.logger,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.rdir_client = RdirClient({"namespace": namespace},
logger=self.logger)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = CacheDict(cache_size)
self.running_tasks = {}
self.running_lock = Semaphore(1)
self.result_queue = LightQueue(concurrency)
self.running = True
self.run_time = 0
# Set of targets which must be checked again, to confirm
# or deny the issues reported by previous passes.
self.delayed_targets = dict()
# Minimum time in error and number of confirmations of the error
# before triggering a reconstruction action.
self.min_time_in_error = min_time_in_error
self.required_confirmations = required_confirmations
class Checker(object):
def __init__(self,
namespace,
concurrency=50,
error_file=None,
rebuild_file=None,
check_xattr=True,
limit_listings=0,
request_attempts=1,
logger=None,
verbose=False,
check_hash=False,
min_time_in_error=0.0,
required_confirmations=0,
beanstalkd_addr=None,
beanstalkd_tube=BlobRebuilder.DEFAULT_BEANSTALKD_WORKER_TUBE,
cache_size=2**24,
**_kwargs):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.error_sender = None
self.check_xattr = bool(check_xattr)
self.check_hash = bool(check_hash)
self.logger = logger or get_logger(
{'namespace': namespace}, name='integrity', verbose=verbose)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
outfile = open(self.error_file, 'a')
self.error_writer = csv.writer(outfile, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
self.fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(self.fd, delimiter='|')
if beanstalkd_addr:
self.error_sender = BeanstalkdSender(beanstalkd_addr,
beanstalkd_tube, self.logger)
self.api = ObjectStorageApi(namespace,
logger=self.logger,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.rdir_client = RdirClient({"namespace": namespace},
logger=self.logger)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = CacheDict(cache_size)
self.running_tasks = {}
self.running_lock = Semaphore(1)
self.result_queue = LightQueue(concurrency)
self.running = True
self.run_time = 0
# Set of targets which must be checked again, to confirm
# or deny the issues reported by previous passes.
self.delayed_targets = dict()
# Minimum time in error and number of confirmations of the error
# before triggering a reconstruction action.
self.min_time_in_error = min_time_in_error
self.required_confirmations = required_confirmations
def reset_stats(self):
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
def _spawn(self, func, target, *args, **kwargs):
"""
Spawn a task on the internal GreenPool.
Discards the task if the pool is no more running.
"""
if self.running:
return self.pool.spawn(func, target, *args, **kwargs)
self.logger.info("Discarding %s", target)
return None
def _spawn_n(self, func, target, *args, **kwargs):
"""
Spawn a task on the internal GreenPool, do not wait for the result.
Discards the task if the pool is no more running.
"""
if self.running:
return self.pool.spawn_n(func, target, *args, **kwargs)
self.logger.info("Discarding %s", target)
return None
def complete_target_from_chunk_metadata(self, target, xattr_meta):
"""
Complete a Target object from metadata found in chunk's extended
attributes. In case the "fullpath" is not available, try to read
legacy metadata, and maybe ask meta1 to resolve the CID into
account and container names.
"""
# pylint: disable=unbalanced-tuple-unpacking
try:
acct, ct, path, vers, content_id = \
decode_fullpath(xattr_meta['full_path'])
target.account = acct
target.container = ct
target.obj = path
target.content_id = content_id
target.version = vers
except KeyError:
# No fullpath header, try legacy headers
if 'content_path' in xattr_meta:
target.obj = xattr_meta['content_path']
if 'content_id' in xattr_meta:
target.content_id = xattr_meta['content_id']
if 'content_version' in xattr_meta:
target.version = xattr_meta['content_version']
cid = xattr_meta.get('container_id')
if cid:
try:
md = self.api.directory.show(cid=cid)
acct = md.get('account')
ct = md.get('name')
if acct:
target.account = acct
if ct:
target.container = ct
except Exception as err:
self.logger.warn(
"Failed to resolve CID %s into account "
"and container names: %s", cid, err)
def recover_and_complete_object_meta(self, target, chunk):
_, rawx_service, chunk_id = chunk.rsplit('/', 2)
# 1. Fetch chunk list from rdir (could be cached).
# Unfortunately we cannot seek for a chunk ID.
entries = [
x for x in self.rdir_client.chunk_fetch(rawx_service, limit=-1)
if x[2] == chunk_id
]
if not entries:
self.logger.warn('Chunk %s not found in rdir' % chunk_id)
return
elif len(entries) > 1:
self.logger.info('Chunk %s appears in %d objects', chunk_id,
len(entries))
# 2. Find content and container IDs
target.cid, target.content_id = entries[0][0:2]
meta = self.api.object_get_properties(None,
None,
None,
cid=target.cid,
content=target.content_id)
target.obj = meta['name']
target.version = meta['version']
target.account, target.container = self.api.resolve_cid(target.cid)
def send_result(self, target, errors=None, irreparable=False):
"""
Put an item in the result queue.
"""
# TODO(FVE): send to an external queue.
target.append_result(ItemResult(errors, irreparable))
self.result_queue.put(target)
def send_chunk_job(self, target, irreparable=False):
"""
Send a "content broken" event, to trigger the
reconstruction of the chunk.
"""
item = (self.api.namespace, target.cid, target.content_id,
target.chunk)
ev_dict = BlobRebuilder.task_event_from_item(item)
if irreparable:
ev_dict['data']['irreparable'] = irreparable
job = json.dumps(ev_dict)
self.error_sender.send_job(job)
self.error_sender.job_done() # Don't expect any response
def write_error(self, target, irreparable=False):
if not self.error_file:
return
error = list()
if irreparable:
error.append(IRREPARABLE_PREFIX)
error.append(target.account)
if target.container:
error.append(target.container)
if target.obj:
error.append(target.obj)
if target.chunk:
error.append(target.chunk)
self.error_writer.writerow(error)
def write_rebuilder_input(self, target, irreparable=False):
error = list()
if irreparable:
error.append(IRREPARABLE_PREFIX)
error.append(target.cid)
# FIXME(FVE): ensure we always resolve content_id,
# or pass object version along with object name.
error.append(target.content_id or target.obj)
error.append(target.chunk)
self.rebuild_writer.writerow(error)
def write_chunk_error(self, target, chunk=None, irreparable=False):
if chunk is not None:
target = target.copy()
target.chunk = chunk
self.write_error(target, irreparable=irreparable)
if self.rebuild_file:
self.write_rebuilder_input(target, irreparable=irreparable)
if self.error_sender:
self.send_chunk_job(target, irreparable=irreparable)
def _check_chunk_xattr(self, target, obj_meta, xattr_meta):
"""
Check coherency of chunk extended attributes with object metadata.
:returns: a list of errors
"""
errors = list()
# Composed position -> erasure coding
attr_prefix = 'meta' if '.' in obj_meta['pos'] else ''
attr_key = attr_prefix + 'chunk_size'
if str(obj_meta['size']) != xattr_meta.get(attr_key):
errors.append(
"'%s' xattr (%s) differs from size in meta2 (%s)" %
(attr_key, xattr_meta.get(attr_key), obj_meta['size']))
attr_key = attr_prefix + 'chunk_hash'
if obj_meta['hash'] != xattr_meta.get(attr_key):
errors.append(
"'%s' xattr (%s) differs from hash in meta2 (%s)" %
(attr_key, xattr_meta.get(attr_key), obj_meta['hash']))
return errors
def _check_chunk(self, target):
"""
Execute various checks on a chunk:
- does it appear in object's chunk list?
- is it reachable?
- are its extended attributes coherent?
:returns: the list of errors encountered,
and the chunk's owner object metadata.
"""
chunk = target.chunk
errors = list()
obj_meta = None
xattr_meta = None
cached = self._get_cached_or_lock(chunk)
if cached is not None:
return cached + (True, )
self.logger.debug('Checking chunk "%s"', target)
try:
xattr_meta = self.api.blob_client.chunk_head(
chunk, xattr=self.check_xattr, check_hash=self.check_hash)
except exc.NotFound as err:
self.chunk_not_found += 1
errors.append('Not found: %s' % (err, ))
except exc.FaultyChunk as err:
self.chunk_exceptions += 1
errors.append('Faulty: %r' % (err, ))
except Exception as err:
self.chunk_exceptions += 1
errors.append('Check failed: %s' % (err, ))
if not target.obj:
if xattr_meta:
self.complete_target_from_chunk_metadata(target, xattr_meta)
else:
self.recover_and_complete_object_meta(target, chunk)
if target.obj:
obj_listing, obj_meta = self.check_obj(target.copy_object())
if chunk not in obj_listing:
errors.append('Missing from object listing')
db_meta = dict()
else:
db_meta = obj_listing[chunk]
if db_meta and xattr_meta and self.check_xattr:
errors.extend(
self._check_chunk_xattr(target, db_meta, xattr_meta))
self.list_cache[chunk] = errors, obj_meta
self._unlock(chunk)
# Do not send errors directly, let the caller do it.
# Indeed, it may want to check if the chunks can be repaired or not.
self.chunks_checked += 1
return errors, obj_meta, False
def check_chunk(self, target):
errors, _obj_meta, from_cache = self._check_chunk(target)
# If the result comes from the cache, we already reported it.
if not from_cache:
self.send_result(target, errors, target.irreparable)
return errors
def _check_metachunk(self, target, stg_met, pos, chunks, recurse=0):
"""
Check that a metachunk has the right number of chunks.
:returns: the list of errors
"""
required = stg_met.expected_chunks
errors = list()
chunk_results = list()
if len(chunks) < required:
missing_chunks = required - len(chunks)
if stg_met.ec:
subs = {x['num'] for x in chunks}
for sub in range(required):
if sub not in subs:
chkt = target.copy()
chkt.chunk = '%d.%d' % (pos, sub)
err = "Missing chunk at position %s" % chkt.chunk
chunk_results.append((chkt, [err], False))
errors.append(err)
else:
for _ in range(missing_chunks):
chkt = target.copy()
chkt.chunk = '%d.%d' % (pos, sub)
err = "Missing chunk at position %d" % pos
chunk_results.append((chkt, [err], False))
errors.append(err)
if recurse > 0:
for chunk in chunks:
tcopy = target.copy()
tcopy.chunk = chunk['url']
chunk_errors, _, from_cache = self._check_chunk(tcopy)
chunk_results.append((tcopy, chunk_errors, from_cache))
if chunk_errors:
errors.append("Unusable chunk %s at position %s" %
(chunk['url'], chunk['pos']))
irreparable = required - len(errors) < stg_met.min_chunks_to_read
if irreparable:
errors.append(
"Unavailable metachunk at position %s "
"(%d/%d chunks available, %d/%d required)" %
(pos, required - len(errors), stg_met.expected_chunks,
stg_met.min_chunks_to_read, stg_met.expected_chunks))
for tgt, errs, from_cache in chunk_results:
# If the result comes from the cache, we already reported it.
if not from_cache:
self.send_result(tgt, errs, irreparable)
# Since the "metachunk" is not an official item type,
# this method does not report errors itself. Errors will
# be reported as object errors.
return errors
def _check_obj_policy(self, target, obj_meta, chunks, recurse=0):
"""
Check that the list of chunks of an object matches
the object's storage policy.
:returns: the list of errors encountered
"""
stg_met = STORAGE_METHODS.load(obj_meta['chunk_method'])
chunks_by_pos = _sort_chunks(chunks, stg_met.ec)
tasks = list()
for pos, pchunks in iteritems(chunks_by_pos):
tasks.append((pos,
self._spawn(self._check_metachunk,
target.copy(),
stg_met,
pos,
pchunks,
recurse=recurse)))
errors = list()
for pos, task in tasks:
if not task and not self.running:
errors.append("Pos %d skipped: checker is exiting" % pos)
continue
try:
errors.extend(task.wait())
except Exception as err:
errors.append("Check failed: pos %d: %s" % (pos, err))
return errors
def check_obj_versions(self, target, versions, recurse=0):
"""
Run checks of all versions of the targeted object in parallel.
"""
tasks = list()
for ov in versions:
tcopy = target.copy_object()
tcopy.content_id = ov['id']
tcopy.version = str(ov['version'])
tasks.append((tcopy.version,
self._spawn(self.check_obj, tcopy, recurse=recurse)))
errors = list()
for version, task in tasks:
if not task and not self.running:
errors.append("Version %s skipped: checker is exiting" %
version)
continue
try:
task.wait()
except Exception as err:
errors.append("Check failed: version %s: %s" % (version, err))
if errors:
# Send a result with the target without version to tell
# we were not able to check all versions of the object.
self.send_result(target, errors)
def _load_obj_meta(self, target, errors):
"""
Load object metadata and chunks.
:param target: which object to check.
:param errors: list of errors that will be appended
in case any error occurs.
:returns: a tuple with object metadata and a list of chunks.
"""
try:
return self.api.object_locate(target.account,
target.container,
target.obj,
version=target.version,
properties=False)
except exc.NoSuchObject as err:
self.object_not_found += 1
errors.append('Not found: %s' % (err, ))
except Exception as err:
self.object_exceptions += 1
errors.append('Check failed: %s' % (err, ))
return None, []
def _get_cached_or_lock(self, lock_key):
# If something is running, wait for it
with self.running_lock:
event = self.running_tasks.get(lock_key)
if event:
event.wait()
event = None
# Maybe get a cached result
if lock_key in self.list_cache:
return self.list_cache[lock_key]
# No cached result, try to compute the thing ourselves
while True:
with self.running_lock:
# Another check while locked
if lock_key in self.list_cache:
return self.list_cache[lock_key]
# Still nothing cached
event = self.running_tasks.get(lock_key)
if event is None:
self.running_tasks[lock_key] = Event()
return None
event.wait()
def _unlock(self, lock_key):
with self.running_lock:
event = self.running_tasks[lock_key]
del self.running_tasks[lock_key]
event.send(True)
def check_obj(self, target, recurse=0):
"""
Check one object version.
If no version is specified, all versions of the object will be checked.
:returns: the result of the check of the most recent version,
or the one that is explicitly targeted.
"""
account = target.account
container = target.container
obj = target.obj
vers = target.version # can be None
cached = self._get_cached_or_lock((account, container, obj, vers))
if cached is not None:
return cached
self.logger.info('Checking object "%s"', target)
container_listing, _ = self.check_container(target.copy_container())
errors = list()
if obj not in container_listing:
errors.append('Missing from container listing')
# checksum = None
else:
versions = container_listing[obj]
if vers is None:
if target.content_id is None:
# No version specified, check all versions
self.check_obj_versions(target.copy_object(),
versions,
recurse=recurse)
# Now return the cached result of the most recent version
target.content_id = versions[0]['id']
target.version = str(versions[0]['version'])
res = self.check_obj(target, recurse=0)
self._unlock((account, container, obj, vers))
return res
else:
for ov in versions:
if ov['id'] == target.content_id:
vers = str(ov['version'])
target.version = vers
break
else:
errors.append('Missing from container listing')
# TODO check checksum match
# checksum = container_listing[obj]['hash']
pass
meta, chunks = self._load_obj_meta(target, errors)
chunk_listing = {c['url']: c for c in chunks}
if meta:
if target.content_id is None:
target.content_id = meta['id']
if target.version is None:
target.version = str(meta['version'])
self.list_cache[(account, container, obj, vers)] = \
(chunk_listing, meta)
self.objects_checked += 1
self._unlock((account, container, obj, vers))
# Skip the check if we could not locate the object
if meta:
errors.extend(
self._check_obj_policy(target, meta, chunks, recurse=recurse))
self.send_result(target, errors)
return chunk_listing, meta
def check_container(self, target, recurse=0):
account = target.account
container = target.container
cached = self._get_cached_or_lock((account, container))
if cached is not None:
return cached
self.logger.info('Checking container "%s"', target)
account_listing = self.check_account(target.copy_account())
errors = list()
if container not in account_listing:
errors.append('Missing from account listing')
marker = None
results = []
ct_meta = dict()
extra_args = dict()
if self.limit_listings > 1 and target.obj:
# When we are explicitly checking one object, start the listing
# where this object is supposed to be. Do not use a limit,
# but an end marker, in order to fetch all versions of the object.
extra_args['prefix'] = target.obj
extra_args['end_marker'] = target.obj + '\x00' # HACK
while True:
try:
resp = self.api.object_list(account,
container,
marker=marker,
versions=True,
**extra_args)
except exc.NoSuchContainer as err:
self.container_not_found += 1
errors.append('Not found: %s' % (err, ))
break
except Exception as err:
self.container_exceptions += 1
errors.append('Check failed: %s' % (err, ))
break
truncated = resp.get('truncated', False)
if truncated:
marker = resp['next_marker']
if resp['objects']:
# safeguard, probably useless
if not marker:
marker = resp['objects'][-1]['name']
results.extend(resp['objects'])
if not truncated or self.limit_listings > 1:
break
else:
ct_meta = resp
ct_meta.pop('objects')
break
container_listing = dict()
# Save all object versions, with the most recent first
for obj in results:
container_listing.setdefault(obj['name'], list()).append(obj)
for versions in container_listing.values():
versions.sort(key=lambda o: o['version'], reverse=True)
if self.limit_listings <= 1:
# We just listed the whole container, keep the result in a cache
self.containers_checked += 1
self.list_cache[(account, container)] = container_listing, ct_meta
self._unlock((account, container))
if recurse > 0:
for obj_vers in container_listing.values():
for obj in obj_vers:
tcopy = target.copy_object()
tcopy.obj = obj['name']
tcopy.content_id = obj['id']
tcopy.version = str(obj['version'])
self._spawn_n(self.check_obj, tcopy, recurse - 1)
self.send_result(target, errors)
return container_listing, ct_meta
def check_account(self, target, recurse=0):
account = target.account
cached = self._get_cached_or_lock(account)
if cached is not None:
return cached
self.logger.info('Checking account "%s"', target)
errors = list()
marker = None
results = []
extra_args = dict()
if self.limit_listings > 0 and target.container:
# When we are explicitly checking one container, start the listing
# where this container is supposed to be, and list only one
# container.
extra_args['prefix'] = target.container
extra_args['limit'] = 1
while True:
try:
resp = self.api.container_list(account,
marker=marker,
**extra_args)
except Exception as err:
self.account_exceptions += 1
errors.append('Check failed: %s' % (err, ))
break
if resp:
marker = resp[-1][0]
results.extend(resp)
if self.limit_listings > 0:
break
else:
break
containers = dict()
for container in results:
# Name, number of objects, number of bytes
containers[container[0]] = (container[1], container[2])
if self.limit_listings <= 0:
# We just listed the whole account, keep the result in a cache
self.accounts_checked += 1
self.list_cache[account] = containers
self._unlock(account)
if recurse > 0:
for container in containers:
tcopy = target.copy_account()
tcopy.container = container
self._spawn_n(self.check_container, tcopy, recurse - 1)
self.send_result(target, errors)
return containers
def check(self, target, recurse=0):
if target.type == 'chunk':
self._spawn_n(self.check_chunk, target)
elif target.type == 'object':
self._spawn_n(self.check_obj, target, recurse)
elif target.type == 'container':
self._spawn_n(self.check_container, target, recurse)
else:
self._spawn_n(self.check_account, target, recurse)
def check_all_accounts(self, recurse=0):
all_accounts = self.api.account_list()
for acct in all_accounts:
self.check(Target(acct), recurse=recurse)
def fetch_results(self, rate_limiter=None):
while self.running and not self.result_queue.empty():
res = self.result_queue.get(True)
yield res
# Rate limiting is done on the result queue for now.
# Someday we could implement a submission queue instead of
# letting each worker submit tasks to the pool, and do
# the rate limiting on this queue.
if rate_limiter is not None:
self.run_time = rate_limiter(self.run_time)
def merge_with_delayed_target(self, target):
"""
Merge the specified target with a delayed one.
:returns: the delayed target, if there is one, with an error log
including the errors of the new target. Return the new target
otherwise.
"""
tkey = repr(target)
prev_target = self.delayed_targets.get(tkey, target)
if prev_target is not target:
errors = dict(prev_target.error_log)
errors.update(target.error_log)
prev_target.error_log = sorted(errors.items())
return prev_target
def log_result(self, target):
"""
Log a check result, if it shows errors. Dispatch the errors to the
appropriate destinations (log files, queues, etc.).
"""
# The result may come from a new target, or from an old target
# we checked another time, or both.
target = self.merge_with_delayed_target(target)
if target.has_errors:
time_in_error, confirmations = target.time_in_error()
if (time_in_error < self.min_time_in_error
or confirmations < self.required_confirmations):
self.logger.info("Delaying check for %s, %d/%d confirmations",
target, confirmations,
self.required_confirmations)
self.delayed_targets[repr(target)] = target
else:
if target.type == 'chunk':
self.logger.info(
"Writing error for %s, %d/%d confirmations", target,
confirmations, self.required_confirmations)
self.write_chunk_error(target,
irreparable=target.irreparable)
else:
self.write_error(target, irreparable=target.irreparable)
self.delayed_targets.pop(repr(target), None)
self.logger.warn(
'%s:%s\n%s', target,
' irreparable' if target.irreparable else '',
target.latest_error_result().errors_to_str(err_format=' %s'))
def run(self, rate_limiter=None):
"""
Fetch results and write logs until all jobs have finished.
:returns: a generator yielding check results.
"""
while self.running and (self.pool.running() + self.pool.waiting()):
for result in self.fetch_results(rate_limiter):
self.log_result(result)
yield result
sleep(0.1)
if self.running:
self.pool.waitall()
# No rate limiting
for result in self.fetch_results():
self.log_result(result)
yield result
self.list_cache = CacheDict(self.list_cache.size)
def stop(self):
self.logger.info("Stopping")
self.running = False
def report(self):
success = True
def _report_stat(name, stat):
print("{0:18}: {1}".format(name, stat))
print()
print('Report')
_report_stat("Accounts checked", self.accounts_checked)
if self.account_not_found:
success = False
_report_stat("Missing accounts", self.account_not_found)
if self.account_exceptions:
success = False
_report_stat("Exceptions", self.account_exceptions)
print()
_report_stat("Containers checked", self.containers_checked)
if self.container_not_found:
success = False
_report_stat("Missing containers", self.container_not_found)
if self.container_exceptions:
success = False
_report_stat("Exceptions", self.container_exceptions)
print()
_report_stat("Objects checked", self.objects_checked)
if self.object_not_found:
success = False
_report_stat("Missing objects", self.object_not_found)
if self.object_exceptions:
success = False
_report_stat("Exceptions", self.object_exceptions)
print()
_report_stat("Chunks checked", self.chunks_checked)
if self.chunk_not_found:
success = False
_report_stat("Missing chunks", self.chunk_not_found)
if self.chunk_exceptions:
success = False
_report_stat("Exceptions", self.chunk_exceptions)
return success
class Checker(object):
def __init__(self,
namespace,
concurrency=50,
error_file=None,
rebuild_file=None,
full=True,
limit_listings=0,
request_attempts=1,
logger=None,
verbose=False,
integrity=False):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.full = bool(full)
self.integrity = bool(integrity)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
outfile = open(self.error_file, 'a')
self.error_writer = csv.writer(outfile, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(fd, delimiter='|')
self.logger = logger or get_logger(
{'namespace': namespace}, name='integrity', verbose=verbose)
self.api = ObjectStorageApi(namespace,
logger=self.logger,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = {}
self.running = {}
self.result_queue = Queue()
def complete_target_from_chunk_metadata(self, target, xattr_meta):
"""
Complete a Target object from metadata found in chunk's extended
attributes. In case the "fullpath" is not available, try to read
legacy metadata, and maybe ask meta1 to resolve the CID into
account and container names.
"""
# pylint: disable=unbalanced-tuple-unpacking
try:
acct, ct, path, vers, content_id = \
decode_fullpath(xattr_meta['full_path'])
target.account = acct
target.container = ct
target.obj = path
target.content_id = content_id
target.version = vers
except KeyError:
# No fullpath header, try legacy headers
if 'content_path' in xattr_meta:
target.obj = xattr_meta['content_path']
if 'content_id' in xattr_meta:
target.content_id = xattr_meta['content_id']
if 'content_version' in xattr_meta:
target.version = xattr_meta['content_version']
cid = xattr_meta.get('container_id')
if cid:
try:
md = self.api.directory.show(cid=cid)
acct = md.get('account')
ct = md.get('name')
if acct:
target.account = acct
if ct:
target.container = ct
except Exception as err:
self.logger.warn(
"Failed to resolve CID %s into account "
"and container names: %s", cid, err)
def send_result(self, target, errors=None):
"""
Put an item in the result queue.
"""
# TODO(FVE): send to an external queue.
self.result_queue.put(ItemResult(target, errors))
def write_error(self, target, irreparable=False):
if not self.error_file:
return
error = list()
if irreparable:
error.append('#IRREPARABLE')
error.append(target.account)
if target.container:
error.append(target.container)
if target.obj:
error.append(target.obj)
if target.chunk:
error.append(target.chunk)
self.error_writer.writerow(error)
def write_rebuilder_input(self, target, irreparable=False):
# FIXME(FVE): cid can be computed from account and container names
ct_meta = self.list_cache[(target.account, target.container)][1]
try:
cid = ct_meta['system']['sys.name'].split('.', 1)[0]
except KeyError:
cid = ct_meta['properties']['sys.name'].split('.', 1)[0]
error = list()
if irreparable:
error.append('#IRREPARABLE')
error.append(cid)
# FIXME(FVE): ensure we always resolve content_id,
# or pass object version along with object name.
error.append(target.content_id or target.obj)
error.append(target.chunk)
self.rebuild_writer.writerow(error)
def write_chunk_error(self, target, chunk=None, irreparable=False):
if chunk is not None:
target = target.copy()
target.chunk = chunk
self.write_error(target, irreparable=irreparable)
if self.rebuild_file:
self.write_rebuilder_input(target, irreparable=irreparable)
def _check_chunk_xattr(self, target, obj_meta, xattr_meta):
"""
Check coherency of chunk extended attributes with object metadata.
:returns: a list of errors
"""
errors = list()
# Composed position -> erasure coding
attr_prefix = 'meta' if '.' in obj_meta['pos'] else ''
attr_key = attr_prefix + 'chunk_size'
if str(obj_meta['size']) != xattr_meta.get(attr_key):
errors.append(
"'%s' xattr (%s) differs from size in meta2 (%s)" %
(attr_key, xattr_meta.get(attr_key), obj_meta['size']))
attr_key = attr_prefix + 'chunk_hash'
if obj_meta['hash'] != xattr_meta.get(attr_key):
errors.append(
"'%s' xattr (%s) differs from hash in meta2 (%s)" %
(attr_key, xattr_meta.get(attr_key), obj_meta['hash']))
return errors
def _check_chunk(self, target):
"""
Execute various checks on a chunk:
- does it appear in object's chunk list?
- is it reachable?
- are its extended attributes coherent?
:returns: the list of errors encountered,
and the chunk's owner object metadata.
"""
chunk = target.chunk
errors = list()
obj_meta = None
xattr_meta = None
try:
xattr_meta = self.api.blob_client.chunk_head(
chunk, xattr=self.full, check_hash=self.integrity)
except exc.NotFound as err:
self.chunk_not_found += 1
errors.append('Not found: %s' % (err, ))
except exc.FaultyChunk as err:
self.chunk_exceptions += 1
errors.append('Faulty: %r' % (err, ))
except Exception as err:
self.chunk_exceptions += 1
errors.append('Check failed: %s' % (err, ))
if not target.obj and xattr_meta:
self.complete_target_from_chunk_metadata(target, xattr_meta)
if target.obj:
obj_listing, obj_meta = self.check_obj(target.copy_object())
if chunk not in obj_listing:
errors.append('Missing from object listing')
db_meta = dict()
else:
db_meta = obj_listing[chunk]
if db_meta and xattr_meta and self.full:
errors.extend(
self._check_chunk_xattr(target, db_meta, xattr_meta))
self.send_result(target, errors)
self.chunks_checked += 1
return errors, obj_meta
def check_chunk(self, target):
errors, _obj_meta = self._check_chunk(target)
return errors
def _check_metachunk(self, target, stg_met, pos, chunks, recurse=0):
"""
Check that a metachunk has the right number of chunks.
:returns: the list of errors
"""
required = stg_met.expected_chunks
errors = list()
if len(chunks) < required:
missing_chunks = required - len(chunks)
if stg_met.ec:
subs = {x['num'] for x in chunks}
for sub in range(required):
if sub not in subs:
errors.append("Missing chunk at position %d.%d" %
(pos, sub))
else:
for _ in range(missing_chunks):
errors.append("Missing chunk at position %d" % pos)
if recurse > 0:
for chunk in chunks:
tcopy = target.copy()
tcopy.chunk = chunk['url']
chunk_errors, _ = self._check_chunk(tcopy)
if chunk_errors:
# The errors have already been reported by _check_chunk,
# but we must count this chunk among the unusable chunks
# of the current metachunk.
errors.append("Unusable chunk %s at position %s" %
(chunk['url'], chunk['pos']))
irreparable = required - len(errors) < stg_met.min_chunks_to_read
if irreparable:
errors.append(
"Unavailable metachunk at position %s (%d/%d chunks)" %
(pos, required - len(errors), stg_met.expected_chunks))
# Since the "metachunk" is not an official item type,
# this method does not report errors itself. Errors will
# be reported as object errors.
return errors
def _check_obj_policy(self, target, obj_meta, chunks, recurse=0):
"""
Check that the list of chunks of an object matches
the object's storage policy.
:returns: the list of errors encountered
"""
stg_met = STORAGE_METHODS.load(obj_meta['chunk_method'])
chunks_by_pos = _sort_chunks(chunks, stg_met.ec)
tasks = list()
for pos, chunks in chunks_by_pos.iteritems():
tasks.append((pos,
self.pool.spawn(self._check_metachunk,
target.copy(),
stg_met,
pos,
chunks,
recurse=recurse)))
errors = list()
for pos, task in tasks:
try:
errors.extend(task.wait())
except Exception as err:
errors.append("Check failed: pos %d: %s" % (pos, err))
return errors
def check_obj_versions(self, target, versions, recurse=0):
"""
Run checks of all versions of the targeted object in parallel.
"""
tasks = list()
for ov in versions:
tcopy = target.copy_object()
tcopy.content_id = ov['id']
tcopy.version = str(ov['version'])
tasks.append((tcopy.version,
self.pool.spawn(self.check_obj,
tcopy,
recurse=recurse)))
errors = list()
for version, task in tasks:
try:
task.wait()
except Exception as err:
errors.append("Check failed: version %s: %s" % (version, err))
if errors:
# Send a result with the target without version to tell
# we were not able to check all versions of the object.
self.send_result(target, errors)
def _load_obj_meta(self, target, errors):
"""
Load object metadata and chunks.
:param target: which object to check.
:param errors: list of errors that will be appended
in case any error occurs.
:returns: a tuple with object metadata and a list of chunks.
"""
try:
return self.api.object_locate(target.account,
target.container,
target.obj,
version=target.version,
properties=False)
except exc.NoSuchObject as err:
self.object_not_found += 1
errors.append('Not found: %s' % (err, ))
except Exception as err:
self.object_exceptions += 1
errors.append('Check failed: %s' % (err, ))
return None, []
def check_obj(self, target, recurse=0):
"""
Check one object version.
If no version is specified, all versions of the object will be checked.
:returns: the result of the check of the most recent version,
or the one that is explicitly targeted.
"""
account = target.account
container = target.container
obj = target.obj
vers = target.version # can be None
if (account, container, obj, vers) in self.running:
self.running[(account, container, obj, vers)].wait()
if (account, container, obj, vers) in self.list_cache:
return self.list_cache[(account, container, obj, vers)]
self.running[(account, container, obj, vers)] = Event()
self.logger.info('Checking object "%s"', target)
container_listing, _ = self.check_container(target.copy_container())
errors = list()
if obj not in container_listing:
errors.append('Missing from container listing')
# checksum = None
else:
versions = container_listing[obj]
if vers is None:
if target.content_id is None:
# No version specified, check all versions
self.check_obj_versions(target.copy_object(),
versions,
recurse=recurse)
# Now return the cached result of the most recent version
target.content_id = versions[0]['id']
target.version = str(versions[0]['version'])
res = self.check_obj(target, recurse=0)
self.running[(account, container, obj, vers)].send(True)
del self.running[(account, container, obj, vers)]
return res
else:
for ov in versions:
if ov['id'] == target.content_id:
vers = str(ov['version'])
target.version = vers
break
else:
errors.append('Missing from container listing')
# TODO check checksum match
# checksum = container_listing[obj]['hash']
pass
meta, chunks = self._load_obj_meta(target, errors)
chunk_listing = {c['url']: c for c in chunks}
if meta:
self.list_cache[(account, container, obj, vers)] = \
(chunk_listing, meta)
self.objects_checked += 1
self.running[(account, container, obj, vers)].send(True)
del self.running[(account, container, obj, vers)]
# Skip the check if we could not locate the object
if meta:
errors.extend(
self._check_obj_policy(target, meta, chunks, recurse=recurse))
self.send_result(target, errors)
return chunk_listing, meta
def check_container(self, target, recurse=0):
account = target.account
container = target.container
if (account, container) in self.running:
self.running[(account, container)].wait()
if (account, container) in self.list_cache:
return self.list_cache[(account, container)]
self.running[(account, container)] = Event()
self.logger.info('Checking container "%s"', target)
account_listing = self.check_account(target.copy_account())
errors = list()
if container not in account_listing:
errors.append('Missing from account listing')
marker = None
results = []
ct_meta = dict()
extra_args = dict()
if self.limit_listings > 1 and target.obj:
# When we are explicitly checking one object, start the listing
# where this object is supposed to be. Do not use a limit,
# but an end marker, in order to fetch all versions of the object.
extra_args['prefix'] = target.obj
extra_args['end_marker'] = target.obj + '\x00' # HACK
while True:
try:
resp = self.api.object_list(account,
container,
marker=marker,
versions=True,
**extra_args)
except exc.NoSuchContainer as err:
self.container_not_found += 1
errors.append('Not found: %s' % (err, ))
break
except Exception as err:
self.container_exceptions += 1
errors.append('Check failed: %s' % (err, ))
break
if resp.get('truncated', False):
marker = resp['next_marker']
if resp['objects']:
# safeguard, probably useless
if not marker:
marker = resp['objects'][-1]['name']
results.extend(resp['objects'])
if self.limit_listings > 1:
break
else:
ct_meta = resp
ct_meta.pop('objects')
break
container_listing = dict()
# Save all object versions, with the most recent first
for obj in results:
container_listing.setdefault(obj['name'], list()).append(obj)
for versions in container_listing.values():
versions.sort(key=lambda o: o['version'], reverse=True)
if self.limit_listings <= 1:
# We just listed the whole container, keep the result in a cache
self.containers_checked += 1
self.list_cache[(account, container)] = container_listing, ct_meta
self.running[(account, container)].send(True)
del self.running[(account, container)]
if recurse > 0:
for obj_vers in container_listing.values():
for obj in obj_vers:
tcopy = target.copy_object()
tcopy.obj = obj['name']
tcopy.content_id = obj['id']
tcopy.version = str(obj['version'])
self.pool.spawn_n(self.check_obj, tcopy, recurse - 1)
self.send_result(target, errors)
return container_listing, ct_meta
def check_account(self, target, recurse=0):
account = target.account
if account in self.running:
self.running[account].wait()
if account in self.list_cache:
return self.list_cache[account]
self.running[account] = Event()
self.logger.info('Checking account "%s"', target)
errors = list()
marker = None
results = []
extra_args = dict()
if self.limit_listings > 0 and target.container:
# When we are explicitly checking one container, start the listing
# where this container is supposed to be, and list only one
# container.
extra_args['prefix'] = target.container
extra_args['limit'] = 1
while True:
try:
resp = self.api.container_list(account,
marker=marker,
**extra_args)
except Exception as err:
self.account_exceptions += 1
errors.append('Check failed: %s' % (err, ))
break
if resp:
marker = resp[-1][0]
results.extend(resp)
if self.limit_listings > 0:
break
else:
break
containers = dict()
for container in results:
# Name, number of objects, number of bytes
containers[container[0]] = (container[1], container[2])
if self.limit_listings <= 0:
# We just listed the whole account, keep the result in a cache
self.accounts_checked += 1
self.list_cache[account] = containers
self.running[account].send(True)
del self.running[account]
if recurse > 0:
for container in containers:
tcopy = target.copy_account()
tcopy.container = container
self.pool.spawn_n(self.check_container, tcopy, recurse - 1)
self.send_result(target, errors)
return containers
def check(self, target, recurse=0):
if target.type == 'chunk':
self.pool.spawn_n(self.check_chunk, target)
elif target.type == 'object':
self.pool.spawn_n(self.check_obj, target, recurse)
elif target.type == 'container':
self.pool.spawn_n(self.check_container, target, recurse)
else:
self.pool.spawn_n(self.check_account, target, recurse)
def fetch_results(self):
while not self.result_queue.empty():
res = self.result_queue.get(True)
yield res
def log_result(self, result):
if result.errors:
if result.target.type == 'chunk':
# FIXME(FVE): check error criticity
# and set the irreparable flag.
self.write_chunk_error(result.target)
else:
self.write_error(result.target)
self.logger.warn('%s:\n%s', result.target,
result.errors_to_str(err_format=' %s'))
def run(self):
"""
Fetch results and write logs until all jobs have finished.
:returns: a generator yielding check results.
"""
while self.pool.running() + self.pool.waiting():
for result in self.fetch_results():
self.log_result(result)
yield result
sleep(0.1)
self.pool.waitall()
for result in self.fetch_results():
self.log_result(result)
yield result
def report(self):
success = True
def _report_stat(name, stat):
print("{0:18}: {1}".format(name, stat))
print()
print('Report')
_report_stat("Accounts checked", self.accounts_checked)
if self.account_not_found:
success = False
_report_stat("Missing accounts", self.account_not_found)
if self.account_exceptions:
success = False
_report_stat("Exceptions", self.account_exceptions)
print()
_report_stat("Containers checked", self.containers_checked)
if self.container_not_found:
success = False
_report_stat("Missing containers", self.container_not_found)
if self.container_exceptions:
success = False
_report_stat("Exceptions", self.container_exceptions)
print()
_report_stat("Objects checked", self.objects_checked)
if self.object_not_found:
success = False
_report_stat("Missing objects", self.object_not_found)
if self.object_exceptions:
success = False
_report_stat("Exceptions", self.object_exceptions)
print()
_report_stat("Chunks checked", self.chunks_checked)
if self.chunk_not_found:
success = False
_report_stat("Missing chunks", self.chunk_not_found)
if self.chunk_exceptions:
success = False
_report_stat("Exceptions", self.chunk_exceptions)
return success
class Checker(object):
def __init__(self, namespace, concurrency=50,
error_file=None, rebuild_file=None, full=True,
limit_listings=0, request_attempts=1):
self.pool = GreenPool(concurrency)
self.error_file = error_file
self.full = bool(full)
# Optimisation for when we are only checking one object
# or one container.
# 0 -> do not limit
# 1 -> limit account listings (list of containers)
# 2 -> limit container listings (list of objects)
self.limit_listings = limit_listings
if self.error_file:
f = open(self.error_file, 'a')
self.error_writer = csv.writer(f, delimiter=' ')
self.rebuild_file = rebuild_file
if self.rebuild_file:
fd = open(self.rebuild_file, 'a')
self.rebuild_writer = csv.writer(fd, delimiter='|')
conf = {'namespace': namespace}
self.account_client = AccountClient(
conf,
max_retries=request_attempts - 1)
self.container_client = ContainerClient(
conf,
max_retries=request_attempts - 1,
request_attempts=request_attempts)
self.blob_client = BlobClient(conf=conf)
self.accounts_checked = 0
self.containers_checked = 0
self.objects_checked = 0
self.chunks_checked = 0
self.account_not_found = 0
self.container_not_found = 0
self.object_not_found = 0
self.chunk_not_found = 0
self.account_exceptions = 0
self.container_exceptions = 0
self.object_exceptions = 0
self.chunk_exceptions = 0
self.list_cache = {}
self.running = {}
def write_error(self, target):
error = [target.account]
if target.container:
error.append(target.container)
if target.obj:
error.append(target.obj)
if target.chunk:
error.append(target.chunk)
self.error_writer.writerow(error)
def write_rebuilder_input(self, target, obj_meta, ct_meta):
try:
cid = ct_meta['system']['sys.name'].split('.', 1)[0]
except KeyError:
cid = ct_meta['properties']['sys.name'].split('.', 1)[0]
self.rebuild_writer.writerow((cid, obj_meta['id'], target.chunk))
def write_chunk_error(self, target, obj_meta, chunk=None):
if chunk is not None:
target = target.copy()
target.chunk = chunk
if self.error_file:
self.write_error(target)
if self.rebuild_file:
self.write_rebuilder_input(
target, obj_meta,
self.list_cache[(target.account, target.container)][1])
def _check_chunk_xattr(self, target, obj_meta, xattr_meta):
error = False
# Composed position -> erasure coding
attr_prefix = 'meta' if '.' in obj_meta['pos'] else ''
attr_key = attr_prefix + 'chunk_size'
if str(obj_meta['size']) != xattr_meta.get(attr_key):
print(" Chunk %s '%s' xattr (%s) "
"differs from size in meta2 (%s)" %
(target, attr_key, xattr_meta.get(attr_key),
obj_meta['size']))
error = True
attr_key = attr_prefix + 'chunk_hash'
if obj_meta['hash'] != xattr_meta.get(attr_key):
print(" Chunk %s '%s' xattr (%s) "
"differs from hash in meta2 (%s)" %
(target, attr_key, xattr_meta.get(attr_key),
obj_meta['hash']))
error = True
return error
def check_chunk(self, target):
chunk = target.chunk
obj_listing, obj_meta = self.check_obj(target)
error = False
if chunk not in obj_listing:
print(' Chunk %s missing from object listing' % target)
error = True
db_meta = dict()
else:
db_meta = obj_listing[chunk]
try:
xattr_meta = self.blob_client.chunk_head(chunk, xattr=self.full)
except exc.NotFound as e:
self.chunk_not_found += 1
error = True
print(' Not found chunk "%s": %s' % (target, str(e)))
except Exception as e:
self.chunk_exceptions += 1
error = True
print(' Exception chunk "%s": %s' % (target, str(e)))
else:
if db_meta and self.full:
error = self._check_chunk_xattr(target, db_meta, xattr_meta)
if error:
self.write_chunk_error(target, obj_meta)
self.chunks_checked += 1
def check_obj_policy(self, target, obj_meta, chunks):
"""
Check that the list of chunks of an object matches
the object's storage policy.
"""
stg_met = STORAGE_METHODS.load(obj_meta['chunk_method'])
chunks_by_pos = _sort_chunks(chunks, stg_met.ec)
if stg_met.ec:
required = stg_met.ec_nb_data + stg_met.ec_nb_parity
else:
required = stg_met.nb_copy
for pos, clist in chunks_by_pos.iteritems():
if len(clist) < required:
print(' Missing %d chunks at position %s of %s' % (
required - len(clist), pos, target))
if stg_met.ec:
subs = {x['num'] for x in clist}
for sub in range(required):
if sub not in subs:
self.write_chunk_error(target, obj_meta,
'%d.%d' % (pos, sub))
else:
self.write_chunk_error(target, obj_meta, str(pos))
def check_obj(self, target, recurse=False):
account = target.account
container = target.container
obj = target.obj
if (account, container, obj) in self.running:
self.running[(account, container, obj)].wait()
if (account, container, obj) in self.list_cache:
return self.list_cache[(account, container, obj)]
self.running[(account, container, obj)] = Event()
print('Checking object "%s"' % target)
container_listing, ct_meta = self.check_container(target)
error = False
if obj not in container_listing:
print(' Object %s missing from container listing' % target)
error = True
# checksum = None
else:
# TODO check checksum match
# checksum = container_listing[obj]['hash']
pass
results = []
meta = dict()
try:
meta, results = self.container_client.content_locate(
account=account, reference=container, path=obj,
properties=False)
except exc.NotFound as e:
self.object_not_found += 1
error = True
print(' Not found object "%s": %s' % (target, str(e)))
except Exception as e:
self.object_exceptions += 1
error = True
print(' Exception object "%s": %s' % (target, str(e)))
chunk_listing = dict()
for chunk in results:
chunk_listing[chunk['url']] = chunk
# Skip the check if we could not locate the object
if meta:
self.check_obj_policy(target.copy(), meta, results)
self.list_cache[(account, container, obj)] = (chunk_listing, meta)
self.objects_checked += 1
self.running[(account, container, obj)].send(True)
del self.running[(account, container, obj)]
if recurse:
for chunk in chunk_listing:
t = target.copy()
t.chunk = chunk
self.pool.spawn_n(self.check_chunk, t)
if error and self.error_file:
self.write_error(target)
return chunk_listing, meta
def check_container(self, target, recurse=False):
account = target.account
container = target.container
if (account, container) in self.running:
self.running[(account, container)].wait()
if (account, container) in self.list_cache:
return self.list_cache[(account, container)]
self.running[(account, container)] = Event()
print('Checking container "%s"' % target)
account_listing = self.check_account(target)
error = False
if container not in account_listing:
error = True
print(' Container %s missing from account listing' % target)
marker = None
results = []
ct_meta = dict()
extra_args = dict()
if self.limit_listings > 1 and target.obj:
# When we are explicitly checking one object, start the listing
# where this object is supposed to be, and list only one object.
extra_args['prefix'] = target.obj
extra_args['limit'] = 1
while True:
try:
_, resp = self.container_client.content_list(
account=account, reference=container, marker=marker,
**extra_args)
except exc.NotFound as e:
self.container_not_found += 1
error = True
print(' Not found container "%s": %s' % (target, str(e)))
break
except Exception as e:
self.container_exceptions += 1
error = True
print(' Exception container "%s": %s' % (target, str(e)))
break
if resp['objects']:
marker = resp['objects'][-1]['name']
results.extend(resp['objects'])
if self.limit_listings > 1:
break
else:
ct_meta = resp
ct_meta.pop('objects')
break
container_listing = dict()
for obj in results:
container_listing[obj['name']] = obj
if self.limit_listings <= 1:
# We just listed the whole container, keep the result in a cache
self.containers_checked += 1
self.list_cache[(account, container)] = container_listing, ct_meta
self.running[(account, container)].send(True)
del self.running[(account, container)]
if recurse:
for obj in container_listing:
t = target.copy()
t.obj = obj
self.pool.spawn_n(self.check_obj, t, True)
if error and self.error_file:
self.write_error(target)
return container_listing, ct_meta
def check_account(self, target, recurse=False):
account = target.account
if account in self.running:
self.running[account].wait()
if account in self.list_cache:
return self.list_cache[account]
self.running[account] = Event()
print('Checking account "%s"' % target)
error = False
marker = None
results = []
extra_args = dict()
if self.limit_listings > 0 and target.container:
# When we are explicitly checking one container, start the listing
# where this container is supposed to be, and list only one
# container.
extra_args['prefix'] = target.container
extra_args['limit'] = 1
while True:
try:
resp = self.account_client.container_list(
account, marker=marker, **extra_args)
except Exception as e:
self.account_exceptions += 1
error = True
print(' Exception account "%s": %s' % (target, str(e)))
break
if resp['listing']:
marker = resp['listing'][-1][0]
results.extend(resp['listing'])
if self.limit_listings > 0:
break
else:
break
containers = dict()
for e in results:
containers[e[0]] = (e[1], e[2])
if self.limit_listings <= 0:
# We just listed the whole account, keep the result in a cache
self.accounts_checked += 1
self.list_cache[account] = containers
self.running[account].send(True)
del self.running[account]
if recurse:
for container in containers:
t = target.copy()
t.container = container
self.pool.spawn_n(self.check_container, t, True)
if error and self.error_file:
self.write_error(target)
return containers
def check(self, target):
if target.chunk and target.obj and target.container:
self.pool.spawn_n(self.check_chunk, target)
elif target.obj and target.container:
self.pool.spawn_n(self.check_obj, target, True)
elif target.container:
self.pool.spawn_n(self.check_container, target, True)
else:
self.pool.spawn_n(self.check_account, target, True)
def wait(self):
self.pool.waitall()
def report(self):
success = True
def _report_stat(name, stat):
print("{0:18}: {1}".format(name, stat))
print()
print('Report')
_report_stat("Accounts checked", self.accounts_checked)
if self.account_not_found:
success = False
_report_stat("Missing accounts", self.account_not_found)
if self.account_exceptions:
success = False
_report_stat("Exceptions", self.account_exceptions)
print()
_report_stat("Containers checked", self.containers_checked)
if self.container_not_found:
success = False
_report_stat("Missing containers", self.container_not_found)
if self.container_exceptions:
success = False
_report_stat("Exceptions", self.container_exceptions)
print()
_report_stat("Objects checked", self.objects_checked)
if self.object_not_found:
success = False
_report_stat("Missing objects", self.object_not_found)
if self.object_exceptions:
success = False
_report_stat("Exceptions", self.object_exceptions)
print()
_report_stat("Chunks checked", self.chunks_checked)
if self.chunk_not_found:
success = False
_report_stat("Missing chunks", self.chunk_not_found)
if self.chunk_exceptions:
success = False
_report_stat("Exceptions", self.chunk_exceptions)
return success