def POST(self, req):
"""
Handles Delete Multiple Objects.
"""
def object_key_iter(elem):
for obj in elem.iterchildren('Object'):
key = obj.find('./Key').text
if not key:
raise UserKeyMustBeSpecified()
version = obj.find('./VersionId')
if version is not None:
version = version.text
yield key, version
max_body_size = min(
# FWIW, AWS limits multideletes to 1000 keys, and swift limits
# object names to 1024 bytes (by default). Add a factor of two to
# allow some slop.
2 * self.conf.max_multi_delete_objects * MAX_OBJECT_NAME_LENGTH,
# But, don't let operators shoot themselves in the foot
10 * 1024 * 1024)
try:
xml = req.xml(max_body_size)
if not xml:
raise MissingRequestBodyError()
req.check_md5(xml)
elem = fromstring(xml, 'Delete', self.logger)
quiet = elem.find('./Quiet')
if quiet is not None and quiet.text.lower() == 'true':
self.quiet = True
else:
self.quiet = False
delete_list = list(object_key_iter(elem))
if len(delete_list) > self.conf.max_multi_delete_objects:
raise MalformedXML()
except (XMLSyntaxError, DocumentInvalid):
raise MalformedXML()
except ErrorResponse:
raise
except Exception as e:
self.logger.error(e)
raise
elem = Element('DeleteResult')
# check bucket existence
try:
req.get_response(self.app, 'HEAD')
except AccessDenied as error:
body = self._gen_error_body(error, elem, delete_list)
return HTTPOk(body=body)
if any(version is not None for _key, version in delete_list):
# TODO: support deleting specific versions of objects
raise S3NotImplemented()
def do_delete(base_req, key, version):
req = copy.copy(base_req)
req.environ = copy.copy(base_req.environ)
req.object_name = key
try:
query = req.gen_multipart_manifest_delete_query(self.app)
resp = req.get_response(self.app,
method='DELETE',
query=query,
headers={'Accept': 'application/json'})
# Have to read the response to actually do the SLO delete
if query:
try:
delete_result = json.loads(resp.body)
if delete_result['Errors']:
# NB: bulk includes 404s in "Number Not Found",
# not "Errors"
msg_parts = [delete_result['Response Status']]
msg_parts.extend(
'%s: %s' % (obj, status)
for obj, status in delete_result['Errors'])
return key, {
'code': 'SLODeleteError',
'message': '\n'.join(msg_parts)
}
# else, all good
except (ValueError, TypeError, KeyError):
# Logs get all the gory details
self.logger.exception(
'Could not parse SLO delete response: %r',
resp.body)
# Client gets something more generic
return key, {
'code': 'SLODeleteError',
'message': 'Unexpected swift response'
}
except NoSuchKey:
pass
except ErrorResponse as e:
return key, {'code': e.__class__.__name__, 'message': e._msg}
return key, None
with StreamingPile(self.conf.multi_delete_concurrency) as pile:
for key, err in pile.asyncstarmap(
do_delete,
((req, key, version) for key, version in delete_list)):
if err:
error = SubElement(elem, 'Error')
SubElement(error, 'Key').text = key
SubElement(error, 'Code').text = err['code']
SubElement(error, 'Message').text = err['message']
elif not self.quiet:
deleted = SubElement(elem, 'Deleted')
SubElement(deleted, 'Key').text = key
body = tostring(elem)
return HTTPOk(body=body)
def handle_delete_iter(self,
req,
objs_to_delete=None,
user_agent='BulkDelete',
swift_source='BD',
out_content_type='text/plain'):
"""
A generator that can be assigned to a swob Response's app_iter which,
when iterated over, will delete the objects specified in request body.
Will occasionally yield whitespace while request is being processed.
When the request is completed will yield a response body that can be
parsed to determine success. See above documentation for details.
:params req: a swob Request
:params objs_to_delete: a list of dictionaries that specifies the
(native string) objects to be deleted. If None, uses
self.get_objs_to_delete to query request.
"""
last_yield = time()
if out_content_type and out_content_type.endswith('/xml'):
to_yield = b'<?xml version="1.0" encoding="UTF-8"?>\n'
else:
to_yield = b' '
separator = b''
failed_files = []
resp_dict = {
'Response Status': HTTPOk().status,
'Response Body': '',
'Number Deleted': 0,
'Number Not Found': 0
}
req.environ['eventlet.minimum_write_chunk_size'] = 0
try:
if not out_content_type:
raise HTTPNotAcceptable(request=req)
try:
vrs, account, _junk = req.split_path(2, 3, True)
except ValueError:
raise HTTPNotFound(request=req)
vrs = wsgi_to_str(vrs)
account = wsgi_to_str(account)
incoming_format = req.headers.get('Content-Type')
if incoming_format and \
not incoming_format.startswith('text/plain'):
# For now only accept newline separated object names
raise HTTPNotAcceptable(request=req)
if objs_to_delete is None:
objs_to_delete = self.get_objs_to_delete(req)
failed_file_response = {'type': HTTPBadRequest}
def delete_filter(predicate, objs_to_delete):
for obj_to_delete in objs_to_delete:
obj_name = obj_to_delete['name']
if not obj_name:
continue
if not predicate(obj_name):
continue
if obj_to_delete.get('error'):
if obj_to_delete['error']['code'] == HTTP_NOT_FOUND:
resp_dict['Number Not Found'] += 1
else:
failed_files.append([
wsgi_quote(str_to_wsgi(obj_name)),
obj_to_delete['error']['message']
])
continue
delete_path = '/'.join(
['', vrs, account,
obj_name.lstrip('/')])
if not constraints.check_utf8(delete_path):
failed_files.append([
wsgi_quote(str_to_wsgi(obj_name)),
HTTPPreconditionFailed().status
])
continue
yield (obj_name, delete_path,
obj_to_delete.get('version_id'))
def objs_then_containers(objs_to_delete):
# process all objects first
yield delete_filter(lambda name: '/' in name.strip('/'),
objs_to_delete)
# followed by containers
yield delete_filter(lambda name: '/' not in name.strip('/'),
objs_to_delete)
def do_delete(obj_name, delete_path, version_id):
delete_obj_req = make_subrequest(
req.environ,
method='DELETE',
path=wsgi_quote(str_to_wsgi(delete_path)),
headers={'X-Auth-Token': req.headers.get('X-Auth-Token')},
body='',
agent='%(orig)s ' + user_agent,
swift_source=swift_source)
if version_id is None:
delete_obj_req.params = {}
else:
delete_obj_req.params = {'version-id': version_id}
return (delete_obj_req.get_response(self.app), obj_name, 0)
with StreamingPile(self.delete_concurrency) as pile:
for names_to_delete in objs_then_containers(objs_to_delete):
for resp, obj_name, retry in pile.asyncstarmap(
do_delete, names_to_delete):
if last_yield + self.yield_frequency < time():
last_yield = time()
yield to_yield
to_yield, separator = b' ', b'\r\n\r\n'
self._process_delete(resp, pile, obj_name, resp_dict,
failed_files,
failed_file_response, retry)
if len(failed_files) >= self.max_failed_deletes:
# Abort, but drain off the in-progress deletes
for resp, obj_name, retry in pile:
if last_yield + self.yield_frequency < time():
last_yield = time()
yield to_yield
to_yield, separator = b' ', b'\r\n\r\n'
# Don't pass in the pile, as we shouldn't retry
self._process_delete(resp, None, obj_name,
resp_dict, failed_files,
failed_file_response,
retry)
msg = 'Max delete failures exceeded'
raise HTTPBadRequest(msg)
if failed_files:
resp_dict['Response Status'] = \
failed_file_response['type']().status
elif not (resp_dict['Number Deleted']
or resp_dict['Number Not Found']):
resp_dict['Response Status'] = HTTPBadRequest().status
resp_dict['Response Body'] = 'Invalid bulk delete.'
except HTTPException as err:
resp_dict['Response Status'] = err.status
resp_dict['Response Body'] = err.body.decode('utf-8')
except Exception:
self.logger.exception('Error in bulk delete.')
resp_dict['Response Status'] = HTTPServerError().status
yield separator + get_response_body(out_content_type, resp_dict,
failed_files, 'delete')
def handle_multipart_put(self, req, start_response):
"""
Will handle the PUT of a SLO manifest.
Heads every object in manifest to check if is valid and if so will
save a manifest generated from the user input. Uses WSGIContext to
call self and start_response and returns a WSGI iterator.
:params req: a swob.Request with an obj in path
:raises: HttpException on errors
"""
try:
vrs, account, container, obj = req.split_path(1, 4, True)
except ValueError:
return self.app(req.environ, start_response)
if req.content_length > self.max_manifest_size:
raise HTTPRequestEntityTooLarge("Manifest File > %d bytes" %
self.max_manifest_size)
if req.headers.get('X-Copy-From'):
raise HTTPMethodNotAllowed(
'Multipart Manifest PUTs cannot be COPY requests')
if req.content_length is None and \
req.headers.get('transfer-encoding', '').lower() != 'chunked':
raise HTTPLengthRequired(request=req)
parsed_data = parse_and_validate_input(
req.body_file.read(self.max_manifest_size), req.path)
problem_segments = []
if len(parsed_data) > self.max_manifest_segments:
raise HTTPRequestEntityTooLarge(
'Number of segments must be <= %d' %
self.max_manifest_segments)
total_size = 0
out_content_type = req.accept.best_match(ACCEPTABLE_FORMATS)
if not out_content_type:
out_content_type = 'text/plain'
data_for_storage = []
path2indices = defaultdict(list)
for index, seg_dict in enumerate(parsed_data):
path2indices[seg_dict['path']].append(index)
def do_head(obj_name):
obj_path = '/'.join(
['', vrs, account,
get_valid_utf8_str(obj_name).lstrip('/')])
sub_req = make_subrequest(
req.environ,
path=obj_path + '?', # kill the query string
method='HEAD',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent='%(orig)s SLO MultipartPUT',
swift_source='SLO')
return obj_name, sub_req.get_response(self)
def validate_seg_dict(seg_dict, head_seg_resp):
if not head_seg_resp.is_success:
problem_segments.append(
[quote(obj_name), head_seg_resp.status])
return 0, None
segment_length = head_seg_resp.content_length
if seg_dict.get('range'):
# Since we now know the length, we can normalize the
# range. We know that there is exactly one range
# requested since we checked that earlier in
# parse_and_validate_input().
ranges = seg_dict['range'].ranges_for_length(
head_seg_resp.content_length)
if not ranges:
problem_segments.append(
[quote(obj_name), 'Unsatisfiable Range'])
elif ranges == [(0, head_seg_resp.content_length)]:
# Just one range, and it exactly matches the object.
# Why'd we do this again?
del seg_dict['range']
segment_length = head_seg_resp.content_length
else:
rng = ranges[0]
seg_dict['range'] = '%d-%d' % (rng[0], rng[1] - 1)
segment_length = rng[1] - rng[0]
if segment_length < 1:
problem_segments.append([
quote(obj_name),
'Too small; each segment must be at least 1 byte.'
])
if seg_dict['size_bytes'] is not None and \
seg_dict['size_bytes'] != head_seg_resp.content_length:
problem_segments.append([quote(obj_name), 'Size Mismatch'])
if seg_dict['etag'] is not None and \
seg_dict['etag'] != head_seg_resp.etag:
problem_segments.append([quote(obj_name), 'Etag Mismatch'])
if head_seg_resp.last_modified:
last_modified = head_seg_resp.last_modified
else:
# shouldn't happen
last_modified = datetime.now()
last_modified_formatted = \
last_modified.strftime('%Y-%m-%dT%H:%M:%S.%f')
seg_data = {
'name': '/' + seg_dict['path'].lstrip('/'),
'bytes': head_seg_resp.content_length,
'hash': head_seg_resp.etag,
'content_type': head_seg_resp.content_type,
'last_modified': last_modified_formatted
}
if seg_dict.get('range'):
seg_data['range'] = seg_dict['range']
if config_true_value(
head_seg_resp.headers.get('X-Static-Large-Object')):
seg_data['sub_slo'] = True
return segment_length, seg_data
data_for_storage = [None] * len(parsed_data)
with StreamingPile(self.concurrency) as pile:
for obj_name, resp in pile.asyncstarmap(
do_head, ((path, ) for path in path2indices)):
for i in path2indices[obj_name]:
segment_length, seg_data = validate_seg_dict(
parsed_data[i], resp)
data_for_storage[i] = seg_data
total_size += segment_length
if problem_segments:
resp_body = get_response_body(out_content_type, {},
problem_segments)
raise HTTPBadRequest(resp_body, content_type=out_content_type)
slo_etag = md5()
for seg_data in data_for_storage:
if seg_data.get('range'):
slo_etag.update('%s:%s;' %
(seg_data['hash'], seg_data['range']))
else:
slo_etag.update(seg_data['hash'])
slo_etag = slo_etag.hexdigest()
req.headers.update({
SYSMETA_SLO_ETAG: slo_etag,
SYSMETA_SLO_SIZE: total_size,
'X-Static-Large-Object': 'True',
})
json_data = json.dumps(data_for_storage)
if six.PY3:
json_data = json_data.encode('utf-8')
req.body = json_data
env = req.environ
if not env.get('CONTENT_TYPE'):
guessed_type, _junk = mimetypes.guess_type(req.path_info)
env['CONTENT_TYPE'] = guessed_type or 'application/octet-stream'
env['swift.content_type_overridden'] = True
env['CONTENT_TYPE'] += ";swift_bytes=%d" % total_size
def start_response_wrapper(status, headers, exc_info=None):
for i, (header, _value) in enumerate(headers):
if header.lower() == 'etag':
headers[i] = ('Etag', '"%s"' % slo_etag)
break
return start_response(status, headers, exc_info)
return self.app(env, start_response_wrapper)
def POST(self, req):
"""
Handles Delete Multiple Objects.
"""
def object_key_iter(elem):
for obj in elem.iterchildren('Object'):
key = obj.find('./Key').text
if not key:
raise UserKeyMustBeSpecified()
version = obj.find('./VersionId')
if version is not None:
version = version.text
yield key, version
max_body_size = min(
# FWIW, AWS limits multideletes to 1000 keys, and swift limits
# object names to 1024 bytes (by default). Add a factor of two to
# allow some slop.
2 * self.conf.max_multi_delete_objects * MAX_OBJECT_NAME_LENGTH,
# But, don't let operators shoot themselves in the foot
10 * 1024 * 1024)
try:
xml = req.xml(max_body_size)
if not xml:
raise MissingRequestBodyError()
req.check_md5(xml)
elem = fromstring(xml, 'Delete', self.logger)
quiet = elem.find('./Quiet')
if quiet is not None and quiet.text.lower() == 'true':
self.quiet = True
else:
self.quiet = False
delete_list = list(object_key_iter(elem))
if len(delete_list) > self.conf.max_multi_delete_objects:
raise MalformedXML()
except (XMLSyntaxError, DocumentInvalid):
raise MalformedXML()
except ErrorResponse:
raise
except Exception as e:
self.logger.error(e)
raise
elem = Element('DeleteResult')
# check bucket existence
try:
req.get_response(self.app, 'HEAD')
except AccessDenied as error:
body = self._gen_error_body(error, elem, delete_list)
return HTTPOk(body=body)
if any(version is not None for _key, version in delete_list):
# TODO: support deleting specific versions of objects
raise S3NotImplemented()
def do_delete(base_req, key, version):
req = copy.copy(base_req)
req.environ = copy.copy(base_req.environ)
req.object_name = key
try:
query = req.gen_multipart_manifest_delete_query(self.app)
req.get_response(self.app, method='DELETE', query=query)
except NoSuchKey:
pass
except ErrorResponse as e:
return key, {'code': e.__class__.__name__, 'message': e._msg}
return key, None
with StreamingPile(self.conf.multi_delete_concurrency) as pile:
for key, err in pile.asyncstarmap(
do_delete,
((req, key, version) for key, version in delete_list)):
if err:
error = SubElement(elem, 'Error')
SubElement(error, 'Key').text = key
SubElement(error, 'Code').text = err['code']
SubElement(error, 'Message').text = err['message']
elif not self.quiet:
deleted = SubElement(elem, 'Deleted')
SubElement(deleted, 'Key').text = key
body = tostring(elem)
return HTTPOk(body=body)