393a3bd947
This reverts commit e74b45d0eb
.
This re-adds the non-deterministic sort logic but avoids using
boolean columns to work around the related bug.
Original commit msg:
With the switch to subquery loading, we need to eliminate the
possibility for non-deterministic results in the event of LIMIT
queries caused by non-unique SORT BY parameters.
This always adds on a set of unique keys to sort by when sorting
params are present. By placing them at the end, they take the lower
priority over the user-provided sort strings.
1. http://docs.sqlalchemy.org/en/latest/orm/loading_relationships.html#subqueryload-ordering
2. http://docs.sqlalchemy.org/en/latest/faq/ormconfiguration.html#faq-subqueryload-limit-sort
Related-Bug: #1656947
Change-Id: I503c65560d970427be93b39898f4e7acfc4d3e1f
283 lines
13 KiB
Python
283 lines
13 KiB
Python
# Copyright (c) 2014 OpenStack Foundation.
|
|
# All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License"); you may
|
|
# not use this file except in compliance with the License. You may obtain
|
|
# a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
|
|
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
|
|
# License for the specific language governing permissions and limitations
|
|
# under the License.
|
|
|
|
import weakref
|
|
|
|
from neutron_lib.db import utils as db_utils
|
|
from oslo_db.sqlalchemy import utils as sa_utils
|
|
import six
|
|
from sqlalchemy import and_
|
|
from sqlalchemy.ext import associationproxy
|
|
from sqlalchemy import or_
|
|
from sqlalchemy import sql
|
|
|
|
from neutron.api.v2 import attributes
|
|
from neutron.db import _utils as ndb_utils
|
|
|
|
|
|
# TODO(HenryG): Remove these when available in neutron-lib
|
|
safe_creation = ndb_utils.safe_creation
|
|
model_query_scope = ndb_utils.model_query_scope_is_project
|
|
model_query = ndb_utils.model_query
|
|
resource_fields = ndb_utils.resource_fields
|
|
|
|
|
|
class CommonDbMixin(object):
|
|
"""Common methods used in core and service plugins."""
|
|
# Plugins, mixin classes implementing extension will register
|
|
# hooks into the dict below for "augmenting" the "core way" of
|
|
# building a query for retrieving objects from a model class.
|
|
# To this aim, the register_model_query_hook and unregister_query_hook
|
|
# from this class should be invoked
|
|
_model_query_hooks = {}
|
|
|
|
# This dictionary will store methods for extending attributes of
|
|
# api resources. Mixins can use this dict for adding their own methods
|
|
# TODO(salvatore-orlando): Avoid using class-level variables
|
|
_dict_extend_functions = {}
|
|
|
|
@classmethod
|
|
def register_model_query_hook(cls, model, name, query_hook, filter_hook,
|
|
result_filters=None):
|
|
"""Register a hook to be invoked when a query is executed.
|
|
|
|
Add the hooks to the _model_query_hooks dict. Models are the keys
|
|
of this dict, whereas the value is another dict mapping hook names to
|
|
callables performing the hook.
|
|
Each hook has a "query" component, used to build the query expression
|
|
and a "filter" component, which is used to build the filter expression.
|
|
|
|
Query hooks take as input the query being built and return a
|
|
transformed query expression.
|
|
|
|
Filter hooks take as input the filter expression being built and return
|
|
a transformed filter expression
|
|
"""
|
|
cls._model_query_hooks.setdefault(model, {})[name] = {
|
|
'query': query_hook, 'filter': filter_hook,
|
|
'result_filters': result_filters}
|
|
|
|
@classmethod
|
|
def register_dict_extend_funcs(cls, resource, funcs):
|
|
cls._dict_extend_functions.setdefault(resource, []).extend(funcs)
|
|
|
|
@property
|
|
def safe_reference(self):
|
|
"""Return a weakref to the instance.
|
|
|
|
Minimize the potential for the instance persisting
|
|
unnecessarily in memory by returning a weakref proxy that
|
|
won't prevent deallocation.
|
|
"""
|
|
return weakref.proxy(self)
|
|
|
|
# TODO(HenryG): Remove this when available in neutron-lib
|
|
def model_query_scope(self, context, model):
|
|
return ndb_utils.model_query_scope_is_project(context, model)
|
|
|
|
def _model_query(self, context, model):
|
|
query = context.session.query(model)
|
|
# define basic filter condition for model query
|
|
query_filter = None
|
|
if ndb_utils.model_query_scope_is_project(context, model):
|
|
if hasattr(model, 'rbac_entries'):
|
|
query = query.outerjoin(model.rbac_entries)
|
|
rbac_model = model.rbac_entries.property.mapper.class_
|
|
query_filter = (
|
|
(model.tenant_id == context.tenant_id) |
|
|
((rbac_model.action == 'access_as_shared') &
|
|
((rbac_model.target_tenant == context.tenant_id) |
|
|
(rbac_model.target_tenant == '*'))))
|
|
elif hasattr(model, 'shared'):
|
|
query_filter = ((model.tenant_id == context.tenant_id) |
|
|
(model.shared == sql.true()))
|
|
else:
|
|
query_filter = (model.tenant_id == context.tenant_id)
|
|
# Execute query hooks registered from mixins and plugins
|
|
for _name, hooks in six.iteritems(self._model_query_hooks.get(model,
|
|
{})):
|
|
query_hook = hooks.get('query')
|
|
if isinstance(query_hook, six.string_types):
|
|
query_hook = getattr(self, query_hook, None)
|
|
if query_hook:
|
|
query = query_hook(context, model, query)
|
|
|
|
filter_hook = hooks.get('filter')
|
|
if isinstance(filter_hook, six.string_types):
|
|
filter_hook = getattr(self, filter_hook, None)
|
|
if filter_hook:
|
|
query_filter = filter_hook(context, model, query_filter)
|
|
|
|
# NOTE(salvatore-orlando): 'if query_filter' will try to evaluate the
|
|
# condition, raising an exception
|
|
if query_filter is not None:
|
|
query = query.filter(query_filter)
|
|
return query
|
|
|
|
# TODO(HenryG): Remove this when available in neutron-lib
|
|
def _fields(self, resource, fields):
|
|
return ndb_utils.resource_fields(resource, fields)
|
|
|
|
def _get_by_id(self, context, model, id):
|
|
query = self._model_query(context, model)
|
|
return query.filter(model.id == id).one()
|
|
|
|
def _apply_filters_to_query(self, query, model, filters, context=None):
|
|
if filters:
|
|
for key, value in six.iteritems(filters):
|
|
column = getattr(model, key, None)
|
|
# NOTE(kevinbenton): if column is a hybrid property that
|
|
# references another expression, attempting to convert to
|
|
# a boolean will fail so we must compare to None.
|
|
# See "An Important Expression Language Gotcha" in:
|
|
# docs.sqlalchemy.org/en/rel_0_9/changelog/migration_06.html
|
|
if column is not None:
|
|
if not value:
|
|
query = query.filter(sql.false())
|
|
return query
|
|
if isinstance(column, associationproxy.AssociationProxy):
|
|
# association proxies don't support in_ so we have to
|
|
# do multiple equals matches
|
|
query = query.filter(
|
|
or_(*[column == v for v in value]))
|
|
else:
|
|
query = query.filter(column.in_(value))
|
|
elif key == 'shared' and hasattr(model, 'rbac_entries'):
|
|
# translate a filter on shared into a query against the
|
|
# object's rbac entries
|
|
rbac = model.rbac_entries.property.mapper.class_
|
|
matches = [rbac.target_tenant == '*']
|
|
if context:
|
|
matches.append(rbac.target_tenant == context.tenant_id)
|
|
# any 'access_as_shared' records that match the
|
|
# wildcard or requesting tenant
|
|
is_shared = and_(rbac.action == 'access_as_shared',
|
|
or_(*matches))
|
|
if not value[0]:
|
|
# NOTE(kevinbenton): we need to find objects that don't
|
|
# have an entry that matches the criteria above so
|
|
# we use a subquery to exclude them.
|
|
# We can't just filter the inverse of the query above
|
|
# because that will still give us a network shared to
|
|
# our tenant (or wildcard) if it's shared to another
|
|
# tenant.
|
|
# This is the column joining the table to rbac via
|
|
# the object_id. We can't just use model.id because
|
|
# subnets join on network.id so we have to inspect the
|
|
# relationship.
|
|
join_cols = model.rbac_entries.property.local_columns
|
|
oid_col = list(join_cols)[0]
|
|
is_shared = ~oid_col.in_(
|
|
query.session.query(rbac.object_id).
|
|
filter(is_shared)
|
|
)
|
|
elif (not context or
|
|
not self.model_query_scope(context, model)):
|
|
# we only want to join if we aren't using the subquery
|
|
# and if we aren't already joined because this is a
|
|
# scoped query
|
|
query = query.outerjoin(model.rbac_entries)
|
|
query = query.filter(is_shared)
|
|
for _nam, hooks in six.iteritems(self._model_query_hooks.get(model,
|
|
{})):
|
|
result_filter = hooks.get('result_filters', None)
|
|
if isinstance(result_filter, six.string_types):
|
|
result_filter = getattr(self, result_filter, None)
|
|
|
|
if result_filter:
|
|
query = result_filter(query, filters)
|
|
return query
|
|
|
|
def _apply_dict_extend_functions(self, resource_type,
|
|
response, db_object):
|
|
for func in self._dict_extend_functions.get(
|
|
resource_type, []):
|
|
args = (response, db_object)
|
|
if isinstance(func, six.string_types):
|
|
func = getattr(self, func, None)
|
|
else:
|
|
# must call unbound method - use self as 1st argument
|
|
args = (self,) + args
|
|
if func:
|
|
func(*args)
|
|
|
|
def _get_collection_query(self, context, model, filters=None,
|
|
sorts=None, limit=None, marker_obj=None,
|
|
page_reverse=False):
|
|
collection = self._model_query(context, model)
|
|
collection = self._apply_filters_to_query(collection, model, filters,
|
|
context)
|
|
if sorts:
|
|
sort_keys = db_utils.get_and_validate_sort_keys(sorts, model)
|
|
sort_dirs = db_utils.get_sort_dirs(sorts, page_reverse)
|
|
if limit:
|
|
# we always want deterministic results for limit subqueries
|
|
# so add unique keys to limit queries when present.
|
|
# (http://docs.sqlalchemy.org/en/latest/orm/
|
|
# loading_relationships.html#subqueryload-ordering)
|
|
# (http://docs.sqlalchemy.org/en/latest/faq/
|
|
# ormconfiguration.html#faq-subqueryload-limit-sort)
|
|
for k in self._unique_keys(model, marker_obj):
|
|
if k not in sort_keys:
|
|
sort_keys.append(k)
|
|
sort_dirs.append('asc')
|
|
collection = sa_utils.paginate_query(collection, model, limit,
|
|
marker=marker_obj,
|
|
sort_keys=sort_keys,
|
|
sort_dirs=sort_dirs)
|
|
return collection
|
|
|
|
def _unique_keys(self, model, marker_obj):
|
|
# just grab first set of unique keys and use them.
|
|
# if model has no unqiue sets, 'paginate_query' will
|
|
# warn if sorting is unstable
|
|
uk_sets = sa_utils.get_unique_keys(model)
|
|
for kset in uk_sets:
|
|
for k in kset:
|
|
if marker_obj and isinstance(getattr(marker_obj, k), bool):
|
|
# TODO(kevinbenton): workaround for bug/1656947.
|
|
# we can't use boolean cols until that bug is fixed. return
|
|
# first entry in uk_sets once that bug is resolved
|
|
break
|
|
else:
|
|
return kset
|
|
return []
|
|
|
|
def _get_collection(self, context, model, dict_func, filters=None,
|
|
fields=None, sorts=None, limit=None, marker_obj=None,
|
|
page_reverse=False):
|
|
query = self._get_collection_query(context, model, filters=filters,
|
|
sorts=sorts,
|
|
limit=limit,
|
|
marker_obj=marker_obj,
|
|
page_reverse=page_reverse)
|
|
items = [attributes.populate_project_info(dict_func(c, fields))
|
|
for c in query]
|
|
if limit and page_reverse:
|
|
items.reverse()
|
|
return items
|
|
|
|
def _get_collection_count(self, context, model, filters=None):
|
|
return self._get_collection_query(context, model, filters).count()
|
|
|
|
def _get_marker_obj(self, context, resource, limit, marker):
|
|
if limit and marker:
|
|
return getattr(self, '_get_%s' % resource)(context, marker)
|
|
return None
|
|
|
|
# TODO(HenryG): Remove this when available in neutron-lib
|
|
def _filter_non_model_columns(self, data, model):
|
|
return ndb_utils.filter_non_model_columns(data, model)
|