openstack/deb-cinder
Code Issues Proposed changes
ad052ea99a
deb-cinder/cinder/objects/base.py
John Griffith 106c14a84b Replication v2.1 (Cheesecake) 
This focuses the replication work on a specific use case,
and eliminates some of the ambiguity in earlier versions.

Additionally this implementation addresses needs for
devices that do replication based on the whole backend-device
or on Pools.

Use case:
  DR scenario, where a storage device is rendered inoperable.
  This implementation allows the preservation of user data
  for those volumes that are of type replication-enabled.

  The goal is NOT to make failures completely transparent
  but instead to preserve data access while an Admin tries
  to rebuild/recover his/her cloud.

It's very important to note that we're no longer interested in
dealing with replication in Cinder at a Volume level.  The concept
of have "some" volumes failover, and "others" left behind, proved
to not only be overly complex and difficult to implement, but we
never identified a concrete use-case where one would use failover
in a scenario where some volumes would stay and be accessible on
a primary but other may be moved and accessed via a secondary.

In this model, it's host/backend based.  So when you failover,
you're failing over an entire backend.  We heavily leverage
existing resources, specifically services, and capabilities.

Implements: blueprint replication-update

Change-Id: If862bcd18515098639f94a8294a8e44e1358c52a
2016-02-26 13:15:19 -07:00

424 lines
17 KiB
Python
Raw Blame History

# Copyright 2015 IBM Corp.
#
# 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.
"""Cinder common internal object model"""
import contextlib
import datetime
from oslo_log import log as logging
from oslo_utils import versionutils
from oslo_versionedobjects import base
from oslo_versionedobjects import fields
from cinder import db
from cinder import exception
from cinder.i18n import _
from cinder import objects
LOG = logging.getLogger('object')
remotable = base.remotable
remotable_classmethod = base.remotable_classmethod
obj_make_list = base.obj_make_list
class CinderObjectVersionsHistory(dict):
"""Helper class that maintains objects version history.
Current state of object versions is aggregated in a single version number
that explicitily identifies a set of object versions. That way a service
is able to report what objects it supports using a single string and all
the newer services will know exactly what that mean for a single object.
"""
def __init__(self):
super(CinderObjectVersionsHistory, self).__init__()
# NOTE(dulek): This is our pre-history and a starting point - Liberty.
# We want Mitaka to be able to talk to Liberty services, so we need to
# handle backporting to these objects versions (although I don't expect
# we've made a lot of incompatible changes inside the objects).
#
# If an object doesn't exist in Liberty, RPC API compatibility layer
# shouldn't send it or convert it to a dictionary.
#
# Please note that we do not need to add similar entires for each
# release. Liberty is here just for historical reasons.
self.versions = ['liberty']
self['liberty'] = {
'Backup': '1.1',
'BackupImport': '1.1',
'BackupList': '1.0',
'ConsistencyGroup': '1.1',
'ConsistencyGroupList': '1.0',
'Service': '1.0',
'ServiceList': '1.0',
'Snapshot': '1.0',
'SnapshotList': '1.0',
'Volume': '1.1',
'VolumeAttachment': '1.0',
'VolumeAttachmentList': '1.0',
'VolumeList': '1.1',
'VolumeType': '1.0',
'VolumeTypeList': '1.0',
}
def get_current(self):
return self.versions[-1]
def get_current_versions(self):
return self[self.get_current()]
def add(self, ver, updates):
self[ver] = self[self.get_current()].copy()
self.versions.append(ver)
self[ver].update(updates)
OBJ_VERSIONS = CinderObjectVersionsHistory()
# NOTE(dulek): You should add a new version here each time you bump a version
# of any object. As a second parameter you need to specify only what changed.
#
# When dropping backward compatibility with an OpenStack release we can rework
# this and remove some history while keeping the versions order.
OBJ_VERSIONS.add('1.0', {'Backup': '1.3', 'BackupImport': '1.3',
'CGSnapshot': '1.0', 'CGSnapshotList': '1.0',
'ConsistencyGroup': '1.2',
'ConsistencyGroupList': '1.1', 'Service': '1.1',
'Volume': '1.3', 'VolumeTypeList': '1.1'})
OBJ_VERSIONS.add('1.1', {'Service': '1.2', 'ServiceList': '1.1'})
OBJ_VERSIONS.add('1.2', {'Backup': '1.4', 'BackupImport': '1.4'})
OBJ_VERSIONS.add('1.3', {'Service': '1.3'})
class CinderObjectRegistry(base.VersionedObjectRegistry):
def registration_hook(self, cls, index):
setattr(objects, cls.obj_name(), cls)
# For Versioned Object Classes that have a model store the model in
# a Class attribute named model
try:
cls.model = db.get_model_for_versioned_object(cls)
except (ImportError, AttributeError):
pass
class CinderObject(base.VersionedObject):
# NOTE(thangp): OBJ_PROJECT_NAMESPACE needs to be set so that nova,
# cinder, and other objects can exist on the same bus and be distinguished
# from one another.
OBJ_PROJECT_NAMESPACE = 'cinder'
# NOTE(thangp): As more objects are added to cinder, each object should
# have a custom map of version compatibility. This just anchors the base
# version compatibility.
VERSION_COMPATIBILITY = {'7.0.0': '1.0'}
Not = db.Not
Case = db.Case
def cinder_obj_get_changes(self):
"""Returns a dict of changed fields with tz unaware datetimes.
Any timezone aware datetime field will be converted to UTC timezone
and returned as timezone unaware datetime.
This will allow us to pass these fields directly to a db update
method as they can't have timezone information.
"""
# Get dirtied/changed fields
changes = self.obj_get_changes()
# Look for datetime objects that contain timezone information
for k, v in changes.items():
if isinstance(v, datetime.datetime) and v.tzinfo:
# Remove timezone information and adjust the time according to
# the timezone information's offset.
changes[k] = v.replace(tzinfo=None) - v.utcoffset()
# Return modified dict
return changes
@base.remotable_classmethod
def get_by_id(cls, context, id, *args, **kwargs):
# To get by id we need to have a model and for the model to
# have an id field
if 'id' not in cls.fields:
msg = (_('VersionedObject %s cannot retrieve object by id.') %
(cls.obj_name()))
raise NotImplementedError(msg)
model = db.get_model_for_versioned_object(cls)
orm_obj = db.get_by_id(context, model, id, *args, **kwargs)
kargs = {}
if hasattr(cls, 'DEFAULT_EXPECTED_ATTR'):
kargs = {'expected_attrs': getattr(cls, 'DEFAULT_EXPECTED_ATTR')}
return cls._from_db_object(context, cls(context), orm_obj, **kargs)
def conditional_update(self, values, expected_values=None, filters=(),
save_all=False, session=None, reflect_changes=True):
"""Compare-and-swap update.
A conditional object update that, unlike normal update, will SAVE
the contents of the update to the DB.
Update will only occur in the DB and the object if conditions are
met.
If no expected_values are passed in we will default to make sure
that all fields have not been changed in the DB. Since we cannot
know the original value in the DB for dirty fields in the object
those will be excluded.
We have 4 different condition types we can use in expected_values:
- Equality: {'status': 'available'}
- Inequality: {'status': vol_obj.Not('deleting')}
- In range: {'status': ['available', 'error']
- Not in range: {'status': vol_obj.Not(['in-use', 'attaching'])
Method accepts additional filters, which are basically anything that
can be passed to a sqlalchemy query's filter method, for example:
[~sql.exists().where(models.Volume.id == models.Snapshot.volume_id)]
We can select values based on conditions using Case objects in the
'values' argument. For example:
has_snapshot_filter = sql.exists().where(
models.Snapshot.volume_id == models.Volume.id)
case_values = volume.Case([(has_snapshot_filter, 'has-snapshot')],
else_='no-snapshot')
volume.conditional_update({'status': case_values},
{'status': 'available'}))
And we can use DB fields using model class attribute for example to
store previous status in the corresponding field even though we
don't know which value is in the db from those we allowed:
volume.conditional_update({'status': 'deleting',
'previous_status': volume.model.status},
{'status': ('available', 'error')})
:param values: Dictionary of key-values to update in the DB.
:param expected_values: Dictionary of conditions that must be met
for the update to be executed.
:param filters: Iterable with additional filters
:param save_all: Object may have changes that are not in the DB,
this will say whether we want those changes saved
as well.
:param session: Session to use for the update
:param reflect_changes: If we want changes made in the database to
be reflected in the versioned object. This
may mean in some cases that we have to
reload the object from the database.
:returns number of db rows that were updated, which can be used as a
boolean, since it will be 0 if we couldn't update the DB
and 1 if we could, because we are using unique index id.
"""
if 'id' not in self.fields:
msg = (_('VersionedObject %s does not support conditional update.')
% (self.obj_name()))
raise NotImplementedError(msg)
# If no conditions are set we will require object in DB to be unchanged
if expected_values is None:
changes = self.obj_what_changed()
expected = {key: getattr(self, key)
for key in self.fields.keys()
if self.obj_attr_is_set(key) and key not in changes and
key not in self.OPTIONAL_FIELDS}
else:
# Set the id in expected_values to limit conditional update to only
# change this object
expected = expected_values.copy()
expected['id'] = self.id
# If we want to save any additional changes the object has besides the
# ones referred in values
if save_all:
changes = self.cinder_obj_get_changes()
changes.update(values)
values = changes
result = db.conditional_update(self._context, self.model, values,
expected, filters)
# If we were able to update the DB then we need to update this object
# as well to reflect new DB contents and clear the object's dirty flags
# for those fields.
if result and reflect_changes:
# If we have used a Case, a db field or an expression in values we
# don't know which value was used, so we need to read the object
# back from the DB
if any(isinstance(v, self.Case) or db.is_orm_value(v)
for v in values.values()):
# Read back object from DB
obj = type(self).get_by_id(self._context, self.id)
db_values = obj.obj_to_primitive()['versioned_object.data']
# Only update fields were changes were requested
values = {field: db_values[field]
for field, value in values.items()}
# NOTE(geguileo): We don't use update method because our objects
# will eventually move away from VersionedObjectDictCompat
for key, value in values.items():
setattr(self, key, value)
self.obj_reset_changes(values.keys())
return result
def refresh(self):
# To refresh we need to have a model and for the model to have an id
# field
if 'id' not in self.fields:
msg = (_('VersionedObject %s cannot retrieve object by id.') %
(self.obj_name()))
raise NotImplementedError(msg)
current = self.get_by_id(self._context, self.id)
for field in self.fields:
# Only update attributes that are already set. We do not want to
# unexpectedly trigger a lazy-load.
if self.obj_attr_is_set(field):
if self[field] != current[field]:
self[field] = current[field]
self.obj_reset_changes()
def __contains__(self, name):
# We're using obj_extra_fields to provide aliases for some fields while
# in transition period. This override is to make these aliases pass
# "'foo' in obj" tests.
return name in self.obj_extra_fields or super(CinderObject,
self).__contains__(name)
class CinderObjectDictCompat(base.VersionedObjectDictCompat):
"""Mix-in to provide dictionary key access compat.
If an object needs to support attribute access using
dictionary items instead of object attributes, inherit
from this class. This should only be used as a temporary
measure until all callers are converted to use modern
attribute access.
NOTE(berrange) This class will eventually be deleted.
"""
def get(self, key, value=base._NotSpecifiedSentinel):
"""For backwards-compatibility with dict-based objects.
NOTE(danms): May be removed in the future.
"""
if key not in self.obj_fields:
# NOTE(jdg): There are a number of places where we rely on the
# old dictionary version and do a get(xxx, None).
# The following preserves that compatibility but in
# the future we'll remove this shim altogether so don't
# rely on it.
LOG.debug('Cinder object %(object_name)s has no '
'attribute named: %(attribute_name)s',
{'object_name': self.__class__.__name__,
'attribute_name': key})
return None
if (value != base._NotSpecifiedSentinel and
not self.obj_attr_is_set(key)):
return value
else:
try:
return getattr(self, key)
except (exception.ObjectActionError, NotImplementedError):
# Exception when haven't set a value for non-lazy
# loadable attribute, but to mimic typical dict 'get'
# behavior we should still return None
return None
class CinderPersistentObject(object):
"""Mixin class for Persistent objects.
This adds the fields that we use in common for all persistent objects.
"""
fields = {
'created_at': fields.DateTimeField(nullable=True),
'updated_at': fields.DateTimeField(nullable=True),
'deleted_at': fields.DateTimeField(nullable=True),
'deleted': fields.BooleanField(default=False),
}
@contextlib.contextmanager
def obj_as_admin(self):
"""Context manager to make an object call as an admin.
This temporarily modifies the context embedded in an object to
be elevated() and restores it after the call completes. Example
usage:
with obj.obj_as_admin():
obj.save()
"""
if self._context is None:
raise exception.OrphanedObjectError(method='obj_as_admin',
objtype=self.obj_name())
original_context = self._context
self._context = self._context.elevated()
try:
yield
finally:
self._context = original_context
class CinderComparableObject(base.ComparableVersionedObject):
def __eq__(self, obj):
if hasattr(obj, 'obj_to_primitive'):
return self.obj_to_primitive() == obj.obj_to_primitive()
return False
class ObjectListBase(base.ObjectListBase):
pass
class CinderObjectSerializer(base.VersionedObjectSerializer):
OBJ_BASE_CLASS = CinderObject
def __init__(self, version_cap=None):
super(CinderObjectSerializer, self).__init__()
self.version_cap = version_cap
def _get_capped_obj_version(self, obj):
objname = obj.obj_name()
version_dict = OBJ_VERSIONS.get(self.version_cap, {})
version_cap = version_dict.get(objname, None)
if version_cap:
cap_tuple = versionutils.convert_version_to_tuple(version_cap)
obj_tuple = versionutils.convert_version_to_tuple(obj.VERSION)
if cap_tuple > obj_tuple:
# NOTE(dulek): Do not set version cap to be higher than actual
# object version as we don't support "forwardporting" of
# objects. If service will receive an object that's too old it
# should handle it explicitly.
version_cap = None
return version_cap
def serialize_entity(self, context, entity):
if isinstance(entity, (tuple, list, set, dict)):
entity = self._process_iterable(context, self.serialize_entity,
entity)
elif (hasattr(entity, 'obj_to_primitive') and
callable(entity.obj_to_primitive)):
# NOTE(dulek): Backport outgoing object to the capped version.
backport_ver = self._get_capped_obj_version(entity)
entity = entity.obj_to_primitive(backport_ver)
return entity
View Git Blame Copy Permalink