Bring back the separate web-side entity cache
Prior to changeset 635cfac73d28 "[repoapi] fold ClientConnection into
Connection", we had two entity caches: one on the client/dbapi/web side,
and one on the server/repo side. This is a waste, but it is actually
needed as long as we have the magic _cw attribute on entities which must
sometimes be a request and sometimes a cnx. Removing the duplication
caused weird problems with entity._cw alternating between both types of
objects, which is unexpected by both the repo and the web sides.
We add an entity cache on ConnectionCubicWebRequestBase, separate from
the Connection's, and bring back the
_cw_update_attr_cache/dont-cache-attrs mechanism, to let the
server/edition code let caches know which attributes have been modified
Entity.as_rset can be cached again, as ResultSet no longer modifies the
entity when fetching it from a cache.
Contrary to the pre-3.21 code, _cw_update_attr_cache now handles web
requests and connections in the same way (otherwise the cache ends up
with wrong values if a hook modifies attributes), but dont-cache-attrs
is never set for (inlined) relations.
Closes #6863543
# copyright 2014 LOGILAB S.A. (Paris, FRANCE), all rights reserved.
# contact http://www.logilab.fr/ -- mailto:contact@logilab.fr
#
# This file is part of CubicWeb.
#
# CubicWeb is free software: you can redistribute it and/or modify it under the
# terms of the GNU Lesser General Public License as published by the Free
# Software Foundation, either version 2.1 of the License, or (at your option)
# any later version.
#
# CubicWeb is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
# details.
#
# You should have received a copy of the GNU Lesser General Public License along
# with CubicWeb. If not, see <http://www.gnu.org/licenses/>.
"""Hooks for synchronizing computed attributes"""
__docformat__ = "restructuredtext en"
from cubicweb import _
from collections import defaultdict
from rql import nodes
from cubicweb.server import hook
class RecomputeAttributeOperation(hook.DataOperationMixIn, hook.Operation):
"""Operation to recompute caches of computed attribute at commit time,
depending on what's have been modified in the transaction and avoiding to
recompute twice the same attribute
"""
containercls = dict
def add_data(self, computed_attribute, eid=None):
try:
self._container[computed_attribute].add(eid)
except KeyError:
self._container[computed_attribute] = set((eid,))
def precommit_event(self):
for computed_attribute_rdef, eids in self.get_data().items():
attr = computed_attribute_rdef.rtype
formula = computed_attribute_rdef.formula
select = self.cnx.repo.vreg.rqlhelper.parse(formula).children[0]
xvar = select.get_variable('X')
select.add_selected(xvar, index=0)
select.add_group_var(xvar, index=0)
if None in eids:
select.add_type_restriction(xvar, computed_attribute_rdef.subject)
else:
select.add_eid_restriction(xvar, eids)
update_rql = 'SET X %s %%(value)s WHERE X eid %%(x)s' % attr
for eid, value in self.cnx.execute(select.as_string()):
self.cnx.execute(update_rql, {'value': value, 'x': eid})
class EntityWithCACreatedHook(hook.Hook):
"""When creating an entity that has some computed attribute, those
attributes have to be computed.
Concret class of this hook are generated at registration time by
introspecting the schema.
"""
__abstract__ = True
events = ('after_add_entity',)
# list of computed attribute rdefs that have to be recomputed
computed_attributes = None
def __call__(self):
for rdef in self.computed_attributes:
RecomputeAttributeOperation.get_instance(self._cw).add_data(
rdef, self.entity.eid)
class RelationInvolvedInCAModifiedHook(hook.Hook):
"""When some relation used in a computed attribute is updated, those
attributes have to be recomputed.
Concret class of this hook are generated at registration time by
introspecting the schema.
"""
__abstract__ = True
events = ('after_add_relation', 'before_delete_relation')
# list of (computed attribute rdef, optimize_on) that have to be recomputed
optimized_computed_attributes = None
def __call__(self):
for rdef, optimize_on in self.optimized_computed_attributes:
if optimize_on is None:
eid = None
else:
eid = getattr(self, optimize_on)
RecomputeAttributeOperation.get_instance(self._cw).add_data(rdef, eid)
class AttributeInvolvedInCAModifiedHook(hook.Hook):
"""When some attribute used in a computed attribute is updated, those
attributes have to be recomputed.
Concret class of this hook are generated at registration time by
introspecting the schema.
"""
__abstract__ = True
events = ('after_update_entity',)
# list of (computed attribute rdef, attributes of this entity type involved)
# that may have to be recomputed
attributes_computed_attributes = None
def __call__(self):
edited_attributes = frozenset(self.entity.cw_edited)
for rdef, used_attributes in self.attributes_computed_attributes.items():
if edited_attributes.intersection(used_attributes):
# XXX optimize if the modified attributes belong to the same
# entity as the computed attribute
RecomputeAttributeOperation.get_instance(self._cw).add_data(rdef)
# code generation at registration time #########################################
def _optimize_on(formula_select, rtype):
"""Given a formula and some rtype, tells whether on update of the given
relation, formula may be recomputed only for rhe relation's subject
('eidfrom' returned), object ('eidto' returned) or None.
Optimizing is only possible when X is used as direct subject/object of this
relation, else we may miss some necessary update.
"""
for rel in formula_select.get_nodes(nodes.Relation):
if rel.r_type == rtype:
sub = rel.get_variable_parts()[0]
obj = rel.get_variable_parts()[1]
if sub.name == 'X':
return 'eidfrom'
elif obj.name == 'X':
return 'eidto'
else:
return None
class _FormulaDependenciesMatrix(object):
"""This class computes and represents the dependencies of computed attributes
towards relations and attributes
"""
def __init__(self, schema):
"""Analyzes the schema to compute the dependencies"""
# entity types holding some computed attribute {etype: [computed rdefs]}
self.computed_attribute_by_etype = defaultdict(list)
# depending entity types {dep. etype: {computed rdef: dep. etype attributes}}
self.computed_attribute_by_etype_attrs = defaultdict(lambda: defaultdict(set))
# depending relations def {dep. rdef: [computed rdefs]
self.computed_attribute_by_relation = defaultdict(list) # by rdef
# Walk through all attributes definitions
for rdef in schema.iter_computed_attributes():
self.computed_attribute_by_etype[rdef.subject.type].append(rdef)
# extract the relations it depends upon - `rdef.formula_select` is
# expected to have been set by finalize_computed_attributes
select = rdef.formula_select
for rel_node in select.get_nodes(nodes.Relation):
if rel_node.is_types_restriction():
continue
rschema = schema.rschema(rel_node.r_type)
lhs, rhs = rel_node.get_variable_parts()
for sol in select.solutions:
subject_etype = sol[lhs.name]
if isinstance(rhs, nodes.VariableRef):
object_etypes = set(sol[rhs.name] for sol in select.solutions)
else:
object_etypes = rschema.objects(subject_etype)
for object_etype in object_etypes:
if rschema.final:
attr_for_computations = self.computed_attribute_by_etype_attrs[subject_etype]
attr_for_computations[rdef].add(rschema.type)
else:
depend_on_rdef = rschema.rdefs[subject_etype, object_etype]
self.computed_attribute_by_relation[depend_on_rdef].append(rdef)
def generate_entity_creation_hooks(self):
for etype, computed_attributes in self.computed_attribute_by_etype.items():
regid = 'computed_attribute.%s_created' % etype
selector = hook.is_instance(etype)
yield type('%sCreatedHook' % etype,
(EntityWithCACreatedHook,),
{'__regid__': regid,
'__select__': hook.Hook.__select__ & selector,
'computed_attributes': computed_attributes})
def generate_relation_change_hooks(self):
for rdef, computed_attributes in self.computed_attribute_by_relation.items():
regid = 'computed_attribute.%s_modified' % rdef.rtype
selector = hook.match_rtype(rdef.rtype.type,
frometypes=(rdef.subject.type,),
toetypes=(rdef.object.type,))
optimized_computed_attributes = []
for computed_rdef in computed_attributes:
optimized_computed_attributes.append(
(computed_rdef,
_optimize_on(computed_rdef.formula_select, rdef.rtype))
)
yield type('%sModifiedHook' % rdef.rtype,
(RelationInvolvedInCAModifiedHook,),
{'__regid__': regid,
'__select__': hook.Hook.__select__ & selector,
'optimized_computed_attributes': optimized_computed_attributes})
def generate_entity_update_hooks(self):
for etype, attributes_computed_attributes in self.computed_attribute_by_etype_attrs.items():
regid = 'computed_attribute.%s_updated' % etype
selector = hook.is_instance(etype)
yield type('%sModifiedHook' % etype,
(AttributeInvolvedInCAModifiedHook,),
{'__regid__': regid,
'__select__': hook.Hook.__select__ & selector,
'attributes_computed_attributes': attributes_computed_attributes})
def registration_callback(vreg):
vreg.register_all(globals().values(), __name__)
dependencies = _FormulaDependenciesMatrix(vreg.schema)
for hook_class in dependencies.generate_entity_creation_hooks():
vreg.register(hook_class)
for hook_class in dependencies.generate_relation_change_hooks():
vreg.register(hook_class)
for hook_class in dependencies.generate_entity_update_hooks():
vreg.register(hook_class)