Fixed #24554 -- Sped up migrations by rendering initial apps when they are first needed

Calling Migration.mutate_state() now also allows to do in_place
mutations in case an intermediate state is thrown away later.

Thanks Anssi Kääriäinen for the idea, Ryan Hall for parts of the patch,
and Claude Paroz and Tim Graham for the review
This commit is contained in:
Markus Holtermann 2015-03-31 23:35:41 +02:00
parent 4e59156c10
commit 57dc8dd3fa
4 changed files with 28 additions and 11 deletions

View File

@ -74,20 +74,27 @@ class MigrationExecutor(object):
migrations_to_run = {m[0] for m in plan} migrations_to_run = {m[0] for m in plan}
# Create the forwards plan Django would follow on an empty database # Create the forwards plan Django would follow on an empty database
full_plan = self.migration_plan(self.loader.graph.leaf_nodes(), clean_start=True) full_plan = self.migration_plan(self.loader.graph.leaf_nodes(), clean_start=True)
# Holds all states right before and right after a migration is applied # Holds all states right before a migration is applied
# if the migration is being run. # if the migration is being run.
states = {} states = {}
state = ProjectState(real_apps=list(self.loader.unmigrated_apps)) state = ProjectState(real_apps=list(self.loader.unmigrated_apps))
if self.progress_callback: if self.progress_callback:
self.progress_callback("render_start") self.progress_callback("render_start")
state.apps # Render all real_apps -- performance critical # Phase 1 -- Store all project states of migrations right before they
# are applied. The first migration that will be applied in phase 2 will
# trigger the rendering of the initial project state. From this time on
# models will be recursively reloaded as explained in
# `django.db.migrations.state.get_related_models_recursive()`.
for migration, _ in full_plan:
do_run = migration in migrations_to_run
if do_run:
if 'apps' not in state.__dict__:
state.apps # Render all real_apps -- performance critical
states[migration] = state.clone()
# Only preserve the state if the migration is being run later
state = migration.mutate_state(state, preserve=do_run)
if self.progress_callback: if self.progress_callback:
self.progress_callback("render_success") self.progress_callback("render_success")
# Phase 1 -- Store all required states
for migration, _ in full_plan:
if migration in migrations_to_run:
states[migration] = state.clone()
state = migration.mutate_state(state) # state is cloned inside
# Phase 2 -- Run the migrations # Phase 2 -- Run the migrations
for migration, backwards in plan: for migration, backwards in plan:
if not backwards: if not backwards:

View File

@ -271,7 +271,7 @@ class MigrationGraph(object):
plan.append(migration) plan.append(migration)
project_state = ProjectState(real_apps=real_apps) project_state = ProjectState(real_apps=real_apps)
for node in plan: for node in plan:
project_state = self.nodes[node].mutate_state(project_state) project_state = self.nodes[node].mutate_state(project_state, preserve=False)
return project_state return project_state
def __contains__(self, node): def __contains__(self, node):

View File

@ -69,12 +69,16 @@ class Migration(object):
def __hash__(self): def __hash__(self):
return hash("%s.%s" % (self.app_label, self.name)) return hash("%s.%s" % (self.app_label, self.name))
def mutate_state(self, project_state): def mutate_state(self, project_state, preserve=True):
""" """
Takes a ProjectState and returns a new one with the migration's Takes a ProjectState and returns a new one with the migration's
operations applied to it. operations applied to it. Preserves the original object state by
default and will return a mutated state from a copy.
""" """
new_state = project_state.clone() new_state = project_state
if preserve:
new_state = project_state.clone()
for operation in self.operations: for operation in self.operations:
operation.state_forwards(self.app_label, new_state) operation.state_forwards(self.app_label, new_state)
return new_state return new_state

View File

@ -34,6 +34,12 @@ def get_related_models_recursive(model):
""" """
Returns all models that have a direct or indirect relationship Returns all models that have a direct or indirect relationship
to the given model. to the given model.
Relationships are either defined by explicit relational fields, like
ForeignKey, ManyToManyField or OneToOneField, or by inheriting from another
model (a superclass is related to its subclasses, but not vice versa). Note,
however, that a model inheriting from a concrete model is also related to
its superclass through the implicit *_ptr OneToOneField on the subclass.
""" """
def _related_models(m): def _related_models(m):
return [ return [