From ab938bf376efe7a93b54e4ca595d3102d04b0080 Mon Sep 17 00:00:00 2001 From: Isaku Yamahata Date: Fri, 27 May 2011 11:10:24 +0900 Subject: compute: implement ec2 stop/start instances This patch implements ec2 stop/start instances with block device mapping support. --- nova/api/ec2/cloud.py | 34 +++++++++++++++++++++++---------- nova/compute/api.py | 39 ++++++++++++++++++++++++++++++++++++++ nova/compute/manager.py | 48 ++++++++++++++++++++++++++++++++++++++++------- nova/db/api.py | 5 +++++ nova/db/sqlalchemy/api.py | 18 ++++++++++++++++++ nova/scheduler/simple.py | 8 +++++++- 6 files changed, 134 insertions(+), 18 deletions(-) diff --git a/nova/api/ec2/cloud.py b/nova/api/ec2/cloud.py index 56b958458..0989a4f40 100644 --- a/nova/api/ec2/cloud.py +++ b/nova/api/ec2/cloud.py @@ -903,33 +903,47 @@ class CloudController(object): return self._format_run_instances(context, instances[0]['reservation_id']) + def _do_instance(self, action, context, ec2_id): + instance_id = ec2utils.ec2_id_to_id(ec2_id) + action(context, instance_id=instance_id) + + def _do_instances(self, action, context, instance_id): + for ec2_id in instance_id: + self._do_instance(action, context, ec2_id) + def terminate_instances(self, context, instance_id, **kwargs): """Terminate each instance in instance_id, which is a list of ec2 ids. instance_id is a kwarg so its name cannot be modified.""" LOG.debug(_("Going to start terminating instances")) - for ec2_id in instance_id: - instance_id = ec2utils.ec2_id_to_id(ec2_id) - self.compute_api.delete(context, instance_id=instance_id) + self._do_instances(self.compute_api.delete, context, instance_id) return True def reboot_instances(self, context, instance_id, **kwargs): """instance_id is a list of instance ids""" LOG.audit(_("Reboot instance %r"), instance_id, context=context) - for ec2_id in instance_id: - instance_id = ec2utils.ec2_id_to_id(ec2_id) - self.compute_api.reboot(context, instance_id=instance_id) + self._do_instances(self.compute_api.reboot, context, instance_id) + return True + + def stop_instances(self, context, instance_id, **kwargs): + """Stop each instance in instace_id""" + LOG.debug(_("Going to stop instnces")) + self._do_instances(self.compute_api.stop, context, instance_id) + return True + + def start_instances(self, context, instance_id, **kwargs): + """Start each instance in instace_id""" + LOG.debug(_("Going to start instnces")) + self._do_instances(self.compute_api.start, context, instance_id) return True def rescue_instance(self, context, instance_id, **kwargs): """This is an extension to the normal ec2_api""" - instance_id = ec2utils.ec2_id_to_id(instance_id) - self.compute_api.rescue(context, instance_id=instance_id) + self._do_instance(self.compute_api.rescue, contect, instnace_id) return True def unrescue_instance(self, context, instance_id, **kwargs): """This is an extension to the normal ec2_api""" - instance_id = ec2utils.ec2_id_to_id(instance_id) - self.compute_api.unrescue(context, instance_id=instance_id) + self._do_instance(self.compute_api.unrescue, context, instance_id) return True def update_instance(self, context, instance_id, **kwargs): diff --git a/nova/compute/api.py b/nova/compute/api.py index af2f23724..6b38e4eb5 100644 --- a/nova/compute/api.py +++ b/nova/compute/api.py @@ -426,6 +426,45 @@ class API(base.Base): terminate_volumes(self.db, context, instance_id) self.db.instance_destroy(context, instance_id) + @scheduler_api.reroute_compute("stop") + def stop(self, context, instance_id): + """Stop an instnace.""" + LOG.debug(_("Going to try to stop %s"), instance_id) + + instance = self._get_instance(context, instance_id, 'stopping') + if not _is_able_to_shutdown(instance, instance_id): + return + + self.update(context, + instance['id'], + state_description='stopping', + state=power_state.NOSTATE, + terminated_at=datetime.datetime.utcnow()) + + host = instance['host'] + if host: + self._cast_compute_message('stop_instance', context, + instance_id, host) + + def start(self, context, instance_id): + """Start an instnace.""" + LOG.debug(_("Going to try to start %s"), instance_id) + instance = self._get_instance(context, instance_id, 'starting') + if instance['state_description'] != 'stopped': + _state_description = instance['state_description'] + LOG.warning(_("Instance %(instance_id)s is not " + "stopped(%(_state_description)s)") % locals()) + return + + # TODO(yamahata): injected_files isn't supported right now. + # It is used only for osapi. not for ec2 api. + # availability_zone isn't used by run_instance. + rpc.cast(context, + FLAGS.scheduler_topic, + {"method": "start_instance", + "args": {"topic": FLAGS.compute_topic, + "instance_id": instance_id}}) + def get(self, context, instance_id): """Get a single instance with the given instance_id.""" rv = self.db.instance_get(context, instance_id) diff --git a/nova/compute/manager.py b/nova/compute/manager.py index 86dac0ff2..5e8429e88 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -269,7 +269,7 @@ class ComputeManager(manager.SchedulerDependentManager): return block_device_mapping - def run_instance(self, context, instance_id, **kwargs): + def _run_instance(self, context, instance_id, **kwargs): """Launch a new instance with specified options.""" context = context.elevated() instance_ref = self.db.instance_get(context, instance_id) @@ -333,13 +333,25 @@ class ComputeManager(manager.SchedulerDependentManager): self._update_launched_at(context, instance_id) self._update_state(context, instance_id) + @exception.wrap_exception + def run_instance(self, context, instance_id, **kwargs): + self._run_instance(context, instance_id, **kwargs) + @exception.wrap_exception @checks_instance_lock - def terminate_instance(self, context, instance_id): - """Terminate an instance on this host.""" + def start_instance(self, context, instance_id): + """Starting an instance on this host.""" + # TODO(yamahata): injected_files isn't supported. + # Anyway OSAPI doesn't support stop/start yet + self._run_instance(context, instance_id) + + def _shutdown_instance(self, context, instance_id, action_str): + """Shutdown an instance on this host.""" context = context.elevated() instance_ref = self.db.instance_get(context, instance_id) - LOG.audit(_("Terminating instance %s"), instance_id, context=context) + LOG.audit(_("%(action_str)s instance %(instance_id)s") % + {'action_str': action_str, 'instance_id': instance_id}, + context=context) fixed_ip = instance_ref.get('fixed_ip') if not FLAGS.stub_network and fixed_ip: @@ -375,16 +387,34 @@ class ComputeManager(manager.SchedulerDependentManager): volumes = instance_ref.get('volumes') or [] for volume in volumes: - self.detach_volume(context, instance_id, volume['id']) - if instance_ref['state'] == power_state.SHUTOFF: + self._detach_volume(context, instance_id, volume['id'], False) + + if (instance_ref['state'] == power_state.SHUTOFF and + instance_ref['state_description'] != 'stopped'): self.db.instance_destroy(context, instance_id) raise exception.Error(_('trying to destroy already destroyed' ' instance: %s') % instance_id) self.driver.destroy(instance_ref) + if action_str == 'Terminating': + terminate_volumes(self.db, context, instance_id) + + @exception.wrap_exception + @checks_instance_lock + def terminate_instance(self, context, instance_id): + """Terminate an instance on this host.""" + self._shutdown_instance(context, instance_id, 'Terminating') + # TODO(ja): should we keep it in a terminated state for a bit? self.db.instance_destroy(context, instance_id) + @exception.wrap_exception + @checks_instance_lock + def stop_instance(self, context, instance_id): + """Stopping an instance on this host.""" + self._shutdown_instance(context, instance_id, 'Stopping') + # instance state will be updated to stopped by _poll_istance_states() + @exception.wrap_exception @checks_instance_lock def rebuild_instance(self, context, instance_id, image_id): @@ -1250,11 +1280,15 @@ class ComputeManager(manager.SchedulerDependentManager): "State=%(db_state)s, so setting state to " "shutoff.") % locals()) vm_state = power_state.SHUTOFF + if db_instance['state_description'] == 'stopping': + self.db.instance_stop(context, db_instance['id']) + continue else: vm_state = vm_instance.state vms_not_found_in_db.remove(name) - if db_instance['state_description'] == 'migrating': + + if (db_instance['state_description'] in ['migrating', 'stopping']): # A situation which db record exists, but no instance" # sometimes occurs while live-migration at src compute, # this case should be ignored. diff --git a/nova/db/api.py b/nova/db/api.py index 858c8eb10..7fb7f336f 100644 --- a/nova/db/api.py +++ b/nova/db/api.py @@ -414,6 +414,11 @@ def instance_destroy(context, instance_id): return IMPL.instance_destroy(context, instance_id) +def instance_stop(context, instance_id): + """Stop the instance or raise if it does not exist.""" + return IMPL.instance_stop(context, instance_id) + + def instance_get(context, instance_id): """Get an instance or raise if it does not exist.""" return IMPL.instance_get(context, instance_id) diff --git a/nova/db/sqlalchemy/api.py b/nova/db/sqlalchemy/api.py index e0904b3c6..be47e2c2e 100644 --- a/nova/db/sqlalchemy/api.py +++ b/nova/db/sqlalchemy/api.py @@ -831,6 +831,24 @@ def instance_destroy(context, instance_id): 'deleted_at': datetime.datetime.utcnow(), 'updated_at': literal_column('updated_at')}) +@require_context +def instance_stop(context, instance_id): + session = get_session() + with session.begin(): + from nova.compute import power_state + session.query(models.Instance).\ + filter_by(id=instance_id).\ + update({'host': None, + 'state': power_state.SHUTOFF, + 'state_description': 'stopped', + 'updated_at': literal_column('updated_at')}) + session.query(models.SecurityGroupInstanceAssociation).\ + filter_by(instance_id=instance_id).\ + update({'updated_at': literal_column('updated_at')}) + session.query(models.InstanceMetadata).\ + filter_by(instance_id=instance_id).\ + update({'updated_at': literal_column('updated_at')}) + @require_context def instance_get(context, instance_id, session=None): diff --git a/nova/scheduler/simple.py b/nova/scheduler/simple.py index dd568d2c6..ccbc79a36 100644 --- a/nova/scheduler/simple.py +++ b/nova/scheduler/simple.py @@ -40,7 +40,7 @@ flags.DEFINE_integer("max_networks", 1000, class SimpleScheduler(chance.ChanceScheduler): """Implements Naive Scheduler that tries to find least loaded host.""" - def schedule_run_instance(self, context, instance_id, *_args, **_kwargs): + def _schedule_instance(self, context, instance_id, *_args, **_kwargs): """Picks a host that is up and has the fewest running instances.""" instance_ref = db.instance_get(context, instance_id) if (instance_ref['availability_zone'] @@ -76,6 +76,12 @@ class SimpleScheduler(chance.ChanceScheduler): " for this request. Is the appropriate" " service running?")) + def schedule_run_instance(self, context, instance_id, *_args, **_kwargs): + return self._schedule_instance(context, instance_id, *_args, **_kwargs) + + def schedule_start_instance(self, context, instance_id, *_args, **_kwargs): + return self._schedule_instance(context, instance_id, *_args, **_kwargs) + def schedule_create_volume(self, context, volume_id, *_args, **_kwargs): """Picks a host that is up and has the fewest volumes.""" volume_ref = db.volume_get(context, volume_id) -- cgit