def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha') % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource resources.initialise() self.env = env or environment.Environment({}) self.parameters = Parameters(self.name, self.t, user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues())
def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=None, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, adopt_stack_data=None, stack_user_project_id=None, created_time=None, updated_time=None, user_creds_id=None, tenant_id=None, validate_parameters=True): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None self._access_allowed_handlers = {} self._db_resources = None self.adopt_stack_data = adopt_stack_data self.stack_user_project_id = stack_user_project_id self.created_time = created_time self.updated_time = updated_time self.user_creds_id = user_creds_id # This will use the provided tenant ID when loading the stack # from the DB or get it from the context for new stacks. self.tenant_id = tenant_id or self.context.tenant_id resources.initialise() self.env = env or environment.Environment({}) self.parameters = self.t.parameters(self.identifier(), user_params=self.env.params) self.parameters.validate(validate_value=validate_parameters, context=context) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[self.t.OUTPUTS]) else: self.outputs = {}
def __init__(self, context, stack_name, tmpl, parameters=None, stack_id=None, state=None, state_description='', timeout_mins=60, resolve_data=True, disable_rollback=True): ''' Initialise from a context, name, Template object and (optionally) Parameters object. The database ID may also be initialised, if the stack is already in the database. ''' if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _("Invalid stack name %s" % stack_name + ", must contain only alphanumeric or " + "\"_-.\" characters, must start with alpha")) self.id = stack_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.state = state self.state_description = state_description self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback resources.initialise() if parameters is None: parameters = Parameters(self.name, self.t) self.parameters = parameters self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues())
def __init__( self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason="", timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, ): """ Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. """ if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _( "Invalid stack name %s" " must contain only alphanumeric or " '"_-." characters, must start with alpha' ) % stack_name ) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None resources.initialise() self.env = env or environment.Environment({}) self.parameters = Parameters(self.name, self.t, user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {}
def validate(self, value, context): if not value: # Don't validate empty key, which can happen when you use a KeyPair # resource return True try: nova_utils.get_keypair(Clients(context).nova(), value) except exception.UserKeyPairMissing: return False else: return True
def __init__(self, context, stack_name, tmpl, parameters=None, stack_id=None, state=None, state_description='', timeout_mins=60, resolve_data=True, disable_rollback=True): ''' Initialise from a context, name, Template object and (optionally) Parameters object. The database ID may also be initialised, if the stack is already in the database. ''' if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.state = state self.state_description = state_description self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback resources.initialise() if parameters is None: parameters = Parameters(self.name, self.t) self.parameters = parameters self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues())
def __init__(self, context, stack_name, tmpl, parameters=None, stack_id=None, state=None, state_description='', timeout_mins=60, resolve_data=True): ''' Initialise from a context, name, Template object and (optionally) Parameters object. The database ID may also be initialised, if the stack is already in the database. ''' self.id = stack_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.state = state self.state_description = state_description self.timeout_mins = timeout_mins if parameters is None: parameters = Parameters(self.name, self.t) self.parameters = parameters if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resources.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues())
class Stack(object): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK ) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK') CREATE_IN_PROGRESS = 'CREATE_IN_PROGRESS' CREATE_FAILED = 'CREATE_FAILED' CREATE_COMPLETE = 'CREATE_COMPLETE' DELETE_IN_PROGRESS = 'DELETE_IN_PROGRESS' DELETE_FAILED = 'DELETE_FAILED' DELETE_COMPLETE = 'DELETE_COMPLETE' UPDATE_IN_PROGRESS = 'UPDATE_IN_PROGRESS' UPDATE_COMPLETE = 'UPDATE_COMPLETE' UPDATE_FAILED = 'UPDATE_FAILED' ROLLBACK_IN_PROGRESS = 'ROLLBACK_IN_PROGRESS' ROLLBACK_COMPLETE = 'ROLLBACK_COMPLETE' ROLLBACK_FAILED = 'ROLLBACK_FAILED' created_time = timestamp.Timestamp(db_api.stack_get, 'created_at') updated_time = timestamp.Timestamp(db_api.stack_get, 'updated_at') _zones = None def __init__(self, context, stack_name, tmpl, parameters=None, stack_id=None, state=None, state_description='', timeout_mins=60, resolve_data=True, disable_rollback=True): ''' Initialise from a context, name, Template object and (optionally) Parameters object. The database ID may also be initialised, if the stack is already in the database. ''' if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.state = state self.state_description = state_description self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback resources.initialise() if parameters is None: parameters = Parameters(self.name, self.t) self.parameters = parameters self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the AWS::StackId pseudo parameter ''' # This can fail if constructor called without a valid context, # as it is in many tests try: stack_arn = self.identifier().arn() except (AttributeError, ValueError, TypeError): logger.warning("Unable to set parameters StackId identifier") else: self.parameters.set_stack_id(stack_arn) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id) if stack is None: message = 'No stack exists with id "%s"' % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) params = Parameters(stack.name, template, stack.parameters) stack = cls(context, stack.name, template, params, stack.id, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback) return stack def store(self, owner=None): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' new_creds = db_api.user_creds_create(self.context) s = { 'name': self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.parameters.user_parameters(), 'owner_id': owner and owner.id, 'user_creds_id': new_creds.id, 'username': self.context.username, 'tenant': self.context.tenant_id, 'status': self.state, 'status_reason': self.state_description, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over this template's resources in the order that they should be started. ''' return iter(self.dependencies) def __reversed__(self): ''' Return an iterator over this template's resources in the order that they should be stopped. ''' return reversed(self.dependencies) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, value): '''Set the resource with the specified name to a specific value.''' self.resources[key] = value def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' return key in self.resources def keys(self): '''Return a list of resource keys for the stack.''' return self.resources.keys() def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s"' % self.name def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.resources.values(): if r.state in ( r.CREATE_IN_PROGRESS, r.CREATE_COMPLETE, r.UPDATE_IN_PROGRESS, r.UPDATE_COMPLETE) and r.FnGetRefId() == refid: return r def validate(self): ''' http://docs.amazonwebservices.com/AWSCloudFormation/latest/\ APIReference/API_ValidateTemplate.html ''' # TODO(sdake) Should return line number of invalid reference for res in self: try: result = res.validate() except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=str(ex)) if result: raise StackValidationFailed(message=result) def state_set(self, new_status, reason): '''Update the stack state in the database.''' self.state = new_status self.state_description = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) stack.update_and_save({'status': new_status, 'status_reason': reason}) def timeout_secs(self): ''' Return the stack creation timeout in seconds, or None if no timeout should be used. ''' if self.timeout_mins is None: return None return self.timeout_mins * 60 def create(self): ''' Create the stack and all of the resources. ''' creator = scheduler.TaskRunner(self.create_task) creator(timeout=self.timeout_secs()) @scheduler.wrappertask def create_task(self): ''' A task to create the stack and all of the resources. ''' self.state_set(self.CREATE_IN_PROGRESS, 'Stack creation started') stack_status = self.CREATE_COMPLETE reason = 'Stack successfully created' res = None def resource_create(r): return r.create create_task = scheduler.DependencyTaskGroup(self.dependencies, resource_create) try: yield create_task() except exception.ResourceFailure as ex: stack_status = self.CREATE_FAILED reason = 'Resource failed: %s' % str(ex) except scheduler.Timeout: stack_status = self.CREATE_FAILED reason = 'Timed out' self.state_set(stack_status, reason) if stack_status == self.CREATE_FAILED and not self.disable_rollback: self.delete(action=self.ROLLBACK) def update(self, newstack, action=UPDATE): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' if action not in (self.UPDATE, self.ROLLBACK): logger.error("Unexpected action %s passed to update!" % action) self.state_set(self.UPDATE_FAILED, "Invalid action %s" % action) return if self.state not in (self.CREATE_COMPLETE, self.UPDATE_COMPLETE, self.ROLLBACK_COMPLETE): if (action == self.ROLLBACK and self.state == self.UPDATE_IN_PROGRESS): logger.debug("Starting update rollback for %s" % self.name) else: if action == self.UPDATE: self.state_set(self.UPDATE_FAILED, 'State invalid for update') else: self.state_set(self.ROLLBACK_FAILED, 'State invalid for rollback') return if action == self.UPDATE: self.state_set(self.UPDATE_IN_PROGRESS, 'Stack update started') else: self.state_set(self.ROLLBACK_IN_PROGRESS, 'Stack rollback started') # cache all the resources runtime data. for r in self: r.cache_template() # Now make the resources match the new stack definition with eventlet.Timeout(self.timeout_secs()) as tmo: try: # First delete any resources which are not in newstack for res in reversed(self): if res.name not in newstack.keys(): logger.debug("resource %s not found in updated stack" % res.name + " definition, deleting") # res.destroy raises exception.ResourceFailure on error res.destroy() del self.resources[res.name] self.dependencies = self._get_dependencies( self.resources.itervalues()) # Then create any which are defined in newstack but not self for res in newstack: if res.name not in self.keys(): logger.debug("resource %s not found in current stack" % res.name + " definition, adding") res.stack = self self[res.name] = res self.dependencies = self._get_dependencies( self.resources.itervalues()) # res.create raises exception.ResourceFailure on error scheduler.TaskRunner(res.create)() # Now (the hard part :) update existing resources # The Resource base class allows equality-test of resources, # based on the parsed template snippet for the resource. # If this test fails, we call the underlying resource.update # # Currently many resources have a default handle_update method # which raises exception.ResourceReplace # optionally they may implement non-interruptive logic and # return UPDATE_COMPLETE. If resources do not implement the # handle_update method at all, update will fail. for res in newstack: # Compare resolved pre/post update resource snippets, # note the new resource snippet is resolved in the context # of the existing stack (which is the stack being updated) old_snippet = self[res.name].parsed_template(cached=True) new_snippet = self.resolve_runtime_data(res.t) if old_snippet != new_snippet: # res.update raises exception.ResourceFailure on error # or exception.ResourceReplace if update requires # replacement try: self[res.name].update(new_snippet) except resource.UpdateReplace: # Resource requires replacement for update self[res.name].destroy() res.stack = self self[res.name] = res self.dependencies = self._get_dependencies( self.resources.itervalues()) scheduler.TaskRunner(res.create)() else: logger.info("Resource %s for stack %s updated" % (res.name, self.name)) if action == self.UPDATE: stack_status = self.UPDATE_COMPLETE reason = 'Stack successfully updated' else: stack_status = self.ROLLBACK_COMPLETE reason = 'Stack rollback completed' except eventlet.Timeout as t: if t is tmo: stack_status = self.UPDATE_FAILED reason = 'Timed out waiting for %s' % str(res) else: # not my timeout raise except exception.ResourceFailure as e: reason = str(e) or "Error : %s" % type(e) if action == self.UPDATE: stack_status = self.UPDATE_FAILED # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if self.disable_rollback: stack_status = self.UPDATE_FAILED else: oldstack = Stack(self.context, self.name, self.t, self.parameters) self.update(oldstack, action=self.ROLLBACK) return else: stack_status = self.ROLLBACK_FAILED self.state_set(stack_status, reason) # flip the template & parameters to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t self.parameters = newstack.parameters template_outputs = self.t[template.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action == self.DELETE: self.state_set(self.DELETE_IN_PROGRESS, 'Stack deletion started') elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_IN_PROGRESS, 'Stack rollback started') else: logger.error("Unexpected action %s passed to delete!" % action) self.state_set(self.DELETE_FAILED, "Invalid action %s" % action) return failures = [] for res in reversed(self): try: res.destroy() except exception.ResourceFailure as ex: logger.error('Failed to delete %s error: %s' % (str(res), str(ex))) failures.append(str(res)) if failures: if action == self.DELETE: self.state_set(self.DELETE_FAILED, 'Failed to delete ' + ', '.join(failures)) elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_FAILED, 'Failed to rollback ' + ', '.join(failures)) else: if action == self.DELETE: self.state_set(self.DELETE_COMPLETE, 'Deleted successfully') elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_COMPLETE, 'Rollback completed') db_api.stack_delete(self.context, self.id) self.id = None def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') return self.resolve_runtime_data(value) def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: res.destroy() except exception.ResourceFailure as ex: failed = True logger.error('delete: %s' % str(ex)) for res in deps: if not failed: try: scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception('create') failed = True else: res.state_set(res.CREATE_FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list(detailed=False)] return self._zones def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)
class Stack(collections.Mapping): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME, ADOPT) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME', 'ADOPT') STATUSES = (IN_PROGRESS, FAILED, COMPLETE) = ('IN_PROGRESS', 'FAILED', 'COMPLETE') _zones = None def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=None, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, adopt_stack_data=None, stack_user_project_id=None, created_time=None, updated_time=None, user_creds_id=None, tenant_id=None, validate_parameters=True, use_stored_context=False): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha') % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None self._access_allowed_handlers = {} self._db_resources = None self.adopt_stack_data = adopt_stack_data self.stack_user_project_id = stack_user_project_id self.created_time = created_time self.updated_time = updated_time self.user_creds_id = user_creds_id if use_stored_context: self.context = self.stored_context() self.clients = Clients(self.context) # This will use the provided tenant ID when loading the stack # from the DB or get it from the context for new stacks. self.tenant_id = tenant_id or self.context.tenant_id resources.initialise() self.env = env or environment.Environment({}) self.parameters = self.t.parameters(self.identifier(), user_params=self.env.params) self.parameters.validate(validate_value=validate_parameters, context=self.context) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[self.t.OUTPUTS]) else: self.outputs = {} def stored_context(self): if self.user_creds_id: creds = db_api.user_creds_get(self.user_creds_id) # Maintain request_id from self.context so we retain tracability # in situations where servicing a request requires switching from # the request context to the stored context creds['request_id'] = self.context.request_id # We don't store roles in the user_creds table, so disable the # policy check for admin by setting is_admin=False. creds['is_admin'] = False return common_context.RequestContext.from_dict(creds) else: msg = _("Attempt to use stored_context with no user_creds") raise exception.Error(msg) @property def resources(self): if self._resources is None: self._resources = dict( (name, resource.Resource(name, data, self)) for (name, data) in self.t.resource_definitions(self).items()) # There is no need to continue storing the db resources # after resource creation self._db_resources = None return self._resources def db_resource_get(self, name): if not self.id: return None if self._db_resources is None: try: self._db_resources = db_api.resource_get_all_by_stack( self.context, self.id) except exception.NotFound: return None return self._db_resources.get(name) @property def dependencies(self): if self._dependencies is None: self._dependencies = self._get_dependencies( self.resources.itervalues()) return self._dependencies def reset_dependencies(self): self._dependencies = None @property def root_stack(self): ''' Return the root stack if this is nested (otherwise return self). ''' if (self.parent_resource and self.parent_resource.stack): return self.parent_resource.stack.root_stack return self def total_resources(self): ''' Return the total number of resources in a stack, including nested stacks below. ''' def total_nested(res): get_nested = getattr(res, 'nested', None) if callable(get_nested): nested_stack = get_nested() if nested_stack is not None: return nested_stack.total_resources() return 0 return len(self) + sum(total_nested(res) for res in self.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the StackId pseudo parameter ''' if not self.parameters.set_stack_id(self.identifier()): LOG.warning(_("Unable to set parameters StackId identifier")) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, parent_resource=None, show_deleted=True, use_stored_context=False): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted, eager_load=True) if stack is None: message = _('No stack exists with id "%s"') % str(stack_id) raise exception.NotFound(message) return cls._from_db(context, stack, parent_resource=parent_resource, use_stored_context=use_stored_context) @classmethod def load_all(cls, context, limit=None, marker=None, sort_keys=None, sort_dir=None, filters=None, tenant_safe=True, show_deleted=False, resolve_data=True): stacks = db_api.stack_get_all(context, limit, sort_keys, marker, sort_dir, filters, tenant_safe, show_deleted) or [] for stack in stacks: yield cls._from_db(context, stack, resolve_data=resolve_data) @classmethod def _from_db(cls, context, stack, parent_resource=None, resolve_data=True, use_stored_context=False): template = Template.load(context, stack.raw_template_id, stack.raw_template) env = environment.Environment(stack.parameters) return cls(context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id, stack_user_project_id=stack.stack_user_project_id, created_time=stack.created_at, updated_time=stack.updated_at, user_creds_id=stack.user_creds_id, tenant_id=stack.tenant, validate_parameters=False, use_stored_context=use_stored_context) def store(self, backup=False): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' s = { 'name': self._backup_name() if backup else self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.env.user_env_as_dict(), 'owner_id': self.owner_id, 'username': self.context.username, 'tenant': self.tenant_id, 'action': self.action, 'status': self.status, 'status_reason': self.status_reason, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, 'stack_user_project_id': self.stack_user_project_id, 'updated_at': self.updated_time, 'user_creds_id': self.user_creds_id } if self.id: db_api.stack_update(self.context, self.id, s) else: if not self.user_creds_id: # Create a context containing a trust_id and trustor_user_id # if trusts are enabled if cfg.CONF.deferred_auth_method == 'trusts': keystone = self.clients.client('keystone') trust_ctx = keystone.create_trust_context() new_creds = db_api.user_creds_create(trust_ctx) else: new_creds = db_api.user_creds_create(self.context) s['user_creds_id'] = new_creds.id self.user_creds_id = new_creds.id new_s = db_api.stack_create(self.context, s) self.id = new_s.id self.created_time = new_s.created_at self._set_param_stackid() return self.id def _backup_name(self): return '%s*' % self.name def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over the resource names. ''' return iter(self.resources) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def add_resource(self, resource): '''Insert the given resource into the stack.''' template = resource.stack.t resource.stack = self definition = resource.t.reparse(self, template) resource.t = definition resource.reparse() self.resources[resource.name] = resource self.t.add_resource(definition) if self.t.id is not None: self.t.store(self.context) def remove_resource(self, resource_name): '''Remove the resource with the specified name.''' del self.resources[resource_name] self.t.remove_resource(resource_name) if self.t.id is not None: self.t.store(self.context) def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' if self._resources is not None: return key in self.resources else: return key in self.t[self.t.RESOURCES] def __eq__(self, other): ''' Compare two Stacks for equality. Stacks are considered equal only if they are identical. ''' return self is other def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s" [%s]' % (self.name, self.id) def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.values(): if r.state in ((r.INIT, r.COMPLETE), (r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.RESUME, r.IN_PROGRESS), (r.RESUME, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE)) and r.FnGetRefId() == refid: return r def register_access_allowed_handler(self, credential_id, handler): ''' Register a function which determines whether the credentials with a give ID can have access to a named resource. ''' assert callable(handler), 'Handler is not callable' self._access_allowed_handlers[credential_id] = handler def access_allowed(self, credential_id, resource_name): ''' Returns True if the credential_id is authorised to access the resource with the specified resource_name. ''' if not self.resources: # this also triggers lazy-loading of resources # so is required for register_access_allowed_handler # to be called return False handler = self._access_allowed_handlers.get(credential_id) return handler and handler(resource_name) def validate(self): ''' Validates the template. ''' # TODO(sdake) Should return line number of invalid reference # validate overall template (top-level structure) self.t.validate() # Validate Parameter Groups parameter_groups = ParameterGroups(self.t) parameter_groups.validate() # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.keys()) if dup_names: LOG.debug("Duplicate names %s" % dup_names) raise StackValidationFailed(message=_("Duplicate names %s") % dup_names) for res in self.dependencies: try: result = res.validate() except exception.HeatException as ex: LOG.exception(ex) raise ex except Exception as ex: LOG.exception(ex) raise StackValidationFailed( message=strutils.safe_decode(six.text_type(ex))) if result: raise StackValidationFailed(message=result) for val in self.outputs.values(): snippet = val.get('Value', '') try: function.validate(snippet) except Exception as ex: reason = 'Output validation error: %s' % six.text_type(ex) raise StackValidationFailed(message=reason) def requires_deferred_auth(self): ''' Returns whether this stack may need to perform API requests during its lifecycle using the configured deferred authentication method. ''' return any(res.requires_deferred_auth for res in self.values()) def state_set(self, action, status, reason): '''Update the stack state in the database.''' if action not in self.ACTIONS: raise ValueError(_("Invalid action %s") % action) if status not in self.STATUSES: raise ValueError(_("Invalid status %s") % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) if stack is not None: stack.update_and_save({ 'action': action, 'status': status, 'status_reason': reason }) msg = _('Stack %(action)s %(status)s (%(name)s): %(reason)s') LOG.info( msg % { 'action': action, 'status': status, 'name': self.name, 'reason': reason }) notification.send(self) @property def state(self): '''Returns state, tuple of action, status.''' return (self.action, self.status) def timeout_secs(self): ''' Return the stack action timeout in seconds. ''' if self.timeout_mins is None: return cfg.CONF.stack_action_timeout return self.timeout_mins * 60 def preview_resources(self): ''' Preview the stack with all of the resources. ''' return [resource.preview() for resource in self.resources.itervalues()] def create(self): ''' Create the stack and all of the resources. ''' def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def _adopt_kwargs(self, resource): data = self.adopt_stack_data if not data or not data.get('resources'): return {'resource_data': None} return {'resource_data': data['resources'].get(resource.name)} @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): ''' A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specified by reverse ''' self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, '%s' % action_l) # If a local _$action_kwargs function exists, call it to get the # action specific argument list, otherwise an empty arg list handle_kwargs = getattr(self, '_%s_kwargs' % action_l, lambda x: {}) return handle(**handle_kwargs(r)) action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action, six.text_type(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def _backup_stack(self, create_if_missing=True): ''' Get a Stack containing any in-progress resources from the previous stack state prior to an update. ''' s = db_api.stack_get_by_name_and_owner_id(self.context, self._backup_name(), owner_id=self.id) if s is not None: LOG.debug('Loaded existing backup stack') return self.load(self.context, stack=s) elif create_if_missing: prev = type(self)(self.context, self.name, copy.deepcopy(self.t), self.env, owner_id=self.id) prev.store(backup=True) LOG.debug('Created new backup stack') return prev else: return None def adopt(self): ''' Adopt a stack (create stack with all the existing resources). ''' def rollback(): if not self.disable_rollback and self.state == (self.ADOPT, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.ADOPT, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def update(self, newstack): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' self.updated_time = datetime.utcnow() updater = scheduler.TaskRunner(self.update_task, newstack) updater() @scheduler.wrappertask def update_task(self, newstack, action=UPDATE): if action not in (self.UPDATE, self.ROLLBACK): LOG.error(_("Unexpected action %s passed to update!") % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if (action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS)): LOG.debug("Starting update rollback for %s" % self.name) else: self.state_set(action, self.FAILED, 'State invalid for %s' % action) return self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) oldstack = Stack(self.context, self.name, copy.deepcopy(self.t), self.env) backup_stack = self._backup_stack() try: update_task = update.StackUpdate(self, newstack, backup_stack, rollback=action == self.ROLLBACK) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters self.t.files = newstack.t.files self.disable_rollback = newstack.disable_rollback self.timeout_mins = newstack.timeout_mins self._set_param_stackid() try: updater.start(timeout=self.timeout_secs()) yield while not updater.step(): yield finally: self.reset_dependencies() if action == self.UPDATE: reason = 'Stack successfully updated' else: reason = 'Stack rollback completed' stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = 'Timed out' except exception.ResourceFailure as e: reason = six.text_type(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: yield self.update_task(oldstack, action=self.ROLLBACK) return else: LOG.debug('Deleting backup stack') backup_stack.delete(backup=True) # flip the template to the newstack values self.t = newstack.t template_outputs = self.t[self.t.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) # Don't use state_set to do only one update query and avoid race # condition with the COMPLETE status self.action = action self.status = stack_status self.status_reason = reason self.store() notification.send(self) def delete(self, action=DELETE, backup=False): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action not in (self.DELETE, self.ROLLBACK): LOG.error(_("Unexpected action %s passed to delete!") % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) backup_stack = self._backup_stack(False) if backup_stack: for key, backup_resource in backup_stack.resources.items(): # If UpdateReplace is failed, we must restore backup_resource # to existing_stack in case of it may have dependencies in # these stacks. current_resource is the resource that just # created and failed, so put into the backup_stack to delete # anyway. backup_resource_id = backup_resource.resource_id current_resource = self.resources[key] current_resource_id = current_resource.resource_id if backup_resource_id: child_failed = False for child in self.dependencies[current_resource]: # If child resource failed to update, current_resource # should be replaced to resolve dependencies. But this # is not fundamental solution. If there are update # failer and success resources in the children, cannot # delete the stack. if (child.status == child.FAILED and child.action == child.CREATE): child_failed = True if (current_resource.status == current_resource.FAILED or child_failed): # Stack class owns dependencies as set of resource's # objects, so we switch members of the resource that is # needed to delete it. self.resources[key].resource_id = backup_resource_id self.resources[ key].properties = backup_resource.properties backup_stack.resources[ key].resource_id = current_resource_id backup_stack.resources[ key].properties = current_resource.properties backup_stack.delete(backup=True) if backup_stack.status != backup_stack.COMPLETE: errs = backup_stack.status_reason failure = 'Error deleting backup resources: %s' % errs self.state_set(action, self.FAILED, 'Failed to %s : %s' % (action, failure)) return action_task = scheduler.DependencyTaskGroup(self.dependencies, resource.Resource.destroy, reverse=True) try: scheduler.TaskRunner(action_task)(timeout=self.timeout_secs()) except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action, six.text_type(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() if stack_status != self.FAILED and not backup: # Cleanup stored user_creds so they aren't accessible via # the soft-deleted stack which remains in the DB if self.user_creds_id: user_creds = db_api.user_creds_get(self.user_creds_id) # If we created a trust, delete it if user_creds is not None: trust_id = user_creds.get('trust_id') if trust_id: try: self.clients.client('keystone').delete_trust( trust_id) except Exception as ex: LOG.exception(ex) stack_status = self.FAILED reason = ("Error deleting trust: %s" % six.text_type(ex)) # Delete the stored credentials try: db_api.user_creds_delete(self.context, self.user_creds_id) except exception.NotFound: LOG.info( _("Tried to delete user_creds that do not exist " "(stack=%(stack)s user_creds_id=%(uc)s)") % { 'stack': self.id, 'uc': self.user_creds_id }) try: self.user_creds_id = None self.store() except exception.NotFound: LOG.info( _("Tried to store a stack that does not exist " "%s ") % self.id) # If the stack has a domain project, delete it if self.stack_user_project_id: try: keystone = self.clients.client('keystone') keystone.delete_stack_domain_project( project_id=self.stack_user_project_id) except Exception as ex: LOG.exception(ex) stack_status = self.FAILED reason = "Error deleting project: %s" % six.text_type(ex) try: self.state_set(action, stack_status, reason) except exception.NotFound: LOG.info( _("Tried to delete stack that does not exist " "%s ") % self.id) if stack_status != self.FAILED: # delete the stack try: db_api.stack_delete(self.context, self.id) except exception.NotFound: LOG.info( _("Tried to delete stack that does not exist " "%s ") % self.id) self.id = None def suspend(self): ''' Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. ''' # No need to suspend if the stack has been suspended if self.state == (self.SUSPEND, self.COMPLETE): LOG.info(_('%s is already suspended') % str(self)) return sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): ''' Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. ''' # No need to resume if the stack has been resumed if self.state == (self.RESUME, self.COMPLETE): LOG.info(_('%s is already resumed') % str(self)) return sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') try: return function.resolve(value) except Exception as ex: self.outputs[key]['error_msg'] = six.text_type(ex) return None def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: scheduler.TaskRunner(res.destroy)() except exception.ResourceFailure as ex: failed = True LOG.error(_('delete: %s') % ex) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: LOG.exception(_('create')) failed = True else: res.state_set(res.CREATE, res.FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): nova = self.clients.client('nova') if self._zones is None: self._zones = [ zone.zoneName for zone in nova.availability_zones.list(detailed=False) ] return self._zones def set_stack_user_project_id(self, project_id): self.stack_user_project_id = project_id self.store() def prepare_abandon(self): return { 'name': self.name, 'id': self.id, 'action': self.action, 'status': self.status, 'template': self.t.t, 'resources': dict((res.name, res.prepare_abandon()) for res in self.resources.values()) } def resolve_static_data(self, snippet): return self.t.parse(self, snippet) def resolve_runtime_data(self, snippet): """DEPRECATED. Use heat.engine.function.resolve() instead.""" warnings.warn( 'Stack.resolve_runtime_data() is deprecated. ' 'Use heat.engine.function.resolve() instead', DeprecationWarning) return function.resolve(snippet) def reset_resource_attributes(self): # nothing is cached if no resources exist if not self._resources: return # a change in some resource may have side-effects in the attributes # of other resources, so ensure that attributes are re-calculated for res in self.resources.itervalues(): res.attributes.reset_resolved_values()
class Stack(collections.Mapping): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME, ADOPT ) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME', 'ADOPT') STATUSES = (IN_PROGRESS, FAILED, COMPLETE ) = ('IN_PROGRESS', 'FAILED', 'COMPLETE') _zones = None def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=None, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, adopt_stack_data=None, stack_user_project_id=None, created_time=None, updated_time=None, user_creds_id=None, tenant_id=None, validate_parameters=True): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None self._access_allowed_handlers = {} self._db_resources = None self.adopt_stack_data = adopt_stack_data self.stack_user_project_id = stack_user_project_id self.created_time = created_time self.updated_time = updated_time self.user_creds_id = user_creds_id # This will use the provided tenant ID when loading the stack # from the DB or get it from the context for new stacks. self.tenant_id = tenant_id or self.context.tenant_id resources.initialise() self.env = env or environment.Environment({}) self.parameters = self.t.parameters(self.identifier(), user_params=self.env.params) self.parameters.validate(validate_value=validate_parameters, context=context) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[self.t.OUTPUTS]) else: self.outputs = {} def stored_context(self): if self.user_creds_id: creds = db_api.user_creds_get(self.user_creds_id) # Maintain request_id from self.context so we retain tracability # in situations where servicing a request requires switching from # the request context to the stored context creds['request_id'] = self.context.request_id # We don't store roles in the user_creds table, so disable the # policy check for admin by setting is_admin=False. creds['is_admin'] = False return common_context.RequestContext.from_dict(creds) else: msg = _("Attempt to use stored_context with no user_creds") raise exception.Error(msg) @property def resources(self): if self._resources is None: self._resources = dict((name, resource.Resource(name, data, self)) for (name, data) in self.t.resource_definitions(self).items()) # There is no need to continue storing the db resources # after resource creation self._db_resources = None return self._resources def db_resource_get(self, name): if not self.id: return None if self._db_resources is None: try: self._db_resources = db_api.resource_get_all_by_stack( self.context, self.id) except exception.NotFound: return None return self._db_resources.get(name) @property def dependencies(self): if self._dependencies is None: self._dependencies = self._get_dependencies( self.resources.itervalues()) return self._dependencies def reset_dependencies(self): self._dependencies = None @property def root_stack(self): ''' Return the root stack if this is nested (otherwise return self). ''' if (self.parent_resource and self.parent_resource.stack): return self.parent_resource.stack.root_stack return self def total_resources(self): ''' Return the total number of resources in a stack, including nested stacks below. ''' def total_nested(res): get_nested = getattr(res, 'nested', None) if callable(get_nested): nested_stack = get_nested() if nested_stack is not None: return nested_stack.total_resources() return 0 return len(self) + sum(total_nested(res) for res in self.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the StackId pseudo parameter ''' if not self.parameters.set_stack_id(self.identifier()): LOG.warning(_("Unable to set parameters StackId identifier")) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True, parent_resource=None, show_deleted=True): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted, eager_load=True) if stack is None: message = _('No stack exists with id "%s"') % str(stack_id) raise exception.NotFound(message) template = Template.load( context, stack.raw_template_id, stack.raw_template) env = environment.Environment(stack.parameters) stack = cls(context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id, stack_user_project_id=stack.stack_user_project_id, created_time=stack.created_at, updated_time=stack.updated_at, user_creds_id=stack.user_creds_id, tenant_id=stack.tenant, validate_parameters=False) return stack def store(self, backup=False): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' s = { 'name': self._backup_name() if backup else self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.env.user_env_as_dict(), 'owner_id': self.owner_id, 'username': self.context.username, 'tenant': self.tenant_id, 'action': self.action, 'status': self.status, 'status_reason': self.status_reason, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, 'stack_user_project_id': self.stack_user_project_id, 'updated_at': self.updated_time, 'user_creds_id': self.user_creds_id } if self.id: db_api.stack_update(self.context, self.id, s) else: if not self.user_creds_id: # Create a context containing a trust_id and trustor_user_id # if trusts are enabled if cfg.CONF.deferred_auth_method == 'trusts': trust_ctx = self.clients.keystone().create_trust_context() new_creds = db_api.user_creds_create(trust_ctx) else: new_creds = db_api.user_creds_create(self.context) s['user_creds_id'] = new_creds.id self.user_creds_id = new_creds.id new_s = db_api.stack_create(self.context, s) self.id = new_s.id self.created_time = new_s.created_at self._set_param_stackid() return self.id def _backup_name(self): return '%s*' % self.name def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over the resource names. ''' return iter(self.resources) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, resource): '''Set the resource with the specified name to a specific value.''' template = resource.stack.t resource.stack = self resource.t = template.resource_definitions(self)[key] resource.reparse() self.resources[key] = resource def __delitem__(self, key): '''Remove the resource with the specified name.''' del self.resources[key] def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' if self._resources is not None: return key in self.resources else: return key in self.t[self.t.RESOURCES] def __eq__(self, other): ''' Compare two Stacks for equality. Stacks are considered equal only if they are identical. ''' return self is other def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s" [%s]' % (self.name, self.id) def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.values(): if r.state in ( (r.INIT, r.COMPLETE), (r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.RESUME, r.IN_PROGRESS), (r.RESUME, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE)) and r.FnGetRefId() == refid: return r def register_access_allowed_handler(self, credential_id, handler): ''' Register a function which determines whether the credentials with a give ID can have access to a named resource. ''' assert callable(handler), 'Handler is not callable' self._access_allowed_handlers[credential_id] = handler def access_allowed(self, credential_id, resource_name): ''' Returns True if the credential_id is authorised to access the resource with the specified resource_name. ''' if not self.resources: # this also triggers lazy-loading of resources # so is required for register_access_allowed_handler # to be called return False handler = self._access_allowed_handlers.get(credential_id) return handler and handler(resource_name) def validate(self): ''' Validates the template. ''' # TODO(sdake) Should return line number of invalid reference # validate overall template (top-level structure) self.t.validate() # Validate Parameter Groups parameter_groups = ParameterGroups(self.t) parameter_groups.validate() # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.keys()) if dup_names: LOG.debug("Duplicate names %s" % dup_names) raise StackValidationFailed(message=_("Duplicate names %s") % dup_names) for res in self.dependencies: try: result = res.validate() except exception.HeatException as ex: LOG.exception(ex) raise ex except Exception as ex: LOG.exception(ex) raise StackValidationFailed(message=strutils.safe_decode( six.text_type(ex))) if result: raise StackValidationFailed(message=result) for val in self.outputs.values(): snippet = val.get('Value', '') try: function.validate(snippet) except Exception as ex: reason = 'Output validation error: %s' % six.text_type(ex) raise StackValidationFailed(message=reason) def requires_deferred_auth(self): ''' Returns whether this stack may need to perform API requests during its lifecycle using the configured deferred authentication method. ''' return any(res.requires_deferred_auth for res in self.values()) def state_set(self, action, status, reason): '''Update the stack state in the database.''' if action not in self.ACTIONS: raise ValueError(_("Invalid action %s") % action) if status not in self.STATUSES: raise ValueError(_("Invalid status %s") % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) if stack is not None: stack.update_and_save({'action': action, 'status': status, 'status_reason': reason}) msg = _('Stack %(action)s %(status)s (%(name)s): %(reason)s') LOG.info(msg % {'action': action, 'status': status, 'name': self.name, 'reason': reason}) notification.send(self) @property def state(self): '''Returns state, tuple of action, status.''' return (self.action, self.status) def timeout_secs(self): ''' Return the stack action timeout in seconds. ''' if self.timeout_mins is None: return cfg.CONF.stack_action_timeout return self.timeout_mins * 60 def preview_resources(self): ''' Preview the stack with all of the resources. ''' return [resource.preview() for resource in self.resources.itervalues()] def create(self): ''' Create the stack and all of the resources. ''' def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def _adopt_kwargs(self, resource): data = self.adopt_stack_data if not data or not data.get('resources'): return {'resource_data': None} return {'resource_data': data['resources'].get(resource.name)} @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): ''' A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specfifed by reverse ''' self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, '%s' % action_l) # If a local _$action_kwargs function exists, call it to get the # action specific argument list, otherwise an empty arg list handle_kwargs = getattr(self, '_%s_kwargs' % action_l, lambda x: {}) return handle(**handle_kwargs(r)) action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action, six.text_type(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def _backup_stack(self, create_if_missing=True): ''' Get a Stack containing any in-progress resources from the previous stack state prior to an update. ''' s = db_api.stack_get_by_name_and_owner_id(self.context, self._backup_name(), owner_id=self.id) if s is not None: LOG.debug('Loaded existing backup stack') return self.load(self.context, stack=s) elif create_if_missing: templ = Template.load(self.context, self.t.id) templ.files = copy.deepcopy(self.t.files) prev = type(self)(self.context, self.name, templ, self.env, owner_id=self.id) prev.store(backup=True) LOG.debug('Created new backup stack') return prev else: return None def adopt(self): ''' Adopt a stack (create stack with all the existing resources). ''' def rollback(): if not self.disable_rollback and self.state == (self.ADOPT, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner( self.stack_task, action=self.ADOPT, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def update(self, newstack): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' self.updated_time = datetime.utcnow() updater = scheduler.TaskRunner(self.update_task, newstack) updater() @scheduler.wrappertask def update_task(self, newstack, action=UPDATE): if action not in (self.UPDATE, self.ROLLBACK): LOG.error(_("Unexpected action %s passed to update!") % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if (action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS)): LOG.debug("Starting update rollback for %s" % self.name) else: self.state_set(action, self.FAILED, 'State invalid for %s' % action) return self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) oldstack = Stack(self.context, self.name, self.t, self.env) backup_stack = self._backup_stack() try: update_task = update.StackUpdate(self, newstack, backup_stack, rollback=action == self.ROLLBACK) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters self.t.files = newstack.t.files self.disable_rollback = newstack.disable_rollback self.timeout_mins = newstack.timeout_mins self._set_param_stackid() try: updater.start(timeout=self.timeout_secs()) yield while not updater.step(): yield finally: self.reset_dependencies() if action == self.UPDATE: reason = 'Stack successfully updated' else: reason = 'Stack rollback completed' stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = 'Timed out' except exception.ResourceFailure as e: reason = six.text_type(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: yield self.update_task(oldstack, action=self.ROLLBACK) return else: LOG.debug('Deleting backup stack') backup_stack.delete(backup=True) # flip the template to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t template_outputs = self.t[self.t.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) # Don't use state_set to do only one update query and avoid race # condition with the COMPLETE status self.action = action self.status = stack_status self.status_reason = reason self.store() notification.send(self) def delete(self, action=DELETE, backup=False): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action not in (self.DELETE, self.ROLLBACK): LOG.error(_("Unexpected action %s passed to delete!") % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) backup_stack = self._backup_stack(False) if backup_stack: for key, backup_resource in backup_stack.resources.items(): # If UpdateReplace is failed, we must restore backup_resource # to existing_stack in case of it may have dependencies in # these stacks. current_resource is the resource that just # created and failed, so put into the backup_stack to delete # anyway. backup_resource_id = backup_resource.resource_id current_resource = self.resources[key] current_resource_id = current_resource.resource_id if backup_resource_id: child_failed = False for child in self.dependencies[current_resource]: # If child resource failed to update, current_resource # should be replaced to resolve dependencies. But this # is not fundamental solution. If there are update # failer and success resources in the children, cannot # delete the stack. if (child.status == child.FAILED and child.action == child.CREATE): child_failed = True if (current_resource.status == current_resource.FAILED or child_failed): # Stack class owns dependencies as set of resource's # objects, so we switch members of the resource that is # needed to delete it. self.resources[key].resource_id = backup_resource_id self.resources[ key].properties = backup_resource.properties backup_stack.resources[ key].resource_id = current_resource_id backup_stack.resources[ key].properties = current_resource.properties backup_stack.delete(backup=True) if backup_stack.status != backup_stack.COMPLETE: errs = backup_stack.status_reason failure = 'Error deleting backup resources: %s' % errs self.state_set(action, self.FAILED, 'Failed to %s : %s' % (action, failure)) return action_task = scheduler.DependencyTaskGroup(self.dependencies, resource.Resource.destroy, reverse=True) try: scheduler.TaskRunner(action_task)(timeout=self.timeout_secs()) except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action, six.text_type(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() if stack_status != self.FAILED and not backup: # Cleanup stored user_creds so they aren't accessible via # the soft-deleted stack which remains in the DB if self.user_creds_id: user_creds = db_api.user_creds_get(self.user_creds_id) # If we created a trust, delete it if user_creds is not None: trust_id = user_creds.get('trust_id') if trust_id: try: self.clients.keystone().delete_trust(trust_id) except Exception as ex: LOG.exception(ex) stack_status = self.FAILED reason = ("Error deleting trust: %s" % six.text_type(ex)) # Delete the stored credentials try: db_api.user_creds_delete(self.context, self.user_creds_id) except exception.NotFound: LOG.info(_("Tried to delete user_creds that do not exist " "(stack=%(stack)s user_creds_id=%(uc)s)") % {'stack': self.id, 'uc': self.user_creds_id}) self.user_creds_id = None self.store() # If the stack has a domain project, delete it if self.stack_user_project_id: try: self.clients.keystone().delete_stack_domain_project( project_id=self.stack_user_project_id) except Exception as ex: LOG.exception(ex) stack_status = self.FAILED reason = "Error deleting project: %s" % six.text_type(ex) self.state_set(action, stack_status, reason) if stack_status != self.FAILED: # delete the stack db_api.stack_delete(self.context, self.id) self.id = None def suspend(self): ''' Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. ''' # No need to suspend if the stack has been suspended if self.state == (self.SUSPEND, self.COMPLETE): LOG.info(_('%s is already suspended') % str(self)) return sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): ''' Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. ''' # No need to resume if the stack has been resumed if self.state == (self.RESUME, self.COMPLETE): LOG.info(_('%s is already resumed') % str(self)) return sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') try: return function.resolve(value) except Exception as ex: self.outputs[key]['error_msg'] = six.text_type(ex) return None def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: scheduler.TaskRunner(res.destroy)() except exception.ResourceFailure as ex: failed = True LOG.error(_('delete: %s') % ex) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: LOG.exception(_('create')) failed = True else: res.state_set(res.CREATE, res.FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list(detailed=False)] return self._zones def set_stack_user_project_id(self, project_id): self.stack_user_project_id = project_id self.store() def prepare_abandon(self): return { 'name': self.name, 'id': self.id, 'action': self.action, 'status': self.status, 'template': self.t.t, 'resources': dict((res.name, res.prepare_abandon()) for res in self.resources.values()) } def resolve_static_data(self, snippet): return self.t.parse(self, snippet) def resolve_runtime_data(self, snippet): """DEPRECATED. Use heat.engine.function.resolve() instead.""" warnings.warn('Stack.resolve_runtime_data() is deprecated. ' 'Use heat.engine.function.resolve() instead', DeprecationWarning) return function.resolve(snippet) def reset_resource_attributes(self): # nothing is cached if no resources exist if not self._resources: return # a change in some resource may have side-effects in the attributes # of other resources, so ensure that attributes are re-calculated for res in self.resources.itervalues(): res.attributes.reset_resolved_values()
class Stack(object): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK') CREATE_IN_PROGRESS = 'CREATE_IN_PROGRESS' CREATE_FAILED = 'CREATE_FAILED' CREATE_COMPLETE = 'CREATE_COMPLETE' DELETE_IN_PROGRESS = 'DELETE_IN_PROGRESS' DELETE_FAILED = 'DELETE_FAILED' DELETE_COMPLETE = 'DELETE_COMPLETE' UPDATE_IN_PROGRESS = 'UPDATE_IN_PROGRESS' UPDATE_COMPLETE = 'UPDATE_COMPLETE' UPDATE_FAILED = 'UPDATE_FAILED' ROLLBACK_IN_PROGRESS = 'ROLLBACK_IN_PROGRESS' ROLLBACK_COMPLETE = 'ROLLBACK_COMPLETE' ROLLBACK_FAILED = 'ROLLBACK_FAILED' created_time = timestamp.Timestamp(db_api.stack_get, 'created_at') updated_time = timestamp.Timestamp(db_api.stack_get, 'updated_at') _zones = None def __init__(self, context, stack_name, tmpl, parameters=None, stack_id=None, state=None, state_description='', timeout_mins=60, resolve_data=True, disable_rollback=True): ''' Initialise from a context, name, Template object and (optionally) Parameters object. The database ID may also be initialised, if the stack is already in the database. ''' if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _("Invalid stack name %s" % stack_name + ", must contain only alphanumeric or " + "\"_-.\" characters, must start with alpha")) self.id = stack_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.state = state self.state_description = state_description self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback resources.initialise() if parameters is None: parameters = Parameters(self.name, self.t) self.parameters = parameters self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the AWS::StackId pseudo parameter ''' # This can fail if constructor called without a valid context, # as it is in many tests try: stack_arn = self.identifier().arn() except (AttributeError, ValueError, TypeError): logger.warning("Unable to set parameters StackId identifier") else: self.parameters.set_stack_id(stack_arn) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True): '''Retrieve a Stack from the database''' if stack is None: stack = db_api.stack_get(context, stack_id) if stack is None: message = 'No stack exists with id "%s"' % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) params = Parameters(stack.name, template, stack.parameters) stack = cls(context, stack.name, template, params, stack.id, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback) return stack def store(self, owner=None): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' new_creds = db_api.user_creds_create(self.context) s = { 'name': self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.parameters.user_parameters(), 'owner_id': owner and owner.id, 'user_creds_id': new_creds.id, 'username': self.context.username, 'tenant': self.context.tenant_id, 'status': self.state, 'status_reason': self.state_description, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over this template's resources in the order that they should be started. ''' return iter(self.dependencies) def __reversed__(self): ''' Return an iterator over this template's resources in the order that they should be stopped. ''' return reversed(self.dependencies) def __len__(self): '''Return the number of resources''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, value): '''Set the resource with the specified name to a specific value''' self.resources[key] = value def __contains__(self, key): '''Determine whether the stack contains the specified resource''' return key in self.resources def keys(self): '''Return a list of resource keys for the stack''' return self.resources.keys() def __str__(self): '''Return a human-readable string representation of the stack''' return 'Stack "%s"' % self.name def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.resources.values(): if r.state in (r.CREATE_IN_PROGRESS, r.CREATE_COMPLETE, r.UPDATE_IN_PROGRESS, r.UPDATE_COMPLETE) and r.FnGetRefId() == refid: return r def validate(self): ''' http://docs.amazonwebservices.com/AWSCloudFormation/latest/\ APIReference/API_ValidateTemplate.html ''' # TODO(sdake) Should return line number of invalid reference for res in self: try: result = res.validate() except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=str(ex)) if result: raise StackValidationFailed(message=result) def state_set(self, new_status, reason): '''Update the stack state in the database''' self.state = new_status self.state_description = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) stack.update_and_save({'status': new_status, 'status_reason': reason}) def create(self): ''' Create the stack and all of the resources. Creation will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' self.state_set(self.CREATE_IN_PROGRESS, 'Stack creation started') stack_status = self.CREATE_COMPLETE reason = 'Stack successfully created' res = None with eventlet.Timeout(self.timeout_mins * 60) as tmo: try: for res in self: if stack_status != self.CREATE_FAILED: try: scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: stack_status = self.CREATE_FAILED reason = 'Resource %s failed with: %s' % (str(res), str(ex)) else: res.state_set(res.CREATE_FAILED, 'Stack creation aborted') except eventlet.Timeout as t: if t is tmo: stack_status = self.CREATE_FAILED reason = 'Timed out waiting for %s' % str(res) else: # not my timeout raise self.state_set(stack_status, reason) if stack_status == self.CREATE_FAILED and not self.disable_rollback: self.delete(action=self.ROLLBACK) def update(self, newstack, action=UPDATE): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' if action not in (self.UPDATE, self.ROLLBACK): logger.error("Unexpected action %s passed to update!" % action) self.state_set(self.UPDATE_FAILED, "Invalid action %s" % action) return if self.state not in (self.CREATE_COMPLETE, self.UPDATE_COMPLETE, self.ROLLBACK_COMPLETE): if (action == self.ROLLBACK and self.state == self.UPDATE_IN_PROGRESS): logger.debug("Starting update rollback for %s" % self.name) else: if action == self.UPDATE: self.state_set(self.UPDATE_FAILED, 'State invalid for update') else: self.state_set(self.ROLLBACK_FAILED, 'State invalid for rollback') return if action == self.UPDATE: self.state_set(self.UPDATE_IN_PROGRESS, 'Stack update started') else: self.state_set(self.ROLLBACK_IN_PROGRESS, 'Stack rollback started') # cache all the resources runtime data. for r in self: r.cache_template() # Now make the resources match the new stack definition with eventlet.Timeout(self.timeout_mins * 60) as tmo: try: # First delete any resources which are not in newstack for res in reversed(self): if not res.name in newstack.keys(): logger.debug("resource %s not found in updated stack" % res.name + " definition, deleting") try: res.destroy() except exception.ResourceFailure as ex: logger.error("Failed to remove %s : %s" % (res.name, str(ex))) raise exception.ResourceUpdateFailed( resource_name=res.name) else: del self.resources[res.name] self.dependencies = self._get_dependencies( self.resources.itervalues()) # Then create any which are defined in newstack but not self for res in newstack: if not res.name in self.keys(): logger.debug("resource %s not found in current stack" % res.name + " definition, adding") res.stack = self self[res.name] = res self.dependencies = self._get_dependencies( self.resources.itervalues()) try: scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.error("Failed to add %s : %s" % (res.name, str(ex))) raise exception.ResourceUpdateFailed( resource_name=res.name) # Now (the hard part :) update existing resources # The Resource base class allows equality-test of resources, # based on the parsed template snippet for the resource. # If this test fails, we call the underlying resource.update # # FIXME : Implement proper update logic for the resources # AWS define three update strategies, applied depending # on the resource and what is being updated within a # resource : # - Update with no interruption # - Update with some interruption # - Update requires replacement # # Currently all resource have a default handle_update method # which returns "requires replacement" (res.UPDATE_REPLACE) for res in newstack: # Compare resolved pre/post update resource snippets, # note the new resource snippet is resolved in the context # of the existing stack (which is the stack being updated) old_snippet = self[res.name].parsed_template(cached=True) new_snippet = self.resolve_runtime_data(res.t) if old_snippet != new_snippet: # Can fail if underlying resource class does not # implement update logic or update requires replacement retval = self[res.name].update(new_snippet) if retval == self[res.name].UPDATE_COMPLETE: logger.info("Resource %s for stack %s updated" % (res.name, self.name)) elif retval == self[res.name].UPDATE_REPLACE: logger.info("Resource %s for stack %s" % (res.name, self.name) + " update requires replacement") # Resource requires replacement for update try: self[res.name].destroy() except exception.ResourceFailure as ex: logger.error("Failed to delete %s : %s" % (res.name, str(ex))) raise exception.ResourceUpdateFailed( resource_name=res.name) else: res.stack = self self[res.name] = res self.dependencies = self._get_dependencies( self.resources.itervalues()) try: scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.error("Failed to create %s : %s" % (res.name, str(ex))) raise exception.ResourceUpdateFailed( resource_name=res.name) else: logger.error("Failed to %s %s" % (action, res.name)) raise exception.ResourceUpdateFailed( resource_name=res.name) if action == self.UPDATE: stack_status = self.UPDATE_COMPLETE reason = 'Stack successfully updated' else: stack_status = self.ROLLBACK_COMPLETE reason = 'Stack rollback completed' except eventlet.Timeout as t: if t is tmo: stack_status = self.UPDATE_FAILED reason = 'Timed out waiting for %s' % str(res) else: # not my timeout raise except exception.ResourceUpdateFailed as e: reason = str(e) or "Error : %s" % type(e) if action == self.UPDATE: stack_status = self.UPDATE_FAILED # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state # Note - ensure nothing after the "flip the template..." # section above can raise ResourceUpdateFailed or this # will not work ;) if self.disable_rollback: stack_status = self.UPDATE_FAILED else: oldstack = Stack(self.context, self.name, self.t, self.parameters) self.update(oldstack, action=self.ROLLBACK) return else: stack_status = self.ROLLBACK_FAILED self.state_set(stack_status, reason) # flip the template & parameters to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t self.parameters = newstack.parameters template_outputs = self.t[template.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action == self.DELETE: self.state_set(self.DELETE_IN_PROGRESS, 'Stack deletion started') elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_IN_PROGRESS, 'Stack rollback started') else: logger.error("Unexpected action %s passed to delete!" % action) self.state_set(self.DELETE_FAILED, "Invalid action %s" % action) return failures = [] for res in reversed(self): try: res.destroy() except exception.ResourceFailure as ex: logger.error('Failed to delete %s error: %s' % (str(res), str(ex))) failures.append(str(res)) if failures: if action == self.DELETE: self.state_set(self.DELETE_FAILED, 'Failed to delete ' + ', '.join(failures)) elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_FAILED, 'Failed to rollback ' + ', '.join(failures)) else: if action == self.DELETE: self.state_set(self.DELETE_COMPLETE, 'Deleted successfully') elif action == self.ROLLBACK: self.state_set(self.ROLLBACK_COMPLETE, 'Rollback completed') db_api.stack_delete(self.context, self.id) self.id = None def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') return self.resolve_runtime_data(value) def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: res.destroy() except exception.ResourceFailure as ex: failed = True logger.error('delete: %s' % str(ex)) for res in deps: if not failed: try: scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception('create') failed = True else: res.state_set(res.CREATE_FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list( detailed=False) ] return self._zones def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)
class Stack(object): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME ) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME') STATUSES = (IN_PROGRESS, FAILED, COMPLETE ) = ('IN_PROGRESS', 'FAILED', 'COMPLETE') created_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), 'created_at') updated_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), 'updated_at') _zones = None def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None resources.initialise() self.env = env or environment.Environment({}) self.parameters = Parameters(self.name, self.t, user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} @property def resources(self): if self._resources is None: template_resources = self.t[template.RESOURCES] self._resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) return self._resources @property def dependencies(self): if self._dependencies is None: self._dependencies = self._get_dependencies( self.resources.itervalues()) return self._dependencies def reset_dependencies(self): self._dependencies = None @property def root_stack(self): ''' Return the root stack if this is nested (otherwise return self). ''' if (self.parent_resource and self.parent_resource.stack): return self.parent_resource.stack.root_stack return self def total_resources(self): ''' Total number of resources in a stack, including nested stacks below. ''' total = 0 for res in iter(self.resources.values()): if hasattr(res, 'nested') and res.nested(): total += res.nested().total_resources() total += 1 return total def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the AWS::StackId pseudo parameter ''' # This can fail if constructor called without a valid context, # as it is in many tests try: stack_arn = self.identifier().arn() except (AttributeError, ValueError, TypeError): logger.warning("Unable to set parameters StackId identifier") else: self.parameters.set_stack_id(stack_arn) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True, parent_resource=None, show_deleted=True): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted) if stack is None: message = 'No stack exists with id "%s"' % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) env = environment.Environment(stack.parameters) stack = cls(context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id) return stack def store(self, backup=False): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' s = { 'name': self._backup_name() if backup else self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.env.user_env_as_dict(), 'owner_id': self.owner_id, 'username': self.context.username, 'tenant': self.context.tenant_id, 'action': self.action, 'status': self.status, 'status_reason': self.status_reason, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: # Create a context containing a trust_id and trustor_user_id # if trusts are enabled if cfg.CONF.deferred_auth_method == 'trusts': trust_context = self.clients.keystone().create_trust_context() new_creds = db_api.user_creds_create(trust_context) else: new_creds = db_api.user_creds_create(self.context) s['user_creds_id'] = new_creds.id new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def _backup_name(self): return '%s*' % self.name def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over this template's resources in the order that they should be started. ''' return iter(self.dependencies) def __reversed__(self): ''' Return an iterator over this template's resources in the order that they should be stopped. ''' return reversed(self.dependencies) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, value): '''Set the resource with the specified name to a specific value.''' self.resources[key] = value def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' return key in self.resources def keys(self): '''Return a list of resource keys for the stack.''' return self.resources.keys() def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s"' % self.name def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.resources.values(): if r.state in ( (r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.RESUME, r.IN_PROGRESS), (r.RESUME, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE)) and r.FnGetRefId() == refid: return r def validate(self): ''' http://docs.amazonwebservices.com/AWSCloudFormation/latest/\ APIReference/API_ValidateTemplate.html ''' # TODO(sdake) Should return line number of invalid reference # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.resources.keys()) if dup_names: logger.debug("Duplicate names %s" % dup_names) raise StackValidationFailed(message="Duplicate names %s" % dup_names) for res in self: try: result = res.validate() except exception.Error as ex: logger.exception(ex) raise ex except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=str(ex)) if result: raise StackValidationFailed(message=result) def requires_deferred_auth(self): ''' Returns whether this stack may need to perform API requests during its lifecycle using the configured deferred authentication method. ''' return any(res.requires_deferred_auth for res in self) def state_set(self, action, status, reason): '''Update the stack state in the database.''' if action not in self.ACTIONS: raise ValueError("Invalid action %s" % action) if status not in self.STATUSES: raise ValueError("Invalid status %s" % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) stack.update_and_save({'action': action, 'status': status, 'status_reason': reason}) @property def state(self): '''Returns state, tuple of action, status.''' return (self.action, self.status) def timeout_secs(self): ''' Return the stack creation timeout in seconds, or None if no timeout should be used. ''' if self.timeout_mins is None: return None return self.timeout_mins * 60 def create(self): ''' Create the stack and all of the resources. ''' def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): ''' A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specfifed by reverse ''' self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action.lower() res = None def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, '%s' % action_l) return handle() action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def _backup_stack(self, create_if_missing=True): ''' Get a Stack containing any in-progress resources from the previous stack state prior to an update. ''' s = db_api.stack_get_by_name(self.context, self._backup_name(), owner_id=self.id) if s is not None: logger.debug('Loaded existing backup stack') return self.load(self.context, stack=s) elif create_if_missing: prev = type(self)(self.context, self.name, self.t, self.env, owner_id=self.id) prev.store(backup=True) logger.debug('Created new backup stack') return prev else: return None def update(self, newstack): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' updater = scheduler.TaskRunner(self.update_task, newstack) updater() @scheduler.wrappertask def update_task(self, newstack, action=UPDATE): if action not in (self.UPDATE, self.ROLLBACK): logger.error("Unexpected action %s passed to update!" % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if (action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS)): logger.debug("Starting update rollback for %s" % self.name) else: self.state_set(action, self.FAILED, 'State invalid for %s' % action) return self.state_set(self.UPDATE, self.IN_PROGRESS, 'Stack %s started' % action) oldstack = Stack(self.context, self.name, self.t, self.env) backup_stack = self._backup_stack() try: update_task = update.StackUpdate(self, newstack, backup_stack, rollback=action == self.ROLLBACK) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters try: updater.start(timeout=self.timeout_secs()) yield while not updater.step(): yield finally: self.reset_dependencies() if action == self.UPDATE: reason = 'Stack successfully updated' else: reason = 'Stack rollback completed' stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = 'Timed out' except exception.ResourceFailure as e: reason = str(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: yield self.update_task(oldstack, action=self.ROLLBACK) return else: logger.debug('Deleting backup stack') backup_stack.delete(backup=True) self.state_set(action, stack_status, reason) # flip the template to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t template_outputs = self.t[template.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE, backup=False): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action not in (self.DELETE, self.ROLLBACK): logger.error("Unexpected action %s passed to delete!" % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action.lower() self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) backup_stack = self._backup_stack(False) if backup_stack is not None: backup_stack.delete(backup=True) if backup_stack.status != backup_stack.COMPLETE: errs = backup_stack.status_reason failure = 'Error deleting backup resources: %s' % errs self.state_set(action, self.FAILED, 'Failed to %s : %s' % (action, failure)) return action_task = scheduler.DependencyTaskGroup(self.dependencies, resource.Resource.destroy, reverse=True) try: scheduler.TaskRunner(action_task)(timeout=self.timeout_secs()) except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() if stack_status != self.FAILED and not backup: # If we created a trust, delete it stack = db_api.stack_get(self.context, self.id) user_creds = db_api.user_creds_get(stack.user_creds_id) trust_id = user_creds.get('trust_id') if trust_id: try: self.clients.keystone().delete_trust(trust_id) except Exception as ex: logger.exception(ex) stack_status = self.FAILED reason = "Error deleting trust: %s" % str(ex) self.state_set(action, stack_status, reason) if stack_status != self.FAILED: # delete the stack db_api.stack_delete(self.context, self.id) self.id = None def suspend(self): ''' Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): ''' Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') return self.resolve_runtime_data(value) def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: scheduler.TaskRunner(res.destroy)() except exception.ResourceFailure as ex: failed = True logger.error('delete: %s' % str(ex)) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception('create') failed = True else: res.state_set(res.CREATE, res.FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list(detailed=False)] return self._zones def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)
class Stack(object): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME') STATUSES = (IN_PROGRESS, FAILED, COMPLETE) = ('IN_PROGRESS', 'FAILED', 'COMPLETE') created_time = timestamp.Timestamp( functools.partial(db_api.stack_get, show_deleted=True), 'created_at') updated_time = timestamp.Timestamp( functools.partial(db_api.stack_get, show_deleted=True), 'updated_at') _zones = None def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha') % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource resources.initialise() self.env = env or environment.Environment({}) self.parameters = Parameters(self.name, self.t, user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} template_resources = self.t[template.RESOURCES] self.resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) self.dependencies = self._get_dependencies(self.resources.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the AWS::StackId pseudo parameter ''' # This can fail if constructor called without a valid context, # as it is in many tests try: stack_arn = self.identifier().arn() except (AttributeError, ValueError, TypeError): logger.warning("Unable to set parameters StackId identifier") else: self.parameters.set_stack_id(stack_arn) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True, parent_resource=None, show_deleted=True): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted) if stack is None: message = 'No stack exists with id "%s"' % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) env = environment.Environment(stack.parameters) stack = cls(context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id) return stack def store(self): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' new_creds = db_api.user_creds_create(self.context) s = { 'name': self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.env.user_env_as_dict(), 'owner_id': self.owner_id, 'user_creds_id': new_creds.id, 'username': self.context.username, 'tenant': self.context.tenant_id, 'action': self.action, 'status': self.status, 'status_reason': self.status_reason, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over this template's resources in the order that they should be started. ''' return iter(self.dependencies) def __reversed__(self): ''' Return an iterator over this template's resources in the order that they should be stopped. ''' return reversed(self.dependencies) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, value): '''Set the resource with the specified name to a specific value.''' self.resources[key] = value def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' return key in self.resources def keys(self): '''Return a list of resource keys for the stack.''' return self.resources.keys() def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s"' % self.name def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.resources.values(): if r.state in ((r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE)) and r.FnGetRefId() == refid: return r def validate(self): ''' http://docs.amazonwebservices.com/AWSCloudFormation/latest/\ APIReference/API_ValidateTemplate.html ''' # TODO(sdake) Should return line number of invalid reference # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.resources.keys()) if dup_names: logger.debug("Duplicate names %s" % dup_names) raise StackValidationFailed(message="Duplicate names %s" % dup_names) for res in self: try: result = res.validate() except ServerError as ex: logger.exception(ex) raise ex except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=str(ex)) if result: raise StackValidationFailed(message=result) def state_set(self, action, status, reason): '''Update the stack state in the database.''' if action not in self.ACTIONS: raise ValueError("Invalid action %s" % action) if status not in self.STATUSES: raise ValueError("Invalid status %s" % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) stack.update_and_save({ 'action': action, 'status': status, 'status_reason': reason }) @property def state(self): '''Returns state, tuple of action, status.''' return (self.action, self.status) def timeout_secs(self): ''' Return the stack creation timeout in seconds, or None if no timeout should be used. ''' if self.timeout_mins is None: return None return self.timeout_mins * 60 def create(self): ''' Create the stack and all of the resources. ''' def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): ''' A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specfifed by reverse ''' self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action.lower() res = None def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, '%s' % action_l) return handle() action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def update(self, newstack, action=UPDATE): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' if action not in (self.UPDATE, self.ROLLBACK): logger.error("Unexpected action %s passed to update!" % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if (action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS)): logger.debug("Starting update rollback for %s" % self.name) else: self.state_set(action, self.FAILED, 'State invalid for %s' % action) return self.state_set(self.UPDATE, self.IN_PROGRESS, 'Stack %s started' % action) oldstack = Stack(self.context, self.name, self.t, self.env) try: update_task = update.StackUpdate(self, newstack, oldstack) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters try: updater(timeout=self.timeout_secs()) finally: cur_deps = self._get_dependencies(self.resources.itervalues()) self.dependencies = cur_deps if action == self.UPDATE: reason = 'Stack successfully updated' else: reason = 'Stack rollback completed' stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = 'Timed out' except exception.ResourceFailure as e: reason = str(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: self.update(oldstack, action=self.ROLLBACK) return self.state_set(action, stack_status, reason) # flip the template to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t template_outputs = self.t[template.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action not in (self.DELETE, self.ROLLBACK): logger.error("Unexpected action %s passed to delete!" % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) failures = [] for res in reversed(self): try: res.destroy() except exception.ResourceFailure as ex: logger.error('Failed to delete %s error: %s' % (str(res), str(ex))) failures.append(str(res)) if failures: self.state_set(action, self.FAILED, 'Failed to %s : %s' % (action, ', '.join(failures))) else: self.state_set(action, self.COMPLETE, '%s completed' % action) db_api.stack_delete(self.context, self.id) self.id = None def suspend(self): ''' Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): ''' Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') return self.resolve_runtime_data(value) def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: res.destroy() except exception.ResourceFailure as ex: failed = True logger.error('delete: %s' % str(ex)) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception('create') failed = True else: res.state_set(res.CREATE, res.FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list( detailed=False) ] return self._zones def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)
class Stack(collections.Mapping): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME, ADOPT ) = ('CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME', 'ADOPT') STATUSES = (IN_PROGRESS, FAILED, COMPLETE ) = ('IN_PROGRESS', 'FAILED', 'COMPLETE') created_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), 'created_at') updated_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), 'updated_at') _zones = None def __init__(self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason='', timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, adopt_stack_data=None): ''' Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. ''' if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError(_('Invalid stack name %s' ' must contain only alphanumeric or ' '\"_-.\" characters, must start with alpha' ) % stack_name) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None self._access_allowed_handlers = {} self.adopt_stack_data = adopt_stack_data resources.initialise() self.env = env or environment.Environment({}) self.parameters = self.t.parameters(self.identifier(), user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[self.t.OUTPUTS]) else: self.outputs = {} @property def resources(self): if self._resources is None: template_resources = self.t[self.t.RESOURCES] self._resources = dict((name, resource.Resource(name, data, self)) for (name, data) in template_resources.items()) return self._resources @property def dependencies(self): if self._dependencies is None: self._dependencies = self._get_dependencies( self.resources.itervalues()) return self._dependencies def reset_dependencies(self): self._dependencies = None @property def root_stack(self): ''' Return the root stack if this is nested (otherwise return self). ''' if (self.parent_resource and self.parent_resource.stack): return self.parent_resource.stack.root_stack return self def total_resources(self): ''' Return the total number of resources in a stack, including nested stacks below. ''' def total_nested(res): get_nested = getattr(res, 'nested', None) if callable(get_nested): nested_stack = get_nested() if nested_stack is not None: return nested_stack.total_resources() return 0 return len(self) + sum(total_nested(res) for res in self.itervalues()) def _set_param_stackid(self): ''' Update self.parameters with the current ARN which is then provided via the Parameters class as the StackId pseudo parameter ''' if not self.parameters.set_stack_id(self.identifier()): logger.warning(_("Unable to set parameters StackId identifier")) @staticmethod def _get_dependencies(resources): '''Return the dependency graph for a list of resources.''' deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True, parent_resource=None, show_deleted=True): '''Retrieve a Stack from the database.''' if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted) if stack is None: message = _('No stack exists with id "%s"') % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) env = environment.Environment(stack.parameters) stack = cls(context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id) return stack def store(self, backup=False): ''' Store the stack in the database and return its ID If self.id is set, we update the existing stack ''' s = { 'name': self._backup_name() if backup else self.name, 'raw_template_id': self.t.store(self.context), 'parameters': self.env.user_env_as_dict(), 'owner_id': self.owner_id, 'username': self.context.username, 'tenant': self.context.tenant_id, 'action': self.action, 'status': self.status, 'status_reason': self.status_reason, 'timeout': self.timeout_mins, 'disable_rollback': self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: # Create a context containing a trust_id and trustor_user_id # if trusts are enabled if cfg.CONF.deferred_auth_method == 'trusts': trust_context = self.clients.keystone().create_trust_context() new_creds = db_api.user_creds_create(trust_context) else: new_creds = db_api.user_creds_create(self.context) s['user_creds_id'] = new_creds.id new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def _backup_name(self): return '%s*' % self.name def identifier(self): ''' Return an identifier for this stack. ''' return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): ''' Return an iterator over the resource names. ''' return iter(self.resources) def __len__(self): '''Return the number of resources.''' return len(self.resources) def __getitem__(self, key): '''Get the resource with the specified name.''' return self.resources[key] def __setitem__(self, key, resource): '''Set the resource with the specified name to a specific value.''' resource.stack = self self.resources[key] = resource def __delitem__(self, key): '''Remove the resource with the specified name.''' del self.resources[key] def __contains__(self, key): '''Determine whether the stack contains the specified resource.''' return key in self.resources def __eq__(self, other): ''' Compare two Stacks for equality. Stacks are considered equal only if they are identical. ''' return self is other def __str__(self): '''Return a human-readable string representation of the stack.''' return 'Stack "%s" [%s]' % (self.name, self.id) def resource_by_refid(self, refid): ''' Return the resource in this stack with the specified refid, or None if not found ''' for r in self.values(): if r.state in ( (r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.RESUME, r.IN_PROGRESS), (r.RESUME, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE)) and r.FnGetRefId() == refid: return r def register_access_allowed_handler(self, credential_id, handler): ''' Register a function which determines whether the credentials with a give ID can have access to a named resource. ''' assert callable(handler), 'Handler is not callable' self._access_allowed_handlers[credential_id] = handler def access_allowed(self, credential_id, resource_name): ''' Returns True if the credential_id is authorised to access the resource with the specified resource_name. ''' if not self.resources: # this also triggers lazy-loading of resources # so is required for register_access_allowed_handler # to be called return False handler = self._access_allowed_handlers.get(credential_id) return handler and handler(resource_name) def validate(self): ''' Validates the template. ''' # TODO(sdake) Should return line number of invalid reference # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.keys()) if dup_names: logger.debug(_("Duplicate names %s") % dup_names) raise StackValidationFailed(message=_("Duplicate names %s") % dup_names) for res in self.dependencies: try: result = res.validate() except exception.Error as ex: logger.exception(ex) raise ex except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=strutils.safe_decode( six.text_type(ex))) if result: raise StackValidationFailed(message=result) def requires_deferred_auth(self): ''' Returns whether this stack may need to perform API requests during its lifecycle using the configured deferred authentication method. ''' return any(res.requires_deferred_auth for res in self.values()) def state_set(self, action, status, reason): '''Update the stack state in the database.''' if action not in self.ACTIONS: raise ValueError(_("Invalid action %s") % action) if status not in self.STATUSES: raise ValueError(_("Invalid status %s") % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) if stack is not None: stack.update_and_save({'action': action, 'status': status, 'status_reason': reason}) notification.send(self) @property def state(self): '''Returns state, tuple of action, status.''' return (self.action, self.status) def timeout_secs(self): ''' Return the stack creation timeout in seconds, or None if no timeout should be used. ''' if self.timeout_mins is None: return None return self.timeout_mins * 60 def create(self): ''' Create the stack and all of the resources. ''' def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def _adopt_kwargs(self, resource): data = self.adopt_stack_data if not data or not data.get('resources'): return {'resource_data': None} return {'resource_data': data['resources'].get(resource.name)} @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): ''' A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specfifed by reverse ''' self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action.lower() def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, '%s' % action_l) # If a local _$action_kwargs function exists, call it to get the # action specific argument list, otherwise an empty arg list handle_kwargs = getattr(self, '_%s_kwargs' % action_l, lambda x: {}) return handle(**handle_kwargs(r)) action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def _backup_stack(self, create_if_missing=True): ''' Get a Stack containing any in-progress resources from the previous stack state prior to an update. ''' s = db_api.stack_get_by_name_and_owner_id(self.context, self._backup_name(), owner_id=self.id) if s is not None: logger.debug(_('Loaded existing backup stack')) return self.load(self.context, stack=s) elif create_if_missing: templ = Template.load(self.context, self.t.id) templ.files = copy.deepcopy(self.t.files) prev = type(self)(self.context, self.name, templ, self.env, owner_id=self.id) prev.store(backup=True) logger.debug(_('Created new backup stack')) return prev else: return None def adopt(self): ''' Adopt a stack (create stack with all the existing resources). ''' def rollback(): if not self.disable_rollback and self.state == (self.ADOPT, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner( self.stack_task, action=self.ADOPT, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) def update(self, newstack): ''' Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor ''' updater = scheduler.TaskRunner(self.update_task, newstack) updater() @scheduler.wrappertask def update_task(self, newstack, action=UPDATE): if action not in (self.UPDATE, self.ROLLBACK): logger.error(_("Unexpected action %s passed to update!") % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if (action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS)): logger.debug(_("Starting update rollback for %s") % self.name) else: self.state_set(action, self.FAILED, 'State invalid for %s' % action) return self.state_set(self.UPDATE, self.IN_PROGRESS, 'Stack %s started' % action) oldstack = Stack(self.context, self.name, self.t, self.env) backup_stack = self._backup_stack() try: update_task = update.StackUpdate(self, newstack, backup_stack, rollback=action == self.ROLLBACK) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters self.t.files = newstack.t.files self._set_param_stackid() try: updater.start(timeout=self.timeout_secs()) yield while not updater.step(): yield finally: self.reset_dependencies() if action == self.UPDATE: reason = 'Stack successfully updated' else: reason = 'Stack rollback completed' stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = 'Timed out' except exception.ResourceFailure as e: reason = str(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: yield self.update_task(oldstack, action=self.ROLLBACK) return else: logger.debug(_('Deleting backup stack')) backup_stack.delete(backup=True) self.state_set(action, stack_status, reason) # flip the template to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t template_outputs = self.t[self.t.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE, backup=False): ''' Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. ''' if action not in (self.DELETE, self.ROLLBACK): logger.error(_("Unexpected action %s passed to delete!") % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return stack_status = self.COMPLETE reason = 'Stack %s completed successfully' % action.lower() self.state_set(action, self.IN_PROGRESS, 'Stack %s started' % action) backup_stack = self._backup_stack(False) if backup_stack is not None: backup_stack.delete(backup=True) if backup_stack.status != backup_stack.COMPLETE: errs = backup_stack.status_reason failure = 'Error deleting backup resources: %s' % errs self.state_set(action, self.FAILED, 'Failed to %s : %s' % (action, failure)) return action_task = scheduler.DependencyTaskGroup(self.dependencies, resource.Resource.destroy, reverse=True) try: scheduler.TaskRunner(action_task)(timeout=self.timeout_secs()) except exception.ResourceFailure as ex: stack_status = self.FAILED reason = 'Resource %s failed: %s' % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = '%s timed out' % action.title() if stack_status != self.FAILED and not backup: # If we created a trust, delete it stack = db_api.stack_get(self.context, self.id) user_creds = db_api.user_creds_get(stack.user_creds_id) trust_id = user_creds.get('trust_id') if trust_id: try: self.clients.keystone().delete_trust(trust_id) except Exception as ex: logger.exception(ex) stack_status = self.FAILED reason = "Error deleting trust: %s" % str(ex) self.state_set(action, stack_status, reason) if stack_status != self.FAILED: # delete the stack db_api.stack_delete(self.context, self.id) self.id = None def suspend(self): ''' Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): ''' Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. ''' sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): ''' Get the value of the specified stack output. ''' value = self.outputs[key].get('Value', '') return self.resolve_runtime_data(value) def restart_resource(self, resource_name): ''' stop resource_name and all that depend on it start resource_name and all that depend on it ''' deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: scheduler.TaskRunner(res.destroy)() except exception.ResourceFailure as ex: failed = True logger.error(_('delete: %s') % str(ex)) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception(_('create')) failed = True else: res.state_set(res.CREATE, res.FAILED, 'Resource restart aborted') # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [ zone.zoneName for zone in self.clients.nova().availability_zones.list(detailed=False)] return self._zones def set_deletion_policy(self, policy): for res in self.resources.values(): res.set_deletion_policy(policy) def get_abandon_data(self): return { 'name': self.name, 'id': self.id, 'action': self.action, 'status': self.status, 'template': self.t.t, 'resources': dict((res.name, res.get_abandon_data()) for res in self.resources.values()) } def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)
class Stack(collections.Mapping): ACTIONS = (CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME) = ( "CREATE", "DELETE", "UPDATE", "ROLLBACK", "SUSPEND", "RESUME", ) STATUSES = (IN_PROGRESS, FAILED, COMPLETE) = ("IN_PROGRESS", "FAILED", "COMPLETE") created_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), "created_at") updated_time = timestamp.Timestamp(functools.partial(db_api.stack_get, show_deleted=True), "updated_at") _zones = None def __init__( self, context, stack_name, tmpl, env=None, stack_id=None, action=None, status=None, status_reason="", timeout_mins=60, resolve_data=True, disable_rollback=True, parent_resource=None, owner_id=None, ): """ Initialise from a context, name, Template object and (optionally) Environment object. The database ID may also be initialised, if the stack is already in the database. """ if owner_id is None: if re.match("[a-zA-Z][a-zA-Z0-9_.-]*$", stack_name) is None: raise ValueError( _( "Invalid stack name %s" " must contain only alphanumeric or " '"_-." characters, must start with alpha' ) % stack_name ) self.id = stack_id self.owner_id = owner_id self.context = context self.clients = Clients(context) self.t = tmpl self.name = stack_name self.action = action self.status = status self.status_reason = status_reason self.timeout_mins = timeout_mins self.disable_rollback = disable_rollback self.parent_resource = parent_resource self._resources = None self._dependencies = None resources.initialise() self.env = env or environment.Environment({}) self.parameters = Parameters(self.name, self.t, user_params=self.env.params) self._set_param_stackid() if resolve_data: self.outputs = self.resolve_static_data(self.t[template.OUTPUTS]) else: self.outputs = {} @property def resources(self): if self._resources is None: template_resources = self.t[template.RESOURCES] self._resources = dict( (name, resource.Resource(name, data, self)) for (name, data) in template_resources.items() ) return self._resources @property def dependencies(self): if self._dependencies is None: self._dependencies = self._get_dependencies(self.resources.itervalues()) return self._dependencies def reset_dependencies(self): self._dependencies = None @property def root_stack(self): """ Return the root stack if this is nested (otherwise return self). """ if self.parent_resource and self.parent_resource.stack: return self.parent_resource.stack.root_stack return self def total_resources(self): """ Return the total number of resources in a stack, including nested stacks below. """ def total_nested(res): get_nested = getattr(res, "nested", None) if callable(get_nested): nested_stack = get_nested() if nested_stack is not None: return nested_stack.total_resources() return 0 return len(self) + sum(total_nested(res) for res in self.itervalues()) def _set_param_stackid(self): """ Update self.parameters with the current ARN which is then provided via the Parameters class as the AWS::StackId pseudo parameter """ # This can fail if constructor called without a valid context, # as it is in many tests try: stack_arn = self.identifier().arn() except (AttributeError, ValueError, TypeError): logger.warning(_("Unable to set parameters StackId identifier")) else: self.parameters.set_stack_id(stack_arn) @staticmethod def _get_dependencies(resources): """Return the dependency graph for a list of resources.""" deps = dependencies.Dependencies() for resource in resources: resource.add_dependencies(deps) return deps @classmethod def load(cls, context, stack_id=None, stack=None, resolve_data=True, parent_resource=None, show_deleted=True): """Retrieve a Stack from the database.""" if stack is None: stack = db_api.stack_get(context, stack_id, show_deleted=show_deleted) if stack is None: message = _('No stack exists with id "%s"') % str(stack_id) raise exception.NotFound(message) template = Template.load(context, stack.raw_template_id) env = environment.Environment(stack.parameters) stack = cls( context, stack.name, template, env, stack.id, stack.action, stack.status, stack.status_reason, stack.timeout, resolve_data, stack.disable_rollback, parent_resource, owner_id=stack.owner_id, ) return stack def store(self, backup=False): """ Store the stack in the database and return its ID If self.id is set, we update the existing stack """ s = { "name": self._backup_name() if backup else self.name, "raw_template_id": self.t.store(self.context), "parameters": self.env.user_env_as_dict(), "owner_id": self.owner_id, "username": self.context.username, "tenant": self.context.tenant_id, "action": self.action, "status": self.status, "status_reason": self.status_reason, "timeout": self.timeout_mins, "disable_rollback": self.disable_rollback, } if self.id: db_api.stack_update(self.context, self.id, s) else: # Create a context containing a trust_id and trustor_user_id # if trusts are enabled if cfg.CONF.deferred_auth_method == "trusts": trust_context = self.clients.keystone().create_trust_context() new_creds = db_api.user_creds_create(trust_context) else: new_creds = db_api.user_creds_create(self.context) s["user_creds_id"] = new_creds.id new_s = db_api.stack_create(self.context, s) self.id = new_s.id self._set_param_stackid() return self.id def _backup_name(self): return "%s*" % self.name def identifier(self): """ Return an identifier for this stack. """ return identifier.HeatIdentifier(self.context.tenant_id, self.name, self.id) def __iter__(self): """ Return an iterator over the resource names. """ return iter(self.resources) def __len__(self): """Return the number of resources.""" return len(self.resources) def __getitem__(self, key): """Get the resource with the specified name.""" return self.resources[key] def __setitem__(self, key, resource): """Set the resource with the specified name to a specific value.""" resource.stack = self self.resources[key] = resource def __delitem__(self, key): """Remove the resource with the specified name.""" del self.resources[key] def __contains__(self, key): """Determine whether the stack contains the specified resource.""" return key in self.resources def __eq__(self, other): """ Compare two Stacks for equality. Stacks are considered equal only if they are identical. """ return self is other def __str__(self): """Return a human-readable string representation of the stack.""" return 'Stack "%s"' % self.name def resource_by_refid(self, refid): """ Return the resource in this stack with the specified refid, or None if not found """ for r in self.values(): if ( r.state in ( (r.CREATE, r.IN_PROGRESS), (r.CREATE, r.COMPLETE), (r.RESUME, r.IN_PROGRESS), (r.RESUME, r.COMPLETE), (r.UPDATE, r.IN_PROGRESS), (r.UPDATE, r.COMPLETE), ) and r.FnGetRefId() == refid ): return r def validate(self): """ Validates the template. """ # TODO(sdake) Should return line number of invalid reference # Check duplicate names between parameters and resources dup_names = set(self.parameters.keys()) & set(self.keys()) if dup_names: logger.debug(_("Duplicate names %s") % dup_names) raise StackValidationFailed(message=_("Duplicate names %s") % dup_names) for res in self.dependencies: try: result = res.validate() except exception.Error as ex: logger.exception(ex) raise ex except Exception as ex: logger.exception(ex) raise StackValidationFailed(message=strutils.safe_decode(six.text_type(ex))) if result: raise StackValidationFailed(message=result) def requires_deferred_auth(self): """ Returns whether this stack may need to perform API requests during its lifecycle using the configured deferred authentication method. """ return any(res.requires_deferred_auth for res in self.values()) def state_set(self, action, status, reason): """Update the stack state in the database.""" if action not in self.ACTIONS: raise ValueError(_("Invalid action %s") % action) if status not in self.STATUSES: raise ValueError(_("Invalid status %s") % status) self.action = action self.status = status self.status_reason = reason if self.id is None: return stack = db_api.stack_get(self.context, self.id) stack.update_and_save({"action": action, "status": status, "status_reason": reason}) notification.send(self) @property def state(self): """Returns state, tuple of action, status.""" return (self.action, self.status) def timeout_secs(self): """ Return the stack creation timeout in seconds, or None if no timeout should be used. """ if self.timeout_mins is None: return None return self.timeout_mins * 60 def create(self): """ Create the stack and all of the resources. """ def rollback(): if not self.disable_rollback and self.state == (self.CREATE, self.FAILED): self.delete(action=self.ROLLBACK) creator = scheduler.TaskRunner(self.stack_task, action=self.CREATE, reverse=False, post_func=rollback) creator(timeout=self.timeout_secs()) @scheduler.wrappertask def stack_task(self, action, reverse=False, post_func=None): """ A task to perform an action on the stack and all of the resources in forward or reverse dependency order as specfifed by reverse """ self.state_set(action, self.IN_PROGRESS, "Stack %s started" % action) stack_status = self.COMPLETE reason = "Stack %s completed successfully" % action.lower() res = None def resource_action(r): # Find e.g resource.create and call it action_l = action.lower() handle = getattr(r, "%s" % action_l) return handle() action_task = scheduler.DependencyTaskGroup(self.dependencies, resource_action, reverse) try: yield action_task() except exception.ResourceFailure as ex: stack_status = self.FAILED reason = "Resource %s failed: %s" % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = "%s timed out" % action.title() self.state_set(action, stack_status, reason) if callable(post_func): post_func() def _backup_stack(self, create_if_missing=True): """ Get a Stack containing any in-progress resources from the previous stack state prior to an update. """ s = db_api.stack_get_by_name(self.context, self._backup_name(), owner_id=self.id) if s is not None: logger.debug(_("Loaded existing backup stack")) return self.load(self.context, stack=s) elif create_if_missing: templ = Template.load(self.context, self.t.id) templ.files = copy.deepcopy(self.t.files) prev = type(self)(self.context, self.name, templ, self.env, owner_id=self.id) prev.store(backup=True) logger.debug(_("Created new backup stack")) return prev else: return None def update(self, newstack): """ Compare the current stack with newstack, and where necessary create/update/delete the resources until this stack aligns with newstack. Note update of existing stack resources depends on update being implemented in the underlying resource types Update will fail if it exceeds the specified timeout. The default is 60 minutes, set in the constructor """ updater = scheduler.TaskRunner(self.update_task, newstack) updater() @scheduler.wrappertask def update_task(self, newstack, action=UPDATE): if action not in (self.UPDATE, self.ROLLBACK): logger.error(_("Unexpected action %s passed to update!") % action) self.state_set(self.UPDATE, self.FAILED, "Invalid action %s" % action) return if self.status != self.COMPLETE: if action == self.ROLLBACK and self.state == (self.UPDATE, self.IN_PROGRESS): logger.debug(_("Starting update rollback for %s") % self.name) else: self.state_set(action, self.FAILED, "State invalid for %s" % action) return self.state_set(self.UPDATE, self.IN_PROGRESS, "Stack %s started" % action) oldstack = Stack(self.context, self.name, self.t, self.env) backup_stack = self._backup_stack() try: update_task = update.StackUpdate(self, newstack, backup_stack, rollback=action == self.ROLLBACK) updater = scheduler.TaskRunner(update_task) self.env = newstack.env self.parameters = newstack.parameters self.t.files = newstack.t.files self._set_param_stackid() try: updater.start(timeout=self.timeout_secs()) yield while not updater.step(): yield finally: self.reset_dependencies() if action == self.UPDATE: reason = "Stack successfully updated" else: reason = "Stack rollback completed" stack_status = self.COMPLETE except scheduler.Timeout: stack_status = self.FAILED reason = "Timed out" except exception.ResourceFailure as e: reason = str(e) stack_status = self.FAILED if action == self.UPDATE: # If rollback is enabled, we do another update, with the # existing template, so we roll back to the original state if not self.disable_rollback: yield self.update_task(oldstack, action=self.ROLLBACK) return else: logger.debug(_("Deleting backup stack")) backup_stack.delete() self.state_set(action, stack_status, reason) # flip the template to the newstack values # Note we do this on success and failure, so the current # stack resources are stored, even if one is in a failed # state (otherwise we won't remove them on delete) self.t = newstack.t template_outputs = self.t[template.OUTPUTS] self.outputs = self.resolve_static_data(template_outputs) self.store() def delete(self, action=DELETE): """ Delete all of the resources, and then the stack itself. The action parameter is used to differentiate between a user initiated delete and an automatic stack rollback after a failed create, which amount to the same thing, but the states are recorded differently. """ if action not in (self.DELETE, self.ROLLBACK): logger.error(_("Unexpected action %s passed to delete!") % action) self.state_set(self.DELETE, self.FAILED, "Invalid action %s" % action) return stack_status = self.COMPLETE reason = "Stack %s completed successfully" % action.lower() self.state_set(action, self.IN_PROGRESS, "Stack %s started" % action) backup_stack = self._backup_stack(False) if backup_stack is not None: backup_stack.delete() if backup_stack.status != backup_stack.COMPLETE: errs = backup_stack.status_reason failure = "Error deleting backup resources: %s" % errs self.state_set(action, self.FAILED, "Failed to %s : %s" % (action, failure)) return action_task = scheduler.DependencyTaskGroup(self.dependencies, resource.Resource.destroy, reverse=True) try: scheduler.TaskRunner(action_task)(timeout=self.timeout_secs()) except exception.ResourceFailure as ex: stack_status = self.FAILED reason = "Resource %s failed: %s" % (action.lower(), str(ex)) except scheduler.Timeout: stack_status = self.FAILED reason = "%s timed out" % action.title() if stack_status != self.FAILED: # If we created a trust, delete it stack = db_api.stack_get(self.context, self.id) user_creds = db_api.user_creds_get(stack.user_creds_id) trust_id = user_creds.get("trust_id") if trust_id: try: self.clients.keystone().delete_trust(trust_id) except Exception as ex: logger.exception(ex) stack_status = self.FAILED reason = "Error deleting trust: %s" % str(ex) self.state_set(action, stack_status, reason) if stack_status != self.FAILED: # delete the stack db_api.stack_delete(self.context, self.id) self.id = None def suspend(self): """ Suspend the stack, which invokes handle_suspend for all stack resources waits for all resources to become SUSPEND_COMPLETE then declares the stack SUSPEND_COMPLETE. Note the default implementation for all resources is to do nothing other than move to SUSPEND_COMPLETE, so the resources must implement handle_suspend for this to have any effect. """ sus_task = scheduler.TaskRunner(self.stack_task, action=self.SUSPEND, reverse=True) sus_task(timeout=self.timeout_secs()) def resume(self): """ Resume the stack, which invokes handle_resume for all stack resources waits for all resources to become RESUME_COMPLETE then declares the stack RESUME_COMPLETE. Note the default implementation for all resources is to do nothing other than move to RESUME_COMPLETE, so the resources must implement handle_resume for this to have any effect. """ sus_task = scheduler.TaskRunner(self.stack_task, action=self.RESUME, reverse=False) sus_task(timeout=self.timeout_secs()) def output(self, key): """ Get the value of the specified stack output. """ value = self.outputs[key].get("Value", "") return self.resolve_runtime_data(value) def restart_resource(self, resource_name): """ stop resource_name and all that depend on it start resource_name and all that depend on it """ deps = self.dependencies[self[resource_name]] failed = False for res in reversed(deps): try: scheduler.TaskRunner(res.destroy)() except exception.ResourceFailure as ex: failed = True logger.error(_("delete: %s") % str(ex)) for res in deps: if not failed: try: res.state_reset() scheduler.TaskRunner(res.create)() except exception.ResourceFailure as ex: logger.exception(_("create")) failed = True else: res.state_set(res.CREATE, res.FAILED, "Resource restart aborted") # TODO(asalkeld) if any of this fails we Should # restart the whole stack def get_availability_zones(self): if self._zones is None: self._zones = [zone.zoneName for zone in self.clients.nova().availability_zones.list(detailed=False)] return self._zones def set_deletion_policy(self, policy): for res in self.resources.values(): res.set_deletion_policy(policy) def get_abandon_data(self): return { "name": self.name, "id": self.id, "action": self.action, "status": self.status, "template": self.t.t, "resources": dict((res.name, res.get_abandon_data()) for res in self.resources.values()), } def resolve_static_data(self, snippet): return resolve_static_data(self.t, self, self.parameters, snippet) def resolve_runtime_data(self, snippet): return resolve_runtime_data(self.t, self.resources, snippet)