#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
import collections
import copy
import datetime
import functools
import itertools
import re
from oslo_config import cfg
from oslo_log import log as logging
from oslo_utils import encodeutils
from oslo_utils import excutils
from oslo_utils import timeutils as oslo_timeutils
from oslo_utils import uuidutils
from osprofiler import profiler
import six
from heat.common import context as common_context
from heat.common import environment_format as env_fmt
from heat.common import exception
from heat.common.i18n import _
from heat.common.i18n import _LE
from heat.common.i18n import _LI
from heat.common.i18n import _LW
from heat.common import identifier
from heat.common import lifecycle_plugin_utils
from heat.common import timeutils
from heat.engine import dependencies
from heat.engine import environment
from heat.engine import event
from heat.engine import function
from heat.engine.notification import stack as notification
from heat.engine import parameter_groups as param_groups
from heat.engine import resource
from heat.engine import resources
from heat.engine import scheduler
from heat.engine import sync_point
from heat.engine import template as tmpl
from heat.engine import update
from heat.objects import raw_template as raw_template_object
from heat.objects import resource as resource_objects
from heat.objects import snapshot as snapshot_object
from heat.objects import stack as stack_object
from heat.objects import stack_tag as stack_tag_object
from heat.objects import user_creds as ucreds_object
from heat.rpc import api as rpc_api
from heat.rpc import worker_client as rpc_worker_client
cfg.CONF.import_opt('error_wait_time', 'heat.common.config')
LOG = logging.getLogger(__name__)
[docs]class ForcedCancel(BaseException):
"""Exception raised to cancel task execution."""
def __init__(self, with_rollback=True):
self.with_rollback = with_rollback
def __str__(self):
return "Operation cancelled"
[docs]def reset_state_on_error(func):
@six.wraps(func)
def handle_exceptions(stack, *args, **kwargs):
errmsg = None
try:
return func(stack, *args, **kwargs)
except BaseException as exc:
with excutils.save_and_reraise_exception():
errmsg = six.text_type(exc)
LOG.error(_LE('Unexpected exception in %(func)s: %(msg)s'),
{'func': func.__name__, 'msg': errmsg})
finally:
if stack.status == stack.IN_PROGRESS:
msg = _("Unexpected returning while IN_PROGRESS.")
stack.state_set(stack.action, stack.FAILED,
errmsg if errmsg is not None else msg)
assert errmsg is not None, "Returned while IN_PROGRESS."
return handle_exceptions
@six.python_2_unicode_compatible
[docs]class Stack(collections.Mapping):
ACTIONS = (
CREATE, DELETE, UPDATE, ROLLBACK, SUSPEND, RESUME, ADOPT,
SNAPSHOT, CHECK, RESTORE
) = (
'CREATE', 'DELETE', 'UPDATE', 'ROLLBACK', 'SUSPEND', 'RESUME', 'ADOPT',
'SNAPSHOT', 'CHECK', 'RESTORE'
)
STATUSES = (IN_PROGRESS, FAILED, COMPLETE
) = ('IN_PROGRESS', 'FAILED', 'COMPLETE')
_zones = None
def __init__(self, context, stack_name, tmpl,
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,
use_stored_context=False, username=None,
nested_depth=0, strict_validate=True, convergence=False,
current_traversal=None, tags=None, prev_raw_template_id=None,
current_deps=None, cache_data=None, resource_validate=True,
service_check_defer=False):
"""Initialise the Stack.
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.
Creating a stack with cache_data creates a lightweight stack which
will not load any resources from the database and resolve the
functions from the cache_data specified.
"""
def _validate_stack_name(name):
try:
if not re.match("[a-zA-Z][a-zA-Z0-9_.-]{0,254}$", name):
message = _('Invalid stack name %s must contain '
'only alphanumeric or \"_-.\" characters, '
'must start with alpha and must be 255 '
'characters or less.') % name
raise exception.StackValidationFailed(message=message)
except TypeError:
message = _('Invalid stack name %s, must be a string') % name
raise exception.StackValidationFailed(message=message)
if owner_id is None:
_validate_stack_name(stack_name)
self.id = stack_id
self.owner_id = owner_id
self.context = context
self.t = tmpl
self.name = stack_name
self.action = (self.ADOPT if adopt_stack_data else
self.CREATE if action is None else action)
self.status = self.IN_PROGRESS if status is None else status
self.status_reason = status_reason
self.timeout_mins = timeout_mins
self.disable_rollback = disable_rollback
self.parent_resource_name = parent_resource
self._parent_stack = None
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
self.nested_depth = nested_depth
self.convergence = convergence
self.current_traversal = current_traversal
self.tags = tags
self.prev_raw_template_id = prev_raw_template_id
self.current_deps = current_deps
self.cache_data = cache_data
self._worker_client = None
self._convg_deps = None
self.thread_group_mgr = None
# strict_validate can be used to disable value validation
# in the resource properties schema, this is useful when
# performing validation when properties reference attributes
# for not-yet-created resources (which return None)
self.strict_validate = strict_validate
# resource_validate can be used to disable resource plugin subclass
# validate methods, which is useful when you want to validate
# template integrity but some parameters may not be provided
# at all, thus we can't yet reference property values such as is
# commonly done in plugin validate() methods
self.resource_validate = resource_validate
# service_check_defer can be used to defer the validation of service
# availability for a given resource, which helps to create the resource
# dependency tree completely when respective service is not available,
# especially during template_validate
self.service_check_defer = service_check_defer
if use_stored_context:
self.context = self.stored_context()
self.context.roles = self.context.clients.client(
'keystone').auth_ref.role_names
self.context.user_domain = self.context.clients.client(
'keystone').auth_ref.user_domain_id
self.context.project_domain = self.context.clients.client(
'keystone').auth_ref.project_domain_id
self.clients = self.context.clients
# 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
self.username = username or self.context.username
resources.initialise()
self.parameters = self.t.parameters(
self.identifier(),
user_params=self.env.params,
param_defaults=self.env.param_defaults)
self._set_param_stackid()
if resolve_data:
self.outputs = self.resolve_static_data(self.t[self.t.OUTPUTS])
else:
self.outputs = {}
@property
[docs] def worker_client(self):
"""Return a client for making engine RPC calls."""
if not self._worker_client:
self._worker_client = rpc_worker_client.WorkerClient()
return self._worker_client
@property
[docs] def env(self):
"""This is a helper to allow resources to access stack.env."""
return self.t.env
@property
[docs] def parent_resource(self):
"""Dynamically load up the parent_resource.
Note: this should only be used by "Fn::ResourceFacade"
"""
if self._parent_stack is None:
# we need both parent name and owner id.
if self.parent_resource_name is None or self.owner_id is None:
return None
try:
owner = self.load(self.context, stack_id=self.owner_id)
except exception.NotFound:
return None
self._parent_stack = owner
return self._parent_stack[self.parent_resource_name]
[docs] def set_parent_stack(self, parent_stack):
self._parent_stack = parent_stack
[docs] def stored_context(self):
if self.user_creds_id:
creds_obj = ucreds_object.UserCreds.get_by_id(self.user_creds_id)
# Maintain request_id from self.context so we retain traceability
# in situations where servicing a request requires switching from
# the request context to the stored context
creds = creds_obj.obj_to_primitive()["versioned_object.data"]
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
[docs] def resources(self):
return self._find_resources()
def _find_resources(self, filters=None):
if self._resources is None:
res_defns = self.t.resource_definitions(self)
if not filters:
self._resources = dict((name,
resource.Resource(name, data, self))
for (name, data) in res_defns.items())
else:
self._resources = dict()
self._db_resources = dict()
for rsc in six.itervalues(
resource_objects.Resource.get_all_by_stack(
self.context, self.id, True, filters)):
self._db_resources[rsc.name] = rsc
res = resource.Resource(rsc.name,
res_defns[rsc.name],
self)
self._resources[rsc.name] = res
# There is no need to continue storing the db resources
# after resource creation
self._db_resources = None
return self._resources
[docs] def iter_resources(self, nested_depth=0, filters=None):
"""Iterates over all the resources in a stack.
Iterating includes nested stacks up to `nested_depth` levels below.
"""
for res in six.itervalues(self._find_resources(filters)):
yield res
if not res.has_nested() or nested_depth == 0:
continue
nested_stack = res.nested()
for nested_res in nested_stack.iter_resources(nested_depth - 1):
yield nested_res
def _db_resources_get(self, key_id=False):
try:
return resource_objects.Resource.get_all_by_stack(
self.context, self.id, key_id)
except exception.NotFound:
return None
[docs] def db_resource_get(self, name):
if not self.id:
return None
if self._db_resources is None:
self._db_resources = self._db_resources_get()
return self._db_resources.get(name) if self._db_resources else None
@property
[docs] def dependencies(self):
if self._dependencies is None:
self._dependencies = self._get_dependencies(
six.itervalues(self.resources),
ignore_errors=self.id is not None)
return self._dependencies
[docs] def reset_dependencies(self):
self._dependencies = None
[docs] def root_stack_id(self):
if not self.owner_id:
return self.id
return stack_object.Stack.get_root_id(self.context, self.owner_id)
[docs] def object_path_in_stack(self):
"""Return stack resources and stacks in path from the root stack.
If this is not nested return (None, self), else return stack resources
and stacks in path from the root stack and including this stack.
:returns: a list of (stack_resource, stack) tuples.
"""
if self.parent_resource and self.parent_resource.stack:
path = self.parent_resource.stack.object_path_in_stack()
path.extend([(self.parent_resource, self)])
return path
return [(None, self)]
[docs] def path_in_stack(self):
"""Return tuples of names in path from the root stack.
If this is not nested return (None, self.name), else return tuples of
names (stack_resource.name, stack.name) in path from the root stack and
including this stack.
:returns: a list of (string, string) tuples.
"""
opis = self.object_path_in_stack()
return [(stckres.name if stckres else None,
stck.name if stck else None) for stckres, stck in opis]
[docs] def total_resources(self, stack_id=None):
"""Return the total number of resources in a stack.
Includes nested stacks below.
"""
if not stack_id:
stack_id = self.id
return stack_object.Stack.count_total_resources(self.context, stack_id)
def _set_param_stackid(self):
"""Update self.parameters with the current ARN.
The ARN is then provided via the Parameters class as the StackId pseudo
parameter.
"""
if not self.parameters.set_stack_id(self.identifier()):
LOG.warning(_LW("Unable to set parameters StackId identifier"))
@staticmethod
[docs] def get_dep_attrs(resources, outputs, resource_name):
"""Return the attributes of the specified resource that are referenced.
Return an iterator over any attributes of the specified resource that
are referenced.
"""
attr_lists = itertools.chain((res.dep_attrs(resource_name)
for res in resources),
(function.dep_attrs(out.get('Value', ''),
resource_name)
for out in six.itervalues(outputs)))
return set(itertools.chain.from_iterable(attr_lists))
@staticmethod
def _get_dependencies(resources, ignore_errors=True):
"""Return the dependency graph for a list of resources."""
deps = dependencies.Dependencies()
for res in resources:
res.add_explicit_dependencies(deps)
try:
res.add_dependencies(deps)
except Exception as exc:
if not ignore_errors:
raise
else:
LOG.warning(_LW('Ignoring error adding implicit '
'dependencies for %(res)s: %(err)s') %
{'res': six.text_type(res),
'err': six.text_type(exc)})
return deps
@classmethod
[docs] def load(cls, context, stack_id=None, stack=None, show_deleted=True,
use_stored_context=False, force_reload=False, cache_data=None,
resolve_data=True):
"""Retrieve a Stack from the database."""
if stack is None:
stack = stack_object.Stack.get_by_id(
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)
if force_reload:
stack.refresh()
return cls._from_db(context, stack,
use_stored_context=use_stored_context,
cache_data=cache_data, resolve_data=resolve_data)
@classmethod
[docs] 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,
show_nested=False, show_hidden=False, tags=None,
tags_any=None, not_tags=None, not_tags_any=None):
stacks = stack_object.Stack.get_all(
context,
limit,
sort_keys,
marker,
sort_dir,
filters,
tenant_safe,
show_deleted,
show_nested,
show_hidden,
tags,
tags_any,
not_tags,
not_tags_any) or []
for stack in stacks:
try:
yield cls._from_db(context, stack, resolve_data=resolve_data)
except exception.NotFound:
# We're in a different transaction than the get_all, so a stack
# returned above can be deleted by the time we try to load it.
pass
@classmethod
def _from_db(cls, context, stack, resolve_data=True,
use_stored_context=False, cache_data=None):
template = tmpl.Template.load(
context, stack.raw_template_id, stack.raw_template)
tags = None
if stack.tags:
tags = [t.tag for t in stack.tags]
return cls(context, stack.name, template,
stack_id=stack.id,
action=stack.action, status=stack.status,
status_reason=stack.status_reason,
timeout_mins=stack.timeout,
resolve_data=resolve_data,
disable_rollback=stack.disable_rollback,
parent_resource=stack.parent_resource_name,
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,
use_stored_context=use_stored_context,
username=stack.username, convergence=stack.convergence,
current_traversal=stack.current_traversal, tags=tags,
prev_raw_template_id=stack.prev_raw_template_id,
current_deps=stack.current_deps, cache_data=cache_data)
[docs] def get_kwargs_for_cloning(self, keep_status=False, only_db=False):
"""Get common kwargs for calling Stack() for cloning.
The point of this method is to reduce the number of places that we
need to update when a kwarg to Stack.__init__() is modified. It
is otherwise easy to forget an option and cause some unexpected
error if this option is lost.
Note:
- This doesn't return the args(name, template) but only the kwargs.
- We often want to start 'fresh' so don't want to maintain the old
status, action and status_reason.
- We sometimes only want the DB attributes.
"""
stack = {
'owner_id': self.owner_id,
'username': self.username,
'disable_rollback': self.disable_rollback,
'stack_user_project_id': self.stack_user_project_id,
'user_creds_id': self.user_creds_id,
'nested_depth': self.nested_depth,
'convergence': self.convergence,
'current_traversal': self.current_traversal,
'prev_raw_template_id': self.prev_raw_template_id,
'current_deps': self.current_deps
}
if keep_status:
stack.update({
'action': self.action,
'status': self.status,
'status_reason': self.status_reason})
if only_db:
stack['parent_resource_name'] = self.parent_resource_name
stack['tenant'] = self.tenant_id
stack['timeout'] = self.timeout_mins
else:
stack['parent_resource'] = self.parent_resource_name
stack['tenant_id'] = self.tenant_id
stack['timeout_mins'] = self.timeout_mins
stack['strict_validate'] = self.strict_validate
return stack
@profiler.trace('Stack.store', hide_args=False)
[docs] def store(self, backup=False, exp_trvsl=None):
"""Store the stack in the database and return its ID.
If self.id is set, we update the existing stack.
"""
s = self.get_kwargs_for_cloning(keep_status=True, only_db=True)
s['name'] = self.name
s['backup'] = backup
s['updated_at'] = self.updated_time
if self.t.id is None:
stack_object.Stack.encrypt_hidden_parameters(self.t)
s['raw_template_id'] = self.t.store(self.context)
else:
s['raw_template_id'] = self.t.id
if self.id:
if exp_trvsl is None:
exp_trvsl = self.current_traversal
if self.convergence:
# do things differently for convergence
updated = stack_object.Stack.select_and_update(
self.context, self.id, s, exp_trvsl=exp_trvsl)
if not updated:
return None
else:
stack_object.Stack.update_by_id(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 = ucreds_object.UserCreds.create(trust_ctx)
else:
new_creds = ucreds_object.UserCreds.create(self.context)
s['user_creds_id'] = new_creds.id
self.user_creds_id = new_creds.id
if self.convergence:
# create a traversal ID
self.current_traversal = uuidutils.generate_uuid()
s['current_traversal'] = self.current_traversal
new_s = stack_object.Stack.create(self.context, s)
self.id = new_s.id
self.created_time = new_s.created_at
if self.tags:
stack_tag_object.StackTagList.set(self.context, self.id, self.tags)
self._set_param_stackid()
return self.id
def _backup_name(self):
return '%s*' % self.name
[docs] 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]
[docs] 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)
if resource.action == resource.INIT:
resource._store()
[docs] 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."""
text = 'Stack "%s" [%s]' % (self.name, self.id)
return six.text_type(text)
[docs] def resource_by_refid(self, refid):
"""Return the resource in this stack with the specified refid.
:returns: resource in this stack with the specified refid, or None if
not found.
"""
for r in six.itervalues(self):
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 or r.name == refid)):
return r
[docs] def register_access_allowed_handler(self, credential_id, handler):
"""Register an authorization handler function.
Register a function which determines whether the credentials with a
given ID can have access to a named resource.
"""
assert callable(handler), 'Handler is not callable'
self._access_allowed_handlers[credential_id] = handler
[docs] def access_allowed(self, credential_id, resource_name):
"""Is credential_id authorised to access resource by 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)
@profiler.trace('Stack.validate', hide_args=False)
[docs] def validate(self, ignorable_errors=None, validate_by_deps=True):
"""Validates the stack."""
# TODO(sdake) Should return line number of invalid reference
# validate overall template (top-level structure)
self.t.validate()
# Validate parameters
self.parameters.validate(context=self.context,
validate_value=self.strict_validate)
# Validate Parameter Groups
parameter_groups = param_groups.ParameterGroups(self.t)
parameter_groups.validate()
# Validate types of sections in ResourceDefinitions
self.t.validate_resource_definitions(self)
# Check duplicate names between parameters and resources
dup_names = (set(six.iterkeys(self.parameters)) &
set(six.iterkeys(self)))
if dup_names:
LOG.debug("Duplicate names %s" % dup_names)
raise exception.StackValidationFailed(
message=_("Duplicate names %s") % dup_names)
if validate_by_deps:
iter_rsc = self.dependencies
else:
iter_rsc = six.itervalues(self.resources)
for res in iter_rsc:
try:
if self.resource_validate:
result = res.validate()
else:
result = res.validate_template()
except exception.HeatException as ex:
LOG.debug('%s', ex)
if ignorable_errors and ex.error_code in ignorable_errors:
result = None
else:
raise ex
except AssertionError:
raise
except Exception as ex:
LOG.exception(_LE("Exception: %s"), ex)
raise exception.StackValidationFailed(
message=encodeutils.safe_decode(six.text_type(ex)))
if result:
raise exception.StackValidationFailed(message=result)
for key, val in self.outputs.items():
if not isinstance(val, collections.Mapping):
message = _('Outputs must contain Output. '
'Found a [%s] instead') % type(val)
raise exception.StackValidationFailed(
error='Output validation error',
path=[self.t.OUTPUTS],
message=message)
try:
if not val or 'Value' not in val:
message = _('Each Output must contain '
'a Value key.')
raise exception.StackValidationFailed(
error='Output validation error',
path=[self.t.OUTPUTS, key],
message=message)
function.validate(val.get('Value'))
except exception.StackValidationFailed as ex:
raise
except AssertionError:
raise
except Exception as ex:
raise exception.StackValidationFailed(
error='Output validation error',
path=[self.t.OUTPUTS, key,
self.t.get_section_name('Value')],
message=six.text_type(ex))
[docs] def requires_deferred_auth(self):
"""Determine whether to perform API requests with deferred auth.
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 six.itervalues(self))
def _add_event(self, action, status, reason):
"""Add a state change event to the database."""
ev = event.Event(self.context, self, action, status, reason,
self.id, {},
self.name, 'OS::Heat::Stack')
ev.store()
self.dispatch_event(ev)
[docs] def dispatch_event(self, ev):
def _dispatch(ctx, sinks, ev):
try:
for sink in sinks:
sink.consume(ctx, ev)
except Exception as e:
LOG.debug('Got error sending events %s' % e)
if self.thread_group_mgr is not None:
self.thread_group_mgr.start(self.id, _dispatch,
self.context,
self.env.get_event_sinks(),
ev.as_dict())
@profiler.trace('Stack.state_set', hide_args=False)
[docs] def state_set(self, action, status, reason):
"""Update the stack state."""
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.convergence and action in (self.UPDATE, self.DELETE,
self.CREATE, self.ADOPT):
# if convergence and stack operation is create/update/delete,
# stack lock is not used, hence persist state
updated = self._persist_state()
if not updated:
# Possibly failed concurrent update
LOG.warning(_LW("Failed to set state of stack %(name)s with"
" traversal ID %(trvsl_id)s, to"
" %(action)s_%(status)s"),
{'name': self.name,
'trvsl_id': self.current_traversal,
'action': action, 'status': status})
return updated
# Persist state to db only if status == IN_PROGRESS
# or action == UPDATE/DELETE/ROLLBACK. Else, it would
# be done before releasing the stack lock.
if status == self.IN_PROGRESS or action in (
self.UPDATE, self.DELETE, self.ROLLBACK):
self._persist_state()
def _persist_state(self):
"""Persist stack state to database"""
if self.id is None:
return
stack = stack_object.Stack.get_by_id(self.context, self.id)
if stack is not None:
values = {'action': self.action,
'status': self.status,
'status_reason': self.status_reason}
self._send_notification_and_add_event()
if self.convergence:
# do things differently for convergence
updated = stack_object.Stack.select_and_update(
self.context, self.id, values,
exp_trvsl=self.current_traversal)
return updated
else:
stack.update_and_save(values)
def _send_notification_and_add_event(self):
notification.send(self)
self._add_event(self.action, self.status, self.status_reason)
LOG.info(_LI('Stack %(action)s %(status)s (%(name)s): '
'%(reason)s'),
{'action': self.action,
'status': self.status,
'name': self.name,
'reason': self.status_reason})
[docs] def persist_state_and_release_lock(self, engine_id):
"""Persist stack state to database and release stack lock"""
if self.id is None:
return
stack = stack_object.Stack.get_by_id(self.context, self.id)
if stack is not None:
values = {'action': self.action,
'status': self.status,
'status_reason': self.status_reason}
self._send_notification_and_add_event()
stack.persist_state_and_release_lock(self.context, self.id,
engine_id, values)
@property
[docs] def state(self):
"""Returns state, tuple of action, status."""
return (self.action, self.status)
[docs] 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
[docs] def preview_resources(self):
"""Preview the stack with all of the resources."""
return [resource.preview()
for resource in six.itervalues(self.resources)]
def _store_resources(self):
for r in reversed(self.dependencies):
if r.action == r.INIT:
r._store()
@profiler.trace('Stack.create', hide_args=False)
@reset_state_on_error
[docs] 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)
self._store_resources()
creator = scheduler.TaskRunner(
self.stack_task, action=self.CREATE,
reverse=False, post_func=rollback,
error_wait_time=cfg.CONF.error_wait_time)
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
[docs] def stack_task(self, action, reverse=False, post_func=None,
error_wait_time=None,
aggregate_exceptions=False, pre_completion_func=None):
"""A task to perform an action on the stack.
All of the resources are traversed in forward or reverse dependency
order.
:param action: action that should be executed with stack resources
:param reverse: define if action on the resources need to be executed
in reverse order (resources - first and then res dependencies )
:param post_func: function that need to be executed after
action complete on the stack
:param error_wait_time: time to wait before cancelling all execution
threads when an error occurred
:param aggregate_exceptions: define if exceptions should be aggregated
:param pre_completion_func: function that need to be executed right
before action completion. Uses stack ,action, status and reason as
input parameters
"""
try:
lifecycle_plugin_utils.do_pre_ops(self.context, self,
None, action)
except Exception as e:
self.state_set(action, self.FAILED, e.args[0] if e.args else
'Failed stack pre-ops: %s' % six.text_type(e))
if callable(post_func):
post_func()
return
self.state_set(action, self.IN_PROGRESS,
'Stack %s started' % action)
stack_status = self.COMPLETE
reason = 'Stack %s completed successfully' % action
action_method = action.lower()
# 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_method,
lambda x: {})
@functools.wraps(getattr(resource.Resource, action_method))
def resource_action(r):
# Find e.g resource.create and call it
handle = getattr(r, action_method)
return handle(**handle_kwargs(r))
action_task = scheduler.DependencyTaskGroup(
self.dependencies,
resource_action,
reverse,
error_wait_time=error_wait_time,
aggregate_exceptions=aggregate_exceptions)
try:
yield action_task()
except scheduler.Timeout:
stack_status = self.FAILED
reason = '%s timed out' % action.title()
except Exception as ex:
# We use a catch-all here to ensure any raised exceptions
# make the stack fail. This is necessary for when
# aggregate_exceptions is false, as in that case we don't get
# ExceptionGroup, but the raw exception.
# see scheduler.py line 395-399
stack_status = self.FAILED
reason = 'Resource %s failed: %s' % (action, six.text_type(ex))
if pre_completion_func:
pre_completion_func(self, action, stack_status, reason)
self.state_set(action, stack_status, reason)
if callable(post_func):
post_func()
lifecycle_plugin_utils.do_post_ops(self.context, self, None, action,
(self.status == self.FAILED))
@profiler.trace('Stack.check', hide_args=False)
@reset_state_on_error
[docs] def check(self):
self.updated_time = oslo_timeutils.utcnow()
checker = scheduler.TaskRunner(
self.stack_task, self.CHECK,
post_func=self.supports_check_action,
error_wait_time=cfg.CONF.error_wait_time,
aggregate_exceptions=True)
checker()
[docs] def supports_check_action(self):
def is_supported(stack, res):
if res.has_nested():
return res.nested().supports_check_action()
else:
return hasattr(res, 'handle_%s' % self.CHECK.lower())
supported = [is_supported(self, res)
for res in six.itervalues(self.resources)]
if not all(supported):
msg = ". '%s' not fully supported (see resources)" % self.CHECK
reason = self.status_reason + msg
self.state_set(self.CHECK, self.status, reason)
return all(supported)
@profiler.trace('Stack._backup_stack', hide_args=False)
def _backup_stack(self, create_if_missing=True):
"""Backup the stack.
Get a Stack containing any in-progress resources from the previous
stack state prior to an update.
"""
s = stack_object.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:
kwargs = self.get_kwargs_for_cloning()
kwargs['owner_id'] = self.id
del(kwargs['prev_raw_template_id'])
prev = type(self)(self.context, self._backup_name(),
copy.deepcopy(self.t),
**kwargs)
prev.store(backup=True)
LOG.debug('Created new backup stack')
return prev
else:
return None
@profiler.trace('Stack.adopt', hide_args=False)
@reset_state_on_error
[docs] def adopt(self):
"""Adopt existing resources into a new stack."""
def rollback():
if not self.disable_rollback and self.state == (self.ADOPT,
self.FAILED):
# enter the same flow as abandon and just delete the stack
for res in six.itervalues(self.resources):
res.abandon_in_progress = True
self.delete(action=self.ROLLBACK, abandon=True)
creator = scheduler.TaskRunner(
self.stack_task,
action=self.ADOPT,
reverse=False,
error_wait_time=cfg.CONF.error_wait_time,
post_func=rollback)
creator(timeout=self.timeout_secs())
@profiler.trace('Stack.update', hide_args=False)
@reset_state_on_error
[docs] def update(self, newstack, event=None):
"""Update the stack.
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 = oslo_timeutils.utcnow()
updater = scheduler.TaskRunner(self.update_task, newstack,
event=event)
updater()
@profiler.trace('Stack.converge_stack', hide_args=False)
[docs] def converge_stack(self, template, action=UPDATE, new_stack=None):
"""Update the stack template and trigger convergence for resources."""
if action not in [self.CREATE, self.ADOPT]:
# no back-up template for create action
self.prev_raw_template_id = getattr(self.t, 'id', None)
# switch template and reset dependencies
self.t = template
self.reset_dependencies()
self._resources = None
if action is not self.CREATE:
self.updated_time = oslo_timeutils.utcnow()
if new_stack is not None:
self.disable_rollback = new_stack.disable_rollback
self.timeout_mins = new_stack.timeout_mins
self._set_param_stackid()
self.tags = new_stack.tags
if new_stack.tags:
stack_tag_object.StackTagList.set(self.context, self.id,
new_stack.tags)
else:
stack_tag_object.StackTagList.delete(self.context, self.id)
self.action = action
self.status = self.IN_PROGRESS
self.status_reason = 'Stack %s started' % self.action
# generate new traversal and store
previous_traversal = self.current_traversal
self.current_traversal = uuidutils.generate_uuid()
# we expect to update the stack having previous traversal ID
stack_id = self.store(exp_trvsl=previous_traversal)
if stack_id is None:
LOG.warning(_LW("Failed to store stack %(name)s with traversal "
"ID %(trvsl_id)s, aborting stack %(action)s"),
{'name': self.name, 'trvsl_id': previous_traversal,
'action': self.action})
return
self._send_notification_and_add_event()
# delete the prev traversal sync_points
if previous_traversal:
sync_point.delete_all(self.context, self.id, previous_traversal)
# TODO(later): lifecycle_plugin_utils.do_pre_ops
self._converge_create_or_update()
def _converge_create_or_update(self):
current_resources = self._update_or_store_resources()
self._compute_convg_dependencies(self.ext_rsrcs_db, self.dependencies,
current_resources)
# Store list of edges
self.current_deps = {
'edges': [[rqr, rqd] for rqr, rqd in
self.convergence_dependencies.graph().edges()]}
stack_id = self.store()
if stack_id is None:
# Failed concurrent update
LOG.warning(_LW("Failed to store stack %(name)s with traversal "
"ID %(trvsl_id)s, aborting stack %(action)s"),
{'name': self.name, 'trvsl_id': self.current_traversal,
'action': self.action})
return
LOG.info(_LI('convergence_dependencies: %s'),
self.convergence_dependencies)
# create sync_points for resources in DB
for rsrc_id, is_update in self.convergence_dependencies:
sync_point.create(self.context, rsrc_id,
self.current_traversal, is_update,
self.id)
# create sync_point entry for stack
sync_point.create(
self.context, self.id, self.current_traversal, True, self.id)
leaves = set(self.convergence_dependencies.leaves())
if not any(leaves):
self.mark_complete()
else:
for rsrc_id, is_update in self.convergence_dependencies.leaves():
if is_update:
LOG.info(_LI("Triggering resource %s for update"), rsrc_id)
else:
LOG.info(_LI("Triggering resource %s for cleanup"),
rsrc_id)
input_data = sync_point.serialize_input_data({})
self.worker_client.check_resource(self.context, rsrc_id,
self.current_traversal,
input_data, is_update,
self.adopt_stack_data)
[docs] def rollback(self):
old_tmpl_id = self.prev_raw_template_id
if old_tmpl_id is None:
rollback_tmpl = tmpl.Template.create_empty_template(
version=self.t.version)
else:
rollback_tmpl = tmpl.Template.load(self.context, old_tmpl_id)
self.prev_raw_template_id = None
stack_id = self.store()
if stack_id is None:
# Failed concurrent update
LOG.warning(_LW("Failed to store stack %(name)s with traversal"
" ID %(trvsl_id)s, not trigerring rollback."),
{'name': self.name,
'trvsl_id': self.current_traversal})
return
self.converge_stack(rollback_tmpl, action=self.ROLLBACK)
def _get_best_existing_rsrc_db(self, rsrc_name):
candidate = None
if self.ext_rsrcs_db:
for id, ext_rsrc in self.ext_rsrcs_db.items():
if ext_rsrc.name != rsrc_name:
continue
if ext_rsrc.current_template_id == self.t.id:
# Rollback where the previous resource still exists
candidate = ext_rsrc
break
elif (ext_rsrc.current_template_id ==
self.prev_raw_template_id):
# Current resource is otherwise a good candidate
candidate = ext_rsrc
elif candidate is None:
# In multiple concurrent updates, if candidate is not
# found in current/previous template, it could be found
# in old tmpl.
candidate = ext_rsrc
return candidate
def _update_or_store_resources(self):
self.ext_rsrcs_db = self._db_resources_get(key_id=True)
curr_name_translated_dep = self.dependencies.translate(lambda res:
res.name)
rsrcs = {}
def update_needed_by(res):
new_requirers = set(
rsrcs[rsrc_name].id for rsrc_name in
curr_name_translated_dep.required_by(res.name)
)
old_requirers = set(res.needed_by) if res.needed_by else set()
needed_by = old_requirers | new_requirers
res.needed_by = list(needed_by)
for rsrc in reversed(self.dependencies):
existing_rsrc_db = self._get_best_existing_rsrc_db(rsrc.name)
if existing_rsrc_db is None:
update_needed_by(rsrc)
rsrc.current_template_id = self.t.id
rsrc._store()
rsrcs[rsrc.name] = rsrc
else:
update_needed_by(existing_rsrc_db)
resource.Resource.set_needed_by(
existing_rsrc_db, existing_rsrc_db.needed_by
)
rsrcs[existing_rsrc_db.name] = existing_rsrc_db
return rsrcs
def _compute_convg_dependencies(self, existing_resources,
current_template_deps, current_resources):
def make_graph_key(rsrc):
return current_resources[rsrc.name].id, True
dep = current_template_deps.translate(make_graph_key)
if existing_resources:
for rsrc_id, rsrc in existing_resources.items():
dep += (rsrc_id, False), None
for requirement in rsrc.requires:
if requirement in existing_resources:
dep += (requirement, False), (rsrc_id, False)
if rsrc.replaces in existing_resources:
dep += (rsrc.replaces, False), (rsrc_id, False)
if (rsrc.id, True) in dep:
dep += (rsrc_id, False), (rsrc_id, True)
self._convg_deps = dep
@property
[docs] def convergence_dependencies(self):
if self._convg_deps is None:
current_deps = ([tuple(i), (tuple(j) if j is not None else None)]
for i, j in self.current_deps['edges'])
self._convg_deps = dependencies.Dependencies(edges=current_deps)
return self._convg_deps
@scheduler.wrappertask
[docs] def update_task(self, newstack, action=UPDATE, event=None):
if action not in (self.UPDATE, self.ROLLBACK, self.RESTORE):
LOG.error(_LE("Unexpected action %s passed to update!"), action)
self.state_set(self.UPDATE, self.FAILED,
"Invalid action %s" % action)
return
try:
lifecycle_plugin_utils.do_pre_ops(self.context, self,
newstack, action)
except Exception as e:
self.state_set(action, self.FAILED, e.args[0] if e.args else
'Failed stack pre-ops: %s' % six.text_type(e))
return
if self.status == self.IN_PROGRESS:
if action == self.ROLLBACK:
LOG.debug("Starting update rollback for %s" % self.name)
else:
self.state_set(action, self.FAILED,
'State invalid for %s' % action)
return
# Save a copy of the new template. To avoid two DB writes
# we store the ID at the same time as the action/status
prev_tmpl_id = self.prev_raw_template_id
bu_tmpl = copy.deepcopy(newstack.t)
self.prev_raw_template_id = bu_tmpl.store()
self.action = action
self.status = self.IN_PROGRESS
self.status_reason = 'Stack %s started' % action
self._send_notification_and_add_event()
self.store()
if prev_tmpl_id is not None:
raw_template_object.RawTemplate.delete(self.context, prev_tmpl_id)
if action == self.UPDATE:
# Oldstack is useless when the action is not UPDATE , so we don't
# need to build it, this can avoid some unexpected errors.
kwargs = self.get_kwargs_for_cloning()
oldstack = Stack(self.context, self.name, copy.deepcopy(self.t),
**kwargs)
backup_stack = self._backup_stack()
existing_params = environment.Environment({env_fmt.PARAMETERS:
self.t.env.params})
previous_template_id = None
should_rollback = False
update_task = update.StackUpdate(
self, newstack, backup_stack,
rollback=action == self.ROLLBACK,
error_wait_time=cfg.CONF.error_wait_time)
try:
updater = scheduler.TaskRunner(update_task)
self.parameters = newstack.parameters
self.t.files = newstack.t.files
self.t.env = newstack.t.env
self.disable_rollback = newstack.disable_rollback
self.timeout_mins = newstack.timeout_mins
self._set_param_stackid()
self.tags = newstack.tags
if newstack.tags:
stack_tag_object.StackTagList.set(self.context, self.id,
newstack.tags)
else:
stack_tag_object.StackTagList.delete(self.context, self.id)
try:
updater.start(timeout=self.timeout_secs())
yield
while not updater.step():
if event is None or not event.ready():
yield
else:
message = event.wait()
self._message_parser(message)
finally:
self.reset_dependencies()
if action in (self.UPDATE, self.RESTORE, self.ROLLBACK):
self.status_reason = 'Stack %s completed successfully' % action
self.status = self.COMPLETE
except scheduler.Timeout:
self.status = self.FAILED
self.status_reason = 'Timed out'
except (ForcedCancel, Exception) as e:
# If rollback is enabled when resource failure occurred,
# we do another update, with the existing template,
# so we roll back to the original state
should_rollback = self._update_exception_handler(e, action,
update_task)
if should_rollback:
yield self.update_task(oldstack, action=self.ROLLBACK)
except BaseException as e:
with excutils.save_and_reraise_exception():
self._update_exception_handler(e, action, update_task)
else:
LOG.debug('Deleting backup stack')
backup_stack.delete(backup=True)
# flip the template to the newstack values
previous_template_id = self.t.id
self.t = newstack.t
template_outputs = self.t[self.t.OUTPUTS]
self.outputs = self.resolve_static_data(template_outputs)
finally:
if should_rollback:
# Already handled in rollback task
return
# Don't use state_set to do only one update query and avoid race
# condition with the COMPLETE status
self.action = action
self._send_notification_and_add_event()
if self.status == self.FAILED:
# Since template was incrementally updated based on existing
# and new stack resources, we should have user params of both.
existing_params.load(newstack.t.env.user_env_as_dict())
self.t.env = existing_params
self.t.store(self.context)
backup_stack.t.env = existing_params
backup_stack.t.store(self.context)
self.store()
if previous_template_id is not None:
raw_template_object.RawTemplate.delete(self.context,
previous_template_id)
lifecycle_plugin_utils.do_post_ops(self.context, self,
newstack, action,
(self.status == self.FAILED))
def _update_exception_handler(self, exc, action, update_task):
"""Handle exceptions in update_task.
Decide if we should cancel tasks or not. Also decide if we should
rollback or not, depend on disable rollback flag if force rollback flag
not triggered.
:returns: a boolean for require rollback flag.
"""
self.status_reason = six.text_type(exc)
self.status = self.FAILED
if action != self.UPDATE:
return False
if isinstance(exc, ForcedCancel):
update_task.updater.cancel_all()
return exc.with_rollback or not self.disable_rollback
elif isinstance(exc, exception.ResourceFailure):
return not self.disable_rollback
else:
return False
def _message_parser(self, message):
if message == rpc_api.THREAD_CANCEL:
raise ForcedCancel(with_rollback=False)
elif message == rpc_api.THREAD_CANCEL_WITH_ROLLBACK:
raise ForcedCancel(with_rollback=True)
def _delete_backup_stack(self, stack):
# Delete resources in the backup stack referred to by 'stack'
def failed(child):
return (child.action == child.CREATE and
child.status in (child.FAILED, child.IN_PROGRESS))
def copy_data(source_res, destination_res):
if source_res.data():
for key, val in six.iteritems(source_res.data()):
destination_res.data_set(key, val)
for key, backup_res in stack.resources.items():
# If UpdateReplace is failed, we must restore backup_res
# to existing_stack in case of it may have dependencies in
# these stacks. curr_res is the resource that just
# created and failed, so put into the stack to delete anyway.
backup_res_id = backup_res.resource_id
curr_res = self.resources.get(key)
if backup_res_id is not None and curr_res is not None:
curr_res_id = curr_res.resource_id
if (any(failed(child) for child in
self.dependencies[curr_res]) or
curr_res.status in
(curr_res.FAILED, curr_res.IN_PROGRESS)):
# If child resource failed to update, curr_res
# 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.
# 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_res_id
self.resources[key].properties = backup_res.properties
copy_data(backup_res, self.resources[key])
stack.resources[key].resource_id = curr_res_id
stack.resources[key].properties = curr_res.properties
copy_data(curr_res, stack.resources[key])
stack.delete(backup=True)
def _try_get_user_creds(self):
# There are cases where the user_creds cannot be returned
# due to credentials truncated when being saved to DB.
# Ignore this error instead of blocking stack deletion.
user_creds = None
try:
user_creds = ucreds_object.UserCreds.get_by_id(self.user_creds_id)
except exception.Error as err:
LOG.exception(err)
pass
return user_creds
def _delete_credentials(self, stack_status, reason, abandon):
# Cleanup stored user_creds so they aren't accessible via
# the soft-deleted stack which remains in the DB
# The stack_status and reason passed in are current values, which
# may get rewritten and returned from this method
if self.user_creds_id:
user_creds = self._try_get_user_creds()
# If we created a trust, delete it
if user_creds is not None:
trust_id = user_creds.get('trust_id')
if trust_id:
try:
# If the trustor doesn't match the context user the
# we have to use the stored context to cleanup the
# trust, as although the user evidently has
# permission to delete the stack, they don't have
# rights to delete the trust unless an admin
trustor_id = user_creds.get('trustor_user_id')
if self.context.user_id != trustor_id:
LOG.debug("Context user_id doesn't match "
"trustor, using stored context")
sc = self.stored_context()
sc.clients.client('keystone').delete_trust(
trust_id)
else:
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:
ucreds_object.UserCreds.delete(self.context,
self.user_creds_id)
except exception.NotFound:
LOG.info(_LI("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(_LI("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 and not abandon:
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)
return stack_status, reason
@profiler.trace('Stack.delete', hide_args=False)
@reset_state_on_error
[docs] def delete(self, action=DELETE, backup=False, abandon=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.
Note abandon is a delete where all resources have been set to a
RETAIN deletion policy, but we also don't want to delete anything
required for those resources, e.g the stack_user_project.
"""
if action not in (self.DELETE, self.ROLLBACK):
LOG.error(_LE("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:
self._delete_backup_stack(backup_stack)
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
snapshots = snapshot_object.Snapshot.get_all(self.context,
self.id)
for snapshot in snapshots:
self.delete_snapshot(snapshot)
snapshot_object.Snapshot.delete(self.context, snapshot.id)
if not backup:
try:
lifecycle_plugin_utils.do_pre_ops(self.context, self,
None, action)
except Exception as e:
self.state_set(action, self.FAILED,
e.args[0] if e.args else
'Failed stack pre-ops: %s' % six.text_type(e))
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 the stack delete succeeded, this is not a backup stack and it's
# not a nested stack, we should delete the credentials
if stack_status != self.FAILED and not backup and not self.owner_id:
stack_status, reason = self._delete_credentials(stack_status,
reason,
abandon)
try:
self.state_set(action, stack_status, reason)
except exception.NotFound:
LOG.info(_LI("Tried to delete stack that does not exist "
"%s "), self.id)
if not backup:
lifecycle_plugin_utils.do_post_ops(self.context, self,
None, action,
(self.status == self.FAILED))
if stack_status != self.FAILED:
# delete the stack
try:
stack_object.Stack.delete(self.context, self.id)
except exception.NotFound:
LOG.info(_LI("Tried to delete stack that does not exist "
"%s "), self.id)
self.id = None
@profiler.trace('Stack.suspend', hide_args=False)
@reset_state_on_error
[docs] def suspend(self):
"""Suspend the stack.
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(_LI('%s is already suspended'), six.text_type(self))
return
self.updated_time = oslo_timeutils.utcnow()
sus_task = scheduler.TaskRunner(
self.stack_task,
action=self.SUSPEND,
reverse=True,
error_wait_time=cfg.CONF.error_wait_time)
sus_task(timeout=self.timeout_secs())
@profiler.trace('Stack.resume', hide_args=False)
@reset_state_on_error
[docs] def resume(self):
"""Resume the stack.
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(_LI('%s is already resumed'), six.text_type(self))
return
self.updated_time = oslo_timeutils.utcnow()
sus_task = scheduler.TaskRunner(
self.stack_task,
action=self.RESUME,
reverse=False,
error_wait_time=cfg.CONF.error_wait_time)
sus_task(timeout=self.timeout_secs())
@profiler.trace('Stack.snapshot', hide_args=False)
@reset_state_on_error
[docs] def snapshot(self, save_snapshot_func):
"""Snapshot the stack, invoking handle_snapshot on all resources."""
self.updated_time = oslo_timeutils.utcnow()
sus_task = scheduler.TaskRunner(
self.stack_task,
action=self.SNAPSHOT,
reverse=False,
error_wait_time=cfg.CONF.error_wait_time,
pre_completion_func=save_snapshot_func)
sus_task(timeout=self.timeout_secs())
@profiler.trace('Stack.delete_snapshot', hide_args=False)
[docs] def delete_snapshot(self, snapshot):
"""Remove a snapshot from the backends."""
for name, rsrc in six.iteritems(self.resources):
snapshot_data = snapshot.data
if snapshot_data:
data = snapshot.data['resources'].get(name)
scheduler.TaskRunner(rsrc.delete_snapshot, data)()
@profiler.trace('Stack.restore', hide_args=False)
@reset_state_on_error
[docs] def restore(self, snapshot):
"""Restore the given snapshot.
Invokes handle_restore on all resources.
"""
self.updated_time = oslo_timeutils.utcnow()
env = environment.Environment(snapshot.data['environment'])
files = snapshot.data['files']
template = tmpl.Template(snapshot.data['template'],
env=env, files=files)
newstack = self.__class__(self.context, self.name, template,
timeout_mins=self.timeout_mins,
disable_rollback=self.disable_rollback)
for name, defn in six.iteritems(
template.resource_definitions(newstack)):
rsrc = resource.Resource(name, defn, self)
data = snapshot.data['resources'].get(name)
handle_restore = getattr(rsrc, 'handle_restore', None)
if callable(handle_restore):
defn = handle_restore(defn, data)
template.add_resource(defn, name)
newstack.parameters.set_stack_id(self.identifier())
updater = scheduler.TaskRunner(self.update_task, newstack,
action=self.RESTORE)
updater()
@profiler.trace('Stack.output', hide_args=False)
[docs] 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
[docs] def restart_resource(self, resource_name):
"""Restart the resource specified by 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(_LE('Resource %(name)s delete failed: %(ex)s'),
{'name': res.name, 'ex': ex})
for res in deps:
if not failed:
try:
res.state_reset()
scheduler.TaskRunner(res.create)()
except exception.ResourceFailure as ex:
LOG.exception(_LE('Resource %(name)s create failed: '
'%(ex)s'), {'name': res.name, 'ex': ex})
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
[docs] 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
[docs] def set_stack_user_project_id(self, project_id):
self.stack_user_project_id = project_id
self.store()
@profiler.trace('Stack.create_stack_user_project_id', hide_args=False)
[docs] def create_stack_user_project_id(self):
project_id = self.clients.client(
'keystone').create_stack_domain_project(self.id)
self.set_stack_user_project_id(project_id)
@profiler.trace('Stack.prepare_abandon', hide_args=False)
[docs] def prepare_abandon(self):
return {
'name': self.name,
'id': self.id,
'action': self.action,
'environment': self.env.user_env_as_dict(),
'files': self.t.files,
'status': self.status,
'template': self.t.t,
'resources': dict((res.name, res.prepare_abandon())
for res in six.itervalues(self.resources)),
'project_id': self.tenant_id,
'stack_user_project_id': self.stack_user_project_id,
'tags': self.tags,
}
[docs] def resolve_static_data(self, snippet):
try:
return self.t.parse(self, snippet)
except AssertionError:
raise
except Exception as ex:
raise exception.StackValidationFailed(
message=encodeutils.safe_decode(six.text_type(ex)))
[docs] 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 six.itervalues(self.resources):
res.attributes.reset_resolved_values()
[docs] def has_cache_data(self, resource_name):
return (self.cache_data is not None and
self.cache_data.get(resource_name) is not None)
[docs] def cache_data_reference_id(self, resource_name):
return self.cache_data.get(
resource_name, {}).get('reference_id')
[docs] def cache_data_resource_attribute(self, resource_name, attribute_key):
return self.cache_data.get(
resource_name, {}).get('attrs', {}).get(attribute_key)
[docs] def cache_data_resource_all_attributes(self, resource_name):
attrs = self.cache_data.get(resource_name, {}).get('attributes', {})
return attrs
[docs] def mark_complete(self):
"""Mark the update as complete.
This currently occurs when all resources have been updated; there may
still be resources being cleaned up, but the Stack should now be in
service.
"""
LOG.info(_LI('[%(name)s(%(id)s)] update traversal %(tid)s complete'),
{'name': self.name, 'id': self.id,
'tid': self.current_traversal})
reason = 'Stack %s completed successfully' % self.action
updated = self.state_set(self.action, self.COMPLETE, reason)
if not updated:
return
self.purge_db()
[docs] def purge_db(self):
"""Cleanup database after stack has completed/failed.
1. Delete previous raw template if stack completes successfully.
2. Deletes all sync points. They are no longer needed after stack
has completed/failed.
3. Delete the stack if the action is DELETE.
"""
if (self.prev_raw_template_id is not None and
self.status != self.FAILED):
prev_tmpl_id = self.prev_raw_template_id
self.prev_raw_template_id = None
stack_id = self.store()
if stack_id is None:
# Failed concurrent update
LOG.warning(_LW("Failed to store stack %(name)s with traversal"
" ID %(trvsl_id)s, aborting stack purge"),
{'name': self.name,
'trvsl_id': self.current_traversal})
return
raw_template_object.RawTemplate.delete(self.context, prev_tmpl_id)
sync_point.delete_all(self.context, self.id, self.current_traversal)
if (self.action, self.status) == (self.DELETE, self.COMPLETE):
if not self.owner_id:
status, reason = self._delete_credentials(
self.status,
self.status_reason,
False)
if status == self.FAILED:
# something wrong when delete credentials, set FAILED
self.state_set(self.action, status, reason)
return
try:
stack_object.Stack.delete(self.context, self.id)
except exception.NotFound:
pass
[docs] def time_elapsed(self):
"""Time elapsed in seconds since the stack operation started."""
start_time = timeutils.round_to_seconds(self.updated_time or
self.created_time)
nowish = timeutils.round_to_seconds(datetime.datetime.utcnow())
return (nowish - start_time).seconds
[docs] def time_remaining(self):
"""Time left before stack times out."""
return self.timeout_secs() - self.time_elapsed()
[docs] def has_timed_out(self):
"""Returns True if this stack has timed-out."""
if self.status == self.IN_PROGRESS:
return self.time_elapsed() > self.timeout_secs()
return False