heat/heat/engine/scheduler.py

472 lines
15 KiB
Python

#
# 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 functools
import itertools
import sys
from time import time as wallclock
import types
import eventlet
from heat.openstack.common import excutils
from heat.openstack.common.gettextutils import _
from heat.openstack.common import log as logging
logger = logging.getLogger(__name__)
# Whether TaskRunner._sleep actually does an eventlet sleep when called.
ENABLE_SLEEP = True
def task_description(task):
"""
Return a human-readable string description of a task suitable for logging
the status of the task.
"""
name = task.__name__ if hasattr(task, '__name__') else None
if isinstance(task, types.MethodType):
if name is not None and hasattr(task, '__self__'):
return '%s from %s' % (name, task.__self__)
elif isinstance(task, types.FunctionType):
if name is not None:
return str(name)
return repr(task)
class Timeout(BaseException):
"""
Timeout exception, raised within a task when it has exceeded its allotted
(wallclock) running time.
This allows the task to perform any necessary cleanup, as well as use a
different exception to notify the controlling task if appropriate. If the
task suppresses the exception altogether, it will be cancelled but the
controlling task will not be notified of the timeout.
"""
def __init__(self, task_runner, timeout):
"""
Initialise with the TaskRunner and a timeout period in seconds.
"""
message = _('%s Timed out') % str(task_runner)
super(Timeout, self).__init__(message)
# Note that we don't attempt to handle leap seconds or large clock
# jumps here. The latter are assumed to be rare and the former
# negligible in the context of the timeout. Time zone adjustments,
# Daylight Savings and the like *are* handled. PEP 418 adds a proper
# monotonic clock, but only in Python 3.3.
self._endtime = wallclock() + timeout
def expired(self):
return wallclock() > self._endtime
class ExceptionGroup(Exception):
'''
Container for multiple exceptions.
This exception is used by DependencyTaskGroup when the flag
aggregate_exceptions is set to True and it's re-raised again when all tasks
are finished. This way it can be caught later on so that the individual
exceptions can be acted upon.
'''
def __init__(self, exceptions=None):
if exceptions is None:
exceptions = list()
self.exceptions = list(exceptions)
def __str__(self):
return str(map(str, self.exceptions))
def __unicode__(self):
return unicode(map(str, self.exceptions))
class TaskRunner(object):
"""
Wrapper for a resumable task (co-routine).
"""
def __init__(self, task, *args, **kwargs):
"""
Initialise with a task function, and arguments to be passed to it when
it is started.
The task function may be a co-routine that yields control flow between
steps.
"""
assert callable(task), "Task is not callable"
self._task = task
self._args = args
self._kwargs = kwargs
self._runner = None
self._done = False
self._timeout = None
self.name = task_description(task)
def __str__(self):
"""Return a human-readable string representation of the task."""
return 'Task %s' % self.name
def _sleep(self, wait_time):
"""Sleep for the specified number of seconds."""
if ENABLE_SLEEP and wait_time is not None:
logger.debug(_('%s sleeping') % str(self))
eventlet.sleep(wait_time)
def __call__(self, wait_time=1, timeout=None):
"""
Start and run the task to completion.
The task will sleep for `wait_time` seconds between steps. To avoid
sleeping, pass `None` for `wait_time`.
"""
self.start(timeout=timeout)
self.run_to_completion(wait_time=wait_time)
def start(self, timeout=None):
"""
Initialise the task and run its first step.
If a timeout is specified, any attempt to step the task after that
number of seconds has elapsed will result in a Timeout being
raised inside the task.
"""
assert self._runner is None, "Task already started"
logger.debug(_('%s starting') % str(self))
if timeout is not None:
self._timeout = Timeout(self, timeout)
result = self._task(*self._args, **self._kwargs)
if isinstance(result, types.GeneratorType):
self._runner = result
self.step()
else:
self._runner = False
self._done = True
logger.debug(_('%s done (not resumable)') % str(self))
def step(self):
"""
Run another step of the task, and return True if the task is complete;
False otherwise.
"""
if not self.done():
assert self._runner is not None, "Task not started"
if self._timeout is not None and self._timeout.expired():
logger.info(_('%s timed out') % str(self))
try:
self._runner.throw(self._timeout)
except StopIteration:
self._done = True
else:
# Clean up in case task swallows exception without exiting
self.cancel()
else:
logger.debug(_('%s running') % str(self))
try:
next(self._runner)
except StopIteration:
self._done = True
logger.debug(_('%s complete') % str(self))
return self._done
def run_to_completion(self, wait_time=1):
"""
Run the task to completion.
The task will sleep for `wait_time` seconds between steps. To avoid
sleeping, pass `None` for `wait_time`.
"""
while not self.step():
self._sleep(wait_time)
def cancel(self):
"""Cancel the task and mark it as done."""
if not self.done():
logger.debug(_('%s cancelled') % str(self))
try:
if self.started():
self._runner.close()
finally:
self._done = True
def started(self):
"""Return True if the task has been started."""
return self._runner is not None
def done(self):
"""Return True if the task is complete."""
return self._done
def __nonzero__(self):
"""Return True if there are steps remaining."""
return not self.done()
def wrappertask(task):
"""
Decorator for a task that needs to drive a subtask.
This is essentially a replacement for the Python 3-only "yield from"
keyword (PEP 380), using the "yield" keyword that is supported in
Python 2. For example::
@wrappertask
def parent_task(self):
self.setup()
yield self.child_task()
self.cleanup()
"""
@functools.wraps(task)
def wrapper(*args, **kwargs):
parent = task(*args, **kwargs)
subtask = next(parent)
while True:
try:
if subtask is not None:
subtask_running = True
try:
step = next(subtask)
except StopIteration:
subtask_running = False
while subtask_running:
try:
yield step
except GeneratorExit as ex:
subtask.close()
raise ex
except:
try:
step = subtask.throw(*sys.exc_info())
except StopIteration:
subtask_running = False
else:
try:
step = next(subtask)
except StopIteration:
subtask_running = False
else:
yield
except GeneratorExit as ex:
parent.close()
raise ex
except:
subtask = parent.throw(*sys.exc_info())
else:
subtask = next(parent)
return wrapper
class DependencyTaskGroup(object):
"""
A task which manages a group of subtasks that have ordering dependencies.
"""
def __init__(self, dependencies, task=lambda o: o(),
reverse=False, name=None, aggregate_exceptions=False):
"""
Initialise with the task dependencies and (optionally) a task to run on
each.
If no task is supplied, it is assumed that the tasks are stored
directly in the dependency tree. If a task is supplied, the object
stored in the dependency tree is passed as an argument.
If aggregate_exceptions is set to True, then all the tasks will be run
and any raised exceptions will be stored to be re-raised after all
tasks are done.
"""
self._runners = dict((o, TaskRunner(task, o)) for o in dependencies)
self._graph = dependencies.graph(reverse=reverse)
self.aggregate_exceptions = aggregate_exceptions
if name is None:
name = '(%s) %s' % (getattr(task, '__name__',
task_description(task)),
str(dependencies))
self.name = name
def __repr__(self):
"""Return a string representation of the task."""
return '%s(%s)' % (type(self).__name__, self.name)
def __call__(self):
"""Return a co-routine which runs the task group."""
raised_exceptions = []
try:
while any(self._runners.itervalues()):
try:
for k, r in self._ready():
r.start()
yield
for k, r in self._running():
if r.step():
del self._graph[k]
except Exception as e:
self._cancel_recursively(k, r)
if not self.aggregate_exceptions:
raise
raised_exceptions.append(e)
except:
with excutils.save_and_reraise_exception():
for r in self._runners.itervalues():
r.cancel()
if raised_exceptions:
raise ExceptionGroup(raised_exceptions)
def _cancel_recursively(self, key, runner):
runner.cancel()
node = self._graph[key]
for dependent_node in node.required_by():
node_runner = self._runners[dependent_node]
self._cancel_recursively(dependent_node, node_runner)
del self._graph[key]
def _ready(self):
"""
Iterate over all subtasks that are ready to start - i.e. all their
dependencies have been satisfied but they have not yet been started.
"""
for k, n in self._graph.iteritems():
if not n:
runner = self._runners[k]
if not runner.started():
yield k, runner
def _running(self):
"""
Iterate over all subtasks that are currently running - i.e. they have
been started but have not yet completed.
"""
running = lambda (k, r): k in self._graph and r.started()
return itertools.ifilter(running, self._runners.iteritems())
class PollingTaskGroup(object):
"""
A task which manages a group of subtasks.
When the task is started, all of its subtasks are also started. The task
completes when all subtasks are complete.
Once started, the subtasks are assumed to be only polling for completion
of an asynchronous operation, so no attempt is made to give them equal
scheduling slots.
"""
def __init__(self, tasks, name=None):
"""Initialise with a list of tasks."""
self._tasks = list(tasks)
if name is None:
name = ', '.join(task_description(t) for t in self._tasks)
self.name = name
@staticmethod
def _args(arg_lists):
"""Return a list containing the positional args for each subtask."""
return zip(*arg_lists)
@staticmethod
def _kwargs(kwarg_lists):
"""Return a list containing the keyword args for each subtask."""
keygroups = (itertools.izip(itertools.repeat(name),
arglist)
for name, arglist in kwarg_lists.iteritems())
return [dict(kwargs) for kwargs in itertools.izip(*keygroups)]
@classmethod
def from_task_with_args(cls, task, *arg_lists, **kwarg_lists):
"""
Return a new PollingTaskGroup where each subtask is identical except
for the arguments passed to it.
Each argument to use should be passed as a list (or iterable) of values
such that one is passed in the corresponding position for each subtask.
The number of subtasks spawned depends on the length of the argument
lists. For example:
PollingTaskGroup.from_task_with_args(my_task,
[1, 2, 3],
alpha=['a', 'b', 'c'])
will start three TaskRunners that will run:
my_task(1, alpha='a')
my_task(2, alpha='b')
my_task(3, alpha='c')
respectively.
If multiple arguments are supplied, each list should be of the same
length. In the case of any discrepancy, the length of the shortest
argument list will be used, and any extra arguments discarded.
"""
args_list = cls._args(arg_lists)
kwargs_list = cls._kwargs(kwarg_lists)
if kwarg_lists and not arg_lists:
args_list = [[]] * len(kwargs_list)
elif arg_lists and not kwarg_lists:
kwargs_list = [{}] * len(args_list)
task_args = itertools.izip(args_list, kwargs_list)
tasks = (functools.partial(task, *a, **kwa) for a, kwa in task_args)
return cls(tasks, name=task_description(task))
def __repr__(self):
"""Return a string representation of the task group."""
return '%s(%s)' % (type(self).__name__, self.name)
def __call__(self):
"""Return a co-routine which runs the task group."""
runners = [TaskRunner(t) for t in self._tasks]
try:
for r in runners:
r.start()
while runners:
yield
runners = list(itertools.dropwhile(lambda r: r.step(),
runners))
except:
with excutils.save_and_reraise_exception():
for r in runners:
r.cancel()