StructuredStateManagementDetails

Details

In order to implement of this new orchestration layer the following key concepts must be built into the design from the start.

1. Atomic task units.
2. Combining atomic task units into a workflow.
3. Task resumption.
4. Task rollback.
5. Task tracking.
6. Resource locking.
7. Workflow sharding/ownership.
8. Simplicity (allowing for extension and verifiability).
9. Tolerance to upgrades.
10. Task cancellation (nice to have)

Atomic task units

Why it matters

Tasks that are created (either via code or other operation) must be atomic so that the task as a unit can be said to have completed or the task as a unit can be said to have failed. This allows for said task to be rolled back as a single unit.

Extra credit: It is also useful to be able to be able to accurately track exactly what tasks have been applied to a given workflow, which is inherently useful for correct status tracking (and is directly tied to how resumption is done).

How it will be addressed

States and state transition which previously were in the run_instance path of nova will need to be refactored into clearly defined tasks (likely with an apply() method and a rollback() method). These tasks will be split up so that each task performs a clearly defined and understandable single piece of work in an atomic manner (aka not one big task that does many different things) wherever possible. Note: that this will also help make testing of said task easier (since it will have a clear set of inputs and a clear set of expected outputs/side-effects, of which the rollback() method should undo), something which is not possible without this kind of task refactoring work.

For example this is the task base class created in the prototype:

class Task(base.Base):
    __metaclass__ = abc.ABCMeta

    def __init__(self):
        super(State, self).__init__()

    def __str__(self):
        return "Task: %s" % (self.__class__.__name__)

    @abc.abstractmethod
    def apply(self, context, *args, **kwargs):
        # Do whatever is needed to fulfill this task...
        #
        # Return a useful object for others to use, this object
        # must be serializable so that it can be given back in-case of reversions.
        raise NotImplementedError()

    # Gets any context information (nova context) and the result
    # that this function returned when its apply succeeded (if it succeeded)
    # so that said task can examine what it produced and undo whatever it 
    # produced. The chain which caused this exception is also provided as well
    # as the exception itself and the state which causes this reversion.
    def revert(self, context, result, chain, excp, cause):
        pass

Combining atomic tasks into a workflow

Why it matters

A workflow in nova can be typically organized into a series of smaller tasks/states which can be applied as a single unit (and rolled back as a single unit). Since we need to be able to run individual states in a trackable manner (and we shouldn't duplicate this code in multiple places) a concept of a chain of tasks (for the mostly linear nova case) is needed to be able to run and rollback that set of tasks as a group.

How it will be addressed

The following was created to address the issue of running (and rolling back) multiple states as a single unit.

Note: to start a linear unit is fine.

This is the 'chain' class created in the prototype:

class WorkflowChain(object):
    def __init__(self, name, tolerant=False, parents=None):
        # What tasks I have done that may need to be undone on failure.
        self.reversions = []
        self.name = name
        self.tolerant = tolerant
        # The order in which tasks will run is controlled by this - note its only linear.
        self.tasks = OrderedDict()
        self.results = OrderedDict()  # Useful for accessing the result of a given tasks after running...
        # A parent set of tasks that will need to be rolled back if this one fails.
        self.parents = parents
        # This functor is needed to be able to fetch results for tasks which have already occurred.
        #
        # It is required since we need to be able to rollback said tasks if it has already completed elsewhere
        # and one of the requirements we have put in place is that one rollback a task is given the result 
        # it returned when its apply method was called. 
        self.result_fetcher = None
        # This is a default tracker + other listeners which will always be notified when tasks start/complete/error. 
        #
        # They are useful for altering some external system that should be notified when a state changes.
        self.change_tracker = None
        self.listeners = []

    def __setitem__(self, name, performer):
        self.tasks[name] = performer

    def __getitem__(self, name):
        return self.results[name]

    def run(self, context, *args, **kwargs):
        for (name, performer) in self.tasks.items():
            try:
                self._on_state_start(context, performer, name)
                # See if we have already ran this... (resumption!)
                result = None
                if self.result_fetcher:
                    result = self.result_fetcher(context, name, self)
                if result is None:
                    result = performer.apply(context, *args, **kwargs)
                # Keep a pristine copy of the result in the results table
                # so that if said result is altered by other further states
                # the one here will not be. 
                #
                # Note: python is by reference objects, so someone else could screw with this,
                # which would be bad if we need to rollback and a result we created was modified by someone else...
                self.results[name] = copy.deepcopy(result)
                self._on_state_finish(context, performer, name, result)
            except Exception as ex:
                with excutils.save_and_reraise_exception():
                    try:
                        self._on_state_error(context, name, ex)
                    except:
                        pass
                    cause = (name, performer, (args, kwargs))
                    self.rollback(context, name, self, ex, cause)
        return self

    def _on_state_error(self, context, name, ex):
        if self.change_tracker:
            self.change_tracker(context, ERRORED, name, self)
        for i in self.listeners:
            i.notify(context, ERRORED, name, self, error=ex)

    def _on_state_start(self, context, performer, name):
        if self.change_tracker:
            self.change_tracker(context, STARTING, name, self)
        for i in self.listeners:
            i.notify(context, STARTING, name, self)

    def _on_state_finish(self, context, performer, name, result):
        # If a future state fails we need to ensure that we
        # revert the one we just finished.
        self.reversions.append((name, performer))
        if self.change_tracker:
            self.change_tracker(context, COMPLETED, name, self,
                                result=result.to_dict())
        for i in self.listeners:
            i.notify(context, COMPLETED, name, self, result=result)

    def rollback(self, context, name, chain=None, ex=None, cause=None):
        if chain is None:
            chain = self
        for (i, (name, performer)) in enumerate(reversed(self.reversions)):
            try:
                performer.revert(context, self.results[name], chain, ex, cause)
            except excp.NovaException:
                # Ex: WARN: Failed rolling back stage 1 (validate_request) of
                # chain validation due to nova exception
                # WARN: Failed rolling back stage 2 (create_db_entry) of
                # chain init_db_entry due to nova exception
                msg = _("Failed rolling back stage %s (%s)"
                        " of chain %s due to nova exception.")
                LOG.warn(msg, (i + 1), performer.name, self.name)
                if not self.tolerant:
                    # This will log a msg AND re-raise the Nova exception if
                    # the chain does not tolerate exceptions
                    raise
            except Exception:
                # Ex: WARN: Failed rolling back stage 1 (validate_request) of
                # chain validation due to unknown exception
                # WARN: Failed rolling back stage 2 (create_db_entry) of
                # chain init_db_entry due to unknown exception
                msg = _("Failed rolling back stage %s (%s)"
                        " of chain %s, due to unknown exception.")
                LOG.warn(msg, (i + 1), performer.name, self.name)
                if not self.tolerant:
                    # Log a msg AND re-raise the generic Exception if the
                    # Chain does not tolerate exceptions
                    raise
        if self.parents:
            # Rollback any parents chains
            for p in self.parents:
                p.rollback(context, name, chain, ex, cause)

Task resumption

Why it matters

Key to the ability to horizontal scaling your orchestration units is the ability to be able to resume the work of a failed orchestration unit on another orchestration unit. This makes it so that these orchestration units can survive individual program failure and not put the tasks they were working on into a ERROR like state, which should not be needed in a transactional workflow oriented system.

How it will be addressed

To start a task log will need to be created and when each task is executed that log will need to be updated accordingly. Note that it is not enough to store the name of the task that completed by enough metadata must also be stored with that log entry to be able to correctly revert that task. Typically in nova there is a concept of a resource that is being modified by each task, at the end of each task we can store the resource (which will pick up the modifications of that resource automatically).

'Issues:' the admin_password entry is a part of this resource, it needs to be stored securely (if we store it at all in the task log metadata).