ironic/doc/source/dev/architecture.rst
Dmitry Tantsur 3429e3824c Switch to Futurist library for asynchronous execution and periodic tasks
This change switches the conductor to using Futurist library executor and
periodic tasks worker instead of oslo.service periodic tasks. This allows
running periodic tasks in parallel and relying on more standard interfaces
(futures, executors) when dealing with asynchronous execution.

A green thread executor is used instead of using an eventlet green pool
directly. The maximum number of workers is taken from the existing
workers_pool_size configuration option, and no tasks are allowed
to be enqueued to mimic the previous behaviour (might be lifted later).

The periodic tasks worker is using the same executor, and its main loop thread
is also running on it. For this reason minimum value for workers_pool_size
is now 3: periodic task main loop, keep alive thread and at least one thread for
other tasks. A special decorator for driver-periodic tasks is now deprecated,
as the generic decorator can be used there as well.

Closes-Bug: #1526277
Change-Id: I57bf7cebfb6db805b6c521bacfef2993b16ce1ee
2016-02-24 15:44:05 +01:00

4.9 KiB

System Architecture

High Level description

An Ironic deployment will be composed of the following components:

  • An admin-only RESTful API service, by which privileged users, such as cloud operators and other services within the cloud control plane, may interact with the managed bare metal servers.

  • A Conductor service, which does the bulk of the work. Functionality is exposed via the API service. The Conductor and API services communicate via RPC.

  • A Database and DB API for storing the state of the Conductor and Drivers.

  • A Deployment Ramdisk or Deployment Agent, which provide control over the hardware which is not available remotely to the Conductor. A ramdisk should be built which contains one of these agents, eg. with diskimage-builder. This ramdisk can be booted on-demand.

    • NOTE: The agent is never run inside a tenant instance.

Drivers

The internal driver API provides a consistent interface between the Conductor service and the driver implementations. A driver is defined by a class inheriting from the BaseDriver class, defining certain interfaces; each interface is an instance of the relevant driver module.

For example, a fake driver class might look like this:

class FakePower(base.PowerInterface):
    def get_properties(self):
        return {}

    def validate(self, task):
        pass

    def get_power_state(self, task):
        return states.NOSTATE

    def set_power_state(self, task, power_state):
        pass

    def reboot(self, task):
        pass

class FakeDriver(base.BaseDriver):
    def __init__(self):
        self.power = FakePower()

There are three categories of driver interfaces:

  • Core interfaces provide the essential functionality for Ironic within OpenStack, and may be depended upon by other services. All drivers must implement these interfaces. The Core interfaces are power and deploy.
  • Standard interfaces provide functionality beyond the needs of OpenStack, but which have been standardized across all drivers and becomes part of Ironic's API. If a driver implements this interface, it must adhere to the standard. This is presented to encourage vendors to work together with the Ironic project and implement common features in a consistent way, thus reducing the burden on consumers of the API. The Standard interfaces are management, console, boot, inspect, and raid.
  • The Vendor interface allows an exemption to the API contract when a vendor wishes to expose unique functionality provided by their hardware and is unable to do so within the Core or Standard interfaces. In this case, Ironic will merely relay the message from the API service to the appropriate driver.

Driver-Specific Periodic Tasks

Drivers may run their own periodic tasks, i.e. actions run repeatedly after a certain amount of time. Such task is created by decorating a method on the driver itself or on any interface with periodic decorator, e.g.

from futurist import periodics

class FakePower(base.PowerInterface):
    @periodics.periodic(spacing=42)
    def task(self, manager, context):
        pass  # do something

class FakeDriver(base.BaseDriver):
    def __init__(self):
        self.power = FakePower()

    @periodics.periodic(spacing=42)
    def task2(self, manager, context):
        pass  # do something

Here the spacing argument is a period in seconds for a given periodic task. For example 'spacing=5' means every 5 seconds.

Message Routing

Each Conductor registers itself in the database upon start-up, and periodically updates the timestamp of its record. Contained within this registration is a list of the drivers which this Conductor instance supports. This allows all services to maintain a consistent view of which Conductors and which drivers are available at all times.

Based on their respective driver, all nodes are mapped across the set of available Conductors using a consistent hashing algorithm. Node-specific tasks are dispatched from the API tier to the appropriate conductor using conductor-specific RPC channels. As Conductor instances join or leave the cluster, nodes may be remapped to different Conductors, thus triggering various driver actions such as take-over or clean-up.