Skip to content

Latest commit

 

History

History
106 lines (78 loc) · 3.74 KB

README.md

File metadata and controls

106 lines (78 loc) · 3.74 KB
Raw

Autoscaling Lifecycle

A library to handle aws events.

This library is based on the transitions state machine implementation.

Base concepts

Event

This library can handle the following AWS events:

  • Autoscaling events (aws.autoscaling)
  • EC2 Command Status-change Notification events (aws.ssm)
  • Scheduled events (aws.events)

To use this library, you need to implement it in a lambda function that is configured as a target of a cloudwatch rule which collects events of those types.

Each event is than treated as an attempt to transition a resource from its current state to a specific destination state.

Model

The model contains a configuration of possible transitions and implements the corresponding tasks to actually perform the transitions. It takes an event as a parameter from which it loads the resource to operate on and gets initialized with the current state of that resource.

It will be notified of any state changes that are performed and is responsible to persist the resulting state back to the underlying resource.

Base implemenation for handling autoscaling lifecycle events

@todo create new class LifecycleModel (based on Model) and move lifecycle specific methods to the new class

This library contains an abstract implementation of a model which focuses on autoscaling lifecycle events. To make autoscaling instances available to this model, they need to be registered to a dynamodb table containing the following fields:

  • Ident: the instance id
  • ItemStatus: new or pending
  • InstanceIp: the private ip of that instance

If an instance is new or pending, the model will wait for it to reach the initial state (the very first source state in the configuration). Usually both, registering a node and updating the state (ItemStatus) to the initial state is done through a cloud init script. In that case the initial state might be finished_cloud_init which is set in the very last line of the cloud init script.

Once a new instance has reached its initial state or an existing node has been loaded, the model initialisation is complete and can now be used by the LifecycleHandler.

As a base implementation, this model already provides tasks

A common configuration of an autoscaling model might contain the following transitions:

  • finished_cloud_init => join_cluster
  • join_cluster => joined_cluster
  • joined_cluster => updating_dns
  • updating_dns => dns_updated
  • dns_updated => completing_lifecycle
  • completing_lifecycle => ready

Please subclass this model and add a transition configuration by implementing Model.get_transitions() and the corresponding task methods.

LifecycleHandler

The LifecycleHandler is the heart of this library. It initializes a new state machine using the transitions it gets from a model and dispatches the corresponding triggers for the current state.

@todo document failure handling and stop conditions @todo rename to Dispatcher

Configuration

A transition is configured as a dict with the following format:

{
    'source': 'state1',
    'dest': 'state2',
    'triggers': [
        {
            'name': 'do_something',
        }
    ]
}

This means: To transition a resource from state1 to state2, the trigger do_something needs to be called.

A trigger is a list of tasks (methods implemented by the model). These tasks may contain conditions and can be configured to run before or after (or both) the state change gets persisted to the model.

{
    'source': 'state1',
    'dest': 'state2',
    'triggers': [
        {
            'name': 'do_something',
            'conditions': [<method_in_model>],
            'after': [<method_in_model>],
            'before': [<method_in_model>],
        }
    ]
}