Proposal: State & failure handlers could be Tasks themselves

[peterroelants]

Currently

Currently a Task's on_failure or state_handlers should be regular methods (Callable) with a specified signature.

Proposed

Expand state handlers from regular methods to being Prefect Tasks themselves.

Why

Allowing state/failure handlers could provide the benefit of Prefect's ways for dealing with "negative engineering", such as retries, for those state handlers.

For example, I have a few state handlers that write a notification to a Kafka topic to signal failure, however calling the Kafka broker could be via an unreliable network, and retries are sometimes required.

Other potential benefits:

• Providing base Prefect Tasks for dealing with state change or with failure where the run method has the required signature.
• Could be clearer than using Triggers (which need explicit condition handling) to deal with failures.
• Would allow for more complex failure changes with a flow/DAG of tasks to deal with a certain failure.
• State handling becomes an explicit part of the flow and could be visualised and tracked with existing Prefect tools.