Durable Actors - explore actor framework integration

[affandar]

Overview

⚠️ The ideas here have not been vetted in detail yet. They are meant to get the creative juices flowing and serve as a starting point for deeper exploration.

Explore integrating duroxide with actor frameworks to create durable actors — actors with persistent state, reliable message delivery, and crash recovery semantics.

Why Durable Actors?

Traditional actor frameworks are in-memory:

• Actor state is lost on crash
• Messages in flight are lost
• Supervision restarts actors from scratch
• No replay of message history

Combining duroxide's durability with the actor model could provide:

• Persistent state that survives crashes
• Durable mailbox with exactly-once delivery
• Replay recovery from message history
• Durable supervision decisions

Concepts to Explore

Concept 1: Actor as Orchestration

• Each actor is a long-running orchestration with continue_as_new
• Messages via external events
• Simple mapping to existing duroxide primitives

Concept 2: Actor System as Provider Extension

• Extend provider with actor-specific operations (mailboxes, state storage)
• Dedicated actor dispatcher

Concept 3: Hybrid Model

• In-memory actor runtime for performance
• Event sourcing to backing orchestration for durability
• Checkpointing to avoid full replay

Concept 4: Virtual Actors (Orleans-style)

• Actors addressable by ID, no explicit lifecycle
• Activate on demand, deactivate after idle
• Location transparent

Concept 5: Supervision Trees with Durable Recovery

• Erlang/Akka-style supervision (OneForOne, OneForAll, RestForOne)
• Supervision decisions recorded in history
• Durable supervision tree structure

Recommended Starting Point

Start with Actor as Orchestration to prove the concept with minimal effort, then evolve based on learnings.

See: proposals/durable-actors.md