How limbo-io/fluxion works
Project type and purpose
• Fluxion is a full-stack, distributed task scheduling platform composed of three primary logical layers:
a. Broker (fluxion-server) – the control plane that maintains cluster membership, scheduling logic, and task dispatch.
b. Worker (fluxion-worker) – the execution plane where tasks/jobs actually run.
c. Console/API (fluxion-server-start) – a RESTful façade for operators to manage tasks, workflows, triggers, etc.
• It uses a relational database (via JPA + Flyway) for persistent state and a custom RPC layer (fluxion-remote) for inter-node communication.
File structure
• Top-level modules:
• fluxion-common – shared utilities (thread pools, JSON, hashing, reflection).
• fluxion-remote – RPC framework (core abstractions, Netty transport, optional Vert.x).
• fluxion-server – Broker logic, broken into “core” (CQRS services, scheduling, broker tasks, DAOs) and “start” (SpringBoot application, controllers, converters, API layer).
• fluxion-worker – Worker logic (“core”), a demo application, and a Spring-Boot starter for easy integration.
• fluxion-test – unit/integration tests.
• Pattern indications:
• CQRS (infrastructure/cqrs), DAG-based workflows (infrastructure/dag), plugin RPC transport, repository/command/query packages.
• Modular separation (API vs core), clear service/DAO layering.
README for additional insights
• Scheduling types (fixed rate, fixed delay, cron).
• Load-balancing strategies (random, round robin, LRU, LFU, consistent hash, manual).
• Node-filtering (tags, executors, capacity).
• Execution types (single job vs workflow DAG), task types (normal vs broadcast).