Implementing a Lightweight Distributed Scheduling Solution to Replace TBSchedule

Our project originally used the open‑source distributed scheduling framework TBSchedule, which depends on ZooKeeper; due to ZooKeeper instability and lack of maintenance, the scheduler occasionally suffered task stops, loss, or non‑execution, especially problematic during high‑traffic events like the 618 promotion.

To achieve stable operation with minimal modification cost, we designed a lightweight scheduling solution that can be degraded seamlessly.

Features of the original TBSchedule we leveraged:

Cluster and distributed support

Flexible task sharding

Dynamic service scaling and resource reclamation

Configurable and adjustable thread count per task

Our implementation goals:

Decentralize scheduling so each subsystem manages its own tasks, improving stability.

Replace ZooKeeper with Redis as the registration center and for dynamic sharding.

Use a thread‑pool instead of timer for thread control.

Automatically invoke TBSchedule internal methods to keep existing interfaces unchanged, avoiding code changes in business systems.

Support dynamic thread‑count adjustment to increase processing capacity.

Solution implementation: (image omitted)

Results:

Quick integration via XML configuration and jar dependency.

Core TBSchedule functionalities are fully replicated with no migration cost.

Combined with ducc configuration and an emergency plan, the system supports manual or automatic degradation, ensuring seamless operation during promotions.

Deployed online with stable dynamic sharding, timely heartbeat checks, and on‑demand degradation, preventing order backlog caused by ZooKeeper fluctuations.

Future plans:

Adopt Quartz (or EasyJob) as the trigger to support richer scheduling configurations, fully replacing TBSchedule.

Provide a visual monitoring interface for task execution status.

END