Development Roadmap (v0.5.0)

[cheyang]

Here is the development roadmap for v0.5.0. Contributions and feedback are welcome.

Upgrades & Scalability

• [x] In-Place Upgrades @veophi @Syspretor Enhance support for component updates without pod recreation (building on v0.4.0 foundations).
• [x] Group-Level Partition Upgrades @ZYecho11 Enable granular partition upgrades at the group level for controlled deployments.
• [x] RBG ControllerRevision Support @bcfre Implement historical state tracking for RoleBasedGroup objects.
• [ ] Flexible Topology Scheduling
  Support Multi-level Topology Scheduling, Hard/Soft Constraint Separation, Weighted Preferences #47

Quality Assurance

• [x] End-to-End Testing Framework @gujingit Establish full test coverage to ensure stability during rapid iterations.

Runtime Evolution

• [x] Engine Runtime Open-Sourcing @gujingit @TrafalgarZZZ Open-source core runtime components to enable community contributions Prepare documentation and contribution guidelines.

Examples & Demos

• [ ] Mooncake Deployment Examples @Syspretor Provide reference implementation for Mooncake integration.
• [ ] KV cache offloading Examples @Syspretor Provide examples for KV cache offloading by Mooncake & hicache.
• [ ] SGLang Router @gujingit Provide examples for SGLang router.

Continuations from v0.4.0

The following items will be prioritized if not completed in v0.4.0:

• [ ] Orchestrated Upgrade Order
• [ ] Multi-Level Gang Scheduling @gujingit
• [ ] Configurable Failure Policies
• [ ] Runtime-less Service Discovery @cheyang

[ZYecho11]

/assign Group-Level Partition Upgrades and Orchestrated Upgrade Order