Fluid 1.1.0 Roadmap

[cheyang]

Fluid 1.1.0 Roadmap

Version: 1.1.0
Target Release Date: GA in Q1 2026

1. Vision

Transform Fluid into a more extensible, pluggable, and operator-friendly caching and data acceleration platform. Meet production-grade cluster requirements including node restriction awareness, runtime adjustment without service interruption, in-place upgrade capabilities, and rapid integration of multiple engines.

2. Core Objectives

1. Unified Cache Runtime Framework: Define universal Cache Runtime interfaces and standard APIs to facilitate rapid integration of new caching engines (e.g., CubeFS, DragonFly), with standardized compatibility layers for existing engines (Alluxio, Vineyard, JuiceFS).
2. Runtime Dynamic Configuration: Allow real-time modification of caching and client configurations without rebuilding Dataset resources, reducing operational overhead and risks.
3. Support Kubernetes Node-Specific Restrictions: Enable core Fluid components (CSI DaemonSet) to dynamically recognize and adhere to node-level constraints。
4. In-Place Update Support for Runtime Workers: Implement rolling/hot replacement upgrades to minimize maintenance windows and reduce data service disruption.
5. Native Sidecar Default Strategy: Enable native sidecar mode by default while maintaining backward compatibility for legacy Kubernetes versions (preserving legacy behavior/compatibility mode).

3. Key Features & Deliverables (Prioritized)

High Priority

• Unified Cache Runtime Framework

  • Design and expose standardized Cache Runtime interfaces (lifecycle management, cache I/O, metadata operations, flush/writeback, metrics exposure).
  • Deliver reference adapters (provide Curvine/ubeFS/DragonFly adapter blueprint/PoC).

• Runtime Dynamic Configuration (No Dataset Rebuild)

  • Define RuntimeConfig CR (or extend existing Dataset/Runtime CRDs) to support runtime modifications (caching policies, client configs, I/O throttling, etc.).
  • Implement progressive configuration rollout (applying changes incrementally without rebuilding Dataset), with change preview and rollback support.

• Node-Specific Restrictions Support

  • CSI DaemonSet can read node annotations/labels or API-returned restriction policies, enforcing constraints during pod scheduling/volume attachment (e.g., blocking sidecar deployment, limiting pod counts/resource caps).
  • Components dynamically adapt to node restriction changes (e.g., migrate workloads or pause scheduling to restricted nodes).
  • e2e test coverage for common scenarios: node upgrades introducing restrictions, node label toggling.

• In-Place Update for Runtime Workers

  • Implement in-place upgrade workflow in Runtime Controller: pull new image → health probe verification → traffic switch → safe termination of old instances.

• Native Sidecar Default Mode + Backward Compatibility

  • Enable Native Sidecar mode by default on supported K8s versions to simplify deployment and enhance performance.
  • Provide compatibility path for legacy K8s versions (fallback to init-container or legacy sidecar injection), with automatic selection or operator-configurable override.