v0.4.5

Highlights

• Added one-command Helm OCI installation for Kubernetes and K3s clusters.
• Added automatic discovery of Flannel VXLAN and containerd runtime paths.
• Published version-aligned FNCache runtime image and Helm Chart.
• Completed Kubernetes Pod lifecycle handling, endpoint cleanup, IP reuse protection, and resync recovery.
• Completed the real two-node functional acceleration closure with fail-safe fallback to Flannel.
• Improved health checks, readiness reporting, heartbeat handling, and recovery after agent or dependency failures.

Installation

helm upgrade --install fncache oci://ghcr.io/fncache-project/charts/oncache --version 0.4.5 --namespace fncache-system --create-namespace --wait

Published Artifacts

• Runtime image: ghcr.io/fncache-project/fncache-agent:v0.4.5
• Helm Chart: oci://ghcr.io/fncache-project/charts/oncache, version 0.4.5
• Image digest: sha256:08323be78f1f14c6adf635c50124e0aaa16d2eb1a91068e521de794ae4cb41c0

Scope

This release targets Kubernetes or K3s clusters using Flannel VXLAN, containerd, IPv4, and amd64 nodes.

Limitations

This remains an experimental pre-production release. Performance, capacity, compatibility-matrix, fault-injection, and 24-hour soak validation are not included in this release.

v0.4.0

This is the first pre-production-ready release of ONCache-Flannel. The agent has a complete state machine, health epoch-bound heartbeat, multi-layer health checks, safe change publishing, graceful shutdown, and full observability. It is designed for controlled pre-production environments where it can run long-term for validation. Full validation is planned for M5.

v0.3.0

This is the first runnable release of ONCache-Flannel. The agent now watches Kubernetes Pod and Node events, automatically attaches eBPF programs to local Pods, cleans up when Pods are deleted, and handles IP reuse, migration, and remote endpoint invalidation. Users can install it with a single Helm command and benefit from automatic acceleration without manual intervention.

v0.2.0

This release delivers the node-local control plane of ONCache-Flannel. The Go agent can now discover the node environment, load the eBPF data plane, attach TC filters, scan actual kernel state, and reconcile desired vs. actual state — all without depending on Kubernetes events.

v0.1.0

This is the first milestone release of FNCache. It delivers a buildable eBPF data plane with fail-safe bypass control. The four TC programs and six BPF maps compile into a single ELF, share a unified ABI, and safely bypass when disabled, misconfigured, or when the agent is not present.