Secure Containers — supported production profile
Name: fluxvm-sc-supported (informal).
Status: GA RuntimeClass with documented boundaries with documented scope boundaries.
This is the profile buyers can run in production without soft language. Outside this matrix, treat the workload as lab-only.
Supported matrix
| Constraint | Supported | Notes |
|---|---|---|
| VMM | QEMU (primary); Cloud Hypervisor allowed | Set FLUXVM_CONTAINER_BACKEND=qemu or cloud-hypervisor |
| Firecracker | Block-share path only | No live virtio-fs; Pod shares packed to ext4 at launch; needs FLUXVM_CONTAINER_KERNEL |
| CNI primary | Cilium, Calico, or Flannel | Auto-detect via FLUXVM_CONTAINER_CNI_PROVIDER; datapath auto (direct with bridge fallback) |
| Multus | Secondary NICs opt-in | Hybrid bridge default; FLUXVM_CONTAINER_CNI_MULTUS_DATAPATH=direct|auto when ready |
| hostPath | Allowlist / broker only | FLUXVM_HOSTPATH_ALLOW and/or FLUXVM_HOSTPATH_BROKER=1 (QEMU/CH hotplug) |
| Guest image | guestkit-baked Fedora or Ubuntu SC guest | Prefer signed catalog entries; token inject via guestkit (no virt-customize in the FluxVM path) |
| Policy | In-cluster NetworkPolicy → Sentinel | Host TC + optional guest cgroup_skb mirror; opt-in remote seccomp NOTIFY RPC (mode=remote) |
| Isolation extras | Opt-in CLONE_NEWUSER, SELinux linux.mountLabel, seccomp NOTIFY | Live evidence under docs/benchmarks/evidence/sc-live-*.txt |
| TEE / confidential | Out of profile | SC is isolation without TEE. SNP/TDX + attestation stays Ragnarok |
Non-goals (keep)
- CDI /
virtctl - Claiming in-tree virtio-fs is as deep as Cloud Hypervisor under load (prefer CH for production shared-FS)
- Claiming SC as confidential / TEE
Evidence
Live lab pack (dated): benchmarks/evidence/sc-hotcake-bundle-20260926.txt
Flip runbook: secure-containers-flip-runtimeclass.md
Rollup: secure-containers.md