TL;DR — Multus CNI is a meta-plugin that lets you attach multiple network interfaces to a single Kubernetes pod. For AI workloads this is essential: the primary interface handles cluster networking (via Cilium/Calico), while secondary interfaces connect directly to high-speed RDMA/InfiniBand fabrics for GPU-to-GPU collective communication — separate data and control planes in one pod.
What it is
Multus is a CNI meta-plugin — it doesn't provide networking itself, but lets you attach multiple CNI plugins to a single pod, each providing a separate network interface. It's the standard way to give a pod both a regular cluster network (eth0) and additional high-speed interfaces (e.g., an SR-IOV VF for RDMA). In the AI Native landscape it's in AI Native Infra › Network.
Why it exists
Kubernetes gives every pod exactly one network interface by default. But GPU training pods need two networks: the cluster network for API calls, health checks, and metrics — and a separate high-bandwidth, low-latency network (InfiniBand, RoCE) for NCCL all-reduce and parameter sync. Without Multus, you can't attach both to the same pod.
Fig 1 — Multus attaches two interfaces: cluster network for control, InfiniBand for GPU traffic.
How it works
Multus runs as a DaemonSet and intercepts CNI calls from the kubelet. For each pod, it first invokes the primary CNI (e.g., Cilium) to create eth0, then iterates over any NetworkAttachmentDefinition annotations on the pod to invoke secondary CNIs — each one creates an additional interface inside the pod's network namespace. The pod sees multiple NICs, each with its own IP and routes.
Key features
- Multiple interfaces — attach any number of secondary NICs to a pod.
- CNI-agnostic — works with any primary CNI (Cilium, Calico, Flannel).
- NetworkAttachmentDefinition CRD — declarative config for secondary networks.
- SR-IOV integration — the primary use case for AI: attach SR-IOV VFs for RDMA.
- Thick plugin mode — v4.x supports a thick plugin (client/server) for reliability and event-driven attachment.
Quick start
Install Multus, then define a secondary network and annotate your pod:
# install Multus
kubectl apply -f https://raw.githubusercontent.com/k8snetworkplumbingwg/multus-cni/master/deployments/multus-daemonset-thick.yml# NetworkAttachmentDefinition for SR-IOV
apiVersion: k8s.cni.cncf.io/v1
kind: NetworkAttachmentDefinition
metadata:
name: sriov-rdma
annotations:
k8s.v1.cni.cncf.io/resourceName: intel.com/sriov_rdma
spec:
config: '{"type":"sriov","vlan":100,"ipam":{"type":"whereabouts","range":"10.56.0.0/16"}}'
---
# pod annotation to attach it
metadata:
annotations:
k8s.v1.cni.cncf.io/networks: sriov-rdmaThe pod gets eth0 (primary CNI) + net1 (SR-IOV RDMA interface).
When to use, when to skip
Use it whenever a pod needs more than one network — the primary scenario for AI is attaching RDMA/InfiniBand interfaces alongside the cluster network for GPU training. Also used for storage networks, management networks, or telco NFV workloads.
Skip it if your pods only need one network interface — single-NIC is the Kubernetes default and requires no extra tooling.
vs / alongside
| Tool | Role | Note |
|---|---|---|
| Multus | Multi-NIC meta-plugin | The orchestrator for multiple interfaces |
| Cilium | Primary CNI | Handles eth0 / cluster networking |
| SR-IOV CNI | SR-IOV interface plugin | Invoked by Multus for RDMA VFs |
| RDMA | GPU data plane protocol | Runs over the SR-IOV interface |
References
- multus-cni — source and docs.
- Quick start guide — installation and first attachment.
- Thick plugin — v4 architecture.
Extra reads
- Kubernetes networking with Multus — Red Hat walkthrough.
- NVIDIA networking + Multus — GPU cluster networking.
Verified against Multus CNI docs (github.com/k8snetworkplumbingwg), May 2026.
Where Multus fits: the mental model
Multus is a foundational infrastructure service that moves, stores, connects, or distributes AI assets. The useful question is not simply “can it run the demo?” It is whether the component gives your team a clear ownership boundary, predictable failure behavior, and enough evidence to operate changes safely. Treat it as one replaceable layer in a larger system rather than letting it quietly become the architecture.
Start by drawing the request and data path. Mark where untrusted input enters, where identity is checked, where durable state changes, and where retries can repeat work. That diagram tells you which guarantees belong to Multus and which still belong to your application, platform, cloud provider, or database. The distinction matters during incidents: a healthy process is not proof that the end-to-end task is correct.
Core concepts you should understand first
The vocabulary below is more important than any single SDK method. It lets application engineers, platform engineers, security reviewers, and incident responders describe the same system without confusing a framework feature with an end-to-end guarantee.
| Concept | Meaning in this layer | Design question |
|---|---|---|
| Data plane | The hot path that carries bytes, packets, objects, or artifacts. | Write down how Multus represents or enforces this before production. |
| Control plane | APIs and controllers that configure, place, authorize, and observe the data plane. | Write down how Multus represents or enforces this before production. |
| Consistency | What a reader may observe during concurrent writes, replication, or failure. | Write down how Multus represents or enforces this before production. |
| Locality | Keeping compute near data or devices to reduce latency, egress, and cross-zone traffic. | Write down how Multus represents or enforces this before production. |
| Identity | A workload or human principal used to authenticate and authorize every operation. | Write down how Multus represents or enforces this before production. |
| Recovery objective | The measured RPO and RTO for metadata and data, not merely the presence of replicas. | Write down how Multus represents or enforces this before production. |
From quick start to a production deployment
The earlier quick start proves that the package or service runs. Production readiness is a different exercise. Build the smallest vertical slice that crosses every real boundary—identity, network, persistence, upstream provider, telemetry, and rollback—before broadening the feature set.
- Pin the compatibility envelope. Record the Multus release, language/runtime version, client SDK version, model or backend version, and—where applicable—Kubernetes API or driver requirements. Use a lock file, immutable image digest, or chart version; floating “latest” tags prevent repeatable rollback.
- Define contracts before configuration. Write the accepted input, successful output, error classes, timeout, idempotency behavior, and ownership of durable state. Validate at the boundary so corrupt work fails early instead of surfacing deep in a workflow.
- Create separate development, staging, and production identities. Do not copy a broad personal API key into every environment. Prefer workload identity or short-lived credentials, scope access by tenant and operation, and verify denial cases as part of deployment.
- Add bounded failure behavior. Every remote call needs a deadline. Retry only transient, idempotent operations with exponential backoff and jitter. Set concurrency and queue limits so an upstream slowdown becomes controlled backpressure rather than resource exhaustion.
- Instrument the complete path. Emit a correlation ID, component and release version, duration, outcome, retry count, and resource or cost dimensions. Keep sensitive prompt, document, and credential values out of ordinary logs.
- Ship through a reversible rollout. Run compatibility and regression tests, deploy to a canary or isolated workload, compare service-level indicators, then increase exposure. Preserve the previous artifact and configuration until rollback has been exercised.
Production configuration checklist
- Pin artifacts by version and, where possible, digest.
- Set connect, request, and total workflow deadlines.
- Bound retries, concurrency, queue length, and payload size.
- Separate read-only operations from mutations.
- Use idempotency keys for replayable mutations.
- Persist canonical state outside disposable workers.
- Encrypt traffic and durable data with managed keys.
- Redact secrets, tokens, prompts, and personal data.
- Apply per-tenant quotas and authorization filters.
- Expose readiness separately from process liveness.
- Back up metadata and test restore, not only backup.
- Document owner, escalation path, RPO, and RTO.
Failure modes and the response you should design
| Failure mode | What you observe | Engineering response |
|---|---|---|
| Metadata loss | Data exists but indexes, configuration, or ownership are gone. | Back up metadata separately and test restore to an isolated environment. |
| Cross-zone cost | A correct design creates unexpected egress and latency. | Make placement and traffic locality visible in cost and SLO dashboards. |
| Credential leak | Static secrets are copied into images or manifests. | Use workload identity, rotation, scoped roles, and secret scanning. |
| Capacity cliff | A quota, inode, object count, route, or device limit is reached. | Alert on forecasted exhaustion and document hard limits. |
| Split configuration | Nodes run incompatible policy or protocol versions. | Use staged rollouts and explicit version-skew rules. |
| Untested restore | Backups succeed but cannot recreate a working service. | Run scheduled restore drills and measure RPO/RTO. |
Turn these rows into runbook entries with an alert, first diagnostic query, safe mitigation, and escalation owner. Test at least one failure in staging every release cycle. If the system cannot be forced into a failure safely, it is usually not yet observable or isolated enough.
Security, privacy, and tenant isolation
Place Multus in a threat model, not just an architecture diagram. Identify human users, workload identities, administrators, upstream services, model providers, artifact registries, and data stores. For each edge, document authentication, authorization, encryption, audit evidence, and the consequence of credential compromise.
Apply least privilege at the operation and resource level. A component that only retrieves documents should not be able to delete the index; an evaluation worker should not inherit production mutation credentials; a model-serving pod should not need cluster-admin. In multi-tenant systems, enforce the tenant boundary before retrieval or execution and include tenant identity in quotas and audit events. Never rely on a prompt instruction, namespace string supplied by the client, or UI filtering as authorization.
Decide what data is permitted in telemetry. Prompts, retrieved chunks, tool arguments, model responses, notebooks, and traces can contain secrets or regulated data. Redact close to collection, keep high-sensitivity payload capture opt-in, encrypt exports, restrict support access, and give each class an explicit retention period. Verify deletion across caches, replicas, indexes, backups, and derived evaluation datasets.
Observability and service-level objectives
A useful dashboard follows the user-visible unit of work and then decomposes it by component, release, tenant tier, backend, and failure class. Start with these signals for Multus:
- availability and error rate — graph both rate and distribution, then compare with the previous release and traffic mix.
- p50/p95/p99 latency — graph both rate and distribution, then compare with the previous release and traffic mix.
- throughput and saturation — graph both rate and distribution, then compare with the previous release and traffic mix.
- replication or synchronization lag — graph both rate and distribution, then compare with the previous release and traffic mix.
- capacity and growth rate — graph both rate and distribution, then compare with the previous release and traffic mix.
- recovery time in drills — graph both rate and distribution, then compare with the previous release and traffic mix.
Choose an SLO at the boundary your users experience, such as “99% of accepted tasks complete correctly within five minutes over 28 days.” Availability alone is insufficient for AI systems because a fast but incorrect or ungrounded result is still a failure. Pair latency and completion objectives with a reviewed quality or policy indicator. Page on rapid error-budget burn; use tickets for slow capacity trends.
Testing and release strategy
Use four layers. Unit tests cover deterministic adapters, schemas, policy, and error mapping without a live external service. Contract tests exercise the pinned integration boundary—API, CLI, SDK, protocol, or ephemeral service—and verify its exact surface. Scenario tests exercise representative end-to-end cases, including permissions and state. Load and resilience tests establish saturation, queue behavior, retry amplification, and recovery after dependency loss.
Keep a small blocking suite for every commit and a broader scheduled suite for expensive or probabilistic checks. Store results with the application version, Multus version, configuration hash, model/backend version, dataset version, and random seed. A score without that provenance cannot explain a regression. Before upgrading, read the migration notes, run both versions against the same replay set, and explicitly test rollback across any schema or state transition.
How to decide whether Multus is the right tool
| Question | Evidence to collect | Red flag |
|---|---|---|
| Does it remove a real constraint? | A measured bottleneck, missing guarantee, or repeated custom component. | Adoption is based only on a demo or feature count. |
| Can the team operate it? | Named owner, upgrade path, alerts, runbooks, backup, restore, and on-call skills. | Only the original prototype author understands failure behavior. |
| Is the interface portable? | Your domain contracts wrap vendor-specific APIs; data and state have an export path. | Business objects are inseparable from framework internals. |
| Does it meet the envelope? | Benchmarks using your payloads, concurrency, topology, quality bar, and cost model. | Published benchmark hardware or workload does not resemble production. |
| Is failure affordable? | Tested degraded mode, bounded blast radius, rollback, RPO, and RTO. | A component outage blocks unrelated tenants or irreversible actions. |
Prefer the smallest component that satisfies the required guarantees. A provider SDK, relational table, background job, or standard Kubernetes controller is often better than another platform when the workload is small and predictable. Choose Multus when its specific abstraction removes sustained engineering work and the team is willing to own its lifecycle.
A focused 90-minute validation lab
- Minutes 0–15: run the documented quick start in a disposable environment with pinned dependencies. Save the exact commands and a known-good input/output fixture.
- Minutes 15–35: replace the toy input with one representative case from your system. Add schema validation, a deadline, and a correlation ID.
- Minutes 35–55: force invalid credentials, a timeout, malformed input, and one dependency failure. Record the observed errors and whether retries are safe.
- Minutes 55–75: run a small concurrency test and capture latency, throughput, saturation, and unit cost. Do not extrapolate beyond the tested range.
- Minutes 75–90: write the adoption decision: required guarantees met, open risks, owner, next experiment, and the simplest credible alternative.
Frequently asked questions
Should we standardize on Multus for every team?
Standardize the contracts, telemetry, security controls, and release evidence first. Standardizing one implementation is useful only when workloads share requirements and a platform team owns upgrades and support.
Can we use the hosted version and skip operations work?
Hosted service removes part of the control-plane burden, not architecture ownership. You still own identity, tenant isolation, data classification, quotas, dependency failure, observability, export, and an exit plan.
What should be pinned for reproducibility?
Pin the tool/server, client SDK, runtime, configuration, model or backend, container image digest, and test dataset. Record these values with every benchmark and evaluation result.
When is a proof of concept ready for production?
After representative success and failure tests pass, sensitive data paths are approved, limits and SLOs are defined, telemetry and runbooks exist, restore or rollback is rehearsed, and an accountable owner accepts the remaining risk.
Official sources and freshness
This guide was reviewed for architecture and operational guidance on 10 July 2026. Projects evolve quickly: verify installation syntax, supported versions, feature maturity, and upgrade notes against the exact release you deploy.