How do you secure a production Kubernetes platform shared by many teams?
Short answer
Isolate tenants with namespaces, RBAC and network policies (or separate clusters for high sensitivity), enforce admission policies (non-root, read-only filesystems, no privileged pods, signed images), use workload identity instead of static credentials, manage secrets externally, keep clusters patched, restrict API server access, and monitor runtime behaviour.
Detailed answer
- Control plane: private API endpoint, SSO-based access, audit logs, regular upgrades.
- Tenancy: namespace per team; RBAC least privilege; ResourceQuotas and LimitRanges; separate clusters for regulated workloads.
- Admission control: Pod Security Standards (restricted), Kyverno/OPA Gatekeeper policies, image signature verification (Sigstore cosign), allowed registries.
- Network: default-deny NetworkPolicies; egress control; mTLS via mesh where needed.
- Identity and secrets: workload identity (IRSA/Pod Identity, Workload Identity), External Secrets, etcd encryption.
- Supply chain: scanned and minimal base images, SBOMs.
- Runtime: Falco/cloud runtime threat detection, node hardening, auto-repair.
What the interviewer expects
Layered controls across control plane, workloads, network, identity and supply chain.
Common mistakes to avoid
Cluster-admin for everyone, flat network, secrets in plain ConfigMaps, no image provenance.