Kubernetes — overview¶
Kubernetes already provides what Docker needs extra parts for: readiness-driven routing
(Ingress + Service + readinessProbe), rolling updates with surge/unavailable budgets, and
operators for the database and Vault. The core sees exactly the same environment variables as on
Docker — nothing in it is Kubernetes-specific.
Object model¶
| Object | Purpose | Settings |
|---|---|---|
Deployment lyndrix-core (replicas 2–3) |
API instances | LYNDRIX_CLUSTER_MODE=cluster, LYNDRIX_INSTANCE_ROLES=api,leader, LYNDRIX_UI_ENGINE=react, LYNDRIX_TRUSTED_PROXIES=<ingress pod CIDR> (pods see the ingress' pod IP; * only behind a NetworkPolicy, see Ingress), LYNDRIX_DRAIN_SECONDS=10, LYNDRIX_INSTANCE_ID from metadata.name via the downward API |
Deployment lyndrix-runner (replicas 1) |
IaC Orchestrator host | LYNDRIX_INSTANCE_ROLES=api,runner; nodeSelector to a node with Docker (or DinD) until runners become Kubernetes Jobs |
Deployment lyndrix-ui |
frontend | LYNDRIX_BACKEND_URL=http://lyndrix-core (the Service) |
Service lyndrix-core |
load balancing | ClusterIP; readiness decides membership |
Ingress |
TLS, routing | one host; path rule for the runner's plugin prefix → lyndrix-runner Service |
Secret |
STORAGE_SECRET, LYNDRIX_MASTER_KEY, vault_keys.enc, DB password |
mounted/env on every pod |
PersistentVolumeClaim RWX (Longhorn, NFS) |
/data/storage |
shared files until the storage driver ships |
PodDisruptionBudget |
keep ≥ 1 ready during node drains | minAvailable: 1 |
Probes:
livenessProbe: { httpGet: { path: /api/live, port: 8081 }, periodSeconds: 10 }
readinessProbe: { httpGet: { path: /api/ready, port: 8081 }, periodSeconds: 5, failureThreshold: 2 }
startupProbe: { httpGet: { path: /api/live, port: 8081 }, failureThreshold: 30, periodSeconds: 5 }
terminationGracePeriodSeconds: 30 # ≥ LYNDRIX_DRAIN_SECONDS + 15
Database and Vault on Kubernetes¶
- Database: the MariaDB operator (
MariaDBCRD withgalera:and aMaxScaleCRD) gives Galera with a single-writer endpoint, backups as CRDs and rolling updates; or an external database viaDB_HOST. Contract: Database. - Vault/OpenBao: the official Helm charts with Raft (
server.ha.raft) on a block storage class; or an external Vault. Contract: Secrets.
Known limits¶
- No Helm chart is shipped yet; Reference manifests has the verified plain manifests for a Kustomize directory, the object model above is the spec for the chart.
- Orchestrator runners need a Docker daemon; running them as Kubernetes Jobs is a redesign that is out of scope for this milestone.
See also Ingress & load balancing and Rolling updates.