openova/platform/velero

Velero

Kubernetes backup/restore for disaster recovery. Per-host-cluster infrastructure (see docs/PLATFORM-TECH-STACK.md §3.5) — runs on every host cluster Catalyst manages. Backups land in the velero-backups bucket on SeaweedFS, which is Catalyst's unified S3 encapsulation layer; SeaweedFS's cold-tier policy automatically transitions backup objects to the configured cloud archival backend (Cloudflare R2 / AWS S3 / Hetzner Object Storage / etc.) so backups survive cluster failure without any direct cloud-S3 call from Velero itself.

Status: Accepted | Updated: 2026-04-28

---

Overview

Velero provides Kubernetes-native backup. All Velero output goes to the same single S3 endpoint — seaweedfs.storage.svc:8333, bucket velero-backups. SeaweedFS handles the rest: hot-tier in-cluster for fast restore of recent backups; cold-tier in cloud archival storage for backups beyond the configured warm-window.

flowchart TB
    subgraph K8s["Kubernetes Cluster"]
        Velero[Velero]
        Apps[Applications]
        PVs[Persistent Volumes]
    end

    subgraph SW["SeaweedFS (in-cluster S3 encapsulation)"]
        Bucket[velero-backups bucket]
        TierMgr[Tier Manager]
    end

    subgraph Archival["Cloud archive backend (cold tier)"]
        R2[Cloudflare R2]
        S3[AWS S3]
        GCS[GCP GCS]
        Hetzner[Hetzner Object Storage]
        OCI[OCI Object Storage]
    end

    Apps --> Velero
    PVs --> Velero
    Velero -->|"Backup"| Bucket
    Bucket --> TierMgr
    TierMgr -->|"After warm window"| Archival

---

Why route through SeaweedFS

Property	Direct cloud-S3 calls	Through SeaweedFS encapsulation
Number of S3 endpoints in Catalyst components	N (one per consumer × cloud)	1 (seaweedfs.storage.svc:8333)
Hot-restore latency for recent backups	Cloud round-trip	Near-zero (in-cluster cache)
Audit / lifecycle / encryption boundary	Per-component	One central boundary
Air-gap deployment	Requires direct cloud reachability	Works with SeaweedFS-only mode (see SRE §7)

Backups survive cluster failure because SeaweedFS's cold tier is the cloud archival backend, not the in-cluster volumes. Even if the entire host cluster is destroyed, backups beyond the warm window already live in the cold backend (R2 / Glacier / etc.) and a restoring SeaweedFS can read them through.

---

Storage Backend Options

Provider	Availability	Egress Fees	Notes
Cloud Provider Storage	Default	Varies	Hetzner, OCI, Huawei OBS
Cloudflare R2	Always available	Free	Zero egress, multi-cloud friendly
AWS S3	Available	$0.09/GB	Full featured
GCP GCS	Available	$0.12/GB	Full featured

Default: Cloud provider's object storage (Hetzner Object Storage, OCI Object Storage, etc.)

Alternative: Cloudflare R2 for zero egress fees, useful for multi-cloud or egress-heavy scenarios.

---

Configuration

Cloudflare R2 (Zero Egress)

apiVersion: velero.io/v1
kind: BackupStorageLocation
metadata:
  name: r2-backup
  namespace: velero
spec:
  provider: aws
  bucket: <org>-backups
  config:
    region: auto
    s3ForcePathStyle: "true"
    s3Url: https://<account-id>.r2.cloudflarestorage.com
  credential:
    name: r2-credentials
    key: cloud

AWS S3

apiVersion: velero.io/v1
kind: BackupStorageLocation
metadata:
  name: s3-backup
  namespace: velero
spec:
  provider: aws
  bucket: <org>-backups
  config:
    region: us-east-1
  credential:
    name: aws-credentials
    key: cloud

GCP GCS

apiVersion: velero.io/v1
kind: BackupStorageLocation
metadata:
  name: gcs-backup
  namespace: velero
spec:
  provider: gcp
  bucket: <org>-backups
  credential:
    name: gcp-credentials
    key: cloud

---

Backup Schedule

apiVersion: velero.io/v1
kind: Schedule
metadata:
  name: daily-backup
  namespace: velero
spec:
  schedule: "0 2 * * *"  # Daily at 2 AM
  template:
    includedNamespaces:
      - "*"
    excludedNamespaces:
      - velero
      - kube-system
    includedResources:
      - "*"
    excludedResources:
      - events
      - events.events.k8s.io
    storageLocation: r2-backup
    ttl: 720h  # 30 days

Backup Strategy

Resource	Schedule	Retention
All namespaces	Daily 2 AM	30 days
Databases (labels)	Hourly	7 days
Secrets	Daily	90 days
PVs (snapshots)	Daily	14 days

---

Multi-Region Backup

flowchart TB
    subgraph Region1["Region 1"]
        V1[Velero]
        K1[Kubernetes]
    end

    subgraph Region2["Region 2"]
        V2[Velero]
        K2[Kubernetes]
    end

    subgraph Archival["Archival S3"]
        Bucket[Shared Bucket<br/>or Cross-Region Replication]
    end

    V1 -->|"Backup"| Bucket
    V2 -->|"Backup"| Bucket
    Bucket -->|"Restore"| V1
    Bucket -->|"Restore"| V2

Both regions can:

• Backup to same bucket (different prefixes)
• Restore from either region's backups
• Use for cross-region disaster recovery

---

Restore Procedure

sequenceDiagram
    participant Op as Operator
    participant Velero as Velero
    participant S3 as Archival S3
    participant K8s as Kubernetes

    Op->>Velero: velero restore create
    Velero->>S3: Fetch backup
    S3->>Velero: Return backup data
    Velero->>K8s: Restore resources
    Velero->>K8s: Restore PV data
    K8s->>Op: Restoration complete

Commands

# List available backups
velero backup get

# Restore entire backup
velero restore create --from-backup daily-backup-20260116

# Restore specific namespace
velero restore create --from-backup daily-backup-20260116 \
  --include-namespaces databases

# Restore to different namespace
velero restore create --from-backup daily-backup-20260116 \
  --include-namespaces databases \
  --namespace-mappings databases:databases-restored

---

Operations

Check Backup Status

# List backups
velero backup get

# Describe specific backup
velero backup describe daily-backup-20260116

# Check backup logs
velero backup logs daily-backup-20260116

Verify Backup Location

# Check backup storage locations
velero backup-location get

# Verify connection
velero backup-location check r2-backup

Manual Backup

# Create manual backup
velero backup create manual-backup-$(date +%Y%m%d)

# Backup specific namespace
velero backup create db-backup-$(date +%Y%m%d) \
  --include-namespaces databases

---

Consequences

Positive:

• K8s-native backup
• Flexible storage backends
• Zero egress with Cloudflare R2
• Cross-region restore capability
• Incremental backups

Negative:

• Requires external S3 (by design)
• PV backup requires CSI snapshots
• Large restores take time

---

Part of OpenOva