Quick facts

Form
VM on br30 (30.30.30.14:9000)
Buckets (live)
terrakube-state, vault-snapshots-logical (versioned, 14d non-current expiry), vault-replicator-tools (anonymous-read; mc + musl jq + libonig.so.5)
Buckets (planned)
jenkins-rke2-backups, nexus-blobs
Service accounts
vault-replicator-dc (PUT/GET/DELETE on vault-snapshots-logical/dc/*), vault-replicator-dr (GET on vault-snapshots-logical/dc/*)

What it is

Stable S3 endpoint outside both clusters so it's available regardless of which side of DC/DR is current. Application object data on the platform itself uses Longhorn PVCs.

As of 2026-05-06 it also acts as the carrier for Vault cross-cluster sync: DC's vault-replicator-export CronJob ships a logical-export tarball to vault-snapshots-logical/dc/ hourly, DR's import CronJob fetches latest.tar.gz every 4 h and replays into DR Vault. The original vault-snapshots Raft-snapshot bucket plan was abandoned because per-cluster Shamir keyrings can't decrypt each other's master keys (see Vault page).

Architecture

Add diagrams or topology notes here — how this component sits relative to DC/DR, what replicates, and what speaks to it. Mermaid blocks render as plain text today; embed images under /assets/ when needed.

Configuration

Add chart values overrides, important env vars, OIDC client config, secret references. Link to the source files in infra/gitops-rke2 rather than copying YAML wholesale — the repo is the source of truth.

Operations

Add runbook notes: backup, restore, common troubleshooting, dashboards, on-call cheatsheet entries.

Failover

Add DC/DR cutover/cutback notes: edge HAProxy backend name, healthcheck path, measured cutover/cutback times, smoke-test commands.

References