Provisioning
Self-hosting. This page is for operating your own Excalibase install. On Excalibase Cloud the platform runs the provisioner for you: create, back up and restore projects in Studio. The routes below belong to your install's control plane, not to
api.excalibase.io.
Built with Go + native client-go. No kubectl shell-outs. Provisions Postgres (15, 16 or 17) and DocumentDB projects.
Orchestrates best-of-breed Kubernetes operators to provide a unified REST API for multi-database provisioning, backup/restore, and lifecycle management.
Status
| Database | Status | Operator |
|---|---|---|
| PostgreSQL 15, 16, 17 | Available | CloudNativePG |
| DocumentDB (MongoDB-compatible, on Postgres) | Available | Platform addon |
| MySQL 8.4 | Behind a feature flag | MOCO |
MySQL projects run on the MOCO operator, one cluster per project. On your own install they are behind the mysql feature (features.mysql and mysql.enabled in the platform chart); they are not switched on for Excalibase Cloud yet. A MySQL project's end users live in an auth database inside the project's own instance, so they are backed up and restored with its data, as on Postgres.
What It Does
Instead of manually applying Kubernetes CRDs, you call a single REST endpoint. The provisioner handles:
- Namespace creation and isolation per tenant/project
- CRD deployment via native
client-godynamic client (no kubectl) - Credential generation with AES-256-GCM encryption
- Continuous WAL archiving and backup scheduling
- Point-in-time recovery (PITR)
- Per-pod CPU/memory monitoring via Kubernetes metrics-server
- pg_stat_statements performance insights
Architecture
┌────────────────────────────────────────┐
│ Go Server (27MB binary, <1s startup) │
├────────────────────────────────────────┤
│ chi router → Handler → Service │
│ ↓ │
│ Provisioner (Strategy pattern) │
│ ↓ │
│ client-go (dynamic + typed clients) │
│ ↓ │
│ K8s API → CloudNativePG Operator │
└────────────────────────────────────────┘
- No kubectl process spawning — all K8s operations use
client-gonatively - No Fabric8/JVM — 27MB static Go binary vs 200MB+ JVM
- 85%+ test coverage — unit tests with mock K8s client + k3s integration tests
8-Stage Provisioning Pipeline
Each stage is tracked independently and reported via the API.
PostgreSQL provisioning pipeline
| Stage | What Happens |
|---|---|
VALIDATING | Validate request, check project name uniqueness |
NAMESPACE_CREATION | Create isolated Kubernetes namespace {orgId}-{projectName} |
CRD_DEPLOYMENT | Apply CloudNativePG Cluster CRD via client-go dynamic client |
WAITING_FOR_READY | Poll all pods (primary + replicas) until Running and Ready |
CREDENTIAL_GEN | Extract credentials from auto-created K8s Secret, encrypt with AES-256-GCM |
BACKUP_CONFIG | Deploy ScheduledBackup CRD with WAL archiving to S3 |
METRICS_SETUP | Enable PodMonitor for STANDARD/ENTERPRISE tiers |
COMPLETED | Instance marked ACTIVE, credentials available |
Tiers
| Tier | Instances | HA | Storage | Memory | CPU | PodMonitor |
|---|---|---|---|---|---|---|
| FREE | 1 | No | 5Gi | 512Mi | 0.5 | No |
| STANDARD | 3 | Yes (1 primary + 2 standbys) | 50Gi | 4Gi | 2 | Yes |
| ENTERPRISE | 5 | Yes (1 primary + 4 standbys) | 500Gi | 16Gi | 4 | Yes |
STANDARD and ENTERPRISE use synchronous replication: a write is acknowledged only once a standby has it, so an acknowledged write survives a failover. Minor version upgrades use a switchover.
Quick Start
See the Quick Start guide to get a database running in 5 minutes.