418cead036
Closes the user-reported regression where a deployment id created at 12:57
vanished after the catalyst-api Pod restarted ~6 times for image rolls,
leaving the wizard with "Unreachable / SSE connection closed before
completion / Deployment id 5cd1bceaaacb71f6". The in-memory sync.Map was
the entire authority on deployment state, so every Pod restart wiped it.
Changes
-------
internal/store: new flat-file JSON store, one file per deployment id under
/var/lib/catalyst/deployments/<id>.json. Atomic writes via temp-file +
fsync + rename. Walks the directory at startup to repopulate sync.Map.
Per-record schema with credentials redacted (HetznerToken, DynadotAPIKey,
DynadotAPISecret, RegistrarToken — all replaced by "<redacted>"; PDM
reservation token persists, it is a per-deployment opaque identifier, not
a credential).
handler: every event flows through recordEventAndPersist which appends to
the durable buffer AND rewrites the on-disk record under the lock. The
final terminal-state mutation (Status, Result, FinishedAt) gets one more
Save after the producer channel closes. Retry path mirrors the same
persist-after-event semantics. CreateDeployment persists the row before
launching the goroutine so a Pod kill in the spawn gap still leaves a
reachable record.
Restart semantics: in-flight statuses (pending / provisioning /
tofu-applying / flux-bootstrapping) are rewritten to "failed" with an
operator-facing error pointing at the orphaned Hetzner resources tagged
catalyst-deployment-id=<id>; the wizard's FailureCard renders instead of
showing a stuck progress bar forever.
chart: new PersistentVolumeClaim (RWO, 1Gi, default StorageClass) mounted
at /var/lib/catalyst/deployments. Deployment strategy switched to
Recreate because RWO single-attach can't roll. Pod-level fsGroup=65534
with fsGroupChangePolicy=OnRootMismatch so the non-root catalyst-api UID
can write to the PVC without recursive chown on every restart.
Containerfile: pre-create /var/lib/catalyst/deployments at build time
with UID 65534 ownership for the docker bind-mount path used in manual
testing.
Tests
-----
internal/store: secret redaction, save+load round-trip, rewrite over
existing file, garbage-file skipping, unwritable-dir error, path
traversal rejection.
internal/handler: 8 new persistence tests covering CreateDeployment
synchronous Save, restore-on-startup rehydration, in-flight rewrite to
failed, on-disk redaction, per-event append, nil-store no-op, and the
docker-style POST-kill-restart-GET round-trip. waitForGoroutine helper
prevents test cleanup races against the runProvisioning goroutine.
Gates
-----
go vet ./... — clean
go test ./... -count=1 -race — 5 packages pass
npx tsc --noEmit && npm run build — clean (no UI changes)
./node_modules/.bin/vitest run — 153 tests pass
Manual proof
------------
CATALYST_DEPLOYMENTS_DIR=/tmp/catalyst-test PORT=28381 catalyst-api &
curl POST /api/v1/deployments → id a70791c5716bbb13
cat /tmp/catalyst-test/a70791c5716bbb13.json → "<redacted>" present, no leak
kill -9 catalyst-api
CATALYST_DEPLOYMENTS_DIR=/tmp/catalyst-test PORT=28382 catalyst-api &
log: "restored deployments from PVC count=1"
curl GET /api/v1/deployments/a70791c5716bbb13 → 200 with persisted state
Hand-crafted in-flight record (status=tofu-applying) on disk → after
restart, GET surfaces status=failed with the orphaned-resources operator
message; on-disk file rewritten to match.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|---|---|---|
| .claude | ||
| .github | ||
| clusters | ||
| core | ||
| docs | ||
| infra/hetzner | ||
| platform | ||
| products | ||
| tests | ||
| CLAUDE.md | ||
| README.md | ||
OpenOva Catalyst
A self-sufficient Kubernetes-native platform. Published as signed OCI Blueprints. Deployable as your own Sovereign.
Catalyst is the open-source platform built by OpenOva. It turns any Kubernetes cluster into a Sovereign: a self-contained control plane that hosts Organizations, Environments, and Applications via GitOps + Crossplane, with a unified UI/Git/API for users.
Documentation
| Document | What it covers |
|---|---|
docs/GLOSSARY.md |
Canonical terminology — read first |
docs/ARCHITECTURE.md |
Catalyst architecture overview |
docs/IMPLEMENTATION-STATUS.md |
What's built today vs what's design-only — read second |
docs/NAMING-CONVENTION.md |
Naming patterns for every resource type |
docs/PERSONAS-AND-JOURNEYS.md |
Personas × journeys matrix; surfaces |
docs/SECURITY.md |
Identity (SPIFFE + Keycloak), secrets (OpenBao + ESO), rotation, multi-region semantics |
docs/SOVEREIGN-PROVISIONING.md |
How to bring a Sovereign online |
docs/BLUEPRINT-AUTHORING.md |
Writing Blueprints (incl. Crossplane Compositions) |
docs/PLATFORM-TECH-STACK.md |
Every component's role in Catalyst |
docs/SRE.md |
Operating a Sovereign |
docs/BUSINESS-STRATEGY.md |
Product strategy and GTM |
docs/TECHNOLOGY-FORECAST-2027-2030.md |
Component forecast 2027–2030 |
docs/VALIDATION-LOG.md |
Trail of doc-integrity validation passes (audit log) |
Heads-up before reading further: the architecture docs in this repo describe Catalyst's target state. Significant portions are not yet implemented — see
docs/IMPLEMENTATION-STATUS.mdfor what exists today vs what is design.
The model in 60 seconds
OpenOva (the company) publishes Catalyst (the platform).
A deployed Catalyst is called a Sovereign.
A Sovereign has:
- Organizations (multi-tenancy unit)
- Environments (org-scoped, env-typed: prod/stg/uat/dev/poc)
- Applications (installed Blueprints)
- Blueprints (the App Store catalog — public + Org-private)
Users install Applications from Blueprints into Environments.
Blueprints can depend on Blueprints (arbitrary depth).
Each Environment is one Gitea repo + one or more vclusters.
Every state change is a Git commit.
Every UI surface reads from a single CQRS projection.
Same code runs in every Sovereign:
- openova (run by us; SaaS Organizations)
- omantel (run by Omantel; SME Organizations across Oman)
- bankdhofar (run by the bank; internal Organizations)
- your-company (run by you, on infrastructure you choose)
See docs/GLOSSARY.md for every term, docs/ARCHITECTURE.md for the full picture.
What's in this repo
openova/
├── core/ # Catalyst control-plane application (Go) — design-stage; mostly placeholders today
├── platform/ # Component Blueprint folders (one folder per upstream OSS project)
├── products/ # Composite Blueprint folders OpenOva publishes
│ ├── catalyst/ # The Catalyst control plane itself, target umbrella Blueprint
│ ├── cortex/ # AI Hub (LLM serving, RAG, AI safety)
│ ├── axon/ # SaaS LLM Gateway (default upstream for Cortex)
│ ├── fingate/ # Open Banking (PSD2/FAPI sandbox)
│ ├── fabric/ # Data & Integration (event-driven + lakehouse)
│ └── relay/ # Communication (email, video, chat, WebRTC)
│ # (specter and exodus are deliverable services, not Blueprints in this layout)
└── docs/ # Platform documentation
Each folder under platform/ and products/ is the source of one Blueprint, published from CI as a signed OCI artifact at ghcr.io/openova-io/bp-<name>:<semver> (the bp- prefix is added to the OCI artifact name; folder names stay short). Per-folder isolation is provided at the OCI artifact layer, not the Git repo layer — this is a monorepo with per-Blueprint fan-out, not a meta-repo of separate Git repositories. See docs/BLUEPRINT-AUTHORING.md §2 for the folder layout contract.
Today, the 12-component bootstrap kit (cilium, cert-manager, flux, crossplane, sealed-secrets, spire, nats-jetstream, openbao, keycloak, gitea, powerdns + the bp-catalyst-platform umbrella under
products/catalyst/) ships with fullchart/+blueprint.yamlperdocs/IMPLEMENTATION-STATUS.md§7, plusproducts/axon/and theexternal-dnsleaf chart. The remaining 45 platform components and thecortex / fabric / fingate / relayproduct folders are design-stage — README only — until each lands its Blueprint manifest, chart, Compositions, and CI fan-out.
Stack at a glance
| Layer | Technology |
|---|---|
| Container runtime | k3s (k8s-conformant), containerd |
| CNI / Service Mesh | Cilium (eBPF mTLS, L7 policies, Gateway API) |
| GitOps | Flux (per-vcluster, lightweight) |
| Git | Gitea (per-Sovereign, hosts Blueprint mirror + per-Environment repos) |
| IaC for non-K8s | Crossplane (the only IaC; not user-facing) |
| Bootstrap IaC | OpenTofu (one-shot, archived after Phase 0) |
| Multi-tenancy | vcluster (one per Organization per host cluster) |
| Identity (workloads) | SPIFFE/SPIRE (5-min rotating SVIDs, mTLS everywhere) |
| Identity (users) | Keycloak (per-Org for SME, per-Sovereign for corporate) |
| Secrets | OpenBao (Apache 2.0; independent Raft per region, no stretched cluster) + External Secrets Operator |
| Event spine | NATS JetStream (Apache 2.0; pub/sub + KV; per-Org accounts) |
| TLS | cert-manager + Let's Encrypt or corporate CA |
| Policy | Kyverno |
| Supply chain | cosign (Sigstore), Syft + Grype SBOM, Trivy scans |
| Runtime security | Falco (eBPF) |
| Observability | OpenTelemetry → Grafana stack (Alloy + Loki + Mimir + Tempo) |
| WAF | Coraza (OWASP CRS) |
| DNS | PowerDNS authoritative per Sovereign zone + DNSSEC + lua-records (ifurlup, pickclosest); pool-domain-manager allocates pool subdomains and flips parent-zone NS via registrar adapters (Cloudflare / Namecheap / GoDaddy / OVH / Dynadot) — see docs/MULTI-REGION-DNS.md, docs/PLATFORM-POWERDNS.md |
| Backup | Velero (to SeaweedFS, which routes the cold tier to cloud archival S3) |
| Container registry | Harbor |
For the full component list and trends see docs/PLATFORM-TECH-STACK.md and docs/TECHNOLOGY-FORECAST-2027-2030.md.
Cloud providers
| Provider | Status |
|---|---|
| Hetzner Cloud | Available (most-tested path) |
| AWS / GCP / Azure | Crossplane providers available; full path coming |
| Oracle Cloud (OCI) | Crossplane provider available; full path coming |
| Huawei Cloud | Crossplane provider available; full path coming |
All providers reach Catalyst via the same Crossplane abstraction; Sovereign provisioning details per provider are in docs/SOVEREIGN-PROVISIONING.md.
Getting started
Try it (managed)
Visit marketplace.openova.io to install Applications on the openova Sovereign without any infrastructure setup. SaaS journey for SMEs and evaluations.
Run your own Sovereign
1. Provision via catalyst-provisioner.openova.io (managed bootstrap), OR
2. Self-host bp-catalyst-provisioner in your own infrastructure (air-gap path).
Then follow the procedure in docs/SOVEREIGN-PROVISIONING.md.
Build a Blueprint
See docs/BLUEPRINT-AUTHORING.md. A Blueprint is a folder under platform/<name>/ (or products/<name>/) in this monorepo containing blueprint.yaml + manifests (Helm chart or Kustomize base) + (optional) Crossplane Compositions. CI signs each folder's contents and publishes to OCI as ghcr.io/openova-io/bp-<name>:<semver>. Catalyst's blueprint-controller picks it up automatically. Org-private Blueprints follow the same shape inside per-Sovereign Gitea repos.
License
All Blueprints and the Catalyst control plane are open source. Each component carries its own upstream license (typically Apache 2.0, MPL 2.0, or BSD-3); see each component's LICENSE file.
OpenOva charges for support, managed operations, and expert services — never for access to code. See docs/BUSINESS-STRATEGY.md §10.
Contributing
PRs welcome. The contribution path for Blueprints (including Crossplane Compositions) is documented in docs/BLUEPRINT-AUTHORING.md §13. Issues and discussions on GitHub.
Cloud-native is the foundation. Catalyst is how you operate it.