Sprint 1 Complete: Infrastructure Fixes on the Production Cluster
slidev PVC unstuck. SOGo migrated. Storage classes audited across 31 deployments. All smoke tests green.
On June 14, we published the production cluster sprint plan — four sprints to close the gap on six missing services and harden the infrastructure. Today we're reporting that Sprint 1 is complete, and work on Sprint 2 is already under way.
What Was Fixed
slidev PVC: Unstuck
The slidev PersistentVolumeClaim had been stuck in Pending state since initial deployment. The root cause was a missing StorageClass annotation on the PVC — it referenced ceph-rbd-ssd but the spec.storageClassName field was empty, causing the CSI provisioner to skip it entirely.
Fix applied:
- Deleted the stuck PVC (no data loss — slidev is stateless)
- Recreated with explicit
storageClassName: ceph-rbd-ssd - Pod scheduled immediately after PVC bound
- Service responds at
slides.opendesk-edu.org
SOGo Namespace Migration
SOGo was deployed in the demo namespace during initial testing — a leftover from the prototyping phase. We migrated it to opendesk-edu to align with cluster naming conventions.
Migration steps:
- Dumped the SOGo PostgreSQL database from the
demonamespace - Re-deployed the SOGo chart with
namespace: opendesk-edu - Restored the database into the new namespace
- Updated the ingress to point to the new service endpoint
- Removed the old
demonamespace deployment
DNS propagation took ~2 minutes; the service has been stable since.
Storage Class Audit
Every PVC across the 31 deployments in both opendesk and opendesk-edu namespaces was audited for correct StorageClass assignment:
| StorageClass | Purpose | PVCs | Status |
|---|---|---|---|
ceph-rbd-ssd |
Fast RWO (databases, stateful sets) | 18 | ✅ All correct |
ceph-cephfs-hdd-ec |
Slow RWX erasure-coded (files) | 9 | ✅ All correct |
local-path |
Temporary/scratch (K3s default) | 3 | ✅ All correct |
No misconfigurations were found beyond the slidev PVC — confirming the initial deployment templates were sound.
Smoke Tests
A full smoke test suite was run against all 31 deployments:
- All 31 deployments show
READYstatus - All ingresses return HTTP 200 or 302 (SSO redirect)
- Keycloak OIDC login flow works for Nextcloud, OpenProject, XWiki, and BookStack
- K8up backup schedule ran successfully: 33 snapshots pushed to S3
- Prometheus targets all healthy, Grafana dashboards show no critical alerts
Sprint 2: Now Under Way
With infrastructure fixes complete, Sprint 2 targets three missing services:
| Service | Chart Type | Dependencies | Status |
|---|---|---|---|
| Zammad | Local chart | PostgreSQL, Elasticsearch, Redis | 🟡 In progress |
| Overleaf | Upstream chart | Redis, MongoDB | 🟡 In progress |
| KasmVNC | Upstream chart | — | ⏳ Queued |
Zammad deployment is the most complex — it requires three dependent databases and proper ingress routing for its real-time ticket updates. We're adapting the chart from charts-upgrade-v1.12.0/zammad with the same postgresql-subchart-removal pattern used for Etherpad.
What's Next
- Sprint 3 (target: this week): Portal entries (configmap-only chart) and Snipr (Dockerfile build required)
- Sprint 4: Hardening, documentation, and full ingress audit
The production cluster now runs 41 pods across 32 deployments — up from 39 pods and 31 deployments on June 14. We're on track for full production readiness by end of June.
Want to deploy openDesk Edu at your university? See our getting started guide and repository.
