Roadmap
Items are ordered by value per effort for this cluster first, and for other users second. Each lists where the work goes, so any item can be picked up as a project.
Near term
| # |
Item |
Why |
Where |
| 1 |
Platform metrics in Grafana: scrape METRICS_ADDR (uptime metrics are already exported there), add build metrics, and a dashboard |
see builds slowing or queues growing before users do |
a VMServiceScrape, a metrics port on the Service, pipeline, platform |
| 2 |
Preview environments per pull request (pr-42.app.joserod.space) |
try changes before merging |
platform (PR webhooks), render (name suffix), cleanup on close |
Medium term
| Item |
Why |
| Off-site backups: copy Garage's backup bucket to a cloud bucket (Cloudflare R2, Backblaze B2) |
the backups survive losing the whole cluster |
| Canary and A/B releases with ingress-nginx canary weights |
ship to 10% first; automatic rollback on errors |
| Plugin contract: a container that takes JSON in and returns a JSON verdict, at hooks pre-deploy, post-deploy and analysis |
k6 performance gates, chaos experiments (Chaos Mesh), AI checks, all without changing the core |
AI generator behind generate.Generator |
better rendimiento.yaml and Dockerfiles for unusual repos |
| Secrets from an external store (SOPS in git, Vault, 1Password) |
secrets become reproducible and auditable |
| Roles and audit log |
more than one person can use it safely |
HA platform (leader election, LISTEN/NOTIFY events) |
no downtime on node loss; see Scaling |
Scheduled jobs (cron: in the spec, rendered as CronJobs) |
replaces the last hand-written CronJobs |
Long term
| Item |
Why |
| Multi-cluster agents and cloud targets (EKS, GKE, bare metal) |
deploy the same app to the Pis and to the cloud; see Scaling |
| Environments and promotion (dev → staging → prod by digest) |
the same image, promoted once it's proven |
| Cluster provisioning as a plugin (Terraform or Crossplane) |
a new cluster in one click |
| Installable by anyone (Helm chart, multi-arch images, docs site) |
see Open source |
Done
- Repo → live HTTPS URL; GitHub App, onboarding pull requests, detection and templates
- CI on a BuildKit pool across the workers with a registry cache; Railpack Dockerfile-less builds
- The App controller: adoption, pruning, self-healing, rollback, DNS, GPU workloads, LAN IPs
needs: for Postgres, Redis and existing services- An add-on engine (Helm and kustomize from git), with previews, safety gates and Helm hooks, which replaced ArgoCD
- Renovate as an add-on, which replaced the CronJob
- ArgoCD and Jenkins uninstalled (2026-09-28)
- Nightly backups for every volume worth keeping: rendimiento labels the volumes it creates, and the Environment page warns about any volume in use without a recent backup (2026-10-06)
- The book in Spanish, page by page, in the same Cotija style (2026-10-06)
- A new look: Cotija, pueblo mágico. By day whitewashed walls, roof tiles, papel picado and the grana guardapolvo; by night an añil sky, faroles and alebrije colors. Every app is a cempasúchil that withers in an outage and revives when it comes back; petals fall when a release goes live. Movement can be switched off and follows the device's reduce-motion setting (2026-10-06)
- Mexican Spanish: the UI with an EN/ES switch, the platform's messages translated when shown (stored ones too), and emails in the language of
NOTIFY_LANG (2026-10-05)
- Problems resolve themselves when rendimiento sees the fix: a later push accepted, a later email sent (2026-10-05)
- The Problems page: the platform's own warnings and errors, grouped, with a count in the top bar and a banner on the app concerned; an invalid
rendimiento.yaml is a failed run, a red check on GitHub and an email (2026-10-05)
- Apps' databases moved onto
needs: [postgres]: podoi, wellness and stockpulse, each copied with row counts checked, overnight with no outage (2026-10-05)
- Delivery stats on the dashboard: deploys, lead time, change failure rate and time to recover for the last 30 days, with trends against the 30 before and twelve-week trend lines (2026-10-04)
- Object storage moved from MinIO, whose images are no longer free to pull, to Garage: Longhorn's backups and the log archive (2026-10-04)
- Pre-deploy tasks: migrations before the rollout; a failure stops the release (2026-10-02)
- Post-deploy tasks: smoke tests and migrations against the live release, in its namespace; a failure rolls it back (2026-10-02)
- Step logs archived to MinIO (gzip) and deleted after 365 days by a lifecycle rule (2026-10-01)
- Email notifications for failed builds, rollbacks, outages and recoveries (2026-10-01)
- Release verification with automatic rollback: each release is watched for 5 minutes and rolled back to the last good one if it breaks a service (2026-10-01)
- Uptime checks and the Reliability tab: every service checked once a minute, inside the cluster and publicly; uptime, response times and outages charted with release markers (2026-10-01)
- Tasks: commands as CI steps with secrets, for Expo/EAS builds and smoke tests (2026-09-29)
- The Services catalog, Environment checks, dynamic DNS
- Remote tests (
make test-remote), and this book