Grafana runs on PicoPaaS, and we kept the isolation intact
We just got a real, recognizable app running on PicoPaaS: Grafana. It's the public grafana/grafana image (latest stable, 13.2.4), with no custom image build. All of its configuration is env vars, and it stores its data in the platform's managed MySQL.
A real app, no special-casing
Dashboards and settings persist across restarts. We also upgraded it across a major version, 11 to 13, with no downtime. The whole setup is a recipe you can follow: platform.picopaas.xyz/guide/recipes.
What the platform grew along the way
Admins can now set an app's replica count, and the logs, metrics and health views are multi-pod aware: per-pod log sections, and usage measured against the combined capacity of all pods.
Where security drew the line, on purpose
We tried Grafana at two replicas with its clustered alerting, which needs the pods to talk to each other. Our tenant isolation kept them apart, exactly as it's built to. We could have loosened that boundary to make the demo work. We chose not to.
That's a deliberate design decision: multi-replica on PicoPaaS is stateless-only, and clustered or HA apps that need pods to gossip with each other aren't supported, by design. We'd rather hold the isolation boundary firm than bend it for a feature. Grafana runs as a single replica here, and that's the intended shape, not a gap we're working around.
We also ran a fresh security sweep, which came back clean, and we narrowed one internal permission so that if something were ever compromised, the blast radius would be smaller.
The pattern we want to keep: features ship inside the security boundary, not around it.