ragleap-ops

version — loading live from PyPI…

Kubernetes deployment manifests for RagLeap Core — Deployment, Service, PVC and Secret manifests translated from the real docker-compose.yml, live-tested on a kind cluster.

pip install ragleap-ops

Highlights

• Field-by-field translation from the real docker-compose.yml — no invented values
• All 4 pods reached 1/1 Running with zero restarts on a real kind cluster test, run with the earlier image (before 0.6.0). With the default image since 0.6.0, ghcr.io/ragleap/ragleap-app:0.11.0, the app and voice pods have not been run against the chart: 0.6.1 fixed a bug that stopped their containers from being created (runAsNonRoot without a numeric user), and whether they then became Ready was not verified at 0.6.1; the 0.6.2 line below records a live check
• Since 0.6.2 (from its changelog): on kind (a GitHub runner), with chart 0.6.1 and the image overridden by --set, both the app and voice pods became Ready, and the library versions read from inside the running container were pillow=13.0.0 pypdf=6.21.1. The default image is now ghcr.io/ragleap/ragleap-app:0.11.1 (pinned; 0.11.1 is 0.11.0 plus a requirements.lock change only). Readiness is a TCP check on the port, and application behaviour with Pillow 13 and pypdf 6.21.1 was not tested
• Known limitations (from the changelog): the chart does not create the Secret ragleap-app-env that both pods read, so create it yourself or the pods stay in CreateContainerConfigError; and the ragleap-app package was private on 2026-10-20 and still was on 2026-10-25, so the pull secret must have read access to it or the pods stay in ImagePullBackOff
• Sealed Secrets guidance for production (no plaintext secrets in git)

Tech stack

Core dependencies: (no Python runtime dependencies — raw manifests and a Helm chart)

Testing

1 test function across 1 test file, verified directly from the repo.

Like ragleap-observability, this is infrastructure manifests rather than a unit-tested library. Confidence comes from a documented live kind-cluster test (all pods Running, 0 restarts) with the earlier image, not a large pytest suite. The 0.6.0 image change and the 0.6.1 fix were live-tested on kind for the 0.6.2 release (both pods Ready; readiness is a TCP check). For 0.6.0 itself the changelog recorded only that the Helm render differs from 0.5.2 in the two image lines and that helm lint passes.

Quickstart

kubectl apply -f k8s/secrets.yaml
kubectl apply -f k8s/pvcs.yaml
kubectl apply -f k8s/db-deployment.yaml
kubectl apply -f k8s/app-deployment.yaml