Skip to main content

Install the stack on Kind

Create the expected cluster:

kind create cluster --name jafra
kubectl config use-context kind-jafra

Install cert-manager and the complete stack:

./install-jafra.sh --install-cert-manager

For later iterations:

./install-jafra.sh                 # build only missing images
./install-jafra.sh --force-build   # rebuild every component
./install-jafra.sh --deploy-only   # apply manifests without image work

Set KIND_CLUSTER if the cluster has another name. Keep JAFRA_VERSION=0.0.1 for this release.

Deployment sequence​

The installer deliberately orders resources:

  1. verify Docker, kubectl, Kind, and the selected cluster;
  2. install or require cert-manager;
  3. create jafra-system, RBAC, and the controller certificate;
  4. wait for the certificate;
  5. deploy and wait for the controller before registering the webhook;
  6. deploy the analyzer and its 5 GiB PVC;
  7. deploy the agent DaemonSet in log-only;
  8. change the agent to JAFRA_MODE=grpc and wait for restart.

This ordering avoids registering a failing TLS webhook and prevents the agent from streaming before the analyzer is ready.

Published-image installer caveat​

The v0.0.1 pull-jafra.sh pulls images named quay.io/causa-ai-hub/jafra-*:0.0.1, while the Kubernetes manifests refer to quay.io/bharathappali/jafra-*:0.0.1. The script does not rewrite the three manifest image references.

Consequently, loading the pulled tags does not guarantee that Kind uses them; the cluster may try to pull the different manifest names. Prefer install-jafra.sh with the release source and manifest image names. If using published images, retag all three to the manifest names before kind load, or update all three workload image references consistently.

Verify installation​

kubectl get pods -n jafra-system -o wide
kubectl rollout status deployment/jafra-controller -n jafra-system
kubectl rollout status deployment/jafra-analyzer -n jafra-system
kubectl rollout status daemonset/jafra-agent -n jafra-system
kubectl get mutatingwebhookconfiguration jafra-controller
kubectl get certificate -n jafra-system

The controller and analyzer should each have one ready Pod. The agent should have one ready Pod per eligible Linux node.