Activity

1. misohu

   added

   enhancementNew feature or requestNew feature or request

   on Jul 31, 2025
2. syncronize-issues-to-jira commented on Jul 31, 2025

   

   syncronize-issues-to-jirabot

   on Jul 31, 2025 – with Syncronize issues to Jira

   More actions

   Thank you for reporting your feedback to us!

   The internal ticket has been created: https://warthogs.atlassian.net/browse/KF-7809.

   This message was autogenerated
3. misohu commented on Aug 5, 2025

   

   misohu

   on Aug 5, 2025

   Last edited by misohu

   MemberAuthor

   More actions

   Investigation: quay.io/feastdev/feast-operator:0.49.0 Docker Image

   Docker Image

   The image in question is:

   quay.io/feastdev/feast-operator:0.49.0
   

   This image is built from the following Dockerfile in the Feast upstream repository:
   https://github.com/feast-dev/feast/blob/v0.49.0/infra/feast-operator/Dockerfile

   This Docker image contains the Feast Kubernetes Operator — a controller process that manages FeatureStore custom resources in Kubernetes and orchestrates the deployment of Feast services based on declarative specifications.

   How the Image Is Used (Feast Operator)

   The image is used as the core component of the Feast Operator, which runs as a Kubernetes controller and performs the following functions:

   • Watches for custom resources of type FeatureStore in the cluster.
   • Clones Feast project repositories or runs feast init to bootstrap feature stores.
   • Deploys associated services such as:
     • Feature server (online store)
     • Offline store
     • Registry
     • Feast UI (optional)
   • Handles configuration for persistence (PVCs, databases), authentication (RBAC/OIDC), and TLS.
   • Sets up materialization jobs via Kubernetes CronJobs.
   • Reconciles the state automatically when specs change.

   How the CRD Is Used

   The Feast Operator defines and manages a Kubernetes Custom Resource Definition (CRD) called FeatureStore.

   CRD source:
   https://github.com/feast-dev/feast/blob/v0.49.0/infra/feast-operator/bundle/manifests/feast.dev_featurestores.yaml

   This CRD allows users to declaratively define the configuration of an entire Feast feature store deployment in Kubernetes.

   A minimal example:

   apiVersion: feast.dev/v1alpha1
   kind: FeatureStore
   metadata:
     name: my-featurestore
   spec:
     feastProject: fraud_detection
     feastProjectDir:
       git:
         url: https://github.com/my-org/feast-repo.git
         ref: main
     services:
       onlineStore:
         store:
           type: redis
           secretRef:
             name: redis-secret
       offlineStore:
         store:
           type: bigquery
           secretRef:
             name: bq-secret

   Key CRD Sections

   Services
   Under spec.services, users can configure:

   • Feature server (online store)
   • Offline store
   • Registry
   • UI (optional)

   Each can be configured with its own persistence, environment, resource limits, TLS, etc.

   CronJob
   Under spec.cronJob, users can define periodic materialization jobs using Kubernetes CronJob syntax:

   cronJob:
     schedule: "0 * * * *"
     containerConfigs:
       commands: ["feast materialize-incremental ..."]

   Persistence
   Persistence is configurable for each component:

   • File-based with PVCs
   • Store-based with database integrations (Redis, BigQuery, Snowflake, etc.)

   Conclusion

   Although the feast-operator and its image (quay.io/feastdev/feast-operator:0.49.0) provide a Kubernetes-native, declarative, and automated way to manage Feast deployments, we cannot directly use this image in our current setup.

   Reasons:

   • Our implementation uses a custom feast_integrator and Feast UI, which do not conform to the upstream operator model.
   • The operator expects to own the deployment lifecycle and assumes responsibility for cloning repos, initializing projects, and managing services — which would conflict with our current architecture.

   Therefore, adopting the Feast Operator as-is would require significant changes to how we deploy and integrate Feast.

   Note: when designing our feast bundle this solution was not available.
4. misohu commented on Aug 6, 2025

   

   misohu

   on Aug 6, 2025

   MemberAuthor

   More actions

   Testing the Feast Operator with a custom image

   This document outlines the steps to test a custom Feast Operator image.

   1. Clone the Feast repository

   git clone [https://github.com/feast-dev/feast](https://github.com/feast-dev/feast)
   cd feast
   git checkout v0.49.0

   2. Replace the Feast Operator image

   Navigate to the feast/infra/feast-operator/ directory and replace the image in the bundle/manifests/feast-operator.clusterserviceversion.yaml file.

   This script replaces the default image with your desired image

   cd infra/feast-operator/
   IMAGE_TAG="<YOUR_IMAGE_TAG>" # e.g., my-registry/my-feast-operator:latest
   sed -i "s|quay.io/feastdev/feast-operator:0.49.0|${IMAGE_TAG}|g" bundle/manifests/feast-operator.clusterserviceversion.yaml
   

   3. Install Go and Kubernetes

   Make sure you have Go version 1.23 or later installed.
   You can follow the official instructions to install Go:
   https://go.dev/doc/install

   Ensure you have a local Kubernetes cluster running.

   4. Deploy the Feast Operator

   From the feast/infra/feast-operator/ directory, run the following commands to install the CRDs and deploy the operator:

   make install
   make deploy
   

   This will deploy the Feast CRDs and the operator to your Kubernetes cluster.

   5. Verify the Operator Pod

   The Feast Operator should now be running in the feast-operator-system namespace. You can verify this with the following command:

   kubectl get pods -n feast-operator-system
   You should see a pod similar to this:
   NAME                                                READY   STATUS    RESTARTS   AGE
   feast-operator-controller-manager-94fc86679-h59mz   1/1     Running   0          166m
   

   6. Run a sample configuration

   You can now apply one of the sample configurations to create a Feast feature store. For example:

   kubectl apply -f config/samples/v1alpha1_featurestore_git_repopath.yaml
   This should create a new pod in your current namespace, such as:
   feast-sample-git-repopath-7ff67bb5b5-6fjk2
   

   You can verify the pod status with:

   kubectl get pods
   
5. misohu commented on Aug 8, 2025

   

   misohu

   on Aug 8, 2025

   MemberAuthor

   More actions

   Updated Feast Operator Testing Instructions (Simplified Approach)

   After practical testing, the simplest and most effective way to test a custom Feast image is by deploying the upstream Feast operator without modifying the repo, and then manually updating the image using kubectl edit. This avoids the need to recompile or modify manifests manually.

   Steps

   1. Clone and checkout Feast repo (optional, for reference)

      git clone https://github.com/feast-dev/feast
      cd feast
      git checkout v0.49.0
      cd infra/feast-operator/

   2. Ensure prerequisites

      • Go 1.23 installed:

        wget https://go.dev/dl/go1.23.0.linux-amd64.tar.gz
        sudo rm -rf /usr/local/go && sudo tar -C /usr/local -xzf go1.23.0.linux-amd64.tar.gz
        export PATH=$PATH:/usr/local/go/bin
        go version

      • Local Kubernetes cluster (kind, minikube, etc.)

   3. Deploy the upstream Feast Operator

      make install
      make deploy

   4. Edit the deployment to use your custom image

      kubectl edit deploy feast-operator-controller-manager -n feast-operator-system

      In the opened editor, find and replace the line under spec.template.spec.containers.image:

      image: quay.io/feastdev/feast-operator:0.49.0

      with:

      image: <your-custom-image>

   5. Verify that the new image is in use

      kubectl get pods -n feast-operator-system-<id> -o wide

   6. Apply a sample config to trigger the operator

      kubectl apply -f config/samples/v1alpha1_featurestore_git_repopath.yaml

      You should now see a running pod like:

      kubectl get pods
      # feast-sample-git-repopath-xxxxx   1/1   Running   0   1m