| 
 | 1 | + | 
 | 2 | +# Fluss Helm Chart  | 
 | 3 | + | 
 | 4 | +This chart deploys an Apache Fluss cluster on Kubernetes, following Helm best practices.  | 
 | 5 | +It requires a Zookeeper ensemble to be running in the same Kubernetes cluster. In future releases, we may add support for an embedded Zookeeper cluster.  | 
 | 6 | + | 
 | 7 | + | 
 | 8 | +## Development environment   | 
 | 9 | + | 
 | 10 | +| component                                                                      | version |  | 
 | 11 | +| ------------------------------------------------------------------------------ | ------- |  | 
 | 12 | +| [Docker](https://docs.docker.com/)                                             | v28.3.2 |  | 
 | 13 | +| [Minikube](https://minikube.sigs.k8s.io/docs/)                                 | v1.36.0 |  | 
 | 14 | +| [Kubernetes](https://kubernetes.io)                                            | v1.25.3 |  | 
 | 15 | +| [Helm](https://helm.sh)                                                        | v3.18.6 |  | 
 | 16 | +| [Apache Fluss](https://fluss.apache.org/docs/)                                 | v0.8-SNAPSHOT  |  | 
 | 17 | + | 
 | 18 | + | 
 | 19 | +## Image requirements   | 
 | 20 | + | 
 | 21 | +A container image for Fluss is available on DockerHub as `fluss/fluss`. You can use it directly or build your own from this repo. To use your own image you need to build the project with [Maven](https://fluss.apache.org/community/dev/building/) and build it with Docker.  | 
 | 22 | + | 
 | 23 | +The Maven build will create all required artifacts in the `build-target` directory. You need to copy it into the `docker` directory. The Dockerfile in this directory will copy these artifacts into the image.  | 
 | 24 | + | 
 | 25 | +In minikube, you can use the local Docker daemon to build the image without pushing it to a registry:  | 
 | 26 | + | 
 | 27 | +```bash  | 
 | 28 | +eval $(minikube -p minikube docker-env)  | 
 | 29 | +docker build -t fluss/fluss:0.8-SNAPSHOT .  | 
 | 30 | +```  | 
 | 31 | + | 
 | 32 | +## Overview  | 
 | 33 | + | 
 | 34 | +It creates:  | 
 | 35 | +- 1x CoordinatorServer as a StatefulSet with a headless Service (stable per‑pod DNS)  | 
 | 36 | +- 3x TabletServers as a StatefulSet with a headless Service (stable per‑pod DNS)  | 
 | 37 | +- ConfigMap for server.yaml (CoordinatorServer and TabletServers) to override default Fluss configuration  | 
 | 38 | +- Optional PersistentVolumes for data directories  | 
 | 39 | + | 
 | 40 | +## Quick start  | 
 | 41 | + | 
 | 42 | +1) ZooKeeper deployment:  | 
 | 43 | + | 
 | 44 | +To start Zookeeper use Bitnami’s chart or your own deployment. If you have an existing Zookeeper cluster, you can skip this step. Example with Bitnami’s chart:  | 
 | 45 | + | 
 | 46 | +```bash  | 
 | 47 | +helm repo add bitnami https://charts.bitnami.com/bitnami  | 
 | 48 | +helm repo update  | 
 | 49 | +helm install zk bitnami/zookeeper \  | 
 | 50 | +  --set replicaCount=3 \  | 
 | 51 | +  --set auth.enabled=false \  | 
 | 52 | +  --set persistence.size=5Gi  | 
 | 53 | +```  | 
 | 54 | + | 
 | 55 | +2) Default (Zookeeper available in-cluster):  | 
 | 56 | + | 
 | 57 | +```bash  | 
 | 58 | +helm install fluss ./fluss-helm  | 
 | 59 | +```  | 
 | 60 | +With an optional namespace flag `--namespace <your-namespace>` if you want to install it in a specific namespace.  | 
 | 61 | + | 
 | 62 | +This assumes, that Zookeeper is reachable at `zk-zookeeper.<your-namespace>.svc.cluster.local:2181`. If your Zookeeper address is different, you can override it with:  | 
 | 63 | + | 
 | 64 | +```bash  | 
 | 65 | +helm install fluss ./fluss-helm \  | 
 | 66 | +  --set zookeeper.address=<your-zk-address>  | 
 | 67 | +```  | 
 | 68 | + | 
 | 69 | +## Configuration reference  | 
 | 70 | + | 
 | 71 | +Important Fluss options surfaced by the chart:  | 
 | 72 | +- zookeeper.address: CoordinatorServer and TabletServer point to your ZK ensemble.  | 
 | 73 | +- data.dir, remote.data.dir: Local persistent path for data; remote path for snapshots (OSS/HDFS). TabletServers default to a PVC mounted at data.dir.  | 
 | 74 | +- bind.listeners: Where the server actually binds.  | 
 | 75 | +- advertised.listeners: Externally advertised endpoints for clients and intra‑cluster communication. In K8s, advertise stable names.  | 
 | 76 | +- internal.listener.name: Which listener is used for internal communication (defaults to INTERNAL).  | 
 | 77 | +- tablet-server.id: Required to be unique per TabletServer. The chart auto‑derives this from the StatefulSet pod ordinal at runtime.  | 
 | 78 | + | 
 | 79 | + | 
 | 80 | +### Zookeeper and storage  | 
 | 81 | +- zookeeper.address must point to a reachable ensemble.  | 
 | 82 | +- data.dir defaults to /tmp/fluss/data; use a PVC if persistence.enabled=true.  | 
 | 83 | + | 
 | 84 | +## Resource management  | 
 | 85 | + | 
 | 86 | +Set resources with requests/limits as appropriate for production. There are no defaults to make it also run on environments with little resources such as Minikube.  | 
 | 87 | + | 
 | 88 | +## Troubleshooting  | 
 | 89 | +- Image pull errors:  | 
 | 90 | +  - If using a private registry, configure image.pullSecrets and ensure the image repository/tag are correct.  | 
 | 91 | +- Pods not ready: ensure ZooKeeper is reachable and ports 9123 are open.  | 
 | 92 | +- Connection failures: check advertised.listeners configuration and DNS resolution within the cluster by using kubectl exec to get a shell in a pod and test connectivity (using nc).  | 
0 commit comments