Getting Started

Based on the Fluid documentation for the v1.0.x line. Commands assume a working Kubernetes cluster and Helm 3.

Prerequisites

• A Kubernetes cluster with at least one schedulable node.
• kubectl configured against that cluster.
• Helm 3 installed locally.
• A remote data source to mount (an object store bucket, HDFS, or a public dataset URL).

Install

Add the chart repository and install Fluid into its own namespace. This deploys the CRDs and the controllers, CSI driver, and webhook.

helm repo add fluid https://fluid-cloudnative.github.io/charts
helm repo update
helm install fluid fluid/fluid -n fluid-system --create-namespace

A first working setup

The core job is: declare a dataset, give it a runtime, and have an application Pod read it through a cache.

1. Create a Dataset pointing at your under-file-system mount.

   apiVersion: data.fluid.io/v1alpha1
   kind: Dataset
   metadata:
     name: demo
   spec:
     mounts:
       - mountPoint: https://mirrors.bit.edu.cn/apache/spark/
         name: spark

2. Create a same-named AlluxioRuntime. The matching name is what binds them.

   apiVersion: data.fluid.io/v1alpha1
   kind: AlluxioRuntime
   metadata:
     name: demo
   spec:
     replicas: 1
     tieredstore:
       levels:
         - mediumtype: MEM
           path: /dev/shm
           quota: 2Gi

3. Apply both and wait for the dataset to bind.

   kubectl apply -f dataset.yaml -f runtime.yaml
   kubectl get dataset demo

When the runtime controller finishes setup, the dataset reaches the Bound phase and a PersistentVolumeClaim named demo appears. Mount that PVC in an application Pod to read the data through the cache.

Verify it works

kubectl get dataset demo -o jsonpath='{.status.phase}'
kubectl get pvc demo
kubectl get pods -n fluid-system

A healthy setup shows the dataset phase as Bound, a bound demo PVC, and the Fluid controllers, CSI driver, and webhook Pods running in fluid-system. Optionally create a DataLoad resource to prefetch data into the cache before your job runs.

Where to go next

See the Fluid documentation for tiered store tuning, the other runtimes (JuiceFS, JindoCache, ThinRuntime), data operations (DataLoad, DataMigrate, DataBackup), and production concerns such as HA and metrics. The repository's samples/ directory has runnable examples per engine.