Installation
Prerequisites​
- Git
- Kubernetes cluster(version >= 1.18), and support CSI
- kubectl(version >= 1.18)
- Helm(version >= 3.5)
The following documents assume that you have installed all the above requirements.
For the installation and configuration of kubectl, please refer to here.
For the installation and configuration of Helm 3, please refer to here.
How to Deploy​
Install Fluid with Helm​
Create namespace:
$ kubectl create ns fluid-system
Add Fluid repository to Helm repos and keep it up-to-date
$ helm repo add fluid https://fluid-cloudnative.github.io/charts
$ helm repo update
Install Fluid with:
$ helm install fluid fluid/fluid
NAME: fluid
LAST DEPLOYED: Wed May 24 18:17:16 2023
NAMESPACE: default
STATUS: deployed
REVISION: 1
TEST SUITE: None
The command
helm install <RELEASE_NAME> <CHART>takes the release name (fluid) followed by the chart reference (fluid/fluid). The chart reference uses thefluidrepository added above.To install from a local archive, download a versioned
.tgzfrom Fluid Releases and pass its local path as<CHART>.
Upgrade Fluid to the latest version with Helm​
If you have installed an older version of Fluid before, you can use Helm to upgrade it. Before upgrading, it is recommended to ensure that all components in the AlluxioRuntime resource object have been started completely, which is similar to the following state:
$ kubectl get pod
NAME READY STATUS RESTARTS AGE
hbase-fuse-chscz 1/1 Running 0 9h
hbase-fuse-fmhr5 1/1 Running 0 9h
hbase-master-0 2/2 Running 0 9h
hbase-worker-bdbjg 2/2 Running 0 9h
hbase-worker-rznd5 2/2 Running 0 9h
upgrade fluid:
$ helm upgrade fluid fluid/fluid
Release "fluid" has been upgraded. Happy Helming!
NAME: fluid
LAST DEPLOYED: Fri Sep 2 18:54:18 2022
NAMESPACE: default
STATUS: deployed
REVISION: 2
TEST SUITE: None
We recommend you to update Fluid latest version from v0.7. If you have an older version, our suggestion is to reinstall it to ensure everything works fine.
Check Status of Component​
Check CRD used by Fluid:
$ kubectl get crd | grep data.fluid.io
alluxioruntimes.data.fluid.io 2023-05-24T10:14:47Z
databackups.data.fluid.io 2023-05-24T10:14:47Z
dataloads.data.fluid.io 2023-05-24T10:14:47Z
datamigrates.data.fluid.io 2023-05-24T10:28:11Z
datasets.data.fluid.io 2023-05-24T10:14:47Z
efcruntimes.data.fluid.io 2023-05-24T10:28:12Z
goosefsruntimes.data.fluid.io 2023-05-24T10:14:47Z
jindoruntimes.data.fluid.io 2023-05-24T10:14:48Z
juicefsruntimes.data.fluid.io 2023-05-24T10:14:48Z
thinruntimeprofiles.data.fluid.io 2023-05-24T10:28:16Z
thinruntimes.data.fluid.io 2023-05-24T10:28:16Z
Check the status of pods:
$ kubectl get pod -n fluid-system
NAME READY STATUS RESTARTS AGE
csi-nodeplugin-fluid-2scs9 2/2 Running 0 50s
csi-nodeplugin-fluid-7vflb 2/2 Running 0 20s
csi-nodeplugin-fluid-f9xfv 2/2 Running 0 33s
dataset-controller-686d9d9cd6-gk6m6 1/1 Running 0 50s
fluid-crds-upgrade-1.0.0-37e17c6-fp4mm 0/1 Completed 0 74s
fluid-webhook-5bc9dfb9d8-hdvhk 1/1 Running 0 50s
fluidapp-controller-6d4cbdcd88-z7l4c 1/1 Running 0 50s
If the Pod status is as shown above, then Fluid is installed on your Kubernetes cluster successfully!
Check version info of Component​
When csi-nodeplugin, alluxioruntime-controller and dataset-controller start,they will print their own version info into logs.
If you installed with the charts provided by us,their version info will be fully consistent.
If you installed manually, their version info may be not consistent. You can check it with the following command:
$ kubectl exec csi-nodeplugin-fluid-pq2zd -n fluid-system -c plugins fluid-csi version
$ kubectl exec alluxioruntime-controller-66bf8cbdf4-k6cxt -n fluid-system -- alluxioruntime-controller version
$ kubectl exec dataset-controller-558c5c7785-mtgfh -n fluid-system -- dataset-controller version
The output should be like:
BuildDate: 2024-03-02_07:35:18
GitCommit: 50ee8887239f07592ba74af3e14379efc1487c0c
GitTreeState: clean
GoVersion: go1.18.10
Compiler: gc
Platform: linux/amd64
Fluid use cases​
For more use cases about Fluid, please refer to our tutorials:
- Speed Up Accessing Remote Files
- Cache Co-locality for Workload Scheduling
- Accelerate Machine Learning Training with Fluid
Uninstall Fluid​
To uninstall fluid safely, we should check weather Custom Resource Objects about fluid have been deleted completely first:
kubectl get crds -o custom-columns=NAME:.metadata.name | grep data.fluid.io | sed ':t;N;s/\n/,/;b t' | xargs kubectl get --all-namespaces
Before uninstalling the release, run helm list --filter '^fluid$' to find its chart version in the CHART column (for example, 1.0.8 from fluid-1.0.8). Download that chart version from the Fluid Helm repository to get the CRD manifests.
FLUID_CHART_VERSION=1.0.8 # Replace with the installed chart version.
FLUID_CHART_DIR=$(mktemp -d)
helm pull fluid/fluid --version "$FLUID_CHART_VERSION" --untar --destination "$FLUID_CHART_DIR"
Once the download succeeds and all Fluid custom resources have been deleted, uninstall Fluid:
$ helm delete fluid
$ kubectl delete -f "$FLUID_CHART_DIR/fluid/crds"
$ kubectl delete ns fluid-system
The
fluidin commandhelm deletemeans the<RELEASE_NAME>during installation.
Advanced Configuration​
- In some cloud vendors, the default mount root directory
/runtime-mntis not writable, so you have to modify the directory location
helm install fluid fluid/fluid --set runtime.mountRoot=/var/lib/docker/runtime-mnt
- The feature
Fuse Recoveryis not enable by default, to enable this:
helm install fluid fluid/fluid --set csi.featureGates='FuseRecovery=true'
- If your Kubernetes cluster has a custom configured kubelet root directory, please configure the KUBELET_ROOTDIR when installing Fluid with the following command:
KUBELET_ROOTDIR=/data/kubelet # Replace with the node's kubelet root directory.
helm install fluid fluid/fluid \
--set "csi.kubelet.rootDir=${KUBELET_ROOTDIR}" \
--set "csi.kubelet.certDir=${KUBELET_ROOTDIR}/pki"
You can execute the following command on the Kubernetes node to view the --root-dir parameter configuration:
ps -ef | grep $(which kubelet) | grep root-dirIf the above command has no output, the kubelet root path is the default value (/var/lib/kubelet), which is the default value set by Fluid.
- When you install a Kubernetes cluster using Sealer, it by default uses
apiserver.cluster.localas the address of the API Server. At the same time, it writes this address to thekubelet.conffile and the corresponding IP address to thehostsfile. This will cause the Fluid CSI Plugin fail to find the IP address of the API Server. You can set the Fluid CSI Plugin to use hostNetwork via the following command:
# install
helm install fluid --set csi.config.hostNetwork=true fluid/fluid
# upgrade
helm upgrade fluid --set csi.config.hostNetwork=true fluid/fluid
- Under the default configuration, Fluid does not set
spec.resourcesfor the installed Controller Pods. If you are deploying Fluid in a production environment, it is recommended to configurespec.resourcesfor each Controller Pod based on your workload scale. For how to configure for different Controller Pods, you can refer to the examples in thevalues.yamlfile. For instance, if you'd like to modify the resource requests and limits for the Dataset Controller Pod, you should first define a custom values file (e.g., my-resource-file.yaml):
dataset:
resources:
requests:
cpu: 500m
memory: 256Mi
limits:
cpu: 1000m
memory: 512Mi
Override the default resources with the following command:
helm upgrade fluid --install -f my-resource-file.yaml fluid/fluid
- The CacheRuntime controller, which is required by
CacheRuntimeandCacheRuntimeClass(see the CacheRuntime Integration Guide), is not enabled by default. To enable it:
helm install fluid fluid/fluid --set runtime.cacheruntime.enabled=true