Skip to main content
Version: v1.0

Cache Runtime Manually Scaling

Prerequisites​

  • Fluid (version >= 0.5.0)

Please refer to the Fluid installation to complete the installation

Set Up Workspace​

$ mkdir <any-path>/dataset_scale
$ cd <any-path>/dataset_scale

Running​

Create Dataset and AlluxioRuntime resource objects

$ cat << EOF > dataset.yaml
apiVersion: data.fluid.io/v1alpha1
kind: Dataset
metadata:
name: hbase
spec:
mounts:
- mountPoint: https://mirrors.tuna.tsinghua.edu.cn/apache/hbase/stable/
name: hbase
---
apiVersion: data.fluid.io/v1alpha1
kind: AlluxioRuntime
metadata:
name: hbase
spec:
replicas: 1
tieredstore:
levels:
- mediumtype: MEM
path: /dev/shm
quota: 2Gi
high: "0.95"
low: "0.7"
EOF

In the above example, we set AlluxioRuntime.spec.replicas to 1, which means we will start an Alluxio cluster with one worker to cache the data in the dataset.

$ kubectl create -f dataset.yaml
dataset.data.fluid.io/hbase created
alluxioruntime.data.fluid.io/hbase created

After the Alluxio cluster has started normally, we can see that the Dataset and AlluxioRuntime created before are in the following state:

Status of the Alluxio components:

$ kubectl get pod
NAME READY STATUS RESTARTS AGE
hbase-master-0 2/2 Running 0 3m50s
hbase-worker-0 2/2 Running 0 3m15s

Status of the Dataset:

$ kubectl get dataset hbase
NAME UFS TOTAL SIZE CACHED CACHE CAPACITY CACHED PERCENTAGE PHASE AGE
hbase 544.77MiB 0.00B 2.00GiB 0.0% Bound 3m28s

Status of the AlluxioRuntime:

$ kubectl get alluxioruntime hbase -o wide
NAME READY MASTERS DESIRED MASTERS MASTER PHASE READY WORKERS DESIRED WORKERS WORKER PHASE READY FUSES DESIRED FUSES FUSE PHASE AGE
hbase 1 1 Ready 1 1 Ready 0 0 Ready 4m55s

Scale-out Cache Runtime

$ kubectl scale alluxioruntime hbase --replicas=2
alluxioruntime.data.fluid.io/hbase scaled

Directly use the kubectl scale command to complete the scale-out of Cache Runtime. After successfully executing the above command and waiting for a while, you can see that the status of both Dataset and AlluxioRuntime has changed.

A new Alluxio Worker and the corresponding Alluxio Fuse component have been successfully started:

$ kubectl get pod
NAME READY STATUS RESTARTS AGE
hbase-master-0 2/2 Running 0 13m
hbase-worker-1 2/2 Running 0 6m49s
hbase-worker-0 2/2 Running 0 13m

The Cache Capacity in the Dataset changes from 2.00GiB to 4.00GiB, indicating an increase in the available cache capacity of the Dataset:

$ kubectl get dataset hbase
NAME UFS TOTAL SIZE CACHED CACHE CAPACITY CACHED PERCENTAGE PHASE AGE
hbase 544.77MiB 0.00B 4.00GiB 0.0% Bound 15m
$ kubectl get alluxioruntime hbase -o wide
NAME READY MASTERS DESIRED MASTERS MASTER PHASE READY WORKERS DESIRED WORKERS WORKER PHASE READY FUSES DESIRED FUSES FUSE PHASE AGE
hbase 1 1 Ready 2 2 Ready 0 0 Ready 17m

Check the AlluxioRuntime description for the latest scaling information:

$ kubectl describe alluxioruntime hbase
...
Conditions:
...
Last Probe Time: 2021-04-23T07:54:03Z
Last Transition Time: 2021-04-23T07:54:03Z
Message: The workers are scale out.
Reason: Workers scaled out
Status: True
Type: WorkersScaledOut
...
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal Succeed 2m2s AlluxioRuntime Runtime scaled out. current replicas: 2, desired replicas: 2.

Scale-in Cache Runtime

Similar to scale-out, the number of workers in the Runtime can also scale-in using kubectl scale.

$ kubectl scale alluxioruntime hbase --replicas=1
alluxioruntime.data.fluid.io/hbase scaled

After successful execution of the above command, if no application is trying to access the dataset currently, then the Cache Runtime scale-in will be triggered.

Runtime Workers that exceed the specified number of replicas will be stopped:

NAME                 READY   STATUS        RESTARTS   AGE
hbase-master-0 2/2 Running 0 22m
hbase-worker-1 2/2 Terminating 0 17m32s
hbase-worker-0 2/2 Running 0 21m

Dataset's cache capacity (Cache Capacity) is returned to 2.00GiB:

$ kubectl get dataset hbase
NAME UFS TOTAL SIZE CACHED CACHE CAPACITY CACHED PERCENTAGE PHASE AGE
hbase 544.77MiB 0.00B 2.00GiB 0.0% Bound 30m

Note: In the current version of Fluid, there is a delay of a few minutes in the change of the Cache Capacity property when scale-in, so you may not be able to observe the change of this property quickly.

The Ready Workers and Ready Fuses fields in AlluxioRuntime also become 1:

$ kubectl get alluxioruntime hbase -o wide
NAME READY MASTERS DESIRED MASTERS MASTER PHASE READY WORKERS DESIRED WORKERS WORKER PHASE READY FUSES DESIRED FUSES FUSE PHASE AGE
hbase 1 1 Ready 1 1 Ready 0 0 Ready 30m

Check the AlluxioRuntime description for the latest scaling information:

$ kubectl describe alluxioruntime hbase
...
Conditions:
...
Last Probe Time: 2021-04-23T08:00:55Z
Last Transition Time: 2021-04-23T08:00:55Z
Message: The workers scaled in.
Reason: Workers scaled in
Status: True
Type: WorkersScaledIn
...
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal Succeed 6m56s AlluxioRuntime Alluxio runtime scaled out. current replicas: 2, desired replicas: 2.
Normal Succeed 4s AlluxioRuntime Alluxio runtime scaled in. current replicas: 1, desired replicas: 1.

The scaling capability provided by Fluid helps users or cluster administrators to adjust the resources occupied by the dataset cache in a timely manner, reducing the cache capacity of an infrequently used dataset (scale-in) or increasing the cache capacity of a dataset on demand (scale-out) to achieve a more fine-grained resource allocation and improve resource utilization.

Clean your environment​

$ kubectl delete -f dataset.yaml