Add the fluid integration exmaple.

Signed-off-by: Ye Cao <[email protected]>

k8s/examples/fluid/Dockerfile

@@ -0,0 +1,3 @@
+FROM python:3.10
+
+COPY ./configure-vineyard-socket.py /configure-vineyard-socket.py

k8s/examples/fluid/Makefile

@@ -0,0 +1,2 @@
+build-image:
+	docker build -t configure-vineyard-socket:latest -f Dockerfile .

k8s/examples/fluid/README.md

@@ -0,0 +1,101 @@
+## Fluid integration
+
+If you are using [Fluid](https://fluidframework.com/) in your application, now it's a chance to cache your **Python Object** using **Vineyard** based on **Fluid**.
+
+### Prerequisites 
+
+- A kubernetes cluster with version >= 1.25.10. If you don't have one by hand, you can refer to the guide [Initialize Kubernetes Cluster](https://v6d.io/tutorials/kubernetes/using-vineyard-operator.html#step-0-optional-initialize-kubernetes-cluster) to create one.
+- Install the [Vineyardctl](https://v6d.io/notes/developers/build-from-source.html#install-vineyardctl) by following the official guide.
+
+### Install the argo server on Kubernetes
+
+1. Install the argo server on Kubernetes:
+
+```bash
+$ kubectl create namespace argo
+$ kubectl apply -n argo -f https://github.com/argoproj/argo-workflows/releases/download/v3.4.8/install.yaml
+```
+
+2. Check the status of the argo server:
+
+```bash
+$ kubectl get pod -n argo
+NAME                                  READY   STATUS    RESTARTS   AGE
+argo-server-7698c96655-xsd5k          1/1     Running   0          10m
+workflow-controller-b888f4458-ts58f   1/1     Running   0          10m
+```
+
+### Submit the job wihout Vineyard
+
+1. Submit the workflow:
+
+```bash
+$ cd k8s/examples/fluid
+$ argo submit --watch argo_workflow.yaml
+```
+
+### Submit the job with Vineyard
+
+
+1. Install the vineyard deployment:
+
+```bash
+$ vineyardctl deploy vineyard-deployment
+```
+
+2. Install the Fluid:
+
+```bash
+$ kubectl create ns fluid-system
+$ helm repo add fluid https://fluid-cloudnative.github.io/charts
+$ helm repo update
+$ helm install fluid fluid/fluid
+``` 
+
+3. Build the `configure-vineyard-socket` image and load to the cluster.
+
+```bash
+$ cd k8s/examples/fluid && make build-image && kind load docker-image configure-vineyard-socket
+```
+
+4. Install the Vineyard profile:
+
+```bash
+$ kubectl apply -f vineyard_profile.yaml
+```
+
+5. Install the Vineyard Dataset:
+
+```bash
+$ kubectl apply -f dataset.yaml
+```
+
+6. Check the dataset status and make sure the dataset is in `Bound` status as follows.
+
+```bash
+$ kubectl get dataset -A
+NAMESPACE   NAME       UFS TOTAL SIZE   CACHED   CACHE CAPACITY   CACHED PERCENTAGE   PHASE   AGE
+default     vineyard   [Calculating]    N/A      N/A              N/A                 Bound   105s
+```
+
+After that, the vineyard pv and pvc will be created automatically as follows. 
+
+```bash
+$ kubectl get pvc   
+NAME       STATUS   VOLUME             CAPACITY   ACCESS MODES   STORAGECLASS   AGE
+vineyard   Bound    default-vineyard   100Pi      RWX            fluid          87s
+$ kubectl get pv
+NAME               CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS   CLAIM              STORAGECLASS   REASON   AGE
+default-vineyard   100Pi      RWX            Retain           Bound    default/vineyard   fluid                   2m43s
+```
+
+7. Submit the workflow with Vineyard:
+
+```bash
+$ argo submit --watch argo_workflow_with_vineyard.yaml
+```
+
+### result
+
+The execution time of the workflow without Vineyard is 41s.
+

k8s/examples/fluid/argo_workflow.yaml

@@ -0,0 +1,97 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Workflow
+metadata:
+  generateName: mlops-
+spec:
+  entrypoint: dag
+  templates:
+  - name: producer
+    volumes:
+    - name: data
+      hostPath:
+        path: /data
+        type: DirectoryOrCreate
+    container:
+      image: python:3.10
+      volumeMounts:
+      - name: data
+        mountPath: /data
+      command: [bash, -c]
+      args:
+        - |
+          pip install numpy pandas;
+          cat << EOF >> producer.py
+          import time
+
+          import numpy as np
+          import pandas as pd
+
+          def generate_random_dataframe(num_rows):
+              return pd.DataFrame({
+                      'Id': np.random.randint(1, 100000, num_rows),
+                      'TotalRooms': np.random.randint(2, 11, num_rows),
+                      "GarageAge": np.random.randint(1, 31, num_rows),
+                      "RemodAge": np.random.randint(1, 31, num_rows),
+                      "HouseAge": np.random.randint(1, 31, num_rows),
+                      "TotalBath": np.random.randint(1, 5, num_rows),
+                      "TotalPorchSF": np.random.randint(1, 1001, num_rows),
+                      "TotalSF": np.random.randint(1000, 6001, num_rows),
+                      "TotalArea": np.random.randint(1000, 6001, num_rows),
+                      'MoSold': np.random.randint(1, 13, num_rows),
+                      'YrSold': np.random.randint(2006, 2022, num_rows),
+                      'SalePrice': np.random.randint(50000, 800001, num_rows),
+              })
+
+          def producer():
+              st = time.time()
+              print('Generating data....', flush=True)
+              df = generate_random_dataframe(1000000000)
+              ed = time.time()
+              print('##################################')
+              print('Generating data time: ', ed - st, flush=True)
+              st = time.time()
+              print('Serializing data....', flush=True)
+              df.to_pickle('/data/df.pkl')
+              ed = time.time()
+              print('##################################')
+              print('Serializing data time: ', ed - st, flush=True)
+          EOF
+          python producer.py;
+  - name: consumer
+    volumes:
+    - name: data
+      hostPath:
+        path: /data
+        type: DirectoryOrCreate
+    container:
+      image: python:3.10
+      volumeMounts:
+      - name: data
+        mountPath: /data
+      command: [bash, -c]
+      args:
+        - |
+          pip install pandas;
+          cat << EOF >> consumer.py
+          import time
+
+          import pandas as pd
+
+          def consumer():
+              st = time.time()
+              print('Deserializing data....', flush=True)
+              df = pd.read_pickle('/data/df.pkl')
+              ed = time.time()
+              print('##################################')
+              print('Deserializing data time: ', ed - st, flush=True)
+          EOF
+          python consumer.py;
+  - name: dag
+    dag:
+      tasks:
+      - name: producer
+        template: producer
+      - name: consumer
+        template: consumer
+        dependencies:
+          - producer

k8s/examples/fluid/argo_workflow_with_vineyard.yaml

@@ -0,0 +1,95 @@
+apiVersion: argoproj.io/v1alpha1
+kind: Workflow
+metadata:
+  generateName: mlops-
+spec:
+  entrypoint: dag
+  templates:
+  - name: producer
+    volumes:
+    - name: data
+      persistentVolumeClaim:
+        claimName: vineyard
+    container:
+      image: python:3.10
+      volumeMounts:
+      - name: data
+        mountPath: /data
+      command: [bash, -c]
+      args:
+        - |
+          pip install vineyard numpy pandas;
+          cat << EOF >> producer.py
+          import time
+
+          import numpy as np
+          import pandas as pd
+
+          def generate_random_dataframe(num_rows):
+              return pd.DataFrame({
+                      'Id': np.random.randint(1, 100000, num_rows),
+                      'TotalRooms': np.random.randint(2, 11, num_rows),
+                      "GarageAge": np.random.randint(1, 31, num_rows),
+                      "RemodAge": np.random.randint(1, 31, num_rows),
+                      "HouseAge": np.random.randint(1, 31, num_rows),
+                      "TotalBath": np.random.randint(1, 5, num_rows),
+                      "TotalPorchSF": np.random.randint(1, 1001, num_rows),
+                      "TotalSF": np.random.randint(1000, 6001, num_rows),
+                      "TotalArea": np.random.randint(1000, 6001, num_rows),
+                      'MoSold': np.random.randint(1, 13, num_rows),
+                      'YrSold': np.random.randint(2006, 2022, num_rows),
+                      'SalePrice': np.random.randint(50000, 800001, num_rows),
+              })
+
+          def producer()
+              st = time.time()
+              print('Generating data....', flush=True)
+              df = generate_random_dataframe(1000000000)
+              ed = time.time()
+              print('##################################')
+              print('Generating data time: ', ed - st, flush=True)
+              st = time.time()
+              print('Serializing data....', flush=True)
+              vineyard.csi.write(df, '/data/df.pkl')
+              ed = time.time()
+              print('##################################')
+              print('Serializing data time: ', ed - st, flush=True)
+          EOF
+          python producer.py;
+  - name: consumer
+    volumes:
+    - name: data
+      persistentVolumeClaim:
+        claimName: vineyard
+    container:
+      image: python:3.10
+      volumeMounts:
+      - name: data
+        mountPath: /data
+      command: [bash, -c]
+      args:
+        - |
+          pip install vineyard pandas;
+          cat << EOF >> consumer.py
+          import time
+
+          import vineyard
+
+          def consumer()
+              st = time.time()
+              print('Deserializing data....', flush=True)
+              df = vineyard.csi.read('/data/df.pkl')
+              ed = time.time()
+              print('##################################')
+              print('Deserializing data time: ', ed - st, flush=True)
+          EOF
+          python consumer.py;
+  - name: dag
+    dag:
+      tasks:
+      - name: producer
+        template: producer
+      - name: consumer
+        template: consumer
+        dependencies:
+          - producer

k8s/examples/fluid/configure-vineyard-socket.py

@@ -0,0 +1,32 @@
+import json
+
+with open("/etc/fluid/config.json", "r") as f:
+    lines = f.readlines()
+
+rawStr = lines[0]
+print(rawStr)
+
+
+script = """
+#!/bin/sh
+set -ex
+
+mkdir -p $targetPath
+while true; do
+    if [ ! -S "$targetPath/vineyard.sock" ]; then
+        mount --bind $socketPath $targetPath
+    fi
+    sleep 10
+done
+"""
+
+obj = json.loads(rawStr)
+
+with open("mount-vineyard-socket.sh", "w") as f:
+    f.write("targetPath=\"%s\"\n" % obj['targetPath'])
+    if obj['mounts'][0]['mountPoint'].startswith("local://"):
+      f.write("socketPath=\"%s\"\n" % obj['mounts'][0]['mountPoint'][len("local://"):])
+    else:
+      f.write("socketPath=\"%s\"\n" % obj['mounts'][0]['mountPoint'])
+
+    f.write(script)

k8s/examples/fluid/dataset.yaml

@@ -0,0 +1,18 @@
+apiVersion: data.fluid.io/v1alpha1
+kind: Dataset
+metadata:
+  name: vineyard
+spec:
+  mounts:
+  # This directory should be the same as the vineyard socket directory in the vineyard deployment
+  - mountPoint: local:///var/run/vineyard-kubernetes/vineyard-system/vineyardd-sample
+    name: vineyard
+  accessModes:
+  - ReadWriteMany
+---
+apiVersion: data.fluid.io/v1alpha1
+kind: ThinRuntime
+metadata:
+  name: vineyard
+spec:
+  profileName: vineyard-profile

k8s/examples/fluid/vineyard_profile.yaml

@@ -0,0 +1,23 @@
+apiVersion: data.fluid.io/v1alpha1
+kind: ThinRuntimeProfile
+metadata:
+  name: vineyard-profile
+spec:
+  fileSystemType: fuse
+  volumes:
+  - name: vineyard
+    hostPath:
+      # This path should be the same as the vineyard socket path in the vineyard deployment
+      path: /var/run/vineyard-kubernetes/vineyard-system/vineyardd-sample
+      type: DirectoryOrCreate
+  fuse:
+    image: configure-vineyard-socket
+    imageTag: latest
+    imagePullPolicy: IfNotPresent
+    volumeMounts:
+    - name: vineyard-socket
+      mountPath: /var/run/vineyard-kubernetes/vineyard-system/vineyardd-sample
+    command:
+    - sh
+    - -c
+    - "python3 /configure-vineyard-socket.py && chmod u+x ./mount-vineyard-socket.sh && ./mount-vineyard-socket.sh"