CS598 CCC Capstone Project - Machine-Learning-Function-Chaining

Prerequisites - Get Kubernetes Cluster Running

Please check the README.md in the /cluster-resources to follow directions to set up the OpenFaaS system to invoke the functions below.

Function Definitions for FaaS Function Chaining

Here we define 2 functions that will be chained in a Function as a Service platform. The functions will be deployed using OpenFaaS.

Each function has its own folder that contains:

• handler.py: the function code
• requirements.txt: any required dependencies

Each function also has a yaml file that describes a Docker container. To build the function, run:

faas-cli build -f ./<function-name>.yml

Function 1

The first function does image processing, which is to resize an input image to a specified size. The images used here are from ImageNet-1k datasets. Some of the images are stored in an S3 bucket cs598-openfaas. The function will receive an input of filename of an image stored in s3://cs598-openfaas/datasets/, which will be used to download the file from S3. Then the function resize the image to max size (100,100). The edited image is then uploaded to the S3 bucket s3://cs598-openfaas/function-1/

• Input: file name of the image stored in S3 bucket

How to test the function:

1. Open OpenFaaS dashboard. Go to function-1
2. In the Invoke Function section, type in the filename of the image (s3://cs598-openfaas/datasets/) in the Request body as Text. For example: ILSVRC2012_test_00000002.JPEG
3. Click on Invoke button.
4. The output image will be stored in s3://cs598-openfaas/function-1/

Function 2

The second function runs an image classification on the input image based on the ResNet50 model. Similar to function-1, this function receives an input of filename of the edited image from function-1, which is stored in s3://cs598-openfaas/function-1/. The function downloads the image from the bucket, then runs the image classification model on it. The result is stored in s3://cs598-openfaas/function-2/ as a text file.

• Input: updated image filename from S3.

How to test the function:

1. Open OpenFaaS dashboard. Go to function-2
2. In the Invoke Function section, type in the filename of the image (from s3://cs598-openfaas/function-1/) in the Request body as Text. For example: edited_ILSVRC2012_test_00000002.JPEG
3. Click on Invoke button.
4. The output text file containing the predicted label will be stored in s3://cs598-openfaas/function-2/

S3 bucket Auth

1.User name Password Console sign-in URL cs598-team-jaz-console cs598-team-jaz-console https://416999946347.signin.aws.amazon.com/console

Running Experiment

python3 experiment/performance-test.py

Note: Change the url and callback url between using function-async and async

You can also change function-1 to use function-1-flask to gather the custom image metrics