eks-ml

AWS summit seoul 2019 - How to scale your ml job with Amazon EKS

• SlideShare: https://pt.slideshare.net/awskorea/amazon-eks-lg-aws-summit-seoul-2019
• Video: https://www.youtube.com/watch?v=egv2TlfLL1Y&list=PLORxAVAC5fUWyB6Hsk9ibYJHw97k1h6s9&index=46

Scripts description

• setup.sh: setup script before demo
• train.py: model script to train
• train-oom.py: model script to demonstrate OOM error
• example.yaml: example of kubernetes job
• oom.yaml: example of OOM killed job
• experiments.yaml: list of experiment to train
• run-experiments.py: python code to run experiments
• Dockerfile: file to make docker image hongkunyoo/eks-ml
• Dockerfile.oom: file to make OOM training image hongkunyoo/eks-ml:oom

---

1. Configuration

Install awscli & configure ACCESS_KEY and SECRET_KEY. The user should have following permissions.

• EKS full access
• CloudFormation full access
• EC2 full access
• IAM full access

these permissions are quite administrative. please use it at your own risk.

2. Setup

Run setup.sh as sudo user.

sudo ./setup.sh

3. Hello world

Apply first example of job

kubectl apply -f example.yaml

4. Run all

Run all experiments

python run-experiments.py