search

sebranchett / ec2-instance-python

AWS CDK Python example of an EC2 that can be accessed by AWS System Manager and has access to S3 bucket. Configuration shell script can be provided to run on creation
Apache License 2.0
0 stars 1 forks source link
aws cdk ec2-instance python s3-storage shell-script system-manager

readme

Create EC2 Instance in new VPC with Systems Manager enabled

Based on the AWS example here: https://github.com/aws-samples/aws-cdk-examples/tree/master/python/ec2/instance

This adaptation was originally intended for AI researchers who require bare-bones access to GPUs and includes:

EC2 with access from System Manager and to S3

Three alternatives to consider:

WARNING: This architecture uses a public IP for the EC2 instance to access an S3 bucket. If you have sensitive data, please use a VPC endpoint instead. Also, enable encryption for EC2 storage volumes and consider more restrictive assumed roles.

Setting up the environment

This example uses the Python version of AWS CDK (Cloud Development Kit). CDK allows you to create and use infrastructure in the AWS cloud from the command line of your local machine. If you are using Linux or MacOS, you can use a regular terminal window. If you are using a Windows machine, you can use Git Bash (https://gitforwindows.org/) or WSL (https://docs.microsoft.com/en-us/windows/wsl/install).

To set up CDK, read this first: https://docs.aws.amazon.com/cdk/v2/guide/getting_started.html

and then this: https://docs.aws.amazon.com/cdk/v2/guide/work-with.html#work-with-prerequisites

To access your AWS account from the command line, set up Single Sign on: https://tu-delft-ict-innovation.github.io/Cloud4Research/faq/awscli.html#2-directly-from-your-laptop-or-desktop-computer

If you wish to access your EC2 instance with System Manager from the command line (instead of the AWS Console), then read this: https://docs.aws.amazon.com/systems-manager/latest/userguide/session-manager-working-with-install-plugin.html

With everything set up, you can log into your AWS account using Single Sign On (SSO):

aws [--profile=optionalProfileName] sso login

Using this example

Now you are ready to clone this git repository and step into it.

It is advisable to work in a Python virtual environment.

python -m venv .venv  # to create environment in hidden directory .venv/

If you are working on Windows (with Git Bash), activate the virtual environment with:

source .venv/Scripts/activate

otherwise:

source .venv/bin/activate

You will need to activate this environment every time you open a new shell.

Now that you are in the environment, install the requirements (only needs to be done once):

pip install -r requirements.txt

You can use pytest to check that everything is set up properly.

This example connects to the S3 bucket s3://tudelft-results-of-calculations. You do not have access to this bucket, so you will need to create a bucket of your own and change the name in startup.sh and in app.py. It's good practice to restrict bucket access to only those who need it. You will be charged for data downloads. It's also good practice to encrypt the data in a bucket. Don't forget to tick the 'Enable' box.

See the Useful Commands below to bootstrap your AWS CDK environment, synthesize a CloudFormation template from the Python code and deploy the template to your AWS environment.

Now experiment!

You can specify the EC2 instance type, the AMI and the S3 bucket for input/output data in config.yaml.

For the AMI, you can either specify an AMI alias, or you can specify the AMI name and the owner of the AMI. If you are using an AWS Marketplace AMI, you can find the AMI id and alias by clicking 'Continue to Subscribe' and then 'Continue to Configuration'. Make sure you select the correct region.

You can specify the commands that will be run on startup in startup.sh. In the file provided, be sure to change the name of the S3 bucket. Shutting down at the end of the script means that you will no longer be charged for the EC2 instance, though there will be some charges for storage.

docker_example_startup.sh shows an example of how you could run a Docker container in this setup. The image has been 'tar-ed' using:

docker save <image_name> > <image_tar_name>.tar

and then copied to the S3 bucket.

A better solution would be to use Amazon Elastic Container Registry (ECR).

Note that the AMI used in app.py has Docker already installed. You may have to install it yourself, depending on the AMI you choose.

For an example using a GPU, see the files in the 'gpu_example' directory. You may need to increase your GPU quota.

Useful commands