The blog is about...

This is about setup and usage guide for AWS instance for deploying Machine Learning pipeline.

1) EC2 instance Setup

I used MLFlow both locally and on AWS, for tracking. The MLFlow server can be run on a AWS EC2 instance. In this section, I have discussed how to setup EC2 instance for the same.

1.1) Basic Instructions

The first step is to setup an EC2 instance. This can be done via services -> compute -> EC2 -> launch an instance. Setup the instance with the relevant configuration that is needed.

• The name of the instance

• Application and OS image (Ubuntu, is there a better option for developers 😂) with the appropriate version and architecture

• Instance type (the one with the free tier 😅, t2.micro or t3.micro, of course initially)

• Key pair login by create new key pair with key pair type=RSA and private key file format=.pem. A key.pem file will be downloaded. This can be used for the remote login to the EC2 instance.

• Configure storage by selecting the maximum amount of storage available (for free tier)

• Launch instance with default configurations for the rest

1.2) Remote login to the EC2 instance

Go to instances and select the instance that you just created. Use the public IPv4 address for login. The following command can be used for remote login. ssh -i ~/Downloads/key.pem ubuntu@IPv4

I faced one error while trying remote login. The following image shows the error.

This can be resolved with a simple permission change for the key.pem file. This can be done with the following command chmod 600 key.pem. Login to the EC2 instance with a sigh of relief now 😅. The following image shows a successful login.

Once you reach this stage, you can do whatever you want. The next step would obviously be to setup python virtual env and install relevant packages. An important package to install is the awscli for managing the aws services using the command-line.

1.3) Setting up AWS key and secret access key

This can be done via account -> security credentials -> create access key -> copy the access key and secret access key.

1.4) Install awscli and boto3

Run pip install awscli boto3 to install. Run aws configure and enter the access keys. Run aws s3 ls to list the AWS s3 buckets.

1.5) Run MLFlow server

Run the following command to start the MLFlow server.
mlflow server -h 0.0.0.0 -p 5000 --backend-store-uri postgresql://DB_USERNAME:DB_PASSWORD@DB_ENDPOINT:DB_PORT/DB_NAME --default-artifact-root s3://S3_BUCKET_NAME

If you get the below mentioned error,
File "/home/ubuntu/ml_env/lib/python3.12/site-packages/mlflow/cli.py", line 356, in <module> type=click.Choice([e.name for e in importlib.metadata.entry_points().get("mlflow.app", [])]), AttributeError: 'EntryPoints' object has no attribute 'get'

Then make changes to the following file virtual_env/ml_env/lib/python3.12/site-packages/mlflow/cli.py The error is in cli.py, just substitute get("mlflow.app", []) by select(group="mlflow.app")

If you get the following error ModuleNotFoundError: No module named 'psycopg2' then install the psycopg2 using the following command pip install psycopg2-binary

Add an inbound rule for the security group to allow access to port 5000. More info on the inbound rule setup can be found in MLFlow setup on AWS. Now, the MLFlow UI can be accessed in http://EC2_PUBLIC_DNS:5000. This can also be used as the tracking_uri in MLFlow.

2) Configuring an S3 bucket

The model files, requirements and other environment files are stored under artifacts in MLFlow. When using MLFlow, I have tried storing these locally as well as on the Amazon S3 bucket.

The configuration for an S3 bucket can be done via services -> storage -> S3 -> create bucket. Choose a unique bucket name with default settings.

3) Configuring and setting up an instance of RDS

All the metadata for an ML experiment like params, metrics etc. will be stored in a relational database in MLFlow, if we choose to do so. When using MLFlow, I have tried with SQLite database when running locally. For AWS, I have tried with PostgreSQL.

The AWS RDS setup can be done via services -> database -> RDS -> create database.

• Select PostgreSQL as backend database for the MLFlow. Choose the free tier template 😅.

• Choose the database instance identifier name, master username, and setup auto generate password.

• Select appropriate instance configuration and storage required.

• Select appropriate instance configuration and storage required.

• Let the connectivity, tags and monitoring be defaults. Choose password authentication for the database.

• In additional configuration, choose an initial database name.

• Do not forget to make a copy of the username and password to make connection to the database 😅.

• Edit security groups and add an inbound rule for PostgreSQL for the EC2 instance to connect to the database.

• Finally, make a note of the endpoint and port to the database.

Main takeaway

This is not my first time setup / use of EC2 instance and S3 bucket 😅 but a first time configuring MLFlow on AWS by myself and configuring RDS. Although I faced some challenges during setting up this infrastructure on AWS, I had a positive learning experience and learned that it is relatively easy to setup and use an AWS EC2 instance + S3 bucket + RDS (PostgreSQL) for configuring MLFlow! This blog can be used for deploying complex ML pipelines on AWS. To deploying complex MLOps pipelines now 😄.

Next Steps

• Train some complex ML pipelines by using this AWS EC2 + S3 + RDS + MLFlow infrastructure that has been setup.

• Compare this infrastructure setup with AWS Sagemaker. Compare how much effort is required to setup the infrastructure.

• Deploy multiple MLOps projects using complex real world datasets.