Amazon SageMaker MLOps pipeline with PrestoDB

Amazon SageMaker can be used to create end-to-end MLOps pipelines that include steps for data preparation, model training, tuning, evaluation and batch transform. In this repo we show how you can read raw data available in PrestoDB via SageMaker Processing Jobs, then train a binary classification model using SageMaker Training Jobs, tune the model using SageMaker Automatic Model Tuning and then run a batch transform for inference. All these steps are tied together via two SageMaker Pipelines: a training pipeline and a batch inference pipeline. Finally, we also demonstrate deploying the trained model on a SageMaker Endpoint for real-time inference.

Getting started

To run this code follow the prerequisites below and then run the notebooks in the order specified.

Prerequisites

The following prerequisites need to be in place before running this code.

PrestoDB

We will use the built-in datasets available in PrestoDB for this repo. Following the instructions below to setup PrestoDB on an Amazon EC2 instance in your account. If you already have access to a PrestoDB instance then you can skip this section but keep its connection details handy (see the presto section in the config file).

1. Create a security group to limit access to Presto. Create a security group called MyPrestoSG with two inbound rules to only allow access to Presto.

   • Create the first rule to allow inbound traffic on port 8080 to Anywhere-IPv4
   • Create the second rule rule to allow allow inbound traffic on port 22 to your IP only.
   • You should allow all outbound traffic

2. Spin-up an EC2 instance with the following settings. This instance is used to run PrestoDB.

   • AMI: Amazon Linux 2 AMI (HVM)
   • SSD Volume Type – ami-0b1e534a4ff9019e0 (64-bit x86) / ami-0a5c7dec456e07a8d (64-bit Arm)
   • Instance type: t3a.medium
   • Subnet: Pick a public one and assign a public IP
   • IAM role: None
   • EBS: 8 GB gp2
   • Security group: MyPrestoSG

3. Install the JVM and Presto binaries.

   • Once the instance state changes to “running” and status checks are passed. Try to ssh into your EC2 instance with:

     ssh ec2-user@{public-ip} -i {location}
     

   • If everything goes well, you will see the shell of your EC2 instance.

   • Install Presto 330 on the EC2 instance. Presto 330 requires the long-term support version Java 11. So let’s install it. First elevate yourself to root

     sudo su
     

     Then update yum and install Amazon Corretto 11.

     yum update -y
     yum install java-11-amazon-corretto.x86_64
     java --version
     

   • Now download the PrestoDB release binaries into the EC2 instance. You can download the Presto release binaries from the Maven Central Repository with wget. Then extract the archive to a directory named presto-server-330.

     wget https://repo.maven.apache.org/maven2/io/prestosql/presto-server/330/presto-server-330.tar.gz 
     
     tar xvzf presto-server-330.tar.gz 
     
     ls -ltr presto-server-330
     

4. Configure Presto and add a data source. Before we start the Presto daemon, we must first provide a set of configuration files in presto-server-330/etc and add a data source. Go into presto-server-330 and create the etc directory

   cd presto-server-330
   mkdir etc
   

   • Then create the three files using vim or your favorite text editor.

     • Presto logging configuration file etc/config.properties

       coordinator=true
       node-scheduler.include-coordinator=true
       http-server.http.port=8080
       query.max-memory=5GB
       query.max-memory-per-node=1GB
       query.max-total-memory-per-node=2GB
       discovery-server.enabled=true
       discovery.uri=http://localhost:8080 
       

     • Presto node configuration: etc/node.properties

       node.environment=demo 
       

     • JVM configuration: etc/jvm.config

       -server
       -Xmx4G
       -XX:+UseG1GC
       -XX:G1HeapRegionSize=32M
       -XX:+UseGCOverheadLimit
       -XX:+ExplicitGCInvokesConcurrent
       -XX:+HeapDumpOnOutOfMemoryError
       -XX:+ExitOnOutOfMemoryError
       -Djdk.nio.maxCachedBufferSize=2000000
       -Djdk.attach.allowAttachSelf=true 
       

     • Catalog properties file for the TPC-H connector: etc/catalog/tpch.properties

       connector.name=tpch
       

5. Run the PrestoDB daemon. Use the bin/launcher script to start Presto as a foreground process. The script is in the folder presto-server-330

   bin/launcher run
   

   If you’ve set everything up right, Presto will begin printing logs to stdout and stderr. After awhile you should see this line

   INFO        main io.prestosql.server.PrestoServer ======== SERVER STARTED  
   

6. You have a running instance of PrestoDB! Since you launched PrestoDB on a public subnet and enabled 8080 inbound traffic. You can even access the UI at http://{ec2-public-ip}:8080.

IAM Role

The SageMaker execution role used to run this solution should have permissions to launch, list and describes various SageMaker services and artifacts. ***Until a AWS CloudFormation template is provided which creates the role with the requisite IAM permissions, use a SageMaker execution role that AmazonSageMakerFullAccess AWS managed policy for your execution role.

AWS Secrets Manager

Setup a secret in Secrets Manager for the PrestoDB username and password. Call the secret prestodb-credentials and add a username field to to it and a password field to it.

Steps to run

1. Clone the code repo on SageMaker Studio.

2. Edit the config as per PrestoDB connection, IAM role and other pipeline details such as instance types for various pipeline steps etc.

   • Edit the parameter values in the presto section. These parameters define the connectivity to PrestoDB.
   • Edit the parameter values in the aws section. These parameters define the IAM role, bucket name, region and other AWS cloud related parameters.
   • Edit the parameter values in the sections corresponding to the pipeline steps i.e. training_step, tuning_step, transform_step etc. Review all the parameters in these sections carefully and edit them as appropriate for your use-case.
   • Review the parameters in the rest of the sections of the configand edit them if needed.

3. Run the 0_model_training_pipeline notebook to train and tune the ML model and register it with the SageMaker model registry. All the steps in this notebook are executed as part of a training pipeline.

   • This notebook also contains an automatic model approval step that changes the state of the model registered with the model registry from PendingForApproval to Approved state. This step can be removed for prod accounts where manual or some criteria based approval would be required.

4. Run the 1_batch_transform_pipeline notebook to launch the batch inference pipeline that reads data from PrestoDB and runs batch inference on it using the most recent Approved ML model.

5. Run the 2_realtime_inference notebook to deploy the model as a SageMaker endpoint for real-time inference.

Contributing

Please read our contributing guidelines if you'd like to open an issue or submit a pull request.

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.