Table of Contents

• I. Introduction
• II. The NLP Pipeline
• III. Infrastructure Provisioning Workflow
• IV. Conclusion

I. Introduction

• What is our Goal?
• Motivation for the Talk
• Why AWS CLI
• Why Taskfiles
• GitHub Codes supporting this talk

What is our Goal?

Automation in infrastructure provisioning is key to reducing the development time.
No one else can do this better than a developer so that DevOps engineer does less heavy-lifting at the time-crunched release window period

• Create Infrastructure provisioning codes as & when you write your application development codes
• Make every small configuration component of your provisioned services traceable, and version-controlled

In this talk, we will ...

Cover how to build a Python NLP App in Cloud using replicable infrastructure provisioning codes

Give transferable learnings on building Serverless Apps in any cloud infra (here, AWS Cloud infra is chosen).

Be using a combination of CLI commands and Yaml Taskfiles to provision the AWS Infrastructure.

What is in it for you?

By the end of this talk, a developer will

• start loving the potent combination of AWS CLI + Taskfile for provisioning replicable cloud infrastructure

• understand how this AWS CLI approach helps in-depth understanding of the cloud services employed

• reduce the "hell month" for DevOps Engineers during releases (refer adopting DevOps mindset)

Motivation for the Talk

• Is DevOps an afterthought for a developer?
• Does a developer have the same rigor in application code development applied to their infrastructure provisioning codes?

Image Source: https://modelthinkers.com/mental-model/devops-mindset

Why Dev needs to adopt DevOps Principles

• More nimble & responsive product delivery !

The right DevOps culture ultimately makes you deliver better products faster.

Source: https://stackoverflow.blog/2020/06/10/the-rise-of-the-devops-mindset/

“It is not the strongest of the species that survive, nor the most intelligent, but the one most responsive to change.” Charles Darwin

Source: https://faun.pub/some-popular-devops-quotes-and-what-i-learned-from-it-using-in-my-day-to-day-development-7b299ced7884

Why AWS CLI?

The reason why we chose AWS CLI is that

• it is a unified interface to all AWS services and
• it mimics the console way of creating AWS services with the right* level of abstraction
  • meaning, you have control over the CRUD operations
• it is not different from a traditional bash commands that we are highly familiar with

*arguable personal opinion

• What you give in arguments is what is provisioned for you !

/path/where/lambda_codes/located % cat aws_cli_command_for_lambda_creation.bash
#!/bin/bash
aws lambda create-function \
--function-name $1 \
--zip-file fileb://${1}.zip \
--runtime python3.8 \
--role $2 \
--handler lambda_function.lambda_handler \
--timeout 60 \
--memory-size 256 \
--layers $3 \
--architectures x86_64

• And, it gets executed with a just task <task_name>

/path/where/lambda_codes/located % cat Taskfile.yml 
...
...
tasks:
  create_lambda_name:
    cmds:
      - zip -r ${LAMBDA_FUNCTION_NAME}.zip lambda_function.py
      - bash aws_cli_command_for_lambda_creation.bash $LAMBDA_FUNCTION_NAME $IAM_ROLE_ARN $SPACY_LAYER

Why Taskfiles?

• The Taskfiles make the execution of steps so easy. You could repeat it multiple times for various projects.

# how to create IAM policy and roles
/path/where/IAM_Taskfile/is/located/ % task create_policy && task create_role && task attach_role_to_policy

# how lambda function is created
/path/where/LAMBDA_Taskfile/is/located/ % task create_lambda_name && task update_lambda_environment

# how to test the lambda
/path/where/Testing_Taskfile/is/located/ % task run_test_event_1 

{
    "StatusCode": 200, "ExecutedVersion": "$LATEST"
},
{
  "output_bucket_name": "pycon-$USER-nlp-output-bkt", "file_key": "email_1.txt",
  "message": "PII Redaction Pipeline worked successfully"
}

Alternate Options:

• As a developer you could also choose other modes of creation of Cloud Services such as AWS CDK, AWS CloudFormation and AWS Terraform and others.
  • There is a bit of a learning curve involved in all the above frameworks.
• Want to know more about AWS Toolkits? Refer aws-toolkit-aws-cli-sdk-and-cdk-6feab9e746b8

---

---

https://github.com/Toyota-Connected-India/serverless_nlp_app

---

II. The NLP Pipeline

• Goal of the NLP Pipeline
• Sample Input and Output
• Pipeline Architecture
• Workflow of the Pipeline Tasks

Broadly, there are two major types of NLP Pipelines:

1. NLP Pipeline in ML World: Text pre-processing >> Feature Embeddings >> Model Training and Prediction on Numericalized Embeddings
2. NLP Pipeline in Data Engineering World: Transforming raw text data into useful outputs in a sequence of steps.

If an NLP Pipeline could be defined in above 2 major ways, the second definition of Data Engineering based pipeline, is what we will accomplish in this talk.

Goal of the NLP Pipeline

• The pipeline will be capable of redacting sensitive information such as email addresses, phone numbers, and names (PII) from email bodies.
  • We will use examples from Enron Email data to test the pipeline.

Pipeline Architecture

Note: This pipeline is intentionally made simple. Real-world Serverless Pipelines could be much more complex

Sample Input and Output

Sample Input

Output of Simple Lambda (phone and email redacted)

Output of Layer Lambda (names redacted)

Workflow of the Tasks

1. Setting up S3 Buckets

- Create the S3 Trigger bucket (s3_1 in pic), intermediary S3 bucket and Output S3 Bucket (s3_2 in pic) 

GitHub Codes Reference

---

2. Create a "Simple Lambda"

- With no special/ extra packages, in a standard Py3.8 lambda env, 
    - create a lambda that replaces Phone and Email and 
    - test it with a sample csv file

GitHub Codes Reference

---

3. Create a "Layer Lambda"

• 3.A. Create a Spacy Layer inside a amazon/aws-lambda-python:3.8 and publish as a layer
• 3.B. Create a "Layer Lambda" that identifies & replaces names using a SpaCy pre-trained model

GitHub Codes Reference

---

4. Create a StepFunctions Pipeline

• 4.A. Create a Stepfunctions State Machine Json | GitHub Codes Reference
• 4.B. Create a StepFunctions Invoke Lambda and Test it | GitHub Codes Reference

---

5.Test the Whole Pipeline with a simple task command

# set up the temporary AWS credentials
## task executes the task from `Taskfile.yml` in the `/path/to/serverless_nlp_app`
/path/to/serverless_nlp_app/src/aws/2.stepfunctions_invoke_lambda/c.testing  % task run_test_event_1

GitHub Codes Reference

---

III. Infrastructure Provisioning Workflow

Sample IAM Taskfile.yml

Sample Lambda Taskfile.yml

IV. Conclusion

• Treat your infrastructure provisioning codes with the same rigor as your application development codes.
  • This is getting easier with AWS CLI + Taskfile.yml approach
• As many companies adopt a cloud-first approach, ensuring its developers have a DevOps Mindset ensures better software development cycle

---

---

https://github.com/Toyota-Connected-India/serverless_nlp_app