AWS Serverless Application Model Improves Cost Efficiency for Precision Exam’s Infrastructure
Our client provides a web platform for taking exams. The platform empowers teachers nationwide to enroll students, authorize, conduct exams online, and review results.
The existing client product is currently using older technologies and architectures that can be replaced by modern services provided by AWS. Although the system modules are decoupled and provide a robust solution, there were improvements that could be made to the architecture and CI/CD environment to enhance availability and scalability approaches based on AWS services. Due to increases in customer usage, driving new business requirements and in order to improve current infrastructure in the cost-efficient way, it was decided to start development of serverless application within the Proof of Concept stage for further migration to the production-ready solution. That’s why the client requested SoftServe to provide a prototype of a system, built on top of AWS Cloud, using the best-practices approach in architecture and managed AWS Services.
Challenges for the new solution:
- Build a highly-available, scalable and robust system.
- Use AWS managed and serverless services, in order to cover all application requirement.
- Follow AWS best practices for infrastructure design.
- Serverless for local development and testing.
- Continues Integration/Delivery implementation.
- Infrastructure as a Code implementation.
- Application Monitoring and Log aggregation.
Based on the best-practices were chosen AWS Lambda, API Gateway, and DynamoDB as serverless solutions that allow running code, describe RESTful API and store persistent data without any server provisioning and with minimal managing. Amazon Cognito provides user management, sign-up, sign-in, and access control to web apps quickly and easily.
Adhere to the principle of microservices, it was decided to create small lambda functions and avoid huge ones. Lambda functions essentially were designed to have only one class/function inside without various dependencies and repetitive functions. We used AWS Lambda Layers functionality to carry out the repetitive code and shared libraries into the separate essence (layer). In spite of the chosen approach to store a set of lambda code in a git mono-repository, lambdas were designed in the way of independent deployment. Basically, only changed lambdas will be re-deployed without deploying any other unchanged function within the same repository.
On Figure 3 (bellow) shown CI/CD flow of serverless backend with using of AWS CodeCommit, CodeBuild and CodePipline services. Since client requirement was to use BitBucket service as a git repository, we integrated it with AWS CodeCommit for having full-featured CI/CD Pipeline. We have git mono-repository, each lambda function has it is own directory for the codebase. Once changes on CodeCommit trigger AWS CodeBuild, it assembles the application, sends code to S3 bucket and moves CloudFormation template file up the pipeline chain to deploy the changed resources. In order to deploy only that lambda function which code was changed, we set up Maven reproducible builds plugin.
AWS Serverless Application Model (AWS SAM) is a framework that we use for emulating lambdas and API resources locally. It provides the ability to test the code internally and does not upload the code into AWS each time in order to test the changes. Mostly, we use SAM CLI for API Gateway emulation and local function invocation. In its essence, SAM is a CloudFormation library, therefore we have serverless resources (Lambdas, API gateways) described as a code. Moreover, SAM allows us to build serverless applications faster and smoother.
For serverless service monitoring and troubleshooting, we configured AWS CloudWatch and AWS X-Ray which recognize problems and mitigate any risks. With CloudWatch, it easy to monitor key metrics and logs, visualize application and infrastructure stack, create alarms, and correlate metrics and logs to understand and resolve the root cause of performance issues. The AWS X-Ray also provides two powerful features that can improve the efficiency in Active tracing, provides distributed tracing capabilities as well as visual service maps for faster troubleshooting. X-Ray helps to identify performance degradation and quickly understand anomalies, including latency distributions.
SoftServe delivered on time the prototype of the highly-available, robust and scalable serverless system. In tight collaboration with developers, we updated java code to the newer version and wrapped it up to fulfill the necessary serverless criteria. The prototype is ready for further development with no visible red flags for production in perspective.
Within the PoC SoftServe delivered:
- Designing and implementation of such serverless services as AWS Lambda, API Gateway, DynamoDB, CloudFront, S3.
- Log-In and Registration flow with Amazon Cognito.
- Infrastructure as a Code by means of CloudFormation and AWS SAM.
- Serverless local development and testing using AWS SAM.
- Implemented Continues Integration/Delivery for application’s back-end parts.
- Implemented Continues Integration/Delivery for application’s front-end parts.
- Application Monitoring and Log aggregation with CloudWatch and X-RAY.