This website portfolio acts as a spotlight search about me: you can search anything you want to know about me such as "hobbies", "work experience", "favorite ice cream flavor", etc.
During development on my local machine, the following tools and platforms were used:
- Docker container running Elasticsearch instance
- Docker container running Kibana
- Node.js backend with Express.js framework
- React for frontend
When it came to deployment, I considered several routes:
- Use Elastic cloud for Elasticsearch instance and run Node.js backend on AWS EC2 instance; host on GitHub pages
- Use AWS OpenSearch document store, Gateway API and Lambda function for handling requests; host on GitHub pages
- Deploy ELK stack in a kubernetes cluster and run Node.js backend on AWS EC2 instance; host on GitHub pages
I chose to go with the 2nd option, since OpenSearch and Lambda functions are free (for basic configuration) under the AWS free tier. However, this does come at a cost of performance. Another drawback is that I had to re-factor some of the backend code to run as Lambda function.
This web application uses prefix, multi-match queries for a live search results experience.
Elasticsearch was chosen for this project because of their well-documented API as well as the excellent out-of-the-box performance.
Elastic cloud has provided the best performance of the methods that I have tested, but it is less of a long term solution (at the moment). Provisioned concurrency can be used to reduce lambda function response times (https://aws.amazon.com/blogs/compute/new-for-aws-lambda-predictable-start-up-times-with-provisioned-concurrency/), but it is not free. Instead, I have opted to invoke the lambda function on page load to potentially avoid a cold start. I also added a scheduler:
https://docs.aws.amazon.com/lambda/latest/dg/with-eventbridge-scheduler.html
https://docs.aws.amazon.com/eventbridge/latest/userguide/eb-run-lambda-schedule.html
I also increased the resources for the Lambda function to drastically improve performance. The memory is set to 2048 MB and the ephemeral storage is set to 1024 MB. Using an estimate of 100,000 requests per month with a duration of 1500ms per response, the estimated monthly cost is $0: https://calculator.aws/#/estimate?sc_channel=sm&sc_campaign=Support&sc_publisher=REDDIT&sc_country=global&sc_geo=GLOBAL&sc_outcome=AWS%20Support&sc_content=Support&trk=Support&linkId=310626969.