Inherited a front-end e-commerce application, and built and optimized the backend API to store and serve the reviews data for the products on the website. I worked with over 12 million data entries regarding reviews for products.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | <= 50ms | < 1% |
The questions and answers data has over 12 million data entries. I transferred the data from a csv file into a postgreSQL database by performing an ETL process.
Deployed the server and database to AWS, I used a EC2 T2 micro instance for my server. I stress tested using Loader.io, and took the average of three test with randomized product ID.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | 83ms | 0% |
I scaled horizontally and deployed another server. Then, I used Nginx as my load balancer. Interestingly this increased my error rate, so I needed to do further research and optimization.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | 243ms | 51% |
To fix the error problem above I switched to least connection load balancing from round robin load balancing in Nginx. This brought down both my error rate and my latency speed. I still had not reached my goal however, so there was more work to be done.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | 83ms | 0.1% |
Performed content caching in load balancer.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | 61ms | 0% |
To get my latency speed even lower, I evaluated where my testing client (Loader.io) was located and re-deployed my AWS EC2 microinstances in the location of the client.
| Throughput | Latency | Error rate |
|---|---|---|
| 1,000 rps | 1ms | 0% |
Increased throughput to 6,000 rps.
| Throughput | Latency | Error rate |
|---|---|---|
| 6,000 rps | 3ms | 0% |
Ensure postgreSQL is installed in your computer
Copy example.env file and rename to .env, then fill in the variables
npm install
npm run server-dev