Project to compare different backend technologies and how they perform doing same type of tasks.
To have all the servers do the same type of work and compare their performance.
Some of this work is:
- Make multiple calls to a Postgres Database
- Format the data and respond with JSON
The database used will be the same database with the same schema.
The database will easily become the bottleneck. The goal is to see the performance of each backend tech under load.
I would also like to add different types of calls where Redis is also used. This is to see caching benefits of Redis.
It will be a simple model of a Store. A store will have bunch of products. Those products can have many items. I might slowly add more models to it. For now, I will have those.
I will have Rails generate the tables and the models for simplicity. Then use SQL to query from all the different backends.
I am going to setup each of these with their basic tutorial on bootstrapping a project, add a database client, and implement the endpoints. I will write queries in raw SQL format.
I will also build docker containers out of each of them and deploy them to one of my local servers. I will have the database running on a separate server on my local network as well. That way, the database can run in isolation from the web servers. I will also run one server at a time to allow an even playing ground. In the end, it will be impossible to get equal resources on all the containers. So at the 95th percentile, the results might not always be consistent.
Using Ruby 3 and Rails 7. I will use the load_async and use fiber isolation level. I will try it without them as well to see the performance different
I used fastify-cli to generate a fastify project with TypeScript. I am going to use postgres package for database queries. I will also take advantage of JSON Schemas for JSON serialization.
For web server, I am using fiber. For database queries, I will use pgx.
For web server, I am using ktor. For database queries, I will use jasync-sql. It's netty based. I am using the netty engine for Ktor to get Non-Blocking I/O. jasync supports it out of the box.