- This is the gist of what it can do:
- User account registration with email verification.
- Posting questions and answers with media.
- Upvoting/downvoting questions and answers.
- Search and filtering for questions based on timestamp, query, count, and more.
- ⚡ React - a JavaScript library for building UIs.
- 💎 Koa - a next generation web framework for Node.js.
- 🐵 MongoDb - the most popular database for modern apps.
- 🐜 jwt - compact and secure data transmission for stateful services.
- 🔍 Elastic Search - scalable open source search.
- 💂♂️ Kibana - data visualization for Elastic stack.
- 💰 Redis - in-memory data structure store/cache.
- 🚀 Nginx - high performance load balancer.
- 🐰 RabbitMQ - open source message broker.
- ✨ PM2 - Node.js process manager.
- Was able to handle 1000 concurrent users sending a total of 200,000 requests within 10 minutes (averaging ~333 QPS), maintaining a 95th percentile tail latency of 200ms.
- In general, each instance you deploy would consist of all the microservices that are included in the project, which are the folders prefixed with
su-andqu-. - The option is to deploy them separately and cluster each microservice by type on each machine. This may reduce CPU load on instances that share both the backend microservices and the queueing.
- What I recommend is doing a mix of machines that take on a couple of microservices and partially specializing the instances.
- Assuming that you are familiar with Ansible, there are playbooks written to help speed up setting up instances. PM2 is used to deploy and maintain the instances.
- The app will be able to sustain quite a bit of load if all microservices are located on each machine.
- Some specialization, such as moving the search and questions microservices to separate machines may lower the overhead, since these services are often hit the most.
- It is recommended to have a query router for each app instance you have.
- However, shards and config servers should run by itself in its own instance to avoid performance hits.
- Deploying 3 shards, 3 config servers and one query router for each app instance should be sufficient.
- Redis will work best when running by itself on its own instance.
- Clustering is possible, but performance is slower than a single process based on testing.
- The ELK stack can run on a single machine. Ensure that there is at least 4GB of RAM so the write queue limit can be raised in Elastic Search to handle more requests.
Not the best way to do it, but worked nonetheless.
- Built by @derivatives(Ngan Nguyen) and yours truly.