Nikan's Notebook

General Backend Designs

4 min read

Backend Design-Patterns

Backend Function Types

  1. Reactive: executes when something is changed, such as when a file is uploaded, or an entry in the DB is updated
  2. Restful: Traditional restful function ran when a request is made to the server

Folder Structure:

Split file into multiple resource groups (user, login, order, etc…), each resource group will have two folders, Reactive and RESTful folders with each file having one function with the name of the function or the name of its endpoint.

Serverless vs Monolith

https://www.youtube.com/watch?v=90pVRK49AQM

How you do computing for a program:

  1. you own the machine
  2. you do it on the cloud
    1. Managed (S3):
    2. Serverless (Lambda, AWS):
      1. they give you a URL instead of an IP address
        1. bc by definition ur code is not binded to a server
        2. AWS basically refers the request to whatever server instance their Load Balancer algorithm determines
        3. you lose a lot of features, sometimes you wont even have access to node
      2. it is slower than serverful, because serverful is always there but this has to run the algorithm
    3. Monolith (non-managed server)(EC2):
      1. has an IP address
        1. always points to the same machine
      2. you have linux access to the server so you can do anything you want
      3. and you have to reboot if the server crashes or goes down your self, you have to manage everything ur self, ur just basically borrowing actual servers.
      4. you ssh into the cloud computer and just deploy as normal

Monolith vs Microservice

https://www.youtube.com/watch?v=1A9tPOfp6NA Monolith: One service does everything 1. Development Experience: 1. simplest one 2. 2:41 for architecture graph 3. easiest to run and test locally 2. Scalability: 1. can scale vertically(add memory or cpu to machine) 2. not great for horizantally scale 3. Response Time: 1. lowest latency, 2. not many hoops for the request to jump through such as Load Balancers, etc… 4. Reliability: 1. single point of failure 1. if process crashes all of it crashes, even if horizontally scaled since all instances are the same they will eventually crash. 5. Cost: 1. rent a virtual server(EC2) 2. fixed pricing 3. unutilized processing power if not that many users or scaled badly Microservice: multiple different services 1. Development Experience: 1. 7:26 for the graph chart 2. the monolith turns into an API Gateway that just redirects the requests to their Microservice using an HTTP or a GRPC call. 3. you handle the separations using Docker (Kubernetes and Docker). 4. integration tests are hard 5. Code will be nice and clean, but the infrastructure will be complicated 2. Scalability: 1. scale specific services both vertically and horizontally 3. Response Time: 1. since there are internal network calls, generally slower 4. Reliability: 1. failures are isolated per service 5. Cost: 1. deploy managed Kubernetes clusters(AWS EKS) 2. Bin packing problem: how to allocate resources to each Microservice can be hard Serverless: no services, just functions 1. Development Experience: 1. no infrostructure - simple code 2. runtime limitations, like u cant install external libraries 1. websockets and stuff are a pain 2. Scalability 1. scales from 0 to infinity basically 3. Response Time: 1. cold starts, initialiazing cost of the server, since serverless 4. Reliability: 1. failures isolated per function so its good 2. no infrastructure that needs to be managed 5. Cost:

Self Contained Systems- SCS

https://www.youtube.com/watch?v=xkQ9VaNTwxM basically an application that is made up of a lot of different smaller applications, each with their own UI, DB, and microservices, monoliths, etc… there is redundant data copying since they cant access each others data you just copy it over and keep redundant DBs in the overall system. Uses AMQP to handle talks between SCSs asynchronously. kind of like amazon, amazon movies, e-commerce, music, they’re all their own thing, and just come together at the end.

systems are somewhat decoupled and can be scaled independently

Graph View

Backlinks