QueryIdentifierAPI

An API that identifies the type of input text is a sentence, paragraph, question, or URL

Problem statement

Design a server that accepts requests from the user, processes the inputs from the requests, and returns the result to the user.

The server should handle requests from different users parallelly

The user request will contain English text

The server should decide whether the input text is a sentence, paragraph, question, or URL

Problem solving

• here the input text identification task is simple and compute intensive, can be solved using regex match
• api can be developed using various frameworks.
• fastapi - high performance api, built around starlette and pydantic

Design

• to support huge number of concurrent http requests,
  • an api should be deployed on machine with many resources(more cpu core) or
  • multiple replicas of an api should be running behind a good http load-balancer
• the first option is not a viable.
• so choosing 2nd option, with nginx as a load balancer and running multiple replicas of dockerized api behind it.

How to Run

• prerequisites : A system with docker and docker-compose
• steps
  • unzip the code
  • run docker-compose build - to build the images
  • run docker-compose up - to start the services
  • use below curl POST request to test api

CURL post request examples

---

query

curl -X POST \
  'http://172.17.0.1:9998/query' \
  --header 'Accept: */*' \
  --header 'Content-Type: application/json' \
  --data-raw '{"text": "Am i question?"}'

response

{"query_type":"question"}

---

query

curl -X POST \
  'http://172.17.0.1:9998/query' \
  --header 'Accept: */*' \
  --header 'Content-Type: application/json' \
  --data-raw '{"text": "I am sentence."}'

response

{"query_type":"sentence"}

---

query

curl -X POST \
  'http://172.17.0.1:9998/query' \
  --header 'Accept: */*' \
  --header 'Content-Type: application/json' \
  --data-raw '{"text": "I am sentence1. I am part of paragraph."}'

response

{"query_type":"paragraph"}

---

query

curl -X POST \
  'http://172.17.0.1:9998/query' \
  --header 'Accept: */*' \
  --header 'Content-Type: application/json' \
  --data-raw '{"text": "https://github.com/"}'

response

{"query_type":"url"}

Scaling

• To add more replicas
  • add below block in docker-compose.yml

    apireplicanew:
      build: ./src
      ports:
        - '<newport>:80'
    

  • add below block in nginx.conf in nginx/

    upstream app {
        .
        .
        server 172.17.0.1:<newport>;
    }
    

  • build the images again using docker-compose build
  • start the services again docker-compose up

Shortcomings in identification

• some special cases are not covered since it requires special handling
• examples
  • "1.1.1.1" - true identification is IP address but the api will identify as paragraph.