System requirements


Functional:

  • Create short URL
  • Read the short URL and redirect to complete URL
  • Collect metrics for each short-url created



Non-Functional:

  • Unique Short code generated by base62 algo
  • In-memory low-latency to cache short URLs
  • Short-URL is persistent async, returning the code to be used as-soon-as-possible
  • Use load-balancer to scale through multiple instances, adding high availability once instances reach a max load
  • Use a gateway-api to to rate-limit clients before hitting backend



API design

  • POST /short-url -> responsible to create the short-URL, sync request
  • GET /:code -> read the short-url created before and returns/redirect to the complete URL
  • GET /statistics/:code -> provide statistics report by short-URL code (how many access, ...)





High-level design

  • Create endpoint generates a unique short-code and returns it as-soon-as-possible, save on redis, key as code and complete-url as value. We use a Redis INCR command as seed to produce unique global codes.
  • Creation process works in background via queues to write on database in batches.
  • Latest created short-URL as saved on cache, the key is based
  • Read endpoint access Redis cache first, using the idempotent key. If cache miss, then go to database





Detailed component design

  • code-generator-service: used in the create url flow. Generates a unique code, save on redis and return 202 status code with short url.
  • url-writer-service: listen pubsub topic, pull messages in batch and write on postgres in batches
  • url-query-service: read from redis based on code, if cache miss, fetch data from replica postgres database written by url-writer-service




Database design

Postgres database to provide long-term history of short-url:

  • Table short_urls
    • id int sequential (PK)
    • full_url text not null
    • code varchar (20) not null -> index btree
    • reference_id varchar (50) -> unique idempotency key
    • expires_at datetime nullable -> used to ignore on expired read
    • created_at

Postgres database to provide complete report history:

  • Table urls
    • id int sequential (PK)
    • code varchar (20) not null -> index btree
    • read_count bigint default 0
    • created_at