CHIRAG

How are short codes generated?

● secrets random

Six characters from the full alphanumeric set gives about 56.8 billion combinations, which is ample. But size of the space is irrelevant if the generator is predictable. Python’s random is a Mersenne Twister seeded from system time — observe enough outputs and you can reconstruct the state and predict every subsequent code. secrets draws from the OS entropy pool instead. Same code length, same user experience, no enumeration risk.

Why SQLite and not Postgres?

● SQLite Postgres

At this scale Postgres would mean a second process, a second container, a network hop on every query and credentials to manage — real operational weight bought for zero benefit. SQLite is one file with none of that. The important part is that nothing in the codebase is SQLite-specific: everything goes through the ORM, so switching is a connection-string change rather than a rewrite. That is what makes the choice reversible rather than a corner.

Why does the DB dependency yield instead of return?

● yield + finally return

A return hands the session to the handler and the function is done — nothing runs afterwards. If that handler raises, the session is never closed and the connection leaks. Under load that is a slow exhaustion bug that only shows up in production. Yielding inside a try/finally means the framework resumes the generator after the response, and the close runs whether the handler succeeded or blew up.

Why put Nginx in front at all?

● reverse proxy expose the app directly

The application server can speak HTTP perfectly well on its own, so the proxy looks like a spare part until you count what it centralises: TLS termination and renewal, the HTTP-to-HTTPS redirect, static files served without waking the application, and a single public listener instead of one per service. Add a second service later and none of that has to be solved twice.

Why is application data on a volume?

● mounted volume container filesystem

A container’s writable layer dies with the container. Data written there survives exactly until the next rebuild, which turns every deploy into a decision about whether it is worth losing the database. Moving state onto a volume makes rebuilds boring, and boring rebuilds are what let you deploy often enough to keep changes small.