The Engineer's Logs

Software Engineer | Musician | Traveller

🎧 What I Consumed

Zepto launches Atom, giving brands real-time insights into their product performance on the platform.

Databricks ventures into the OLTP space with its acquisition of Neon, an AI-assisted, cloud-native, serverless Postgres engine. With Neon, AI agents can spin up temporary Postgres branches to run, test, and discard without human input.

🎧 What I Learnt

I worked on unit testing for a couple of APIs. Learnt a couple of things:

The flow I followed for setup was:
- Added the project’s root directory to sys.path so imports work everywhere without ugly relative paths.
- For isolation, every test runs against its own in-memory SQLite database.
- The helper fresh_db (an asynccontextmanager in conftest.py) does the heavy lifting. An async context manager bundles “setup → yield resource → teardown” for asynchronous code. It produces an object you use with “async with” anywhere in your application or tests.
  - before yield: creates the SQLite engine, builds all tables, then opens and returns an AsyncSession;
  - after the test exits: disposes the engine, wiping the database.
- db_session is a pytest fixture simply yields fresh_db’s session. A pytest fixture is a reusable, plug-and-play helper that prepares something your test needs and then cleans it up afterward. Using the db_session as an argument inside your test implies:
  - Pytest inspects the test function’s parameter list.
  - It sees a parameter named db_session and looks for a fixture with that name.
  - It runs the fixture, which in turn runs fresh_db, creates a brand-new AsyncSession, and yields it.
  - The yielded session object is fed into the test function as the argument db_session.
  - After the test body completes (pass or fail), execution resumes after the yield inside the fixture, which closes/disposes the SQLite engine—so nothing leaks into the next test.
- Most test files also declare a small seeded_session helper. It calls fresh_db, pre-populates rows that the particular test needs, then yields the session—so every test starts with relevant data.
- During startup we register a few MySQL-style SQL functions with SQLite so that any raw SQL used in the codebase runs unchanged.
- ```
      async test  ───>  requests fixture  ───>  fixture enters context manager
                    ^                       |
                    |                       v
                    <────  uses session  <─── yields session
```
All input and expected data was separated out into a data file for tests. This made it easy to tweak the parameters.
To summarise:
- Pytest finds the file →
- creates a clean SQLite database →
- seeds a few rows →
- runs the business-logic function you’re interested in →
- checks the return values/query results →
- tears everything down

The Engineer's Logs - #3

Comments

Scaling the Engineering Mindset

The Engineer's Logs - #2

More from this blog

The Engineer's Logs - #2

The Engineer’s Logs - #1

Sparking Solutions

Sparks Fly

🔎 This Week’s (Tech and other) Musings

🎧 What I Consumed

🎧 What I Learnt

Command Palette

Comments

Scaling the Engineering Mindset

The Engineer's Logs - #2

More from this blog

🔎 This Week’s (Tech and other) Musings

🎧 What I Consumed

🎧 What I Learnt