Selling groceries online using Postgres, Flask, Docker & Android by Petrus Janse van Rensburg

Startups are exciting. But choosing the right tech-stack can be a daunting task: You want to be able to move quickly right from the start, but also remain flexible as the technical requirements inevitably change.

In this talk I will describe my own experience of building out the tech stack for a mobile e-commerce startup. It will cover: • an overview of the main design challenges • the chosen architecture (database, server-side application layer, API and mobile client) • how I used Flask to build the API • how I used Flask-Admin to build a management console • deploying the backend application using docker-compose.

Transcript

2. THE OPPORTUNITY
4. Internet usage in Africa • 226 million smartphones (end 2015) • set to triple by 2020 http://www.gsma.com/mobileeconomy/africa/
6. THE CHALLENGE
7. 1. Most smartphone users will still be stuck on slow / expensive connections for the foreseeable future.
9. 2. Only around 10 million fixed-line broadband subscriptions (for 1.2 billion people).
10. Existing e-commerce business models don’t do well in these conditions.
11. Existing approach • Access to extremely large inventory. • Find an item by doing:
 Search + Filter + Scroll + Tap for details • Each step requires network
12. Existing approach Example: Amazon’s Android app • 21 Mb data needed to buy 5 items • ~ 240 taps needed to place first order
13. • We sell only groceries to people around Cape Town, through a dedicated Android app. • It’s the first step towards building a new kind of online retailer, that works well in emerging-market conditions. What do we do?
15. Our approach • Only the goods you need for solving a simple, everyday task (e.g. buying groceries for a household). • Dedicated app, with a simple UI that uses little data. • Only in a limited geographic area.
16. Our approach How does it compare? • 21 Mb 0.25 Mb data needed to buy 5 items • Works offline - you only need to be connected when proceeding to checkout. • ~ 240 taps 65 taps needed for first order
18. - Charles Eames “There is always a need for anyone that can do a simple job thoroughly.”
19. Our approach
21. TECH STACK
22. Postgres + PostGIS
23. Flask + SQLAlchemy
26. FlaskAdmin
29. Docker + Docker- compose
30. Vanilla Android
31. Pitfalls
32. 1. Android’s sheer complexity Pitfalls
33. Blank app template starts with: • 48,3 MB on disk • 1 837 files & directories
34. 1. Android’s sheer complexity Be patient with yourself Pitfalls
35. 2. JSON handling in Android Pitfalls
36. 2. JSON handling in Android
 
 use SQLite Pitfalls
37. 3. Keeping sane with all the data models Pitfalls
38. 1. Postgres tables 2. SQLAlchemy model classes 3. JSON API 4. Client-side SQLite tables 5. Client-side Java model classes
39. 3. Keeping sane with all the data models Keep it simple: de-normalise where possible Pitfalls
40. 4. Slow upgrade cycle on Android Pitfalls
41. 4. Slow upgrade cycle on Android Gracefully terminate support for old client versions Pitfalls
42. Thank you! questions? Petrus Janse van Rensburg
 
 @petrusjvr
 
 petrus@timbuktu.io