How Aura Frames Scaled RDS Postgres for Peak Christmas Traffic (#1 in App Store)
Thursday, October 01 · 10:30–11:20
This presentation is a case study about how Aura Frames (https://auraframes.com) expanded from a single primary AWS RDS Postgres instance to 8 in total. Besides the Postgres configuration, we’ll walk through the technical details about the relevant infrastructure components like PgBouncer and Memcached, and which Postgres features were used and not used. Each year, Christmas Day is the biggest day of the year for the company where millions of users use the app to set up hundreds of thousands of new digital frames.
On Christmas Day 2025 the DBs handled a peak of 225K TPS via 10s of thousands of PgBouncer connections, increases of 4-5x or more from normal traffic. With millions of customers using the app, it rose in the ranking of all free apps to #1 in the US and Canadian App Stores. In 2025 in total, more than 1 billion photos were shared to Aura frames globally.
Things weren't so rosy on Christmas Day the year before. We will review scaling challenges from the preceding year that set the stage for significant application data layer refactoring, including unexplained high CPU use during vacuum and a change to how replication slots were used for physical replication.
We’ll walk through the solution for 2025 of “whole table sharding,” and the months of prep, development, verification and rollout that went into it. We'll discuss how additional instances benefit write operations, vacuum, and overall reliability, along with their drawbacks like the loss of joins, foreign key constraints, and the additional infrastructure cost. To wrap up we’ll look ahead to plans the DB engineers have to further expand capacity and reliability ahead of Christmas Day 2026.
More technical details can be found in this blog post: https://andyatkinson.com/postgresql-rds-scaling-aws-christmas-day-peak