Yes, we can turn off the primary: challenges of running analytical platform with 200TB of data at Adjust.

Auditorium 1

11:50 - 12:35

Developer

Adjust, one of the leading international mobile measurement SaaS companies,
stores 200TB of analytical data in PostgreSQL and the amount of data is
constantly growing. This talk introduces Adjust analytical platform and pipeline
and describes how Adjust reliably stores and serves a huge amount of
attribution-related data in PostgreSQL with various hardware-based constraints
and almost no dedicated database on-call. Adjust analytical platform really
benefits from the extensibility of PostgreSQL with the ability to define complex
custom data types and foreign data wrappers (and, as of version 12, pluggable
storages), creating a number of custom data types and several foreign data
wrappers, all of them open-source, I will describe the use-cases for them. I
will explain how Adjust platform can scale well past the present amount of data
and highlight the potential challenges there and some features we are looking
forward in PostgreSQL. Finally, I will answer the title question on how we can
stop PostgreSQL primaries without affecting our customers for events like
database major/minor upgrades, server reboots and so on.

This talk has DBAs, developers, SREs and database reliability engineers in mind,
but will hopefully be useful to other professionals interested in using
PostgreSQL for analytics and PostgreSQL extensibility use cases. Most PostgreSQL
extensions discussed are open-source and available from
https://github.com/adjust