Postgres Thursday s01e04: Dimitri Fontaine. The Art of PostgreSQL; pg_auto_failover
Postgres.TV is a series of online interviews organized by Nikolay Samokhvalov and Ilya Kosmodemiansky
This was one of the more enjoyable online conversations of the pandemic year. Nikolay and Ilya are genuine PostgreSQL experts and the conversation went deep quickly — less “tell us about your book” and more “why did you choose to cover X before Y” and “what is the most misunderstood feature?”
The answer to that last question: window functions. Most readers assume they are an advanced feature for data analysts only. In practice they solve everyday application problems cleanly — ranking, running totals, lead/lag comparisons — and every developer who writes non-trivial SQL should be fluent in them.
The second half covers pg_auto_failover, the HA extension that automates PostgreSQL failover for application developers who want reliable production Postgres without becoming HA experts themselves. The core insight: a monitor process that holds the state machine for each node’s role, so failover is deterministic and observable, not a black-box race between agents.