Building Durable Programs with Postgres
Presented by:
Peter Kraft
Peter is a co-founder of DBOS, helping you build reliable software effortlessly. He loves databases and distributed systems a little too much, so he did a PhD in them, working with Matei Zaharia and Peter Bailis at Stanford.
No video of the event yet, sorry!
Everyone knows you can use Postgres to make data durable. In this talk, we’ll explore how to use Postgres to make your programs durable, so they can survive crashes, restarts and updates without restarting from scratch.
In particular, we’ll explore an emerging technology called durable execution. The idea behind durable execution is to checkpoint a program’s state in a durable store so that if failures occur, it can “reload” from those checkpoints and resume from its last completed step.
But how can we efficiently implement durable execution on top of Postgres? What program state must be captured and how should it be stored and searched? How do we know when to recover programs? How do we implement necessary primitives such as durable queueing and messaging?
In this talk, we’ll address those questions through the design of a Postgres-backed distributed durable execution library. We’ll walk through core mechanisms such as checkpointing, recovery, queues, and messaging, and show how to efficiently implement them in Postgres. We'll then look at the challenges of scaling this approach in practice, including performance overheads, concurrency control, and contention, and discuss techniques for addressing them.
This talk is aimed at Postgres developers interested in reliability and in building systems that must keep running, even when everything else goes wrong.
- Date:
- 2026 April 21 14:30 PDT
- Duration:
- 20 min
- Room:
- San Pedro (Level C)
- Conference:
- Postgres Conference: 2026
- Language:
- Track:
- Dev
- Difficulty:
- Medium