Why Postgres Is the Only Workflow Engine That Actually Works – Forever

Hello everyone. Let’s talk about Postgres being your shiny new savior for durable workflows. Because of course, out of the vast smorgasbord of modern data systems – key-value stores, NoSQL hype machines, bleeding-edge distributed thingamabobs – they went with old reliable Postgres. You know, the database that’s been around longer than most of today’s software engineers have been alive. According to the article, this is not because it’s boring and safe, but because it has technical advantages. Sure. And I eat kale chips for the antioxidants, not because they’re in the pantry and I’m too lazy to buy real food.

Scalable Queues – or Avoiding a Multiplayer Loot Grab

Apparently, database-backed queues are tricky. Who knew? It’s like in an MMO where every player tries to loot the same corpse simultaneously, and then half of them are just left awkwardly mashing F while watching someone else run off with the shiny drops. That’s what happens here: multiple workers try to grab the same workflow, creating contention misery. The article’s fix? Postgres’ magical “SELECT FOR UPDATE SKIP LOCKED” spell. This lets Worker A grab the first N tasks, Worker B grab the next N, and so on – instead of everyone dogpiling like it’s Black Friday at Walmart.

Is this groundbreaking? Not really – it’s just using a feature that’s been there for years. But I’ll give them this: it works, and it transforms a bottleneck into something that can scale across thousands of workers without choking harder than an unpatched Skyrim quest chain.

Observability – Because Flying Blind is a Terrible Idea

The authors pat themselves on the back for making workflows “observable” – as in, you can see what’s going on, in real time. Well, hooray. Postgres’ relational model means you can run SQL queries to see every workflow that errored this week, last month, or during that one ill-fated Friday deployment that everyone pretends didn’t happen. Secondary indexes come into play here, which in doctor terms are like giving your search queries a healthy dose of caffeine to get them moving faster.

Of course, indexes aren’t free – they’re costly to maintain, so you can’t just sprinkle them everywhere like candy. They index only the juicy bits: creation time, executor ID, and workflow status. In gaming terms, they’ve basically picked their “meta loadout” for optimal performance without weighing themselves down with unnecessary junk loot.

Exactly-Once Execution – or How to Stop Double-Charging the Customer

Durable workflows usually give you “at least once” execution, which is tech-speak for “good luck making your code idempotent, buddy.” That means a step might run twice if a worker faceplants mid-run – great if you like duplicate orders and angry accountants. The Postgres-based solution? Wrap your step and the checkpoint write inside the same transaction. If the whole thing goes down mid-process, it rolls back like nothing happened. If it commits, the step is locked in, and you don’t replay it later. Exactly-once semantics achieved. Cue applause.

From a sysadmin’s perspective, this is like having an autosave that actually works instead of corrupting your file – rare, precious, and worth clinging to. And yes, this only works for steps that are purely database operations, but honestly, that’s most of what these workflows are doing anyway.

The Verdict – Legacy Tech Still Rules

So, what’s the grand takeaway here? Postgres, the old warhorse, isn’t just trudging along – it’s still leading the charge. Locking clauses obliterate contention, SQL makes observability effortless, indexes add precision without killing performance, and transactions give you that sweet exactly-once guarantee. It’s not “sexy” tech, but it’s ruthlessly effective, and honestly, I’ll take boring stability over the latest framework-fad-turned-abandonware any day.

If you were hoping for a revolutionary new datastore to change your life, tough luck – you’re still dating Postgres. But at least it’s reliable, buys you dinner first, and doesn’t ghost you when things get tough. Overall impression: good. This is solid engineering dressed in sensible shoes, and the results speak for themselves.

And that, ladies and gentlemen, is entirely my opinion.

Article source: Build durable workflows with Postgres, https://www.dbos.dev/blog/why-postgres-durable-execution