Ask HN: How can I load test PostgreSQL but avoid changing actual data?

1•LawZiL•1y ago

I'm running a load test on my Node.js application and want to simulate realistic write-heavy scenarios to stress test the system — especially the PostgreSQL database.

There's a part of the code that performs UPDATE queries, but during load testing, I don't want these updates to actually change the state of the database. However, I still want the database to experience the full impact of the update — query planning, locking, I/O, WAL writes, etc. — so I can measure how it affects performance under peak load.

Is there a best-practice approach to achieve this?

So far, the most viable option I’ve found is to wrap the UPDATE queries in a transaction and then roll it back — that way, the queries still execute, but the state doesn’t change. Is this the right way to go? Does rollback introduce any performance overhead that would skew my results? Are there better or more realistic alternatives that simulate the full write path without persisting changes?

Thanks in advance!

Comments

sargstuff•1y ago

pgbench[2] / pgreplay-go[0] / python locust[1]

very old school approach:

Misuse sql VIEW[3][4] to generate a temporary parallel table setup. Testing database separate from production database would be 'better', IMHO.

-----

[0] pgreplay-go : https://github.com/gocardless/pgreplay-go

[1] python locust : https://miguel-codes.medium.com/unconventional-load-testing-...

[2] pgbench load testing :

   a) https://medium.com/@c.ucanefe/pgbench-load-test-166bdfb5c75a

   b) https://neon.tech/blog/autoscaling-in-action-postgres-load-testing-with-pgbench

   c) https://hackernoon.com/how-to-create-and-load-test-data-in-postgresql-i41e353m

----------------------------

[3] : a) https://hashrocket.com/blog/posts/materialized-view-strategi...

      b) https://tech.jonathangardner.net/wiki/PostgreSQL/Materialized_Views

      c) https://stackoverflow.com/questions/77603705/what-is-the-use-of-materialized-view-in-postgresql

[4] : https://en.wikipedia.org/wiki/View_(SQL)

LawZiL•1y ago

but again, how can i perform the query, but not actually change the state, i really need the state to keep untouched, but be able to perform the query itself.

MatthiasPortzel•1y ago

> Does rollback introduce any performance overhead that would skew my results?

I would expect it to be the other way around—since the transactions are rolled back and not committed, they would have significantly less performance impact. But I’m working from an academic model of the database.

apothegm•1y ago

Load test against a clone instead of a DB whose contents you care about?

The absence of visible results does not mean the absence of progress

What is 'Havana syndrome,' condition a CIA director reportedly suffered from?

How My Website Works: A Technical Look at the Stack

CanvasUI: HTML-in-Canvas Component Library

Accidentally built a 500-site SEO mesh network and got 3k visits in 14 days

Geekbench 7 launches with Ryzen 7 7700 baseline

Show HN: NvChat – A single-exe Windows client for Nvidia's free LLM API

Tangled – tail CI logs over SSH

Yes, you can copy our eval setup

Liang Wenfeng Investor Exchange Meeting · Audio Transcript

Carbon: Graduating from the Experiment

NeuTTS-2E, an open-source on-device TTS model with 7 controllable emotions

WASM-native Unix sandboxing for the web

The Semiotics of Fear

Where Are All the Solar-Powered Cars?

The forbidden railway: Vienna – Pyongyang (2008)

Trump's Plan for Science: More Money for A.I., Less for Universities

Modern Kubernetes Homelab: GitOps with ArgoCD

Soon everybody will be worried about Chinese LLM's as agent Trojan horses

The Law of Leaky Abstractions (2002)

Google's Design Ethicist Testimony [pdf]

Code review: slapping an AI reviewer on top of an AI author doesn't cut it

OpenAI's models broke free and launched a cyberattack

Show HN: I built a token-per-second counter in Go

Beyond Arc and GSM: Langford Coverage as a Benchmark for ASI

Stupid Slow: The Perceived Speed of Computers (2024)

KTransformers – Flexible LLM Inference Framework

OpenAI's Australian data centre drops water recycling plan

I gave Pi one tool and it ran for eight days

Why creators are hedging against the TikTok algorithm