08 September 2026

42 new posts across 6 database and data engineering sources — 7 worth your time.

One week of Databases & Data, 7 stories, as published.

Alexander Ioffe: How One Extra JOIN Reliably Breaks Postgres

This article details how adding a small 5-row lookup table to a 9-table billing query in Postgres can drastically degrade performance, flipping a 0.3ms nested loop plan to a 467ms hash join. The root cause is identified as `join_collapse_limit = 8`, a default Postgres has shipped since 2005.

Why it matters — It reveals a specific, counter-intuitive performance cliff in PostgreSQL related to `join_collapse_limit`, demonstrating how a seemingly minor query change can lead to significant performance regressions.

Planet PostgreSQL · 08 September 2026 · Read the original →

Postgres

Mikhail Shytsko: What Replica Mode Does Not Switch Off

This post reveals a common misconception about `session_replication_role = replica`, demonstrating that it does not fully disable enforcement for a session, leading to unexpected writes and rejections. It illustrates with an example of an order for a non-existent customer and a negative amount.

Why it matters — It clarifies a critical nuance of `session_replication_role = replica`, preventing data integrity issues or unexpected behavior when attempting to bypass constraints during specific database operations.

Planet PostgreSQL · 08 September 2026 · Read the original →

Data Engineering

Maki Majima: Migration files and Git do not mix

This post argues that migration directories, with their timestamped, append-only schema changes and in-database state tracking, fundamentally conflict with Git's version control model. It highlights that migrations don't inherit Git's protections like review processes, creating two unsynchronized timelines.

Why it matters — It exposes a critical impedance mismatch between database schema migration tools and Git, prompting reflection on how to manage schema changes and their deployment processes more robustly.

Planet PostgreSQL · 08 September 2026 · Read the original →

Analytics & OLAP

How Uken Games reduces observability costs by 87% with ClickHouse

Uken Games achieved an 87% reduction in observability costs by replacing Datadog with an open-source stack built on ClickHouse. This allowed them to store every trace on a single node while significantly cutting expenses.

Why it matters — It provides a concrete example with a significant cost saving (87%) by migrating to an open-source observability stack, demonstrating a viable strategy for managing large-scale data costs.

ClickHouse · 08 September 2026 · Read the original →

Analytics & OLAP

The Agentic Analytics Benchmark: Measuring model accuracy and efficiency in analytical agents

This article presents the Agentic Analytics Benchmark, which tested 28 AI models on 201 real analytics questions from a production data warehouse. It measures correctness, cost, and speed, and provides an open harness for replication.

Why it matters — It offers quantitative benchmark results for AI model performance in analytical tasks, providing data engineers with concrete metrics (correctness, cost, speed) to evaluate and select models for agentic analytics.

ClickHouse · 08 September 2026 · Read the original →

Storage Internals

What is a Neki router?

This post describes the Neki router, a component in sharded databases responsible for determining which shard(s) should execute a query. It explains that the router builds the initial plan based on data topology before Postgres creates its execution plan.

Why it matters — It introduces the concept of a sharding router and its role in distributed database query planning, which is fundamental to understanding how queries are executed and optimized in sharded environments.

PlanetScale · 08 September 2026 · Read the original →

Analytics & OLAP

ClickHouse as a streaming HTTP API

This post describes how to build a streaming HTTP API directly within ClickHouse 26.8, leveraging named handlers, typed parameters, pagination, framing formats, and access control. This allows for direct data serving from the database.

Why it matters — It presents a method to turn an OLAP database into a direct, streaming API endpoint, potentially simplifying data serving architectures and reducing latency for real-time applications.

ClickHouse · 08 September 2026 · Read the original →

The database always has another surprise.

7 stories, every Tuesday

Published here every week. Follow by RSS to get it as it lands.

Subscribe by RSS All issues

← Previous issue Next issue →