<p><strong>TLDR:</strong> A Linux 7.0 kernel regression cuts PostgreSQL throughput roughly in half on Graviton4 hardware, and a fix is not guaranteed before the stable release ships in about two weeks. Meanwhile, Armin Ronacher published five months of production notes on Absurd, a durable execution system built entirely on Postgres that held up well. Netflix detailed the architecture behind their multimodal video search system, fusing Cassandra, Kafka, and Elasticsearch into a second-by-second temporal index. TigerFS, an experimental tool that mounts a PostgreSQL database as a filesystem, also landed this week.</p>
<hr>
<h2>Linux 7.0 kernel regression cuts PostgreSQL throughput in half</h2>
<p>An AWS engineer found that the Linux 7.0 development kernel cuts PostgreSQL throughput roughly in half on Graviton4 hardware. The culprit is Linux 7.0 restricting available kernel preemption modes, which pushes significantly more time into user-space spinlocks. A patch to restore PREEMPT_NONE as the default was proposed, but kernel maintainer Peter Zijlstra pushed back, arguing PostgreSQL should adopt the Restartable Sequences (RSEQ) time slice extension instead. If the patch is rejected and no workaround lands, PostgreSQL users on Linux 7.0 stable — shipping in roughly two weeks — are looking at a serious regression with no easy out.</p>
<hr>
<h2>Five months of Absurd in production</h2>
<p>Armin Ronacher published a detailed retrospective on running Absurd, a durable execution system built entirely on Postgres, in production for five months. The core design held: tasks, steps, checkpoints, and events are unchanged. Additions include decomposed steps via beginStep/completeStep, a CLI tool for debugging, a Go-based web dashboard, and agent integration. The checkpoint-based replay model proved more reliable than deterministic replay, and pull-based scheduling simplified operations. The TypeScript SDK sits at around 1,400 lines versus Temporal’s Python SDK at 170,000 — a pretty clear argument for keeping complexity in SQL. Remaining gaps include no built-in scheduler, no push/webhook model, and limited table partitioning support.</p>
<hr>
<h2>Netflix multimodal video search: Cassandra, Kafka, Elasticsearch</h2>
<p>Netflix published the architecture behind their internal video search system, built to help editorial teams find specific moments across thousands of hours of raw footage. Multiple specialized models handle character recognition, scene detection, and dialogue transcription, with outputs persisted in Cassandra via high-availability pipelines. Kafka-triggered offline jobs fuse detections into one-second temporal buckets, and the enriched records land in Elasticsearch for hybrid queries combining k-NN vector similarity with phrase matching, n-gram tokenization, and Levenshtein fuzzy matching. The temporal bucketing approach for intersecting overlapping signals is worth a close read if you’re building anything similar.</p>
<hr>
<h2>TigerFS mounts PostgreSQL as a filesystem</h2>
<p>TigerFS is an experimental MIT-licensed project that mounts a PostgreSQL database as a filesystem directory, letting you interact with data using standard Unix tools like ls, cat, and grep. It supports two modes: file-first for atomic writes and versioning backed by Postgres, and data-first for exploring existing schemas via filesystem paths that translate to SQL queries. Each file maps to a database row with full ACID guarantees. It mounts via FUSE on Linux and NFS on macOS. Experimental, but the use case for agent workflows that need to avoid custom APIs is real.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Incident workflow benchmarks:</strong> A post on structured data incident workflows cites 30% faster response times, 25-30% productivity gains from automation, and 40-75% error rate reductions — take the numbers with some skepticism since no primary source is named, but the process framing around detection, triage, and post-incident analysis is solid.</li>
<li><strong>Go 1.25 REST API guide:</strong> A detailed walkthrough builds a TODO API using only the standard library and a pure-Go SQLite driver, covering hexagonal architecture, DDD with value objects, and BDD-style TDD — no Gin, no GORM, no code generation.</li>
<li><strong>AI agents and production databases:</strong> A meme about Claude deleting a production database made the rounds, which is funny until it isn’t — worth keeping in mind as agent autonomy over live systems becomes a real design question.</li>
</ul>


Query optimization, schema design, and data store coverage across PostgreSQL, MongoDB, Redis, and more.

Databases Digest

A Linux 7.0 kernel regression halves PostgreSQL throughput on Graviton4 hardware due to restricted kernel preemption modes pushing work into user-space spinlocks, with no guaranteed fix before stable release in ~two weeks. Armin Ronacher shared five months of production notes on Absurd, a durable execution system built on Postgres, highlighting its checkpoint-based replay model and a lean 1,400-line TypeScript SDK. Netflix detailed their multimodal video search architecture fusing Cassandra, Kafka, and Elasticsearch into one-second temporal buckets for editorial search across raw footage. TigerFS, an experimental FUSE/NFS tool that mounts a PostgreSQL database as a filesystem, also launched. Additional items cover incident workflow benchmarks, a Go 1.25 REST API guide using only the standard library, and concerns about AI agents with access to production databases.

Linux 7.0 halves PostgreSQL throughput, Absurd runs five months on Postgres