<p><strong>TLDR:</strong> A detailed PostgreSQL benchmark on AWS RDS shows a single server can sustain 144K writes/second and process 43K durable workflows/second before WAL flush becomes the bottleneck. Honker is a new SQLite extension that adds durable queues, pub/sub, and event streams directly into SQLite with no external broker. Redis 8.6 is now GA in open source, and Redis 8.4 landed on Redis Cloud with atomic slot migration that’s up to 30x faster than the legacy approach. DuckDB’s full-text search extension gets a practical walkthrough, with an honest comparison against Postgres and Elasticsearch.</p>
<hr>
<h2>PostgreSQL write throughput benchmark on AWS RDS</h2>
<p>Someone ran a serious benchmark of PostgreSQL write performance on an AWS RDS db.m7i.24xlarge instance and the numbers are worth knowing. A single Postgres server can sustain 144K writes/second and process 43K durable workflows/second — roughly 4 billion workflows/day — before WAL flush becomes the primary constraint. Queue-backed workflows hit a different ceiling: lock contention at the queue head limits throughput to 12.1K workflows/second, but distributing across multiple queues raises that to 30.6K/second. The benchmark code is open source if you want to run it against your own workload.</p>
<h2>Honker: durable queues and pub/sub inside SQLite</h2>
<p>Honker is a SQLite loadable extension that adds Postgres-style NOTIFY/LISTEN semantics, durable queues, event streams, and a scheduler — no Redis, no broker, no separate daemon. Cross-process wake latency is around 0.7ms via polling SQLite’s PRAGMA data_version every millisecond. The genuinely useful part: because the queue lives in the same SQLite file as your application data, enqueue operations commit atomically in the same transaction as your writes, which eliminates dual-write problems entirely. Bindings exist for Python, Node, Rust, Go, Ruby, Bun, and Elixir.</p>
<h2>Redis 8.6 GA and atomic slot migration in Redis 8.4</h2>
<p>Redis 8.6 is now GA in open source, bringing performance improvements, better Streams reliability, hot key diagnostics, new eviction policies, and native time series gap handling. Redis 8.4 landed on Redis Cloud with Atomic Slot Migration (ASM), which replaces the old key-by-key slot migration with a full-sync replication approach. The numbers are significant: ASM is up to 30x faster (6-8 seconds versus 192-219 seconds), generates 98% fewer client redirects, reduces latency spikes by 40%+, and cuts cluster message overhead by 94%. If you run Redis Cluster and have ever watched a migration crawl, this is the fix.</p>
<h2>DuckDB full-text search: good enough for most exploratory use cases</h2>
<p>A practical walkthrough of DuckDB’s FTS extension covers index creation, BM25 parameter tuning (k₁ and b), stemming, stop words, and conjunctive queries using a real email corpus. The post is honest about what’s missing: no result highlighting equivalent to Postgres’s ts_headline, and no phrase queries. The conclusion — good enough for exploratory use cases — feels accurate. If you’re already using DuckDB for analytics and need lightweight search without standing up Elasticsearch, this is worth knowing about.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Databricks Lakebase branching:</strong> Migrating Backstage from RDS to Lakebase (serverless Postgres on Neon’s copy-on-write architecture) shows a 63 MB database clone completing in 1.09 seconds and point-in-time recovery in under 4 seconds — the argument being that when branching is this cheap, you can eliminate mock objects and catch schema migration issues during development.</li>
<li><strong>Redis native OpenTelemetry metrics:</strong> redis-py v7.3.0+, go-redis v9.18.0+, and node-redis v5.12.0+ now expose connection pool health, command latency, client-side caching stats, and stream lag through the standard OTel pipeline; the command latency metric in redis-py adds ~7-8% overhead, so the grouped opt-in model matters.</li>
<li><strong>SAP metadata in Databricks Unity Catalog:</strong> SAP Business Data Cloud now syncs semantic metadata and governance tags into Unity Catalog via Delta Sharing, enriching cryptic SAP column names like VBAK and KUNNR with business-friendly descriptions and key relationships so Databricks AI tools can query them without manual mapping.</li>
<li><strong>MongoDB Hibernate ORM extension:</strong> Public preview of a Hibernate ORM extension for MongoDB lets Java developers use JPA annotations, HQL, and Criteria Queries against MongoDB’s document model, with HQL automatically translated to aggregation pipelines.</li>
<li><strong>InfluxDB 3 Explorer 1.8:</strong> Adds a Data Subscriptions page for streaming from MQTT, Kafka, and AMQP sources, a line protocol validator with real-time schema previews, and editable retention periods directly in the UI.</li>
<li><strong>ClickHouse on Docker Hardened Images:</strong> Switching to DHI drops medium-severity CVEs from 8 to zero by removing unnecessary packages like wget, curl, and apt, with SLSA Level 3 provenance and non-root execution by default.</li>
<li><strong>Hardwood 1.0.0.Beta2:</strong> Java-based Parquet parser adds VARIANT logical type support, an interactive TUI for exploring file structure, and a reworked page decoding pipeline that cuts a 9.6 GB NYC taxi benchmark from ~2.7s to 2.2s.</li>
<li><strong>EverC EMR on EKS migration:</strong> Migrating a Spark platform from a managed service to EMR on EKS using existing Kubernetes infrastructure reduced platform costs while preserving analyst notebook workflows via EMR Studio.</li>
<li><strong>DBmaestro MCP server:</strong> Connects AI agents to DBmaestro’s database DevOps platform so natural language commands trigger governed workflows within existing RBAC and audit trail constraints, rather than giving agents raw database access.</li>
<li><strong>Database breach governance:</strong> South Africa recorded 2,374 data breach incidents in 2024/2025 with attackers operating undetected for an average of 241 days; POPIA enforcement now carries penalties up to R10 million, making continuous activity monitoring and time-bound privileged access a legal requirement rather than a best practice.</li>
<li><strong>Migflow Rails engine:</strong> A read-only Rails engine that mounts at /migflow and provides a visual dashboard with timeline view, schema diffs, compare mode, an interactive ERD graph, and CI rake tasks for gating pipelines on migration risk scores.</li>
<li><strong>AI on databases (Percona):</strong> The BIRD benchmark puts top AI models at ~82% SQL execution accuracy versus ~93% for human database engineers; the practical takeaway is that AI handles the easy 80% of problems well but the harder 20% still needs human oversight.</li>
<li><strong>Dropbox Magic Pocket compaction redesign:</strong> Two new compaction strategies (L2 for sparse volume consolidation, L3 for extremely underfilled volumes) were added after a Live Coder service inadvertently created volumes using less than 5% of capacity, causing fragmentation across the immutable blob store.</li>
</ul>


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

Databases Digest

A databases-focused roundup covering several notable developments: a PostgreSQL benchmark on AWS RDS showing 144K writes/second with WAL flush as the bottleneck; Honker, a new SQLite extension adding durable queues, pub/sub, and event streams with atomic transactional guarantees and ~0.7ms wake latency; Redis 8.6 going GA with performance and reliability improvements, plus Redis 8.4's Atomic Slot Migration delivering up to 30x faster cluster slot migration; and a practical DuckDB full-text search walkthrough covering BM25 tuning, stemming, and honest limitations. Also covered: Databricks Lakebase branching, Redis native OpenTelemetry metrics, MongoDB Hibernate ORM extension, InfluxDB 3 Explorer 1.8, ClickHouse hardened Docker images, and more.

Postgres hits 144K writes/second in benchmarks, Honker brings durable queues into SQLite