<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>DBExplore blog</title><description>Engineering notes on Postgres observability and governed autonomy.</description><link>https://www.dbexplore.com/</link><language>en</language><item><title>The pgvector index nobody measured</title><link>https://www.dbexplore.com/blog/pgvector-index-outgrew-memory/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/pgvector-index-outgrew-memory/</guid><description>Vector search in Postgres degrades quietly. The index crosses RAM, the build runs on disk, recall decays with churn, and none of it raises an alert.</description><pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate><category>pgvector</category><category>vector-search</category><category>postgres-internals</category></item><item><title>Aurora&apos;s writer endpoint can lie to you</title><link>https://www.dbexplore.com/blog/aurora-writer-endpoint-can-lie-to-you/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/aurora-writer-endpoint-can-lie-to-you/</guid><description>CloudWatch measures instances. Your application talks to endpoints. For the minute those disagree, every instance metric looks healthy while writes fail.</description><pubDate>Wed, 23 Sep 2026 00:00:00 GMT</pubDate><category>aurora</category><category>high-availability</category><category>monitoring</category></item><item><title>PostgreSQL 18 moved the WAL I/O counters</title><link>https://www.dbexplore.com/blog/postgres-18-moved-the-wal-io-counters/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/postgres-18-moved-the-wal-io-counters/</guid><description>pg_stat_wal lost its write and sync columns to pg_stat_io in 18. Nothing errors. The numbers just stop arriving, and most monitoring never notices.</description><pubDate>Wed, 16 Sep 2026 00:00:00 GMT</pubDate><category>postgres-18</category><category>postgres-internals</category><category>monitoring</category></item><item><title>Why we capped ourselves at rung 2</title><link>https://www.dbexplore.com/blog/why-we-capped-ourselves-at-rung-2/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/why-we-capped-ourselves-at-rung-2/</guid><description>We built an action plane and a fail-closed policy gate, then refused to let any action run unattended until it could prove it was reversible. The ladder.</description><pubDate>Thu, 10 Sep 2026 00:00:00 GMT</pubDate><category>governed-autonomy</category><category>action-plane</category><category>policy-gate</category></item><item><title>75% of our query-plan storage was duplicate</title><link>https://www.dbexplore.com/blog/75-percent-of-our-plan-storage-was-duplicate/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/75-percent-of-our-plan-storage-was-duplicate/</guid><description>Thousands of stored plans turned out to be a few hundred distinct shapes. What a structural fingerprint keeps, what it strips, and why regressions depend on it.</description><pubDate>Thu, 27 Aug 2026 00:00:00 GMT</pubDate><category>query-plans</category><category>plan-regression</category><category>postgres-internals</category></item><item><title>What our anomaly engine actually runs</title><link>https://www.dbexplore.com/blog/what-our-anomaly-engine-actually-runs/</link><guid isPermaLink="true">https://www.dbexplore.com/blog/what-our-anomaly-engine-actually-runs/</guid><description>An isolation forest with a fixed contamination rate, density clustering with a sample floor, a threshold-ladder classifier, and heuristics that name a cause.</description><pubDate>Thu, 13 Aug 2026 00:00:00 GMT</pubDate><category>anomaly-detection</category><category>machine-learning</category><category>signatures</category></item></channel></rss>