<p><strong>TLDR:</strong> PostgreSQL 19 is adding <code>DELETE FOR PORTION OF</code> for temporal tables, finally making range-based history edits a single statement instead of a manual close-and-insert dance. Redpanda’s 26.1 release ships profile-guided optimization that cuts p50 latency by ~50% and CPU by ~15%. Snowflake opened up Trust Center Extensions to GA, letting security partners deploy posture management tools natively inside customer accounts. pgEdge released a production-ready MCP server for Postgres that claims 30-50% token reduction when AI agents query your database.</p>
<hr>
<h2>PostgreSQL 19 temporal table syntax</h2>
<p>PostgreSQL 19 introduces <code>UPDATE/DELETE FOR PORTION OF</code>, which lets you modify a specific time range within a row’s history in a single statement. Previously you had to manually close the existing range and insert a new row inside a transaction. The new syntax handles the range splitting automatically, including inserting leftover rows for history outside the targeted bounds. It works for both future and past ranges, which means retroactive corrections to historical data are now straightforward SQL.</p>
<h2>Redpanda PGO ships in 26.1</h2>
<p>Redpanda’s engineering team applied Profile-Guided Optimization to get ~50% reduction in p50 latency and ~15% lower CPU utilization. The core problem was that Redpanda was heavily frontend-bound (51% by Top-Down Microarchitecture Analysis) before PGO, dropping to ~38% after — meaning the bottleneck was instruction fetch and cache pressure, not compute. PGO concentrates hot code paths, which improves instruction cache locality and reduces iTLB pressure. They chose PGO over BOLT for production; the post has the full breakdown if you want to understand why.</p>
<h2>Snowflake Trust Center Extensions go GA</h2>
<p>Snowflake’s Trust Center Extensions are now generally available, built on the Native Application Framework. The pitch is zero data movement: security tools from partners like ALTR, Hunters, OneTrust, and TrustLogix run directly inside your Snowflake account rather than pulling data out. Supported use cases include DSPM, ISPM, UEBA, and SSPM. Extensions install from Snowsight, and an open-source project called SnowSentry provides developer templates.</p>
<h2>pgEdge MCP server for Postgres</h2>
<p>pgEdge released a production-ready MCP server that lets AI agents connect to any Postgres 14+ database without a traditional API layer. It does full schema introspection — primary keys, foreign keys, indexes, constraints — and claims 30-50% token reduction through TSV output, pagination, and context window compaction. It works with Claude Code, Cursor, Windsurf, and VS Code Copilot. Read-only by default, HTTPS/TLS and token auth built in, fully open source under the Postgres license.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Supabase hits 100k GitHub stars:</strong> The project now serves 8 million developers, built on Postgres, PostgREST, pgvector, and Elixir Phoenix — worth noting as a signal of where the Postgres ecosystem is landing for new projects.</li>
<li><strong>pgAdmin 4 v9.14:</strong> 23 bug fixes plus custom LLM provider URLs (LM Studio, EXO, LiteLLM compatible), chat history retention in the AI assistant, and a fix for B-tree index fillfactor defaults.</li>
<li><strong>Hardwood 1.0 Beta:</strong> A new Java-based Parquet parser with S3 backend (custom SigV4, no heavy SDK), predicate push-down, Avro bindings, and a native CLI — minimal dependencies is the explicit design goal, 1.0 final expected within weeks.</li>
<li><strong>MotherDuck Embedded Dives:</strong> Embeddable AI-generated data visualizations that run queries server-side then stream into a browser-local DuckDB-Wasm instance for 5-20ms interaction latency; included in Business plans.</li>
<li><strong>Network egress as a query concern:</strong> A concrete example shows switching from <code>SELECT *</code> to explicit column selection dropping response size from 50KB to 8KB per request — at 1M requests/day that’s ~1.2TB/month and roughly $100/month saved on a single endpoint.</li>
<li><strong>Cursor pagination in Django:</strong> DRF’s built-in <code>CursorPagination</code> delivers constant query time on large datasets; the post covers compound ordering for non-unique fields and why offset pagination still makes sense for admin interfaces.</li>
<li><strong>ElastiCache Serverless IPv6:</strong> AWS added IPv6 and dual-stack support to ElastiCache Serverless across all regions including GovCloud and China, at no extra cost.</li>
<li><strong>Postgres memory explainer from PlanetScale:</strong> High shared_buffers and OS page cache usage is healthy; high RSS is the actual problem — the post walks through using <code>EXPLAIN (ANALYZE, BUFFERS, MEMORY)</code> and PgBouncer in transaction mode to keep RSS in check.</li>
</ul>


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

Databases Digest

A roundup of database and infrastructure news: PostgreSQL 19 introduces DELETE/UPDATE FOR PORTION OF syntax for temporal tables, eliminating the manual close-and-insert pattern for range-based history edits. Redpanda 26.1 ships Profile-Guided Optimization achieving ~50% p50 latency reduction and ~15% CPU savings by improving instruction cache locality. Snowflake's Trust Center Extensions reach GA, enabling security partners to run posture management tools natively inside customer accounts with zero data movement. pgEdge releases a production-ready MCP server for Postgres 14+ with full schema introspection and 30-50% token reduction for AI agents. Also covered: Supabase hitting 100k GitHub stars, pgAdmin 4 v9.14 updates, a new Java Parquet parser, MotherDuck embedded visualizations, network egress optimization, Django cursor pagination, ElastiCache Serverless IPv6, and a Postgres memory usage explainer.

PostgreSQL 19 gets temporal DELETE syntax, Redpanda cuts p50 latency in half with PGO