pgEdge

In PostgreSQL, every write starts in the WAL before it's flushed to disk. CHECKPOINT is the command that forces that flush - settling all pending writes and bringing the database to a known-good state. Until Postgres 19, it had one speed: all of it, right now, block until done. Shaun Thomas covers planned changes to CHECKPOINT within PG19 in his latest PG Phriday - and it's one of those small-sounding additions that suddenly sounds far more relevant once you've watched a write storm take down a busy server. There's two new options coming to checkpoints in the upcoming PostgreSQL major release. MODE SPREAD starts the checkpoint early and distributes the work across the available window instead of hammering through it all at once, while FLUSH_UNLOGGED lets you decide whether unlogged tables are even part of the flush. MODE FAST is still there for when you actually want the blunt instrument - such as before a backup or before pg_upgrade, or between benchmark runs. (And yes, there are benchmarks in the blog!) Happy Phriday: 📖 hubs.ly/Q04pRM4z0 #postgresql #postgres #postgres19 #dba #database #pgEdge #opensource #pg19 #data #dataanalytics #programming #devcommunity #techcommunity #tech

Spock 6 Beta is here. Multi-master replication for PostgreSQL 16 through 19, fully open source under the PostgreSQL license. What's new: ⭐ Custom WAL resource manager for replication progress tracking ⭐ Per-subscription conflict statistics (PG18+) ⭐ JSON audit logging for every replicated change ⭐ Spill-to-disk for multi-gigabyte transactions in exception handling ⭐ Rolling upgrades from Spock 5.x with zero downtime ⭐ Detection and repair for silent node drift ⭐ performance improvements to Zero Downtime for adding nodes & performing rolling major PG version upgrades ⭐ Native logical slot failover on PG17+ ⭐ Cascade replication LSN tracking fixes ⭐ Rebuilt lag tracking with three-position LSN visibility ⭐ Conflict type naming aligned with PostgreSQL 18 ⭐ Automatic resolution cleanup and retention management Antony Pegg wrote the full walkthrough:  hubs.la/Q04pV3Jw0 Check out the full project on GitHub - don't forget to leave a star: hubs.la/Q04pTvLP0 #postgresql #postgres #replication #multimaster #distributedsystems #opensource #database #dba #pgEdge #developers #appdev

The pgEdge AI Knowledgebase Builder now ships per-model database packages instead of a single monolithic DB. If you've been thinking about trying out our open source tooling for agentic AI + Postgres - specifically the MCP Server for PostgreSQL or the AI DBA Workbench - this is a must know for engineering around specific build requirements. Previously, the knowledge base was one SQLite file with pre-computed embeddings for every supported model stored as separate columns - one set per chunk, per model. More content meant more rows. More models meant more columns. The whole thing grew whether you were using those embeddings or not. Now each embedding model gets its own package. `kb-openai-text-embedding-3-small.db` is its own install. Pick your model, pull that package, nothing else comes along for the ride. Multiple packages can sit side by side if you need them. For air-gapped environments, you can pull exactly what your build requires - no bundled extras. Smaller installs. Faster setup. Nothing unused on disk. Changes reflected in upstream projects: 📈 hubs.la/Q04pPQfx0 🤖 hubs.la/Q04pP-Wc0 #postgresql #postgres #rag #embeddings #vectorsearch #opensource #aiengineering #llm #developers #database #appdev #pgEdge

Into Postgres, agentic AI, open source tooling, distributed databases, or just building things with data? We make videos about all of it. 🤖 pgEdge MCP Server for PostgreSQL: query your database through Claude Code, Cursor, or any MCP-compatible client. 15 min, full demo. 🧊 ColdFront data tiering: transparent hot/cold storage, same SQL, no app changes. Analytics on a Postgres table spanning years of data. 🔍 AI DBA Workbench: open source monitoring with AI. We ran real comparisons against Citus, AlloyDB, Amazon RDS, and Azure Flexible Server. 🐘 PostgreSQL 18 deep dive: performance, indexing, replication enhancements. 54 min if you want to go deep. And a lot more: HA strategies, Kubernetes deployments, cloud demos, quick-tip shorts, and "Don't Do This!": common Postgres mistakes, exactly what it sounds like. 🎥 hubs.la/Q04pP6HR0 Still growing... don't forget to subscribe to keep an eye out for new content. 🌱 #postgresql #postgres #agenticai #mcp #opensource #dataengineering #aiengineering #analytics #developertools #database #appdev #tech #learntocode

Great meetup from PostgresEDI in Edinburgh this past Thursday 🐘 July 9th at Paterson's Land, University of Edinburgh - two talks, good crowd, actual sunshine (a rarity). Pat Wright (Redgate) made the case for community involvement as an introvert: practical advice on writing abstracts that get accepted, rehearsing to an empty room before presenting to others, and a challenge to everyone there - meet three new people at your next event. Then Martins Otun (Algonix AI) walked through compliance-first AI retrieval on Postgres and pgvector. The scenario: an AI approves a £3M loan, and six months later the regulator asks why. Which documents, which embedding model, which chunk - and can you reproduce it today? His answer: keep everything in Postgres. Vectors, permissions, document versions, and audit trail all in one place, every retrieval writing its audit record in the same transaction. "Relevance is not evidence." Proud to sponsor the pizza and refreshments. Jimmy Angelakos organizes PostgresEDI alongside Jim Gardner and Denys Rybalchenko. Next meetup is August 13th in Edinburgh - register here: 🎙️ hubs.la/Q04psq3g0 Full recap and slides on Jimmy's blog: 📖 hubs.la/Q04ptpT_0 #postgresql #postgres #edinburgh #scotland #techcommunity #unitedkingdom #opensource #database #techmeetups #ai #pgvector #compliance

ColdFront got coverage in La Revue Tech last week 🐘🇫🇷   The article covers the OLTP/analytics angle well. One thing it doesn't mention... the most important differentiator: cold data in ColdFront is writable. Every other tiering system makes archived data read-only. A GDPR right-to-erasure request against data archived two years ago means: restore to hot, delete, re-archive, verify. With ColdFront: DELETE FROM events WHERE user_id = 42 Same table name, same SQL, whether the data is two days old or two years old. No restore cycle. No application changes.   The architecture behind it: DuckDB running in-process for analytical speed on cold data, Apache Iceberg on any S3-compatible store, stock upstream PostgreSQL — no fork, no proprietary distribution. Distributed write serialization across nodes is TLA+-verified. Storage cost drops roughly 90% by moving data nobody queries from SSD ($0.20–0.50/GB) to object storage ($0.02/GB).   v1.0.0-beta1 is live on GitHub: try it out, leave feedback, and don't forget to ⭐ to keep tabs on the project as it continues to develop... 🔗 hubs.la/Q04ptNsZ0   📰 hubs.la/Q04pt_1J0   #postgresql #postgres #apacheiceberg #duckdb #opensource #dataengineering #basededonnees #ingenieriedesdonnees #stockagedonnees #RGPD #analytique #technologie #informatique #programmation

Postgres Summit US 2026 is coming to New York City - September 30 to October 2, at Convene on Broadway. 2026 marks a milestone: the conference is evolving from Postgres Conf NYC into Postgres Summit US, a new identity that better reflects the national community it's grown into. Three days of talks from contributors, engineers building high-scale systems, finance and tech practitioners, and anyone eager to run their first query. The largest US-based Postgres community conference, organized by the United States PostgreSQL Association. pgEdge is a platinum sponsor this year 🐘 If you're planning to be there - or if you decide to stop by - come find us. Registration is open: 📖 hubs.la/Q04prv980 #postgresql #postgres #postgressummit #newyorkcity #nyc #opensource #database #techevent #techcommunity #devcommunity #data #dev #devcommunity

Chances are you're already running Postgres somewhere. Here's what we've been building on top of it - all open source, all under the PostgreSQL License. 🤖 pgedge-postgres-mcp - MCP server for Postgres. Query your database through Claude Code, Cursor, or any MCP-compatible client. Read-only by default, works with any Postgres 14+. 🔍 pgedge-rag-server - RAG API server using pgvector. Hybrid search, streaming, configurable token budgets. Skip the separate vector database if you're already on Postgres. 🧊 ColdFront - Tier old rows to Apache Iceberg on S3 automatically. Same table name, same SQL, no application changes. Cold data stays writable. v1.0.0-beta1 just shipped. 🕵️ pgedge-anonymizer - Anonymize PII in Postgres databases for dev/test use. 100+ patterns across 19 countries, single atomic transaction, GDPR/CCPA-friendly. If any of this is useful - or if you find something broken - stars, issues, and feedback are all genuinely welcome: 🔗 hubs.la/Q04prCl-0 #postgresql #postgres #opensource #mcp #rag #pgvector #dataengineering #apacheiceberg #developertools #ai

Every Postgres database accumulates history that nobody queries but everybody pays for - event logs, audit trails, user activity going back years, all sitting on SSD storage at full price. ColdFront moves old rows to Apache Iceberg on S3 automatically. Same table name, same SQL, zero application changes. Built by Jimmy Angelakos. v1.0.0-beta1 is live 🎉 What makes it different from other data tiering options: the cold tier is writable. UPDATE and DELETE work on archived rows through the same table name, in normal SQL. Every other tiering option makes cold data read-only - which means a GDPR deletion request against archived rows is usually a restore-then-delete cycle. With ColdFront it's a single SQL statement. Under the hood: 🦆 DuckDB runs in-process for analytic speed on cold data - no separate daemon, no RPC hop ⚙️ Stock upstream Postgres, no fork, confirmed on PG 16/17/18 🔐 Distributed write serialization is TLA+-verified (model's in the repo if you want to audit it) We're in beta and we'd love for you to poke at it. Star it, try it, find bugs, open issues - all of it genuinely helps: 🔗 hubs.la/Q04prQdm0 #postgresql #postgres #opensource #apacheiceberg #datalake #database #dataengineering #duckdb

PGConf Asia 2026 is heading to Hong Kong - November 17-18 at InnoCentre, a FinTech and GreenTech innovation hub operated by Hong Kong Science and Technology Parks Corporation in Kowloon Tong. Two days of talks, workshops, and community time in one of Asia's most connected technology cities. It's also a milestone year: PGConf Asia turns 10 🎉 and PostgreSQL turns 30 🐘. A good year to be in the room. The CFP is open. They're looking for talks in four areas: PostgreSQL engineering (internals, extensions, performance tuning, replication, logical decoding, reliability techniques from people actually running this stuff in production), operations and production (architecture for scale, observability, migrations, disaster recovery, and hard-won lessons from Postgres under load), data platforms and AI (cloud-native deployments, vector databases, AI data pipelines, application patterns built on Postgres), and community and open source (regional user groups, governance, committer stories, the culture that keeps the project moving). If you're running Postgres in production and have a story to tell, this is the audience for it. CFP closes August 31: 🎙️ hubs.la/Q04nM33j0 Two pgEdge folks are helping make it happen: Hari Kiran and Mark Wong are volunteering as conference organizers. Looking forward to seeing what the program shapes up to be. 📖 Full conference details: hubs.la/Q04nLylG0 #postgresql #postgres #pgconfasia #hongkong #opensource #database #conferences #techevent #developers #devcommunity #devs #programming #cfp

A 95% failure rate for enterprise AI projects, per MIT NANDA. IDC frames it differently: only 9% of EMEA organizations have delivered measurable business outcomes from AI. The easy read is that the models are failing. Meredith Shubel's piece in The New Stack makes a different case: the problem is data infrastructure that wasn't built for production, and not enough DBAs to manage what gets deployed. pgEdge co-founder Phillip Merrick is quoted throughout. Most teams prototype on vendor-managed cloud platforms that "make it very easy to get started" - and then hit a wall when production requires deployment flexibility, data sovereignty, governance, and upgrades without downtime. "In the vendor-managed cloud world, the answer to that is almost always no." His answer on infrastructure: Postgres, for its extensibility, open-source licensing, and ability to handle unstructured data, vector embeddings, and geospatial in one place - "the Swiss army knife of databases." Make the right data infrastructure choice early and "you'll have this throughline from prototype to production." The people problem is harder. "You've had this massive, massive step shift in developer productivity. But you have to have some way of managing that on the production side; these databases can't go unmonitored. Who's going to manage them?" Industry predictions say 41% of today's database professionals plan to leave in the next decade. His take: agentic operations needs to catch up with agentic engineering. 📖 hubs.la/Q04nM4y50 #postgresql #postgres #ai #agenticai #database #devops #dba #aiinfrastructure #opensource #aioperations #dev #appdev

Data checksums have been in Postgres since version 9.3. That's 2013. And yet, enormous numbers of production clusters still don't have them enabled. Why? Because enabling checksums on an existing cluster meant a complete outage - run pg_checksums --enable, wait through every page on disk, rebuild your replicas from scratch. On a large database, that's a multi-hour window. And there's no partial credit if something went wrong. Postgres 18 flipped the default to on for new clusters. That helps going forward, but doesn't touch the existing ones. Postgres 19 does. Shaun Thomas's latest PG Phriday covers the new pg_enable_data_checksums() function - a single call that launches a background conversion on a running cluster, WAL-logged, with standbys implicitly updated. And the data_checksums parameter is now an enum, so you can track where the conversion stands. You still pay in background I/O. But "tunable background I/O on a live system" and "a multi-hour outage with no rollback" are not the same cost. Read the full piece: 📖 hubs.la/Q04nvfKV0 #PostgreSQL #Postgres #DBA #DatabaseEngineering #DataIntegrity #OpenSource #pgEdge #Postgres19 #BitRot #DataProtection

pgEdge sponsors PostgresEDI - the Edinburgh PostgreSQL user group organized by Jimmy Angelakos - and the July 9th meetup is coming up sooner than you think. Two talks are scheduled: 🎤 Pat Wright (Redgate) - making the case for why technical skills alone don't set you apart, and how showing up in the community opens doors even if you dread the spotlight. 📊 Martins Otun - building AI retrieval systems for regulated industries with PostgreSQL and pgvector. Auditable, compliant by design, SQL-first. For healthcare, finance, and anywhere a good audit trail isn't optional. Thursday July 9, doors at 6pm, Paterson's Land, University of Edinburgh. Pizza and networking first, talks at 7pm, then across to The Canons' Gait for the post-meetup social. Free to attend - registration required. Register 🎙️ hubs.la/Q04nqf0V0 #PostgreSQL #PostgresEDI #Edinburgh #Postgres #pgvector #AI #RAG #OpenSource #Community #pgEdge #DataEngineering

Ahsan Hadi's spent 20 years watching the same Postgres problems show up at the worst possible time. So he built something he could point at any instance and actually know what was going on. pg-healthcheck runs 180+ checks against live system catalog views - wraparound proximity, slot lag, WAL growth vs. a rolling baseline, visibility map mismatches VACUUM won't catch. The ask subcommand is the interesting part. Describe what you need in plain English - "How safe are we from TXID wraparound?" - and it maps to the right checks and runs them. Ollama by default, no API key, works air-gapped. Open source under the PostgreSQL License. Grab it on GitHub 🔗 hubs.la/Q04nfqLg0 Full writeup ✨ hubs.la/Q04nflNv0 #PostgreSQL #Postgres #OpenSource #DBA #Monitoring #pgEdge #DataEngineering #DevOps #SRE #GitHubRepo #Devs #DevCommunity #Data #Database #DataAnalytics #Monitoring #Metrics #Observability

Build Night #1: The Data Layer for Agentic Apps is in the books. 🛠️ We co-hosted the night with The Frontier Syndicate at AngelList Founders Cafe in San Francisco during Deep Tech Week - a small, intentional room of engineers, founders, architects, and CTOs who are building AI applications on Postgres in production. Mike Josephson started things off by walking through the pgEdge Agentic AI Toolkit: MCP-native database access, hybrid search with pgvector + VectorChord-bm25, RAG, and distributed Postgres for the AI era. The question anchoring the night was a real one - what does it actually take to move an AI agent from demo-mode into a production-grade system that can safely read, write, and act on real data? Codeflash showed how to mount external systems as filesystem context for agents. Open Computer covered agent-native developer infrastructure. Constructive walked through secure-by-default Postgres for AI-native apps. The conversations were technical, the demos were live, and the lobster rolls delivered. It was a good night. 🐘 Thanks to The Frontier Syndicate and AngelList for bringing the right people into the room, and to everyone who came out. #AgenticAI #PostgreSQL #AIEngineering #OpenSource #Postgres #DeepTech #SF2026

Two talks from our team (at @PosetteConf 2026) are on YouTube if you weren't able to catch them live: Hari Kiran's "Past, Present, and Future: Logical Decoding and Replication in PostgreSQL" traces the full arc: from the decoding API in Postgres 9.4, through pglogical, to native replication in recent releases, and then forward to where multi-master distributed Postgres is heading. It covers CDC pipelines, zero-downtime migrations, DDL propagation, and the move away from read-only replicas and single-writer systems. 🎬 youtube.com/watch?v=2XPsRA… @vyruss "LISTEN Carefully: How NOTIFY Can Trip Up Your Database" digs into a production hazard hiding in a feature most people assume is lightweight. NOTIFY acquires a cluster-wide exclusive lock to serialize notifications. Under high concurrency, that's statement timeouts and lock contention that look like they're coming from somewhere else. He walks through a real incident, the investigation, and the fix. 🎬 youtube.com/watch?v=gVsUtg… #PostgreSQL #Postgres #PosetteConf #database #dba #dataengineering #learntech #techtalks @

"In most tiering systems, cold (older) data is read-only, so a GDPR deletion request on archived data means restore-delete-rearchive, which is a half day job. ColdFront's architecture would allow you to UPDATE and DELETE archived rows through one SQL statement." That's Ashish Chaturvedi at HFS Research, quoted in Anirban Ghoshal's piece in InfoWorld today. The article places ColdFront alongside Databricks LTAP, Snowflake pg_lake, and EDB's converged analytics as competing approaches to collapsing the OLTP/OLAP divide for AI workloads. The distinction analysts draw is about where the center of gravity sits. Databricks asks enterprises to adopt a lakehouse as the operational center. EDB keeps Postgres as the source of truth while exposing data through Iceberg for analytical engines. Snowflake writes Postgres data directly into Iceberg so both can query it. ColdFront is the 100% open source solution that treats Iceberg as a transparent storage tier behind Postgres - older data moves there automatically, while applications keep using the same tables and SQL. Read the full piece: 📖 hubs.la/Q04mS3S30 #postgres #postgresql #apacheiceberg #database #databases #dba #dataengineering #ai #datamanagement #oltp #olap #opensource

Postgres 19 finally answers a question that's been hanging since version 10: what do you do when your partition design needs to change? The old approach was five steps: create new tables, copy rows, detach the original, attach the replacements, drop the husk. Serviceable, as Shaun Thomas describes it in his latest PG Phriday - but tedious, error-prone, and subject to various locks throughout. Postgres 19 collapses that into one statement. SPLIT PARTITION breaks a partition into smaller pieces. MERGE PARTITIONS folds them back together. Both are transactional, both work on RANGE and LIST tables. Neither has a CONCURRENTLY variant yet, and both acquire an ACCESS EXCLUSIVE lock on the parent table - not just the partition being modified. That lock matters. For the right use case, this is a genuine quality-of-life improvement. For high-traffic tables with large partitions, it's a maintenance window operation for now. The old table-swap polka is still available if availability is the priority. Hash partitions aren't in scope. The foundation is there. If the history of Postgres partitioning tells you anything, rough edges get filed down release by release. Shaun's full walkthrough - including the DEFAULT partition use case, which is where this really earns its keep: 📖 hubs.la/Q04mQVg10 #postgres #postgresql #databases #sql #dataengineering #database #dba #opensource #softwareengineering #cloudnative