←
pgEdge1,327 posts
pgEdge avatar

pgEdge

@pgEdgeInc

pgEdge is the open source Postgres company for agentic AI and enterprise applications.

◷ Joined March 2023▧ 1,091 Photos and videos
pgEdge avatar
pgEdge@pgEdgeInc·Nov 17

An AI coding agent can stand up a working app and a PostgreSQL database behind it in an afternoon, using whatever defaults get the demo running fastest. That’s a sensible choice for a prototype, and it’s also where those prototypes often run into trouble later, when the app has to handle real traffic and real customer data. In a recent Guest View on SD Times, Dave Page, CTO at pgEdge, PostgreSQL Core Team member and creator of pgAdmin, explains why the database is so often what keeps an AI prototype from reaching production. He covers what slips through when nobody has reviewed those defaults, why choosing a region on a hosting platform doesn’t settle where your data ends up, and what to decide before an agent writes a line of code. If you’re a founder shipping an agent-built app, an engineer reviewing one, or on the SRE or platform team that will end up running it, it’s a short read and well worth the time. Read his full article in SD Times: 📖 hubs.la/Q04FBk200 #PostgreSQL #Postgres #AIAgents #AIEngineering #VibeCoding #Startups #Database #DBA #SRE #PlatformEngineering #SoftwareEngineering #AppDevelopment #DevOps #SDTimes

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 16

Got a plan for when a PostgreSQL cluster gets corrupted? Dr. Ibrar Ahmed, Principal Engineer at pgEdge, is walking through one this Thursday, Nov 19 at 8 AM PT (11 AM ET / 4 PM GMT). He’ll go through recovery step by step, from spotting the damage and stopping the service to restoring from verified backups, point-in-time recovery, and pg_resetwal as the last resort. It’s free. Ibrar’s taking questions live at the end, and everyone who registers gets the recording: hubs.la/Q04FqB-x0 #postgresql #postgres #dba #database #disasterrecovery #webinar

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 13

KubeCon + CloudNativeCon NA 2026 in Salt Lake City has officially ended - we ended up with a whopping 2,991,149 orders written with 0 failed writes in our live demo that showcased a live Postgres node running in Kube, where the network was cut while CloudNativePG restarted the pod and Spock brought the two nodes into alignment. There was great traffic at the event, and it was a very AI focused event through and through. We were glad to be a part of it as attendees and as a sponsor for the event. Missed us in person but still want to get in touch or try out our enterprise-ready solutions for fully open source PostgreSQL + support services? Request a 30 day evaluation for your Kubernetes architecture, anytime: hubs.la/Q04FhfCh0 #postgres #postgreSQL #aiengineering #kubecon #cloudnativecon #saltlakecity #utah #dataengineering #ai #llm #agenticai #agenticengineering #programming #kubernetes #kube #k8s #distributed #highavailability

Post attachmentPost attachmentPost attachmentPost attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 13

If you've missed out on Dr. Ibrar Ahmed's "Eight Steps to Bullet-Proof Database Disaster Recovery" conference talk, he's giving it live (online, free!) this November 19th at 8 AM PST / 11 AM EST. Learn about how to handle everything from start to finish: - early-detection methods based on log analysis, page header inspection, inconsistent WAL records and more - how to handle the actual disaster recovery process itself - practical measures you can take to reduce future risk and so much more. Live Q&A afterwards - register now to attend or save it for watching on-demand later. hubs.la/Q04FgBxL0 #databaseadministrators #dba #postgresql #postgres #dataengineering #data #database #technology #techwebinar #techpodcast #techconference

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 13

In two months, the share of enterprises running open-weight AI models in production climbed from 31% to 42%, according to Futurum’s September survey. Phillip Merrick, pgEdge’s co-founder and CEO, sees data sovereignty as the main driver behind that jump, and in a new VMblog byline he predicts it’ll be the most significant issue in corporate AI and cloud computing in 2027. Agentic applications burn through tokens at a rate chatbots never did, and many enterprises hold sensitive data they can’t upload to a cloud they don’t own and manage themselves. He isn’t writing off frontier models, and for some workloads they’ll stay the right call. Which companies come out ahead in 2027? Not the ones with the biggest model budgets, Phillip argues. The rest of his answer, including one step to take before shadow AI turns into a problem, is in the full piece on VMblog: 📖 hubs.la/Q04FdPc70 #DataSovereignty #AgenticAI #EnterpriseAI #OpenWeightModels #AIInfrastructure #CloudComputing #DataSecurity #Compliance #ShadowAI #2027Predictions

pgEdge avatar
pgEdge@pgEdgeInc·Nov 12

How much can you explain about PostgreSQL data tiering in five minutes, with no laptop and no slides? Quite a lot, it turns out. Jimmy Angelakos gave a lightning talk on pgEdge ColdFront doing exactly that at Community Over Code Glasgow 2026, the Apache Software Foundation's conference. ColdFront is an open-source project Jimmy built that moves rarely read PostgreSQL data to Apache Iceberg on cheaper object storage, while your application keeps querying the same table with plain SQL. The cold tier stays writable, too: an UPDATE or DELETE on an archived row is the same statement you'd run on a recent one. In five minutes he covers tiered and decoupled modes, scaling writes out with active-active Spock (pgEdge's open-source multi-master replication extension), and how Lamport's bakery algorithm, formally verified in TLA+, keeps nodes from colliding on the same Iceberg catalog. His write-up goes further: ColdFront running all 99 TPC-DS queries, why he sees so much potential for the Apache data ecosystem and the PostgreSQL community to build together, and what's next now that 1.0.0-beta3 is out. Watch the talk 🎙️ hubs.la/Q04F9YCH0 Read Jimmy's write-up 📖 hubs.la/Q04F9Yt00 #PostgreSQL #Postgres #ApacheIceberg #CommunityOverCode #ApacheSoftwareFoundation #OpenSource #DataEngineering #DataTiering #DistributedSystems #TLAplus #Glasgow #LightningTalk

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 12

Dr. Ibrar Ahmed, a Principal Engineer at pgEdge, has been working on the PostgreSQL core engine since 2006. On Thursday, Nov 19 he’s presenting “When Postgres Breaks: A Practical Guide to Corruption and Disaster Recovery,” a webinar on what to do when a cluster gets corrupted, whether it’s from failing disks, RAM running short, software bugs, or plain human error. He goes through recovery in the order you’d actually run it, using established commands rather than clever tricks. He'll walk you through how to spot the problem in the logs, checksums, or page headers, then how to stop the service before anything else touches the damaged files. After that he shows you how to restore from verified backups and do point-in-time recovery to just before the corruption, or pull out whatever healthy tables you can when a full rollback isn’t an option. He’ll also cover pg_resetwal, which gets a cluster to start again but costs you recent transactions. That one’s a last resort. The session wraps up with ways to make all of this less likely next time: checksums, regular integrity checks, good backup and write-ahead log archiving habits, and high-availability replicas you can fail over to. The session is free to attend and starts at 8 AM PT (11 AM ET / 4 PM GMT). Ibrar’s sticking around afterwards for a live Q&A, and if the time doesn’t work for you, register anyway to get the recording. hubs.la/Q04F9CWv0 #PostgreSQL #Postgres #DBA #Database #DisasterRecovery #DataIntegrity #BackupAndRecovery #SRE #DevOps #SysAdmin

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 12

KubeCon + CloudNativeCon North America wraps up today here in Salt Lake City, which makes this your last chance to stop by our booth (#366) before the Showcase closes. Looking for high availability solutions for running #PostgreSQL on #Kubernetes? The live demo at our booth takes down a Postgres node running in Kube, cuts the network between two nodes, and showcases how the counter stays at zero failed writes while CloudNativePG restarts the pod and Spock catches both sides back up. (And yes - it's monitoring a real cluster with injected faults!) Take a look, bring your questions, and feel free to talk through your own Postgres on Kubernetes setup - our team is there to help. At the very least, feel free to say hi before you head home. 👋 #KubeCon #CloudNativeCon #SaltLakeCity #Utah #CloudNativePG #Kubernetes #K8s #Postgres

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 10

In our Helm chart, every pgEdge node is its own CloudNativePG Cluster, and at KubeCon + CloudNativeCon North America in Salt Lake City (through Thursday, Nov 12) you can watch a pair of them survive a node failure and a network partition with zero failed writes live - then talk to our team about how to replicate it on your infrastructure. In the demo, when a node goes down, CloudNativePG restarts the pod while the app writes through the other node. Cut the network between them and both sides keep accepting writes, then Spock, our open-source multi-master replication extension, catches them up once the link returns. The chart sets up Spock across the nodes at install time, and it can wire existing CloudNativePG Clusters into a pgEdge cluster as well. Matthew Mols built the demo to be more visual than our usual terminal-heavy videos. Everything runs on one laptop. It’s a local kind cluster, so US-East and EU-West are labels and the orders are a demo workload. The chart is open source at hubs.la/Q04DXpKM0, and the loop is playing at our booth (Booth #366) if you have questions about it and want to talk more. #KubeCon #CloudNativeCon #SaltLakeCity #SLC #Utah #PostgreSQL #CloudNativePG #Kubernetes #OpenSource #HighAvailability

1:21
pgEdge avatar
pgEdge@pgEdgeInc·Nov 10

KubeCon + CloudNativeCon North America is underway at the Salt Palace Convention Center in Salt Lake City, Utah, and it runs through Thursday, Nov 12. The Showcase floor opened to attendees at 10:45 this morning, and Booth #366 is where you’ll find the pgEdge team, ready to say hi, show you live demos, hand out some swag, and talk about your #PostgreSQL on #Kubernetes architecture or plans. We’re here! Look for us in Halls A-C and 1-5 on Level 1, open until 5:15 PM today. Local to the Salt Lake City area and heading downtown to the conference? Come on by. 🐘 #KubeCon #CloudNativeCon #SaltLakeCity #SLC #Utah #Postgres #DataEngineering #Kubernetes #TeamCloudNative

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 6

Keeping years of history in PostgreSQL gets expensive. Disk fills up, queries slow down, and vacuum overhead keeps growing... all for data your application barely touches. 🐘 pgEdge ColdFront is an open source tool that tiers old PostgreSQL partitions to Apache Iceberg on object storage, while a transparent view keeps them queryable from Postgres. Your application never has to change a query. (It's currently in beta.) 🧊 In a new walkthrough, Muhammad Imtiaz sets ColdFront up against every supported storage backend: SeaweedFS for local dev, AWS S3, Google Cloud Storage, and Azure Blob Storage (ADLS Gen2). ☁️ One interesting part: once the archiver runs, the April partition no longer exists in PostgreSQL... so what happens when you SELECT from the table? The April row still comes back, read live from Iceberg, right next to the row inserted moments earlier. One query, both tiers. He also flags the gotchas along the way (like why the AWS S3 storage secret needs null in the endpoint slot), and closes with two bonus sections: running the partitioner on its own, and using the compactor to stop small Parquet files from piling up. Find the full walkthrough on our blog 📖 hubs.la/Q04DFMH50 #PostgreSQL #Postgres #ApacheIceberg #OpenSource #DataEngineering #DataTiering #DuckDB #AWS #GoogleCloud #Azure #ObjectStorage #DevOps

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 5

In a three-node multi-master PostgreSQL cluster, one node came back from an in-place upgrade to PostgreSQL 18 looking healthy to the other two nodes, even though it wasn’t applying any of the changes they sent. That’s one of the surprises in the latest post from Ahsan Hadi and Susan Douglas, a follow-up to an earlier post on rolling major-version upgrades. Rolling upgrades add a new-version node and retire an old one, which “assumes you can stand up an extra server for every node you need to upgrade.” A lot of customers can’t. This time, each node gets upgraded where it sits with pg_upgrade --link (Postgres’s major-version upgrade tool; --link hard-links the data files instead of copying them), one at a time, so that node goes down for a few minutes while the other nodes keep taking writes. The test cluster ran Spock 5.0.13, pgEdge’s open-source multi-master logical replication extension, and the upgrade took one node from 17.10 to 18.6 along with its physical standby, which Patroni manages for high availability. The silent node traces back to something pg_upgrade doesn’t carry over. Resuming Patroni also tripped its system ID check, and PostgreSQL 18’s initdb now enables data checksums by default, which matters if your old cluster runs without them. Coming from 15 or 16 rather than 17? Check what the post says about logical slots before you plan anything. Within, you'll also find the commands and checks between steps, including how to confirm the upgraded node is applying changes again before you touch the next one. Read the full walkthrough: 📖 hubs.la/Q04Dw59X0 #postgresql #postgres #postgresql18 #dba #database #opensource #highavailability #replication #patroni #devops #sre #distributedsystems

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 4

Phillip Merrick says our enterprise Postgres cloud platform, pgEdge Starfleet, lets developers “build at AI speed and then deploy anywhere, without finding themselves at an architectural dead end.” The webinar we ran on October 20 shows how that holds up, and you can still register to watch the recording on demand, free. You can start on pgEdge-hosted cloud, move to your own AWS, Azure or GCP account, or go on-premises and air-gapped with pgEdge Enterprise Postgres, all on the same community Postgres platform. 🐘 Nothing gets replatformed. Gil Browdy, who directs engineering for our cloud team, covers how it’s built. Storage and compute are separated in the file system, with Postgres’s own storage layer left in place, which is what lets branches spin up quickly: a branch is a new Postgres server pointed at copy-on-write blocks, with its own connection string and its own MCP and RAG servers, and it’s charged only for the time it exists. Backups come two ways, hot volume snapshots for a near-instant restore and durable backups that ship the write-ahead log (WAL) to object storage for when the host doesn’t survive. The AI tooling is built in. The MCP server (Model Context Protocol, the standard that lets AI agents talk to external systems like your database) is read-only by default, and the RAG server (retrieval-augmented generation) runs hybrid vector and keyword search over your own tables. Both are open source and free to use on Starfleet. pgEdge SafeSession, an open source Postgres extension, blocks DDL, COPY FROM, GRANT and REVOKE on read-only connections, and IP allow lists start closed, so nothing is reachable until you add a CIDR range, set separately for each database, branch and service. Then Paul Rothrock drops a table on a branch live while the production table stays right where it was. The Q&A covers moving an existing Postgres database over with a dump and restore, how hot and durable restores differ, and where to start with branching. Want all the details? Register free and watch on demand: hubs.la/Q04Dmnm80 #postgresql #postgres #webinar #opensource #database #cloud #devops #aiagents #databasebranching #developers

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Nov 3

You don't need to run a Kubernetes cluster or be a CNCF regular to get something out of KubeCon + CloudNativeCon North America, which opens in Salt Lake City next week (Nov 9-12). If you're a platform engineer or SRE, it's where the tools you run every day get explained by the people who maintain them. If you're a developer or DBA whose app just moved onto Kubernetes, it's where you find out what that changes for your data. And if you lead a team deciding what to adopt next, you can compare notes with 9,000+ attendees from 2,100+ organizations who have already made that call. The program has 400+ sessions, and this year it leans into the infrastructure underneath AI: inference, agentic workloads, GPU scheduling and model serving, alongside platform engineering, observability, security, and running stateful systems like databases on Kubernetes. Monday is co-located events, including Agentics Day (MCP + Agents), Platform Engineering Day and Observability Day. Tuesday through Thursday brings the keynotes, breakouts, Maintainer Tracks and ContribFests. Keynote and breakout recordings should be on the CNCF YouTube channel within about two weeks, but a recording can't answer your follow-up question, and the Maintainer Tracks and ContribFests put you in the same room as the people who write the code. Seating is first come, first served... so it's a good idea to build your agenda before you land. We'll be there too. Come by Booth #366 to talk Postgres on Kubernetes: pgEdge is 100% open-source enterprise Postgres that gives you complete deployment flexibility. And yes, it does work with CloudNativePG, and there's a Helm chart if you'd rather start from the manifests. We'd love to meet you. 🐘 Still deciding? The All-Access pass adds Monday's co-located events, and the KubeCon + CloudNativeCon Only pass covers Tuesday through Thursday plus Monday's Project Lightning Talks. Register here: 🎟️ hubs.la/Q04D9Z1r0 #KubeCon #CloudNativeCon #PostgreSQL #Kubernetes #CloudNative #PlatformEngineering #OpenSource #TeamCloudNative

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Oct 30

The main heap for a 100,000-row PostgreSQL table is 47 MB when TOAST moves the big JSON column out of line and 318 MB when it doesn't (heap only, so the toast table's own storage isn't in that number). TOAST, The Oversized-Attribute Storage Technique, is how Postgres keeps large values in a hidden companion table and leaves a pointer in the row. Shaun Thomas's latest PG Phriday starts with a slow event log he once fought with, and the twist is that the problem wasn't necessarily TOAST but the column itself. Counting rows ran in 26 ms against 91 ms and an UPDATE took 62 ms against 128 ms, because small rows pack densely. Pulling one field out of the JSON took 288,127 buffer accesses against 40,754, because every toasted value means a trip to the toast table. So what do you do when the big column is the problem? Splitting it into its own table behind a LEFT JOIN view dropped the main heap from 47 MB to 5 MB, and the planner skips the join for queries that never ask for the column. Lowering toast_tuple_target won't get you there, since rows under TOAST_TUPLE_THRESHOLD never get toasted. And lz4 compressed nearly 30 times faster than pglz in his test, on very repetitive data, so your ratio will vary. The queries and their output are in the full post: hubs.la/Q04CXnkH0 #postgresql #postgres #pgphriday #dba #database #performance #sql #compression #lz4 #devops #opensource #dataengineering #programming #tech

Post attachment
pgEdge avatar
pgEdge@pgEdgeInc·Oct 30

Should you go to KubeCon + CloudNativeCon North America? A quick checklist. Check any that apply: [ ] You run, or are about to run, stateful workloads on Kubernetes [ ] You’d like to be in a room with 9,000+ people building cloud native infrastructure, across 400+ sessions [ ] You want to meet the maintainers behind the projects you rely on (Maintainer Tracks and ContribFests run all week) [ ] You like a head start: Monday has co-located events like Platform Engineering Day, Observability Day and Cloud Native AI + Inference Day [ ] You’d rather not rent a car: the Salt Palace sits inside the downtown free fare zone, and TRAX gets you from the airport in about 20 minutes Checked even one? Put Salt Lake City, November 9-12, on your calendar. pgEdge will be there too, talking Postgres on Kubernetes with the cloud-native community. Come by the pgEdge table in the sponsor area and say hi. Register: hubs.la/Q04CV9kw0 #KubeCon #CloudNativeCon #PostgreSQL #DevCommunity #SaltLakeCity #Utah #TechConference #Tech #Kubernetes #CNCF #CloudNative #CloudEngineering #CloudOps #Containers #Docker

Post attachment