About
As a seasoned leader in Developer Relations, I'm dedicated to unlocking the potential of…
Articles by Matty
-
Was it the model, or just a fresh pass?
Was it the model, or just a fresh pass?
So I started a newsletter. The short version: I needed a place to keep my writing muscle from atrophying, and somewhere…
2 Comments -
Your database has two bills. You're only tracking one
Your database has two bills. You're only tracking one
The cloud invoice went up 40% last quarter. Finance noticed.
-
Database optimization isn't free. It's just not on the invoice.
Database optimization isn't free. It's just not on the invoice.
I've had the same conversation probably a dozen times. A senior engineer tells me they can fix a performance problem…
-
Five warning signs your database needs different architecture
Five warning signs your database needs different architecture
We've watched engineering teams do everything right and still end up in the same planning meeting six months later…
1 Comment -
Autovacuum Is the Tax You're Always Paying
Autovacuum Is the Tax You're Always Paying
You've had the autovacuum conversation at least twice this quarter. Not because something broke.
1 Comment -
Two Kinds of Database Performance Problems
Two Kinds of Database Performance Problems
I've sat in a lot of post-mortem meetings where someone says "we just need to optimize better." Usually they're wrong.
3 Comments -
Why Postgres Writes 3x More Than You Asked It To
Why Postgres Writes 3x More Than You Asked It To
You're inserting 100 MB/sec of sensor data into Postgres. Your monitoring dashboard is cheerfully reporting 300-400…
7 Comments -
What the ClickHouse Benchmark Doesn't Tell You
What the ClickHouse Benchmark Doesn't Tell You
The benchmarks aren't lying. ClickHouse wins by 10x-100x on typical analytical patterns against vanilla Postgres.
1 Comment -
Why Your Database Migration Gets Harder Every Month You Wait
Why Your Database Migration Gets Harder Every Month You Wait
I've watched the same pattern play out at dozens of teams running high-volume workloads on vanilla Postgres. The…
2 Comments -
Why Your Read Replicas Stopped Helping With Write Latency
Why Your Read Replicas Stopped Helping With Write Latency
The Fix That Worked (Until It Didn't) You added a read replica and watched the metrics improve. Dashboard queries…
3 Comments
Activity
7K followers
-
Matty Stratton shared thisMy blog is older than most of the software running it. 2,630 posts, dating back to 2001, dragged through LiveJournal → WordPress → Hugo → now Astro 5, with every URL preserved byte for byte because breaking a decade-old inbound link out of laziness is not a hill I’m dying on. Today on my day off, I merged three repos (blog, speaking site, dev.to sync tool) into one monorepo using git subtree add, not a squash. Real commit history, not “trust me, the files are here now.” My dev.to sync tool runs on ownership, not direction. Everyone tries to build bidirectional sync between a platform and git, keep both sides in lockstep. Don’t. You get an infinite phantom-diff loop where each side insists it has the real version. A post is dev.to-owned while drafting. Once a human merges the import PR, it’s git-owned forever. Ownership transfers exactly once, on purpose, and never flips back. Also found an unhinged bug: dev.to’s Ruby backend was rejecting my posts with a 422 because a YAML round-trip silently unquoted a timestamp, and Ruby’s parser auto-casts unquoted ISO-8601 strings into a Time object its safe loader then refuses to load. Two YAML implementations, quietly disagreeing about what a bare string is allowed to become. Cool system we’ve all agreed to trust. Full writeup, plus the Netlify gotcha that reads like a failure in your inbox when it isn’t, and the astro check bug that gave me 200 false type errors from an unrelated subproject: https://lnkd.in/gP2W3nvgI Contain Multitudes (and Also Three Git Repos) · Matty StrattonI Contain Multitudes (and Also Three Git Repos) · Matty Stratton
-
Matty Stratton shared thisGood piece going around that traces, PR by PR, how ClickHouse rebuilt its join engine between 2022 and 2026. If your mental model is "ClickHouse can't do joins," it's worth revisiting. https://lnkd.in/gbMp-PfH The short version: in 2020 the criticism was fair. One join algorithm, no disk spilling, no cost-based optimizer. Big join, OOM crash. By 2026 it's six algorithms, automatic join reordering, runtime bloom filters, the works. None of it requires manual tuning. What I like about the writeup is the method. No toy benchmarks. Just the commit history, with PR numbers you can go read yourself. The broader lesson has nothing to do with ClickHouse specifically. We all carry around database opinions that were true once and quietly expired. "Postgres can't scale." "Mongo loses data." "Joins are slow in X." Some of those were accurate in 2018 and are now just folklore we repeat because we never re-checked. Engineering tools move. If you're going to criticize one, criticize the version that ships today, not the one you tried three years ago. Trust me, I've lost arguments to my own outdated takes more than once. What's a database belief you held way past its expiration date? #ClickHouse #Postgres #Databases #DataEngineeringAre ClickHouse JOINs Slow? A 2026 PR-by-PR AnalysisAre ClickHouse JOINs Slow? A 2026 PR-by-PR Analysis
-
Matty Stratton shared thisYour coding agent didn't break production. Your pipeline did. I keep running into the same setup: "passed CI" and "safe for a real user to see" quietly treated as the same gate. Those have never been the same question, not for humans and not for robots. We just used to have enough friction in deploys that nobody noticed the gap. Wrote up how I actually handle this: a hard gate between merged and live, automated checks that run after the deploy has actually landed (not just "build succeeded"), and a revert path that only works because the migrations behind it are boring on purpose. None of this is agent-specific. It's stuff engineering orgs have known for well over a decade. Taking a human out of the loop, or just making them faster with an agent, makes the gap you were quietly covering for impossible to ignore. https://lnkd.in/g6tUe-H2 What's actually gating your releases right now? A person, a flag, a cron job, or vibes?
-
Matty Stratton shared thisA benchmark crossed my feed that I have to share, mostly because it lines up with what we see every day. Three databases, observability data, and the question of which one actually serves a dashboard. https://lnkd.in/gMpPgXWq The headline number: TimescaleDB won 96% of the tests. Filtered log queries in under half a millisecond at 100K records. Fifty concurrent queries holding at 6ms while the columnar option climbed past six seconds. I want to be fair, and so was the author. ClickHouse wins at billion-row analytical scans across months of history. That's real, and it's the workload it was built for. Mongo wins a couple of trace-by-ID lookups at small scale. Nobody got dunked on here. But observability is mostly point lookups and short time ranges hit by a lot of people at once. That is exactly where hypertable partitioning and continuous aggregates earn their keep. It's why TimescaleDB, which powers Tiger Cloud, holds flat under concurrency instead of falling over. Tattoo this on your brain: "Fast" is not a property of a database. It's a property of a database on your workload. So test your workload. Mostly dashboards, or mostly deep analytics? What does your observability query mix actually look like? #Observability #TimescaleDB #Postgres #TimeSeriesI Benchmarked TimescaleDB vs ClickHouse vs MongoDB for Observability Data - The Results Surprised MeI Benchmarked TimescaleDB vs ClickHouse vs MongoDB for Observability Data - The Results Surprised Me
-
Matty Stratton shared thisYour shell config used to have one audience: you. That stopped being true a while ago. If you're running Claude Code, or any coding agent that shells out to run commands on your behalf, your terminal now has a second audience, and it wants completely different defaults than you do. It doesn't want an interactive confirmation before `rm` deletes something. It doesn't want your fancy prompt burning tokens. It doesn't want `thefuck` popping up to "helpfully" suggest a correction. I wrote up a tour through my actual .zshrc, the parts worth explaining and not just the whole file dumped on you. The centerpiece is a block I call AGENT_MODE: it detects when a non-interactive process is running the shell (CI, an agent, whatever) and flips a bunch of defaults to match, force flags on file commands, quiet flags on package managers, no beeps, a flat boring prompt instead of starship. Also in there: pulling secrets out of macOS Keychain instead of plaintext dotfiles, a one-line alias that fixes a genuinely annoying `gh` CLI auth collision, and a shortcut function that spins up Claude Code against a GitHub issue in its own worktree. If you've never audited what your shell actually does when something other than a human is driving it, this is worth ten minutes. https://lnkd.in/gYSDP6Si
-
Matty Stratton reposted thisMatty Stratton reposted thisA 10,000-row metadata table is quietly dragging your largest queries into the wrong performance tier. Not the fact table or the indexes. The metadata table. That's one of the parts of a Postgres workload that don't announce themselves. Our July 1 newsletter covers four more: ▪ What you actually own when you self-host TimescaleDB ▪ Why model quality was never the gate for fleet-scale Physical AI ▪ Why ORDER BY stops being free once your index outgrows shared_buffers ▪ How a 3-person Danish energy startup runs a licensed energy company on 99.3% compression Plus new integration guides for EMQX and Node-RED, and a July 22 hands-on IIoT workshop where you'll build a real MQTT-to-Grafana pipeline. Full edition: https://lnkd.in/ekEDbmSk #PostgreSQL #TimescaleDB #DataEngineering #IIoTWhat you're really owning when you self-host TimescaleDBWhat you're really owning when you self-host TimescaleDB
-
Matty Stratton reposted thisMatty Stratton reposted thisMost engineers think evaluating TimescaleDB is a database decision. By the time it's in production and customers depend on it, they've found out it was a platform ownership decision. Availability, recovery, scalability, security, lifecycle management. None of it originates with the database. It originates with the application, and it lands on your team as systems you have to design, staff, and maintain for as long as the app runs. TimescaleDB is still TimescaleDB. Postgres is still Postgres. What changes is the operational commitment required to deliver the guarantees the application now demands. None of this work is beyond a capable team. The real question is whether owning that platform is the highest-leverage thing your engineers could be doing. https://lnkd.in/gDXucbXc #PostgreSQL #TimescaleDB #PlatformEngineering #TimeSeries
-
Matty Stratton shared thisEverybody's running multi-model AI review workflows now. Write the code with one model, hand it to a different one to review, maybe run it through three and let them argue. And it works. Stuff gets caught. I'm just not convinced the model switch is what's doing the work. Run the same model twice, once to write and once to review with a fresh prompt and a clean context window, and it still catches things. So was it the different model? Or just a second pass with fresh eyes and a different job? We've seen this movie before. "We moved from Chef to Ansible and everything got better" was rarely about Ansible. It was about it being your second attempt, when you were already smarter. That's the main take in the first issue of Uncommitted, my new monthly newsletter. It's not a work thing, just me writing about whatever doesn't have a home elsewhere. This issue also gets into the quietly great parts of Postgres 19, the "AI era" marketing cycle, and a dangerously stacked September concert calendar. Be honest about what's actually doing the lifting. Read the first one, and tell me where I'm wrong.
-
Matty Stratton shared thisA team just wrote up moving their cache layer off Postgres and onto ClickHouse. 20 billion entries a day, 250ms average latency. Reading it, my first reaction was: "yeah, that tracks." https://lnkd.in/gG92pA-8 Here's the thing, they didn't move because Postgres is bad. They moved because their workload collapsed down to exactly one query shape. Look up a cache entry by four key values. That's it. No joins. No transactions. No updates that mattered. When your access pattern is that narrow, a sparse columnar index beats a B-tree that grows with every row. That's not a knock on Postgres. That's picking the right tool for a very specific job. What I appreciated is what they were honest about giving up. Updates became inserts plus async dedup. They built a materialized view just to make main-branch queries behave. They traded operational simplicity for raw lookup speed, on purpose, with eyes open. That's the whole game. Not "which database wins." Which tradeoffs you can live with. Most teams don't have a one-query workload. If yours does, this is a good read on doing the move carefully. If it doesn't, the lesson is the same: know your access pattern before you pick your database. What's the narrowest workload you've ever over-engineered? #Postgres #ClickHouse #DataEngineering #DatabaseDesignHow We Ditched Postgres for ClickHouse to Process 12 Billion Caches Per Day | MomenticHow We Ditched Postgres for ClickHouse to Process 12 Billion Caches Per Day | Momentic
-
Matty Stratton liked thisMatty Stratton liked thisAfter three and a half months of highly intentional "employment discontinuity," the battery is officially at 100%. Since stepping away from the arena in mid-March, I’ve been living deliberately. I dove headfirst into a pile of long-delayed passion projects—wrestling with vintage computing hardware (see the awesome picture of me with Lee Felsenstein - the inventor of the Osborne 1 - as we discuss my unique and well hacked Osborne 1), optimizing home automation logic, and spending time mentoring the next generation of security professionals. I’ve also kept a hand on the pulse, participating in advisory groups and engaging with the security community. It has been exactly the recharge I needed. Now, it’s officially summer. My wife teaches kindergarten and operates on a well-earned school calendar is off for the next seven weeks, so the immediate priority is to enjoy this time together before the autumn rush. That said, the horizon is starting to look interesting again. I’m beginning to think about my next challenge. I am not in a rush, I’m looking for the right fit where I can build, lead, and drive high-leverage impact. If you want to talk strategy, hyper-growth engineering, or just debate the merits of secured infrastructure in a world of Harnesses and Inference, my inbox is open. I’ll be at the Black Hat USA Briefings in Las Vegas this August—ping me if you’d like to reconnect in person. Stay deliberate and precise.
-
Matty Stratton liked thisMatty Stratton liked thisA 10,000-row metadata table is quietly dragging your largest queries into the wrong performance tier. Not the fact table or the indexes. The metadata table. That's one of the parts of a Postgres workload that don't announce themselves. Our July 1 newsletter covers four more: ▪ What you actually own when you self-host TimescaleDB ▪ Why model quality was never the gate for fleet-scale Physical AI ▪ Why ORDER BY stops being free once your index outgrows shared_buffers ▪ How a 3-person Danish energy startup runs a licensed energy company on 99.3% compression Plus new integration guides for EMQX and Node-RED, and a July 22 hands-on IIoT workshop where you'll build a real MQTT-to-Grafana pipeline. Full edition: https://lnkd.in/ekEDbmSk #PostgreSQL #TimescaleDB #DataEngineering #IIoTWhat you're really owning when you self-host TimescaleDBWhat you're really owning when you self-host TimescaleDB
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Licenses & Certifications
-
-
Microsoft Presales Technical Specialist, Private Cloud/Management and Virtualization
Microsoft
Issued -
Microsoft Certified Solutions Associate
Microsoft
Issued -
ITIL Version 3 Foundation
Examination Institute for Information Science
Issued -
Microsoft Certified Technology Specialist ﴾MCTS﴿
Microsoft
Issued -
Microsoft® Certified IT Professional ﴾MCITP﴿
Microsoft
Issued -
Microsoft Certified Systems Administrator
Microsoft
Issued -
Microsoft Certified Systems Engineer
Microsoft
Issued -
Microsoft Certified Professional
Microsoft
Issued
Publications
-
Developing Custom Apps for the Cloud
Better Software Magazine
See publicationWith the cloud providing tremendous freedom like instant deployment of updates, you're definitely going to have to adjust how you develop and deploy apps. Pete and Matt have created a list of things you need to consider when developing apps for the cloud.
-
Housing Service Delivers New Features Faster with Hyper-V-based Cloud Solution
Microsoft.com
See publicationCase study published by Microsoft describing the migration of Apartments.com from the VMWare virtualization environment to a Hyper-V based cloud solution.
Projects
-
DevOpsDays Chicago
-
Arrested DevOps
- Present
-
Website Redesign for Chicago Photography Center
-
Volunteer project to completely revamp web presence for Chicago-based photography training center. Migrated all content from previous site into a WordPress-based CMS using a custom theme. Utilized Event Espresso event management WordPress plugin to manage all class registrations.
Other creators -
Honors & Awards
-
Serena Innovation Award for Orchestrated IT
Serena Software
Languages
-
English
Native or bilingual proficiency
Organizations
-
DevOpsDays
Global Core Organizer
- Present -
DevOpsDays Chicago
Organizer
- Present
Recommendations received
40 people have recommended Matty
Join now to viewOther similar profiles
Explore more posts
-
Grab The Bull Software
Validate partner CSVs from SFTP nightly and load clean data into Airtable, with instant error alerts. Today’s idea automates the intake of files from an SFTP server. Each night, it pulls a CSV file, validates header structure, required fields, and order ID uniqueness. If any issues are found—like missing columns or bad data types—it sends a Slack summary and creates a Jira ticket with details. Clean rows are uploaded directly to Airtable (or Postgres), and logs are retained for audit purposes. Nodes/APIs: • Cron • SFTP • Spreadsheet File • Function • IF • Airtable • Slack • Jira Benefits: • Eliminates manual CSV QA and rework • Prevents bad data from entering downstream systems • Speeds order ingestion Interested in exploring how this n8n workflow could save your company time & money? Leave a comment below or send us a DM. #n8n #automation #nocode
-
Ciroos
Reactive SRE will always be expensive. Here's why. The cost isn't just downtime. It's the engineering hours spent on alert triage, log parsing, and manual investigation -- work that never ends because the model never changes. Faster tools, better runbooks, smarter alerts -- these optimize the fire department. They don't stop the fires and the firefighting tax stays. Proactive SRE changes the economics entirely. Fewer incidents, engineers who can actually build, and infrastructure that gets more resilient over time -- not just better at recovering. For the first time, AI makes this shift genuinely possible. The question isn't whether to move from reactive to proactive. It's when. Full breakdown at the blog in the comments.
3 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content