Mergify

Ever seen a 392 GB Postgres table with indexes bigger than the data? A routine disk alert sent us digging and we found 150+ GB to reclaim: unread blobs, the same JSON repeated across 200M rows, and bloated indexes. Why it matters: fewer bytes mean faster queries, cheaper storage, and smaller backups. We wrote up the investigation and takeaways. 🧹 👉 https://lnkd.in/eUkatmpV

Thought zero-downtime is always the answer? Dropping two columns on our busiest 4M-row table meant swapping the primary key—and the safest move was a deliberate 10-minute outage. I walk through the decision, the steps, and how to know when downtime beats clever migrations. New post: “Swapping a Primary Key on a 4M-Row Table: Why I Took 10 Minutes of Downtime” ⏱️ 👉 https://lnkd.in/eKx9TGNb

Does your monorepo crawl because a single merge queue makes everyone wait? 🚦 We split the line: scopes let us run parallel merge queues, batch related PRs, and reuse CI across them. Unrelated changes land side‑by‑side without stepping on each other. We wrote up exactly how we set it up on our repo—config, gotchas, results. 👉 https://lnkd.in/en2HQbBR

We're hiring a Staff Backend Engineer. The bar that matters most: you're already AI-native, building with agentic coding tools (Claude Code, Cursor) every day, with real judgment about where they help and where they don't. Strong Python in production and PostgreSQL on top of that. You'd own systems thousands of engineering teams rely on to ship faster: test orchestration, the merge queue, the event-processing engine handling millions of CI events. Fully remote, bootstrapped, profitable. No VC pressure. One catch: France only. Link to apply in the comments.

What happens when you spin up a dozen AI coding agents in parallel? 🤖 I started capped at two; now I run twelve—and the bottleneck moved from writing code to reviewing it. Why it matters: once agents scale, you have to scale review bandwidth, CI signals, and merge flow to avoid PR pileups. I wrote about it in "How Agents Work With Parallel Agents." 👉 https://lnkd.in/dN9AeW-3

Would you let an LLM open a PR in your repo? We did—safely. “Fix with Mergify” asks a model to repair a broken .mergify.yml and opens the change as a pull request you can review. The hard part wasn’t the prompt; it was the guardrails: strict schema validation, write‑access gating, and quotas to keep model access in check. 🛡️ We explain the schema validator, write‑access gate, and quota design that keep a model with repo access from doing damage. 👉 https://lnkd.in/eKRdugPy

Ever tried reviewing a 1M‑line PR? 🤯 Bun’s Rust rewrite landed as a single PR: 6,755 commits, over a million lines, most authored by Claude—in nine days. We break down the four merge operations at that scale and why the PR became the bottleneck, not the code. Plus, what a stacked workflow would have changed. 👉 https://lnkd.in/eSuhraRU

Stacked PRs without teaching your team a new git workflow—sounds nice, right? Here’s how we pulled it off: one local branch, a stable trailer that survives every rebase, four git hooks, and unmodified GitHub. You get small, reviewable PRs with less context switching and no process rewrite. We just posted the breakdown—curious to hear your take. 👉 https://lnkd.in/eV2SZdfk

Ever merge two Alembic migrations and get the dreaded multiple-heads error? We stopped hardcoding down_revision and let Git history decide the lineage. Migrations now follow the order code actually landed—no manual stitching, fewer surprises. 🔧 We wrote up how we did it and why it works. 👉 https://lnkd.in/eVqQut9h

Feature branch behind main right before opening a PR? 🔀 Here are the two commands to catch up—and when you should rebase instead of merge (and why). Plus the conflict cases that trip people up every time, so you don’t lose an afternoon. New post: How to merge main into a branch in Git (and when to rebase instead) 👉 https://lnkd.in/eqMdGqmF