You asked for longer rants, so here are longer rants! After almost 10 years I'm back to blogging. Thanks for the encouragement. Link below in the thread 🔻

I find sharding by customer/tenant a far simpler approach than sharding data by micro service. Multi record changes become simple. Less asynchronous hacks like Saga’s to compensate for multiple consistency boundaries for a business operation. Many orgs can go FAR with this

It’s great fun watching Codex run your benchmarks for you and iterate on fixing performance bottlenecks.

I redid my AI coding fun test asking Codex 5.2 Extra High to implement the Accord consensus algorithm like I did a couple months ago with Claude. Claude took 72h and started promising but code base got messy. Could only do 100rps Codex 5.2 Extra High achieved 14k rps in 8h.

Biscuit is a PostgreSQL index that accelerates LIKE and ILIKE wildcard pattern matching using in-memory bitmap indexing. They use Roaring Bitmaps so you know it is going to be good. https://t.co/W6hcH0DexU https://t.co/HybZjJ5Rpt

I published a new library : https://t.co/VnqATwdkVg

Choose wisely when selecting your hosting provider. The one on the left (blue) is a well-known hosting provider. The one on the right (brown) is Hetzner, which is actually one of Coolify Cloud's servers. The CPU usage is the SAME, but one has waaay more CPU time. 🤯 Guess

Reminder there’s Liftbridge that offers similar added features to NATS as JetStream

Jepsen tests NATS 2.12.1 JetStream messaging guarantees. TLDR: Several different types of anomalies. https://t.co/Yr0qlv4h9m I always love reading Kyle’s work. I always learn so much.

I really struggle booking time off. Not cuz I’m a workaholic but cuz I always seem to be on some critical project under urgent deadlines. Many years I only take 1-2 days off here and there. I just booked 7d in a way that gives me 15 days continuous no work. Can’t wait.

Do you like databases? Do you want to hear two database professors rant about them? Do you need one of those professors to have a Turing Award for databases? If yes, then join Mike Stonebraker and I next Wed Dec 10 @ 1:00pm EST for database hot takes: https://t.co/JXGHtesZzC

I enjoy developing on Mac far more than on Windows for last 10 years cuz of all the native Unix under the hood and with AI tools using Unix tools this gets entrenched even further Gaming on Windows is far superior. Like $10k in computers I would pay MSFT $1k for DirectX on Mac

Uber actually moved 2,600+ MySQL clusters from primary-replica to a Paxos-based group replication model. This is going to be a very useful system design post! You might wanna bookmark this. --- 🚖 1. First, What Was Uber Using Before? Uber originally used the classic MySQL

Anyone here deployed TiDB/TiKV? Can you tell me about your experience? I want geodistributed + strict serializability + low latency self-hosted transactional DB that can make do for occasional analytics queries. Right now, I settle for single-master postgres w/ read replicas.

https://t.co/hy0w1O8OQ2

This is pretty cool - Kafka Connect and Debezium at the heart of Datadog's data replication platform

Everyone will have a major outage during a really important moment. It WILL happen someday. So how will you handle it when it’s your company? The _BARE MINIMUM_ is “we will do what’s needed to make next time smoother”

Technology company leaders are like relationship partners. You only know how good or honest they truly are under adversity. Cloudflare? Extremely honest. They face the music. Yesterdays outage? Not honest at all.

@kellabyte @spimescape You don’t write to s3 on the write path. You simply write to a distributed log on nvme. S3 should only be used to store the pages that are applied after processing the wal

CPU as far as I know tiered caching means writes go to store buffer then to the closest cache (L0) before getting to RAM at some later time. I don’t consider a storage system where only reads go to NVMe first but writes go the long route to object storage first as tiered storage

Local NVMe caching on top of object stores is not the same architecture or benefits as local NVMe for transactional writes replicated to another NVMe. I think this point has been lost in the object store debates in the last couple days.