An honest look at three production deployments — one Raft cluster, one CRDT-based store, and a quorum-write SQL fleet — and the operational scars they each left behind.
Six teams, six clever proxy-stacks, one unkillable bug. We trace why so many idle-connection failures map back to a five-line sysctl change you forgot to ship.
An on-the-record postmortem of a multi-region active-active build that almost shipped, and the boring fix that finally did.
A field report on logical-replication, JSONB compression and parallel vacuum across 18 production deployments. The wins are real but smaller than the marketing.
A short essay on why the term has stopped meaning anything useful, and a more precise vocabulary for the four very different problems we hide behind it.
Three engineers, one product, four reorganizations in two years. What stayed, what shipped, what they would do differently.
Push-back on the recent anti-monorepo wave. The cost surface most blog posts skip is real, but it isn’t where you think.