Sleeping at Scale
When hundreds of thousands of Buildkite Agents wake up at once, a simple one-second loop can become a distributed-systems problem. Jitter helps, but where and when you add it matters.
How a flaky test exposed a Redis client use-after-free
A handful of flaky tests turned out to be a use-after-free in the Redis client's C extension. Here's how we tracked it down.
Designing log-navigation tools in the Buildkite MCP server
How we turned raw, messy CI logs into a structured format and a small set of navigation tools so AI agents can debug build failures the way a human would.
Goodbye integers. Hello UUIDv7!
The journey Buildkite took from sequential integer primary keys, through randomly generated UUIDs, to adopting time-ordered UUIDv7 — and the database index tradeoffs along the way.