Observability

The debugging loop is slow. You cannot reproduce the bug locally, so you add a log line and redeploy. Wrong place. You do it again. An agent alone does not fix this: it still has to pick a mechanism, and it has to know what that mechanism costs. This talk is about three open-source approaches that take the rebuild out of that loop. Each acts at a different point in the software lifecycle, and each inherits the strengths and constraints of its intervention point. ...

A benchmark is a measurement system. It can report a precise number while measuring removed work, an unstable sample, or an uncontrolled machine. This talk argues you should trust a Go benchmark only after answering three questions: Is the compiler measuring real work? Is the sample stable enough? Is the difference large relative to the noise? We work through three layers. Compiler honesty covers dead-code elimination, sinks, constant folding, inlining, timer ordering, and the B.Loop construct that fixes the non-terminating-timer case. Statistical interpretation covers repeated samples, benchstat, coefficient of variation, run-count discipline, and the p-hacking traps that inflate false positives. Environment control covers both local machines and CI: Linux frequency and isolation controls, perflock, benchdiff, and the bare-metal runners that shared CI instances are not. ...

Jonathan Hall and Shay Nehmad had me on Cup o’ Go for episode 167, straight after their GopherCon US recap. We spent about forty minutes on instrumentation: what an APM tool is actually measuring, and the three ways you can get telemetry out of a Go program without asking every team to rewrite their handlers. The part I enjoyed arguing about most was blast radius. Manual instrumentation is precise and tedious, and it never finishes. eBPF buys you one agent covering every language on the box, at the cost of running code in the kernel, where a bad program can take the whole host down with it. Compile-time instrumentation rewrites the syntax tree behind go build through the -toolexec hook, which keeps failures scoped to one service and works anywhere Go compiles. People love this dark magic, which I find either reassuring or alarming depending on the day. ...

Note This post was originally published on the OpenTelemetry blog on 2026-07-16. I wrote it for the OpenTelemetry community; this is a mirror on my personal blog with some added commentary. Why this milestone matters to me I have been involved with the Go Compile-Time Instrumentation SIG since it kicked off at the start of 2025, when Alibaba and Datadog decided to pool their efforts into a single vendor-neutral project. Watching it reach v1 last week felt genuinely satisfying — not because “v1” is a magic number, but because it marks the point where we stopped calling it experimental and started calling it something people should actually run in their builds. ...

Every now and then a project surprises you by remembering something you said years ago. This week was one of those weeks. A talk I gave at PromCon Online 2020 — The Zen of Prometheus — has quietly become part of the official Prometheus documentation. I am still sitting with it. Where it started The talk was born in the strangest year of my career. PromCon 2020 was online, like everything else. I was a few years deep into running Prometheus in anger, collecting scars from instrumenting services that didn’t want to be instrumented and writing alerts that kept me up at night for the wrong reasons. I wanted a way to package those lessons that wasn’t another forty-slide deck of bullet points. ...