#observability

The bug that wasn't in the mesh

“Write your root cause as a sentence you'd be willing to have your own logs contradict. Mine did, and the cure was 30 seconds of debug logging — not another rebuild.”

Kill by message, not by ownership — and a node is a state

“Which OS process happens to own a node should have nothing to do with who's allowed to operate on it. A kill is a message, not a syscall.”

Make the cross-mesh view honest

“A dependency graph is only as honest as the spans under it. If a service name is too coarse, nodes merge; if a receiver emits nothing, edges vanish; if you cc the observer, fake edges appear. Fix the emission, not the picture.”

Watching a node boot, then a second mesh with no bridge

“A node that does work without leaving a trace is a bug. Start with the trace; everything else hangs off the spine it gives you.”

What good IoT observability looks like in CloudWatch

“An IoT product without per-device-percentile latency dashboards is a product that doesn't know it's broken.”