Production debugging

Observability starts with questions

Dashboards do not make a system observable. Good questions do.

~5 minTech
scroll
01 · the shape

More charts can mean less clarity

A wall of CPU, memory, request count, and error graphs can still leave you unable to answer why one customer is stuck. Observability is about asking new questions without shipping new code every time.

trace · one checkout

The useful signal is usually already there.

system steady

Systems stories get more interesting when you can feel the pressure move through the stack.

02 · the move

Emit the story of the request

Logs, metrics, and traces should share ids and vocabulary. That lets you move from symptom to path to cause without guessing.

instrumentation.ts

Shared attributes make telemetry joinable.

03 · the check

Start with user pain

Ask what a user would report, then make sure telemetry can answer it: who was affected, when it started, what changed, and which dependency was slow or wrong.

reliability control room
brittle
resilient0
brittle: the system works only while the weather is perfect

Flip the switches and watch the story turn into a tiny operating model.

0signals: logs, metrics, traces
0request id across all signals
0dashboards nobody uses
04 · the takeaway

Telemetry should shorten the mystery

The point is not to collect everything. The point is to make production explain itself faster.

source trail