Observability and on-call

Metrics cost models, sampling, SLOs, alerts, runbooks, postmortems: the difference between having telemetry and being able to debug at 3 AM.

Read in order: later instalments assume the earlier ones. 10 more instalments in the publish queue.

  1. 01 Your trace has a gap exactly where the bug is

    [2026.10.26] • [5 min read]

    Why context stops propagating across thread pools, queues and retries, why the missing span is always in the interesting part, and why a trace that looks complete can still be lying about causality.
TTFB: -- ms LOAD: -- s PAYLOAD: -- kb