Notes
Working notes
Short write-ups from live work. Published as the thinking happens rather than polished afterwards.
- Your disaster recovery plan has probably never been runWhy documented recovery targets are not recovery capability, which dependencies break a failover, and how to produce evidence a client security review will accept.
- Your disaster recovery plan has probably never been runWhy documented recovery targets are not recovery capability, which dependencies break a failover, and how to produce evidence a client security review will accept.
- Resource requests and limits are probably copied from somewhereHow to derive Kubernetes requests and limits from real usage, why CPU limits often cause the incidents they were meant to prevent, and what to do about memory.
- Prometheus, InfluxDB, Datadog or ELK — which, and for whatA practical comparison of four telemetry backends based on running all of them in production, and the workload characteristics that should decide between them.
- Measure the pipeline before you optimise itWhy the average build time lies, how to get per-stage numbers across enough runs to be meaningful, and the four places the time usually turns out to be hiding.