Section · 8 articles
Infrastructure
Operations, networking, observability and the things that break at 3am.
For a separate people-operations application of the same measurement discipline, see remote employee monitoring software.
Logs, Metrics and Traces: What Each One Is Actually For
Three signals with different cost curves and different questions. Which one answers what, and why averages and sampling quietly hide your worst problems.
02What Actually Happens When You Run Out of Memory
The process is rarely killed for the reason people assume. Overcommit, the OOM killer, cgroup limits and swap produce different symptoms.
03Timeouts and Retries: How to Set Them Without Making Outages Worse
Retries turn a slow dependency into an outage, and a missing timeout turns one slow call into a stalled thread pool. The interactions that matter.
04TLS: What Is Actually Verified When a Connection Is Established
The handshake checks a chain of signatures, a name and a date — and several things people assume it checks, it does not. Where the gaps are.
05DNS Is Usually the Problem
The joke is a joke because DNS caches at five layers, fails in ways that look like something else, and honours a TTL nobody checked.
06Containers: What Is Isolated and What Is Not
Namespaces hide things and cgroups limit things — and several important resources are covered by neither. The gaps that cause the confusing failures.
07The Backups Nobody Has Tested
A backup you have not restored is a hypothesis. The failure modes that only appear at restore time, and the number that actually matters.
08Rate Limiting: The Algorithms and When Each One Fits
Four algorithms with different burst behaviour, and the distributed case where the naive implementation lets through twice what you configured.
For primary background on this topic, consult Google Site Reliability Engineering resources.