The first LLM bill is a shock, and the instinct — switch to a cheaper model — is usually the wrong lever. Where the money actually goes: the asymmetry between input and output pricing, the context you resend every single turn, and the caching almost nobody turns on.
You cannot ship an LLM feature without measuring output quality, and hand-grading does not scale. Using an LLM as a judge works — if you know about position bias, self-preference, and the scores that quietly drift. A practical guide to doing it honestly.
Rolling, blue-green, and canary are not a maturity ladder where canary is the grown-up answer. Each trades infrastructure cost, blast radius, and rollback speed differently. How they actually work and which fits your constraints.
Rolling, Blue-Green, Canary: Which Deploy Strategy Fits Read Post »
Requests and limits are the two numbers on every Kubernetes container, and most teams set them by copying whatever the last service used. Why CPU limits often make latency worse, why memory limits are non-negotiable, and how to size them from real usage.
Requests and Limits: The Kubernetes Settings Everyone Gets Wrong Read Post »
A process dies with OutOfMemoryError and restarts, and it will happen again. Heap dump analysis tells you what filled the heap: how to capture the dump automatically at the moment of death, and how to read the dominator tree without drowning.
One page load fires 200 queries and nobody notices until production. The N+1 query problem hides behind clean-looking code. Here is how it happens, how to catch it, and the four fixes with their honest trade-offs.
Battery complaints arrive as one-star reviews saying “drains my battery” with nothing reproducible. How to measure actual drain with Battery Historian and Instruments, and the four causes that account for most of it.
Diagnosing What Your App Does When Nobody Is Looking Read Post »
A teardown of every stage between a click and the first pixel: DNS, TCP, TLS, the HTML preload scanner, render-blocking CSS, parser-blocking scripts, layout, paint, compositing — and which stage your slow page is actually stuck in.
What Actually Happens Between the Click and the Paint Read Post »
EXPLAIN ANALYZE output looks like a wall of numbers. Only four of them matter, it reads inside out, and the gap between estimated and actual rows finds most bad plans on its own.