More posts on cloud cost optimization, FinOps, and engineering.
FinOps
What is FinOps and how do you implement it? The Inform/Optimize/Operate phases, roles, a 90-day plan, Crawl/Walk/Run maturity and the metrics that matter.
FinOps
Fintropy's rule count just crossed 497, up from 470 in May - here's what the new 27 rules check for, and why we built them.
Engineering
Our own cost tool flagged a live AWS load balancer as unused - because a Tier-1 placeholder of 0 was mistaken for a measured zero. Absence of a measurement is not a measurement of absence.
Engineering
Our first production scan caught our own idle Cloud NAT - $32.12/mo for 0 bytes over 7 days - provisioned by Terraform boilerplate but never used by a serverless stack.
Engineering
Google Secret Manager bills every enabled version, per replica, every hour - even when your compute is off. A deploy step with no idempotent upsert quietly stacked ~470 duplicate secret versions in our idle GCP dev account. Here's the three-part fix.
Engineering
When bcrypt 4.0 removed __about__.__version__, passlib's CryptContext started failing our password hashing at runtime. Why we deleted the abstraction and called bcrypt directly.
Engineering
Heavy module-level SDK imports (boto3, numpy, azure, google.cloud) blew past Cloud Run's 60-second cold-start budget and got our Celery worker SIGTERM'd. The lazy-import fix - and the test patches it broke.
Claude
Nuvika Technologies recently joined the Claude Partner Network. A few weeks in, a candid founder's account of what Anthropic's partner onboarding gets right - the documentation, the support, the intent, and the PAM AI assistant built into the portal.
Engineering
A test suite that passed, then failed, then passed - because OneDrive kept syncing node_modules mid-import. How to tell environmental flakes from real bugs, and the vitest flags that fixed it.