Incident write-ups and migration notes from production — Kubernetes, AWS, Terraform, and databases. What broke, what I checked, and what changed.
My site went down with 504s and my first instinct was 'out of memory.' The server was fine — AWS's VPC-origin plumbing had silently stopped forwarding traffic. Here's how I proved it, worked around it, and reverted once AWS fixed it.
I had the feature designed, the prompt written, and the IAM key scoped. The one thing standing between me and a working AI feature was a payment method AWS wouldn't accept. Here's the blocker on the India region — and the model-agnostic escape hatch that let me ship the same day.
XIRR Ledger just crossed 50 real portfolios. Here's why I built it, what the early numbers actually mean for retail investors, and where the product goes next.
We migrated Aurora to RDS with zero downtime — then quietly paid for two databases for two months. The fix wasn't more knowledge. It was a forcing function.
Our Aurora to RDS cutover went perfectly — pgcopydb, logical replication, all of it. Then the app wrote to RDS and every insert exploded. The cause: sequences.
Most of us use the same 20% of Terraform daily. The rest turned up two features I wish I'd known sooner: ephemeral blocks and DynamoDB-free state locking.
An idle public IP sat on my EC2 instance for months, billing me for nothing. The fix was simple — I'd just lacked the confidence to do it without downtime.
A MyISAM → InnoDB migration looked routine until bulk CSV ingestion crawled. The slowdown wasn't a regression to undo — it was the engine doing exactly its job.
Everyone treats Aurora as the premium PostgreSQL choice on AWS. But for regulated healthcare workloads, what mattered was how fast CVE patches actually land.
Practical strategies that meaningfully cut my Kubernetes cluster costs on AWS — right-sizing, cluster autoscaling, spot instances, and storage optimization.