Practical notes from production work: architecture decisions, database performance, reliability, and business impact.
Registering an .edu.ph domain through PHNET requires government recognition papers and an authorized letter on school letterhead. Then I accidentally claimed the domain under a Google Workspace Business account and had to contact support to undo it.
What I actually got from PostHog beyond page views, and why Upstash Redis made more sense than self-hosted Redis for edge rate limiting.
How I structured request filtering in middleware to block path traversal, SQL injection probes, and rate limit abuse without touching the database.
How I structured role-based access across middleware route guards, server action checks, and UI rendering without maintaining six separate permission tables.
What failed at 20K to 30K record uploads, what changed, and why raw MySQL2 became the better production path.
Why individual connections fail under concurrent load and how a proper pool configuration fixed it.
Found 101 queries per page load on a report used every morning. How I brought it down to one.
Processing 20K records one row at a time is a different problem than processing them in batches.
Index selection based on actual query patterns, not guesswork. EXPLAIN first, add second.
Validation, deduplication, batch fetching, bulk writes, and inventory sync in one transaction.
What O(n2) looks like at 20K records and what one Map object does to fix it.
How to use Prisma-generated types and RowDataPacket interfaces with raw MySQL2 queries.
A practical optimization pass using indexes, JOIN refactors, and batch queries on a live multi-brand system.
The client wanted 7 more brands on the same system. The obvious move was 8 forks. Here's why I didn't — and what I hit instead: an ORM that failed silently at 20K records, a server running at 25% CPU, and a manual Excel workflow costing the operations team weeks every month.
How we moved from single-process bottlenecks to stable multi-core utilization with better uptime.
A chunked upload pipeline that prevented crashes and made concurrent uploads reliable.
How we replaced manual weekly reconciliation with transactional updates and audit-safe sync logic.
Practical caching rules that reduced redundant requests and improved day-to-day dashboard responsiveness.
No cloud shortcuts, no managed database. Just hard limits, predictable behavior, and careful operations.
Role-based data access and workflow design that kept one interface useful across five different teams.
A delivery approach focused on bottlenecks, adoption, and measurable outcomes instead of feature count.
New production notes added regularly.
