Writing
Race Conditions in Reactive Microservices: Beyond the Sleep Workaround
A hard-coded sleep in a Lambda callback handler masking a timing race between a DB write and an external notification. Three ways to actually fix it — and why adding a database index doesn't help.
PostgreSQL Autovacuum: The Hidden Bottleneck in High-Churn Tables
The default autovacuum configuration works fine for small tables. For tables with millions of rows and frequent updates, it silently accumulates dead rows until query performance collapses. Here's what I learned diagnosing a production incident on a 9M-row table.
When Adding an Index Makes Things Worse: Write Amplification in PostgreSQL
Not all indexes help. A production investigation found that a 1 GB index on a 9.6M-row table had been maintained on every write for years — used only 166 times in total. Here's how to identify and eliminate write amplification from redundant indexes.
Never Use Integer for External Partner IDs
A partner API spec declared a field as 'Integer'. We implemented it as Java Integer. Months later the partner returned values beyond 2,147,483,647. A production hotfix across three repositories followed. The fix took one evening; the lesson is permanent.