> At a technical level an abstraction will have more call sites to worry about in different contexts, the more wrong the initial abstraction the harder it will

> At a technical level an abstraction will have more call sites to worry about in different contexts, the more wrong the initial abstraction the harder it will be to change. As I recently called it, infrastructure and systems lose agility as they gain dependency and move down the stack. If you have like 1 customer and they have good retries, honestly: fuck everything. Deploy master, in fact, deploy every keystroke to prod. It'll be fine. At the same time, about 30k - 40k FTEs of our B2B customers depend on one of my Postgres instances during business hours and about twice of that during different holiday seasons. Honestly? Nothing touches the system-level settings of these database systems unless we have pondered a change for 2 weeks. And even then we will schedule an approved change over 4 weeks across applicable postgres clusters. The carnage a bad change at this level can cause is ridiculous enough to not be.