We cut our Postgres bill 60% by being boring
- Client
- Ledgerline (spec)
- Year
- 2026
- Niche
- Technical blog
[ The brief ]
Ledgerline (fintech infra) wanted developer credibility, not marketing reach. The spec brief: an engineering post where the confession is the hook and the graphs do the persuading.
[ Craft decisions ]
- →The failure comes first: developers trust posts that start with the mistake.
- →Every optimization names its actual win in dollars or milliseconds, not adjectives.
- →No product pitch until the reader has gotten three usable ideas for free.
[ The sample ]
The $11,000 query
For eight months, our largest RDS instance existed to serve one dashboard, for one team, refreshed every ten seconds, showing numbers that changed hourly. Nobody decided this. It accreted, the way infrastructure bills always accrete, one reasonable-at-the-time choice after another, until finance asks why the database costs more than the engineers.
This post is the audit we should have run a year earlier: three unglamorous changes, zero new technology, and a bill that went from $18.4K to $7.2K a month. No rewrites. No migration war story. Just the boring stuff, measured.
Change #1: we read pg_stat_statements like a bank statement
Sorted by total_exec_time, our top query was the dashboard poll: 41% of total execution time, invoked 2.8M times a month, answer materially unchanged 97% of those times. A 60-second cache, one line of config in front of it, erased 39% of database load. The graph in our monitoring looked like a cliff. We printed it.
Boring optimizations compound. Exciting ones get rolled back.
[ Read it like an editor ]
The confession structure (“nobody decided this”) buys the credibility that lets the numbers land. Developers share posts that make them feel smart, not sold to.