How many database connections should a service with 4 workers

You

How many database connections should a service with 4 workers open?

ChatGPT

Start from cores, not workers. A pool of about twice the database core count is the usual ceiling, so 8 to 16 for a small instance. Beyond that throughput flattens and latency grows, because the connections queue inside postgres instead of in your pool.

Tool steps: 3 (reasoning summary, call to python, execution output)

reasoning summary

Checked the query plan, then compared index options.

call to python

rows = run_sql("EXPLAIN (ANALYZE) SELECT ...")
print(rows[:3])

execution output

Seq Scan on events  (cost=0.00..812431.00 rows=41203122 width=16)
  Rows Removed by Filter: 38996878
Planning Time: 0.214 ms
Execution Time: 11842.663 ms

You

Should I store user settings as JSONB or as real columns?

ChatGPT

Columns for anything you filter or join on, JSONB for the long tail nobody queries. The moment you write a WHERE settings->>'plan' = ... that runs often, that key has earned a column and an index.