As well as the built-in scenarios, pgbench allows you to run a custom “transaction script”; making it a viable alternative (for some scenarios) to driving load testing via your application, e.g. using a locust swarm.
It’s not a full blown programming language (or even a DSL really), but you can for example: insert a row, and then update it:
\set uid random(1, 100000 * :scale)
BEGIN;
INSERT INTO foo (user_id, ...) VALUES (:uid, ...) RETURNING id \gset g_
COMMIT;
\sleep 1s
BEGIN;
UPDATE foo SET bar = 1 WHERE id = :g_id;
COMMIT;
You can then point pgbench at your DB, and this script, with say 20 clients for 5 mins:
docker run -it --rm --network=host -v $PWD:/app -w /app -e PGPASSWORD=postgres postgres:15 pgbench -h ... -U ... -d ... -f foo.sql -c 20 -T 600 -j 8
...
...
number of transactions actually processed: 81752
number of failed transactions: 0 (0.000%)
latency average = 146.670 ms
initial connection time = 560.452 ms
tps = 136.360106 (without initial connection time)
The transaction count here is actually the number of times the script ran (if you have multiple transactions inside it).
Look on my works, ye Mighty, and despair! 