Epic: add extensive enough Vercel+Prisma tests in our release pipeline

[kelvich]

Motivation

We already had two cases where some breaking changes were delivered to the big amount of users who consume us though Vercel:

• incompatibility of the new version of pgbouncer and Prisma:
  • https://github.com/pgbouncer/pgbouncer/issues/974
  • https://github.com/prisma/prisma/issues/21635
  • https://neonstatus.com/incidents/01HDP7BA1R1YE13Q8MWFGV01ZV
• simultaneous use of pgbouncer and http proxy: https://github.com/neondatabase/neon/issues/5714

With some app that uses Prisma and Vercel storage driver we could detect it earlier. It should load database pretty well to use several pgbouncer connection simultaneously (otherwise prepared statements over pgbouncer would work).

DoD

We have some early warning system that involves Vercel, @vercel/postgres and Neon that is able to detect such breaking changes. Also makes sense to check the whole matrix of possible combinations (e.g. direct and pooled connection, direct use of @vercel/postgres and direct use of @neondatabase/serverless in websockets/http/http_with_connection_cache modes).

Implementation ideas

Could be our app deployed on Vercel. As part of the long running tests during the release (similar to postgres endurance tests by @bayandin) we could send the load to that app using smth like k6 for 15-30 min.

Tasks

• [ ]

Other related tasks and Epics

[conradludgate]

We have testodrome testing serverless driver on staging, no vercel or prisma driver yet.