A database pool with a maximum of ten connections sounds fairly restrained. Put it in twenty application instances and the application can now ask for 200 connections. Add workers, migration tooling and a rolling deployment with old and new instances running together, and the little setting in the application config starts looking rather more consequential.
PostgreSQL’s connection settings define a server-wide connection limit, with reserved slots for particular roles and emergency access. Raising that limit also increases some resource allocations. I would work backwards from what the database can support and the capacity the service actually needs, rather than treating a larger number as a free performance improvement.
Count pools per process, not just per deployment. A web server with several worker processes may create several pools inside one instance. Include the maximum intended instance count and the temporary overlap during a rollout. A pool limit is a ceiling rather than a promise that every connection is always open, but all those ceilings still matter when load arrives together. Leave deliberate room for operational access and the other clients sharing the server.
Then measure waiting as well as connections. Requests can spend their time waiting to borrow a connection even when individual queries look quick. Set a bounded wait that fits the request’s deadline, check that connections are returned after failures and inspect how long transactions hold them. A slow external call inside a transaction can keep a scarce connection occupied while the database has nothing useful to do. I would load-test the intended deployment size and watch pool wait time alongside database behaviour before deciding that the answer is more connections.