That is an understandably automatic control feature to keep the DB instance running well since PostgreSQL doesn't handle connection scaling well. The recommendation is to use pgbouncer, which you can also combine with the GCP cloudsql proxy into a separate container and use that way.

We deploy that onto GKE as a service and have our apps use that. Some setup required but cleaner overall.

I assume CloudSQL is google’s alternative to AWS RDS. Less of an excuse to not make the setting user configurable when the competition (AWS) does.

It's a double edged sword. When you don't limit connections and customers push their servers beyond their limits, some of them then ask why Google didn't do anything to prevent the situation from happening. Ideally, this would be solved by having better failure modes but that's not really the case today.

Most product decisions are not set in stone and can be changed if there's sufficient evidence to suggest that they do more harm than good. There's a lot of work that goes on behind the scenes to keep the service running smoothly and unfortunately/fortunately customers are not exposed to them. Some things that may appear to be simple on the surface are not when you take things like supportability into consideration.

Source: Used to work on Cloud SQL

There's no excuse, it's about a different philosophy.

AWS gives you lots of tuning controls and complexity while GCP tends to hide most of it with easier scaling and performance. It'll vary with individual products themselves but it looks like Cloud SQL is designed with that general outlook.

None of the clouds are perfect, the only viable strategy is to use the best parts of them all that fit your needs.

I'm afraid all of you are getting it wrong. The setting is hardcoded to 100.

Check this out - https://issuetracker.google.com/issues/37271935#comment39 and then the update https://issuetracker.google.com/issues/37271935#comment49

This is not a feature. It is a bug. It needs to be at par with RDS.

As actually stated in the comment you linked to, it is not hardcoded but scales with the instance size to protect against an out-of-memory condition. The lack of it being user configurable is a separate issue, and not a bug but rather a design detail of their managed service.

It's a product decision to not make this configurable to make the service supportable. It's not a technical issue.