Sorry for the lack of data related to the configuration, I was in a hurry!
So, both instances are on same region : us-central1 for Cloud Run and us-central1-f for Cloud SQL. (I assume these are the same). I use :postgrex, "0.15.13" & :ecto, "3.7.1" and connect to database through socket socket_dir: System.get_env("POSTGRESQL_SOCKET"). Instance IP assignment is set to Public IP. Tell me if you need more information about this.
Finally found what’s was wrong : I have a process running in background, pulling messages from a GCP PubSub and processing it with some db queries (the ones that were really slow). And it seems that without any incoming traffic from the outside (GCP PubSub pull requests obviously aren’t), the container goes in some kind of sleep mode after a few seconds.
To overcome this, you just need to update your container’s configuration CPU allocation and pricing to CPU is always allocated so the container never goes in “sleep mode” and is always awake to query database without any delay. This was more like a GCP issue than a Elixir one but still, it might be useful for some other ppl.
Nah man, I worked with a lot of stuff but not with CloudSQL. I distinctly remember having used socket_dir once to connect to a local Postgres and it worked. Can’t help you further I am afraid. Must be something in the configuration of CloudSQL or something else.