Moving away from Elixir here, you could also enable logs on Postgres and monitor them as you play around with the suggestions above. I have some notes about this here. Please post your findings!

Yes and yes. But given this is a timeout for the pooled connection this means it’s not bound to a particular query. You can e.g. do Repo.checkout(fn -> Process.sleep(:infinity) end) and you’d get that error without any query involved.

IMO, you have to diagnose this problem from the slow query.

Edit: checkout the below explaination of @LostKobrakai.

If you guess that the root cause is slow queries. You can find the slow queries by using the tools provided by PostgreSQL (which you said above). Or, you can use Ecto directly. Here’s how:

Ecto supports metrics which are reported by :telemetry.
You can log slow queries like this - Output slow Ecto queries to logs · GitHub.

(The gist is picked randomly from the Internet…

IEx has some special ways to deal with exceptions. Generally the caller crashes.

I’m also interested in logging the associated query, if there is one. I thought about adding it to the Postgrex.Protocol DBConnection.ConnectionError error message that gets printed but I haven’t been able to figure out where that’s being generated.

The error which gets returned by ecto is

{:error,
 %DBConnection.ConnectionError{
   message: "connection is closed because of an error, disconnect or timeout",
   severity: :error,
   reason: :error
 }}

Unfortunately at that point we are no longer sure if it’s a timeout anymore (it’s just a generic connection error) which makes it less useful to just wrap all Repo methods, though I might still have to do that.