Document global var / setting idle_timeout solutions for long-running development processes eg. Next.js

[karlhorky]

Hi @porsager! 👋

Since a while now, students at @upleveled have been running into an issue where refreshes of a page (and Fast Refresh) are causing further database connections, and the old connections are not going away. This leads to confusing errors like (also mentioned in #42):

Error: remaining connection slots are reserved for non-replication superuser connections

The teardown / cleanup section in the readme with the prexit example appears to be helpful, but process.exit is not actually called during the development workflow (multiple page refreshes and Fast Refresh hot reloads when a file is changed).

As far as I can tell, there are a couple of different solutions to this:

1. Set a global variable that will persist across page reloads and Fast Refresh (like this: Dev mode hot reload loses database connection vercel/next.js#7811 (comment))
2. Set an idle_timeout that will cause the idle connections to be terminated fairly quickly (eg. after 2 seconds)

Would you be open to me creating a section in the readme addressing this problem with option 1 as a recommended solution for development workflows for these long-running apps? With maybe option 2 as an "alternative" solution?

[porsager]

Hi @karlhorky .. Is this with v1 or v2-beta?

Hehe, scratch that, not relevant, I should read closer :) brb...

[porsager]

Ah ok, I see the issue now. I think this is probably an issue for a lot of libraries using next.js. I'm afraid it might be more confusing if information about the global workaround is added to the Postgres.js documentation - what do you think?

[karlhorky]

Yes, on further consideration, I think documenting Option 1 would be misplaced in the docs here. I have asked if they would accept a PR for me to document it in the Next.js docs instead.

However, for Option 2, I would see value in having a short readme section on the fact Postgres.js does not automatically close connections, including example usage of the idle_timeout option (eg. the conversation from #42) - does that sound reasonable?

[porsager]

Yeah, that would be great! Much welcomed 🙂

[karlhorky]

Ah, thanks for f90eb2c, never ended up circling back around to this one 😨

Quick PR to add an example too: #171