fix: do not retry transactions on 40001 SQL error

[laurenceisla]

Should close #3673.

• Fix error upstream: Support for non-retryable transactions nikita-volkov/hasql-transaction#22
• Update dependencies upstream to newer hasql version: Add support for hasql 1.9 diogob/hasql-notifications#30
• Update PostgREST to newer library versions and use a non-retryable transaction to fix issue (this is being worked on in: chore(deps): update hasql to 1.9.3.1 #4829)
• Add tests

[steve-chavez]

Should we vendor hasql-notifications for now since there's no response for diogob/hasql-notifications#30 ?

I remember we did this before too. It's a single file so no major maintenance problem there IMO.

[wolfgangwalther]

We should not vendor it, for two reasons:

• First of all, we haven't even done our own work, yet. We can't update to hasql 1.8 or 1.9 before the static build is sorted out - and it currently doesn't look like there will be progress soon. The upstream Nixpkgs PR that Build static executable for linux aarch64 (w/ TH) #4193 depends on is likely not going to get merged - and currently Nixpkgs is not even able to build pkgsStatic.libpq anmyore. This only worked by accident. Thus, upgrading to GHC 9.6 and then hasql is quite far out of reach for us right now.
• 3 weeks without reaction is not "no reaction", imho. Especially when there has been no additional ping in the PR, yet, etc.

[laurenceisla]

3 weeks without reaction is not "no reaction", imho. Especially when there has been no additional ping in the PR, yet, etc.

Thanks for the ping suggestion (I should've done this weeks ago). Changes are now merged and released upstream!

First of all, we haven't even done our own work, yet. We can't update to hasql 1.8 or 1.9 before the static build is sorted out - and it currently doesn't look like there will be progress soon.

OK, so this would be the only blocker for this PR for now then. Also some stack shenanigans, but will fix after this is sorted.

[steve-chavez]

@laurenceisla Possible to backport? Would it be hard because it depends on a newer nixpkgs?

[laurenceisla]

Would it be hard because it depends on a newer nixpkgs?

Yes, I'm afraid it will result in a similar case as #4702 (comment), due to the dependencies.