v14 backport for admin server fix

[steve-chavez]

Backport of #5078 plus required dependencies

[wolfgangwalther]

I believe we should not backport this fix. Way too involved with the many refactors that need to be picked and also due to https://github.com/PostgREST/postgrest/pull/5078/changes#r3563574656 - depending on unreleased upstream commits is not something that we should release. We can potentially do it temporarily on main (although that just blocks the next release now, too...), but we should not do it on the back-branch.

[steve-chavez]

Reminder that this is a critical fix for v14, no other workaround but to restart postgREST for solving #5012.

Way too involved with the many refactors that need to be picked and also due to https://github.com/PostgREST/postgrest/pull/5078/changes#r3563574656

I'll not include #5077 then, which requires the other refactors.

depending on unreleased upstream commits is not something that we should release

Warp had a new release, already included on a71ab36.

Given that I see no other arguments as why this shouldn't be merged.

[wolfgangwalther]

Not sure whether this will be easy to get right - the base set of dependencies used on the v14 branch is rather outdated by now (hint! we should release major versions more frequently! hint! :D).

Cabal builds are failing now, due to ram / memory etc.

A shot in the dark would be doing the same as

postgrest/postgrest.cabal

Lines 116 to 118 in 2051358

	-- crypton 1.1.0 moved from `memory` to `ram`, which jose-jwt fails to build with right now.
	-- should be possible to remove this once jose-jwt had a new release.
	, crypton < 1.1.0

, but I have not looked in more depth at the error messages, nor tested it.

[wolfgangwalther]

I'm not opposed to doing this backport in the current form, not at all...

Reminder that this is a critical fix for v14, no other workaround but to restart postgREST for solving #5012.

... but I do want to mention that just because it's a bugfix and even when its possibly critical, does not mean we automatically need to backport it. We can just release a new major version and point those who need the bugfix at that - that's always an option as well.

[steve-chavez]

Cabal builds are failing now, due to ram / memory etc.

Fixed the above by pinning jose-jwt. Any idea why the loadtests fail? Edit: looks like it was only a race condition.

[steve-chavez]

I'll squash the 4 commits here since we need the all of them to be able to compile. Better for history too.

[steve-chavez]

We can just release a new major version and point those who need the bugfix at that - that's always an option as well.

Yeah but it's hard to justify this for customers on production instances that are already impacted by a bug, then tell them to fix it you have to upgrade to a major that has breaking changes and potentially new bugs.

I think we should call v14 closed once #5096 is done.

(hint! we should release major versions more frequently! hint! :D).

Opened #5098 to settle that. I'd have liked more "community features" for this release to motivate users to upgrade, but I guess this is mostly going to be a "operational maturity" release.