v14: fix: Do not clear the schema cache during retries

[taimoorzaeem]

Backport #4869.

[wolfgangwalther]

I'm not sure why this was labeled fix, but I don't think we should backport this. This just seems like a change from one intended behavior to another. Might not be a breaking change, but still.

[steve-chavez]

I'm not sure why this was labeled fix,

We should amend the changelog on #4869, I remember thinking it needed docs but was conflicted by the fix prefix.

[mkleczek]

I'm not sure why this was labeled fix,

We should amend the changelog on #4869, I remember thinking it needed docs but was conflicted by the fix prefix.

But it does fix #4873, doesn't it?

[wolfgangwalther]

But it does fix #4873, doesn't it?

There's multiple angles to look at this:

• Let's consider A statement timeout can void the schema cache #4873 a bug (which I don't really agree with), then this change does indeed fix that bug. But just because it fixes a bug, it doesn't mean that this only fixes a bug. You can fix a bug in many ways - and introducing a significant change in behavior, where both the previous and the new behavior were explicitly intended and we did not communicate it anywhere that we intended differently, is a "heavier" change than just fixing a bug.
• Quite clearly the underlying discussion in that issue showed that this is just one tiny step in a much bigger change of mind on how to approach the schema cache and down-time. This is not an isolated change that just works nicely on its own and does not need any other considerations or future changes. It's part of something much bigger.

No matter whether you call it a fix or not - this is just not material to backport.

[wolfgangwalther]

it needed docs but was conflicted by the fix prefix.

tbh, I'm not sure whether I follow the reasoning of "it's a fix, so no docs needed" ;)

[steve-chavez]

tbh, I'm not sure whether I follow the reasoning of "it's a fix, so no docs needed" ;)

We never done that for regular fixes since they mean a "correction to match already documented behavior", only for critical fixes (like https://github.com/PostgREST/postgrest/releases/tag/v12.2.12) we've added docs; because they're partly an addition (new behavior). So far that has made sense.

We'll likely do the same for #5012.

Edit: added this clarification on 412c6a1

[steve-chavez]

We should amend the changelog on #4869, I remember thinking it needed docs but was conflicted by the fix prefix.

Updated the docs on 6951e13. I only updated a note that was now wrong. It didn't warrant changing the fix to a feature because there wasn't much behavior change.

Quite clearly the underlying discussion in that issue showed that this is just one tiny step in a much bigger change of mind on how to approach the schema cache and down-time

Yeah I think that this is part of treating the "schema cache as a performance optimization", which makes a lot of sense considering some wrong decisions in the past (like passing it to SQL land #1794). This also further validates postgrest-openapi and #5031.