ci: improve previous caches reuse

[develop7]

• Enables setup-nix to try restoring more previous Nix caches, improving cache reuse
• Lets Stack build jobs restore caches from previous runs

[wolfgangwalther]

Has all of this been done in 411cf43 and 43f552d already, so that we can close this?

Edit: Ah, I think there are more cache restore keys here. Is that still relevant or were you just chasing bugs in cache-nix-action v5?

[develop7]

@wolfgangwalther yeah, still relevant. Couldn't wait for the #3190 to be merged.

[wolfgangwalther]

Enables setup-nix to try restoring more previous Nix caches, improving cache reuse

Lets Stack build jobs restore caches from previous runs

Seems like only the stack part is addressed now. Should this be reflected in the commit message, which still mentions both nix and stack caches, or were you missing those changes for the nix part?

[develop7]

Indeed, I've lost extra keys for nix during rebase. Thank you for the heads-up. Let me fix it in a sec.

[wolfgangwalther]

It seems like since we merged this, loadtest is failing with "no space left on device" errors. I suspect this is because the loadtest is actually restoring multiple caches, i.e. test-loadtest and common - and then runs out of diskspace. Is that a possible explanation, @develop7?

[wolfgangwalther]

Hm. Reading the docs it seems that only one of those caches would be restored. Also https://github.com/PostgREST/postgrest/actions/caches tells me that we currently have two caches only - and only for the macos job. Something is entirely wrong here...

[develop7]

Exactly, (while I was toying multi-layer cache idea in my head) only one cache is restored, not according to the docs, but also to the logs — https://github.com/PostgREST/postgrest/actions/runs/7797378199/job/21263851394#step:5:83

My guess is Github cut their disc space allowance for action runners

[wolfgangwalther]

My guess is Github cut their disc space allowance for action runners

Hm. At least not in general. #3211 seems to fix it, but I don't really know why. I just ran the loadtest on main again and it's still failing. So it does seem to be related to the changes that were made in this PR (which started to fail it) and to the changes in my PR (which seems to fix it). Looking at the logs, there is no cache restored on main at all right now.. so that's a bit strange.