error: hash mismatch in fixed-output derivation ‘/nix/store/haq8ky77kr4g5fvwbixinjbr5axhnwx6-cuda_11.8.0_520.61.05_linux.run.drv’:
specified: sha256-kiPErzrr5Ke77Zq9mxY7A6GzS4VfvCtKDRtwasCaWhY=
got: sha256-qNUVrWxgp55at+E9f26AQnKyIU//W73Ugsl5CIIMPQ0=
perhaps the cuda download from nvidia was altered?
What do I do to finish my installation here?
turns out it solved itself awhile later. Not sure if there was an actual update or not?
weird, reinstalling after repartitioning on the same machine I get the same error, and again, it seems to clear up on it’s own after attempting again?!? I don’t really like inconsistent errors, makes me want to test the underlying disks for errors… or memory… my stomach feels upset all of a sudden…
I just had the same problem installing fluent-bit for the last couple of days, and this morning it installed just fine.
I see this commit from 3 days ago changed the hash of the fluent-bit package: fluent-bit: update source hash · NixOS/nixpkgs@c45ceda · GitHub
So my question is - how could the hash change without any other change to the nix package?
The src of that package (and most packages that use fetchFromGitHub) specify a rev. There’s nothing stopping upstream from force-pushing to that rev, changing its hash. Someone with access to Hydra can probably access the cached copy of the previous version of the source and check what changes.
Sooo, this one is the run-file based installer which is something like 4 or 5GB in size and which we don’t support anymore in 24.05/unstable: it’s been renamed into cudaPackages.cudatoolkit-legacy-runfile or something. The reason the FOD sometimes works and sometimes gives the mismatch is likely because nvidia’s servers aren’t reliable enough to transfer the entire 5GB blob. I think we’ve seen this before. This shall remain a mystery because there’s no reason to use the run-file based installation anymore, use normal cudaPackages instead