I did some massive rebuild today because I want to try the latest nvidia driver (Actually I don’t fully understand: why the update to nvidia driver causes nix, ghc, rustc, etc, to be rebuilt?), and I notice some derivations always fail to build due to test failure while building in parallel with other derivations. However they can be successfully built when they are the only derivation being built, currently I found pycurl and gom have such flaky behavior. Do you ever meet this situation on hydra? How do you solve it?
Gonna need some minimal repro steps. Changing the nvidia driver is very much not supposed to cause stuff like GHC to rebuild, so something fishy is going on.
Hi @ElvishJerricco, the massive rebuild is triggered by these commands:
$ cat ~/.nix-defexpr/channels/nixpkgs/.git-revision
4cd2cb43fb3a87f48c1e10bb65aee99d8f24cb9d
$ cp -R --no-preserve=all $(realpath ~/.nix-defexpr/channels/nixpkgs) nixpkgs-nvidia-beta
$ curl -o nvidia.patch https://github.com/NixOS/nixpkgs/commit/c27b21e2c92e1163f19d0d9ca9f50694ee60bc4d.diff
$ patch -p1 -d nixpkgs-nvidia-beta < nvidia.patch
$ nixos-rebuild boot -I nixpkgs=$(pwd)/nixpkgs-nvidia-beta
The flaky test should be caused by lack of computing resource on my local machine, I’m now more curious about the root cause of the massive rebuild.
Funny that I see the same pycurl test failures today in the year 2020, month June. 
Well try build it alone or disable the test, timeout failures are hard to forbidden in a massive rebuild