`--keep-going` does not keep going with remote builders involved #928

New issue

Closed

opened 2025-07-22 11:05:15 +00:00 by raito · 2 comments

raito commented

2025-07-22 11:05:15 +00:00

Owner

Describe the bug

Consider:

{ pkgs ? import <nixpkgs> { } }: 
{
  a = pkgs.runCommand "test" { } ''
    sleep 1; exit 1
  '';

  b = pkgs.runCommand "unfailing" { } ''
    sleep 3
    echo toto > $out
  '';
}

If you do nix-build on such a file with keep-going, you expect that the a failure should not abort the build, including the one of b which should succeed.

Steps To Reproduce

Have remote builders
Use the above reproducer.
Build it with keep-going.
See instantaneous abortion.

Expected behavior

A workaround is to disable the remote builders and would produce the following output.

❯ nix-build --keep-going test.nix --builders ''
these 2 derivations will be built:
  /nix/store/4wa1bkwchh29037ccjjbxi14nwkf45mr-unfailing.drv
  /nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv
building '/nix/store/4wa1bkwchh29037ccjjbxi14nwkf45mr-unfailing.drv'...
building '/nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv'...
error: builder for '/nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv' failed with exit code 1

`nix --version` output

Reported happening by @mweinelt, confirmed by @raito on:

nix (Lix, like Nix) 2.94.0-devpre20250720_2fc47b6
2.94.0-dev20250720162504-raito-edition

Additional context

This is very most likely related to the RPC changes.

## Describe the bug Consider: ```nix { pkgs ? import <nixpkgs> { } }: { a = pkgs.runCommand "test" { } '' sleep 1; exit 1 ''; b = pkgs.runCommand "unfailing" { } '' sleep 3 echo toto > $out ''; } ``` If you do `nix-build` on such a file with keep-going, you expect that the `a` failure should not abort the build, including the one of `b` which should succeed. ## Steps To Reproduce 1. Have remote builders 2. Use the above reproducer. 3. Build it with keep-going. 4. See instantaneous abortion. ## Expected behavior A workaround is to disable the remote builders and would produce the following output. ``` ❯ nix-build --keep-going test.nix --builders '' these 2 derivations will be built: /nix/store/4wa1bkwchh29037ccjjbxi14nwkf45mr-unfailing.drv /nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv building '/nix/store/4wa1bkwchh29037ccjjbxi14nwkf45mr-unfailing.drv'... building '/nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv'... error: builder for '/nix/store/slxpnb0l1q7i9ljflvcdsdzxnhwsply1-test.drv' failed with exit code 1 ``` ## `nix --version` output Reported happening by @mweinelt, confirmed by @raito on: - `nix (Lix, like Nix) 2.94.0-devpre20250720_2fc47b6` - `2.94.0-dev20250720162504-raito-edition` ## Additional context This is very most likely related to the RPC changes.

👍 1

raito added the

labels

2025-07-22 11:05:15 +00:00

mweinelt commented

2025-07-22 11:38:48 +00:00

Member

Works as expected on ca12657a68.

Works as expected on ca12657a6805045a7e81b2ea761e1af66c4ea1a7.

pennae referenced this issue

2025-07-22 12:55:21 +00:00

SSH connection dropping when copying outputs should not abort the entire build tree #922

lix-bot commented

2025-07-22 13:32:38 +00:00

Member

This issue was mentioned on Gerrit on the following CLs:

commit message in cl/3705 ("libstore: fix remote build failures killing everything")

This issue was mentioned on Gerrit on the following CLs: * commit message in [cl/3705](https://gerrit.lix.systems/c/lix/+/3705) ("libstore: fix remote build failures killing everything")