Nixpkgs: unstable channels have not been updated in over 10 days

[markuskowa]

Issue description

Both the nixos-unstable and nixpkgs-unstable have not been updated in over 10 days: https://howoldis.herokuapp.com/
However, there were at least some clean evaluations during that time. Any idea why the channels are stuck?

[jtojnar]

There really were no evaluations that succeeded – see the label at the top of the linked page:

This is an aggregate job: its success or failure is determined entirely by the result of building its constituent jobs. [emphasis mine] The table below shows the status of each constituent job for the 15 most recent builds of the aggregate.

[hedning]

Yeah, nixos-unstable seem to have been broken for one reason or another a while (though there's this eval https://hydra.nixos.org/build/86233229#tabs-constituents where I suspect the failing job was restarted, but the success not picked up).

The nixpkgs-unstable unstable job have been green a few times recently though: https://hydra.nixos.org/job/nixpkgs/trunk/unstable#tabs-status (as I understand it the nixpkgs-unstable channel should advance when this job passes, though I might be mistaken here).

[jtojnar]

python37 seems to have been broken by a patch introduced in https://github.com/NixOS/nixpkgs/pull/42925 not applying after https://github.com/NixOS/nixpkgs/commit/552d8a70d4c47a6e52a8e0b26e3a2e33d5b94ca6 on Darwin. cc @veprbl

[FRidh]

About python3 https://github.com/NixOS/nixpkgs/issues/52821

[FRidh]

I expect the nixpkgs-unstable channel to advance after 2534dddaa96af8faa442ba7ac360966c7990e773 is tested.

[markuskowa]

Some failures on nixos-unstable are more mysterious: https://hydra.nixos.org/build/86233229 ran all tests successfully (all green) but failed a depend installer test (https://hydra.nixos.org/build/86233229/nixlog/34). However, looking at the log of the test it is unclear why the test even failed. Every test within succeeded with exit status 0. Any ideas what goes wrong here?

[vcunat]

That's because the things got restarted and succeeded at the second time. Neither nix nor hydra have a concept of storing multiple logs for the same derivation (unfortunately!), so the logs just get overwritten.

Usually one can simply restart the combined jobs, but sometimes for me it happens that restarting a job only transitions it into "user cancelled" state and I can't do anything about it (that one is mystery for me).

[vcunat]

nixpkgs-unstable has had some successes of the job since the last update, so I gather it's again some "invisible" error happening in the script that updates channels. @edolstra: can you post the error somewhere? (not sure who else can see these)

[vcunat]

nixos-unstable got updated now.

EDIT: nixpkgs-18.09-darwin seems to be in the same situation as nixpkgs-unstable. Maybe the two are closely related.

[orivej]

Can we make nixos-unstable be dependent on pandoc? It is really unfortunate that it has advance with +10000 broken packages, all due to the pandoc failure.

[vcunat]

Maybe we'd better add a threshold on percentage o failures, assuming that's also easy to add. EDIT: well, the tested job is easy enough to control from nixpkgs already, that's an advantage.

[vcunat]

All are up to date now. No idea what was blocking the nixpkgs channels.

[8573]

Isn't nixos-unstable stuck again, since 2018-12-30?

[markuskowa]

It looks some of the installer tests are failing again (?).

[8573]

nixos-unstable did update, but now it's stuck on 2019-01-04.