Nixpkgs: chromium build cycling on Hydra

Created on 30 Oct 2018  路  14Comments  路  Source: NixOS/nixpkgs

The chromium builds are now repeatedly "aborted" on Hydra.nixos.org, and it isn't even clear why. The result is huge waste of resources, and blocked channel due to never finishing.

Examples:

I see it happening on two different machines. @edolstra: any idea how to get some logs from the abortions? Otherwise I can't see how to deal with it. Also /cc @grahamc (Packet machines).

My first guess would be that it exhausts some resource, e.g. /tmp/ space.

blocker

Most helpful comment

We now had multiple successes of the test and chromium itself, on both branches. Thanks everyone!

All 14 comments

Looking at job lists (e.g. trunk), it always started exactly on 69 -> 70 bump, both on master and 18.09, so we could most likely avoid the problem temporarily by reverting that bump.

I now at least manually cancelled most of the jobs, so there aren't too many versions tried at once.

That post seems windows-specific. It's over a year old, and at least at that time I had no such issues when building chromium via nix(os) in background (I haven't tried now yet).

Hm, so it started failing before switching to clang, which is simultaneously good and bad news ;-)
Let's hope, it can be resolved as part of https://github.com/NixOS/nix/issues/2493 as Eelco suggested.

Sorry, that I didn't catch this failure. I have now starred
https://hydra.nixos.org/job/nixos/trunk-combined/nixpkgs.chromium.x86_64-linux and enabled notifications on hydra. That should mean, I get notified of such conditions, right?

Unfortunately, the aborted jobs don't seem to have log files available. Do you have any hint as to what caused it to abort? @vcunat
We don't happen to build chromium on a tmpfs, do we?

I'll keep an eye on https://hydra.nixos.org/build/83226757/nixlog/140/tail
If it goes through, can I re-attempt the version bump right away, or should I wait until the channel finishes?

No, I found no relevant logs.

/tmp(fs): a superficial search in configs suggested that only ike uses it, and that one doesn't have the "big-parallel" feature (so shouldn't get chromium).

OK, that build you linked succeeded :tada: I hope that's it, but we'll see.

The build succeeded but the tests are now failing due to changes in the sandbox outputs.

@bendlas Could you have a look at them?

Here is a screenshot of the VM test that fails:

2018-11-01-132444_1452x695_scrot

The logs from 18.09 before the recent systemd bump: https://gist.github.com/andir/2b64a68bb0bf953ebe02102aca358b44 & hydra log for comparison https://nix-cache.s3.amazonaws.com/log/adaq71i6yc400pm392d4hc04xwl9ick1-vm-test-run-chromium-stable.drv

hydra build: https://hydra.nixos.org/build/83250711

Related problem I see now: https://github.com/NixOS/hydra/issues/591#issuecomment-435125621

I've pushed https://github.com/NixOS/nixpkgs/commit/89ede978a0614fd7215e0aa2c483df142cc79af9 in a desperate attempt to fix timeout issues.

Here's an out-of-space error on packet-t2-4, but those aren't that often.

We now had multiple successes of the test and chromium itself, on both branches. Thanks everyone!

Was this page helpful?
0 / 5 - 0 ratings