| 27 Oct 2022 |
Rick (Mindavi) | Might be worth trying that, if you can find it | 11:33:46 |
magic_rb | maybe im somehow eating those logs somewhere, let me check | 11:33:50 |
magic_rb |
export PATH=${with pkgs; makeBinPath [ hydra-package nettools jq ]}:$PATH
sv -v -w 0 once hydra-init
[[ ! -e ${baseDir}/.init-hydra ]] && exit 1
export PATH=${pkgs.nettools}/bin:$PATH # Hydra runs some variant of `hostname --fqdn`, which BusyBox doesn't support
HOME=~hydra exec chpst ${optionalString cfg.adjustNiceness "-n +5"} -b hydra-evaluator -u hydra:hydra ${hydra-package}/bin/hydra-evaluator
nope, that should be what is doing the eval if I understand Hydra's architecture right | 11:36:48 |
das_j | Are you not using systemd? ๐ | 12:16:54 |
das_j | That looks like linux from the 90s | 12:17:07 |
magic_rb | im not using systemd | 12:19:14 |
magic_rb | and it is runit so your time period guess is correct | 12:19:31 |
das_j | Do you see any logs from the evaluator? because it should log once it starts an eval | 12:20:25 |
magic_rb |
(systems:master) Evaluating...
accepted connection from pid 2869996, user hydra
I only see this in the logs, first line is evaluator second is nix-daemon | 12:21:35 |
das_j | I remember some stdout/stderr magic but I'm not sure if we still have that | 12:22:49 |
@nixinator:nixos.dev | In reply to @magic_rb:matrix.redalder.org Hi! I'm running my own Hydra instance on a non NixOS host essentially and I've never gotten evaluation logs out of it, it just fails as shown here on this link https://hydra.redalder.org/jobset/systems/master#tabs-errors where should I start digging as to why it happens? The module I'm using is heavily based on the NixOS one though so I have no idea why it's misbehaving... I might asking the obvious, but why is this a non nixos machine? There is probably a very good reason for this . | 13:08:54 |
@nixinator:nixos.dev | * I might be asking the obvious, but why is this a non nixos machine? There is probably a very good reason for this . | 13:09:13 |
magic_rb |
In reply to
nixinator
* I might be asking the obvious, but why is this a non nixos machine? There is probably a very good reason for this .
not really tbh, i just decided at one point i want to run it in a container because why not. | 13:10:52 |
magic_rb |
I'm starting to be really surprised this works at all, I don't have git in the module mentioned anywhere. It's not in PATH, neither is Nix which is fascinating | 13:41:32 |
magic_rb |
nix correctly reports a failed eval from the terminal | 13:45:00 |
@nixinator:nixos.dev | This is probably how the container is handling certainly file descriptors, or something is not quite what Perl expects . | 13:55:39 |
magic_rb |
i've no clue, i dont see a reason why it should be misbehaving nor do i see any errors or anything like that | 13:56:48 |
@nixinator:nixos.dev | Without seeing the logs, or reproducing the bug myself there not much I can do. Also beware that some process know when they are being run from the interactive cli or spawned from unit/systemd .. so bear that in mind . | 14:34:39 |
magic_rb | i sent all the logs i have, nothing indicates any failure | 15:04:58 |
@nixinator:nixos.dev | Well darn and blast . Looks like a job for dtrace then , shame they never got the licensing sorted out to use it in Linux, itโs basically the dogs bollocks. | 22:52:20 |
@nixinator:nixos.dev | Systemtap works but is a poor alternative | 22:52:51 |
@nixinator:nixos.dev | https://www.brendangregg.com/dtrace.html#OneLiners | 22:55:38 |
@nixinator:nixos.dev | This has help me nail all sorts of deep rooted bugs . :-) | 22:55:54 |
@nixinator:nixos.dev | (In Solaris) | 22:57:50 |
| 28 Oct 2022 |
magic_rb |
is dtrace a superior version of strace? (strace is really hard to use tbh) | 10:28:24 |
K900 | dtrace is more like BPF tracing in the kernel | 10:29:42 |
K900 | It's much more fine-grained than strace | 10:30:04 |
K900 | https://github.com/iovisor/bpftrace | 10:30:53 |
toonn | It's also a lot harder to get into IIRC. | 11:23:42 |
toonn | But use what works for you. | 11:23:55 |