IRC channel logs

2020-10-13.log

back to list of logs

<zimoun>rekado_: cool that you are working on guile-drmaa. I am eager the results. :-)
<rekado_>guile-drmaa only talks to libdrmaa.so
<rekado_>this libdrmaa.so is provided by slurm
<rekado_>some operations (like run-job) cause libdrmaa.so to interface with a running slurmd instance
<rekado_>and that’s when auth_munge.so (or even auth_none.so) complain about missing symbols
<rekado_>it would be nonsensical for DRMAA applications to also link to libslurm
<rekado_>because the whole point of DRMAA is to provide a scheduler-agnostic abstraction.
<rekado_>all operations that don’t require the scheduler (like creating and editing a job template) work fine.
<zimoun>rekado_, civodul: and does this guile DRMAA scheduler-agnostic asbtraction could be reused to improve the build farm instead of rely on guix-daemon for scheduling, if I read correctly?
<zimoun>hey, I have just seen this: https://con.racket-lang.org/#slot5 by Konrad :-)
<rekado_>zimoun: I always felt that HPC schedulers provide so much configurable infrastructure, which isn’t really the difficult part of scheduling.
<rekado_>the difficult part is policy
<rekado_>and that’s left to the admins
<rekado_>the problem space is very similar to workflow languages, I feel
<rekado_>people obsess about this or that workflow language, but in the end they all do the same and it’s really deployment and debuggability that matters – most of which is left to the user to figure out on their own
<rekado_>so: no, I don’t think DRMAA is going to be useful for the build farm
<civodul>heheh
<civodul>zimoun: Konrad is everywhere!
<civodul>speaking of which, JupyterCon is taking place right now and i thought someone should submit a lightning talk
<civodul>but i lack motivation
*zimoun needs to think about what rekado_ said :-)
<rekado_>better don’t! It might not actually make sense.
<rekado_>haven’t slept much at all the past few days. Baby’s a very poor sleeper these weeks.
<zimoun>civodul: Konrad’s show is on Sat. Cool!
<zimoun>yeah it could be cool for JupyterCon. But I have not using yet guix-jupyter.
<zimoun>rekado_: oh! less time to hack overnight, so.
<zimoun>BTW, WDYT to set by default pararelle-build? to #f? It will fix the reproducibility issue with Haskell
<rekado_>I get maybe one hour each day to do stuff on the computer, but usually I go straight to smell the yaks.
<rekado_>fine by me
<rekado_>I think we should disable it; doing it in the build system seems reasonable
<civodul>rekado_: :-)
<zimoun>patch core-updates?
<civodul>zimoun: you're talking about sequential builds for Haskell stuff only, right?
<zimoun>yes, for Haskell only.
<rekado_>yes, my patch to fix parallel builds actually resulted in lots of Haskell packages to no longer build reproducibly
<civodul>ok
<zimoun>and OCaml because of dune. roptat stopped out the issue
<zimoun>rekado_: the shortest diff to core-updates ;-)