guix-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: RFC: subcommand to pause/resume builds


From: John Soo
Subject: Re: RFC: subcommand to pause/resume builds
Date: Tue, 03 Nov 2020 06:41:02 -0800
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/27.1 (gnu/linux)

Hello!

I want to preface all this by saying this is not a huge priority to me.
pause/resume would just be a nice quality of life improvement.

Ludovic Courtès <ludo@gnu.org> writes:

> First, note that the daemon is unaware of “packages”, it only knows
> about “derivations”.

Agreed. I was thinking mostly about the best interface for a user. But
if derivations are too low-level I can see not being able to use the
package name.

> Second, ‘guix processes’ is nice but it uses low-level heuristics to
> determine what daemon sessions are open, what their clients are, and
> what they’re building; it resorts to heuristics because the daemon as it
> stands doesn’t have a way to communicate its current state.  It works
> well in practice, but still I wouldn’t go too far building around it.
>
> Last, you’d need to send SIGTSTP to the whole process group of the
> build, like so (I think, haven’t tried):
>
>   sudo kill -TSTP -123

Ah right, I got that advice on IRC also but I don't actually know how to
do it. Thanks!

> where 123 is the “SessionPID” shown by ‘guix processes’.  However, doing
> so may affect build results: processes in the build environment might
> handle SIGTSTP specially, which can have side effects.  It’s an
> observable action.

I thought of the side effects after sending the email. Makes sense to
me. Does that mean it is not worth including?  Given that the source and
state of a build do not change while paused, if a build tool changes its
outputs after paused/resumed is it still reproducible?

The other part that would make it difficult is that maybe different
build-systems respond differently to different signals. A simpler to
implement alternative might just be to provide a way to send an
arbitrary signal to the process tree. Given some builds might respond
un-reproducibly to it, a warning could be provided.

In any case, such a subcommand not a high priority to me as a user,
though sometimes I long to pause guix build ungoogled-chromium.

Thanks again,

John



reply via email to

[Prev in Thread] Current Thread [Next in Thread]