On Wed, Jun 30, 2021 at 12:08:44PM +0200, Miroslav Suchý wrote:
Dne 29. 06. 21 v 22:25 Ben Cotton napsal(a):
>When this proposal is implemented, they can instead declaratively
>specify the amount of RAM needed per build thread, e.g.
>
> %limit_build -m 8192
>
>for declaring a build thread should be allocated 8GB of RAM.
Do I understand it correctly that semantics will be: One process
will consume 8GB ram. And because we have only 16 GB RAM on this
host we will limit smp flag to 2 despite having 4 cores. Right?
FYI Copr utilize heavily tmpfs plugin
http://miroslav.suchy.cz/blog/archives/2015/05/28/increase_mock_performan...
we have ~100 GB swap which we use for tmpfs.
This can provide misleading numbers for SwapTotal.
The OP talks exclusively about "RAM". And I think RAM is what we need
to look at: even if there is bountiful swap, the build would be quite
slow if it was used to any significant degree. So we're probably
better of limiting parallelism rather than using all that swap.
(Note that this proposal is about limiting parallelism, i.e. tweaking
an optimization. If instead the macro was about "what are the minimum
memory reqs below which we will reject a build", the rules would be
completely different and we should include swap in the calculation,
because it's better to build slowly than not at all.)
Zbyszek