Re: Staging branch [substitute availability armhf-linux]

[Mathieu Othacehe]

Hey Ludo,

> You seem to imply that the issue is the number of architectures, rather
> than the small number of ARMv7 build machines (now that we disabled
> 32-bit builds on AArch64).  Do I get it right?

Yes my point is that building three specifications on three
architectures, including an emulated one, is already hard for the build
farm, so adding more specifications/architectures seems complex.

Even if we fix the problem raised by Danny, enabling again ARMv7
transparent emulation, without any additional hardware wouldn't fit.

> That was a problem with Cuirass doing ‘build-derivations’ RPCs for
> derivations spanning multiple architectures (the RPC would complete once
> the slowest architecture is done), but maybe that’s no longer the case
> with the new remote builds feature you’ve been working on?

Yes, that's solved by the remote building feature. The workers are
declaring the architectures they support. When they request work, the
remote server picks randomly an architecture and select the most
priority build available. This way the queuing happens at the database
level and not in the guix-daemon itself.

Thanks,

Mathieu