bug#18070: Hydra does not cope well with offline build machines

bug#18070: Hydra does not cope well with offline build machines

[Andreas Enge]

For instance here:
   http://hydra.gnu.org/build/79392

There are lines
lsh: Connect failed, (errno = 0)
guix offload: error: failed to register GC root for '/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' on '#<<build-machine> name: "enge.fr" port: 2222 system: "mips64el-linux" user: "hydra" private-key: "/home/hydra/.lsh/identity" parallel-builds: 2 speed: 0.5 features: ()>' (status: 256)
builder for `/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' failed with exit code 1
@ hook-failed /gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv - 256 builder for `/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' failed with exit code 1
cannot build derivation `/gnu/store/i7xzs82ja6pwa641hdagq32z3m0g6c9y-gd-2.0.33.drv': 1 dependencies couldn't be built

It is normal that there is no connection to enge.fr:2222 because the machine
is offline. But hydra should notice the lack of response and build on another
machine instead.

Andreas

bug#18070: Hydra does not cope well with offline build machines

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> lsh: Connect failed, (errno = 0)
> guix offload: error: failed to register GC root for '/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' on '#<<build-machine> name: "enge.fr" port: 2222 system: "mips64el-linux" user: "hydra" private-key: "/home/hydra/.lsh/identity" parallel-builds: 2 speed: 0.5 features: ()>' (status: 256)
> builder for `/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' failed with exit code 1
> @ hook-failed /gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv - 256 builder for `/gnu/store/8nklnmzj317ahk6ynz7k5748lhma8cgn-GD_2_0_33.tar.xz.drv' failed with exit code 1
> cannot build derivation `/gnu/store/i7xzs82ja6pwa641hdagq32z3m0g6c9y-gd-2.0.33.drv': 1 dependencies couldn't be built
>
> It is normal that there is no connection to enge.fr:2222 because the machine
> is offline. But hydra should notice the lack of response and build on another
> machine instead.

I believe this is now fixed.  The change just makes ‘machine-load’
return +∞ when the connection fails; the effect is to ignore any such
machine.

I’ll deploy it on hydra.gnu.org ASAP.

Thanks,
Ludo’.