bug#44520: Graphical installer issues on 1.2.0.

bug#44520: Graphical installer issues on 1.2.0.

[Mathieu Othacehe]

[-- Attachment #1: Type: text/plain, Size: 529 bytes --]


Hello,

I tried the 1.2.0 graphical installer at ae0fe28, installing Guix System
from a usb drive to another one.

I had three issues:

* The auto partitioning crashed, backtrace as attachment.

* Using manual partitioning as a work-around, I started the installation
and observed that the installer downloaded "python" sources and then
substitutes for the whole bootstrapping chain (also as attachment).

* The substitutes download speed was sometimes super low (~ 50Kb/s).

An unpleasant sunday afternoon testing :(.

Mathieu

[-- Attachment #2: backtrace.jpg --]
[-- Type: image/jpeg, Size: 231630 bytes --]

[-- Attachment #3: config.jpg --]
[-- Type: image/jpeg, Size: 238542 bytes --]

[-- Attachment #4: bootstrap.jpg --]
[-- Type: image/jpeg, Size: 317692 bytes --]

bug#44520: Graphical installer issues on 1.2.0.

[Ludovic Courtès]

Hi,

Mathieu Othacehe <othacehe@gnu.org> skribis:

> I tried the 1.2.0 graphical installer at ae0fe28, installing Guix System
> from a usb drive to another one.
>
> I had three issues:
>
> * The auto partitioning crashed, backtrace as attachment.

Argh.

> * Using manual partitioning as a work-around, I started the installation
> and observed that the installer downloaded "python" sources and then
> substitutes for the whole bootstrapping chain (also as attachment).

Could it be that these get downloaded due to some “debug” output being
pulled (perhaps unnecessarily so but grafting can do that sometimes)?

> * The substitutes download speed was sometimes super low (~ 50Kb/s).

Yes, unfortunately non-offloaded disk image builds (fixed by recent
commits) along with GC running all day long has a terrible impact on
bandwidth.

It should get a bit better at least now that disk image builds will no
longer happen on the head node, but OTOH there’s still a lot of build
activity happening and that definitely increases pressure.

I look forward to the day where using the Build Coordinator we can
better decouple the build farm from the node running ‘guix publish’.

Thanks,
Ludo’.

bug#44520: Graphical installer issues on 1.2.0.

[Mathieu Othacehe]

Hey,

I cannot reproduce the partitioning issue which is even more
frustrating. I guess it was related to the original partitioning of the
USB drive I used. Added some more debug logging with 8c287bb2fb1 in the
meantime.

> Could it be that these get downloaded due to some “debug” output being
> pulled (perhaps unnecessarily so but grafting can do that sometimes)?

I'm not sure yet but I'll run more tests tomorrow.

>> * The substitutes download speed was sometimes super low (~ 50Kb/s).
>
> Yes, unfortunately non-offloaded disk image builds (fixed by recent
> commits) along with GC running all day long has a terrible impact on
> bandwidth.

Oh, right. Thanks for your recent patches on this subject. It should
definitely help. Having berlin doing nothing much that serving
substitutes and coordinating the build machines will solve many issues.

I'm working on that topic and will present my progress during Guix Days!

Thanks,

Mathieu

bug#44520: Graphical installer issues on 1.2.0.

[Mathieu Othacehe]

Hello,

I tried again on commit 86e9e5c using an image built by the CI and an
image built locally. I was not able to reproduce any of the issues I had
initially.

Closing this one until we find a reproducible way to trigger those
issues.

Thanks,

Mathieu