Cuirass job names and package variants

Cuirass job names and package variants

[Ludovic Courtès]

Hello!

Cuirass support in (gnu ci) computes job names as a function of package
names (the ‘job-name’ procedure).  I wanted to build several package
variants using ‘--with-input’, which I expressed in a manifest:

  https://gitlab.inria.fr/guix-hpc/guix-hpc/-/blob/master/manifest#L33-50

However, variants of a given package have the same package name/version,
and thus the same job name.  Apparently Cuirass only takes the first job
with a given name into account and dismisses subsequent ones.

What would be a good way to address this, either in Cuirass or in the
user config?

I thought about changing the name that appears in manifest entries
(without changing the actual package name), but (@ (gnu ci)
manifests->packages) ignores manifest entry names.

Ideas?

Thanks,
Ludo’.

Re: Cuirass job names and package variants

[Leo Prikler]

Am Freitag, den 25.06.2021, 13:44 +0200 schrieb Ludovic Courtès:
> Hello!
> 
> Cuirass support in (gnu ci) computes job names as a function of
> package
> names (the ‘job-name’ procedure).  I wanted to build several package
> variants using ‘--with-input’, which I expressed in a manifest:
> 
>   
> https://gitlab.inria.fr/guix-hpc/guix-hpc/-/blob/master/manifest#L33-50
> 
> However, variants of a given package have the same package
> name/version,
> and thus the same job name.  Apparently Cuirass only takes the first
> job
> with a given name into account and dismisses subsequent ones.
> 
> What would be a good way to address this, either in Cuirass or in the
> user config?
Do we already know – or could we compute – store paths at this point in
time?  It seems as though this could be solved once more by assigning
ids based on hashed inputs.  So each job-name would be something like
(package-name)-(package-version)-(first-bytes-of-package-input-hash).
WDYT?

Re: Cuirass job names and package variants

[Ludovic Courtès]

Leo Prikler <leo.prikler@student.tugraz.at> skribis:

> Am Freitag, den 25.06.2021, 13:44 +0200 schrieb Ludovic Courtès:

[...]

>> However, variants of a given package have the same package
>> name/version,
>> and thus the same job name.  Apparently Cuirass only takes the first
>> job
>> with a given name into account and dismisses subsequent ones.
>> 
>> What would be a good way to address this, either in Cuirass or in the
>> user config?
> Do we already know – or could we compute – store paths at this point in
> time?  It seems as though this could be solved once more by assigning
> ids based on hashed inputs.  So each job-name would be something like
> (package-name)-(package-version)-(first-bytes-of-package-input-hash).
> WDYT?

That job names are human-readable is a feature.  :-)

However, what we could do is have an autoincrement ID or similar (?)
rather than the package name used as the actual key.  Dunno.

Thanks,
Ludo’.

Re: Cuirass job names and package variants

[Leo Prikler]

Am Freitag, den 25.06.2021, 16:30 +0200 schrieb Ludovic Courtès:
> Leo Prikler <leo.prikler@student.tugraz.at> skribis:
> 
> > Am Freitag, den 25.06.2021, 13:44 +0200 schrieb Ludovic Courtès:
> 
> [...]
> 
> > > However, variants of a given package have the same package
> > > name/version,
> > > and thus the same job name.  Apparently Cuirass only takes the
> > > first
> > > job
> > > with a given name into account and dismisses subsequent ones.
> > > 
> > > What would be a good way to address this, either in Cuirass or in
> > > the
> > > user config?
> > Do we already know – or could we compute – store paths at this
> > point in
> > time?  It seems as though this could be solved once more by
> > assigning
> > ids based on hashed inputs.  So each job-name would be something
> > like
> > (package-name)-(package-version)-(first-bytes-of-package-input-
> > hash).
> > WDYT?
> 
> That job names are human-readable is a feature.  :-)
> 
> However, what we could do is have an autoincrement ID or similar (?)
> rather than the package name used as the actual key.  Dunno.
In that case there is nothing to relate jobs to packages, which I'd
assume is also a feature?  We could have job names be unique id +
package name, but since those ids grow larger and larger, i think
they'd soon become unreadable as well.  Maybe add -1, -2, -3,... to the
job names if a job with the same name is already queued like backup
systems do?