Cuirass package build timeouts and max-silent-time [was Re: Merging branch wip-haskell]

[Leo Famulari]

On Wed, Mar 08, 2023 at 08:31:33PM +0100, Lars-Dominik Braun wrote:
> Hello Felix,
> 
> > I see build failures [1] for Ghc 9.2.5 in the Cuirass job set [2] for
> > my own feature branch. [3] They are caused by a one-hour timeout. Did
> > you folks figure out how to extend the timeout limit on Cuirass when
> > working on your own branch? Thanks!
> 
> I don’t see GHC being rebuild in [2] and GHC 9.2 should
> have a higher max-silent-time than 1 hour on master since
> 3bb2078a12d78a13f1e1520fe3705333a74ef6e3 (which is part of your
> branch). So I’m slightly confused.

I suspect that GHC was never built by Cuirass, but that somebody logged
in to the build server and invoked `guix build ghc
--max-silent-time=9`.

There's no logs of a successful build of GHC 9.2.5 available through the
Cuirass web interface, which is a clue that my hunch is correct. Builds
performed "by hand" don't appear in the web interface.

https://ci.guix.gnu.org/search?query=spec%3Amaster+ghc-9.2.5

Are we sure that Cuirass honors the max-silent-time property? If I
remember correctly, it did not do so in the past.

Re: Merging branch wip-haskell

[Development of GNU Guix and the GNU System distribution.]

Hi Lars,

On Wed, Mar 8, 2023 at 11:31 AM Lars-Dominik Braun  wrote:
>
> I don’t see GHC being rebuild

How about here [1] please? Maybe I'm misreading the log:

building path(s)
`/gnu/store/xdl4lyjjai90w21whf7xwvdqh8cf7hy7-cpphs-1.20.9.1-1.cabal'
building path(s)
`/gnu/store/199mrp0p6l9pldpzrpgmkmmfxikga1jw-ghc-9.2.5-src.tar.xz'

and ultimately pandoc failing

cannot build derivation
`/gnu/store/qahn3czw4msmv7lnykawfsngdjq867sn-pandoc-2.19.2.drv': 70
dependencies couldn't be built

as well as the apparently undeclared consumer python-bokeh

cannot build derivation
`/gnu/store/38hpmicc3v4f3a8ihbmas9ii0fz5pysd-python-bokeh-2.4.3.drv':
1 dependencies couldn't be built

with this remark in the middle

   building of `/gnu/store/in0gfsahn1z6ym232wg3cayzim51zng8-ghc-9.2.5.drv'
timed out after 3600 seconds of silence
@ build-failed /gnu/store/in0gfsahn1z6ym232wg3cayzim51zng8-ghc-9.2.5.drv
- timeout

I am totally new to Cuirass and open to an alternative interpretation
of these messages. My apologies for doubling up.

Kind regards,
Felix Lechner

[1] https://ci.guix.gnu.org/build/516127/log/raw

Re: Merging branch wip-haskell

[Lars-Dominik Braun]

Hello Felix,

> I see build failures [1] for Ghc 9.2.5 in the Cuirass job set [2] for
> my own feature branch. [3] They are caused by a one-hour timeout. Did
> you folks figure out how to extend the timeout limit on Cuirass when
> working on your own branch? Thanks!

I don’t see GHC being rebuild in [2] and GHC 9.2 should
have a higher max-silent-time than 1 hour on master since
3bb2078a12d78a13f1e1520fe3705333a74ef6e3 (which is part of your
branch). So I’m slightly confused.

Cheers,
Lars

> [1] https://ci.guix.gnu.org/build/478772/log/raw
> [2] https://ci.guix.gnu.org/eval/262309
> [3] https://git.savannah.gnu.org/cgit/guix.git/log/?h=wip-go-updates

Re: Merging branch wip-haskell

[Development of GNU Guix and the GNU System distribution.]

Hi,

On Mon, Feb 27, 2023 at 3:05 AM Lars-Dominik Braun  wrote:
>
> the branch has been merged into master.

I see build failures [1] for Ghc 9.2.5 in the Cuirass job set [2] for
my own feature branch. [3] They are caused by a one-hour timeout. Did
you folks figure out how to extend the timeout limit on Cuirass when
working on your own branch? Thanks!

Kind regards
Felix Lechner

[1] https://ci.guix.gnu.org/build/478772/log/raw
[2] https://ci.guix.gnu.org/eval/262309
[3] https://git.savannah.gnu.org/cgit/guix.git/log/?h=wip-go-updates

Re: Merging branch wip-haskell

[Andreas Enge]

Am Wed, Feb 22, 2023 at 12:54:03PM +0200 schrieb Efraim Flashner:
> > By the way, when building pandoc for my core-updates profile, I noticed
> > that ghc bootstrapping takes a terribly long time. Might there be a fleeting
> > chance to jump over some older releases?
> I believe in core-updates I've changed it so we go from 8.6 to 8.10. I
> think if we packaged 8.2 we could go from 7.10 -> 8.2 -> 8.6 -> 8.10,
> using 8.2 to replace 8.0 and 8.4.

Good to hear that! Right now I indeed see the following:
7.10.3, 8.0.2, 8.4.4, 8.6.5, 8.10.7, 9.2.5.
There is also a 9.0 (which could be dropped maybe?), but 9.2 is built
from 8.10. And 8.10 is built with 8.6, but inherits from 8.8; to make
things simpler, it would be nice to drop 8.8, but this would imply some
work to get the right differences between the recipe for 8.6 and for 8.10.

Andreas

Re: Merging branch wip-haskell

[Development of GNU Guix and the GNU System distribution.]

Hi Lars,

> the branch has been merged into master.

Thank you for that giant contribution to making Haskell more usable in Guix!

Kind regards
Felix Lechner

Re: Merging branch wip-haskell

[Lars-Dominik Braun]

Hi,

> > the branch has been merged into master.
> \o/ Nice, thank you!
well, GHC fails a single testcase on i686 (which we did not test for
wip-haskell) right now[1], but it’ll take some rounds of building it
locally until I have this sorted out too.

Lars

[1] https://ci.guix.gnu.org/eval/230900?status=failed=0

Re: Merging branch wip-haskell

[Simon Tournier]

Hi,

On Sun, 26 Feb 2023 at 10:47, Lars-Dominik Braun  wrote:

> the branch has been merged into master.

\o/ Nice, thank you!

Cheers,
simo

Re: Merging branch wip-haskell

[Lars-Dominik Braun]

Hi,

> just a heads-up: The long overdue Haskell update is finally rolling in,
> bumping our packages to Stackage release 20.5 and the compiler to 9.2. The
> corresponding issue is #61420.
the branch has been merged into master.

Lars

Re: Merging branch wip-haskell

[Efraim Flashner]

On Wed, Feb 22, 2023 at 10:37:41AM +0100, Andreas Enge wrote:
> Hello Lars,
> 
> Am Wed, Feb 15, 2023 at 11:38:23AM +0100 schrieb zimoun:
> > Feel free to merge. :-)
> 
> +1, please go ahead!
> 
> By the way, when building pandoc for my core-updates profile, I noticed
> that ghc bootstrapping takes a terribly long time. Might there be a fleeting
> chance to jump over some older releases?

I believe in core-updates I've changed it so we go from 8.6 to 8.10. I
think if we packaged 8.2 we could go from 7.10 -> 8.2 -> 8.6 -> 8.10,
using 8.2 to replace 8.0 and 8.4.

-- 
Efraim Flashner  אפרים פלשנר
GPG key = A28B F40C 3E55 1372 662D  14F7 41AA E7DC CA3D 8351
Confidentiality cannot be guaranteed on emails sent or received unencrypted


signature.asc
Description: PGP signature

Re: Merging branch wip-haskell

[Andreas Enge]

Hello Lars,

Am Wed, Feb 15, 2023 at 11:38:23AM +0100 schrieb zimoun:
> Feel free to merge. :-)

+1, please go ahead!

By the way, when building pandoc for my core-updates profile, I noticed
that ghc bootstrapping takes a terribly long time. Might there be a fleeting
chance to jump over some older releases?

Andreas

Re: Merging branch wip-haskell

[zimoun]

Hi Lars,

On Wed, 15 Feb 2023 at 10:01, Lars-Dominik Braun  wrote:

> Is there anything preventing a merge into currently? Can we coordinate
> the merge with some other big world-rebuilding changes waiting to happen
> (apart from core-updates)?

As discussed in #61420 [1], from my small experiments most of the
substitutes are there and after checking about CI status, all seems
green except ganeti and python-pypandoc.

Not all susbtitutes are available after rebasing wip-haskell on the top
of current master (4501a50bbd), but somehow this situation appears to me
unavoidable, sadly.

Well, maybe we could try to build ganeti and python-pypandoc with an
older GHC than 9.2 since Guix provides several GHC version.  Somehow
apply an transformation similar to ’package-with-explicit-python’ for
GHC.  Or I propose to hide them waiting a proper fix.

Feel free to merge. :-)

Cheers,
simon

1: 

Re: Merging branch wip-haskell

[Andreas Enge]

Am Wed, Feb 15, 2023 at 10:45:47AM +0100 schrieb Lars-Dominik Braun:
> There’s a job on CI already: https://ci.guix.gnu.org/jobset/wip-haskell

Okay, then it should be fine to merge; since both build farms are enabled
by default, if there are substitutes at one of them, this is enough.

Thanks!

Andreas

Re: Merging branch wip-haskell

[Lars-Dominik Braun]

Hi Andreas,

> I do not think we should coordinate, this was part of the motivation for
> considering feature branches in the first place, to avoid entanglement
> with different updates.
ah, alright. I’ll give everyone more time for reviews and then just
merge it.

> However, QA has not run yet:
>https://qa.guix.gnu.org/issue/61420
> I assume it will start in a few days, so maybe you could wait a little bit
> until it has run and hopefully provided substitutes on bordeaux?
There’s a job on CI already: https://ci.guix.gnu.org/jobset/wip-haskell

> By the way, there are also branches wip-haskell-updates and
> wip-haskell-updates-2 from 2020; maybe they can be deleted?
Yeah, I guess we can remove them.

Cheers,
Lars

Re: Merging branch wip-haskell

[Andreas Enge]

Hello Lars,

Am Wed, Feb 15, 2023 at 10:01:16AM +0100 schrieb Lars-Dominik Braun:
> just a heads-up: The long overdue Haskell update is finally rolling in,
> bumping our packages to Stackage release 20.5 and the compiler to 9.2. The
> corresponding issue is #61420.

congratulations!

> Is there anything preventing a merge into currently? Can we coordinate
> the merge with some other big world-rebuilding changes waiting to happen
> (apart from core-updates)?

I do not think we should coordinate, this was part of the motivation for
considering feature branches in the first place, to avoid entanglement
with different updates.

However, QA has not run yet:
   https://qa.guix.gnu.org/issue/61420
I assume it will start in a few days, so maybe you could wait a little bit
until it has run and hopefully provided substitutes on bordeaux?

By the way, there are also branches wip-haskell-updates and
wip-haskell-updates-2 from 2020; maybe they can be deleted?

Andreas

Merging branch wip-haskell

[Lars-Dominik Braun]

Hi,

just a heads-up: The long overdue Haskell update is finally rolling in,
bumping our packages to Stackage release 20.5 and the compiler to 9.2. The
corresponding issue is #61420.

Is there anything preventing a merge into currently? Can we coordinate
the merge with some other big world-rebuilding changes waiting to happen
(apart from core-updates)?

Cheers,
Lars