Missing substitutes and TIMEOUT property (Racket, MAME)

Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 432 bytes --]

Hi!

Some packages are frequently without substitutes, like Racket and MAME.
Link between the two?  They both take a long time to build.  So I wonder
if this is not because the build somehow timeouts.

I see that there is a `timeout' property on packages like chromium,
qtwebkit, etc.

Would it make sense to add, say, a 6-hour timeout to the offending packages?

Cheers!

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 67 bytes --]

Friendly ping! :)

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Leo Famulari]

On Tue, Feb 18, 2020 at 03:58:19PM +0100, Pierre Neidhardt wrote:
> Some packages are frequently without substitutes, like Racket and MAME.
> Link between the two?  They both take a long time to build.  So I wonder
> if this is not because the build somehow timeouts.

The logs of the failing builds should say if the failure was caused by
timeout or exceeding the maximum "silent time" (time without any output
from the build).

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 644 bytes --]

Leo Famulari <leo@famulari.name> writes:

> The logs of the failing builds should say if the failure was caused by
> timeout or exceeding the maximum "silent time" (time without any output
> from the build).

If I look at failed builds for Racket or MAME, I get this:

http://ci.guix.gnu.org/build/2290964/details
http://ci.guix.gnu.org/build/2293173/details

It says

--8<---------------cut here---------------start------------->8---
Status	Failed (dependency)
--8<---------------cut here---------------end--------------->8---

and does not have any logs.
Something is fishy.

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Tobias Geerinckx-Rice]

[-- Attachment #1: Type: text/plain, Size: 1094 bytes --]

Pierre,

Pierre Neidhardt 写道：
> Leo Famulari <leo@famulari.name> writes:
>
>> The logs of the failing builds should say if the failure was 
>> caused by
>> timeout or exceeding the maximum "silent time" (time without 
>> any output
>> from the build).
>
> If I look at failed builds for Racket or MAME, I get this:
>
> http://ci.guix.gnu.org/build/2290964/details
> http://ci.guix.gnu.org/build/2293173/details
>
> It says
>
> --8<---------------cut 
> here---------------start------------->8---
> Status	Failed (dependency)
> --8<---------------cut 
> here---------------end--------------->8---

Cuirass should link to the failing dependencies for further 
inspection, but doesn't (yet?).  You have to guess.  It's fun.

> and does not have any logs.

Since these builds were never attempted, they never produced any 
logs.  I think that's the correct approach.

This is also why blindly raising time-outs for these packages 
isn't the solution: there's nothing to time out.

> Something is fishy.

Not fishy, just -ETODO.

Kind regards,

T G-R

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 417 bytes --]

Tobias Geerinckx-Rice <me@tobias.gr> writes:

> Cuirass should link to the failing dependencies for further 
> inspection, but doesn't (yet?).  You have to guess.  It's fun.

Hmmm, not easy considering it builds on my machine :p

> Since these builds were never attempted, they never produced any 
> logs.  I think that's the correct approach.

Makes sense.

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Tobias Geerinckx-Rice]

[-- Attachment #1: Type: text/plain, Size: 436 bytes --]

Tobias Geerinckx-Rice 写道：
> Since these builds were never attempted, they never produced any 
> logs.

Not the same architecture, but here's a staging MAME that was 
built and killed without explanation after a non-magical-looking 
number of seconds:

http://ci.guix.gnu.org/build/2337230/details

I don't understand.  There has to be a better way to query Cuirass 
than endlessly clicking ‘Next’…  :-/

T G-R

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 751 bytes --]

Hello Tobias,

Tobias Geerinckx-Rice <me@tobias.gr> ezt írta (időpont: 2020. márc. 3., Ke
15:35):

> Tobias Geerinckx-Rice 写道：
> > Since these builds were never attempted, they never produced any
> > logs.
>
> Not the same architecture, but here's a staging MAME that was
> built and killed without explanation after a non-magical-looking
> number of seconds:
>
> http://ci.guix.gnu.org/build/2337230/details
>
> I don't understand.  There has to be a better way to query Cuirass
> than endlessly clicking ‘Next’…  :-
>
You can try to get failing dependency information from the guix data
service. Last time I checked that did work. It should list all the failed
dependencies that failed directly.

>
> T G-R
>

[-- Attachment #2: Type: text/html, Size: 1348 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 299 bytes --]

Do you have an example link?

For me:

http://data.guix.gnu.org/repository/1/branch/master/package/mame

and

http://data.guix.gnu.org/revision/1d9b4664e30ce3f79606bf1167a2fca9bf4abba2/package/mame/0.218

only shows "unknown" build status.

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Ludovic Courtès]

Hi!

Pierre Neidhardt <mail@ambrevar.xyz> skribis:

> Some packages are frequently without substitutes, like Racket and MAME.
> Link between the two?  They both take a long time to build.  So I wonder
> if this is not because the build somehow timeouts.
>
> I see that there is a `timeout' property on packages like chromium,
> qtwebkit, etc.

Few people know that this property has absolutely no effect.  :-)

It used to be that Hydra would honor it.  But Cuirass has never honored it.

It’s not clear to me whether Cuirass should honor it, in fact.

At the Guix level, timeouts are per session: you set them for all your
session with ‘set-build-options’.  So if you want them per-derivation,
you basically need to create one session per derivation build; that’s
what Hydra did.

But there’s a good reason for it to be per-session: the admin may want
to be in control of those parameters.  For example, on berlin, we’d
rather make sure machines don’t stay idle for too long.

Thoughts?

Ludo’.

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Pierre Neidhardt]

[-- Attachment #1: Type: text/plain, Size: 679 bytes --]

Ludovic Courtès <ludo@gnu.org> writes:

> But there’s a good reason for it to be per-session: the admin may want
> to be in control of those parameters.  For example, on berlin, we’d
> rather make sure machines don’t stay idle for too long.

This makes sense, but for now it seems that it's hard to know whether
the build failed because of a timeout or something else.
I suppose some improvements should be done on the Cuirass / Data service
fronts.

If we confirm that timeouts really cause the issue, then I suppose the Berlin
admin can increase the timeout so that the offending packages build.

Thoughts?

-- 
Pierre Neidhardt
https://ambrevar.xyz/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Marius Bakke]

[-- Attachment #1: Type: text/plain, Size: 1197 bytes --]

Ludovic Courtès <ludo@gnu.org> writes:

> Hi!
>
> Pierre Neidhardt <mail@ambrevar.xyz> skribis:
>
>> Some packages are frequently without substitutes, like Racket and MAME.
>> Link between the two?  They both take a long time to build.  So I wonder
>> if this is not because the build somehow timeouts.
>>
>> I see that there is a `timeout' property on packages like chromium,
>> qtwebkit, etc.
>
> Few people know that this property has absolutely no effect.  :-)
>
> It used to be that Hydra would honor it.  But Cuirass has never honored it.
>
> It’s not clear to me whether Cuirass should honor it, in fact.
>
> At the Guix level, timeouts are per session: you set them for all your
> session with ‘set-build-options’.  So if you want them per-derivation,
> you basically need to create one session per derivation build; that’s
> what Hydra did.
>
> But there’s a good reason for it to be per-session: the admin may want
> to be in control of those parameters.  For example, on berlin, we’d
> rather make sure machines don’t stay idle for too long.

Maybe we could create a separate jobset for the jobs that require a
longer max-silent-time, IIUC?

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: Missing substitutes and TIMEOUT property (Racket, MAME)

[Ludovic Courtès]

Hi,

Marius Bakke <mbakke@fastmail.com> skribis:

> Ludovic Courtès <ludo@gnu.org> writes:

[...]

>> At the Guix level, timeouts are per session: you set them for all your
>> session with ‘set-build-options’.  So if you want them per-derivation,
>> you basically need to create one session per derivation build; that’s
>> what Hydra did.
>>
>> But there’s a good reason for it to be per-session: the admin may want
>> to be in control of those parameters.  For example, on berlin, we’d
>> rather make sure machines don’t stay idle for too long.
>
> Maybe we could create a separate jobset for the jobs that require a
> longer max-silent-time, IIUC?

Not really, because Cuirass does not use per-jobset sessions.

So really, we’d have to change the timeouts globally for guix-daemon, or
for all of Cuirass.

Perhaps we should start by creating a list of affected package/system
pairs, and from there we could make a more informed decisions.

Thoughts?

Ludo’.