#29634 'max-silent-time' etc. should be per-derivation settings

Message #5 received at submit@debbugs.gnu.org (full text, mbox, reply):

From: Mark H Weaver <mhw@netris.org>

To: bug-guix@gnu.org

Subject: Hydra fails to honor 'max-silent-time' for early guile build

Date: Sat, 09 Dec 2017 21:02:42 -0500

An early 'guile' build on armhf-linux consistently fails on Hydra,
because the 'timeout' and 'max-silent-time' properties are not honored.
For example:

  https://hydra.gnu.org/build/2391845

This is a longstanding issue.  In past core-updates cycles, I have
worked around the problem by rebuilding these early 'guile' derivations
from Hydra's command line, passing --timeout and --max-silent-time
manually, and then restarting the many dependency failures.

It would be good to fix this issue properly.

My guess is that the build parameters are not honored because there is
no Hydra 'job' associated with this early 'guile'.

       Mark

Message #8 received at 29634@debbugs.gnu.org (full text, mbox, reply):

From: ludo@gnu.org (Ludovic Courtès)

To: Mark H Weaver <mhw@netris.org>

Cc: 29634@debbugs.gnu.org

Subject: Re: bug#29634: Hydra fails to honor 'max-silent-time' for early guile build

Date: Mon, 11 Dec 2017 10:40:08 +0100

Mark H Weaver <mhw@netris.org> skribis:

> An early 'guile' build on armhf-linux consistently fails on Hydra,
> because the 'timeout' and 'max-silent-time' properties are not honored.
> For example:
>
>   https://hydra.gnu.org/build/2391845
>
> This is a longstanding issue.  In past core-updates cycles, I have
> worked around the problem by rebuilding these early 'guile' derivations
> from Hydra's command line, passing --timeout and --max-silent-time
> manually, and then restarting the many dependency failures.
>
> It would be good to fix this issue properly.
>
> My guess is that the build parameters are not honored because there is
> no Hydra 'job' associated with this early 'guile'.

Exactly.  Or maybe there’s one but this early ‘guile’ is built as a
dependency of another job.

Ludo’.

Message #11 received at 29634@debbugs.gnu.org (full text, mbox, reply):

From: Mark H Weaver <mhw@netris.org>

To: ludo@gnu.org (Ludovic Courtès)

Cc: 29634@debbugs.gnu.org

Subject: Re: bug#29634: Hydra fails to honor 'max-silent-time' for early guile build

Date: Mon, 11 Dec 2017 21:57:56 -0500

ludo@gnu.org (Ludovic Courtès) writes:

> Mark H Weaver <mhw@netris.org> skribis:
>
>> An early 'guile' build on armhf-linux consistently fails on Hydra,
>> because the 'timeout' and 'max-silent-time' properties are not honored.
>> For example:
>>
>>   https://hydra.gnu.org/build/2391845
>>
>> This is a longstanding issue.  In past core-updates cycles, I have
>> worked around the problem by rebuilding these early 'guile' derivations
>> from Hydra's command line, passing --timeout and --max-silent-time
>> manually, and then restarting the many dependency failures.
>>
>> It would be good to fix this issue properly.
>>
>> My guess is that the build parameters are not honored because there is
>> no Hydra 'job' associated with this early 'guile'.
>
> Exactly.  Or maybe there’s one but this early ‘guile’ is built as a
> dependency of another job.

How do you think we should fix this?

      Mark

Message #14 received at 29634@debbugs.gnu.org (full text, mbox, reply):

From: ludo@gnu.org (Ludovic Courtès)

To: Mark H Weaver <mhw@netris.org>

Cc: 29634@debbugs.gnu.org

Subject: Re: bug#29634: Hydra fails to honor 'max-silent-time' for early guile build

Date: Tue, 12 Dec 2017 10:21:08 +0100

Mark H Weaver <mhw@netris.org> skribis:

> ludo@gnu.org (Ludovic Courtès) writes:
>
>> Mark H Weaver <mhw@netris.org> skribis:
>>
>>> An early 'guile' build on armhf-linux consistently fails on Hydra,
>>> because the 'timeout' and 'max-silent-time' properties are not honored.
>>> For example:
>>>
>>>   https://hydra.gnu.org/build/2391845
>>>
>>> This is a longstanding issue.  In past core-updates cycles, I have
>>> worked around the problem by rebuilding these early 'guile' derivations
>>> from Hydra's command line, passing --timeout and --max-silent-time
>>> manually, and then restarting the many dependency failures.
>>>
>>> It would be good to fix this issue properly.
>>>
>>> My guess is that the build parameters are not honored because there is
>>> no Hydra 'job' associated with this early 'guile'.
>>
>> Exactly.  Or maybe there’s one but this early ‘guile’ is built as a
>> dependency of another job.
>
> How do you think we should fix this?

Of course I’d prefer if Guile compile times were shorter.

Besides, we should probably handle timeouts per-derivation rather than
globally (currently timeout are per-session, see ‘set-build-options’).
We could do that by adding a new special value in the environment
variable map of derivation (as is the case for, e.g.,
“exportReferencesGraph”) that the daemon would honor.  It’s a
rebuild-the-world change though.

Ludo’.

Message #17 received at 29634@debbugs.gnu.org (full text, mbox, reply):

From: Mark H Weaver <mhw@netris.org>

To: ludo@gnu.org (Ludovic Courtès)

Cc: 29634@debbugs.gnu.org

Subject: Re: bug#29634: Hydra fails to honor 'max-silent-time' for early guile build

Date: Tue, 12 Dec 2017 17:12:56 -0500

ludo@gnu.org (Ludovic Courtès) writes:

> Mark H Weaver <mhw@netris.org> skribis:
>
>> ludo@gnu.org (Ludovic Courtès) writes:
>>
>>> Mark H Weaver <mhw@netris.org> skribis:
>>>
>>>> An early 'guile' build on armhf-linux consistently fails on Hydra,
>>>> because the 'timeout' and 'max-silent-time' properties are not honored.
>>>> For example:
>>>>
>>>>   https://hydra.gnu.org/build/2391845
>>>>
>>>> This is a longstanding issue.  In past core-updates cycles, I have
>>>> worked around the problem by rebuilding these early 'guile' derivations
>>>> from Hydra's command line, passing --timeout and --max-silent-time
>>>> manually, and then restarting the many dependency failures.
>>>>
>>>> It would be good to fix this issue properly.
>>>>
>>>> My guess is that the build parameters are not honored because there is
>>>> no Hydra 'job' associated with this early 'guile'.
>>>
>>> Exactly.  Or maybe there’s one but this early ‘guile’ is built as a
>>> dependency of another job.
>>
>> How do you think we should fix this?
>
> Of course I’d prefer if Guile compile times were shorter.
>
> Besides, we should probably handle timeouts per-derivation rather than
> globally (currently timeout are per-session, see ‘set-build-options’).
> We could do that by adding a new special value in the environment
> variable map of derivation (as is the case for, e.g.,
> “exportReferencesGraph”) that the daemon would honor.  It’s a
> rebuild-the-world change though.

Sounds good to me.  For now, I'll rebuild the early guile for armhf
manually.

     Thanks,
       Mark

Send a report that this bug log contains spam.