nar-herder uses 10 GiB of resident memory, 100% CPU on hydra-guix-129

  • Open
  • quality assurance status badge
Details
2 participants
  • Christopher Baines
  • Maxim Cournoyer
Owner
unassigned
Submitted by
Maxim Cournoyer
Severity
normal
M
M
Maxim Cournoyer wrote on 8 Nov 2023 04:51
(name . bug-guix)(address . bug-guix@gnu.org)(name . Christopher Baines)(address . mail@cbaines.net)
87jzqshph2.fsf@gmail.com
Hi,

I was looking at top on the hydra-guix-129 node, which runs nar-herder,
and saw this:

Toggle snippet (3 lines)
4772 nar-her+ 20 0 28.9g 11.5g 100.0 6.1 55,55 S .nar-herder-rea

11.5 GiB of memory seems a bit excessive, no? Its cumulated processing
time is also at the top of the table, and it seems stuck at 100% of CPU
usage.

Is this expected of the nar-herder, or has it gone awry?

--
Thanks,
Maxim
C
C
Christopher Baines wrote on 8 Nov 2023 16:36
(name . Maxim Cournoyer)(address . maxim.cournoyer@gmail.com)(name . bug-guix)(address . bug-guix@gnu.org)
87il6cgsqw.fsf@cbaines.net
Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:

Toggle quote (11 lines)
> I was looking at top on the hydra-guix-129 node, which runs nar-herder,
> and saw this:
>
> 4772 nar-her+ 20 0 28.9g 11.5g 100.0 6.1 55,55 S .nar-herder-rea
>
> 11.5 GiB of memory seems a bit excessive, no? Its cumulated processing
> time is also at the top of the table, and it seems stuck at 100% of CPU
> usage.
>
> Is this expected of the nar-herder, or has it gone awry?

I think hydra-guix-129 was running an older version of the nar-herder,
so I've updated it and I think that's improved the situation.

I've just reconfigured the machine to use a specific commit at the
moment, but it would be good to bump the commit that the Guix package is
using.
-----BEGIN PGP SIGNATURE-----
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=mipM
-----END PGP SIGNATURE-----

M
M
Maxim Cournoyer wrote on 11 Feb 17:27 +0100
Re: bug#66997: nar-herder uses 10 GiB of resident memory, 100% CPU on hydra-guix-129
(name . Christopher Baines)(address . mail@cbaines.net)(address . 66997@debbugs.gnu.org)
87mss7vude.fsf@gmail.com
Hi,

Christopher Baines <mail@cbaines.net> writes:

Toggle quote (20 lines)
> Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:
>
>> I was looking at top on the hydra-guix-129 node, which runs nar-herder,
>> and saw this:
>>
>> 4772 nar-her+ 20 0 28.9g 11.5g 100.0 6.1 55,55 S .nar-herder-rea
>>
>> 11.5 GiB of memory seems a bit excessive, no? Its cumulated processing
>> time is also at the top of the table, and it seems stuck at 100% of CPU
>> usage.
>>
>> Is this expected of the nar-herder, or has it gone awry?
>
> I think hydra-guix-129 was running an older version of the nar-herder,
> so I've updated it and I think that's improved the situation.
>
> I've just reconfigured the machine to use a specific commit at the
> moment, but it would be good to bump the commit that the Guix package is
> using.

The situation appears unresolved; c.f. `top' output on hydra-guix-129:

Toggle snippet (3 lines)
62775 nar-her+ 20 0 903.7g 93.9g 0.0 49.8 22,44 S `- fibers 0

It's gobbled up half of the memory of the machine so far, and will
probably keep growing slowly.

Perhaps the machine should be reconfigured & rebooted, as its generation
appears to be from Nov 23. It's been up for 41 days.

--
Thanks,
Maxim
?