Constant load of 1 on a recent 12-STABLE

[Daniel Ebdrup Jensen]

On Wed, Jun 03, 2020 at 10:29:29PM +0200, Gordon Bergling via freebsd-hackers wrote:
>Hi Allan,
>
>On Wed, Jun 03, 2020 at 03:13:47PM -0400, Allan Jude wrote:
>> On 2020-06-03 06:16, Gordon Bergling via freebsd-hackers wrote:
>> > since a while I am seeing a constant load of 1.00 on 12-STABLE,
>> > but all CPUs are shown as 100% idle in top.
>> >
>> > Has anyone an idea what could caused this?
>> >
>> > The load seems to be somewhat real, since the buildtimes on this
>> > machine for -CURRENT increased from about 2 hours to 3 hours.
>> >
>> > This a virtualized system running on Hyper-V, if that matters.
>> >
>> > Any hints are more then appreciated.
>> >
>> > Kind regards,
>> >
>> > Gordon
>>
>> Try running 'top -SP' and see if that shows a specific CPU being busy,
>> or a specific process using CPU time
>
>Below is the output of 'top -SP'. The only relevant process / thread that is
>relatively constant consumes CPU time seams to be 'zfskern'.
>
>-----------------------------------------------------------------------------
>last pid: 68549;  load averages:  1.10,  1.19,  1.16 up 0+14:59:45  22:17:24
>67 processes:  2 running, 64 sleeping, 1 waiting
>CPU 0:  0.0% user,  0.0% nice,  0.0% system,  0.0% interrupt,  100% idle
>CPU 1:  0.0% user,  0.0% nice,  0.0% system,  0.0% interrupt,  100% idle
>CPU 2:  0.0% user,  0.0% nice,  0.4% system,  0.0% interrupt, 99.6% idle
>CPU 3:  0.0% user,  0.0% nice,  0.0% system,  0.0% interrupt,  100% idle
>Mem: 108M Active, 4160M Inact, 33M Laundry, 3196M Wired, 444M Free
>ARC: 1858M Total, 855M MFU, 138M MRU, 96K Anon, 24M Header, 840M Other
>     461M Compressed, 1039M Uncompressed, 2.25:1 Ratio
>Swap: 2048M Total, 2048M Free
>
>  PID USERNAME    THR PRI NICE   SIZE    RES STATE    C   TIME    WCPU COMMAND
>   11 root          4 155 ki31     0B    64K RUN      0  47.3H 386.10% idle
>    8 root         65  -8    -     0B  1040K t->zth   0 115:39  12.61% zfskern
>-------------------------------------------------------------------------------
>
>The only key performance indicator that is relatively high IMHO, for a
>non-busy system, are the context switches, that vmstat has reported.
>
>-------------------------------------------------------------------------------
>procs  memory       page                    disks     faults         cpu
>r b w  avm   fre   flt  re  pi  po    fr   sr da0 da1   in    sy    cs us sy id
>0 0 0 514G  444M  7877   2   7   0  9595  171   0   0    0  4347 43322 17  2 81
>0 0 0 514G  444M     1   0   0   0     0   44   0   0    0   121 40876  0  0 100
>0 0 0 514G  444M     0   0   0   0     0   40   0   0    0   133 42520  0  0 100
>0 0 0 514G  444M     0   0   0   0     0   40   0   0    0   120 43830  0  0 100
>0 0 0 514G  444M     0   0   0   0     0   40   0   0    0   132 42917  0  0 100
>--------------------------------------------------------------------------------
>
>Any other ideas what could generate that load?
>
>Best regards,
>
>Gordon
>_______________________________________________
>freebsd-hackers at freebsd.org mailing list
>https://lists.freebsd.org/mailman/listinfo/freebsd-hackers
>To unsubscribe, send any mail to "freebsd-hackers-unsubscribe at freebsd.org"

I seem to recall bde@ (may he rest in peace) mentioning that the ULE scheduler 
had some weirdness around sometimes generating a higher load number (one of my 
systems would regularily idle at 0.60, but doesn't do it on 12.1 so I gave up 
trying to debug it) for no apparent reason, and it maybe being linked to how 
WCPU and CPU don't differ on the ULE scheduler?

Have you tried setting the kern.eventtimer.periodic sysctl to 1?

Yours,
Daniel Ebdrup Jensen
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 618 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20200603/04420be4/attachment.sig>