Very fresh (two days ago) 10-current becomes completely unresponsive under load

Alexander Motin mav at FreeBSD.org
Wed Jan 11 00:05:31 UTC 2012


On 01/11/12 01:21, Lev Serebryakov wrote:
> Hello, Lev.
> You wrote 11 января 2012 г., 3:05:29:
>
>>    OH! I have top running right now, when it "hangs". 0% idle time, LA
>>   becomes 20 when it have only 35 processes at all, but there is no specific
>>   process consuming CPU.
>   Ok, it seems, that here is a problem (CPU time :
>
> PID USERNAME PRI NICE   SIZE    RES STATE    TIME   WCPU COMMAND
>   12 root     -16    -     0K     8K sleep   18:35 30.18% ng_queue
>
>   And after that 5 minutes without updates, than normal numbers. It
> seems, that root of problem is:
>
>   (a) netgraph in this version of kernel
>     OR
>   (b) new mpd 5.6 (previous version had 5.5).

I remember no changes in mpd-5.6 that I would expect to cause this. Any 
way it should be trivial to check -- just build 5.5.

What do you have configured in mpd configuration and netgraph at all? 
AFAIR for plain PPPoE it is not very typical to use ng_queue at all as 
it doesn't requires stack unwrapping and at least few years ago stack 
size was sufficient to run all processing in one pass.

-- 
Alexander Motin


More information about the freebsd-net mailing list