Re: pflogd consume CPU

[David H. Gutteridge <dhgutteridge%sympatico.ca@localhost>]

On 2016-02-07, at 10:39 PM, David H. Gutteridge wrote:
> On Sun, Feb 07, 2016 at 23:12:02 +0100, Manuel Bouyer wrote:
>> On Sun, Feb 07, 2016 at 08:06:35PM +0100, Manuel Bouyer wrote:
>>> On Mon, Feb 08, 2016 at 12:06:54AM +0900, Miwa Susumu wrote:
>>>> 2016-02-07 1:15 GMT+09:00 Christos Zoulas <christos%astron.com@localhost>:
>>>>>>>> pflogd process consume CPU.
>>>>>>>> Because of that in load average is too high.
>>>>>>> Can you ktrace it?
>>>>>> this is kdump output.
>>>>>> sakura# kdump ktrace.out
>>>>>>  974      1 pflogd   EMUL  "netbsd"
>>>>>>  974      1 pflogd   CALL  read(3,0xbb912000,0x80000)
>>>>>>  974      1 pflogd   RET   read -1 errno 35 Resource temporarily unavailable
>>>>> And what does 'fstat -p 974' say about fd 3?
>>>> 
>>>> fd 3 is bpf.
>>> 
>>> I've seen something similar with arpwatch (might have been against bpf too),
>>> and with nagios reading a pipe (see my post today on current-users@
>>> about this).
>> 
>> I guess this is a 32bit host (i386 maybe) ?
> 
> I've been encountering this for the past few months on i386, as
> originally reported back in September. I haven't ever seen the same
> issue on evbarm, both architectures running 7.0_STABLE and using PF.
> 
> http://mail-index.netbsd.org/netbsd-users/2015/09/24/msg016939.html

I'm now encountering this on my evbarm machine as well. I'm not sure
why it's now acting up, as nothing's changed concerning the network
and nothing's changed concerning the machine other than releng
pull-ups to the netbsd-7 branch. Probably just co-incidence.

Dave