Netgraph/mpd5 stability issues

[Mike Tancsa]

On 1/17/2011 11:14 AM, Eugene Grosbein wrote:
> On 17.01.2011 17:41, Mike Tancsa wrote:
> 
>>>> I also have stability issues on RELENG_8.
>>>>
>>>> http://www.freebsd.org/cgi/query-pr.cgi?pr=153497
>>>
>>> Do you change net.isr.direct from default 1 to 0 for your box?
>>
>> No, I leave it at 1
>>
>> 	---Mike
> 
> I was experiencing some instability with mpd/netgraph too
> but found workaround: increase netgraph buffers/queues.
> 
> For 4GB RAM, /boot/loader.conf:
> 
> # netgraph queue sizes tuning, see vmstat -z|egrep 'ITEM|NetGraph'
> net.graph.maxdata=65536
> net.graph.maxalloc=65536
> 
> For /etc/sysctl.conf:
> 
> kern.ipc.maxsockbuf=83886080
> net.graph.maxdgram=8388608
> net.graph.recvspace=8388608
> 
> It helps not to overflow netgraph event queues in my case
> and takes away netgraph-related stability problems.


I have only 2G of RAM on i386. On a box that has ipv6 disabled, with 641
ng interfaces right now

# vmstat -z | egrep -i 'item|netg'
ITEM                     SIZE     LIMIT      USED      FREE  REQUESTS
FAILURES
NetGraph items:            36,     4130,        2,      234, 8855752088,
       0
NetGraph data items:       36,      531,        0,      472,
20152691765,        0



This is with 21 days of uptime and its quite stable.  Enabling ipv6
seems to make the problems more likely for me.

I had to restart it as the values I chose were too high and I was getting

negative sbsize for uid = 0


 vmstat -z | egrep -i 'ITEM|Netg'
ITEM                     SIZE     LIMIT      USED      FREE  REQUESTS
FAILURES
NetGraph items:            36,     8201,        0,      236,    25397,
      0
NetGraph data items:       36,     2065,        0,      236,    54671,
      0

I tweaked things down

net.graph.maxdata=2048
net.graph.maxalloc=8192


net.inet.ip.intr_queue_maxlen=500
kern.ipc.maxsockbuf=1048576
net.graph.maxdgram=524288
net.graph.recvspace=524288

I also have

devd_enable="NO"


	---Mike