RELENG_8 pf stack issue (state count spiraling out of control)

[Vlad Galu]

On Tue, May 3, 2011 at 12:12 PM, Vlad Galu <dudu at dudu.ro> wrote:

>
>
> On Tue, May 3, 2011 at 11:31 AM, Vincent Hoffman <vince at unsane.co.uk>wrote:
>
>> On 03/05/2011 10:16, Jeremy Chadwick wrote:
>>
>> <snip lots of data relevant to the discussion but not my answer>
>> > Sadly I don't see a way with bsnmpd(8) to monitor things like interrupt
>> > usage, etc. otherwise I'd be graphing that.  The more monitoring the
>> > better; at least then I could say "wow, interrupts really did shoot
>> > through the roof -- the box went crazy!" and RMA the thing.  :-)
>> >
>> you could use net-mgmt/bsnmp-regex although I dont know what the
>> overhead for that is like.
>>
>
> I use munin for graphing, as it allows easy scripting without using SNMP.
>
> My case is a bit different from Jeremy's. Every once in a while there is a
> sudden traffic spike which impacts pf performance as well. However, the
> graphed figures are nowhere near what I'd consider alarming levels (this box
> has withstood more in the past). I was able to coincidentally log in after
> such a spike and noticed the pfpurge thread eating up about 30% of the CPU
> while using the normal optimization policy. In my case, it could be related
> to another issue I'm seeing on this box - mbuma allocation failures. Here
> are my graphs:
>
> http://dl.dropbox.com/u/14650083/PF/bge_bits_1-week.png
> http://dl.dropbox.com/u/14650083/PF/bge_packets_1-week.png
> http://dl.dropbox.com/u/14650083/PF/bge_stats_1-week.png
> http://dl.dropbox.com/u/14650083/PF/load-week.png
> http://dl.dropbox.com/u/14650083/PF/mbuf_errors-week.png
> http://dl.dropbox.com/u/14650083/PF/mbuf_usage-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_inserts-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_matches-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_removals-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_searches-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_src_limit-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_states-week.png
> http://dl.dropbox.com/u/14650083/PF/pf_synproxy-week.png
>
> I'll wait for the next time the symptom occurs to switch to a stateless
> configuration.
>
>
I forgot to mention this is a UP box using TSC for timekeeping and running
ntpd.

-- /boot/loader.conf --
hint.p4tcc.0.disabled="1"
hint.acpi_throttle.0.disabled="1"
debug.acpi.disabled="timer"
-- /boot/loader.conf --

-- sysctl output --
kern.timecounter.choice: TSC(800) i8254(0) dummy(-1000000)
kern.timecounter.hardware: TSC
-- sysctl output --


-- 
Good, fast & cheap. Pick any two.