On 7/20/22 18:50, Jakub Kicinski wrote:
> On Wed, 20 Jul 2022 17:24:17 +0200 Matthias May wrote:
>> I finally got around to do the previously mentioned selftest for gretap, vxlan and geneve.
>> See the bash-script below.
>>
>> Many of the vxlan/geneve tests are currently failing, with gretap working on net-next
>> because of the fixes i sent.
>> What is the policy on sending selftests that are failing?
>> Are fixes for the failures required in advance?
>>
>> I'm not sure i can fix them.
>> Geneve seems to ignore the 3 upper bits of the DSCP completely.
>>
>> My other concern is:
>> The whole test is... slow.
>> I tried to figure out what takes so long, and the culprit seem to be tcpdump.
>> It just takes ages to start capturing, more so when it is capturing IPv6.
>> Does anyone know of a better way to capture traffic and analyze it afterwards?
>> I used tcpdump because other tests seem to use it, and i guess this is a tool
>> that most everyone has installed (that works with networks).
> 
> Yeah, tcpdump is not great, there's a bunch of flags that make it a
> little less bad (--immediate-mode?)
> 
> Looking at the last test I wrote I ended up with:
> 
> tcpdump --immediate-mode -p -ni dummy0 -w $TMPF -c 4
> sleep 0.05 # wait for tcpdump to start

Thank you for the hint with the immediate-mode, that already shaved over a minute off.

Before:
real	3m40.056s
user	0m4.103s
sys	0m2.434s

After:
real	2m34.438s
user	0m3.955s
sys	0m2.350s

BR
Matthias