Hi Claudio Thank you for your very helpful tips. I probably had not enough sleep this night, and did not saw, that the bgpd process I killed in the morning was not relay dead. I started a new one by hand without consulting ps after kill. And this bgpd was it that run amok. Now I killed it with -9 and all is looking perfect.
So the bug was me and not bgpd ... Regards Matthias On 12/06/12 21:05, Claudio Jeker wrote: > On Tue, Jun 12, 2012 at 09:43:55AM +0200, Matthias Cramer wrote: >> Hi All >> >> I have a newly set up OpenBSD VM which mainly runs bgpd. I have 2 full IPv6 >> and 2 full IPv6 feeds and >> about 20 peerings. The sessions are all up for a bit more than 3 hours now and >> the bgpd session engine uses all the CPU it can get. >> >> In a very similar setup the CPU is under 1%. >> >> It's a 5.1 GENERIC.MP#207 amd64 install >> >> Any Idea where to begin debugging ? >> > > First of all run "bgpctl log verbose". Also try other bgpctl commands to > see if there is something funky. Check the logs. Is only the SE running at > full throttle? > > If there is nothing in the logs that may indicate what the SE is > doing it gets a bit more complex. My normal approach then is to ktrace > the process for a short while and see what syscalls happen on which fds > (fstat is of great help here as well). If it is network traffic then > tcpdump may give some insights as well. Attaching gdb is often not very > helpful on spinning processes unless there is a good understanding where > the loop most probably is (just verifying assumtions). > > Having the SE running at 100% is very uncommon (the RDE is known to > chew a lot of CPU time but the SE is just moving packets back and forth). > -- Matthias Cramer, Erachfeldstrasse 1b, CH-8180 Bülach, Switzerland http://www.freestone.net GnuPG 1024D/2D208250 = DBC6 65B6 7083 1029 781E 3959 B62F DF1C 2D20 8250 [demime 1.01d removed an attachment of type application/pgp-signature which had a name of signature.asc]
