On Mon, Apr 04, 2022 at 03:14:35PM +0200, Laurent CARON wrote: > > Le 01/04/2022 à 14:38, Claudio Jeker a écrit : > > > > The numbers look reasonable with maybe the exception of prefix and BGP > > path attrs. Unless this system is pushing or pulling lots of full feeds to > > peers I would not expect such a high number of prefixes. Also the number > > of path attributes is high but that could again be reasonable if many > > different full feeds are involved. > > Hi Claudio, > > This box is terminating 3 full IPv4 + 3 full IPv6 feeds + a few dozen IX > sessions in addition to 5 IPv4 + 5 IPv6 iBGP connections.
3G is not enough for such a busy system. You need to increase your limit, 5GB is probably enough. > > > I'm not sure why the processes gets killed at around 3GB. Feels like you > > > hit the ulimit. See Stuart's mail about how to look into that. > > > So looking at this output I feel like you somehow created a BGP update > > > loop where one or more systems are constantly sending UPDATEs to each > > > other because the moment the update is processed the route decision > > > changes and flaps back resulting in a withdraw or update. > > I sincerely think it is not related to a BGP update loop because the issue > is only triggered when adding the following filters: > > include "/etc/bgpd/deny-asn.ru.bgpd" > include "/etc/bgpd/deny-asn.by.bgpd" > include "/etc/bgpd/deny-asn.ua.bgpd" > > for a total of 8265 rules > > I'll try to dig further. If you deny asns then please use an as-set instead of individual rules. -- :wq Claudio
